Detecting Gravitational-Wave Transients at $5\sigma$: A Hierarchical Approach

As second-generation gravitational-wave detectors prepare to analyze data at unprecedented sensitivity, there is great interest in searches for unmodeled transients, commonly called bursts. Significant effort has yielded a variety of techniques to identify and characterize such transient signals, and many of these methods have been applied to produce astrophysical results using data from first-generation detectors. However, the computational cost of background estimation remains a challenging problem; it is difficult to claim a $5\sigma$ detection with reasonable computational resources without paying for efficiency with reduced sensitivity. We demonstrate a hierarchical approach to gravitational-wave transient detection, focusing on long-lived signals, which can be used to detect transients with significance in excess of $5\sigma$ using modest computational resources. In particular, we show how previously developed seedless clustering techniques can be applied to large data sets to identify high-significance candidates without having to trade sensitivity for speed.

Figure 1

(a) The FAP versus the single-detector statistic ${\mathrm{SNR}}_{\mathrm{tot}}^{\max (I)}$. The blue curve indicates the distribution of recolored noise for Advanced LIGO at design sensitivity. The vertical lines show the median values for data including an accretion disk instability signal (ADI 2; face-on, optimal orientation) with different colors corresponding to different distances. The dashed black lines show which data are removed via the single-detector cut. (b) The FAP versus the coherent statistic $\mathrm{\Lambda }$. The blue curve indicates the distribution of recolored noise for Advanced LIGO (Hanford and Livingston) at design sensitivity. The vertical lines show the median values for data including an accretion disk instability signal (ADI 2; face-on, optimal orientation) with different colors corresponding to different distances. The dashed black lines indicate the threshold required for $3\sigma$, $4\sigma$, and $5\sigma$ detection in one week of data.