Vertex reconstruction in the TAO experiment

TL;DR

This work tackles precise vertex reconstruction for the TAO detector by comparing a traditional charge-center algorithm (CCA) with a data-driven deep-learning approach (DLA). It develops targeted optimizations for CCA (dual-opening corrections, nonlinear radius corrections, and MCC fitting) and implements two neural architectures, VGG-T and ResNet-T, to map calibrated charge and timing data into 3D vertex positions. Results show that optimized CCA achieves sub-20 mm $R$-resolution at 1 MeV with small bias, while ResNet-T-based DLA reaches $R$-resolution below 12 mm with sub-millimeter bias and improved angular resolutions, with DLA generally outperforming CCA. Both methods meet TAO's vertex-precision requirements and offer complementary strengths, with CCA providing fast, robust startup capabilities and DLA delivering the best overall reconstruction performance and applicability to similar detectors.

Abstract

The Taishan Antineutrino Observatory (TAO) is a tonne-scale gadolinium-doped liquid scintillator satellite experiment of the Jiangmen Underground Neutrino Observatory (JUNO). It is designed to measure the reactor antineutrino energy spectrum with unprecedented energy resolution, better than 2% at 1 MeV. To fully achieve its designed performance, precise vertex reconstruction is crucial. This work reports two distinct vertex reconstruction methods, the charge center algorithm (CCA) and the deep learning algorithm (DLA). We describe the efforts in optimizing and improving these two methods and compare their reconstruction performance. The results show that the CCA and DLA methods can achieve vertex position resolutions better than 20mm (bias<5mm) and 12mm (bias<1.3mm) at 1 MeV, respectively, fully meeting the requirements of the TAO experiment. The reconstruction algorithms developed in this study not only prepare the TAO experiment for its upcoming real data but also hold significant potential for application in other similar experiments.

Figures (20)

• Figure 1: Structural Schematic of the TAO's central detector. The X-axis and Z-axis of the detector are shown in the upper right corner, with the origin at the center of the spherical GdLS volume. The event vertex ($\ast$) is expressed in spherical coordinates as $\bm{r}=(R, \theta, \phi )$, where $R$ is the vertex radius, $\theta$ the polar angle, and $\phi$ the azimuthal angle. The green dashed line represents the ACU calibration path along the Z-axis, and the pink dashed line indicates the CLS calibration path.
• Figure 2: Geometric schematic of photon propagation in a spherical detector. It is assumed that the photons emitted at position $x_0$ propagate in straight lines and are completely detected on the spherical surface.
• Figure 3: Impact of TAO detector's intrinsic openings on the $R$ reconstruction bias, defined as the difference between $R_\text{reco}$ and $R_\text{true}$. Panels (a) and (b) respectively show the results before and after applying the dual-opening correction.
• Figure 4: Schematic diagram of virtual channel construction. Each SiPM has two channels, and the TAO detector has a total of 8048 channels. (a) Distribution of virtual channels in the detector. (b) Charge calculation for virtual channels, with arrows indicating the calculation direction and the dashed box illustrating the interpolation process.
• Figure 5: After calibration with $^{137}$Cs, a two-dimensional distribution of $\mathit{R_\text{true}}$ versus $\mathit{R_\text{reco}}$ was obtained and fitted with a quadratic function (red curve). Dual-opening correction was applied, while the DN remains unprocessed.