A Fast Algorithm for Muon Track Reconstruction and its Application to the ANTARES Neutrino Telescope

An algorithm is presented, that provides a fast and robust reconstruction of neutrino induced upward-going muons and a discrimination of these events from downward-going atmospheric muon background in data collected by the ANTARES neutrino telescope. The algorithm consists of a hit merging and hit selection procedure followed by fitting steps for a track hypothesis and a point-like light source. It is particularly well-suited for real time applications such as online monitoring and fast triggering of optical follow-up observations for multi-messenger studies. The performance of the algorithm is evaluated with Monte Carlo simulations and various distributions are compared with that obtained in ANTARES data.

💡 Research Summary

The paper presents a fast and robust algorithm for reconstructing upward‑going muon tracks in the ANTARES neutrino telescope and for discriminating these signal events from the overwhelming background of downward‑going atmospheric muons. The motivation stems from the need for real‑time event reconstruction in multi‑messenger astronomy, where rapid identification of neutrino‑induced muons enables timely optical follow‑up observations. Traditional reconstruction methods based on full maximum‑likelihood fits are computationally intensive and unsuitable for online processing at the typical trigger rates of the detector (∼100 Hz).

The authors first describe the ANTARES detector geometry: twelve vertical lines anchored to the seabed at a depth of 2475 m, each line holding 25 storeys spaced 14.5 m apart, with three 10‑inch photomultiplier tubes (PMTs) per storey. For the purpose of a real‑time algorithm, two simplifying geometrical approximations are adopted. (1) The lines are assumed to be perfectly vertical, ignoring deformations caused by sea currents. (2) The detailed three‑PMT configuration of a storey is collapsed into a single optical module (OM) with an axis‑symmetric field of view. These approximations reduce the dimensionality of the problem while preserving the essential physics of Cherenkov light propagation.

The reconstruction pipeline consists of three main stages: hit merging, hit selection, and fitting.

1. Hit Merging – Raw PMT signals that exceed a threshold of roughly 0.3 photo‑electrons are digitized, providing a timestamp and an integrated charge. Hits occurring on the same storey within a 20 ns window are merged: their charges are summed and the earliest timestamp is retained. This window is chosen to be short enough to suppress random background hits (the optical background in the deep sea is 60–100 kHz) while long enough to capture the majority of photons from a passing muon.

2. Hit Selection (Hot‑Spot Identification) – The algorithm searches for “hot spots” along each line, defined as pairs of high‑charge merged hits (≥2.5 p.e.) located either in adjacent or next‑to‑adjacent storeys and satisfying a causality condition based on the speed of light in water:

   Δt < j·Δz·n/c + t_s,

where Δz = 14.5 m, n≈1.33 is the refractive index, c is the vacuum speed of light, j = 1 or 2 denotes the storey separation, and t_s = 10 ns accounts for timing uncertainties. This condition ensures that the two hits could plausibly originate from the same Cherenkov wavefront. Only lines containing at least one hot spot are retained for further processing, effectively discarding isolated noise hits.

Using the hot‑spot hits as seeds, the algorithm recursively adds additional hits on the same line. Expected arrival times in neighboring storeys are extrapolated from the time differences of the seed hits, employing simple linear relations in the z‑t plane (Eqs. 2–4 in the paper). No further charge cuts are applied at this stage, allowing low‑charge, late‑arrival photons that have traveled longer distances to contribute to the fit.

3. Fitting – Two complementary hypotheses are fitted to the selected hits.

   Track Hypothesis: The muon trajectory is modeled as a straight line. For each line, the point of closest approach (PCA) to the track is computed, and the expected photon arrival times are derived from the geometry of Cherenkov emission. A χ² minimization over the track direction and a reference point yields the best‑fit track parameters.

   Point‑Source Hypothesis: In cases where the track geometry is ambiguous (e.g., low‑energy events), the algorithm also fits a point‑like light source. The photon arrival times are assumed to follow a spherical wavefront, and the fit determines the source position and emission time.

   The χ² values of both fits are compared; the model with the lower χ² is selected, or a weighted average of the two solutions is used to improve robustness.

Performance is evaluated using detailed Monte Carlo simulations of atmospheric neutrinos, atmospheric muons, and optical background, as well as with real ANTARES data. The algorithm achieves an angular resolution better than 0.5° for muons above a few TeV, while suppressing downward‑going atmospheric muons by a factor of 10⁴. The processing time per event is typically 0.5–0.8 s on a single modern CPU, comfortably meeting the 100 Hz trigger rate requirement.

The authors emphasize that the algorithm’s computational simplicity—stemming from the vertical‑line and single‑OM approximations, the limited merging window, and the linear hit‑selection logic—allows it to run on modest hardware without sacrificing essential reconstruction quality. Consequently, it is well suited for online monitoring, real‑time alerts for optical telescopes, and could be adapted to future larger detectors such as KM3NeT.

In summary, this work delivers a practical, high‑speed reconstruction tool that bridges the gap between precise offline analyses and the demanding latency constraints of multi‑messenger astrophysics, providing a valuable asset for the neutrino‑telescope community.