pith. machine review for the scientific record. sign in

arxiv: 2603.02733 · v2 · submitted 2026-03-03 · ✦ hep-ex

Recognition: 2 theorem links

· Lean Theorem

Two-stage Convolutional Neural Network for pseudo six-dimensional phase space reconstruction

Sayantan Mukherjee , Masao Kuriki , Zachary John Liptak , Hitoshi Hayano , Masakazu Kurata , Nobuhiro Terunuma , Toshiyuki Okugi , Yasuchika Yamamoto
Authors on Pith no claims yet

Pith reviewed 2026-05-15 17:06 UTC · model grok-4.3

classification ✦ hep-ex
keywords convolutional neural networkphase space reconstructionbeam diagnosticssix-dimensional phase spaceparticle acceleratorKEK-ATFelectron gun
0
0 comments X

The pith

A two-stage CNN reconstructs pseudo 6D phase space at the cathode from sixteen transverse screen images.

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

The paper presents a two-stage convolutional neural network trained on accelerator simulations to recover the full six-dimensional beam phase space from a small set of two-dimensional screen images. Sixteen x-y images are acquired at a dispersive location by stepping the RF phase and solenoid field to rotate the phase space, then fed to the network. The output is a set of fifteen pairwise 2D projections that together represent the pseudo 6D distribution right at the cathode surface. When applied to real KEK-ATF injector data the reconstructed time width and spatial spread match independent measurements. This route cuts the number of shots and the post-processing load relative to conventional tomography.

Core claim

The two-stage CNN reconstructs a pseudo 6D phase space distribution at the cathode surface expressed through 15 two-dimensional distributions covering all pairwise coordinate combinations, with time width and spatial spread consistent with measured values at KEK-ATF.

What carries the argument

Two-stage convolutional neural network that maps sixteen transverse x-y screen images taken at controlled phase-space rotation angles into the full set of pairwise 2D phase-space projections.

Load-bearing premise

Simulation data generated with ASTRA for the KEK-ATF injector accurately captures the real beam dynamics so that a network trained on it can correctly interpret experimental images.

What would settle it

A mismatch between the model's predicted time width or rms spatial spread and independent cathode or downstream measurements on the same beam would falsify the reconstruction claim.

read the original abstract

In particle accelerators, broad characterization of the six-dimensional (6D) beam phase space is crucial but difficult to obtain with conventional beam diagnostics. We develop a two-stage convolutional neural network (CNN) that reconstructs the 6D phase space from only sixteen transverse $x-y$ screen images taken at a place with dispersion by different phase space rotation angles. The model is trained with simulation data of KEK-Accelerator Test Facility (ATF) injector with ASTRA. The real-space images in the chicane orbit at the KEK-ATF injector were acquired by varying the RF phase of the RF electron gun and the solenoid magnetic field. From these data, we reconstructed a pseudo 6D phase space distribution at the cathode surface, expressed through 15 two-dimensional (2D) distributions covering all pairwise coordinate combinations. The time width and spatial spread of the electron beam at the cathode showed values consistent with the measured values at KEK-ATF. Compared to existing 6D beam imaging measurement techniques such as tomography, it significantly reduces measurement time and required computational resources, enabling the provision of a more practical 6D phase space measurement method.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 1 minor

Summary. The paper develops a two-stage convolutional neural network (CNN) trained exclusively on ASTRA simulations of the KEK-ATF injector to reconstruct a pseudo-6D phase space distribution at the cathode surface from sixteen transverse x-y screen images acquired at different RF phases and solenoid settings. The output is expressed as fifteen pairwise 2D distributions covering all coordinate combinations; when applied to real experimental images, the reconstructed time width and spatial spread are reported as consistent with independent KEK-ATF measurements. The method is positioned as a practical alternative to tomography that reduces measurement time and computational cost.

Significance. If the simulation-to-reality transfer is reliable, the approach would provide a resource-efficient route to 6D beam characterization in accelerators, addressing a long-standing diagnostic challenge. The two-stage CNN architecture and use of varied rotation angles represent a concrete technical contribution that could be adopted more broadly if the validation limitations are addressed.

major comments (1)
  1. [Results] Results/validation section: The only reported consistency check is that the reconstructed time width and spatial spread match independent measurements. Because no ground-truth 6D distribution exists for the real beam, agreement on these two marginal quantities does not constrain the accuracy of the cross-plane correlations or higher moments that constitute the novel pseudo-6D claim. The central assertion therefore rests on an untested assumption that ASTRA faithfully captures all relevant beam dynamics.
minor comments (1)
  1. [Introduction] The term 'pseudo 6D' is used throughout without an explicit definition; a short clarifying sentence in the introduction would improve readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for highlighting the validation limitations in our results. We agree that agreement on marginal quantities alone does not fully constrain the cross-plane correlations, and we have revised the manuscript to address this explicitly.

read point-by-point responses

  1. Referee: [Results] Results/validation section: The only reported consistency check is that the reconstructed time width and spatial spread match independent measurements. Because no ground-truth 6D distribution exists for the real beam, agreement on these two marginal quantities does not constrain the accuracy of the cross-plane correlations or higher moments that constitute the novel pseudo-6D claim. The central assertion therefore rests on an untested assumption that ASTRA faithfully captures all relevant beam dynamics.

    Authors: We agree that the experimental validation is necessarily limited because no ground-truth 6D distribution is available for the real beam. The reported consistency checks are confined to the time width and spatial spread, which are independently measurable but do not directly test the cross-plane correlations. In the revised manuscript we have added a new paragraph in the Results section that explicitly states this limitation, clarifies that the reconstruction relies on the fidelity of the ASTRA model for the KEK-ATF injector, and notes that the method yields a practical pseudo-6D approximation rather than a fully validated experimental 6D distribution. We also reference prior literature on ASTRA benchmarking for similar gun and solenoid configurations to support the simulation assumptions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external simulation-to-experiment transfer

full rationale

The paper trains a two-stage CNN exclusively on ASTRA-generated simulation data of the KEK-ATF injector to learn a mapping from 16 transverse x-y screen images (taken at varied RF phase and solenoid settings) to 15 pairwise 2D projections of the 6D phase space at the cathode. This learned mapping is then applied to real experimental images. The only reported consistency check is that reconstructed marginal time width and spatial spread match independent measurements. No equations or steps reduce the output to the inputs by construction, no parameters are fitted to a data subset and then called a prediction, and no load-bearing self-citations or uniqueness theorems are invoked. The central claim therefore rests on the (external) assumption that ASTRA captures the relevant beam dynamics sufficiently for generalization, rather than on any internal definitional loop or self-referential fit.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that ASTRA simulations accurately model real accelerator conditions at KEK-ATF and that the CNN generalizes from training data to experimental inputs.

axioms (1)
  • domain assumption Convolutional neural networks can learn mappings from 2D images to higher-dimensional phase space distributions
    Implicit in the use of CNN for this reconstruction task.

pith-pipeline@v0.9.0 · 5534 in / 1063 out tokens · 50006 ms · 2026-05-15T17:06:20.263218+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 6 internal anchors

  1. [1]

    Ties Behnke, James E. Brau, Brian Foster, Juan Fuster, Mike Harrison, James McEwan Paterson, Michael Peskin, Marcel Stanitzki, Nicholas Walker, and Hitoshi Yamamoto, The International Linear Collider Tech- nical Design Report – Volume 1: Executive Summary, Technical Report DESY-13-062, DESY (2013), arXiv:1306.6327

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  2. [2]

    (Springer, Berlin, Heidelberg, 2003)

    Helmut Wiedemann,Synchrotron Radiation, Advanced Texts in Physics. (Springer, Berlin, Heidelberg, 2003)

    work page 2003

  3. [3]

    Zhirong Huang and Kwang-Je Kim, Phys. Rev. ST Accel. Beams,10, 034801 (2007)

    work page 2007

  4. [4]

    Vivek Maradia, Benjamin Clasie, Emma Snively, Katia Parodi, Marco Schwarz, and Marco Durante, Nature Physics,21, 1363–1373 (2025)

    work page 2025

  5. [5]

    J.G. Wang, P. Haldemann, D. Kehne, M. Reiser, D.X. Wang, E. Boggasch, and T. Shea, IEEE Transactions on Electron Devices,37(12), 2622 – 2628 (1990)

    work page 1990

  6. [6]

    Wang, D.X

    J.G. Wang, D.X. Wang, and M. Reiser, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,307(2), 190–194 (1991)

    work page 1991

  7. [7]

    Guzenko, Olivier R

    Benedikt Hermann, Vasily A. Guzenko, Olivier R. Huerzeler, Srdjan Dordevic, Leonardo Verra, Eduard Prat, and Victor Schlott, Physical Review Accelerators and Beams,24(2), 022802 (February 2021)

    work page 2021

  8. [8]

    Wong, Andrei Shishlo, Alexander Aleksandrov, Yun Liu, and Cary Long, Physical Review Accelerators and Beams,25(4), 042801 (April 2022)

    Jonathan C. Wong, Andrei Shishlo, Alexander Aleksandrov, Yun Liu, and Cary Long, Physical Review Accelerators and Beams,25(4), 042801 (April 2022)

    work page 2022

  9. [9]

    McKee, P.G

    C.B. McKee, P.G. O’, and J.M.J. Madey, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment,358(1), 264–267 (1995)

    work page 1995

  10. [10]

    Hancock, M

    S. Hancock, M. Lindroos, E. McIntosh, and M. Metcalf, Computer Physics Communications,118(1), 61–70 (1999)

    work page 1999

  11. [11]

    Stratakis, K

    D. Stratakis, K. Tian, R. A. Kishek, I. Haber, M. Reiser, and P. G. O’Shea, Phys. Plasmas,14, 120703 (2007)

    work page 2007

  12. [12]

    Diktys Stratakis,Tomographic measurement of the phase space distribution of a space-charge-dominated beam, PhD thesis, Maryland U. (2008)

    work page 2008

  13. [13]

    Yakimenko, M

    V. Yakimenko, M. Babzien, I. Ben-Zvi, R. Malone, and X.-J. Wang, Phys. Rev. ST Accel. Beams,6, 122801 (Dec 2003)

    work page 2003

  14. [14]

    K. M. Hock and M. G. Ibison, JINST,8, P02003 (2013)

    work page 2013

  15. [15]

    J. J. Scheins, Tomographic reconstruction of transverse and longitudinal phase space distributions using the maximum entropy algorithm, TESLA Report 2004-08, DESY (May 2004), DESY-TESLA-2004-08

    work page 2004

  16. [16]

    Townsend and Michel Defrise,Image-reconstruction methods in positron tomography, CERN Academic Training Lectures

    David W. Townsend and Michel Defrise,Image-reconstruction methods in positron tomography, CERN Academic Training Lectures. (CERN, Geneva, 1993), CERN Yellow Report CERN-1993-002 (also cited as CERN-93-02); CERN, Geneva, 8–11 Mar 1993

    work page 1993

  17. [17]

    Wong, Andrei Shishlo, Alexander Aleksandrov, Yun Liu, and Cary Long, Phys

    Jonathan C. Wong, Andrei Shishlo, Alexander Aleksandrov, Yun Liu, and Cary Long, Phys. Rev. Accel. Beams, 25, 042801 (Apr 2022)

    work page 2022

  18. [18]

    Cathey, S

    B. Cathey, S. Cousineau, A. Aleksandrov, and A. Zhukov, Physical Review Letters,121(6), 064804 (2018)

    work page 2018

  19. [19]

    Dong Wang, Xiaodong Wang, and Shaohe Lv, Symmetry,11(8) (2019)

    work page 2019

  20. [20]

    Breunig, Hans-Peter Kriegel, and J¨ org Sander, SIGMOD Rec.,28(2), 49–60 (June 1999)

    Mihael Ankerst, Markus M. Breunig, Hans-Peter Kriegel, and J¨ org Sander, SIGMOD Rec.,28(2), 49–60 (June 1999)

    work page 1999

  21. [21]

    Hironobu Fujiyoshi, Tsubasa Hirakawa, and Takayoshi Yamashita, IATSS Research,43(4), 244–252 (2019)

    work page 2019

  22. [22]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, Deep residual learning for image recognition, In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778 (2016)

    work page 2016

  23. [23]

    Pierre Baldi, Autoencoders, unsupervised learning, and deep architectures, InICML Unsupervised and Transfer Learning(2011)

    work page 2011

  24. [24]

    Ryan Roussel, Juan Pablo Gonzalez-Aguilera, Eric Wisniewski, Alexander Ody, Wanming Liu, John Power, Young-Kee Kim, and Auralee Edelen, Phys. Rev. Accel. Beams,27, 094601 (Sep 2024)

    work page 2024

  25. [25]

    Wolski, M

    A. Wolski, M. A. Johnson, M. King, B. L. Militsyn, and P. H. Williams, Phys. Rev. Accel. Beams,25, 122803 (Dec 2022)

    work page 2022

  26. [26]

    Alexander Scheinker, Frederick Cropp, and Daniele Filippetto, Phys. Rev. E,107, 045302 (Apr 2023)

    work page 2023

  27. [27]

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

    work page 2018

  28. [28]

    K M Hock and M G Ibison, Journal of Instrumentation,8(02), P02003 (feb 2013)

    work page 2013

  29. [29]

    Humaidi, Ayad Al-Dujaili, Ye Duan, Omran Al-Shamma, J

    Laith Alzubaidi, Jinglan Zhang, Amjad J. Humaidi, Ayad Al-Dujaili, Ye Duan, Omran Al-Shamma, J. Santa- mar´ ıa, Mohammed A. Fadhel, Muthana Al-Amidie, and Laith Farhan, Journal of Big Data,8, Article number: 53 (December 2021)

    work page 2021

  30. [30]

    Jiuxiang Gu, Zhenhua Wang, Jason Kuen, Lianyang Ma, Amir Shahroudy, Bing Shuai, Ting Liu, Xingxing Wang, Gang Wang, Jianfei Cai, and Tsuhan Chen, Pattern Recognition,77, 354–377 (2018)

    work page 2018

  31. [31]

    Guangle Yao, Tao Lei, and Jiandan Zhong, Pattern Recognition Letters,118, 14–22, Cooperative and Social Robots: Understanding Human Activities and Intentions (2019)

    work page 2019

  32. [32]

    Karen Simonyan and Andrew Zisserman, arXiv preprint arXiv:1409.1556 (2014). 31

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  33. [33]

    Sayantan Mukherjee, Masao Kuriki, Keisuke Date, Zachary J. Liptak, Hitoshi Hayano, Kirk Yamamoto, Naoto Yamamoto, Masafumi Fukuda, Xiuguang Jin, Masakazu Kurata, and Kazuyuki Sakaue, A study for envelope matching of electron bunch for emittance repartitioning experiment in KEK–STF, InProceedings of the 20th Annual Meeting of the Particle Accelerator Socie...

    work page 2023

  34. [34]

    Keiron O’Shea and Ryan Nash, ArXiv e-prints (11 2015)

    work page 2015

  35. [35]

    Floettmann, ASTRA: A space–charge tracking algorithm, Technical report, Deutsches Elektronen- Synchrotron (DESY), Hamburg, Germany (03 2017), Version 3.2 software manual

    K. Floettmann, ASTRA: A space–charge tracking algorithm, Technical report, Deutsches Elektronen- Synchrotron (DESY), Hamburg, Germany (03 2017), Version 3.2 software manual

    work page 2017

  36. [36]

    Yi Fang, Sundar Murugappan, and Karthik Ramani, BMC medical imaging,10, 12 (06 2010)

    work page 2010

  37. [37]

    Umberto Michelucci, An introduction to autoencoders (2022), arXiv:2201.03898

    work page arXiv 2022

  38. [38]

    Abien Fred Agarap, arXiv preprint arXiv:1803.08375 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  39. [39]

    Attention Is All You Need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin, Attention is all you need (2023), arXiv:1706.03762

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  40. [40]

    Yassine Moussaoui, J´ erˆ ome Idier, Anne Laure Lecocq, Catherine Riddell, C´ eline Blondeau, and Adrien Bousse, arXiv preprint arXiv:2503.21825 (2025)

    work page arXiv 2025

  41. [41]

    Hang Zhao, Orazio Gallo, Iuri Frosio, and Jan Kautz, IEEE Transactions on Computational Imaging,3(1), 47–57 (2017)

    work page 2017

  42. [42]

    Yuanjun Luo, Alexander G. Schwing, and Raquel Urtasun, Cosine normalization: Using cosine similarity instead of dot product in neural networks, InProceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS), pages 879–887 (2018)

    work page 2018

  43. [43]

    R. A. Powell, W. E. Spicer, G. B. Fisher, and P. Gregory, Phys. Rev. B,8, 3987–3995 (Oct 1973)

    work page 1973

  44. [44]

    Curtis, and Jorge Nocedal, SIAM Review,60, 223–311 (2018)

    L´ eon Bottou, Frank E. Curtis, and Jorge Nocedal, SIAM Review,60, 223–311 (2018)

    work page 2018

  45. [45]

    Adam: A Method for Stochastic Optimization

    Diederik P. Kingma and Jimmy Ba, Adam: A method for stochastic optimization (2017), arXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  46. [46]

    Cyclical Learning Rates for Training Neural Networks

    Leslie N. Smith, Cyclical learning rates for training neural networks (2017), arXiv:1506.01186

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  47. [47]

    Morgan Kaufmann Publishers Inc

    Ron Kohavi, A study of cross-validation and bootstrap for accuracy estimation and model selection, In Proceedings of the 14th International Joint Conference on Artificial Intelligence - Volume 2, IJCAI’95, page 1137–1143, San Francisco, CA, USA (1995). Morgan Kaufmann Publishers Inc

    work page 1995

  48. [48]

    STONE, Journal of the Royal Statistical Society

    M. STONE, Journal of the Royal Statistical Society. Series B (Methodological),36(2), 111–147 (1974)

    work page 1974

  49. [49]

    Press, Saul A

    William H. Press, Saul A. Teukolsky, William T. Vetterling, and Brian P. Flannery,Numerical Recipes in C: The Art of Scientific Computing, (Cambridge University Press, 2nd edition, 1992)

    work page 1992

  50. [50]

    ATF Collaboration, Aiming for nanobeams: Accelerator Test Facility, Technical Report ATF-E-1227, KEK (2011)

    work page 2011

  51. [51]

    T. Okugi and ATF International Collaboration, Achievement of small beam size at ATF2 beamline, In Proceedings of the 28th International Linear Accelerator Conference (LINAC’16), pages 27–31, East Lansing, MI, USA (September 2016), Contribution MO3A02, doi:10.18429/JACoW-LINAC2016-MO3A02

    work page doi:10.18429/jacow-linac2016-mo3a02 2016

  52. [52]

    K00 K01 K10 K11 # .(B2) At first, the 2×2 sub-matrix ofIat rows 0−1, columns 0−1 is taken, namely,

    K. Popov, A. Aryshev, and N. Terunuma, Journal of Physics: Conference Series,3094(1), 012044 (sep 2025). 32 A Appendix: Experimental images with lens positions−1300µm and −1000µm We used three different zoom lens positions for generating different laser spot sizes at the cathode. In the Sec. 5, the measurement images at chicane corresponding to lens posi-...

    work page 2025