2023-11-29 15:00  P7F Seminar Room

[Journal Club] Deep-Learning-Based UHECR mass reconstruction using the Auger surface detectors

Dr. Anton Prosekin

The main obstacle for identifying the sources of Ultra-high-energy cosmic rays (UHECR) is their deflections by galactic and extragalactic magnetic fields on their way from a source to the Earth. The strength of the deflections depends on the rigidity of the particle (energy-to-charge ratio E/Z), which for the observed heavy composition is significantly different from energy. Therefore it is crucial to determine rigidity or mass (A ~ 2Z) for each primary particle to enable more sophisticated search methods for source identifications.

Several observables of extensive air showers are sensitive to the mass of the primary particle, such as the depth of the shower maximum (Xmax), muon densities (Rmu), electron-to-muon size ratio, and lateral distribution shape parameters of different shower components. The fluorescent detector (FD) can measure Xmax directly, but has small accumulated statistics due to 10 percent duty cycle. Surface detector array with 100 percent duty cycle provides much more detected events, but the reconstruction of Xmax and other relevant parameters is complicated and less reliable. Recent advancements in machine learning techniques allow to apply deep neural networks (DNN) for more accurate reconstructions employing relationships not used in human-designed analyses. Moreover, because of the fluctuations of observables it is difficult to distinguish primary nuclei of different masses based on specific observable. DNN can potentially find correlations between combinations of mass-sensitive observables that have high power in mass separation. In this talk I review recent advancements in mass reconstruction using DNN. Based on  PoS ICRC2023 (2023) 371 and PoS ICRC2023 (2023) 278.