Karlsruhe 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 67: Data, AI, Computing 5 (normalising flows)
T 67.3: Talk
Wednesday, March 6, 2024, 16:30–16:45, Geb. 30.33: MTI
Normalizing Flows to Infer Ultra-High-Energy Cosmic-Ray Source Properties from Surface Detector Measurements at the Pierre Auger Observatory — •Frederik Krieger1, Teresa Bister2, Martin Erdmann1, and Josina Schulte1 — 1III. Physikalisches Institut A, RWTH Aachen University — 2Institute for Mathematics, Astrophysics and Particle Physics, Radboud Universiteit Nijmegen
The energy spectrum and the depth of shower maximum Xmax distributions of ultra-high-energy cosmic rays (UHECRs) are measured at the Pierre Auger Observatory. Since Xmax is correlated to the mass of the primary cosmic ray, these measurements are used to constrain the astrophysical parameters of UHECR source models. These parameters include the spectral index and the maximum energy of the injected spectrum, and the initial mass composition. Owing to the stochastic nature of interactions during propagation, simple inversion of the process from source to Earth is not possible.
For parameter inference, we apply conditional invertible neural networks, a method based on normalizing flows. In comparison to the frequently used Markov Chain Monte Carlo method, they act as likelihood-free estimators. We investigate the influence of higher event statistics of the Xmax distributions, which can now be extracted from the surface detector data of the Pierre Auger Observatory using deep learning. Our results indicate that the increased statistics lead to stronger constraints on the astrophysical parameters and to enhanced sensitivity to experimental systematic effects.
Keywords: Pierre Auger Observatory; ultra-high-energy cosmic rays; astrophysical model; machine learning; parameter inference