Master Thesis defense by Frederik Valentin Hansen

Title:

Combining Two Worlds Of Neutrinos

Identifying And Reconstructing Neutrinos Across Six Orders Of Magnitude In Energy Using Graph Neural Networks In The IceCube Experiment

Abstract:

The IceCube Neutrino Observatory instruments one cubic kilometer of ice, making it capable of detecting the low reactive particle, neutrino. Neutrino oscillation and multi-messenger astronomy, depend on fast and accurate classification and re- construction, of the low and high-energy neutrino events. The open-source Graph Neural Network (GNN) framework GraphNeT, developed for the purpose is shown to be able to reconstruct and classify 9013 low-energy and 71-155 high-energy neu- trino events from 1% of the IceCube data (∼ 61 million events).

The low-energy neutrinos and their backgrounds are well-simulated and abundant, this is less true for the more rare high energy neutrinos and their backgrounds, models for which are also challenged by the large number of out-of-domain events. The neutrino energies and zenith angles have been reconstructed, and it is shown that combining high and low-energy neutrino events into one reconstruction model proves to be advantageous.

The deployment of these GraphNeT models to the full IceCube dataset in Wisconsin has been explored and discussed, as a work in progress.