Neutron source-based event reconstruction algorithm in large liquid scintillator detectors

TL;DR

The paper presents a neutron source–based, data-driven event reconstruction framework for large liquid scintillator detectors, exemplified on JUNO-like geometry. It uses a maximum-likelihood approach combining PMT timing and charge information, with input tables derived from $^{241}$Am$^{13}$C neutron calibration, cosmogenic neutrons, and laser calibration to model timing PDFs, charge maps, and charge PDFs. Vertex position is recovered with about $\pm 4\ \mathrm{cm}$ bias and $\sim 9\ \mathrm{cm}$ resolution at low energy, while energy uniformity across the detector stays below $0.5\%$ and energy resolution remains competitive with JUNO benchmarks. The method achieves effective background rejection via PSD, identifying $\sim$80% of $\alpha$-particle and 45% of fast-neutron backgrounds while preserving a high positron efficiency, and it requires fewer fixed calibration points than traditional approaches, enabling earlier physics analyses and applicability to other detectors.

Abstract

We developed an event reconstruction algorithm, applicable to large liquid scintillator detectors, built primarily upon neutron calibration data. We employ a likelihood method using photon detection time and charge information from individual photomultiplier tubes. Detector response tables in the likelihood function were derived from americium-carbon neutron source events, 2.2~MeV $γ$-ray events from cosmic-ray muon spallation neutrons, and laser calibration events. This algorithm can reconstruct the event position, energy, and also has capability to differentiate particle types for events within the energy range of reactor neutrinos. Using the detector simulation of the Jiangmen Underground Neutrino Observatory (JUNO) experiment as a large liquid scintillator detector example, we demonstrate that the presented reconstruction algorithm has a reconstructed position accuracy within $\pm$4~cm, and a reconstructed energy non-uniformity under 0.5\% throughout the central detector volume. The vertex resolution for positron events at 1~MeV is estimated to be around 9~cm, and the energy resolution is confirmed to be comparable to that in the JUNO official publication. Furthermore, the algorithm can eliminate 80\% (45\%) of $α$-particle (fast-neutron) events while maintaining a positron event selection efficiency of approximately 99\%.

Figures (12)

• Figure 1: A schematic view of the JUNO detector, a 20-kton liquid scintillator volume, contained in a spherical acrylic vessel and viewed by PMTs mounted on the stainless steel support structure. The acrylic vessel and stainless steel structure are connected by 590 acrylic nodes (connecting bars), and the volume outside of the acrylic vessel is filled with purified water. The detector $X$- and $Z$-axes are defined in the top right corner of this figure, and the origin of the coordinates is the center of the spherical liquid scintillator volume. The asterisk mark (*) in the figure represents the event vertex position, $\bm{x}=(R,\Theta,\Phi)$. The relative angle between the vertex position and PMT is expressed as $\theta_{\rm PMT}$. The radius of the spherical liquid scintillator volume is 17.7 m, and the distance between the detector center and PMT surface plane is approximately 19.4 m. PMTs mounted on the outer surface of the stainless steel support structure help to detect cosmic-ray muons from the outside of the detector. These PMTs are not involved in the reconstruction algorithm presented in this study.
• Figure 2: The total charge distributions for the prompt events in the $^{241}$Am$^{13}$C calibration source sample at the detector center. The black histogram represents all of the $^{241}$Am$^{13}$C source events, and other colored histograms are made based on the true particle information from the simulation. The vertical lines denote the event classification criteria as described in the text. The peak around 9000 photoelectrons (p.e.) corresponds to the 4.4 MeV $\gamma$-rays from $^{12}{\rm C}^{*}$ produced by the inelastic scattering between fast-neutrons and $^{12}{\rm C}$ in the liquid scintillator volume. A small fraction of 2.2 MeV $\gamma$-ray events contaminate this prompt event sample following the event time cut, yielding a peak around 4000 p.e.
• Figure 3: Example residual timing distributions for the MCP LPMTs in the $^{241}$Am$^{13}$C source events at $Z=10$ m. Both of the histograms are normalized by their area. The histogram from the fast-neutron sample has a longer tail than the one from the 2.2 MeV $\gamma$-ray sample.
• Figure 4: The black histogram shows the distance between the 2.2 MeV $\gamma$-ray true energy deposition position and $^{241}$Am$^{13}$C source position including the neutron and 2.2 MeV $\gamma$-ray travel length. The blue histogram shows the distance between the true energy deposition and reconstructed vertex positions based on the vertex reconstruction algorithm before applying the iterative timing PDF construction introduced in Section \ref{['subsubsec:timingpdf3']}.
• Figure 5: The residual time distributions of the $^{241}$Am$^{13}$C source events at (0, 0, 16 m) for the Dynode LPMTs. The black histogram shows the original residual time distributions, where the photons' time-of-flight are calculated based on the $^{241}$Am$^{13}$C source position. The green histogram corresponds to the residual time distribution with the iterative method introduced in the text. The red distribution shows the residual time distribution from the electron sample generated at (0, 0, 16 m). All of the histograms are normalized by their area.