Fast waveform generation for gravitational waves using evolutionary algorithms

Quirijn Meijer,Sarah Caudill
DOI: https://doi.org/10.1103/physrevd.110.043035
IF: 5.407
2024-08-28
Physical Review D
Abstract:Gravitational-wave analyses depend heavily on waveforms that model the evolution of compact binary coalescences as seen by observing detectors. In many cases these waveforms are given by waveform approximants, models that approximate the amplitude and phase of the waveform at a set of frequencies. Because of their omnipresence, improving the speed at which approximants can generate waveforms is crucial to accelerating the overall analysis of gravitational-wave detections. An optimization algorithm is proposed that can select at which frequencies in the spectrum an approximant should compute the power of a waveform, and at which frequencies the power can be safely interpolated at a minor loss in accuracy. The algorithm used is an evolutionary algorithm modeled after the principle of natural selection, iterating frequency arrays that perform better at every iteration. As an application, the candidates proposed by the algorithm are used to reconstruct signal-to-noise ratios. It is shown that the imrp henom xphm approximant can be sped up by at least 30% at a loss of at most 2.87% on the drawn samples, measured by the accuracy of the reconstruction of signal-to-noise ratios. The behavior of the algorithm as well as lower bounds on both speedup and error are explored, leading to a proposed proof of concept candidate that obtains a speedup of 46% with a maximum error of 0.5% on a sample of the parameter space used. https://doi.org/10.1103/PhysRevD.110.043035 © 2024 American Physical Society
astronomy & astrophysics,physics, particles & fields
What problem does this paper attempt to address?