No abstract

Machine learning methods are increasingly helping astronomers identify new radio pulsars. However, they require a large amount of labelled data, which is time consuming to produce and biased. Here we describe a Semi-Supervised Generative Adversarial Network (SGAN) which achieves better classification performance than the standard supervised algorithms using majority unlabelled datasets. We achieved an accuracy and mean F-Score of 94.9 per cent trained on only 100 labelled candidates and 5000 unlabelled candidates compared to our standard supervised baseline which scored at 81.1 per cent and 82.7 per cent respectively. Our final model trained on a much larger labelled dataset achieved an accuracy and mean F-score value of 99.2 per cent and a recall rate of 99.7  per cent. This technique allows for high quality classification during the early stages of pulsar surveys on new instruments when limited labelled data is available. We open-source our work along with a new pulsar-candidate dataset produced from the High Time Resolution Universe - South Low Latitude Survey. This dataset has the largest number of pulsar detections of any public dataset and we hope it will be a valuable tool for benchmarking future machine learning models.

We present the results of processing an additional 44 % of the High Time Resolution Universe South Low Latitude (HTRU-S LowLat) pulsar survey, the most sensitive blind pulsar survey of the southern Galactic plane to date. Our partially-coherent segmented acceleration search pipeline is designed to enable the discovery of pulsars in short, highly-accelerated orbits, while our 72-min integration lengths will allow us to discover pulsars at the lower end of the pulsar luminosity distribution. We report the discovery of 40 pulsars, including three millisecond pulsar-white dwarf binary systems (PSRs J1537−5312, J1547−5709 and J1618−4624), a black-widow binary system (PSR J1745−23) and a candidate black-widow binary system (PSR J1727−2951), a glitching pulsar (PSR J1706−4434), an eclipsing binary pulsar with a 1.5-yr orbital period (PSR J1653−45), and a pair of long spin-period binary pulsars which display either nulling or intermittent behaviour (PSRs J1812−15 and J1831−04). We show that the total population of 100 pulsars discovered in the HTRU-S LowLat survey to date represents both an older and lower-luminosity population, and indicates that we have yet to reach the bottom of the luminosity distribution function. We present evaluations of the performance of our search technique and of the overall yield of the survey, considering the 94 % of the survey which we have processed to date. We show that our pulsar yield falls below earlier predictions by approximately 25 % (especially in the case of millisecond pulsars), and discuss explanations for this discrepancy as well as future adaptations in RFI mitigation and searching techniques which may address these shortfalls.

Relativistic binary pulsars orbiting white-dwarfs and neutron stars have already provided excellent tests of gravity. However, despite observational efforts, a pulsar orbiting a black hole has remained elusive. One possible explanation is the extreme Doppler smearing caused by the pulsar’s orbital motion which changes its apparent spin frequency during an observation. The classical solution to this problem has been to assume a constant acceleration/jerk for the entire observation. However, this assumption breaks down when the observation samples a large fraction of the orbit. This limits the length of search observations, and hence their sensitivity. This provides a strong motivation to develop techniques that can find compact binaries in longer observations. Here we present a GPU-based radio pulsar search pipeline that can perform a coherent search for binary pulsars by directly searching over three or five Keplerian parameters using the template-bank algorithm. We compare the sensitivity obtained from our pipeline with acceleration and jerk search pipelines for simulated pulsar-stellar-mass black hole binaries and observations of PSR J0737-3039A. We also discuss the computational feasibility of our pipeline for untargeted pulsar surveys and targeted searches. Our benchmarks indicate that circular orbit searches for P-BH binaries with spin-period Pspin ≥ 20ms covering the 3-10 Tobs regime are feasible for the High Time Resolution Universe pulsar survey. Additionally, an elliptical orbit search in Globular clusters for Pspin ≥ 20ms pulsars orbiting intermediate-mass black holes in the 5-10 Tobs regime is feasible for observations shorter than 2 hours with an eccentricity limit of 0.1.

The observable population of double neutron star (DNS) systems in the Milky Way allow us to understand the nature of supernovae and binary stellar evolution. Until now, all DNS systems in wide orbits (Porb > 1 day) have been found to have orbital eccentricities, e > 0.1. In this paper, we report the discovery of pulsar PSR J1325−6253: a DNS system in a 1.81 day orbit with a surprisingly low eccentricity of just e = 0.064. Through 1.4 yr of dedicated timing with the Parkes radio telescope we have been able to measure its rate of advance of periastron, $\dot{\omega }=0.138^{\circ }\pm 0.002^{\circ }yr^{-1}$. If this induced $\dot{\omega }$ is solely due to general relativity then the total mass of the system is, Msys = 2.57 ± 0.06 M⊙. Assuming an edge-on orbit the minimum companion mass is constrained to be Mc, min > 0.98 M⊙ which implies the pulsar mass is Mp, max < 1.59 M⊙. Its location in the P-$\dot{P}$ diagram suggests that, like other DNS systems, PSR J1325−6253 is a recycled pulsar and if its mass is similar to the known examples (>1.3 M⊙), then the companion neutron star is probably less than ∼1.25 M⊙ and the system is inclined at about 50○-60○. The low eccentricity along with the wide orbit of the system strongly favours a formation scenario involving an ultra-stripped supernova explosion.