Pith. sign in

REVIEW 3 major objections 6 minor 70 references

Neural-network-based longitudinal electric field prediction in nonlinear plasma wakefield accelerators

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read A neural network predicts the on-axis wakefield of a two-bunch plasma accelerator in under 0.1 seconds.

desk verdict An honest, incremental surrogate-model paper for two-bunch PWFA beam-loading that is plausible but under-validated off-optimal; it deserves peer review and a conditional acceptance path, not a desk reject. read the letter →

arxiv 2505.04236 v1 pith:TOHO2EJ6 submitted 2025-05-07 physics.plasm-ph physics.comp-ph

classification physics.plasm-phphysics.comp-ph
keywords plasmawakefieldaccelerationbeamloadingneuralnetworksurrogatelongitudinalelectricfieldparticle-in-cellsimulationenergy-spreadoptimizationtransformerratio
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

The paper tries to show that a fully connected neural network can replace particle-in-cell simulation as the feedback provider when optimizing a two-bunch, beam-driven plasma wakefield accelerator. Given seven beam and box parameters, the network directly outputs the on-axis longitudinal electric field along the wake, and the same network can be inserted into the quasi-Newton optimization loop that tunes the trailing beam's charge until the field inside it is flat, i.e., optimal beam-loading. In the demonstrated case, the optimal normalized charge per unit length is found in under 0.1 seconds, matching the value from particle-in-cell simulation, and the predicted maximum decelerating field, average accelerating field, and transformer ratio agree with simulation to within 2 percent. If this holds beyond the single demonstration, it would make parameter scans and design optimization for plasma wakefield accelerators much cheaper.

What carries the argument

The load-bearing object is the trained fully connected feedforward neural network, with seven input neurons (the normalized charge per unit length of the drive and trailing beams, the two rms beam lengths, the beam separation, the drive-beam center position, and the simulation box length) and 256 output neurons sampling the on-axis longitudinal electric field on the original 512-point grid. The training uses mean squared error loss, rectified linear unit activations, batch normalization, and one thousand epochs; before design quantities are read off, the predicted curves are passed through a smoothing filter. The reduction from 512 to 256 output points is part of the mechanism: it lowers the prediction difficulty while keeping the field smooth enough for the optimization objective, which weights $E_z$ by the trailing-beam current profile. The network carries the argument because it supplies $E_z$ cheaply enough for the quasi-Newton optimizer to iterate without grid-based simulation.

What would settle it

Take the demonstrated case ($\Lambda_d=1.0$, $\sigma_{zd}=1.0$, $\sigma_{zt}=0.25$, $d=4.5$) and evaluate the model at off-optimal charges $\Lambda_t$ spanning, say, 0.2 to 2.8; run the corresponding particle-in-cell simulations and compare the weighted energy-spread objective $F(\Lambda_t)$ from both sources. If the model's objective misorders the optimum or deviates from simulation by more than a few percent away from the trained path, the claim that the surrogate can drive the quasi-Newton optimizer reliably would be settled in the negative.

Watch

Extended reading notes

Core claim

The central discovery is a trained surrogate model: a seven-input, three-hidden-layer, 256-output fully connected network that maps ($\Lambda_d$, $\sigma_{zd}$, $\sigma_{zt}$, $d$, $\Lambda_t$, $C_d$, $L_z$) to the on-axis longitudinal electric field $E_z(\xi)$ in the bubble regime. Trained on 25,335 quasi-static particle-in-cell datasets taken from the iterative path of a previous energy-spread-minimization study, and evaluated on held-out data with $r^2=0.90$, the model reproduces the loaded wakefield including the flattening inside the trailing beam. When used as the evaluation engine inside the quasi-Newton optimization loop for the case $\Lambda_d=1.0$, $\sigma_{zd}=1.0$, $\sigma_{zt}=0.25$, $d=4.5$, it returns the optimal trailing-beam charge $\Lambda_t=1.49$ in less than 0.1 seconds, identical to the value found with particle-in-cell feedback. From the predicted field it also reads off $W_{\rm dec}=0.520$, $W_{\rm acc}=0.482$, and the transformer ratio $R=|W_{\rm acc}/W_{\rm dec}|=0.927$, all within 2 percent of the particle-in-cell values. The paper's claim is that this makes the trained network a viable direct substitute for PIC in both optimization and design-parameter extraction.

Load-bearing premise

The load-bearing premise is that the network's predictions stay accurate at the off-optimal trial points the optimizer visits, even though the training data came mainly from the iterative path toward a single optimal beam-loading condition.

Editorial extensions

If this is right

  • For the demonstrated parameter regime, optimal beam-loading charges can be identified without running any particle-in-cell simulation in the optimization loop, shrinking time per optimization from about 7.6 minutes to under 0.1 seconds.
  • The predicted full wakefield gives design quantities directly, namely the maximum decelerating field, the average accelerating field inside the trailing beam, and the transformer ratio, so designers need not rerun a simulation after the optimization finishes.
  • Because the test case used for optimization was not in the training or validation sets, the same trained model is able to handle unseen configurations within the covered parameter ranges.
  • The approach turns a single-parameter optimization into an almost-free operation, making multi-start optimizations or sensitivity studies around the optimum practical at negligible computational cost.

Reading between the lines

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

  • The training set was harvested from the iterative path toward one optimal loading condition, so the model is most trustworthy in the neighborhood of that optimum; training on uniformly sampled off-optimal configurations would quantify how far the surrogate can stray.
  • Because the network maps continuous inputs to the whole field, its gradients with respect to inputs could be used to optimize several beam parameters simultaneously rather than only the trailing-beam charge.
  • A live experiment could use this surrogate as the fast feedback element in a real-time beam-loading controller, since the sub-0.1-second evaluation time is compatible with inter-pulse tuning in a wakefield facility.
  • The 512-to-256 point downsampling trades tail-region fidelity for trainability; predicting the less-smooth wake tail with more output points, or adding a physics-based regularizer, might extend the model beyond the beam regions.
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.

Referee Report

3 major / 6 minor

Summary. This manuscript presents a fully connected neural network that maps seven beam/plasma parameters (Λd, σzd, d, σzt, Λt, Cd, Lz) to a 256-point on-axis longitudinal electric field profile for a two-bunch beam-driven plasma wakefield accelerator in the bubble regime. The network is trained on 25,335 QuickPIC simulation cases and achieves a test-set coefficient of determination r2=0.90. The authors integrate the trained model into a BFGS optimization loop, replacing QuickPIC feedback, to find the optimal normalized charge per unit length Λt of the trailing beam; for the single demonstrated case (Λd=1.0, σzd=1.0, σzt=0.25, d=4.5) they obtain Λt=1.49 in under 0.1 s, matching the QuickPIC-optimized value and their previous fitting formula. They also use the predicted field, after Savitzky-Golay smoothing, to extract the maximum decelerating field in the drive beam, the average accelerating field in the trailing beam, and the transformer ratio, reporting agreement with PIC within about 2%.

Significance. If the claimed performance generalizes, this is a practically useful surrogate-model application: it reduces a beam-loading optimization from minutes to milliseconds and provides direct visualization of the wakefield, which the existing fitting formulas in Ref. [47] do not. Strengths of the paper include the use of an external quasi-static PIC code (QuickPIC) for training data, a clear description of the network architecture and training procedure, and a consistency check against an independent PIC simulation for the demonstrated case. The main limitation is that the central optimization claim rests on a single parameter set with no statistical evaluation of the surrogate objective at off-optimal trial points, so the magnitude of the claimed speedup is established but its reliability across the stated parameter space is not yet demonstrated.

major comments (3)
  1. [Sec. II B and Sec. III] The paper's central claim—that the neural-network surrogate can replace QuickPIC feedback inside BFGS—requires accurate predictions of the objective F(Λt) at off-optimal trial points along the optimization trajectory. However, Sec. II B states that the training dataset 'contains data from the iterative process of achieving the optimal beam-loading,' and the Conclusion explicitly concedes that the model is 'primarily suitable for giving the wakefield information in the vicinity of the optimal beam-loading conditions.' No experiment in the manuscript evaluates surrogate accuracy for Λt values away from the optimum, and the reported test-set r2=0.90 is for the 256-point Ez profiles, not for F(Λt) or its gradient. I recommend adding a direct test that samples a grid of Λt for several parameter sets, compares the NN-predicted F with QuickPIC-computed F, and demonstrates that BFGS trajectories with the NN converge to the same optimum as BFGS with PIC feedback across multiple initial conditions and parameter sets.
  2. [Sec. II B, Tables II–III] The statistical evidence for the core results is limited to a single optimization case. The test-set r2=0.90 is reported as a single number with no distribution over profiles, no per-case RMSE, and no confidence interval; Tables II and III give no uncertainties for Λt, Wdec, Wacc, or R. Because the BFGS result and the 2% agreement with PIC are the main claims, the manuscript should report error statistics over many held-out cases and, ideally, repeat the BFGS optimization with different random seeds and initial Λt values to show that the <0.1 s optimum is stable. Without this, the agreement in Tables II–III could be a favorable single draw.
  3. [Sec. II B and Sec. III, Fig. 6–7 and Table III] The Savitzky-Golay smoothing applied to the predicted field is used to extract Wdec, Wacc, and R, but the manuscript gives no window size or polynomial order for the smoother and no comparison of the unsmoothed network output. This matters because the derived R from the smoothed prediction differs from the PIC value by 0.018 (1.9%) and from the fitting formula by 0.021 (2.2%), which is slightly more than the stated 'no more than 2%' when comparing the two surrogate-based methods. A sensitivity analysis of R (and of the BFGS optimum) to the smoothing parameters is needed to show that the reported agreement is not partly an artifact of post-processing.
minor comments (6)
  1. [Sec. I] In the paragraph on normalized units, 'velocities to the spreed of light c' should read 'speed of light c.'
  2. [References] Reference [17] lists 'W. P. Leemans, B. Leemans' in the author list; the second name appears to be a duplication error and should be corrected.
  3. [Sec. III, Table III] The sentence 'The results obtained by the three methods are consistent, with no more than 2% differences' is not strictly supported by Table III: the neural-network R=0.927 differs from the fitting-formula R=0.948 by 2.2%.
  4. [Sec. III] The objective function F(Λt) is not numbered, which makes precise reference to it awkward; adding equation numbers throughout the manuscript would improve clarity.
  5. [Sec. II B, Fig. 6] The parameter notation in the subplot labels of Fig. 6 (e.g., 'd = 0.39' rather than 'Λd = 0.39') should be made consistent with the notation used in the text.
  6. [Sec. II A] The manuscript would benefit from reporting the random seed used for data shuffling and initialization, since the 60/20/20 data split and Adam training are stochastic and reproducibility would be improved.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular step identified: the network is trained on external QuickPIC field data and checked against a held-out QuickPIC case; self-citations to Ref. [47] serve as benchmarks, not as the derivation.

full rationale

The derivation chain is not circular. The neural network is a supervised map from seven beam/wake parameters to 256 on-axis Ez values, trained on 25,335 QuickPIC datasets (Sec. II B); no output variable is used as an input, and no fitted constant is renamed as a prediction. The optimal beam-loading charge is obtained by minimizing the weighted RMS field objective F(Lambda_t) with the surrogate, and the result (Lambda_t = 1.49) is checked against a QuickPIC run for a case stated to be absent from the training and validation sets (Sec. III, Table II). The comparisons with the fitting formulas of Ref. [47] are same-author benchmarks, but the load-bearing validation is the independent PIC comparison, so the self-citation is not load-bearing. The Conclusion's caveat that the model is 'primarily suitable for giving the wakefield information in the vicinity of the optimal beam-loading conditions' identifies a generalization risk for off-optimal BFGS trial points; that is a correctness limitation, not an equation-level reduction. No referenced uniqueness theorem or ansatz is imported to force the model choice, and no equation in the paper reduces to its own input by construction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The model is a purely empirical surrogate; all predictive power resides in weights fitted to QuickPIC data. No new physical entities or constants are introduced. The main assumptions are standard for the bubble-regime, tri-Gaussian-beam literature, but they bound the model's applicability to the parameter ranges and conditions listed in Table I.

free parameters (2)
  • Neural network weights and biases = trained on 25,335 QuickPIC samples
    The predictive capability of the model is entirely contained in these fitted parameters; no closed-form physical model is provided.
  • Model hyperparameters (hidden layer sizes 300-150-50, learning rate 1e-4, 1000 epochs) = as reported in Sec. II B
    Chosen by hand to achieve r2=0.90 on the test set; they are not derived from the physics and may not be optimal.
assumptions (4)
  • domain assumption Beams have tri-Gaussian density profiles
    The model inputs and training data are based on this profile (Sec. II B). Real beams may deviate from tri-Gaussian shapes.
  • domain assumption Bubble regime with Rb >> sigma_r, so the beam spot size can be neglected
    The paper assumes sigma_r does not affect the wake when the spot size is much smaller than the bubble radius (Sec. II B, citing Refs. 18 and 62).
  • domain assumption QuickPIC quasi-static simulations provide accurate ground-truth wakefields
    All training labels and validation comparisons come from QuickPIC (Sec. II B). The model inherits any limitations or errors of this simulation code.
  • domain assumption On-axis longitudinal field determines beam energy spread independent of transverse position
    This justifies using only the on-axis Ez for the optimization objective (Sec. III, citing Ref. 18).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural-network-based longitudinal electric field prediction in nonlinear plasma wakefield accelerators." pith.science (2026). https://pith.science/paper/TOHO2EJ6

@misc{pith2026250504236,
  author       = {Pith},
  title        = {Pith review of: Neural-network-based longitudinal electric field prediction in nonlinear plasma wakefield accelerators},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TOHO2EJ6}},
  note         = {Machine review of arXiv:2505.04236}
}
read the original abstract

Plasma wakefield acceleration holds remarkable promise for future advanced accelerators. The design and optimization of plasma-based accelerators typically require particle-in-cell simulations, which can be computationally intensive and time consuming. In this study, we train a neural network model to obtain the on-axis longitudinal electric field distribution directly without conducting particle-in-cell simulations for designing a two-bunch plasma wakefield acceleration stage. By combining the neural network model with an advanced algorithm for achieving the minimal energy spread, the optimal normalized charge per unit length of a trailing beam leading to the optimal beam-loading can be quickly identified. This approach can reduce computation time from around 7.6 minutes in the case of using particle-in-cell simulations to under 0.1 seconds. Moreover, the longitudinal electric field distribution under the optimal beam-loading can be visually observed. Utilizing this model with the beam current profile also enables the direct extraction of design parameters under the optimal beam-loading, including the maximum decelerating electric field within the drive beam, the average accelerating electric field within the trailing beam and the transformer ratio. This model has the potential to significantly improve the efficiency of designing and optimizing the beam-driven plasma wakefield accelerators.

Figures

Figures reproduced from arXiv: 2505.04236 by the authors.

Figure 1
Figure 1. Before model training, it is common to randomize the data and divide it into train [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Schematic of a beam-driven plasma wakefield acceleration. The beam in the front is the [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Axial distribution of (a) the decelerating wakefield within the drive beam and (b) the [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Schematic of the neural network model for the longitudinal wakefield prediction. The [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. The curves of MSE loss for the training dataset (blue line) and the validation dataset (red [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. (a)-(d) show the axial distribution of the longitudinal wakefield from the PIC simulation [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Schematic of the beam current profiles and the (predicted) longitudinal wakefield distri [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

70 extracted references · 65 canonical work pages

  1. [47]

    X. Wang, J. Gao, Q. Su, J. Wang, D. Li, M. Zeng, W. Lu, W. B. Mori, C. Joshi, and W. An, Plasma Physics and Controlled Fusion 64, 065007 (2022)

  2. [1]

    Scheinker, A

    A. Scheinker, A. Edelen, D. Bohler, C. Emma, and A. Lutman, Phys. Rev. Lett. 121, 044801 (2018)

  3. [2]

    C. Emma, A. Edelen, M. J. Hogan, B. O’Shea, G. White, and V. Yakimenko, Phys. Rev. Accel. Beams 21, 112802 (2018)

  4. [3]

    S. C. Leemann, S. Liu, A. Hexemer, M. A. Marcus, C. N. Melton, H. Nishimura, and C. Sun, Phys. Rev. Lett. 123, 194801 (2019). 15

  5. [4]

    Rescic, R

    M. Rescic, R. Seviour, and W. Blokland, Nucl. Instrum. Methods Phys. Res. Sect. A 955, 163240 (2020)

  6. [5]

    Scheinker, S

    A. Scheinker, S. Hirlaender, et al. , AIP Adv. 10, 055320 (2020)

  7. [6]

    Ramjiawan, S

    R. Ramjiawan, S. D¨ obert, J. Farmer, E. Gschwendtner, F. M. Velotti, L. Verra, G. Z. Della Porta, V. Bencini, and P. N. Burrows, Phys. Rev. Accel. Beams 25, 101602 (2022)

  8. [7]

    Roussel, A

    R. Roussel, A. Edelen, C. Mayes, D. Ratner, J. P. Gonzalez-Aguilera, S. Kim, E. Wisniewski, and J. Power, Phys. Rev. Lett. 130, 145001 (2023)

Show all 70 references
  1. [8]

    Goswami, S

    K. Goswami, S. Prasad, N. Mallick, R. Sahoo, and G. B. Mohanty, Phys. Rev. D 110, 034017 (2024)

  2. [9]

    Tajima and J

    T. Tajima and J. M. Dawson, Phys. Rev. Lett. 43, 267 (1979)

  3. [10]

    P. Chen, J. M. Dawson, R. W. Huff, and T. Katsouleas, Phys. Rev. Lett. 54, 693 (1985)

  4. [11]

    C. E. Clayton, B. E. Blue, et al. , Phys. Rev. Lett. 88, 154801 (2002)

  5. [12]

    B. E. Blue, C. E. Clayton, et al. , Phys. Rev. Lett. 90, 214801 (2003)

  6. [13]

    Muggli, B

    P. Muggli, B. E. Blue, et al. , Phys. Rev. Lett. 93, 014802 (2004)

  7. [14]

    C. G. R. Geddes, C. Toth, et al. , Nature 431, 538 (2004)

  8. [15]

    S. P. D. Mangles, C. D. Murphy, et al. , Nature 431, 535 (2004)

  9. [16]

    Faure, Y

    J. Faure, Y. Glinec, et al. , Nature 431, 541 (2004)

  10. [17]

    W. P. Leemans, B. Leemans, A. J. Gonsalves, C. T´ oth, K. Nakamura, C. G. R. Geddes, E. Esarey, C. B. Schroeder, and S. M. Hooker, Nat. Phys. 2, 696 (2006)

  11. [18]

    W. Lu, C. Huang, M. Zhou, W. B. Mori, and T. Katsouleas, Phys. Rev. Lett. 96, 165002 (2006)

  12. [19]

    Huang, W

    C. Huang, W. Lu, M. Zhou, C. E. Clayton, C. Joshi, W. B. Mori, P. Muggli, S. Deng, E. Oz, T. Katsouleas, M. J. Hogan, I. Blumenfeld, F. J. Decker, R. Ischebeck, R. H. Iverson, N. A. Kirby, and D. Walz, Phys. Rev. Lett. 99, 255001 (2007)

  13. [20]

    Blumenfeld, C

    I. Blumenfeld, C. E. Clayton, F.-J. Decker, M. J. Hogan, C. Huang, R. Ischebeck, R. Iverson, C. Joshi, T. Katsouleas, N. Kirby, et al. , Nature 445, 741 (2007)

  14. [21]

    Tzoufras, W

    M. Tzoufras, W. Lu, F. S. Tsung, C. Huang, W. B. Mori, T. Katsouleas, J. Vieira, R. A. Fonseca, and L. O. Silva, Phys. Rev. Lett. 101, 145002 (2008)

  15. [22]

    Litos, E

    M. Litos, E. Adli, W. An, C. Clarke, C. E. Clayton, S. Corde, J. Delahaye, R. England, A. Fisher, J. Frederico, et al. , Nature 515, 92 (2014)

  16. [23]

    Corde, E

    S. Corde, E. Adli, J. Allen, W. An, C. Clarke, C. Clayton, J. Delahaye, J. Frederico, S. Gessner, 16 S. Green, et al. , Nature 524, 442 (2015)

  17. [24]

    A. J. K. Gonsalves, J. Daniels, et al. , Phys. Rev. Lett. 122, 084801 (2019)

  18. [25]

    X. Zhu, M. Chen, S. Weng, T.-P. Yu, W. Wang, F. He, Z.-M. Sheng, P. McKenna, D. A. Jaroszynski, and J. Zhang, Sci. Adv. 6, eaaz7240 (2020)

  19. [26]

    Roussel, G

    R. Roussel, G. Andonian, J. B. Rosenzweig, and S. S. Baturin, Phys. Rev. Accel. Beams 23, 121303 (2020)

  20. [27]

    Y. Wu, J. Hua, Z. Zhou, J. Zhang, S. Liu, B. Peng, Y. Fang, X. Ning, Z. Nie, F. Li, C. Zhang, C.-H. Pai, Y. Du, W. Lu, W. B. Mori, and C. Joshi, Nat. Phys. 17, 801 (2021)

  21. [28]

    C. A. Lindstrøm, J. M. Garland, et al. , Phys. Rev. Lett. 126, 014801 (2021)

  22. [29]

    Ferran Pousa, I

    A. Ferran Pousa, I. Agapov, S. A. Antipov, R. W. Assmann, R. Brinkmann, S. Jalas, M. Kirchen, W. P. Leemans, A. R. Maier, A. Martinez de la Ossa, J. Osterhoff, and M. Thevenet, Phys. Rev. Lett. 129, 094801 (2022)

  23. [30]

    Bohlen, T

    S. Bohlen, T. Brummer, F. Gruner, C. A. Lindstrøm, M. Meisel, T. Staufer, M. J. V. Streeter, M. C. Veale, J. C. Wood, R. D’Arcy, K. Poder, and J. Osterhoff, Phys. Rev. Lett.129, 244801 (2022)

  24. [31]

    D’Arcy, J

    R. D’Arcy, J. Beinortaite, S. Diederichs, G. Boyle, B. Foster, M. J. Garland, P. G. Caminal, C. A. Lindstrøm, G. Loisch, S. Schreiber, S. Schroder, R. J. Shalloo, M. Thevenet, S. Wesch, M. Wing, and J. Osterhoff, Nature 603, 58 (2022)

  25. [32]

    Knetsch, I

    A. Knetsch, I. A. Andriyash, M. Gilljohann, O. Kononenko, A. Matheron, Y. Mankovska, P. San Miguel Claveria, V. Zakharova, E. Adli, and S. Corde, Phys. Rev. Lett. 131, 135001 (2023)

  26. [33]

    Picksley, J

    A. Picksley, J. Stackhouse, C. Benedetti, K. Nakamura, H. E. Tsai, R. Li, B. Miao, J. E. Shrock, E. Rockafellow, H. M. Milchberg, C. B. Schroeder, J. van Tilborg, E. Esarey, C. G. R. Geddes, and A. J. Gonsalves, Phys. Rev. Lett. 133, 255001 (2024)

  27. [34]

    Shalloo, S

    R. Shalloo, S. Dann, J. Gruse, C. Underwood, A. Antoine, C. Arran, M. Backhouse, C. Baird, M. Balcazar, N. Bourgeois, et al. , Nat. Commun. 11, 6355 (2020)

  28. [35]

    Kirchen, S

    M. Kirchen, S. Jalas, P. Messner, P. Winkler, T. Eichner, L. Hubner, T. Hulsenbusch, L. Jeppe, T. Parikh, M. Schnepp, and A. R. Maier, Phys. Rev. Lett. 126, 174801 (2021)

  29. [36]

    J. Lin, Q. Qian, J. Murphy, A. Hsu, A. Hero, Y. Ma, A. G. R. Thomas, and K. Krushelnick, Phys. Plasmas 28, 083102 (2021)

  30. [37]

    Jalas, M

    S. Jalas, M. Kirchen, P. Messner, P. Winkler, L. H¨ ubner, J. Dirkwinkel, M. Schnepp, R. Lehe, 17 and A. R. Maier, Phys. Rev. Lett. 126, 104801 (2021)

  31. [38]

    Irshad, S

    F. Irshad, S. Karsch, and A. D¨ opp, Phys. Rev. Res. 5, 013063 (2023)

  32. [39]

    Y. Qin, J. Ma, M. Jiang, C. Dong, H. Fu, L. Wang, W. Cheng, and Y. Jin, Phys. Rev. Res. 5, 033079 (2023)

  33. [40]

    Ferran Pousa, S

    A. Ferran Pousa, S. Jalas, M. Kirchen, A. Martinez de la Ossa, M. Th´ evenet, S. Hudson, J. Larson, A. Huebl, J.-L. Vay, and R. Lehe, Phys. Rev. Accel. Beams 26, 084601 (2023)

  34. [41]

    Roussel, A

    R. Roussel, A. L. Edelen, et al. , Phys. Rev. Accel. Beams 27, 084801 (2024)

  35. [42]

    Roussel, J

    R. Roussel, J. P. Gonzalez-Aguilera, E. Wisniewski, A. Ody, W. Liu, J. Power, Y.-K. Kim, and A. Edelen, Phys. Rev. Accel. Beams 27, 094601 (2024)

  36. [43]

    D. R. Jones, M. Schonlau, and W. J. Welch, J. Glob. optim. 13, 455 (1998)

  37. [44]

    Gurney, An introduction to neural networks (CRC press, 2018)

    K. Gurney, An introduction to neural networks (CRC press, 2018)

  38. [45]

    C. M. Bishop, Rev. Sci. Instrum. 65, 1803 (1994)

  39. [46]

    C. G. Broyden, Math. Comput. 21, 368 (1967)

  40. [48]

    Gordienko and A

    S. Gordienko and A. Pukhov, Phys. Plasmas 12, 515 (2005)

  41. [49]

    W. Lu, M. Tzoufras, C. Joshi, F. S. Tsung, W. B. Mori, J. Vieira, R. A. Fonseca, and L. O. Silva, Phys. Rev. ST Accel. Beams 10, 061301 (2007)

  42. [50]

    Pompili, R

    R. Pompili, R. Alesini, et al. , Nat. Phys. 17, 499 (2021)

  43. [51]

    X. Xu, T. N. Dalichaouch, J. Liu, Q. Ma, J. Pierce, K. Miller, X. Yan, and W. B. Mori, Phys. Rev. Accel. Beams 26, 111302 (2023)

  44. [52]

    J. Gu, Z. Wang, J. Kuen, L. Ma, A. Shahroudy, B. Shuai, T. Liu, X. Wang, G. Wang, J. Cai, and T. Chen, Pattern Recognit. 77, 354 (2018)

  45. [53]

    Medsker and L

    L. Medsker and L. Jain, Recurrent neural networks: design and applications (CRC Press, 1999)

  46. [54]

    Sharma, S

    S. Sharma, S. Sharma, and A. Athaiya, Towards Data Sci. 6, 310 (2017)

  47. [55]

    J. Li, J. Cheng, J. Shi, and F. Huang, in Advances in Computer Science and Information Engineering, edited by D. Jin and S. Lin (Springer Berlin Heidelberg, Berlin, Heidelberg, 2012) pp. 553–558

  48. [56]

    Ruder, arXiv preprint arXiv:1609.04747 (2016)

    S. Ruder, arXiv preprint arXiv:1609.04747 (2016)

  49. [57]

    Zhang, A

    S. Zhang, A. Choromanska, and Y. LeCun, in Advances in neural information processing 18 systems, Vol. 28 (Montreal, Canada, 2015)

  50. [58]

    Kingma and J

    D. Kingma and J. Ba, in International Conference on Learning Representations (2015)

  51. [59]

    Bjorck, C

    J. Bjorck, C. Gomes, B. Selman, and K. Q. Weinberger, in 32nd Conference on Neural Information Processing System (Montr´ eal, Canada, 2018)

  52. [60]

    G. I. Diaz, A. Fokoue-Nkoutche, G. Nannicini, and H. Samulowitz, IBM J. Res. Dev. 61, 9:1 (2017)

  53. [61]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, et al. , J. Mach. Learn. Res. 12, 2825 (2011)

  54. [62]

    W. Lu, C. Huang, M. M. Zhou, W. B. Mori, and T. Katsouleas, Phys. Plasmas 12, 063101 (2005)

  55. [63]

    W. An, V. K. Decyk, et al. , J. Comput. Phys. 250, 165 (2013)

  56. [64]

    M. J. Hogan, T. O. Raubenheimer, A. Seryi, P. Muggli, T. Katsouleas, C. Huang, W. Lu, W. An, K. A. Marsh, W. B. Mori, C. E. Clayton, and C. Joshi, New J. Phys. 12, 055030 (2010)

  57. [65]

    Joshi, E

    C. Joshi, E. Adli, et al. , Plasma Phys. Control. Fusion 60, 034001 (2018)

  58. [66]

    D’Arcy, A

    R. D’Arcy, A. Aschikhin, et al. , Phil. Trans. R. Soc. A 377, 20180392 (2019)

  59. [67]

    Joshi, S

    C. Joshi, S. Corde, and W. B. Mori, Phys. Plasmas 27, 070602 (2020)

  60. [68]

    Paszke, S

    A. Paszke, S. Gross, et al. , in 33rd Conference on Neural Information Processing Systems (Vancouver, Canada, 2019)

  61. [69]

    Bai, in SHS Web of Conferences (EDP Sciences, 2022)

    Y. Bai, in SHS Web of Conferences (EDP Sciences, 2022)

  62. [70]

    W. H. Press and S. A. Teukolsky, Comput. Phys. 4, 669 (1990). 19

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.