Background rejection with neural network for KamLAND-Zen

Atsuto Takeuchi
2020 Zenodo  
KamLAND-Zen is a neutrinoless double beta decay $(0\nu\beta\beta)$ search experiment using $^{136}$Xe. Taking advantages of the low-background environment of KamLAND, we realize the most sensitive $0\nu\beta\beta$ search. While $0\nu\beta\beta$ is a pure $\beta$ event, the backgrounds such as $^{214}$Bi and spallation products emit $\gamma$-rays. Therefore, particle identification(PID) is effective to improve the sensitivity. In this study, we develop a PID method with a neural network focusing
more » ... on difference of scintillation timing property between $\beta$ and $\gamma$. It rejects gamma backgrounds based on hit-timing spectrum of PMTs. We applied the method to hit-timing spectrum of MC and data of KamLAND-Zen400 Phase1, where $^{110m}$Ag (gamma event) was the most dominant background. We found that this method could reject 60% of gamma backgrounds and had a potential of ~10% improvement of its limit.
doi:10.5281/zenodo.4122868 fatcat:r77dowyth5ftxev4zi65wcqowu