Using Gravitational Wave Search Pipelines to Better Understand Glitches

The detection of gravitational waves would not be possible without search pipelines analyzing the output data of the LIGO and Virgo interferometers. The presence of transient noise, commonly referred to as glitches, alongside gravitational wave signals in the strain data makes it impossible for search pipelines to find genuine gravitational wave signals without also analyzing noise. Thus, it is critical to understand how gravitational wave search pipelines interact with transient noise. We present a machine learning method for identifying which search pipeline candidates are caused by transient noise, and apply this method to offline triggers of the GstLAL search pipeline in the second half of LIGO’s third observing run. We examine the parameters of triggers caused by different types of glitches and find that different types of transient noise appear uniquely in the relevant pipeline parameter spaces, and that the pipeline triggers caused by noise have characteristic differences between the LIGO Hanford and LIGO Livingston detectors. We further conclude that certain types of glitches over-represent themselves in the eyes of the pipeline, while the impact of other types is comparatively small. This method also allows us to identify which types of transient noise produce triggers that the pipeline believes to be strong candidates for genuine gravitational waves. We conclude that this method successfully identifies pipeline triggers caused by transient noise, and that the resulting analysis of these triggers can inform experts on the most damaging noise types, potentially resulting in focused efforts to mitigate the most harmful glitches and new pipeline design considerations.