| 研究実績の概要 |
Richard Calland has developed a new method for reconstructing neutrino interaction events in the Super-Kamiokande detector using graphics processing units (GPUs). The method tests the charge and arrival time likelihoods for different hypotheses of the event topology and particle kinematics. To find the maximum likelihood, Richard developed a novel trans-dimensional reversible jump Markov chain Monte Carlo algorithm. Given the large number of photo-multiplier tubes (PMTs) in Super-Kamiokande (~11,000), it was necessary to use GPUs to calculate the charge and arrival time likelihoods for each PMT at each step in the change. Using this research grant, Richard built a GPU cluster at Kavli IPMU to carry out his work. By using the 1000-2000 logical cores on each GPU, the calculation time could be improved by a factor of 50 compared to using CPUs. Richard also showed the improved ability of the reversible jump Markov chain Monte Carlo method to reconstruct event topologies with a large number of particles. His work will be applied to the reconstruction of multi-particle events that are critical for the determination of the neutrino mass ordering using atmospheric neutrino interactions in Super-Kamiokande. Richards was invited to present his work at the PhyStat-nu workshop on statistics and event classification methods in neutrino experiments in a talk titled, "Statistical Issues in Neutrino Event Reconstruction.
|
