A study of an energy-dependent anisotropy of cosmic rays beyond the GZK cut-off with deep neural networks

TL;DR

This paper enhances a deep neural network method for analyzing ultra-high-energy cosmic ray arrival directions by incorporating energy as a variable, improving the analysis of anisotropy beyond the GZK cut-off.

Contribution

It introduces an energy-inclusive neural network architecture for better analysis of cosmic ray anisotropy, accounting for energy uncertainties in experimental data.

Findings

The new neural network effectively analyzes mock UHECR maps.

Inclusion of energy improves classification robustness.

Method outlines potential for further refinement.

Abstract

In this letter, we present an update of a method for analysing arrival directions of ultra-high-energy cosmic rays (UHECRs) above the Greisen--Zatsepin--Kuz'min cut-off with a deep convolutional neural network developed originally in Kalashev, Pshirkov, Zotov (2020). Namely, we introduce energy as another variable employed in the analysis. This allows us to take into account the intrinsic uncertainties in energy of primary cosmic rays present in any experiment, which were not taken into account in the previous study, without any loss of quality of the classifier. We present the architecture of the new neural network, results of its application to mock maps of UHECR arrival directions and outline possible directions of a further improvement of the method.