GPU-Accelerated Event Reconstruction for the COMET Phase-I Experiment

Authors:

Abstract
A parallelization scheme with GPU is introduced for event reconstruction in the cylindrical drift chamber of COMET Phase-I experiment. The experiment is aiming for a discovery of charged lepton flavor violation by observing 104.97 MeV electrons from neutrinoless muon to electron conversion in muonic atoms. Helical trajectories of signal electrons, inside the cylindrical drift chamber, should be reconstructed with a good momentum resolution to distinguish them from background electrons. However, events with multiple helix turns require huge amount of computation in classifying hits for each turn. The computing was accelerated by parallelizing track extrapolations in which a modified least square estimator is calculated for each seed based on the residuals between the hits and the extrapolated track representing a specific turn. The hits whose distances to the track are less than a cutoff value were classified as hits belonging to the track. The event reconstruction algorithm was applied to the multiple turn events of electrons from 90 MeV to 104.97 MeV, where we confirmed a consistency in the momentum resolution for each energy. A performance comparison between the GPUs and parallelized CPU threads is also presented with technical aspects: For a specific GPU model, the computing speed increased by an order of magnitude.

This publication has 11 references indexed in Scilit: