A scalable random forest regressor for combining neutron-star equation of state measurements: A case study with GW170817 and GW190425

TL;DR

This paper introduces a scalable machine learning approach using a random forest regressor to efficiently combine gravitational-wave data from neutron star mergers GW170817 and GW190425, improving constraints on neutron-star equations of state.

Contribution

The authors develop a machine learning-based method to interpolate likelihoods, enabling rapid and scalable combination of multiple gravitational-wave measurements for neutron star EOS constraints.

Findings

Constrained neutron star radius to approximately 11.6 km at 90% confidence.

Estimated pressure at twice nuclear saturation density with uncertainties.

Method allows for easy combination of multiple gravitational-wave signals.

Abstract

Gravitational-wave observations of binary neutron star coalescences constrain the neutron-star equation of state by enabling measurement of the tidal deformation of each neutron star. This deformation is determined by the tidal deformability parameter $\Lambda$, which was constrained using the first binary neutron star gravitational-wave observation, GW170817. Now, with the measurement of the second binary neutron star, GW190425, we can combine different gravitational-wave measurements to obtain tighter constraints on the neutron-star equation of state. In this paper, we combine data from GW170817 and GW190425 to place constraints on the neutron-star equation of state. To facilitate this calculation, we derive interpolated marginalized likelihoods for each event using a machine learning algorithm. These likelihoods, which we make publicly available, allow for results from multiple gravitational-wave signals to be easily combined. Using these new data products, we find that the radius of a fiducial 1.4 $M_\odot$ neutron star is constrained to $11.6^{+1.6}_{-0.9}$ km at 90% confidence and the pressure at twice the nuclear saturation density is constrained to $3.1^{+3.1}_{-1.3}\times10^{34}$ dyne/cm$^2$ at 90% confidence. This result is dominated by GW170817 and is consistent with findings from other works.