Searching for gravitational-wave transients with a qualitative signal model: Seedless clustering strategies

Abstract

Gravitational-wave bursts are observable as bright clusters of pixels in spectrograms of strain power. Clustering algorithms can be used to identify candidate gravitational-wave events. Clusters are often identified by grouping together seed pixels in which the power exceeds some threshold. If the gravitational-wave signal is long-lived, however, the excess power may be spread out over many pixels, none of which are bright enough to become seeds. Without seeds, the problem of detection through clustering becomes more complicated. In this paper, we investigate seedless clustering algorithms in searches for long-lived narrow-band gravitational-wave bursts. Using four astrophysically motivated test waveforms, we compare a seedless clustering algorithm to two algorithms using seeds. We find that the seedless algorithm can detect gravitational-wave signals (at a fixed false-alarm and false-dismissal rate) at distances between 1:5-2Â those achieved with the seed-based clustering algorithms, corresponding to significantly increased detection volumes: 4:2-7:4Â . This improvement in sensitivity may extend the reach of second-generation detectors such as Advanced LIGO and Advanced Virgo deeper into astrophysically interesting distances.

Prospects for searches for long-duration gravitational-waves without time slides

“…One algorithm used to search for long-duration gravitational waves is seedless clustering, which integrate along many different paths in spectrograms. This algorithm is sensitive to signals that can be well-approximated by parameterized curves, and the advantage of seedless clustering is most pronounced for long and weak signals [15][16][17][18]. We have previously shown how seedless clustering algorithms can be used to significantly enhance the sensitivity of searches for signals of this type [15].…”

“…This algorithm is sensitive to signals that can be well-approximated by parameterized curves, and the advantage of seedless clustering is most pronounced for long and weak signals [15][16][17][18]. We have previously shown how seedless clustering algorithms can be used to significantly enhance the sensitivity of searches for signals of this type [15]. Although seedless clustering algorithms are embarrassingly parallel [31], and therefore computations can be performed on graphical processor units, seedless clustering searches are still limited by computation of the noise background.…”

“…For a single event, therefore, 1,744,278 time-slides must be performed to determine whether an event has an SNR exceeding a 5σ threshold. The difficulty in reaching these levels is due to computationally intensive calculations like the matched filter used in compact binary coalescence [24], calculation of the coherent SNR [25] used in burst searches, and potentially seedless clustering [15][16][17][18] in a long-duration transient search.…”

“…There are a number of papers contained in the literature about the sensitivity of gravitational-wave detectors to long-duration gravitational waves [6,[15][16][17][18]. In general, the sensitiv-ity studies have been performed by running the analysis on 1,000 f t-maps to reach a FAP of 0.1%, and the sensitivity to various waveform models are computed relative to this number.…”

“…Uncertain models exist for more exotic sources of long-lived transients, including emission from rotational instabilities in protoneutron stars [6][7][8][9] and black-hole accretion disk instabilities [10][11][12]. When a matched filter search is not possible, searches for unmodeled long-lived transients [13][14][15][16][17][18] can be employed.…”