-
质子照相是探测等离子体中场分布的一个有效方法。然而由于电磁场结构的复杂性，从质子束的成像反演电磁场分布是极其困难的，往往需要对场结构做一些对称性假设。借助机器学习的方法，我们开展了不依赖对称性假设的复杂质子成像反演磁场分布的研究。我们以数十个Weibel不稳定性产生的磁场作为基元构建出的复杂磁场为反演目标，通过GEANT4计算了50000份质子照相图像，搭建并训练了一个包含三层卷积层的轻量化卷积神经网络，在质子成像的散焦区实现对具有数个不对称焦散和显著通量聚集的复杂质子成像的三维（3D）磁场反演。在场反演过程中，涉及到对包含场结构、位置、旋转等信息在内的80个参数的预测。结果显示预测值平均绝对百分比误差仅为8.5%，反演获得的磁场所计算对应的质子成像图与原测试图像具有较好的一致性。该研究表明建立更为精准的空间网格点等方法反演更一般电磁场是可能的，为强激光等离子体作用中的电磁场反演提供了一个可行的方法。
-
关键词:
- 质子照相 /
- 深度学习 /
- 磁场反演 /
- 激光等离子体相互作用
Proton radiography is an effective technique for diagnosing field distributions in plasmas. However, due to the complexity of electromagnetic field structures, reconstructing electromagnetic fields from proton radiographs is extremely challenging and often requires some simplified symmetry assumptions about the fields. Here, we present a machine learning approach to reconstruct three-dimensional (3D) magnetic field distributions from complex proton radiographs without relying on such assumptions.
To enable this, we construct the target 3D magnetic fields by linearly superposing multiple elementary magnetic structures generated from the Weibel instability. Each element is characterized by eight parameters—structural parameters (a, b, B0), spatial coordinates (x0, y0, z0), and rotation angles (θ, ϕ)—resulting in 80 degrees of freedom in total. Parameters were uniformly sampled within ±25% of their baseline values, and a dataset of 50,000 magnetic field–proton radiograph pairs was generated through forward simulation using GEANT4. All proton radiographs reside in the caustic regime, exhibiting multiple asymmetric caustics and significant flux concentrations.
A lightweight three-layer convolutional neural network (CNN) was designed for the reconstruction task. The network consists of an input layer, three convolutional modules (the first two following a ”convolution–batch normalization–max pooling” cascaded structure, and the third is simplified to a single convolutional layer), a flattening layer, a dropout layer, and an output layer. Bayesian optimization was applied to determine the optimal hyperparameters. The model was trained on 40000 samples, with 5000 samples for validation and 5000 for testing.
On the test set, the CNN achieves a mean absolute percentage error (MAPE) of 8.5% in predicting the 80 magnetic parameters, below the 12.9% random-guessing threshold. Prediction errors for most parameters follow near-zero-mean Gaussian distributions, with relative standard deviations under 6%. The reconstructed fields show high spatial agreement with the reference fields, and corresponding proton images match the originals with a cosine similarity of 0.89.
This study demonstrates that our CNN-based proton radiography reconstruction method can effectively reconstruct complex 3D magnetic fields without symmetry assumptions or manual parameter tuning, offering a novel tool for diagnosing electromagnetic fields in high-intensity laser-plasma interactions. Future work may incorporate multi-angle proton radiography and transfer learning from experimental data to enhance the method’s practicality and robustness.-
Keywords:
- Proton radiography /
- Deep learning /
- Magnetic field reconstruction /
- Laserplasma interactions
-
[1] Drake R P 2018 High-Energy-Density Physics: Foundation of Inertial Fusion and Experimental Astrophysics. Second edition. edn. (Cham: Springer Nature), pp 12–16
[2] Priest E, Forbes T 2000 Magnetic Reconnection: MHD Theory and Applications (Oxford: Cambridge University Press), pp 1–6
[3] Caprioli D, Spitkovsky A 2014 Astrophys. J. 783 91
[4] Balbus S A, Hawley J F 1998 Rev. Mod. Phys. 70 1
[5] Blandford R, Eichler D 1987 Phys. Rep. 154 1
[6] Song Y, Srinivasan B 2020 Radiat. Eff. Defects Solids 175 1009
[7] García-Rubio F, Betti R, Sanz J, Aluie H 2021 Physics of Plasmas 28 012103
[8] Samtaney R 2003 Phys. Fluids 1994 15 L53
[9] Ryu D, Jones T W, Frank A 2000 Astrophys. J. 545 475
[10] Cui Y, Yang X H, Ma Y Y, Zhang G B, Xu B H, Chen Z H, Li Z, Shao F Q, Zhang J 2024 High Power Laser Science and Engineering 12 e24
[11] Slutz S A, Herrmann M C, Vesey R A, Sefkow A B, Sinars D B, Rovang D C, Peterson K J, Cuneo M E 2010 Phys. Plasmas 17 056303
[12] Kugland N L, Ryutov D D, Plechaty C, Ross J S, Park H S 2012 Rev. Sci. Instrum. 83 101301
[13] Clark E L, Krushelnick K, Davies J R, Zepf M, Tatarakis M, Beg F N, Machacek A, Norreys P A, Santala M I K, Watts I, Dangor A E 2000 Phys. Rev. Lett. 84 670
[14] Maksimchuk A, Gu S, Flippo K, Umstadter D, Bychenkov V Y 2000 Phys. Rev. Lett. 84 4108
[15] Snavely R A, Key M H, Hatchett S P, Cowan T E, Roth M, Phillips T W, Stoyer M A, Henry E A, Sangster T C, Singh M S, Wilks S C, MacKinnon A, Offenberger A, Pennington D M, Yasuike K, Langdon A B, Lasinski B F, Johnson J, Perry M D, Campbell E M 2000 Phys. Rev. Lett. 85 2945
[16] Teng J, Zhu B, Wang J, Hong W, Yan Y H, Zhao Z Q, Cao L F, Gu Y Q 2013 Acta Phys. Sin. 62 114103 (in Chinese) [滕建, 朱斌, 王剑, 洪伟, 闫永宏, 赵宗清, 曹磊峰, 谷渝秋 2013 物理学报 62 114103]
[17] Li C K, Séguin F H, Frenje J A, Rygg J R, Petrasso R D, Town R P J, Amendt P A, Hatchett S P, Landen O L, Mackinnon A J, Patel P K, Smalyuk V A, Sangster T C, Knauer J P 2006 Physical Review Letters 97 135003
[18] Manuel M J E, Zylstra A B, Rinderknecht H G, Casey D T, Rosenberg M J, Sinenian N, Li C K, Frenje J A, Séguin F H, Petrasso R D 2012 Review of Scientific Instruments 83 063506
[19] Bott A F A, Graziani C, Tzeferacos P, White T G, Lamb D Q, Gregori G, Schekochihin A A 2017 J. Plasma Phys. 83 905830614
[20] Fryxell B, Olson K, Ricker P, Timmes F X, Zingale M, Lamb D Q, MacNeice P, Rosner R, Truran J W, Tufo H 2000 Astrophys. J. Suppl. Ser. 131 273
[21] Tzeferacos P, Fatenejad M, Flocke N, Graziani C, Gregori G, Lamb D Q, Lee D, Meinecke J, Scopatz A, Weide K 2015 High Energy Density Phys. 17 24
[22] Chittenden J P, Lebedev S V, Jennings C A, Bland S N, Ciardi A 2004 Plasma Phys. Control. Fusion 46 B457
[23] Romagnani L, Fuchs J, Borghesi M, Antici P, Audebert P, Ceccherini F, Cowan T, Grismayer T, Kar S, Macchi A, Mora P, Pretzler G, Schiavi A, Toncian T, Willi O 2005 Phys. Rev. Lett. 95 195001.1
[24] Vay J L 2008 Phys. Plasmas 15 056701
[25] Schaeffer D B, Bott A F A, Borghesi M, Flippo K A, Fox W, Fuchs J, Li C K, Séguin F H, Park H S, Tzeferacos P, Willingale L 2023 Rev. Mod. Phys. 95 1
[26] Lu Y C, Li S T, Li H, Flippo K A, Barnak D, Birkel A, Lahmann B, Li C K, Rasmus A M, Kelso K, Zylstra A, Liang E, Tzeferacos P, Lamb D 2020 Physics of Plasmas 27 012303
[27] Huntington C M, Fiuza F, Ross J S, Zylstra A B, Drake R P, Froula D H, Gregori G, Kugland N L, Kuranz C C, Levy M C, Li C K, Meinecke J, Morita T, Petrasso R, Plechaty C, Remington B A, Ryutov D D, Sakawa Y, Spitkovsky A, Takabe H, Park H S 2015 Nat. Phys. 11 173
[28] Cecchetti C A, Borghesi M, Fuchs J, Schurtz G, Kar S, Macchi A, Romagnani L, Wilson P A, Antici P, Jung R, Osterholtz J, Pipahl C A, Willi O, Schiavi A, Notley M, Neely D 2009 Physics of Plasmas 16 043102
[29] Graziani C, Tzeferacos P, Lamb D Q, Li C K 2017 Rev. Sci. Instrum. 88 123507
[30] Kasim M F, Ceurvorst L, Ratan N, Sadler J, Chen N, Sävert A, Trines R, Bingham R, Burrows P N, Kaluza M C, Norreys P 2017 Phys. Rev. E 95 023306
[31] Sulman M M, Williams J F, Russell R D 2011 Appl. Numer. Math. 61 298
[32] Schaeffer D B, Fox W, Follett R K, Fiksel G, Li C K, Matteucci J, Bhattacharjee A, Germaschewski K 2019 Phys. Rev. Lett. 122 245001
[33] Campbell P T, Walsh C A, Russell B K, Chittenden J P, Crilly A, Fiksel G, Nilson P M, Thomas A G R, Krushelnick K, Willingale L 2020 Phys. Rev. Lett. 125 1
[34] Song H Y, Wu F Y, Sheng Z M, Zhang J 2023 Physics of Plasmas 30 092707
[35] Liu D J, Tan Y X, Khoram E, Yu Z F 2018 ACS photonics 5 1365
[36] Chen N F Y, Kasim M F, Ceurvorst L, Ratan N, Sadler J, Levy M C, Trines R, Bingham R, Norreys P 2017 Phys. Rev. E 95 043305
[37] Fried B D 1959 The Physics of Fluids 2 337
[38] Weibel E S 1959 Physical Review Letters 2 83
[39] Biermann L, Schlüter A 1951 Phys. Rev. 82 863
[40] Stamper J A, Papadopoulos K, Sudan R N, Dean S O, McLean E A, Dawson J M 1971 Physical Review Letters 26 1012
[41] Levy M C, Ryutov D D, Wilks S C, Ross J S, Huntington C M, Fiuza F, Martinez D A, Kugland N L, Baring M G, Park H S 2015 Rev. Sci. Instrum. 86 033302
[42] Bai Z W, Zhou Z C, Zhao J J, Li X L, Li Z Y, Xiong F Y, Yang H K, Zhang Y Y, Xu Z Q J 2025 ArXiv:2506.04805v1 [cs.LG]
[43] Chollet F （translated by Zhang L） 2021 Deep Learning with Python. Second edition. edn. (New York: Manning), pp 233–235 (in Chinese) [弗朗索瓦 C 著 (张亮译) 2022 Python 深度学习 (第 2 版) (北京：人民邮电出版社) 第 233—235 页]
计量
- 文章访问数: 16
- PDF下载量: 0
- 被引次数: 0
下载:
