Hyper-Kamiokande (HK) is a next-generation neutrino experiment with a large-scale water-Cherenkov far detector approved in Japan. Its physics program addresses some of the most challenging questions in fundamental physics like the precise measurement of the neutrino oscillation parameters (solar, atmospheric, accelerator), the investigation of astrophysical neutrino sources (supernovae and Diffuse Supernova Bursts (DSNB)), and the search for proton and exotic nucleon decays.
Since over a decade, HK’s predecessor, Super-Kamiokande, has proven the importance of neutron-tagging in a large variety of measurements, improving the limits of DSNB and proton-decay searches, and enhancing the sensitivity to the atmospheric oscillation parameters.
Neutrons produced in the interaction of an HK event thermalize and are eventually captured by hydrogen, emitting a 2.2 MeV photon. This signal is too weak for HK’s trigger threshold; therefore, the delayed neutron signal is searched by scanning all the hit PMTs after the prompt signal.
The developed method in this study feeds this information into a neural network, providing as output which of the hit PMTs are more likely to have received the neutron capture signal. This not only improves the candidate selection efficiency and purity, but also provides valuable information about the hit PMTs, identifying the most relevant ones for the subsequent fitting process.
This new technique improves the primary selection of neutron signals from 58% to 75% compared to the usual procedure based on a number-of-hits threshold.