Systematic errors in searches for nanohertz gravitational waves

TL;DR

This paper examines whether systematic errors in pulsar timing noise models could bias gravitational wave detection, finding that current methods are robust and may even underestimate the true significance of the signal.

Contribution

It demonstrates that plausible noise model misspecifications do not falsely produce gravitational wave signals and that current detection significance is likely reliable.

Findings

Quasi-resampling underestimates significance in pure noise datasets

Systematic errors are unlikely to bias gravitational wave evidence

Reported gravitational wave evidence is robust against common noise model misspecifications

Abstract

A number of pulsar timing arrays have recently reported preliminary evidence for the existence of a nanohertz frequency gravitational-wave background. These analyses rely on detailed noise analyses, which are inherently complex due to the many astrophysical and instrumental factors that contribute to the pulsar noise budget. We investigate whether realistic systematic errors, stemming from misspecified noise models that fail to capture salient features of the pulsar timing noise, could bias the evidence for gravitational waves. We consider two plausible forms of misspecification: small unmodeled jumps and unmodeled chromatic noise. Using simulated data, we calculate the distribution of the commonly used optimal statistic with no signal present and using plausibly misspecified noise models. By comparing the optimal statistic distribution with the distribution created using ``quasi-resampling'' techniques (such as sky scrambles and phase shifts), we endeavor to determine the extent to which plausible misspecification might lead to a false positive. The results are reassuring: we find that quasi-resampling techniques tend to underestimate the significance of pure-noise datasets. We conclude that recent reported evidence for a nanohertz gravitational-wave background is likely robust to the most obvious sources of systematic errors; if anything, the significance of the signal is potentially underestimated.