GRAPE: Genetic Routine for Astronomical Period Estimation

TL;DR

GRAPE is a genetic algorithm designed for improved period estimation in variable astrophysical objects, outperforming traditional methods especially for sinusoidal and sawtooth signals in survey data with artifacts.

Contribution

The paper introduces GRAPE, a novel genetic algorithm optimized with a Bayesian Lomb-Scargle fitness function for more accurate period estimation in survey data.

Findings

GRAPE outperforms periodograms on sinusoidal and sawtooth light curves.

Performance degrades with non-sinusoidal signals but remains effective.

Runtime efficiency is achieved for light curves with over 500 data points.

Abstract

Period estimation is an important task in the classification of many variable astrophysical objects. Here we present GRAPE: Genetic Routine for Astronomical Period Estimation, a genetic algorithm optimised for the processing of survey data with spurious and aliased artefacts. It uses a Bayesian Generalised Lomb-Scargle (BGLS) fitness function designed for use with the Skycam survey conducted at the Liverpool Telescope. We construct a set of simulated light curves using both regular and Skycam survey cadence with four types of signal: sinusoidal, sawtooth, symmetric eclipsing binary and eccentric eclipsing binary. We apply GRAPE and a BGLS periodogram to this data and show that the performance of GRAPE is superior to the periodogram on sinusoidal and sawtooth light curves with relative hit rate improvement of 18.2% and 6.4% respectively. The symmetric and eccentric eclipsing binary light curves have similar performance on both methods. We show the Skycam cadence is sufficient to correctly estimate the period for all of the sinusoidal shape light curves although this degrades with increased non-sinusoidal shape with sawtooth, symmetric binary and eccentric binary light curves down by 20%, 30% and 35% respectively. The runtime of GRAPE demonstrates that light curves with more than 500-1000 data points achieve similar performance in less computing time. The GRAPE performance can be matched by a frequency spectrum with an oversampled fine-tuning grid at the cost of almost doubling the runtime. Finally, we propose improvements which will extend this method to the detection of quasi-periodic signals and the use of multiband light curves.