Machine learning approach to the background reduction in singly charged cosmic-ray isotope measurements with AMS-02

Abstract

Studying the isotopic composition of single-charge cosmic rays (CRs) provides essential data to investigate the CR propagation processes in our Galaxy. While current measurements are rare above 4 GeV/nucleon, the Alpha Magnetic Spectrometer (AMS-02) is able to measure the isotopic fluxes up to 10 GeV/n by combining the momentum measured by the silicon tracker with the precise measurements of the velocity provided by its Ring Imaging Cherenkov Detector (RICH). The correct measurement of the particles’ velocity is essential for identifying isotopes through their mass. This is particularly challenging for single-charge particles due to the low number of photons they produce in their Cherenkov rings, which makes the reconstruction easily disrupted by noise. Hence, identifying the sources and cleaning the sample from the background is essential for ensuring the quality of the rings. In this paper, we propose a novel approach to track the events whose mass is misidentified due to interactions inside the AMS-02 detector. Based on the actual location of these interactions, we propose a novel strategy to mitigate the background effectively and with high efficiency, which includes using cut-based selection criteria and a multivariate estimator based on the signals detected by the RICH.