Improved techniques for pulsar data analysis

Abstract

Pulsars are rapidly rotating, strongly magnetised neutron stars with spin periods ranging from few milliseconds to tens of seconds. They emit beams of electromagnetic radiation that are predominantly detected as periodic pulsations at the radio end of the spectrum. Since the first pulsar detection in 1967, there have been close to 3000 such systems discovered and they have proven to be excellent tools for a variety of scientific applications. This includes areas in fundamental physics ranging from testing theories of gravity to implications for condensed matter physics via understanding the equation of state. Despite the large number of discoveries, there exist several open questions related to pulsar astronomy that can be addressed by pushing the boundary of pulsar discoveries beyond the current population census. Chapter 1 gives a detailed description of pulsar properties and motivates the need to search for pulsars. Chapter 2 describes the telescope signal chain and the methodology behind searching for new pulsars using radio telescopes. It also motivates the advantage of using interferometers for conducting large scale pulsar surveys, the scope for which has been, to date, limited. <br /> Chapter 3 presents the pulsar searching infrastructure developed for MeerKAT and introduces the Transients and Pulsars with MeerKAT (TRAPUM) survey. This survey is targeted at sources where the probability of finding new radio pulsars is high. The TRAPUM infrastructure provides a large field of view via a multi-beam beamformer that is capable of synthesising up to 864 coherent beams using the full array of MeerKAT dishes. The highlight is the use of open source tools like Kubernetes to build an automated cloud computing setup that can process multi-beam data in quasi real-time. In order to test the efficiency of the processing pipelines, a multi-beam search was conducted on globular cluster Terzan 5 and the results are discussed. Although no new pulsars were discovered, an upper limit of ~ 18 μJy was placed on the flux density of a possible new pulsar in the cluster. This limit holds for binary pulsars in circular orbits with orbital periods above 5 hours and whose companion masses are below 4 Msun. Besides this, advantages of the multi-beam Terzan 5 data are put forth for conducting future searches on this globular cluster. <br /> Chapter 4 introduces the MPIfR Galactic Plane survey, which is the most sensitive Galactic plane survey conducted in the Southern Hemisphere. This multi-purpose survey serves a wide range of scientific fields including pulsars, transients, Galactic magnetic fields, continuum and spectral line studies. A major part of the survey is aimed at using the S-Band receivers (operating between 1.7 to 3.5 GHz) built at the MPIfR to probe deeper into the Galactic plane as well as the Galactic centre region. From the pulsar searching aspect, simulations show that 476±23 and 220±15 new pulsars are expected to be discovered at L-Band (1.284 GHz) and S-Band (2.4 GHz). A 24-hour pilot survey was conducted at L-Band (centered at 1284 MHz) on survey fields that have been previously imaged by MeerKAT. While candidate discoveries await confirmation, the known pulsar redetections show reasonable detection significance as expected from the radiometer equation. The survey coverage achieved assuming a minimum fractional telescope gain of 0.5 is 78 per cent. <br /> Chapter 5 describes a collaborative effort between MPIfR and multinational software company SAP for pulsar candidate classification on a large scale. Tool-kits were developed for deployment of pulsar candidate classifiers on a large scale using the SAP Data Intelligence platform. As a use-case, 3.2 million candidates generated from the High Time Resolution Universe South low-lat survey were put through a pipeline developed using Data Intelligence containing multiple redetections of 60 known pulsars. The pipeline uses publicly available machine learning models (trained in different ways on different data sets) to generate scores for each candidate. The score indicates the likelihood of a candidate being a pulsar. Based on standard machine learning evaluation metrics, the models were compared against each other to test their robustness. The most robust model is shown to be a classifier whose training set contained pulsar detections with a wide range of spin periods and duty cycles. This model is reported to have an area under the receiver operating characteristic (ROC) curve of 0.988, which for an ideal classifier would be 1.0. <br /> Chapter 6 presents a detailed analysis of the pulse profile instability of PSR J1022+1001, a pulsar that is regularly timed by pulsar timing arrays with the aim of detecting nanohertz gravitational waves. The pulsar has a rich and conflicting history of results pertaining to its purported profile instability and its likely cause. In this analysis, data spanning 20 years including two different recording instruments from the Effelsberg telescope and a recording instrument from the Parkes telescopes were used. Previously used methods of characterising the profile stability were modified and adapted to improve their robustness. The results show a significant amount of long term instability in the pulse shape. By accounting for instrumental effects like polarisation miscalibration and propagation effects like interstellar scintillation, up to 25 per cent of the instability could be explained. This indicates that the observed instability is most likely intrinsic to the pulsar.