High precision modeling of germanium detector waveforms using Bayesian machine learning Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 21, 2019
  • Shanks, Benjamin
    • Affiliation: College of Arts and Sciences, Department of Physics and Astronomy
  • The universe as we see it today is dominated by matter, but the Standard Model of particle physics cannot explain why so little antimatter remains. If the neutrino is its own antiparticle - a so-called Majorana particle - lepton number must be violated, which is a key component of theories that explain the observed matter-antimatter asymmetry. Neutrinoless double-beta decay (0νββ), a hypothetical radioactive decay in certain nuclei, is the only experimentally accessible signature that can prove if neutrinos are Majorana in nature. But if it exists, 0νββ must be exceedingly rare, with current half-life limits over 1025 years. Measuring a process with such a faint signal requires extraordinary efforts to eliminate backgrounds. The Majorana Demonstrator is a search for 0νββ of germanium-76 in an array of germanium detectors, with the goal of "demonstrating" backgrounds low enough to justify building a larger experiment with ~ 1 tonne of isotope. Reducing backgrounds even further will be critical to the discovery potential of a tonne scale experiment. One powerful method to reject background is pulse shape discrimination, which uses the shape of measured detector signals to differentiate between background and candidate 0νββ events. With a better understanding of pulse shapes from our detectors, we may be able to improve the discrimination efficiency. We have developed a detailed model of signal formation in germanium detectors, where the shape depends sensitively on characteristics specific to each individual detector crystal. To train the parameters for specific crystals in the Demonstrator, we have implemented a Bayesian machine learning algorithm which is able to infer detector characteristics using only standard calibration waveforms. This model is accurate to the level of parts per thousand of the signal amplitude, is able to discriminate against common background events, and has even shown some ability to estimate the position of origin for signals inside the detector.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Henning, Reyco
  • Drut, Joaquin
  • Green, Matthew
  • Radford, David
  • Wilkerson, John
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2017

This work has no parents.