Speaker
Description
As nuclear physics moves solidly into the 21st century, we are in need of high-coverage, high-rate detection of events. The Solenoidal Large Intensity Device SoLID seeks to fill that need for a large-acceptance, high-luminosity device in Hall A of Jefferson Lab. In keeping with the theme of modernization, the SoLID collaboration has sought to bring cutting-edge technologies and methods into the standard operating procedures, including the use of Machine Learning ML for data analysis. In order to characterize the electromagnetic calorimeter modules for SoLID, the collaboration has conducted a beam test study to validate our hardware. A critical component of this study is establishing SoLID’s ability to distinguish between different particles, or particle identification PID. During this beam test, we have worked to bridge the gap between simulation and data in order to improve the accuracy of GEANT-based simulation and provide insight into analyzing our data. By accomplishing this, we explore traditional PID methods, as well as ML PID methods for analyzing beam test data. During benchmarking of PID performance using simulation, we show that ML algorithms outperform traditional cuts with higher electron efficiency and pion rejection for a range of momentum bins and trigger settings. We additionally show ML PID methods allow for complex combinations of models for individual detectors without the characteristic loss in efficiency associated with traditional methods. After further validation, we seek to apply ML and traditional PID methods to beam test data to complete the proof of concept and laid foundational work for ML PID methods for a fully-realized SoLID, thus bringing this collaboration onto the same level as CLAS and GlueX.
