A glitch in the millisecond pulsar J0900$-$3144

TL;DR

This study reports the detection of a small glitch in the millisecond pulsar PSR J0900-3144, using a ~14-year, multi-telescope PTA dataset. The glitch is modelled with a standard spin-frequency step and spin-down change, embedded in a comprehensive noise model that includes achromatic red noise and chromatic dispersion measure variations, analysed via Bayesian methods. The results yield a fractional frequency step of $\Delta \nu_g/\nu = 1.15(13) \times 10^{-12}$ and a spin-down change of $\Delta \dot{\nu}_g/\dot{\nu} = -6.3(79) \times 10^{-4}$, with strong Bayesian support for the glitch over a no-glitch model, and simulations show that unmodelled glitches can bias red-noise inferences. The findings lead to an updated MSP glitch rate of $2.5 \pm 1.4$ glitches per 1000 pulsar-years, and highlight the importance of incorporating glitch models in PTA analyses to avoid dow-weighting pulsars in gravitational wave background searches.

Abstract

We report the detection of a glitch in the millisecond pulsar (MSP) PSR J0900$-$3144, which is included in the European, MeerKAT and Parkes pulsar timing array experiments. The dataset combines observations from the MeerKAT, Nançay, Lovell, and Murriyang telescopes, spanning a total baseline of approximately 14 years. The glitch occurred on MJD~59942(17), with a measured fractional spin frequency step of $Δν_g / ν=1.15(13) \times 10^{-12}$. This event represents the third glitch detected in a MSP, following those in PSRs B1821$-$24A and J0613$-$0200. Although smaller in amplitude than the previous two, the glitch in PSR J0900$-$3144 is of a comparable order of magnitude. The updated MSP glitch rate is $2.5(1)\times 10^{-3}$ glitches per pulsar per year, which suggests it is likely current PTAs will detect another MSP glitch within five years. Using simulations, we demonstrate that such small glitches can go undetected, especially in short datasets such as those from new PTAs, and can bias the inferred achromatic noise model parameters, potentially leading to the down-weighting of the pulsar in gravitational wave background searches.

Figures (8)

• Figure 1: Frequency and time coverage of the different telescopes used in this study. Each point represents an individual time of arrival (ToA) measurement. Each observation has one ToA per sub-band, see Table \ref{['table:telescope_data']} for further details.
• Figure 2: Comparison of timing residuals and spin parameter evolution for PSR J0900$-$3144. The top panel shows the frequency averaged pre-fit residuals, weighted by the uncertainty in ToA and the white noise parameters for each observing system for each observing epoch, clearly highlighting the glitch event. The second panel presents the frequency averaged uncertainty-weighted post-fit residuals after subtracting the glitch model. The third and fourth panels show an estimate of the deviations from the model pulsar spin frequency and frequency derivative, as obtained from fitting a quadratic to the short sections of the residuals in the top panel. $\Delta{\nu}$ is computed over 200-day windows, and $\Delta\dot{\nu}$ over 600-day windows. The vertical dashed line marks the glitch epoch.
• Figure 3: Post-fit timing residuals from the MeerKAT-only dataset, with and without including the glitch in the model. For clarity, the residuals from each observation have been combined to a single frequency-averaged residual, weighted by the ToA uncertainty and the white noise parameters.
• Figure 4: Comparison of red noise parameter posteriors between glitch and non-glitch models for different datasets. The unmodelled glitch has the effect of biasing both amplitude and spectral exponent of red noise.
• Figure 5: Bayesian evidence in favour of the glitch model as a function of total observation span and glitch amplitude. The horizontal dashed line indicates a Bayes factor of 10, broadly considered to be a significant detection.