TL;DR
This paper reports on the most sensitive search to date for neutrinoless double-beta decay (0νββ) in Xenon-136, using the complete dataset from the EXO-200 experiment. By employing a deep neural network (DNN) for improved background discrimination and enhancing detector performance, the experiment achieved unprecedented sensitivity, setting a new lower limit on the 0νββ half-life.
What Problem Does This Paper Address?
The search for neutrinoless double-beta decay (0νββ) is a cornerstone of modern particle physics. Double-beta decay is a rare nuclear process where a nucleus transforms into another by emitting two electrons. The standard model predicts a version of this decay that also emits two antineutrinos (2νββ), which has been observed. However, a hypothetical variant, 0νββ, would involve the emission of no antineutrinos. The observation of 0νββ would have profound implications: it would prove that neutrinos are their own antiparticles (Majorana particles) and violate lepton number conservation, pointing towards physics beyond the Standard Model. It would also provide a direct way to constrain the absolute mass of neutrinos. While many experiments are searching for this elusive decay, current limits on its half-life are in the range of 10²⁵ to 10²⁶ years, meaning the decay is incredibly rare, if it occurs at all. This paper addresses the challenge of pushing these limits further by improving both the sensitivity of the detector and the analysis techniques.
Methods
The EXO-200 experiment utilizes a large liquid xenon (LXe) time projection chamber (TPC) enriched in the isotope Xenon-136 (¹³⁶Xe). The key to their improved analysis lies in two main areas:
- Deep Neural Network (DNN) for Background Discrimination: The primary challenge in these searches is distinguishing the faint signal of 0νββ from a much larger background of other radioactive decays. The authors developed a DNN that analyzes the detailed topological information of events within the LXe TPC. This DNN takes "greyscale images" derived from the U-wire signals (which capture ionization) and uses them to classify events as either signal-like (0νββ) or background-like. This approach significantly outperforms previous methods that relied on simpler event classifications like single-site (SS) vs. multi-site (MS) events.
- Detector Upgrades and Analysis Refinements: The experiment underwent upgrades between "Phase I" and "Phase II" data taking, including an increase in the electric field within the TPC. This improved the energy resolution of the detector. Furthermore, the analysis introduced new criteria for event reconstruction, such as relaxing time cuts and including events with partial U-wire signal information, which increased the signal detection efficiency.
Datasets
The analysis uses data collected by the EXO-200 experiment over its operational period, encompassing both "Phase I" (Sept. 2011 - Feb. 2014) and "Phase II" (May 2016 - Dec. 2018). This combined dataset represents a total exposure of 234.1 kg·yr of ¹³⁶Xe.
For training and validation of the DNN, simulated Monte Carlo (MC) data were generated. These MC simulations modeled both potential 0νββ signal events and various background sources, including radioactive decays from detector materials and external sources. The MC simulations were validated against calibration data from external gamma sources (²²⁸Th, ²²⁶Ra, ⁶⁰Co) and background-subtracted 2νββ data.
Key Insights and Ideas
- Topology is King: The success of the DNN highlights the power of using detailed event topology within a TPC to reject backgrounds. Simply looking at the total energy is insufficient; understanding how the energy is deposited provides crucial discriminating power.
- DNNs as a Game Changer: The DNN approach offers a significant (~25%) improvement in sensitivity compared to previous analyses that relied on simpler event characteristics. This demonstrates the growing impact of advanced machine learning techniques in rare event searches.
- Synergy of Detector and Analysis: The combination of detector upgrades (improving energy resolution) and sophisticated analysis techniques (DNN) is essential for pushing the frontiers of sensitivity in low-background physics.
- The Elusive Nature of 0νββ: The fact that no statistically significant signal was observed, despite the increased sensitivity, reinforces the extremely low probability of this decay occurring, or the very long half-life it possesses.
Key Results
- Increased Signal Detection Efficiency: The signal detection efficiency was improved from 80.8% in previous analyses to 96.4±3.0% with the upgraded detector and analysis.
- Improved Energy Resolution: The energy resolution at the ¹³⁶Xe 0νββ Q-value (2457.83 keV) was enhanced from σ/E = 1.23% to 1.15±0.02%.
- Median 90% CL Sensitivity: The median 90% confidence level (CL) sensitivity for the 0νββ half-life is 5.0 × 10²⁵ yr.
- Lower Limit on Half-Life: No statistically significant evidence for 0νββ was observed, leading to a lower limit on the ¹³⁶Xe 0νββ half-life of 3.5 × 10²⁵ yr at the 90% CL.
- Constraining Neutrino Mass: This result corresponds to an upper limit on the effective Majorana neutrino mass of (mββ) < (93-286) meV, depending on the nuclear matrix elements used.
Limitations and Future Work
- Systematic Errors: While significant efforts were made to control and quantify systematic errors (e.g., background model uncertainties, spectral shape agreement, signal detection efficiency), these remain a limiting factor in achieving even greater sensitivity. The paper details contributions from various sources, with background errors and spectral shape agreement being the most significant.
- Future Experiments: The paper explicitly mentions the promise of nEXO, a planned tonne-scale successor to EXO-200. This next-generation experiment aims to achieve a sensitivity to the 0νββ half-life of ~10²⁸ yr, which would probe the "normal" neutrino mass hierarchy and potentially discover 0νββ if it occurs within this range. The success of the DNN approach in EXO-200 provides a strong foundation for its application in nEXO.
Summary updated