Speaker
Description
Ultra-high-energy (UHE) photons are expected to originate in the extragalactic and Galactic propagation of ultra-high-energy cosmic rays (UHECRs), and their flux depends on UHECR characteristics.
UHE photons could be generated as well in superheavy dark matter decays.
Bounds on the UHE photon flux thus provide crucial constraints on cosmic-ray acceleration, propagation, composition, beyond-Standard-Model production mechanisms, and quantum gravity models.
Cosmic-ray detectors, such as the 3000 km$^2$ surface array (SD-1500) of the Pierre Auger Observatory, can observe UHE photons with energies above 10 EeV.
Their mean free path allows probing distances up to a few Mpc.
Unlike hadron-induced showers, photon-induced showers penetrate deeper in the atmosphere and have a steeper lateral distribution function, since secondary particles belong almost exclusively to the electromagnetic component.
In this work we present a convolutional neural network ensemble that improves photon-to-hadron separation beyond existing classical approaches by analyzing the temporal and spatial structure of signals recorded by the SD-1500.
As no clear candidates for UHE photons have been identified before, simulations must be used to study typical trigger patterns and observables used for discrimination.
We also evaluate possible systematic uncertainties due to the imperfect simulation of air showers and detector effects using an independent test set and a burn sample consisting of 2 % of the available data.
We present results from the unblinded analysis using 19 years of data collected during the Phase I of the Pierre Auger Observatory and discuss their implications.