Neutrino Reconstruction in TRIDENT Based on Graph Neural Network

TL;DR

This work addresses precise directional reconstruction of high-energy neutrinos in the TRIDENT detector by introducing a graph neural network (GNN) reconstruction method that treats triggered hDOMs as graph nodes and connects them via k-nearest-neighbor edges. The EdgeConv-based GNN computes direction estimates for both shower-like $\nu_e$ CC events and track-like $\nu_\mu$ CC events, with the graph represented as $G=\{pos_i, x_i, e_{ij}, u\}$ and edge updates $e_{ij}=\phi_\theta(u, x_i, x_j - x_i)$. The approach achieves a median angular error of ~1.3 degrees for shower-like events at $100\,\mathrm{TeV}$—significantly better than the ~1.7 degrees from a traditional likelihood method—and ~0.1 degrees for track-like events at high energy, comparable to likelihood-based methods. This demonstrates the potential of GNNs to enhance neutrino direction reconstruction in large-volume detectors, and paves the way for extending to energy reconstruction and improved robustness against detector uncertainties.

Abstract

TRopIcal DEep-sea Neutrino Telescope (TRIDENT) is a next-generation neutrino telescope to be located in the South China Sea. With a large detector volume and the use of advanced hybrid digital optical modules (hDOMs), TRIDENT aims to discover multiple astrophysical neutrino sources and probe all-flavor neutrino physics. The reconstruction resolution of primary neutrinos is on the critical path to these scientific goals. We have developed a novel reconstruction method based on graph neural network (GNN) for TRIDENT. In this paper, we present the reconstruction performance of the GNN-based approach on both track- and shower-like neutrino events in TRIDENT.

Figures (5)

• Figure 1: Top view of TRIDENT detectors.
• Figure 2: Architecture of GNN used in this study is shown on the left plot. The detailed structure of EdgeConv block is illustrated on the right plot.
• Figure 3: The angular error for 100 TeV $\nu_e$ CC events. The red line represents the median angular error. The orange and blue lines exhibit the 68% and 90% quantiles.
• Figure 4: Muon emits Cherenkov photon at $\hbox{\boldmath$r$}_i$ and triggers DOM$_i$.
• Figure 5: The angular resolution of $\nu_\mu$ CC events as a function of energy. The median angle between the reconstructed track and the true direction of $\mu$ and $\nu_\mu$ is visualized by the green and red line, respectively. Color bands exhibits the 68% and 90% quantiles. Black line represents the median angle between direction of $\mu$ and $\nu_\mu$.