Incorporating Inelasticity Reconstruction into Neutrino Mass Ordering Studies with IceCube

This paper proposes using new machine-learning-based inelasticity reconstruction algorithms to statistically separate neutrinos from antineutrinos, thereby enhancing the sensitivity of IceCube and IceCube Upgrade detectors to the neutrino mass ordering.

The Gist

The Cosmic Sorting Machine: How IceCube is Using "Splatter Patterns" to Solve a Universe Mystery

Imagine you are standing in a dark room, and someone is throwing two different types of balls at you from behind a curtain. One type is a tennis ball (neutrinos), and the other is a heavy medicine ball (antineutrinos).

You can’t see the people throwing them, and you can’t see the balls themselves. All you can see are the "splatter patterns" they leave on the floor when they hit. Your goal is to figure out a fundamental rule of the universe: Is the "gravity" of the Earth pulling on the tennis balls differently than it pulls on the medicine balls?

This is essentially what scientists at the IceCube Neutrino Observatory are trying to do. Here is the breakdown of their latest discovery.

---

1. The Mystery: The Neutrino Identity Crisis

The universe is filled with tiny, ghostly particles called neutrinos. There are two main "flavors" of them: neutrinos and antineutrinos. They are almost impossible to catch, but when they hit the Antarctic ice where IceCube is buried, they leave a tiny flash of light.

Scientists want to know the Neutrino Mass Ordering (NMO). Think of this as knowing whether the "family tree" of neutrinos is organized in a "Normal" way or an "Inverted" way. This isn't just a trivia question; it’s a key to understanding how the entire universe was built.

The problem? Neutrinos and antineutrinos look almost identical to our detectors. It’s like trying to tell the difference between a left-handed and a right-handed person in a pitch-black room just by the sound of their footsteps.

2. The Secret Weapon: "Inelasticity" (The Splatter Factor)

The researchers found a clever way to tell them apart. They looked at something called inelasticity.

In our ball analogy, when a tennis ball hits the floor, it might bounce a little. But when a heavy medicine ball hits, it creates a massive, messy explosion of dust and debris.

In physics, when a neutrino hits an atom, it transfers some of its energy to the surrounding material, creating a "hadronic cascade" (a messy splash of particles).

• Neutrinos tend to make a big, energetic splash.
• Antineutrinos tend to make a much smaller, more polite splash.

By measuring the size of this "splash" (the inelasticity), scientists can statistically guess: "That big splash was probably a neutrino, and that tiny splash was probably an antineutrino."

3. The Tool: Artificial Intelligence "Detectives"

Because these splashes are incredibly subtle and messy, humans can't calculate them by hand. The researchers built two high-tech "AI Detectives":

• For the old detector (DeepCore): They used Convolutional Neural Networks (the same kind of AI that helps self-driving cars recognize stop signs) to look at the patterns of light.
• For the new, upgraded detector (IceCube Upgrade): They used a Graph Neural Network, which is like an AI that can look at a complex web of connections to understand the "shape" of the event.

4. The Result: A Sharper Lens

By adding this "splash measurement" to their data, the scientists added a fourth dimension to their search. Instead of just looking at Energy, Direction, Flavor, and Time, they added Splash Size.

The result? It worked! By using this new "splatter" information, the IceCube team significantly improved their ability to distinguish between the "Normal" and "Inverted" mass orderings. It’s like going from looking at the universe through a blurry window to looking through a high-definition telescope.

Summary in a Nutshell

Scientists are using AI to study the "messiness" of particle collisions in the Antarctic ice. By measuring how much energy is "splattered" during a collision, they can tell neutrinos and antineutrinos apart, helping them solve one of the biggest mysteries in physics: the fundamental ordering of the building blocks of our universe.

Technical Summary

Technical Summary: Incorporating Inelasticity Reconstruction into Neutrino Mass Ordering Studies with IceCube

Problem Statement

The determination of the Neutrino Mass Ordering (NMO)—whether the mass eigenstates follow a "normal" or "inverted" hierarchy—is a fundamental goal in particle physics. The IceCube Neutrino Observatory can probe NMO by observing how Earth's matter affects the oscillation of atmospheric neutrinos. Because the matter effect (resonance) affects neutrinos and antineutrinos differently, and IceCube detects more neutrinos than antineutrinos, the NMO can be inferred from the intensity of these signals.

However, a significant challenge in this measurement is the inability to perfectly distinguish between neutrinos and antineutrinos. While they have different oscillation patterns, they are often processed together in standard analyses. This paper addresses the need to statistically separate these two populations to enhance NMO sensitivity.

Methodology

The researchers leverage inelasticity ($y$), which is the fraction of neutrino energy transferred to a hadronic cascade during a nucleon interaction. Due to the chirality of the weak interaction, antineutrinos are suppressed at high inelasticity compared to neutrinos. By reconstructing $y$, the collaboration can use it as a fourth observable (alongside energy, direction, and flavor) to better separate the two species.

1. Reconstruction Algorithms:
Two distinct machine-learning approaches were developed for different detector configurations:

• IceCube DeepCore (IC86): An ensemble of three 2D Convolutional Neural Networks (CNNs) was used. Each CNN was trained with different parameters (energy spectra, cuts, and containment radii) to optimize performance across different energy ranges. The outputs of these three CNNs were then aggregated using a Boosted Decision Tree (BDT) to produce the final reconstructed inelasticity.
• IceCube Upgrade (IC93): A Graph Neural Network (GNN), specifically the DynEdge model, was employed. This model uses the spatial positions of Digital Optical Modules (DOMs) and the arrival times of detected photons as inputs.

2. Sensitivity Calculation:
The reconstructed inelasticity was integrated into the event binning of existing oscillation analyses. The researchers used a modified $\chi^2$ metric (via the PISA software package) to calculate NMO sensitivity ($\eta_\sigma$), accounting for systematic uncertainties in neutrino flux, cross-sections, ice properties, and detector response.

Key Contributions

• Development of New ML Tools: The creation of specialized CNN-based and GNN-based reconstruction algorithms tailored to the specific geometries and energy thresholds of IC86 and IC93.
• Multi-Observable Analysis Framework: The successful integration of inelasticity as a statistical tool to break the degeneracy between neutrino and antineutrino signals.
• Optimization of Training Strategies: The demonstration that using an ensemble of CNNs with varying training constraints (e.g., flat vs. nominal energy spectra) outperforms single-network approaches.

Results

• Reconstruction Performance: The algorithms show improved accuracy as true energy increases. While performance is limited below 20–30 GeV (where reconstructions tend to cluster near the training mean), the quality is sufficient for physics analysis. The IC93 (Upgrade) reconstruction shows superior performance at lower energies compared to IC86.
• NMO Sensitivity Gains: Incorporating inelasticity leads to a measurable increase in NMO sensitivity for both the IceCube DeepCore and the upcoming IceCube Upgrade. This improvement holds true for both Normal Ordering (NO) and Inverted Ordering (IO).
• Specific Improvements: The most significant sensitivity boost was observed for the Normal Ordering in the second octant of $\theta_{23}$, which aligns with current global best-fit values from NuFIT 6.0.

Significance

This work demonstrates that even "imperfect" reconstruction of a secondary observable like inelasticity can provide a statistically significant boost to the discovery potential of a major neutrino observatory. By adding this dimension to the analysis, IceCube can more effectively exploit the matter effect in Earth, bringing the community closer to resolving the neutrino mass ordering. This methodology sets a precedent for using machine learning to extract subtle physical signatures from complex detector data in high-energy neutrino astronomy.