Development of machine learning analyses with graph neural network for the WASA-FRS experiment

The WASA-FRS experiment aims to reveal the nature of light A hypernuclei with heavy-ion beams. The lifetimes of hypernuclei are measured precisely from their decay lengths and kinematics. To reconstruct a p(- )track emitted from hypernuclear decay, track finding is an important issue. In this study, a machine learning analysis method with a graph neural network (GNN), which is a powerful tool for deducing the connection between data nodes, was developed to obtain track associations from numerous combinations of hit information provided in detectors based on a Monte Carlo simulation. An efficiency of 98% was achieved for tracking p(-) mesons using the developed GNN model. The GNN model can also estimate the charge and momentum of the particles of interest. More than 99.9% of the negative charged particles were correctly identified with a momentum accuracy of 6.3%.

Automated summary

The WASA-FRS experiment aims to reveal the nature of light A hypernuclei with heavy-ion beams. The lifetimes of hypernuclei are measured precisely from their decay lengths and kinematics. In this study, a machine learning analysis method with a graph neural network (GNN) was developed to obtain track associations and accurately identify p(-) mesons emitted from hypernuclear decay. The developed GNN model achieved a 98% tracking efficiency and accurately estimated the charge and momentum of the particles of interest, with a 99.9% identification rate for negative charged particles and a momentum accuracy of 6.3%.