From NS observations to nuclear matter properties: a machine learning approach

TL;DR

This paper demonstrates that Bayesian neural networks can reliably infer nuclear matter properties from neutron star mass-radius observations, providing uncertainty estimates and effective predictions within the model's training domain.

Contribution

The study introduces a Bayesian neural network approach for extracting nuclear matter properties from observational data, incorporating uncertainty quantification and testing on simulated and real data.

Findings

BNNs accurately predict nuclear matter properties within training data range.

The model estimated specific parameters with associated uncertainties from real pulsar data.

Results show BNNs as a promising tool for nuclear physics inference from astrophysical observations.

Abstract

This study is devoted to the inference problem of extracting the nuclear matter properties directly from a set of mass-radius observations. We employ Bayesian neural networks (BNNs), which is a probabilistic model capable of estimating the uncertainties associated with its predictions. To simulate different noise levels on the $M(R)$ observations, we create three different sets of mock data. Our results show BNNs as an accurate and reliable tool for predicting the nuclear matter properties whenever the true values are not completely outside the training dataset statistics, i.e., if the model is not heavily dependent on its extrapolating capacities. Using real mass-radius pulsar data, the model predicted, for instance, $L_{\text{sym}}=39.80\pm17.52 $ MeV and $K_{\text{sym}}=-101.67\pm62.86 $ MeV ($2\sigma$ interval). Our study provides a valuable inference framework when new NS data becomes available.