[Paper Review] The GstLAL Search Analysis Methods for Compact Binary Mergers in Advanced LIGO's Second and Advanced Virgo's First Observing Runs

S. Sachdev, Sarah Caudill+27 more|arXiv (Cornell University)|Jan 24, 2019

Pulsars and Gravitational Waves ResearchPhysics and Astronomy83 citations

TL;DR

This paper describes updates to the GstLAL pipeline for detecting compact binary mergers during LIGO's second observing run (O2) and Virgo's first observing run, including latency reductions, data handling improvements, and expanded parameter space.

ABSTRACT

After their successful first observing run (September 12, 2015 - January 12, 2016), the Advanced LIGO detectors were upgraded to increase their sensitivity for the second observing run (November 30, 2016 - August 26, 2017). The Advanced Virgo detector joined the second observing run on August 1, 2017. We discuss the updates that happened during this period in the GstLAL-based inspiral pipeline, which is used to detect gravitational waves from the coalescence of compact binaries both in low latency and an offline configuration. These updates include deployment of a zero-latency whitening filter to reduce the over-all latency of the pipeline by up to 32 seconds, incorporation of the Virgo data stream in the analysis, introduction of a single-detector search to analyze data from the periods when only one of the detectors is running, addition of new parameters to the likelihood ratio ranking statistic, increase in the parameter space of the search, and introduction of a template mass-dependent glitch-excision thresholding method.

Motivation & Objective

• Motivate the need for rapid, reliable gravitational-wave detections to enable multi-messenger astronomy and timely alerts.
• Describe the GstLAL pipeline updates implemented for Advanced LIGO's O2 and Virgo's first run to improve latency, sensitivity, and parameter coverage.
• Explain the methodological changes in waveform modeling, template bank construction, data conditioning, and statistical ranking used to distinguish signals from noise.

Proposed method

• Use a matched-filtering approach with a bank of GR-based waveforms spanning intrinsic parameters (masses and spins) and extrinsic parameters (distance, orientation).
• Maximize the complex SNR over time and phase using a time-domain implementation and a LLOID-inspired reduced-filter set.
• Implement a zero-latency whitening filter to reduce online latency by replacing a 16–32 s PSD-based whitening with a minimum-phase time-domain approximation.
• Group templates into sub-banks (bar{θ} bins) using a high-mass duration-based scheme to better model noise, and adapt the template density in high-mass regions.
• Compute a multi-dimensional log-likelihood-ratio statistic to rank candidates based on signal and noise models; construct background with Monte Carlo sampling of single-detector triggers for FAR estimation.
• Introduce single-detector ranking for periods with only one detector online and adjust binning and PDFs for SNR and ξ² to accommodate O2 configuration.

Experimental results

Research questions

• RQ1 How can the GstLAL pipeline be enhanced to achieve lower latency and higher sensitivity for compact binary mergers during O2?
• RQ2 What template-bank and data-conditioning changes are needed to accurately model noise and improve background estimation in the expanded O2 parameter space?
• RQ3 How should the likelihood-ratio ranking be adapted to include single-detector triggers and additional parameters when multiple detectors are online?
• RQ4 What impact does a zero-latency whitening filter have on online detection latency and consistency with offline analysis?

Key findings

• Latency was reduced by approximately 40 seconds in O2 with the zero-latency whitening filter.
• The template bank was expanded to 2–400 M_sun total mass, with high-mass grouping based on template duration rather than chirp mass and effective spin for better background modeling.
• A linear gating scheme for data conditioning was introduced to reduce glitches for high-mass templates without gating out real signals.
• Single-detector triggers were incorporated into online ranking, expanding the analysis beyond coincident multi-detector coincidences.
• The background and signal models were refined with updated SNR-ξ² PDFs and binning to improve false-alarm probability estimation for O2.