⚛️ high-energy experiments

Searches for Non-Standard Neutrino Oscillations with the IceCube Neutrino Observatory

This thesis utilizes 10.67 years of IceCube atmospheric neutrino data to search for an unstable sterile neutrino model, finding no evidence for such decay while significantly constraining the parameter space preferred by short-baseline anomalies, and simultaneously lays the groundwork for future antineutrino resonance searches through improved cross-section calculations, machine-learning reconstruction, and enhanced event selection.

Original authors: Philip L. R. Weigel

Published 2026-09-14

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Philip L. R. Weigel

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Deep within the ice of the South Pole, a massive detector listens for ghostly particles called neutrinos. These particles are produced in the nuclear fires of the sun, in the crushing depths of the atmosphere, and in the violent explosions of distant stars. They are unique because they rarely interact with anything; they can pass through the entire Earth without stopping. For decades, scientists have used these particles to test the Standard Model of physics, the rulebook that describes how the universe works at its most fundamental level. While the model has been incredibly successful, it has a few stubborn gaps. One of the most puzzling involves a mismatch between what we see in short-distance experiments and what we see in long-distance ones. Some experiments suggest there is a fourth type of neutrino, a "sterile" one that does not interact with matter at all, while others see no sign of it. This contradiction has left the scientific community searching for a solution that can explain the strange signals without breaking the rest of physics.

A researcher led by Philip Weigel at the Massachusetts Institute of Technology has taken a fresh look at this problem using ten years of data from the IceCube Neutrino Observatory. Instead of looking for a stable fourth neutrino, they tested a more complex idea: what if this extra neutrino exists but is unstable, decaying into invisible particles before it can be detected? This hypothesis was attractive because it could explain the short-distance anomalies while avoiding the strict limits set by cosmology, which suggest the universe should not contain too many of these extra particles. The researcher analyzed high-energy neutrinos that traveled thousands of kilometers through the Earth, looking for the specific signature that would appear if these heavy neutrinos were decaying on their journey.

The analysis found no evidence to support the idea of a decaying sterile neutrino. In fact, the data showed no preference for sterile decay over the standard no-decay 3+1 hypothesis. The researcher compared their observations against a vast range of possible scenarios and found that the results matched the standard expectation of three neutrino types perfectly. They were able to exclude most of the parameter space that previous global fits to short-baseline data had favored at 90% confidence level, effectively ruling out this specific explanation for the anomalies in the majority of cases. The study did not find a new particle, but it did provide a powerful constraint, showing that if these anomalies are real, they cannot be solved by a simple unstable neutrino model across the preferred regions.

To prepare for future searches, the researcher also built a new set of tools to improve how IceCube sees these particles. One major hurdle in studying neutrinos is that the detector cannot easily tell the difference between a neutrino and its antimatter twin, the antineutrino, on an event-by-event basis. This distinction is crucial because some theories predict that sterile neutrinos would cause antineutrinos to disappear more often than neutrinos. The researcher developed a new method using machine learning to estimate a property called "inelasticity," which measures how much energy is transferred during a collision. By analyzing the patterns of light left behind by the particles, they could statistically separate neutrinos from antineutrinos with much greater precision than before. This breakthrough, combined with a new event selection process that doubled the number of useful signals captured, lays the groundwork for the next generation of searches.

The work also involved a complete recalculation of how neutrinos interact with matter at high energies. These interactions are the foundation for predicting what the detector should see. The researcher updated these calculations to include the latest data from particle accelerators and to account for complex effects that were previously ignored, such as the behavior of heavy quarks and the influence of the Earth's density. These improved calculations reduced the uncertainties in the predictions, making the detector's measurements more reliable. By refining the theoretical models and the data analysis techniques, the researcher has created a more sensitive instrument for the future.

While the search for a decaying sterile neutrino came up empty, the effort has clarified the landscape of the problem. The short-baseline anomalies remain, but the simplest non-minimal explanations have been excluded. The tools developed in this study, particularly the ability to distinguish between neutrinos and antineutrinos and the improved understanding of their interactions, will allow IceCube to continue probing these mysteries with greater accuracy. The path forward is now clearer: the anomalies are real, but the solution likely lies in a more complex form of new physics than the unstable neutrino model proposed here. The work stands as a testament to the power of patience and precision in science, where ruling out a possibility is just as important as finding a new one.

Technical Summary: Searches for Non-Standard Neutrino Oscillations with the IceCube Neutrino Observatory

Problem Statement
Short-baseline neutrino anomalies observed by LSND, MiniBooNE, and gallium-based experiments (SAGE, GALLEX, BEST) suggest the existence of neutrino oscillations at a mass-squared splitting scale of $\Delta m^2 \sim 1 \text{ eV}^2$ . These anomalies are difficult to reconcile within the standard three-flavor oscillation framework. The minimal extension, a "3+1" model introducing a sterile neutrino mass eigenstate ( $\nu_4$ ), faces significant tension with null results from reactor and long-baseline disappearance experiments, as well as cosmological constraints on the effective number of relativistic species ( $N_{\text{eff}}$ ) and the sum of neutrino masses ( $\sum m_\nu$ ).

To resolve these tensions, this thesis investigates a non-minimal extension where the heavy sterile mass eigenstate is unstable and decays into two invisible particles ( $\nu_4 \to \phi + \psi$ ). This "unstable sterile neutrino" model preserves the oscillation-like phenomenology required to explain short-baseline anomalies while suppressing sterile neutrino thermalization in the early universe, thereby alleviating cosmological constraints. The primary goal is to search for signatures of this decay using high-energy atmospheric neutrinos detected by the IceCube Neutrino Observatory.

Methodology
The analysis utilizes 10.67 years of high-energy atmospheric neutrino data from IceCube, selecting events with reconstructed energies between 500 GeV and 100 TeV. The methodology relies on three pillars:

Theoretical Framework: The search targets the resonant disappearance of antineutrinos ( $\bar{\nu}_\mu$ ) induced by matter effects as they traverse the Earth. In the unstable sterile model, the decay width ( $\Gamma_4 \propto g^2$ ) modifies the survival probability, leading to a distinct energy and zenith angle dependence compared to the standard 3+1 no-decay hypothesis.
Improved Cross-Section Calculations: A comprehensive calculation of neutrino-nucleon and neutrino-nucleus deep-inelastic scattering (DIS) cross sections was developed. This includes:
- Next-to-next-to-leading order (NNLO) calculations for free nucleons and next-to-leading order (NLO) for nuclear targets.
- Implementation of the FONLL-C mass scheme to handle heavy quark (charm, bottom, top) production.
- Inclusion of target mass corrections, low- $Q^2$ phenomenological parameterizations (CKMT + PCAC), and small- $x$ resummation.
- Explicit inclusion of final-state radiation (FSR) effects.
- These calculations provide precise inelasticity ( $y$ ) distributions, which are critical for the statistical separation of neutrinos and antineutrinos.
Advanced Reconstruction and Event Selection:
- Machine Learning: New reconstruction techniques utilizing Transformer and MAMBA (State-Space Model) architectures were developed. These models process raw pulse series data to reconstruct energy, direction, and inelasticity with higher precision and lower computational cost than previous likelihood-based or CNN methods.
- Event Selection: A new multi-level event selection was implemented to replace the MEOWS-2023 selection. It employs transformer-based background classifiers and angular uncertainty metrics to reduce atmospheric muon backgrounds while doubling the signal efficiency compared to the previous iteration.
- Analysis Framework: The fitting framework was optimized with GPU-accelerated likelihood evaluations (using adjoint-mode gradient computation) and warm-starting techniques. These improvements significantly reduce the computational cost of fitting the full parameter space, which previously required nearly 4.5 CPU-years for a single seed.

Key Results

Unstable Sterile Neutrino Search: The analysis found no preference for the unstable sterile neutrino hypothesis over the standard 3+1 no-decay hypothesis. The best-fit point corresponds to the no-decay limit ( $g^2 = 0$ ).
Exclusion Limits: The study excludes most of the parameter space preferred by global fits to short-baseline data at the 90% confidence level (C.L.). Specifically, the analysis places strong constraints on the coupling constant $g^2$ and the mixing angle $\theta_{24}$ , ruling out the regions where the unstable decay model was expected to resolve the tension between short-baseline anomalies and cosmological data.
Reconstruction Performance: The new MAMBA-based reconstruction achieved a root mean square error (RMSE) of 0.13 for inelasticity, comparable to Transformer models but with significantly lower computational complexity ( $O(L)$ scaling). The new event selection increased the signal efficiency by a factor of 2.3 while maintaining high purity, with negligible atmospheric muon contamination ( $<0.005\%$ ).
Cross-Section Uncertainties: The updated cross-section calculations reduced uncertainties on the total CC DIS cross section to $<2\%$ for free nucleons and $<5\%$ for oxygen targets in the energy range relevant to IceCube ( $E_\nu \lesssim 10 \text{ PeV}$ ).

Significance
This work provides a stringent test of the unstable sterile neutrino model, a leading candidate for explaining short-baseline anomalies without violating cosmological bounds. By excluding the parameter space favored by previous global fits, the thesis suggests that the unstable decay mechanism is unlikely to be the sole explanation for the observed anomalies.

Furthermore, the thesis establishes the infrastructure for the next generation of sterile neutrino searches at IceCube. The development of statistical $\nu/\bar{\nu}$ separation via inelasticity reconstruction, combined with improved cross-section predictions and event selection, enables future searches for the resonant disappearance of antineutrinos—a unique signature of eV-scale sterile neutrinos. These tools are not limited to sterile neutrino searches but also enhance broader IceCube physics analyses, including measurements of the astrophysical neutrino flux and tests of Standard Model neutrino interactions. The work narrows the viable solution space for short-baseline anomalies and advances the experimental methodology required to resolve them.

More like this