Probing neutrino emission at GeV energies from astrophysical transient events with the IceCube Neutrino Observatory

Multi-messenger astronomy leverages diverse signals (electromagnetic waves, gravitational waves, neutrinos) to understand astrophysical phenomena. Neutrinos, with their negligible interaction cross-section and neutral charge, offer a unique perspective, traversing obstacles that impede other messengers. Their detection signifies hadronic reactions, providing insights inaccessible through conventional astronomy. Neutrino sources span a wide energy range, with solar neutrinos originating from nuclear reactions in the sun's core (below MeV energies), while IceCube has detected TeV-PeV neutrinos from sources like TXS 0506+056 and NGC 1068, attributed to proton-photon interactions. GeV neutrinos, stemming from proton-proton and proton-neutron collisions, are predicted from sources capable of accelerating protons and neutrons, with Gamma-Ray Bursts (GRBs) being prime candidates. Several detectors, including IceCube, Super-Kamiokande, and KM3NeT, search for astrophysical neutrinos. Super-Kamiokande, optimized for MeV-GeV neutrinos, has provided upper limits, but its limited size restricts sensitivity. IceCube, with its km³ volume, offers the potential to detect fainter sources. IceCube's DeepCore, with highly sensitive PMTs, enables GeV neutrino observation. Atmospheric neutrinos, though themselves scientifically interesting, form a significant background. By comparing neutrino events during transient events (like GRBs) to background periods, the presence of astrophysical neutrinos can be inferred.

The paper cites several key studies. The detection of solar neutrinos with energies below MeV is referenced [1]. IceCube's detection of astrophysical neutrinos in the TeV-PeV range from TXS 0506+056 [3] and NGC 1068 [4] are discussed. The potential for GRB neutrino emission below the photosphere is mentioned [5]. Super-Kamiokande's upper limits on astrophysical neutrinos in the GeV range are presented [6]. Prior IceCube work using the ELOWEN selection for GeV neutrino searches focusing on solar flares [7] and binary mergers [12, 13] is also mentioned. The exceptionally bright GRB 221009A [14] is highlighted, along with multi-messenger analyses of this event [15, 16] encompassing a wide energy range.

The IceCube Neutrino Observatory's DeepCore, a densely instrumented region, is employed to search for GeV neutrinos. The ELOWEN selection, designed to isolate 0.5-5 GeV events, employs several filtering steps to reduce background noise, primarily from atmospheric muons and detector noise. These steps include restricting events to DeepCore, applying constraints on the number of triggered optical modules, using NoiseEngine to identify correlated hits [10], and applying data quality selections based on interaction point depth, total charge, and hit timing. The resulting data rate is significantly reduced (from 1400 Hz to 0.02 Hz). The final sample retains 40% of simulated neutrinos following a standard -2 spectrum. For transient events, the number of neutrino candidates during the event is compared to a background sample from off-time periods. For GRB 221009A, two time windows (1000 s and 2200 s centered around the detection time) were analyzed, using a background sample of 2600 time windows without transient events. Data quality checks were performed to ensure the reliability of the GRB observation period. Similar time windows (1000s, and for binary neutron star mergers, 3s) are used for binary merger searches [12, 13]. Improvements to ELOWEN include machine learning for enhanced noise filtering and the development of zenith direction reconstruction at GeV energies. t-SNE, an unsupervised machine learning algorithm, was used to visualize the separation between neutrinos and noise in the parameter space. A boosted decision tree (BDT) was employed for improved NoiseEngine filtering, optimizing for signal-to-noise ratio. Zenith direction reconstruction is achieved using two BDTS trained on upward and downward-going neutrinos, leveraging differences in hit patterns on a DeepCore string.

The search for GeV neutrinos from GRB 221009A yielded no significant deviation from the expected background, with p-values of 0.79 (1000 s window) and 0.81 (2200 s window). 90% confidence level upper limits on the time-integrated all-flavor neutrino fluence and flux (at 1 GeV) were derived. Similar upper limits are set for neutrinos from binary mergers. The analysis of the t-SNE plot of filtering variables reveals clusters dominated by simulated neutrinos, suggesting potential improvements in the filtering process. The boosted decision tree improves the NoiseEngine's ability to distinguish between neutrinos and noise. Zenith direction reconstruction achieves a balanced accuracy of 77%, although 40% of neutrino events are not classified due to low hit counts or horizontal directions. This suggests that with further development, direction reconstruction could enhance the signal-to-noise ratio.

The absence of a significant neutrino signal from GRB 221009A, despite its brightness, does not rule out GeV neutrino emission from GRBs. The upper limits derived constrain models of GRB neutrino production. The improvements to ELOWEN, particularly the enhanced noise filtering and the novel zenith direction reconstruction, hold significant promise for future searches. Improved sensitivity will allow the exploration of a wider range of transient events and potentially reveal previously undetectable low-energy neutrino sources.

IceCube's capacity for observing GeV neutrinos has been demonstrated through transient searches, such as the one conducted for GRB 221009A. Ongoing work to refine noise filtering and introduce zenith direction reconstruction will significantly improve sensitivity. Future analyses with these enhancements will enable deeper exploration of transient events and enhance our understanding of astrophysical neutrino emission.

The current ELOWEN selection lacks full directional reconstruction at GeV energies, limiting the ability to pinpoint sources. While significant progress has been made in noise reduction, some background noise remains, potentially affecting the sensitivity of the search. The analysis is limited to a specific energy range and spectral index.