Search templates for gravitational waves from inspiraling binaries: Choice of template spacing

TL;DR

This paper uses differential geometry to optimize the spacing and number of waveform templates for gravitational wave searches from inspiraling binaries, balancing detection efficiency and computational feasibility.

Contribution

It extends the formalism for template placement using differential geometry, providing estimates for template spacing, count, and computational requirements for on-line searches.

Findings

Estimated the number of templates needed for effective detection.

Quantified the computational power required for the search.

Provided a framework valid for arbitrary noise spectra.

Abstract

It has long been known that the search for gravitational waves from inspiraling binaries must be aided by signal processing methods, and that the technique of matched filtering is the most likely to be used for the detection of inspiral ``chirps''. This means that the output of an interferometer must be cross- correlated with many waveform templates to dig out faint signals buried in the noise. The templates are characterized by several parameters which vary continuously over some finite range; but because the amount of computing power available to perform the cross-correlations of the search templates with the output is finite, the actual templates used must be picked with certain discrete values of these parameters, which will have some finite spacing. If the spacing is too small, the number of templates (and therefore the computing power) needed to perform an on-line search becomes prohibitive; if the spacing is too large, many chirps will not be detected because the values of their parameters lie too far from those of the nearest template. In this paper I use differential geometry to extend the earlier formalism of Sathyaprakash and Dhurandhar to estimate the template spacing, number of templates, and computing power required for a single-pass, on-line search of the output of a single interferometer in terms of the fraction of events lost to parameter discretization. My formalism obtains these results with little numerical computation, and is valid for arbitrary noise spectra and template parameterizations. I find that the computing power needed for the most computationally intensive, reasonable search is of order several hundred Gigaflops. This will be feasible by the time LIGO is operational, but it is worth pursuing methods of reducing this number.