Pulse shape discrimination for GERDA Phase I data

TL;DR

This paper presents pulse shape discrimination algorithms for GERDA Phase I data, improving background rejection and signal retention in the search for neutrinoless double beta decay using germanium detectors.

Contribution

It introduces novel pulse shape analysis methods, including neural network and likelihood approaches, tailored for both BEGe and semi-coaxial germanium detectors in GERDA Phase I.

Findings

High background rejection efficiency achieved, especially with BEGe detectors.

Neural network analysis retains 90% of DEP events and rejects half of background events.

Consistent efficiency estimates across different analysis methods.

Abstract

The GERDA experiment located at the LNGS searches for neutrinoless double beta (0\nu\beta\beta) decay of ^{76}Ge using germanium diodes as source and detector. In Phase I of the experiment eight semi-coaxial and five BEGe type detectors have been deployed. The latter type is used in this field of research for the first time. All detectors are made from material with enriched ^{76}Ge fraction. The experimental sensitivity can be improved by analyzing the pulse shape of the detector signals with the aim to reject background events. This paper documents the algorithms developed before the data of Phase I were unblinded. The double escape peak (DEP) and Compton edge events of 2.615 MeV \gamma\ rays from ^{208}Tl decays as well as 2\nu\beta\beta\ decays of ^{76}Ge are used as proxies for 0\nu\beta\beta\ decay. For BEGe detectors the chosen selection is based on a single pulse shape parameter. It accepts 0.92$\pm$0.02 of signal-like events while about 80% of the background events at Q_{\beta\beta}=2039 keV are rejected. For semi-coaxial detectors three analyses are developed. The one based on an artificial neural network is used for the search of 0\nu\beta\beta\ decay. It retains 90% of DEP events and rejects about half of the events around Q_{\beta\beta}. The 2\nu\beta\beta\ events have an efficiency of 0.85\pm0.02 and the one for 0\nu\beta\beta\ decays is estimated to be 0.90^{+0.05}_{-0.09}. A second analysis uses a likelihood approach trained on Compton edge events. The third approach uses two pulse shape parameters. The latter two methods confirm the classification of the neural network since about 90% of the data events rejected by the neural network are also removed by both of them. In general, the selection efficiency extracted from DEP events agrees well with those determined from Compton edge events or from 2\nu\beta\beta\ decays.