Search for Diffuse Supernova Neutrino Background in the Full KamLAND Dataset with Neural-Network-Based Event Classification

Abstract

We report a search for the diffuse supernova neutrino background (DSNB) with the KamLAND detector, targeting electron antineutrinos via inverse beta decay in the neutrino energy range of 8.3 to 30.8 MeV. Using liquid-scintillator exposures of 9.02 kton-year for 8.3 to 9.3 MeV and 9.42 kton-year for 9.3 to 30.8 MeV, we observe seven candidate events after applying a new deep-neural-network-based event classification technique. This result is consistent with the background-only expectation of 16.2 plus or minus 9.4 events, including systematic uncertainties associated with the neural-network selection. A spectral analysis of the energy and radial distributions finds no significant excess attributable to the DSNB. We therefore set 90 percent confidence-level upper limits on the DSNB flux of 38 to 43 per square centimeter per second, depending on the assumed DSNB model. We also derive model-independent 90 percent confidence-level upper limits on the electron-antineutrino flux, obtaining some of the most stringent constraints below 13.3 MeV. Beyond the DSNB search itself, this work demonstrates neural-network-based event classification as a promising approach for suppressing neutron-associated backgrounds in liquid-scintillator neutrino detectors.