Development of the machine learning-based online trigger algorithms for heavy flavor events selection in sPHENIX experiment

Abstract

The sPHENIX experiment at the Relativistic Heavy Ion Collider (RHIC) is designed to investigate the physics of the strongly coupled Quark–Gluon Plasma (QGP). Heavy-flavor hadrons serve as key probes of the QGP produced during heavy-ion collisions. However, the current readout event rate of the sPHENIX detector is limited to 15 kHz, due to constraints in the readout system of the outer calorimeter detectors. To improve the trigger efficiency of heavy-flavor events, a dedicated online trigger system is imperative. This paper presents the design of machine learning-based online trigger algorithms aimed at selecting heavy-flavor quark events in the sPHENIX experiment. The algorithms, primarily based on Multi-Layer Perceptrons, Graph Neural Networks, and Bipartite Graph Neural Networks, use only spatial track hit information from fast silicon detectors as input. The performance of these algorithms are evaluated using simulation datasets, with charm-quark event samples generated via Monte Carlo for p+p collisions at $\sqrt{s}$ = 200 GeV. The results demonstrate that the system achieves a track reconstruction efficiency of 0.945. Additionally, the precision and recall values for identifying heavy-flavor events are both around 0.79.