Abstract: Pulse shape discrimination (PSD) is widely used in particle and nuclear physics. Specifically in liquid scintillator detectors, PSD facilitates the classification of different particle types based on their energy deposition patterns. This technique is particularly valuable for studies of the Diffuse Supernova Neutrino Background (DSNB), nucleon decay, and dark matter searches. This paper presents a detailed investigation of the PSD technique, applied in the DSNB search performed with the Jiangmen Underground Neutrino Observatory (JUNO). Instead of using conventional cut-and-count methods, we employ methods based on Boosted Decision Trees and Neural Networks and compare their capability to distinguish the DSNB signals from the atmospheric neutrino neutral-current background events. The two methods demonstrate comparable performance, resulting in a 50% to 80% improvement in signal efficiency compared to a previous study performed for JUNO. Moreover, we study the dependence of the PSD performance on the visible energy and final state composition of the events and find a significant dependence on the presence/absence of $^{11}$C. Finally, we evaluate the impact of the detector effects (photon propagation, PMT dark noise, and waveform reconstruction) on the PSD performance.
Introduction: The "Needle in a Haystack" Problem
The Diffuse Supernova Neutrino Background (DSNB) represents the cumulative flux of neutrinos from all past core-collapse supernovae in the history of the universe. Detecting this signal would provide unprecedented insights into star formation history and neutrino physics. However, identifying DSNB neutrinos at the Jiangmen Underground Neutrino Observatory (JUNO) is a formidable challenge.
The expected DSNB signal is extremely low—approximately 2–3 events per year. In contrast, neutral-current (NC) interactions from atmospheric neutrinos create a significant background of roughly 60 events per year. These background events often mimic the signal, requiring sophisticated discrimination techniques to separate the positron signal from the cocktail of neutrons, protons, and alpha particles produced by atmospheric neutrinos.
Methodology: Pulse Shape Discrimination (PSD)
To address this background, researchers utilize Pulse Shape Discrimination (PSD). This technique exploits the fact that different particles excite liquid scintillators differently. Heavier, highly ionizing particles (such as protons in the background) produce a higher fraction of "slow" light compared to lighter particles (such as positrons in the signal).
This study moves beyond traditional "cut-and-count" methods by implementing two advanced machine learning approaches:
- Boosted Decision Trees (BDT): This model utilizes nine physical variables, including pulse rising/falling times, decay constants, and the cubic radius ($R^3$) of the event position to account for Time-of-Flight biases.
- Neural Networks (NN): A multi-layer perceptron that processes the raw, binned timing profile of light hits, allowing the model to autonomously learn discriminating features.
Key Insights and Results
The Role of $^{11}$C Mimicry
A critical finding of the study is the role of $^{11}$C. When atmospheric neutrinos interact with Carbon-12, they can produce $^{11}$C and high-energy neutrons. The resulting gamma rays are highly "electron-like," making these specific background events the most difficult to reject. As shown in the background breakdown, $^{11}$C-related channels constitute a significant portion of the post-selection background.
Performance Gains
Both the BDT and NN methods achieved approximately 80% signal efficiency while maintaining a background leakage of only 1%. This represents a 50–80% improvement over the baseline established in previous JUNO studies. While the two models performed comparably, the Neural Network showed a slight advantage at lower energies (below 20 MeV).
Detector Effects
The study also quantified the impact of realistic detector conditions. While theoretical "truth-level" information would allow for near 100% efficiency, the introduction of PMT dark noise, photon propagation, and Transit Time Spread (TTS) reduces this to the final ~80% benchmark. Interestingly, waveform reconstruction was found to sharpen the timing distribution peak for the signal, as electronics effectively "pick" the first photon in a pile-up, enhancing the timing difference between fast and slow signals.
Conclusion and Future Outlook
This simulation-based study demonstrates that machine learning-driven PSD can significantly enhance JUNO's sensitivity to the DSNB. Once JUNO becomes operational (expected late 2024), these models will require validation using real-world "tagged" data—such as background events identified through the decay of $^{11}$C—to calibrate the algorithms. Continued improvements in vertex reconstruction resolution are expected to further refine PSD performance, bringing us closer to detecting the elusive glow of the supernova background.
Summary updated