Simulation-based inference of radio millisecond pulsars in globular clusters

TL;DR

This paper introduces a novel likelihood-free inference method using neural networks to estimate the MSP population in globular clusters, overcoming limitations of traditional likelihood-based approaches and improving parameter constraints with future data.

Contribution

The study presents a new likelihood-free inference technique based on Marginal Neural Ratio Estimation for better MSP population estimates in GCs, demonstrated on Terzan 5.

Findings

Likelihood-free method yields well-behaved posteriors.

Adding diffuse radio emission data improves estimates.

Lowering detection thresholds enhances parameter constraints.

Abstract

Millisecond pulsars (MSPs) are abundant in globular clusters (GCs), which offer favorable environments for their creation. While the advent of recent, powerful facilities led to a rapid increase in MSP discoveries in GCs through pulsation searches, detection biases persist. In this work, we investigate the ability of current and future detections in GCs to constrain the parameters of the MSP population in GCs through a careful study of their luminosity function. Parameters of interest are the number of MSPs hosted by a GC, as well as the mean and the width of their luminosity function, which are typically affected by large uncertainties. While, as we show, likelihood-based studies can lead to ill-behaved posterior on the size of the MSP population, we introduce a novel, likelihood-free analysis, based on Marginal Neural Ratio Estimation, which consistently produces well-behaved posteriors. We focus on the GC Terzan 5, which currently counts 48 detected MSPs. We find that about 158 MSPs should be hosted in this GC, but the uncertainty on this number remains large. We explore the performance of our new method on simulated Terzan 5-like datasets mimicking possible future observational outcomes. We find that significant improvement on the posteriors can be obtained by adding a reliable measurement of the diffuse radio emission of the GC to the analysis or by improving the detection threshold of current radio pulsation surveys by at least a factor two.