Classification of high energy muon bundles and single muons from the southern sky in IceCube

TL;DR

This paper presents a machine learning-based method to distinguish between atmospheric muon bundles and single muons in IceCube, improving the identification of astrophysical neutrinos from the southern sky.

Contribution

It introduces a novel classification approach using lateral and longitudinal event features with a Boosted Decision Tree trained on simulations.

Findings

Effective separation of muon bundles and single muons achieved

Enhanced identification of neutrino-induced events from the southern sky

Method improves background rejection in IceCube data analysis

Abstract

The IceCube Neutrino Observatory, located at the geographic South Pole, uses the glacial ice volume to detect astrophysical neutrinos. Detection of the neutrinos from the northern sky provides the opportunity to use a large effective volume. However, as the cross-section increases with energy, most high-energy neutrinos are absorbed by the Earth. On the other hand, probing down-going PeV neutrinos from the southern sky becomes challenging because of the large cosmic ray induced muon backgrounds. This contribution presents a method for classifying atmospheric muon bundles and single muons by analyzing the lateral and longitudinal characteristics of through-going track-like events from the southern sky. Muons generated in cosmic ray air showers form muon bundles, exhibiting a lateral spread spanning tens of meters within IceCube. We explore the time residual feature for the observed Cherenkov light to separate bundles from single muons. We also utilize energy losses along the track, which lead to fluctuations in the light intensity: bundles follow a smooth pattern, whereas single muons are more stochastic. A Boosted Decision Tree algorithm is trained on simulated, well-reconstructed cosmic ray and neutrino events to classify neutrino-induced single muons and cosmic ray bundles.