A hybrid approach to event reconstruction for atmospheric Cherenkov Telescopes combining machine learning and likelihood fitting

Abstract

The imaging atmospheric Cherenkov technique provides potentially the highest angular resolution achievable in astronomy at energies above the X-ray waveband. High-resolution measurements provide the key to progress on many of the major questions in high energy astrophysics, including the sites and mechanisms of particle acceleration to PeV energies. The huge potential of the next-generation CTA observatory in this regard can be realised with the help of improved algorithms for the reconstruction of the air-shower direction and energy.

Hybrid methods combining maximum-likelihood-fitting techniques with neural networks represent a particularly promising approach and have recently been successfully applied for the reconstruction of astrophysical neutrinos. Here, we present the FreePACT algorithm, a hybrid reconstruction method for IACTs. In this, making use of the neural ratio estimation technique from the field of likelihood-free inference, the analytical likelihood used in traditional image likelihood fitting is replaced by a neural network that approximates the charge probability density function for each pixel in the camera.

The performance of this improved algorithm is demonstrated using simulations of the planned CTA southern array. For this setupFreePACT provides significant performance improvements over analytical likelihood techniques, with improvements in angular and energy resolution of 25% or more over a wide energy range and an angular resolution as low as $\text{40}\prime \prime$ at energies above $50\mathrm{TeV}$ for observations at $20°$ zenith angle. It also yields more accurate estimations of the uncertainties on the reconstructed parameters and significantly speeds up the reconstruction compared to analytical likelihood techniques while showing the same stability with respect to changes in the observation conditions. Therefore, the FreePACT method is a promising upgrade over the current state-of-the-art likelihood event reconstruction techniques.