Poissonian Analysis of Glitches Observed in the LIGO Gravitational Wave Interferometers

TL;DR

The paper tests whether LIGO glitches follow a Poisson process by segmenting data into fixed-time boxes of length $t_0$, comparing observed glitch counts to Poisson predictions using $R^2$, and validating with Poisson-generated data. It analyzes Gravity Spy–classified glitches from LIGO O3a/O3b across both detectors, identifying morphologies that are Poisson-like (e.g., Blip, Koi Fish, Extremely Loud) and those that are not (e.g., Scattered Light, Whistle, 1400Ripples, Fast Scattering), with results dependent on detector and run. A key methodological contribution is the optimization of $t_0$ to maximize Poisson-fit quality, yielding $\mu \approx 1.526$ and $t_0 = \frac{\mu}{\lambda}$. The findings provide a quantitative basis to categorize glitches by their degree of Poissonness, informing targeted mitigation strategies and aiding interpretation of their physical origins. Overall, the work offers a practical framework for distinguishing Poissonian from non-Poissonian glitch classes in gravitational-wave data.

Abstract

This work investigates the temporal distribution of glitches detected by LIGO, focusing on the morphological classification provided by the Gravity Spy project. Starting from the hypothesis that these events follow a Poisson process, we developed a statistical methodology to evaluate the agreement between the empirical distribution of glitches and an ideal Poisson model, using the coefficient of determination ($R^2$) as the main metric. The analysis was applied to real data from the LIGO detectors in Livingston and Hanford throughout the O3 run, as well as to synthetic datasets generated from pure Poisson distributions. The results show that while several morphologies exhibit good agreement with the proposed model, classes such as 1400Ripples, Fast Scattering, and Power Line display significant deviations ($R^2 \leq 0.6$), suggesting that their origins do not strictly follow Poissonian statistics. In some cases, a dependence on the detector or the observing run was also observed. This analysis provides a quantitative basis for distinguishing glitch classes based on their degree of "Poissonness", potentially supporting the development of more effective glitch mitigation strategies in gravitational wave detector data.

Figures (30)

• Figure 1: Behavior of $R^2$ as a function of the Poisson peak position and box length $t_0$.
• Figure 2: Spectrograms of some of the analyzed glitch morphologies, corresponding respectively to the classes Scattered Light, Koi Fish, Scratchy, Extremely Loud, Tomte and Whistle.
• Figure 3: Frequency distributions of selected glitch morphologies observed in different interferometers and observing runs.
• Figure 4: Histograms of Blip and Koi Fish glitches during O3a in the L1 detector.
• Figure 5: Histograms of Fast Scattering and Scattered Light glitches during O3a in the L1 detector.