The False Alarms induced by Gaussian Noise in Gravitational Wave Detectors

TL;DR

This paper investigates the correlation properties of Gaussian noise in gravitational wave detectors, deriving a new method to accurately estimate false alarm rates by accounting for time correlations in the matched filter SNR.

Contribution

It introduces a first-principles calculation of false alarm rates considering time correlations, improving detection thresholds in GW data analysis.

Findings

Gaussian noise exhibits significant time correlation affecting false alarm rates.

The effective sampling rate depends on template parameters and noise PSD.

Application to GWTC-3 candidates suggests some events could be noise fluctuations.

Abstract

Gaussian noise is an irreducible component of the background in gravitational wave (GW) detectors. Although stationary Gaussian noise is uncorrelated in frequencies, we show that there is an important correlation in time when looking at the matched filter signal to noise ratio (SNR) of a template, with a typical autocorrelation time that depends on the template and the shape of the noise power spectral density (PSD). Taking this correlation into account, we compute from first principles the false alarm rate (FAR) of a template in Gaussian noise, defined as the number of occurrences per unit time that the template's matched filter SNR goes over a threshold $\rho$. We find that the Gaussian FAR can be well approximated by the usual expression for uncorrelated noise, if we replace the sampling rate by an effective sampling rate that depends on the parameters of the template, the noise PSD and the threshold $\rho$. This results in a minimum SNR threshold that has to be demanded to a given GW trigger, if we want to keep events generated from Gaussian noise below a certain FAR. We extend the formalism to multiple detectors and to the analysis of GW events. We apply our method to the GW candidates added in the GWTC-3 catalog, and discuss the possibility that GW200308\_173609 and GW200322\_091133 could be generated by Gaussian noise fluctuations.