Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 93: Neutrino physics without accelerators IV
T 93.8: Talk
Thursday, March 18, 2021, 17:45–18:00, Tr
Vertex Reconstruction using Graph Convolutional Networks in Double Chooz — •Markus Bachlechner, Thilo Birkenfeld, Philipp Soldin, Achim Stahl, Alexandros Tsagkarakis, and Christopher Wiebusch — RWTH Aachen University - Physics Institute III B, Aachen, Germany
Double Chooz is a reactor anti-neutrino disappearance experiment, which took data from 2011 until the end of 2017. The main purpose was the precise measurement of the neutrino mixing angle θ13 with two identical liquid scintillator detectors. Neutrinos are detected via the signature of the inverse beta decay (IBD), which is characterized by a prompt signal from a positron and a delayed signal from neutron capture. The random association of uncorrelated events caused by natural radioactivity and the β-n decay of 9Li produced by atmospheric muons are two major backgrounds. The discriminations between signal and background are based on either the spatial distance between the prompt and delayed like events or the proximity to a preceding muon track. A precise vertex reconstruction is thus important for reducing the background and improving the measurement of θ13. In this talk an approach via Graph Convolutional Networks (GCNs), which can adapt to the complex geometry and specific physical features of the detector, is presented. By using such versatile deep learning technique, the current maximum likelihood based reconstruction is outperformed.