Gaussian mixture model and population synthesis of radio pulsars

TL;DR

This paper evaluates the effectiveness of Gaussian mixture models in identifying distinct sub-populations of radio pulsars and finds limitations in their stability and discriminative power.

Contribution

The study extends previous GMM applications to synthetic pulsar populations, assessing their ability to distinguish physically different groups and models.

Findings

GMM is oversensitive to parameter variations for rapidly evolving pulsars

GMM does not reliably identify sub-populations with different initial conditions

GMM fails to effectively discriminate between different population synthesis models

Abstract

Recently, Lee et al. used Gaussian mixture models (GMM) to study the radio pulsar population. In the distribution of normal pulsars in the P-dotP plane, they found four clusters. We develop this approach further and apply it to different synthetic pulsar populations in order to determine whether the method can effectively select groups of sources that are physically different. We check several combinations of initial conditions as well as the models of pulsar evolution and the selection effects. We find that in the case of rapidly evolving objects, the GMM is oversensitive to parameter variations and does not produce stable results. We conclude that the method does not help much to identify the sub-populations with different initial parameters or/and evolutionary paths. For the same reason, the GMM does not discriminate effectively between theoretical population synthesis models of normal radio pulsars.