First constraints on point-like astrophysical sources using Baikal-GVD muon neutrino events

Baikal-GVD is a new-generation neutrino telescope currently under construction in Lake Baikal, Russia. With an instrumented volume already at 0.7 km$^3$, Baikal-GVD is currently the largest neutrino telescope in the Northern hemisphere. A sub-degree angular resolution, made possible thanks to high purity of Baikal water, further enhances Baikal-GVD sensitivity to cosmic neutrino sources. In this work, we employ track-like events collected from the partially completed detector between April 2019 and March 2024 to search for muon neutrino fluxes from 92 astrophysical objects of interest. For this, a $χ^2$-based track reconstruction method is used along with a cut-based analysis. The analysis uses upward-going muons only, providing coverage for declinations between -90$^\circ$ and +38$^\circ$. No significant excess has been found, so upper limits are reported. The obtained limits are competitive with those set by ANTARES and KM3NeT. We briefly comment on a possible low-significance indication of an excess from the direction of Westerlund 1. This work sets a major milestone on the way to full-scale scientific exploitation of Baikal-GVD data.

💡 Research Summary

The Baikal‑GVD neutrino telescope, currently under construction in Lake Baikal, Russia, has reached an instrumented volume of about 0.7 km³ (4212 optical modules on 117 strings) and thus represents the largest detector in the Northern Hemisphere. Thanks to the excellent optical properties of the lake water, the array achieves a sub‑degree angular resolution (≈0.5° on average, down to ≈0.2° for the longest tracks), which is crucial for point‑source searches.

In this work the authors analyse track‑like events recorded between April 2019 and March 2024 while the detector comprised 11 clusters (30 cluster‑years, equivalent to ≈1.5 years of a full‑scale 1 km³ detector). They select only upward‑going muon tracks (zenith > 90°) to suppress the overwhelming background of downgoing atmospheric muons. Residual mis‑reconstructed muons are further reduced with a Boosted Decision Tree (BDT) classifier; a cut on the BDT score (BDTHE > 0.25) limits atmospheric‑muon contamination to <5 %. The final sample contains 988 neutrino‑candidate events with reconstructed energies from ~100 GeV to ~1 PeV and a median energy of ~1 TeV.

A catalogue of 92 astrophysical objects was compiled, including the well‑known neutrino emitters TXS 0506+056 and NGC 1068, the IceCube high‑energy event KM3‑230213A, the Galactic centre, several very‑high‑energy gamma‑ray sources, and a selection of Seyfert galaxies and other promising candidates. For each source a circular search region of 2° radius was defined, deliberately larger than the typical angular resolution to maximise sensitivity in the low‑background Poisson regime (the analysis is optimized for an E⁻² spectrum). Event counts inside each cone were compared with a background estimate obtained by scrambling right ascension.

No source shows a statistically significant excess after accounting for trial factors. Upper limits on the muon‑neutrino plus antineutrino flux were derived using the Feldman‑Cousins approach for three assumed spectral indices (γ = 2, 2.5, 3.2). The resulting 90 % confidence‑level limits are comparable to those published by the ANTARES 15‑year analysis, despite the much shorter exposure, demonstrating the competitiveness of Baikal‑GVD.

The most intriguing hint is found for the young massive stellar cluster Westerlund 1. Within the 2° search cone three events are observed, whereas the expected background is 0.98 events, corresponding to a pre‑trial p‑value of 0.0036 (≈2.7σ). After accounting for the 92 trials the significance drops below discovery level, but the result is noted as a low‑significance indication. Westerlund 1 is associated with the TeV gamma‑ray source HESS J1646‑458 and has been proposed as a potential cosmic‑ray accelerator; however, converting the observed neutrino flux into a gamma‑ray flux under a hadronic scenario would overshoot the measured HESS flux, suggesting that the neutrino contribution, if any, is modest.

Systematic uncertainties are dominated by the detector efficiency, which is known to within ±35 % (tending toward under‑prediction). No systematic error is applied to the quoted limits. Angular‑resolution uncertainties have negligible impact because the search cone is much larger than the resolution. The analysis uses only single‑cluster tracks and a simple cut‑and‑count method; it does not exploit multi‑cluster events or a full likelihood that incorporates event energy and angular distance to the source. Consequently, the present sensitivity is limited but already impressive.

Looking forward, the Baikal‑GVD array will reach its design volume of 1 km³ (≈20 clusters) by ~2028, increasing the exposure by roughly an order of magnitude. Incorporating multi‑cluster reconstructions, refined BDTs, and likelihood‑based point‑source searches will further improve sensitivity, potentially allowing Baikal‑GVD to detect or tightly constrain neutrino emission from both Galactic and extragalactic sources. This work thus marks a significant milestone, delivering the first point‑source limits from Baikal‑GVD and establishing a solid foundation for future high‑energy neutrino astronomy in the Northern Hemisphere.