Inference of the Mass Composition of Cosmic Rays with energies between 3 and 100 EeV using the data of the Pierre Auger Observatory and Deep Learning

Located in Argentina, the Pierre Auger Observatory is the largest cosmic-ray observatory on Earth. The Observatory is a hybrid detector employing different detection principles to observe multiple components of air showers. The core part of the detector is the Surface Detector (SD), which comprises $1600$ water-Cherenkov detectors with $1.5\,\mathrm{km}$ spacing in an area of $3000\,\mathrm{km}^2$. The highly sensitive Fluorescence Detector (FD) overlooks the area above the SD. Since the FD can only operate on nights with good atmospheric conditions and low moon fraction, its duty cycle is limited to approximately 15%. The indirect nature of measurements of the Pierre Auger Observatory poses several challenges. For example, estimating the mass of a primary cosmic ray. The atmospheric depth of the shower maximum $X_\mathrm{max}$ is a mass-sensitive observable. The FD measures the $X_\mathrm{max}$ directly, but the statistic is limited by the duty cycle. On the contrary, the SD of the Pierre Auger Observatory, operating almost at 100% duty cycle, allows for a significant increase in the data. In this contribution, we present the $X_\mathrm{max}$ reconstruction based on deep neural networks that extends the energy range and statistics. We probe the energy evolution of the mean and standard deviation of the reconstructed $X_\mathrm{max}$, which reflects the changes in the mass composition. The features found in the average $X_\mathrm{max}$ rate suggest a heavier and purer mass composition with increasing energy.