GWtuna: Trawling through the data to find Gravitational Waves with Optuna and Jax

TL;DR

GWtuna introduces a novel, fast gravitational-wave search method using Optuna and JAX, employing black box optimization algorithms to efficiently identify binary neutron star mergers in data.

Contribution

The paper presents the first template bank free search for gravitational waves using Optuna's optimization algorithms, significantly reducing search time and computational resources.

Findings

TPE identifies binary neutron star mergers in 1 second median

CMA-ES recovers SNR in 48 seconds median

The method reduces computational cost compared to traditional searches

Abstract

GWtuna is a fast gravitational-wave search prototype built on Optuna (optimisation software library) and JAX (accelerator-orientated array computation library) [1, 2]. Using Optuna, we introduce black box optimisation algorithms and evolutionary strategy algorithms to the gravitational-wave community. Tree-structured Parzen Estimator (TPE) and Covariance Matrix Adaption Evolution Strategy (CMA-ES) have been used to create the first template bank free search and used to identify binary neutron star mergers. TPE can identify a binary neutron star merger in 1 second (median value) and less than 1000 matched-filter evaluations when 512 seconds of data is searched over. A stopping algorithm is used to curtail the TPE search if the signal-to-noise ratio (SNR) threshold has been reached, or the SNR has not improved in 500 evaluations. If the SNR threshold is surpassed, CMA-ES is used to recover the SNR and the template parameters in 9,000 matched filter iterations taking 48 seconds (median value). GWtuna showcases alternatives to the standard template bank search and therefore has the potential to revolutionise the future of gravitational-wave data analysis.