Effects of periodicity in observation scheduling on parameter estimation of pulsar glitches

TL;DR

This paper investigates how periodic observation scheduling affects pulsar glitch parameter estimation, revealing biases in traditional methods and proposing models that improve glitch detection and measurement accuracy.

Contribution

It introduces a systematic analysis of degeneracies caused by periodic scheduling and evaluates new approaches like hidden Markov models for unbiased glitch estimation.

Findings

Bias towards small glitches in traditional timing methods.

Local frequency estimates reduce bias.

Hidden Markov models detect multiple solutions and improve glitch size recovery.

Abstract

In certain pulsar timing experiments, where observations are scheduled approximately periodically (e.g. daily), timing models with significantly different frequencies (including but not limited to glitch models with different frequency increments) return near-equivalent timing residuals. The average scheduling aperiodicity divided by the phase error due to time-of-arrival uncertainties is a useful indicator of when the degeneracy is important. Synthetic data are used to explore the effect of this degeneracy systematically. It is found that phase-coherent tempo2 or temponest-based approaches are biased sometimes toward reporting small glitch sizes regardless of the true glitch size. Local estimates of the spin frequency alleviate this bias. A hidden Markov model is free from bias towards small glitches and announces explicitly the existence of multiple glitch solutions but sometimes fails to recover the correct glitch size. Two glitches in the UTMOST public data release are re-assessed, one in PSR J1709$-$4429 at MJD 58178 and the other in PSR J1452$-$6036 at MJD 58600. The estimated fractional frequency jump in PSR J1709$-$4429 is revised upward from $\Delta f/f = (54.6\pm 1.0) \times 10^{-9}$ to $\Delta f/f = (2432.2 \pm 0.1) \times 10^{-9}$ with the aid of additional data from the Parkes radio telescope. We find that the available UTMOST data for PSR J1452$-$6036 are consistent with $\Delta f/f = 270 \times 10^{-9} + N/(fT)$ with $N = 0,1,2$, where $T \approx 1\,\text{sidereal day}$ is the observation scheduling period. Data from the Parkes radio telescope can be included, and the $N = 0$ case is selected unambiguously with a combined dataset.