Novel Search Techniques to Detect Pulsar Black Hole Binaries in Radio Observations

Abstract

Pulsars are rapidly rotating highly magnetised neutron stars that emit beams of electromagnetic radiation from their magnetic poles. These compact objects are unique as they are one of the densest forms of matter known in the Universe. Discovering more pulsars are crucial as they have a wide range of scientific applications from studies of strong-field tests of Gravity, neutron star physics, condensed matter physics and cosmology being a few examples. While all pulsars have their own applications, the bulk of this thesis will concentrate on the techniques involved in finding new relativistic binary pulsars particularly the yet to be detected pulsar black-hole binary which can be used to test General Relativity and alternate theories of gravity in the quasi-stationary strong-field regime. In Chapter 1, I lay the foundations of this work by expanding on the basics of pulsar phenomenology, its formation, the diverse types of pulsars already known and their properties in the radio regime. In Chapter 2, I describe the various processes involved in finding a pulsar in a modern pulsar search pipeline. This is followed by a brief history of all the notable past pulsar surveys which provides context to the current generation HTRU-South low latitude (LOWLAT) pulsar survey which is the primary dataset analysed during this PhD thesis. Past searches for Pulsar - Stellar Mass Black Hole (PSR‐BH) binaries in LOWLAT assume that the pulsar has a constant acceleration during the course of an observation. However, this assumption breaks down when the observation samples a large fraction of the orbit. This limits the length of search observations for finding compact binaries, and hence their sensitivity. <br /> In Chapter 3, I expand on my comprehensive search for PSR‐BH binaries in circular orbits in LOWLAT using the template-bank algorithm and use it to search for recycled and unrecycled PSR-BH binaries in compact orbits. This is currently the most sensitive search for PSR‐BH binaries done in Galactic-plane observations in the southern hemisphere. I demonstrate the extra sensitivity factor of 2-2.5 gained from our search compared to previous searches in the same data for PSR‐BH binaries with orbital periods in the 6-12 hours range. I also give details about a new GPU pipeline that was developed during this PhD which accelerated our search analysis. Additionally, I give details about our 20 new pulsar discoveries including a new millisecond pulsar J1743-24 which is a rare intermediate spin-period pulsar in a 70.7 day orbit around a light companion star. I also present updated timing solutions of PSR J1753−2819 ‐ a pulsar similar to PSR J1743−24 but in a much shorter orbit of 9.3 hours and it's likely that the formation of both these binary pulsars cannot be explained by standard binary stellar evolutionary models. I conclude this chapter by using our non-detections of PSR-BH binaries to place limits on short orbital period PSR-BH binaries near the Galactic-Plane (|b| < 3.5&#176;). Our results indicate that the existence of nearby (d &#x2264; 1 kpc) PSR-BH binaries with circular orbits and orbital period range of 4-24 hours is highly unlikely. The possibility of PSR-BH binaries having significantly eccentric orbits or circular orbits shorter than 4 hours cannot currently be ruled out due to them being outside our search range. <br /> In Chapter 4, I describe a novel Machine-learning (ML) pulsar candidate classifier using Semi-Supervised Generative Adversarial Networks (SGAN) which achieved better classification performance than the standard supervised algorithms commonly used in the literature using majority unlabelled datasets. This is the first implementation of a Semi-Supervised ML classifier for pulsar candidate classification in literature. The SGAN pipeline achieved an accuracy and mean F-Score of 94.9 &#37; trained on only 100 labelled candidates and 5000 unlabelled candidates compared to our standard supervised baseline which scored at 81.1 &#37; and 82.7 &#37; respectively. This pipeline played a pivotal role in reducing the number of pulsar candidates that needed to be inspected by eye which aided us in finding our earlier mentioned pulsar discoveries. We also describe in detail why this would be a promising solution in the early stages of future pulsar surveys when limited labelled data is available. <br /> In Chapter 5, I push the envelope of sensitivity that can be regained from a binary search pipeline further by describing the first fully coherent radio pulsar search pipeline that can search across all Five Keplerian Parameters. I compare the performance of this pipeline to standard techniques used in literature like acceleration and jerk searches and describe the feasibility of this approach for targeted and blind pulsar surveys in the near future. I demonstrate that a five-parameter search for pulsars with spin-period P_spin &#x2265; 10 ms orbiting intermediate-mass black holes (IMBH: M &#x223C; 10²&#8208;10⁴ M_&#x02299;) in Globular clusters with orbital periods in the P_orb = 5&#8208;10 T_obs regime is feasible for observations shorter than 2 hours with an eccentricity limit of 0.1 in the template bank. This is a region in the binary orbital phase-space that cannot be fully explored by other search techniques. Finally, in Chapter 6, I summarise the relevant findings of this thesis and describe some possible future research paths that can be undertaken based on my results.