Development of source-dependent gamma-ray image parameters using Generative Adversarial Network (GAN) for the MACE telescope and its implications

Abstract

In the field of Very High Energy (VHE) gamma-ray astronomy using the Imaging Atmospheric Cherenkov Technique (IACT), the main challenge is to elicit the faint gamma-ray signal from a huge deluge of Cosmic Ray (CR) signals. For this purpose, an extensive simulation database of γ-ray-induced extensive air shower (EAS) images across a wide energy range (30 GeV - a few 10 TeVs) must be generated. Moreover, different celestial γ-ray sources differ in flux and power-law spectral indices.

However, the position of the gamma-ray events in the image parameter hyperspace changes as per the gamma-ray source spectral profile. Hence, depending on the gamma-ray sources' spectral indices, the gamma domain cuts are subject to change. The gamma domain cuts proposed for a particular source (the Crab Nebula, for example) applied to a source of different spectral indices will inevitably result in an erroneous flux calculation of the source. The Generative Adversarial Network (GAN) is a potent unsupervised and generative Machine Learning (ML) tool to generate synthetic gamma-ray image parameters for sources with different spectral indices. This article will discuss the rationale behind developing a source-dependent database generation using the GAN and highlight its importance for the MACE telescope.