Scalable data-analysis framework for long-duration gravitational waves from compact binaries using short Fourier transforms

Abstract

We introduce a framework based on short Fourier transforms (SFTs) to analyze long-duration gravitational wave signals from compact binaries. Targeted systems include binary neutron stars observed by third-generation ground-based detectors and massive black hole binaries observed by the LISA space mission. In short, ours is an extremely fast, scalable, and parallelizable implementation of the gravitational wave inner product, a core operation of gravitational wave matched filtering. By operating on disjoint data segments, SFTs allow for efficient handling of noise nonstationarities, data gaps, and detector-induced signal modulations. We present a pilot application to early warning problems in both ground- and space-based next-generation detectors. Overall, SFTs reduce the computing cost of evaluating an inner product by three to five orders of magnitude, depending on the specific application, with respect to a nonoptimized approach. We release public tools to operate using the SFT framework, including a vectorized and hardware-accelerated reimplementation of a time-domain waveform. The inner product is the key building block of all gravitational wave data treatments; by speeding up this low-level element so massively, SFTs provide an extremely promising solution for current and future gravitational wave data-analysis problems.