WavePID: Studies of DOM-level waveform timing for track vs. cascade discrimination in IceCube at 5-100 GeV

TL;DR

This work tackles track-vs-cascade discrimination for IceCube in the $5-100~\mathrm{GeV}$ regime, where sparse light deposits challenge traditional topology-based methods. It introduces WavePID, a template-based, likelihood-ratio discriminator built from waveform timing-aware observables, and validates it on Monte Carlo and 11.1 years of data. WavePID achieves a notable improvement in cascade purity (≈5 percentage points at a fixed 20% down-selection) while preserving data–MC agreement within detector systematics, demonstrating the value of waveform-level timing in low-energy reconstruction. The approach is compact, interpretable, and offers pathways for further gains with IceCube hardware upgrades and broader waveform-based PID strategies.

Abstract

The IceCube Neutrino Observatory is a cubic-kilometer Cherenkov detector embedded in the Antarctic ice at the South Pole. Its densely instrumented sub-array and dedicated low-energy analyses provide sensitivity to neutrinos in the 5-100 GeV range, enabling precision studies of neutrino oscillations and searches for new physics. This work focuses specifically on this low-energy regime, where sparse hit patterns limit the performance of topology-based reconstruction and classification methods. We introduce Waveform-based Particle Identification (WavePID), a statistically rigorous and interpretable likelihood-ratio discriminator for track-cascade separation, built from Monte Carlo templates in timing-aware, physics-motivated observables and validated through dedicated simulations. Applied to both Monte Carlo and 11.1 years of IceCube data, WavePID suggests improved cascade purity by about 5 percentage points at a fixed 20% down-selection rate relative to the current leading cascade selection, while maintaining Data-MC agreement within detector systematics. The approach is compact and robust to sparse observations, demonstrating the value of waveform-level timing for low-energy reconstruction.

Figures (4)

• Figure 1: Distribution of the WavePID TS for true tracks and cascades. Shaded bands indicate one-standard-deviation detector systematics. Data are overlaid for comparison.
• Figure 2: Cascade purity gain and selection efficiency from the WavePID TS within the DynEdge cascade bin for events with $10 \leq N_{\text{hits}} < 20$ hit DOMs (on the nominal MC set).
• Figure 3: PATDs for $\mu$ (left) and $e$ (right) resolved by parent process.
• Figure 4: Early-photon fraction at $80~\mathrm{GeV}$ as a function of distance. Shaded bands indicate $1\sigma$ variations.