Neural Network-Driven Electric and Magnetic Field Reconstruction for Ion Radiography

Chun-Sung Jao^1*, Y.-C. Chen², F. Nikaido³, Y.-L. Liu⁴, K. Sakai^3,5, T. Minami³, S. Isayama⁶, Y. Abe^3,7, Y. Kuramitsu^3,7

¹Department of Physics, National Cheng Kung University, Taiwan
²National Center for High-Performance Computing, National Applied Research Laboratories, Taiwan
³Graduate School of Engineering, Osaka University, Japan
⁴Institute of Space and Plasma Sciences, National Cheng Kung University, Taiwan
⁵National Institute for Fusion Science, Japan
⁶Department of Earth System Science and Technology, Kyushu University, Japan
⁷Institute of Laser Engineering, Osaka University, Japan

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

Within controlled experimental setups, particularly when employing lasers to interact with solid or gaseous targets, a wide array of phenomena can be observed in laboratory settings. These phenomena encompass shock formation, plasma instability, and magnetic reconnection. A key facet of these experiments revolves around investigating the behavior of electromagnetic fields within laboratory plasmas. To aid in the measurement of electromagnetic fields, scientists utilize a method known as ion radiography, also referred to as proton imaging or proton reflectometry. Ion radiography involves the utilization of high-energy protons generated from a point-like source. In simpler terms, these protons are directed through the plasma system. As they traverse the plasma, they experience alterations in the paths of electromagnetic fields. Eventually, these protons reach a detector, where they generate an image that unveils the patterns of electromagnetic fields within the plasma. However, a significant challenge in proton imaging lies in ascertaining the path-integrated electromagnetic fields based on the observed proton fluence. In this context, as part of our involvement in experiments led by Osaka University, we aim to employ neural network techniques for the reconstruction of electric and magnetic fields. Our goal is to develop a neural network capable of reconstructing the configurations of electric and magnetic fields based on ion images, utilizing simulation results as our training dataset.

Keywords: Laborotory astrophysics, High-energy-density (HED) plasmas, Neural Network, Machine learning, Ion radiography