Pith. sign in

REVIEW 2 major objections 3 minor 86 references

Deep networks can predict, from the local magnetic field an ion experiences across just a few gyrations at a collisionless shock, whether that ion will be accelerated — with >90% accuracy in hybrid plasma simulations.

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 ↗

T0 review · deepseek-v4-flash

2026-08-03 20:55 UTC pith:LDH4ZEYM

load-bearing objection The >90% accuracy likely reflects label leakage: recognizing the ~45% of nonthermal particles that have already crossed the energy threshold inside the input window is enough to reach ~90% accuracy, so the paper's predictive claim is not yet demonstrated. the 2 major comments →

arxiv 2511.17363 v2 pith:LDH4ZEYM submitted 2025-11-21 astro-ph.HE astro-ph.IM

Deep Learning Analysis of Ions Accelerated at Shocks

Paxson Swierc , Damiano Caprioli , Luca Orusa , Miha Cernetic This is my paper
classification astro-ph.HE astro-ph.IM
keywords deep learningparticle accelerationcollisionless shocksdiffusive shock accelerationshock drift accelerationhybrid simulationsconvolutional neural networkautoencoder
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.

This paper asks whether a deep network can look at the short stretch of magnetic field an ion experiences while first meeting a collisionless shock and decide, before the ion's energy history is known, whether it will be boosted into the nonthermal population. The answer is yes: in hybrid simulations, a convolutional network classifies shock-accelerated ions with >90% accuracy using only local field time series. The amount of data needed differs by shock geometry — roughly one gyro-period for perpendicular shocks, about three for parallel — which independently matches the idea that the first few shock interactions seal an ion's fate. The paper also shows an autoencoder can compress and reconstruct these field and momentum tracks, which it sees as a step toward embedding kinetic injection physics inside cheaper fluid codes. If this holds, it gives a data-driven route to a practical, physical prescription for which particles get injected into shock acceleration.

Core claim

The central claim is that ion injection into shock acceleration is readable from the locally experienced magnetic field during the first shock encounter. For a perpendicular shock (shock drift acceleration only), about one gyro-period of B-field time series yields >94% test accuracy separating thermal from suprathermal ions. For a quasi-parallel shock (diffusive shock acceleration), about three gyro-periods are required to reach >90% accuracy separating nonthermal ions (E/Esh >10) from the rest. The authors argue this difference is not arbitrary: it corroborates earlier trajectory-level studies finding the first 2–3 gyrations — the SDA bootstrap — determine whether an ion escapes upstream an

What carries the argument

The load-bearing object is a convolutional neural network trained on time series of the three components of the local magnetic field (or electric field) sampled along each tracked ion's trajectory as it first encounters the shock. The network maps that short field history to the probability that the ion ends its simulation above a kinetic-energy threshold; for the parallel shock the labels are thermal+suprathermal versus nonthermal (E/Esh >10), for the perpendicular shock thermal versus suprathermal (E/Esh >2). A companion multilayer perceptron fed twelve hand-picked statistics per component shows most of the signal survives aggressive compression. An autoencoder with the same convolutional

Load-bearing premise

The target label is a finite-time energy threshold: an ion counts as nonthermal only if it exceeds E/Esh = 10 by the end of the 1000 omega_c^-1 run, so ions that would be accelerated later are silently labeled thermal; the reported >90% accuracy is accuracy against this finite-time label, not against a particle's true eventual fate.

What would settle it

Take the same quasi-parallel shock setup but extend the run (or relabel particles by whether they eventually join the power-law tail, tracked until the spectrum is converged) and retrain or evaluate the CNN on the first three gyro-periods of B-field data. If accuracy falls materially below 90%, the claimed predictability of DSA injection is an artifact of the finite-time energy threshold; if accuracy survives, the result is robust. A cheaper check: train on the perpendicular dataset but restrict the input to the first quarter gyro-period; the paper's own curve predicts accuracy should drop to

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

If this is right

  • For quasi-parallel shocks, at least ~3 gyro-periods of local B-field data are needed to classify injection with >90% accuracy; shorter windows degrade sharply, placing a physical timescale on the injection decision.
  • For perpendicular shocks, 0.5–0.75 gyro-period suffices for >90% accuracy, consistent with a single SDA interaction deciding suprathermal status.
  • An MLP using only 12 statistics per component still reaches ~85–93% accuracy, so coarse or downsampled field measurements may be enough to tag injected particles in practice.
  • Autoencoder compression of field and momentum time series works best for accelerated particles, suggesting a low-dimensional latent description of the acceleration process exists.
  • These results motivate a sub-grid model: a network trained on kinetic simulations could prescribe ion injection inside MHD-PIC or hybrid fluid treatments, replacing hand-tuned injection recipes.

Where Pith is reading between the lines

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

  • One testable extension: train the same classifier on electron tracks from full-PIC simulations; if a similar short-window field history predicts electron injection, the method gives a quantitative handle on the still-unresolved electron-to-proton injection ratio.
  • The CNN versus MLP accuracy gap implies simple field statistics carry most of the predictive information; finding which statistics (e.g., peak field, gyro-phase at re-encounter) drive the decision could yield an analytic injection criterion rather than a black box.
  • Because the input is localized in time and space, the classifier could be run inside a simulation as particles cross the shock, flagging likely injectees on the fly without waiting for the run to end — enabling injection-efficiency statistics over much larger parameter spaces.
  • If the three-gyro window is truly universal for quasi-parallel shocks, then shock simulations that do not resolve at least three ion gyrations cannot be trusted for injection statistics; the result gives a resolution floor for cheaper fluid-kinetic codes.

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

2 major / 3 minor

Summary. The manuscript applies deep learning to classify ions accelerated at collisionless shocks in hybrid simulations. Two data sets are used: a 3D perpendicular shock (M=100) where only shock drift acceleration occurs, and a 2D parallel shock (M=10) where diffusive shock acceleration produces a nonthermal tail. Ions are labeled as thermal, suprathermal, or nonthermal according to their final energy relative to E_sh at the end of the simulation. A CNN is trained to perform binary classification (thermal vs. accelerated) using the first 86 (perpendicular) or 238 (parallel) time steps of momentum, local magnetic field, or local electric field time series after the shock encounter. The central claims are that a CNN can predict injection with >94% accuracy for the perpendicular case using about one gyro-period of magnetic-field data, and with >90% accuracy for the parallel case using about three gyro-periods, and that this supports the earlier trajectory-based conclusion that the first 2–3 gyrations determine injection into DSA. An MLP with manual statistics and an autoencoder are also explored.

Significance. If the claims are correct, this is a valuable proof-of-principle: it suggests that local electromagnetic-field time series contain information about which particles become accelerated, and it points toward a route for constructing sub-grid kinetic models in fluid simulations. The paper has genuine strengths: a held-out test split is used, multiple metrics (accuracy, precision, recall, F1, ROC-AUC) are reported, three architectures are compared, and the finite-time label limitation is explicitly acknowledged. The work is exploratory rather than a definitive physical measurement, but it is the kind of first step that could open a useful line of research. However, the central quantitative claim of prospective prediction is not yet established because the classification labels are partly realized within the input window.

major comments (2)
  1. [§4.2, Table 2, Figure 6] The headline claim that the CNN 'predicts' injection is weakened by label realization within the input window. The paper states that approximately 55% of nonthermal particles have not yet reached E/E_sh>10 within the 238-step input, which implies that about 45% already have. With a test set in which nonthermal particles are roughly one third of the sample, a trivial rule that classifies only the already-nonthermal particles as positive and everything else as thermal already gives about 0.67 + 0.15 = 0.82 accuracy; adding a modest ability to guess half of the remaining nonthermal particles reaches about 0.91. The reported 91.7% accuracy for B-field input is therefore fully compatible with a network that detects already-accelerated particles rather than predicting future injection. Please report accuracy, precision, recall, and AUC separately for the subset of particles whose E/E_sh first
  2. [§4.1, Table 1, Figure 4] The same issue applies to the perpendicular case. SDA occurs within roughly one gyration, and the 86-step input window is chosen to be about one gyro-period. It is likely that a substantial fraction of suprathermal particles (E/E_sh>2) already satisfy the label criterion before the end of the window. The accuracy of 94.7% for B-field input may therefore be partly retrospective. To establish that the model is genuinely sensitive to the initial interaction rather than to the current energy of the particle, the accuracy should be stratified by when the E/E_sh>2 threshold is first crossed. This is especially important because the paper uses the time-dependence of accuracy in Figure 4 to argue that the first 0.5–0.75 gyrations are decisive; if the correctly classified particles are precisely those that have already been accelerated, that argument is circular.
minor comments (3)
  1. [§2.2 / §4.2] The text says the data are 'evenly split' among thermal, suprathermal, and nonthermal particles, but the classification problem merges thermal and suprathermal into one class. This makes the test-set prior approximately 2:1 rather than the physically relevant highly imbalanced ratio mentioned in §4.2. Please state the exact class fractions in the test set and discuss how the reported accuracy would change under the natural shock abundance of nonthermal particles.
  2. [§4.1, Figures 3 and 5] The models are selected by early stopping on validation loss ('best epoch'), and hyperparameters are tuned on validation data with Optuna. No confidence intervals, bootstrap errors, or multiple-seed results are provided. Given that the central numbers are point estimates, the absence of uncertainty quantification makes it difficult to judge whether differences of a few percent are meaningful. Please add bootstrap confidence intervals and, ideally, results for a few random seeds.
  3. [§5, Figure 8] The autoencoder results are exploratory and the authors acknowledge overfitting and limited hyperparameter tuning. Still, the sMAPE distributions are shown without error bars, and the claim that 'some models recover morphology of momentum time series they never had access to' is based on mean sMAPE values that are close to the 100 threshold. A brief quantitative statement of what sMAPE level would correspond to useful reconstruction (beyond the visual examples) would strengthen this section.

Circularity Check

0 steps flagged

Genuine held-out classification; no circular reduction found

full rationale

The paper's central claim is a supervised classification test, not a derivation: the CNN is trained on 80% of particle tracks (with 5% validation and 15% held-out test) to map time series of the local B field to a final-energy label (E/E_sh > 10 for the parallel case). The label is defined in Section 2.2 from the simulation endpoint, and the input is the earlier field history; no equation constructs the label from the input, and no fitted parameter is relabeled as a prediction. The accuracy-versus-window-length curves in Figures 4 and 6 are measured on test data, so the 1-vs-3 gyro-period result is not algebraically forced by the choice of input. References [33,39] (same group) are used to set the nonthermal threshold and to frame the gyro-period timescale, but the classifier's performance does not reduce to those citations; this is at most an interpretive self-citation, not load-bearing. The paper explicitly notes that ~55% of nonthermal particles have not crossed E/E_sh >= 10 within the 238-step input window, and acknowledges the rest are not purely prospective; that is a validity/leakage caveat rather than circularity. No circular step meets the 'exhibit the reduction' bar.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claim depends on hand-chosen thresholds (B>=5B0, energy class boundaries) and on a finite-time proxy for injection. No new physical entities are introduced. The machine-learning hyperparameters are standard and tuned via Optuna on validation data, but they are not the main source of circularity.

free parameters (3)
  • Shock encounter threshold = B >= 5 B0
    Hand-chosen threshold to define the start of the input time series (Section 6.1). Not fitted to data, but directly determines what 'first interaction' means.
  • Time series length = 86 steps (perp), 238 steps (parallel)
    Chosen to correspond to ~1 and ~3 gyro-periods. The parallel length is justified by the accuracy-vs-length curve (Fig. 6), so it is effectively selected to maximize accuracy.
  • Classification energy thresholds = E/E_sh = 2 and 10
    Boundaries between thermal/suprathermal/nonthermal, adapted from Refs. [33,39,64]. Not fitted, but they define the labels the models learn to predict.
axioms (4)
  • domain assumption The hybrid code dHybridR correctly simulates collisionless shock acceleration relevant to ion injection.
    All data come from this code; no convergence checks or validation against independent simulation codes are presented.
  • domain assumption A particle's final energy class at the end of the simulation is a valid proxy for whether it is injected into the acceleration process.
    Labels are defined by final E/E_sh thresholds; the paper acknowledges this is a finite-time proxy, and 55% of nonthermal particles have not crossed the threshold within the input window.
  • domain assumption The local magnetic field along the particle's trajectory during the first 1-3 gyrations contains sufficient information about injection.
    This is the core premise of the classification task; the paper's high accuracy supports it, but it is not independently established against other possible inputs.
  • standard math CNN/MLP/autoencoder training on a random split of particles generalizes to unseen particles from the same simulation.
    Standard supervised-learning assumption; random splitting ensures independence but not generalization across shock parameters.

pith-pipeline@v1.3.0-alltime-deepseek · 15726 in / 10052 out tokens · 97975 ms · 2026-08-03T20:55:18.916002+00:00 · methodology

0 comments
read the original abstract

We study the application of deep learning techniques to the analysis and classification of ions accelerated at collisionless shocks in hybrid (kinetic ions--fluid electrons) simulations. Ions were classified as thermal, suprathermal, or non-thermal, depending on the energy they achieved and the acceleration regime they fell under. These classifications were used to train deep learning models to predict which particles are injected into the acceleration process with high accuracy (>90%), using only time series of the local magnetic or electric field they experienced during their initial interaction with the shock. An autoencoder architecture was also tested, for which time series of various parameters were reconstructed from encoded representations. This study shows the potential of applying machine learning techniques to extract physical insights from kinetic plasma simulations and sets the groundwork for future applications, including the construction of sub-grid models in fluid approaches.

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

86 extracted references · 32 linked inside Pith

  1. [1]

    Krymskii,A regular mechanism for the acceleration of charged particles on the front of a shock wave, inAkademiia Nauk SSSR Doklady, vol

    G. Krymskii,A regular mechanism for the acceleration of charged particles on the front of a shock wave, inAkademiia Nauk SSSR Doklady, vol. 234, pp. 1306–1308, 1977

    1977

  2. [2]

    Bell,The acceleration of cosmic rays in shock fronts–i,Monthly Notices of the Royal Astronomical Society182(1978) 147

    A. Bell,The acceleration of cosmic rays in shock fronts–i,Monthly Notices of the Royal Astronomical Society182(1978) 147

    1978

  3. [3]

    Blandford and J.P

    R.D. Blandford and J.P. Ostriker,Particle acceleration by astrophysical shocks,Astrophysical Journal, Part 2-Letters to the Editor, vol. 221, Apr. 1, 1978, p. L29-L32.221(1978) L29

    1978

  4. [4]

    Axford,Acceleration of cosmic rays by shock waves, inInvited Papers: Vol

    W. Axford,Acceleration of cosmic rays by shock waves, inInvited Papers: Vol. 12, pp. 155–203, Springer (1977)

    1977

  5. [5]

    Morlino and D

    G. Morlino and D. Caprioli,Strong evidence for hadron acceleration in tycho’s supernova remnant,Astronomy & Astrophysics538(2012) A81

    2012

  6. [6]

    Caprioli,Cosmic-ray acceleration in supernova remnants: non-linear theory revised,Journal of cosmology and astroparticle physics2012(2012) 038

    D. Caprioli,Cosmic-ray acceleration in supernova remnants: non-linear theory revised,Journal of cosmology and astroparticle physics2012(2012) 038

    2012

  7. [7]

    Hillas,TOPICAL REVIEW: Can diffusive shock acceleration in supernova remnants account for high-energy galactic cosmic rays?,Journal of Physics G Nuclear Physics31(2005) 95

    A.M. Hillas,TOPICAL REVIEW: Can diffusive shock acceleration in supernova remnants account for high-energy galactic cosmic rays?,Journal of Physics G Nuclear Physics31(2005) 95

    2005

  8. [8]

    Blasi,The origin of galactic cosmic rays,A&ARv21(2013) 70 [1311.7346]

    P. Blasi,The origin of galactic cosmic rays,A&ARv21(2013) 70 [1311.7346]

    Pith/arXiv arXiv 2013

  9. [9]

    Zirakashvili and V.S

    V.N. Zirakashvili and V.S. Ptuskin,Diffusive Shock Acceleration with Magnetic Amplification by Nonresonant Streaming Instability in Supernova Remnants,ApJ678(2008) 939 [0801.4488]

    Pith/arXiv arXiv 2008

  10. [10]

    Caprioli, P

    D. Caprioli, P. Blasi, E. Amato and M. Vietri,Dynamical Effects of Self-Generated Magnetic Fields in Cosmic-Ray-modified Shocks,ApJ679(2008) L139 [0804.2884]

    Pith/arXiv arXiv 2008

  11. [11]

    Caprioli, P

    D. Caprioli, P. Blasi, E. Amato and M. Vietri,Dynamical feedback of self-generated magnetic fields in cosmic ray modified shocks,MNRAS395(2009) 895 [0807.4261]

    Pith/arXiv arXiv 2009

  12. [12]

    Caprioli,Cosmic-ray acceleration in supernova remnants: non-linear theory revised,JCAP 7(2012) 38 [1206.1360]

    D. Caprioli,Cosmic-ray acceleration in supernova remnants: non-linear theory revised,JCAP 7(2012) 38 [1206.1360]

    Pith/arXiv arXiv 2012

  13. [13]

    Haggerty and D

    C.C. Haggerty and D. Caprioli,Kinetic Simulations of Cosmic-Ray-modified Shocks. I. Hydrodynamics,ApJ905(2020) 1 [2008.12308]

    Pith/arXiv arXiv 2020

  14. [14]

    Caprioli, C.C

    D. Caprioli, C.C. Haggerty and P. Blasi,Kinetic Simulations of Cosmic-Ray-modified Shocks. II. Particle Spectra,ApJ905(2020) 2 [2009.00007]

    Pith/arXiv arXiv 2020

  15. [15]

    Caprioli,Cosmic-ray acceleration and propagation, in34th International Cosmic Ray Conference (ICRC2015), A.S

    D. Caprioli,Cosmic-ray acceleration and propagation, in34th International Cosmic Ray Conference (ICRC2015), A.S. Borisov, V.G. Denisova, Z.M. Guseva, E.A. Kanevskaya, M.G. Kogan, A.E. Morozov et al., eds., vol. 34 ofInternational Cosmic Ray Conference, p. 8, July, 2015 [1510.07042]

    Pith/arXiv arXiv 2015

  16. [16]

    Diesing and D

    R. Diesing and D. Caprioli,Steep Cosmic-Ray Spectra with Revised Diffusive Shock Acceleration,ApJ922(2021) 1 [2107.08520]. – 17 –

    Pith/arXiv arXiv 2021

  17. [17]

    Reville and A.R

    B. Reville and A.R. Bell,A filamentation instability for streaming cosmic rays,MNRAS419 (2012) 2433 [1109.5690]

    Pith/arXiv arXiv 2012

  18. [18]

    Caprioli and A

    D. Caprioli and A. Spitkovsky,Simulations of ion acceleration at non-relativistic shocks. i. acceleration efficiency,The Astrophysical Journal783(2014) 91

    2014

  19. [20]

    Birdsall and A.B

    C.K. Birdsall and A.B. Langdon,Particle simulation techniques, inComputer Applications in Plasma Science and Engineering, pp. 7–41, Springer (1991)

    1991

  20. [21]

    A. Bell, A. Robinson, M. Sherlock, R. Kingham and W. Rozmus,Fast electron transport in laser-produced plasmas and the kalos code for solution of the vlasov–fokker–planck equation, Plasma Physics and controlled fusion48(2006) R37

    2006

  21. [22]

    Amano and M

    T. Amano and M. Hoshino,Electron injection at high mach number quasi-perpendicular shocks: surfing and drift acceleration,The Astrophysical Journal661(2007) 190

    2007

  22. [23]

    Spitkovsky,Simulations of relativistic collisionless shocks: shock structure and particle acceleration, inAstrophysical Sources of High Energy Particles and Radiation, T

    A. Spitkovsky,Simulations of relativistic collisionless shocks: shock structure and particle acceleration, inAstrophysical Sources of High Energy Particles and Radiation, T. Bulik, B. Rudak and G. Madejski, eds., vol. 801 ofAmerican Institute of Physics Conference Series, pp. 345–350, Nov., 2005, DOI [astro-ph/0603211]

    Pith/arXiv arXiv 2005

  23. [24]

    Sironi and A

    L. Sironi and A. Spitkovsky,Particle Acceleration in Relativistic Magnetized Collisionless Electron-Ion Shocks,ApJ726(2011) 75 [1009.0024]

    Pith/arXiv arXiv 2011

  24. [25]

    Niemiec, M

    J. Niemiec, M. Pohl, A. Bret and V. Wieland,Nonrelativistic parallel shocks in unmagnetized and weakly magnetized plasmas,The Astrophysical Journal759(2012) 73

    2012

  25. [26]

    Riquelme and A

    M.A. Riquelme and A. Spitkovsky,Electron injection by whistler waves in non-relativistic shocks,The Astrophysical Journal733(2011) 63

    2011

  26. [27]

    Palmroth, U

    M. Palmroth, U. Ganse, Y. Pfau-Kempf, M. Battarbee, L. Turc, T. Brito et al.,Vlasov methods in space physics and astrophysics,Living reviews in computational astrophysics4(2018) 1

    2018

  27. [28]

    Lipatov,The hybrid multiscale simulation technology: an introduction with application to astrophysical and laboratory plasmas, Springer Science & Business Media (2013)

    A.S. Lipatov,The hybrid multiscale simulation technology: an introduction with application to astrophysical and laboratory plasmas, Springer Science & Business Media (2013)

    2013

  28. [29]

    Winske,Hybrid simulation codes with application to shocks and upstream waves,Space Science Reviews42(1985) 53

    D. Winske,Hybrid simulation codes with application to shocks and upstream waves,Space Science Reviews42(1985) 53

    1985

  29. [30]

    Giacalone,Large-scale hybrid simulations of particle acceleration at a parallel shock,The Astrophysical Journal609(2004) 452

    J. Giacalone,Large-scale hybrid simulations of particle acceleration at a parallel shock,The Astrophysical Journal609(2004) 452

    2004

  30. [31]

    Gargaté, R

    L. Gargaté, R. Bingham, R.A. Fonseca and L.O. Silva,dhybrid: A massively parallel code for hybrid simulations of space plasmas,Computer Physics Communications176(2007) 419

    2007

  31. [32]

    Guo and J

    F. Guo and J. Giacalone,The acceleration of thermal protons at parallel collisionless shocks: three-dimensional hybrid simulations,The Astrophysical Journal773(2013) 158

    2013

  32. [33]

    Caprioli, A.-R

    D. Caprioli, A.-R. Pop and A. Spitkovsky,Simulations and theory of ion injection at non-relativistic collisionless shocks,The Astrophysical Journal Letters798(2014) L28

    2014

  33. [34]

    Caprioli, C.C

    D. Caprioli, C.C. Haggerty and P. Blasi,Kinetic simulations of cosmic-ray-modified shocks. ii. particle spectra,The Astrophysical Journal905(2020) 2

    2020

  34. [35]

    Orusa and D

    L. Orusa and D. Caprioli,Fast particle acceleration in 3d hybrid simulations of quasiperpendicular shocks,Physical Review Letters131(2023) 095201

    2023

  35. [36]

    Orusa, D

    L. Orusa, D. Caprioli, L. Sironi and A. Spitkovsky,The role of three-dimensional effects on ion injection and acceleration in perpendicular shocks,arXiv preprint arXiv:2507.13436(2025) . – 18 –

    Pith/arXiv arXiv 2025

  36. [37]

    Orusa and V

    L. Orusa and V. Valenzuela-Villaseca,Criteria for ion acceleration in laboratory magnetized quasi-perpendicular collisionless shocks: When are 2D simulations enough?,Physics of Plasmas 32(2025) 052901 [2503.00163]

    Pith/arXiv arXiv 2025

  37. [38]

    Caprioli, L

    D. Caprioli, L. Orusa, M. Cernetic, C.C. Haggerty and B. Ostler,Acceleration of heavy ions at nonrelativistic collisionless shocks,The Astrophysical Journal Letters993(2025) L1

    2025

  38. [39]

    Caprioli, A

    D. Caprioli, A. Pop and A. Spitkovsky,Simulations and Theory of Ion Injection at Non-relativistic Collisionless Shocks,ApJ798(2015) 28 [1409.8291]

    Pith/arXiv arXiv 2015

  39. [40]

    Ball and D

    L. Ball and D. Melrose,Shock drift acceleration of electrons,Publications of the Astronomical Society of Australia18(2001) 361

    2001

  40. [41]

    Bell, K.M

    A.R. Bell, K.M. Schure and B. Reville,Cosmic ray acceleration at oblique shocks,MNRAS418 (2011) 1208 [1108.0582]

    Pith/arXiv arXiv 2011

  41. [42]

    J. Park, D. Caprioli and A. Spitkovsky,Simultaneous Acceleration of Protons and Electrons at Nonrelativistic Quasiparallel Collisionless Shocks,Physical Review Letters114(2015) 085003 [1412.0672]

    Pith/arXiv arXiv 2015

  42. [43]

    Gupta, D

    S. Gupta, D. Caprioli and A. Spitkovsky,Electron Acceleration at Quasi-parallel Nonrelativistic Shocks: A 1D Kinetic Survey,ApJ976(2024) 10 [2408.16071]

    Pith/arXiv arXiv 2024

  43. [44]

    Gupta, D

    S. Gupta, D. Caprioli and A. Spitkovsky,Speed-dependent Threshold for Electron Injection into Diffusive Shock Acceleration,arXiv e-prints(2025) arXiv:2506.09134 [2506.09134]

    Pith/arXiv arXiv 2025

  44. [45]

    Rodríguez, I

    J.-V. Rodríguez, I. Rodríguez-Rodríguez and W.L. Woo,On the application of machine learning in astronomy and astrophysics: A text-mining-based scientometric analysis,Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery12(2022) e1476

    2022

  45. [46]

    G. Li, Z. Lu, J. Wang and Z. Wang,Machine learning in stellar astronomy: Progress up to 2024,arXiv preprint arXiv:2502.15300(2025)

    Pith/arXiv arXiv 2024

  46. [47]

    Gharat and Y

    S. Gharat and Y. Dandawate,Galaxy classification: a deep learning approach for classifying sloan digital sky survey images,Monthly Notices of the Royal Astronomical Society511(2022) 5120

    2022

  47. [48]

    Soo, I.Y.K.A

    J.Y. Soo, I.Y.K.A. Shuaili and I.M. Pathi,Machine learning applications in astrophysics: Photometric redshift estimation, inAIP Conference Proceedings, vol. 2756, AIP Publishing, 2023

    2023

  48. [49]

    Stein, J

    G. Stein, J. Blaum, P. Harrington, T. Medan and Z. Lukić,Mining for strong gravitational lenses with self-supervised learning,The Astrophysical Journal932(2022) 107

    2022

  49. [50]

    Khullar, B

    G. Khullar, B. Nord, A. Ćiprijanović, J. Poh and F. Xu,Digs: deep inference of galaxy spectra with neural posterior estimation,Machine Learning: Science and Technology3(2022) 04LT04

    2022

  50. [51]

    Ćiprijanović, D

    A. Ćiprijanović, D. Kafkes, K. Downey, S. Jenkins, G.N. Perdue, S. Madireddy et al., Deepmerge–ii. building robust deep learning algorithms for merging galaxy identification across domains,Monthly Notices of the Royal Astronomical Society506(2021) 677

    2021

  51. [52]

    Dvorkin, S

    C. Dvorkin, S. Mishra-Sharma, B. Nord, V.A. Villar, C. Avestruz, K. Bechtol et al.,Machine learning and cosmology,arXiv preprint arXiv:2203.08056(2022)

    Pith/arXiv arXiv 2022

  52. [53]

    Moriwaki, T

    K. Moriwaki, T. Nishimichi and N. Yoshida,Machine learning for observational cosmology, Reports on Progress in Physics86(2023) 076901

    2023

  53. [54]

    Erdmann, J

    M. Erdmann, J. Glombitza and D. Walz,A deep learning-based reconstruction of cosmic ray-induced air showers,Astroparticle Physics97(2018) 46

    2018

  54. [55]

    Alvarado, T

    A. Alvarado, T. Capistrán, I. Torres, J. SacahuÍ and R. Alfaro,Cosmic-ray energy reconstruction using machine learning techniques,arXiv preprint arXiv:2310.06938(2023)

    Pith/arXiv arXiv 2023

  55. [56]

    C. Xu, C. McCully, B. Dong, D.A. Howell and P. Sen,Cosmic-conn: a cosmic-ray detection deep-learning framework, data set, and toolkit,The Astrophysical Journal942(2023) 73. – 19 –

    2023

  56. [57]

    Kasim, D

    M.F. Kasim, D. Watson-Parris, L. Deaconu, S. Oliver, P. Hatfield, D.H. Froula et al.,Building high accuracy emulators for scientific simulations with deep neural architecture search,Machine Learning: Science and Technology3(2021) 015013

    2021

  57. [58]

    Peterson, K.D

    J.L. Peterson, K.D. Humbird, J.E. Field, S.T. Brandon, S.H. Langer, R.C. Nora et al.,Zonal flow generation in inertial confinement fusion implosions,Physics of Plasmas24(2017)

    2017

  58. [59]

    J. Kwan, K. Heitmann, S. Habib, N. Padmanabhan, E. Lawrence, H. Finkel et al.,Cosmic emulation: fast predictions for the galaxy power spectrum,The Astrophysical Journal810 (2015) 35

    2015

  59. [60]

    Brockherde, L

    F. Brockherde, L. Vogt, L. Li, M.E. Tuckerman, K. Burke and K.-R. Müller,Bypassing the kohn-sham equations with machine learning,Nature communications8(2017) 872

    2017

  60. [61]

    Torralba Paz, A

    G. Torralba Paz, A. Bohdan and J. Niemiec,Neural networks for the analysis of traced particles in kinetic plasma simulations,AIP Advances15(2025)

    2025

  61. [62]

    Haggerty and D

    C.C. Haggerty and D. Caprioli,dhybridr: A hybrid particle-in-cell code including relativistic ion dynamics,The Astrophysical Journal887(2019) 165

    2019

  62. [63]

    Caprioli, H

    D. Caprioli, H. Zhang and A. Spitkovsky,Diffusive shock re-acceleration,JPP(2018) [1801.01510]

    Pith/arXiv arXiv 2018

  63. [64]

    Johlander, M

    A. Johlander, M. Battarbee, A. Vaivads, L. Turc, Y. Pfau-Kempf, U. Ganse et al.,Ion Acceleration Efficiency at the Earth’s Bow Shock: Observations and Simulation Results,ApJ 914(2021) 82

    2021

  64. [65]

    LeCun, L

    Y. LeCun, L. Bottou, Y. Bengio and P. Haffner,Gradient-based learning applied to document recognition,Proceedings of the IEEE86(2002) 2278

    2002

  65. [66]

    Z. Li, F. Liu, W. Yang, S. Peng and J. Zhou,A survey of convolutional neural networks: analysis, applications, and prospects,IEEE transactions on neural networks and learning systems33(2021) 6999

    2021

  66. [67]

    Krichen,Convolutional neural networks: A survey,Computers12(2023) 151

    M. Krichen,Convolutional neural networks: A survey,Computers12(2023) 151

    2023

  67. [68]

    Nair and G.E

    V. Nair and G.E. Hinton,Rectified linear units improve restricted boltzmann machines, in Proceedings of the 27th international conference on machine learning (ICML-10), pp. 807–814, 2010

    2010

  68. [69]

    Berahmand, F

    K. Berahmand, F. Daneshfar, E.S. Salehi, Y. Li and Y. Xu,Autoencoders and their applications in machine learning: a survey,Artificial intelligence review57(2024) 28

    2024

  69. [70]

    Park, J.C

    J. Park, J.C. Workman, E.G. Blackman, C. Ren and R. Siller,Particle-in-cell simulations of particle energization from low Mach number fast mode shocks,Physics of Plasmas19(2012) 062904 [1203.0074]

    Pith/arXiv arXiv 2012

  70. [71]

    Kingma and J

    D.P. Kingma and J. Ba,Adam: A method for stochastic optimization,arXiv preprint arXiv:1412.6980(2014)

    Pith/arXiv arXiv 2014

  71. [72]

    Akiba, S

    T. Akiba, S. Sano, T. Yanase, T. Ohta and M. Koyama,Optuna: A next-generation hyperparameter optimization framework, inProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2019

    2019

  72. [73]

    Hochreiter and J

    S. Hochreiter and J. Schmidhuber,Long short-term memory,Neural computation9(1997) 1735

    1997

  73. [74]

    Thomas, D

    V.A. Thomas, D. Winske and N. Omidi,Re-forming supercritical quasi-parallel shocks. I - One- and two-dimensional simulations,JGR95(1990) 18809

    1990

  74. [75]

    Amano and M

    T. Amano and M. Hoshino,Electron Injection at High Mach Number Quasi-perpendicular Shocks: Surfing and Drift Acceleration,ApJ661(2007) 190 [arXiv:astro-ph/0612204]. – 20 –

    Pith/arXiv arXiv 2007

  75. [76]

    X. Guo, L. Sironi and R. Narayan,Non-thermal electron acceleration in low mach number collisionless shocks. i. particle energy spectra and acceleration mechanism,ApJ794(2014) 153 [1406.5190]

    Pith/arXiv arXiv 2014

  76. [77]

    X. Guo, L. Sironi and R. Narayan,Non-thermal electron acceleration in low mach number collisionless shocks. ii. firehose-mediated fermi acceleration and its dependence on pre-shock conditions,ApJ797(2014) 47 [1409.7393]

    Pith/arXiv arXiv 2014

  77. [78]

    R. Xu, A. Spitkovsky and D. Caprioli,Electron Acceleration in One-dimensional Nonrelativistic Quasi-perpendicular Collisionless Shocks,ApJ897(2020) L41 [1908.07890]

    Pith/arXiv arXiv 2020

  78. [79]

    Bohdan, J

    A. Bohdan, J. Niemiec, M. Pohl, Y. Matsumoto, T. Amano and M. Hoshino,Kinetic Simulations of Nonrelativistic Perpendicular Shocks of Young Supernova Remnants. I. Electron Shock-surfing Acceleration,ApJ878(2019) 5 [1904.13153]

    Pith/arXiv arXiv 2019

  79. [80]

    Marcowith, G

    A. Marcowith, G. Ferrand, M. Grech, Z. Meliani, I. Plotnikov and R. Walder,Multi-scale simulations of particle acceleration in astrophysical systems,Living Reviews in Computational Astrophysics6(2020) 1 [2002.09411]

    Pith/arXiv arXiv 2020

  80. [81]

    Brunetti and T.W

    G. Brunetti and T.W. Jones,Cosmic rays in galaxy clusters and their nonthermal emission, International Journal of Modern Physics D23(2014) 1430007 [1401.7519]

    Pith/arXiv arXiv 2014

Showing first 80 references.