Integrating Machine Learning with Physics Constraints for Electromagnetic Field Reconstruction

Chun-Sung Jao^1*, Chiung-Yin Chang², Yen-Chen Chen¹, Kun-Han Lee¹, Hsiao-Hsuan Lin¹, Han Hu¹, Akira Mizuta^3,4, Kentaro Sakai⁵, Hsiang-Yi Karen Yang², Tsung-Che Tsai¹, Yasuhiro Kuramitsu^6,7,8

¹National Center for High-Performance Computing, National Applied Research Laboratories, Hsinchu, Taiwan
²Institute of Astronomy, National Tsing Hua University, Hsinchu, Taiwan
³Astrophysical Big Bang Laboratory, RIKEN Pioneering Research Institute, Saitama, Japan
⁴RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences, Saitama, Japan
⁵National Institute for Fusion Science, Gifu, Japan
⁶Graduate School of Engineering, Osaka University, Osaka, Japan
⁷Institute of Laser Engineering, Osaka University, Osaka, Japan
⁸Kansai Photon Science Institute, National Institutes for Quantum Science and Technology, Kyoto, Japan

* Presenter:Chun-Sung Jao, email:csjao899@gmail.com

Machine learning (ML) is increasingly being explored as a tool for analyzing complex phenomena in high-energy-density plasma experiments. In this study, we explore how ML can aid in reconstructing electromagnetic fields from ion radiography data while respecting physical laws. For example, the laws of charged particle motion can be incorporated into the training process to guide the model. By combining data-driven learning with physics knowledge, our approach improves accuracy and produces physically interpretable results. We also discuss the challenges encountered and the insights gained, aiming to advance understanding of how ML can complement traditional physics-based methods in plasma diagnostics.

Keywords: High-energy-density plasma, Ion radiography, Electromagnetic field reconstruction, Machine learning