Probing the energy and mass dependence of cosmic-ray anisotropy with graph neural networks in IceCube

Cosmic rays are energetic particles thought to originate from explosive astrophysical events like supernovae and quasars. Upon their collision with Earth's atmosphere, a cascade of secondary particles is induced and is measurable by instruments like the IceCube Neutrino Observatory—a large ground-based detector array located at the geographic South Pole. The collected data for a given cascade is used to reconstruct information about its parent "primary" particle, such as its incident energy, arrival direction, and mass. Recently, an energy-dependent anisotropy in cosmic-ray arrival directions across the Southern Hemisphere (using IceCube data) has suggested an influence of features in the local interstellar medium on cosmic-ray flux in the TeV-PeV range. Leveraging modern machine learning techniques, we seek to probe with higher precision both the energy and mass dependence of the cosmic-ray arrival direction distribution. We expect, using a graph neural network-based reconstruction framework, to increase our understanding of the phenomenology responsible for current anisotropic observations. Knowing additionally the mass dependence of preferred cosmic-ray arrival directions will shed light on the role of turbulent interstellar magnetic fields in shaping cosmic-ray propagation within our local galactic environment.