Detecting compact binary coalescences with seedless clustering

Compact binary coalescences are a promising source of gravitational waves for second-generation interferometric gravitational-wave detectors. Although matched filtering is the optimal search method for well-modeled systems, alternative detection strategies can be used to guard against theoretical errors (e.g., involving new physics and/or assumptions about spin or eccentricity) while providing a measure of redundancy. In a previous paper, we showed how “seedless clustering” can be used to detect long-lived gravitational-wave transients in both targeted and all-sky searches. In this paper, we apply seedless clustering to the problem of low-mass ($M_{\text{total}}\le 10M_{\odot }$) compact binary coalescences for both spinning and eccentric systems. We show that seedless clustering provides a robust and computationally efficient method for detecting low-mass compact binaries.

Figure 1

The plot on the left shows $\rho (t;f)$ for a simulated eccentric ($\epsilon =0.2$) BNS signal injected on top of Monte Carlo detector noise. The component masses are $1.4M_{\odot }$. The chirping signal appears as a faintly visible track of lighter-than-average pixels. The horizontal lines are frequency notches to remove instrumental artifacts. On the right is the recovery obtained with seedless clustering. The signal is recovered with a $\mathrm{FAP}<0.1\%$.

Figure 2

The plot on the left shows $\rho (t;f)$ during a recent LIGO engineering run, in which data from a LIGO subsystem—not sensitive to GW strain—are recolored to produce semirealistic detector noise. On the right is the seedless clustering recovery, which is able to detect the injected signal with high confidence $\mathrm{FAP}<0.1\%$, despite the relatively poor data quality. (Though it is not immediately apparent from these plots, five segments are identified as characteristic of nonstationary noise [11].) The signal is recovered with $\mathrm{FAP}<0.1\%$.

Figure 3

Data from a recent LIGO engineering run, in which data from a LIGO subsystem—not sensitive to GW strain—are recolored to produce semirealistic detector noise. A simulated binary neutron star signal has been added to the data. The top plot shows a wavelet transform of single-detector auto-power [13], currently in wide use for diagnostics. The injected waveform, which ends at $t=250$, is difficult to make out by eye. The middle plot shows a spectrogram of $\rho (t;f)$. The injection, though faint, is visible between 200–250 s. The bottom plot shows the reconstructed track using seedless clustering; $\mathrm{FAP}<0.1\%$.