Practical considerations in modeling the low light response of photomultiplier tubes in large batch testing

Abstract Photomultiplier tubes continue to be a reliable, cost-effective means of detecting light produced by the interaction of subatomic particles with detectors. For detectors where the expected light yield is modest, characterizing the low light response of the tube is of paramount importance. Several phenomenological models addressing this issue exist. This paper presents side-by-side comparison between three such approaches as they arose from a large scale testing of tubes to be used by a Ring Imaging Cherenkov detector at Jefferson Lab. The main characteristics of the tubes, such as the gain, were found to be consistent within the expected uncertainties for all models considered. Leveraging the extensive nature of the study, a machine learning algorithm based on an artificial neural network capable of obtaining the tube characteristics directly from the raw ADC data was developed and trained. The trained neural network produced results fully compatible with the three models considered, with substantial savings in both computation time and experimenter overhead.