The transient sky observed with the Parkes radio

Emily Brook Petroff

Presented in fulfillment of the requirements of the degree of Doctor of Philosophy

February, 2016

Faculty of Science, Engineering, and Technology Swinburne University

i

Toute la sagesse humaine sera dans ces deux mots: attendre et espérer.

All of human wisdom is summed up in these two words: wait and hope. —The Count of Monte Cristo, Aléxandre Dumas ii Abstract This thesis focuses on the study of time-variable phenomena relating to and fast radio bursts (FRBs). Pulsars are rapidly rotating neutron that produce radio emission at their magnetic poles and are observed throughout the Galaxy. The source of FRBs remains a mystery – their high measures may imply an extragalactic and possibly cosmological origin; however, their progenitor sources and distances have yet to be verified. We first present the results of a 6-year study of 168 young pulsars to search for changes in the electron density along the line of sight through temporal variations in the dispersion measure. Only four pulsars exhibited detectable variations over the period of the study; it is argued that these variations are due to the movement of ionized material local to the pulsar. Our upper limits on DM variations in the other pulsars are consistent with the scattering predicted by current models of turbulence in the free electron density along these lines of sight through the (ISM). We also present new results of a search for single pulses from Fast Radio Bursts (FRBs), including a full analysis of the data from the High Time Resolution Universe (HTRU) sur- vey at intermediate and high Galactic latitudes. No new FRBs were found in the intermedi- ate latitude survey and five new bursts were found at high latitudes. The unexpected dearth at intermediate latitudes is found to be inconsistent (with 99% confidence) with an isotropic distribution using previously published rates. From the 9 FRBs at high latitude an all-sky rate can be derived with the largest sample of FRBs to date of R (F 0.6Jyms)> FRB +4.2 3 1 1 +3.2 3 5.7 2.7 10 (95%) FRBs sky day ,orRFRB(F 2Jyms)=2.5 1.6 10 (95%) ⇥ ⇥ 1 1 FRBs sky day . Although lower than previously published estimates, these rates are still inconsistent with results at intermediate latitudes. The FRBs from the HTRU survey were re-observed in a detailed follow-up campaign to place limits on possible repeating sources. No repetition was detected from any of the FRBs; however, a new burst was discovered in real-time during these observations: FRB 140514. information for the burst was preserved by the real-time search pipeline at Parkes, and the burst was found to be 21 7% circularly polarized. Multi- ± follow-up was also performed and no variability was detected in the field related to the burst. These observations placed the first limits on an afterglow. The full search of the HTRU intermediate and high latitude survey also resulted in the discovery of 50 “perytons”, seemingly dispersed terrestrial signals of unknown origin. We ⇠ present conclusive evidence that perytons are caused by on-site microwave ovens producing sparks in a non-linear shut down phase. Based on the properties of the perytons and FRBs iii we conclude that the observed FRBs cannot be produced by the microwave ovens on site and an astrophysical origin remains highly favored. iv v Acknowledgments

It’s difficult to find the words to fully express my gratitude to all the people who deserve it. There were a lot of people who helped make this happen from when I decided I wanted a PhD when I was twelve. Early thanks are due to Bill Lamb, Rosa Hemphill, Catherine Garland, and all the people at Oregon Episcopal School who let me do my own thing as well as to Cindy Blaha and everyone in the Carleton College physics department. I owe my love of pulsars and my introduction to to the one and only Joel Weisberg. Thank you, Joel, for working patiently with me for over three years at Carleton and for introducing me to Australia and to Parkes. I would never be here writing this if it weren’t for you. Before going any further enormous and heartfelt thanks go to my amazing supervisors: Willem van Straten, Simon Johnston, and Matthew Bailes. Willem – thank you for pro- viding support, encouragement, counsel, and wisdom over the last three years. Your good judgment has helped me navigate the world of research and made me a better person. Simon – thank you for giving me one of the greatest gifts a supervisor ever could, the space and freedom to speak my mind. I’ve been learning from you since my first trip to the ATNF back in 2009 and it has been an honour. Matthew – thank you for showing me what quality research looks like and giving me something to aspire to. Wisdom doesn’t just come from supervisors. I’m grateful to the friends who I have had along the way who have taught me in one way or another. Jonathan, thank you for the book recommendations, the debates, the coding lessons, the chocolate, and saving me from my terrible cooking. Dave, thank you for your endless positivity, your encyclopedic movie knowledge, and for basically carrying our team at trivia (bonus points for your specific Trek theme song knowledge). Jonathan and Dave, thanks for making Scotch Saturday happen. Tyler, thank you for being a friend through thick and thin, laughing with me, and showing me how a good thesis is done. Rebecca, thank you for being you, for having an infectious amount of happiness about the world around you and for picking me up whenever I got down. Evan, thank you for your counsel and for being willing to listen when asked and give advice when greatly needed. Ewan, thank you for always encouraging me to do my best, push outside my comfort zone, not be afraid, and for reminding me that things will turn out OK. Enormous thanks to Andrew for answering my questions – from the embarrassingly simple to the ridiculously technical. And a great big thank you to the rest of the pulsar group for everything along the way: Paul, Stefan, Pablo, Fabian, Manisha, Shivani, Vivek, vi

Vikram, Damien and Ian. I would also like to acknowledge all the mentors who have given me advice, support, direction, and encouragement in one way or another during these past three years (in no particular order): Katie Mack, Bryan Gaensler, Brian Schmidt, Elaine Sadler, Tamara Davis, Naomi McClure-Griffiths, Michael Childress, Fang Yuan, Chris Blake, Alan Duffy, Virginia Kilborn, Karl Glazebrook, Jeff Cooke, Michael Murphy, Sarah Maddison, Tyler Pritchard, George Hobbs, Antonia Rowlinson, Dick Manchester, Keith Bannister, Tara Murphy, Kate Gunn, Sue Lester, Elizabeth Thackray, John Sarkissian, John Reynolds, Phil Edwards, Brett Presig, Mal Smith, JP Macquart, Cath Trott, Ron Ekers, Ben Stap- pers, Andrea Possenti, Jasson Hessels, Aris Karastergiou, Sarah Burke-Spolaor, David Champion, and Michael Kramer. Swinburne, CSIRO, CAASTRO, and Parkes have been supportive and fantastic envi- ronments in which to work and I wish I could thank the entire community individually. You’ve all encouraged, motivated, inspired, and impressed me during my time here. Believe it or not I would also like to thank those out there who caused me pain, both mental and physical – the ones who said I’d never make it. Proving you wrong has been one of the most rewarding experiences of my life. Natasha, thank you for being my best friend for 10 years and reminding me every time I come back why Portland is home. Norma, thank you for being my biffle, my other Wonder Twin, and my robot unicorn. Special thanks to Hallie, Becca, Bizou, Tug, and Sanny, even though you’re not people and you’ll never read this. Pakey, thank you for listening; I aspire to be as good of a person as you. Erica, just thank you. Thank you for being my sister. I’d especially like to thank my mom and dad for supporting me from the very beginning, giving me all a daughter could want, and always being on the other side of the door, or the phone, or the Skype call when I wanted to quit and said I couldn’t do it, saying “Yes you can”. I love you both more than anything. vii viii Declaration

The work presented in this thesis has been carried out in the Centre for Astrophysics & Supercomputing at Swinburne University of Technology (Hawthorn, VIC), the Australia Telescope National Facility/CSIRO Astronomy and Space Sciences (Marsfield, NSW) and the CSIRO Parkes (Parkes, NSW) between 2012 and 2015. This thesis contains no material that has been accepted for the award of any other degree or diploma. To the best of my knowledge, this thesis contains no material previously published or written by another author, except where due reference is made in the text of the thesis. The content of the chapters listed below has appeared in refereed journals. Minor alterations have been made to the published papers in order to maintain argument continuity and consistency of spelling and style.

Chapter 3 has been published in Monthly Notices of the Royal Astronomical Society, • 435, 1610, 2013, as “Dispersion measure variations in a sample of 168 pulsars”, au- thored by E. Petroff, M. J. Keith, S. Johnston, W. van Straten, and R. M. Shannon.

Chapter 4 has been published in Astrophysical Journal Letters, 789,L26,2014,as • “An absence of fast radio bursts at intermediate galactic latitudes”, authored by E. Petroff, W. van Straten, S. Johnston, M. Bailes, E. D. Barr, S. D. Bates, N. D. R. Bhat, M. Burgay, S. Burke-Spolaor, D. Champion, P. Coster, C. Flynn, E. F. Keane, M. J. Keith, M. Kramer, L. Levin, C. Ng, A. Possenti, B. W. Stappers, C. Tiburzi, and D. Thornton.

Chapter 5 has been published in Monthly Notices of the Royal Astronomical Society, • 451, 3933, 2015, as “Identifying the source of perytons at the Parkes radio telescope”, authored by E. Petroff, E. F. Keane, E. D. Barr, J. E. Reynolds, J. Sarkissian, P. G. Edwards, J. Stevens, C. Brem, A. Jameson, S. Burke-Spolaor, S. Johnston, N. D. R. Bhat, P. Chandra, S. Kudale, S. Bhandari.

Chapter 6 (excluding Section 6.2) has been published in Monthly Notices of the • Royal Astronomical Society, 454. 457, 2015, as “A survey of FRB fields: Limits on repeatability”, authored by E. Petroff, S. Johnston, E. F. Keane, W. van Straten, M. Bailes, E. D. Barr, B. R. Barsdell, S. Burke-Spolaor, M. Caleb, D. J. Champion, C. Flynn, A. Jameson, M. Kramer, C. Ng, A. Possenti, B. W. Stappers.

Chapter 7 has been published in Monthly Notices of the Royal Astronomical Society, • 477, 246, 2015, as “A real-time : polarization and multi-wavelength ix follow-up”, authored by E. Petroff, M. Bailes, E. D. Barr, B. R. Barsdell, N. D. R. Bhat, F. Bian, S. Burke-Spolaor, M. Caleb, D. Champion, P. Chandra, G. Da Costa, C. Delvaux, C. Flynn, N. Gehrels, J. Greiner, A. Jameson, S. Johnston, M. M. Kasliwal, E. F. Keane, S. Keller, J. Kocz, M. Kramer, G. Leloudas, D. Malesani, J. S. Mulchaey, C. Ng, E. O. Ofek, D. A. Perley, A. Possenti, B. P. Schmidt, Yue Shen, B. W. Stappers, P. Tisserand, W. van Straten, C. Wolf.

Emily Petroff Melbourne, Victoria, Australia 2015 x xi

For my family

Contents

Abstract i

Acknowledgements iv

Declaration vii

List of Figures xvi

List of Tables xviii

1Introduction 1 1.1 Transient radio astronomy ...... 1 1.2 Pulsars ...... 5 1.2.1 Dispersion ...... 8 1.2.2 Scattering ...... 10 1.2.3 Interstellar scintillation ...... 12 1.2.4 Faraday rotation ...... 13 1.2.5 Pulsar searches and rotating radio transients ...... 15 1.3 Fast Radio Bursts ...... 16 1.3.1 Distances, energies, and brightness ...... 18 1.3.2 FRB rates and progenitor theories ...... 20 1.3.3 Pulse propagation ...... 23 1.3.4 Perytons ...... 24 1.4 Thesis outline ...... 25

2TechnologicalIntroduction 27 2.1 Data acquisition ...... 27 2.1.1 The Parkes radio telescope ...... 29 2.2 Detecting single pulses ...... 30 2.2.1 The heimdall pipeline ...... 33 2.2.2 Real-time searches ...... 34 2.3 The High Time Resolution Universe survey ...... 38

3Dispersionmeasurevariationsofyoungpulsars 41 3.1 Introduction ...... 41 3.2 Observations ...... 44

xiii xiv Contents

3.3 Analysis ...... 44 3.4 Results - Detections ...... 46 3.4.1 PSR J0835–4510 ...... 47 3.4.2 PSR J0908–4913 ...... 48 3.4.3 PSR J1824–1945 ...... 50 3.4.4 PSR J1833–0827 ...... 51 3.5 Results - Upper limits ...... 53 3.6 Conclusions ...... 57

4 An Absence of Fast Radio Bursts at Intermediate Latitudes 59 4.1 Introduction ...... 59 4.2 Analysis and Results ...... 60 4.3 Discussion ...... 62 4.3.1 Dispersion in the ISM ...... 62 4.3.2 Scattering in the ISM ...... 63 4.3.3 Sky Temperature ...... 64 4.3.4 Scintillation ...... 64 4.3.5 Sensitivity Map ...... 65 4.3.6 Summary ...... 66 4.4 Conclusions ...... 67

5Identifyingthesourceofperytons 69 5.1 Introduction ...... 69 5.2 Observations ...... 70 5.3 Results ...... 72 5.3.1 Three perytons ...... 72 5.3.2 Prevalence of 2.3 2.5 GHz signals at Parkes ...... 74 5.3.3 Archival perytons ...... 74 5.3.4 Generating perytons ...... 77 5.3.5 The Cluster of 1998 June 23 ...... 79 5.4 Discussion ...... 80 5.5 Relevance to FRBs ...... 81 5.5.1 Differences in observed properties ...... 81 5.5.2 What is FRB 010724? ...... 82 5.5.3 Deciphering new transient events ...... 82 5.6 Conclusions ...... 84 Contents xv

6 Discovery and follow-up of FRBs at high latitudes 85 6.1 Introduction ...... 85 6.2 Search of the HTRU high latitude survey ...... 87 6.2.1 Burst properties ...... 88 6.2.2 Updated FRB rates and Latitude Distribution ...... 91 6.3 Parkes follow-up ...... 92 6.4 Data processing ...... 93 6.5 Follow-up Results ...... 95 6.6 Discussion ...... 96 6.6.1 Total time ...... 96 6.6.2 Multi-day observations of FRB 090625 ...... 96 6.6.3 Follow-up of FRB 140514 ...... 97 6.7 Conclusions ...... 99

7Areal-timefastradioburst 101 7.1 Introduction ...... 101 7.2 Real-Time Transient Pipeline ...... 102 7.3 Parkes real-time detection of FRB 140514 ...... 103 7.4 FRB Follow-up at Other ...... 108 7.4.1 Parkes Radio Telescope ...... 108 7.4.2 Australia Telescope Compact Array ...... 108 7.4.3 Giant Metrewave Radio Telescope ...... 109 7.4.4 Swift X-Ray Telescope ...... 109 7.4.5 Gamma-Ray Burst Optical/Near- Detector ...... 110 7.4.6 Swope Telescope ...... 110 7.4.7 Palomar Transient Factory ...... 110 7.4.8 Magellan Telescope ...... 111 7.4.9 SkyMapper ...... 111 7.4.10 Effelsberg Radio Telescope ...... 111 7.4.11 Keck Spectroscopy ...... 112 7.4.12 Nordic Optical Telescope Spectroscopy ...... 112 7.5 Interpretation and Discussion ...... 112 7.5.1 Polarization ...... 112 7.5.2 Possible connection with FRB 110220 ...... 116 7.5.3 Limits on a varying counterpart ...... 117 7.5.4 Host galaxies ...... 118 xvi Contents

7.6 Conclusions ...... 118

8 Conclusions 121 8.1 Major Findings of the Thesis ...... 121 8.2 Future Directions ...... 123

Bibliography 138

Appendices

AGlossary 139

B Derivation of Bayesian probabilities 141 B.1 Derivation ...... 141 B.2 Justification ...... 142 List of Figures

1.1 The radio transient parameter space ...... 6 1.2 The Galactic distribution of radio pulsars ...... 8 1.3 Spectrum of a radio pulse from a pulsar ...... 9 1.4 Kolmogorov power law of interestellar turbulence ...... 11 1.5 A cartoon of the scatter broadening of a pulse by a thin screen of . . 12 1.6 Spectrum and pulse profile of FRB 010724 (the Lorimer Burst) ...... 17 1.7 Spectrum of a peryton signal detected at the Parkes radio telescope . . . . . 25

2.1 The Parkes multibeam receiver ...... 29 2.2 Recovered signal-to-noise ratio of a pulse using different single pulse search codes...... 32 2.3 Overview plot of heimdall outputs for a single pointing ...... 35 2.4 The heimdall processing pipeline for filterbank data ...... 36

3.1 DM measurements of PSR J0835 4510 over 2000 days ...... 47 3.2 DM measurements of PSR J0835 4510 since 1969 ...... 49 3.3 DM measurements of PSR J0908 4913 ...... 50 3.4 DM measurements of PSR J1824 1945 ...... 51 3.5 DM measurements of PSR J1833 0827 ...... 52 3.6 Upper limits on dDM/dt for pulsars with no detected DM variations . . . . 53

4.1 Galactic effects on a simulated FRB pulse with strong scattering ...... 66 4.2 Galactic effects on a simulated FRB pulse with no scattering ...... 67

5.1 The time– structure of the three January perytons ...... 73 5.2 RFI monitor spectra for the three January perytons ...... 75 5.3 Data from the Parkes and ATCA RFI monitors at the times of the perytons 76 5.4 Histogram of narrow RFI spikes detected at Parkes ...... 76 5.5 A bright peryton detected during tests of the Parkes microwave ovens . . . 78 5.6 Azimuth and elevation positions with direct line of sight between Parkes receiver and the Woolshed microwave oven ...... 79 5.7 FRB and peryton distributions in time of day and DM ...... 83

6.1 Pulse profiles of the 5 new FRBs from the high latitude survey ...... 90 6.2 Probability of detection for repeating progenitors for FRB 090625 ...... 98

xvii xviii List of Figures

7.1 Spectrum and pulse profile of FRB 140514 ...... 106 7.2 Full polarization profile of FRB 140514 ...... 107 7.3 FRB 140514 multi-wavelength couterpart magnitude limits from follow-up observations ...... 119

8.1 All-sky distribution of the known FRBs ...... 124 List of Tables

2.1 Properties of central, inner, and outer rings of the 21-cm multibeam receiver 30 2.2 Survey regions of the low, intermediate, and high latitude components of HTRU South ...... 39

3.1 Pulsars with DM variations over 6 years above 3 levels ...... 46 3.2 Measurements of DM and dDM/dt for all pulsars in the Fermi sample . . . 55

5.1 Properties of the perytons from 2015 January ...... 72

6.1 Repeating progenitor models for FRBs and their timescales ...... 87 6.2 Observed properties of the 9 FRBs from the HTRU high latitude survey . . 89 6.3 Derived cosmological properties of the 9 FRBs from the HTRU high latitude survey ...... 89 6.4 Total hours of follow-up observations for 8 FRBs from HTRU ...... 94 6.5 Summary of targeted observations for FRB 090625 ...... 97

7.1 Observed properties of FRB 140514...... 105 7.2 Derived cosmological properties of FRB 140514 ...... 105 7.3 Follow-up observations conducted at 12 telescopes for FRB 140514 . . . . . 113

xix

1 Introduction

Within the past decade it has become possible to observe time-variable radio sources with unprecedented sensitivity at high time and frequency resolution thanks, in part, to advances in radio astronomical instrumentation and signal processing. The experimental methods of time-variable radio astronomy, such as observations of pulsars and radio transients, provide a window into events and processes not observable through other techniques. In this chapter we describe different types of transient radio sources, some of which will be discussed in-depth in this work, and give some examples of their unique uses as probes of the ionized baryons in the near-vacuum of space. The first section of this chapter details the history of transient radio astronomy and different types of radio transients. The second section discusses the discovery of pulsars and their use as probes of the interstellar medium. The third section introduces the recently discovered fast radio bursts and their exciting potential as probes of the intergalactic medium over cosmological distances.

1.1 Transient radio astronomy

An astronomical transient is an event of short duration relative to the timescales typical of events in the Universe. Different definitions of transients exist but for the purposes of this thesis astronomical transients are defined in the following three ways: sources that appear and disappear on timescales observable by humans that have no known counterpart; sources that appear and disappear on timescales observable by humans that are associated with known one-off events; or sources that emit short-duration emission which may be highly sporadic. Transient phenomena in the skies have been recorded for centuries; early examples in- clude the observation of a ‘’ in the night sky in the year 1054 A.D. The nearby , which created the Crab , is the birthplace of one of the youngest known

1 2 Chapter 1. Introduction neutron stars (Mayall & Oort, 1942). More recently, telescopes probing the entire electro- magnetic spectrum, from gamma-rays and X-rays down to radio , have detected fainter and more distant variable and transient phenomena. Beginning in 1933 with the study of the radio background at 20 MHz by Karl (Jansky, 1933a,b), observations of the radio sky offered a surprising new window into the physics of our own Galaxy and beyond. Radio transients were first observed in the form of non-thermal Solar emission at 10 40 MHz correlating with periods of high sunspot activity (Appleton, 1945). As radio instrumentation improved and larger radio observatories were constructed, sensitivity to the variability of astrophysical sources increased. Refraction of radio frequen- cies in the ’s ionosphere was first observed in the late 1940’s as changes in brightness of stable radio sources over time (Hey et al., 1946). However, radio stars observed with the most advanced radio telescopes of the time were seen to fluctuate in brightness on levels that could not be explained with ionospheric effects. It was realized that the cause of these fluctuations was scintillation, or twinkling, of radio light not within the ionosphere but rather in the ionized material in the Solar System produced by the Sun (Hewish, 1955). The discovery of interplanetary scintillation opened the doors to the sub-discipline of time-domain radio astronomy and the first detections of radio pulses. In 1967 a source was discovered that emitted highly periodic, bright pulses, each lasting only 30 ms the discovery of the first pulsar (Hewish et al., 1968). Pulsars are rapidly rotating neutron stars and sources of strong radio emission when fortuitously oriented towards the observer. Pulsar astronomy has become a rich and vibrant field and pulsars themselves can be used as excellent probes of gravitation (Kramer et al., 2006; Hulse & Taylor, 1975), condensed matter physics (Antoniadis et al., 2013), and the interstellar medium of the Galaxy (Keith et al. 2013, see also Section 1.2). In the 50 years since the discovery of pulsars, several classes of radio transients have been discovered and studied. Some of these other radio sources and their defining properties and timescales include the following:

i. Solar bursts. The Sun is a strong source of transient radio emission emitting several types of bursts labeled Types I V on a variety of timescales. Type I solar bursts last for only 1 second and are caused by plasma radiation during periods of sunspot ⇠ activity; Type III bursts last for several seconds as sub-relativistic electrons accelerate away from the solar surface; and Type II flares – emission that originates in outward propagating shock waves – can continue for up to 30 minutes. Storms of burst activity have also been observed on the Sun that can last for days to weeks (Dulk, 1985). Most 1.1. Transient radio astronomy 3

types of solar bursts are caused by instabilities and variations in the solar plasma (Melrose, 1980) and are typically detected below 200 MHz.

ii. Brown dwarfs. Periodic bright transient emission has been observed for a small number of brown dwarfs and low mass stars (Hallinan et al., 2007, 2008). The emission, caused by the electron cyclotron maser instability, originates at the magnetic poles of the star and is up to 100% circularly polarized. For the known pulsing dwarf stars, pulses last for several minutes and repeat on timescales of 2 3 hours, the rotation period of the star (Hallinan et al., 2008). All radio detections of these pulses have been made between 3 5 GHz (Osten et al., 2009; Route & Wolszczan, 2013). iii. Flare stars. Bursts of incoherent emission have been recorded from nearby M dwarfs for over 50 years (Lovell, 1963); however, more recent observations have revealed radio flares of highly circularly polarized coherent emission at frequencies of 1 GHz (Osten, ⇠ 2008). The coherent flares take place on timescales of a few minutes but are also seen to exhibit temporal variations of the order of a millisecond believed to be intrinsic to the source of the flare on the stellar surface (Osten & Bastian, 2008). The emission mechanism of these coherent flares remains an open question, however the electron cyclotron maser instability observed from brown dwarfs is preferred (Osten, 2008). iv. Radio afterglows from supernovae and gamma-ray bursts. Long-lived radio emission has been detected from supernovae (SNe, Weiler et al., 1982; Sramek et al., 1984) and gamma-ray bursts (GRBs, Kulkarni et al., 1998) in the form of a radio afterglow following the peaked emission in the optical (for SNe, Weiler et al., 1982) and the gamma-ray and X-ray (for GRBs, Galama et al., 1998). Radio emission has been observed in connection with Type Ib/c, and Type II (core collapse) events peaking at various timescales with respect to the optical transient. Type Ib/c radio emission is seen to peak nearly simultaneously with the optical peak, whereas Type II emission peaks months after the optical maximum (e.g. Weiler et al., 2002). The source of the radio emission from supernovae is thought to be the collision of the supernova into the ionized circumstellar medium producing incoherent visible across a range of radio frequencies (Weiler et al., 1982).

Whereas optical brightening of supernovae can be observed for many days before and after the peak, gamma-ray bursts are of much shorter duration; e.g. as a canonical long gamma-ray burst, caused by the rapid collapse of a high-mass star into a (Woosley & Bloom, 2006), is visible only on second timescales at gamma-ray frequen- cies (Weiler et al., 2002). Only a subset of GRBs have observed radio counterparts; 4 Chapter 1. Introduction

based on available data, Chandra & Frail (2012) estimate approximately 31% of GRBs have detectable radio afterglows. Afterglows of GRBs are observed in X-ray, optical, and radio as the columnized jet from the GRB moves through the interstellar medium near the source and produces incoherent synchrotron radiation. As the jet travels further away from the source the energy of the emission decreases resulting in an af- terglow spectrum that peaks first in X-rays then optical and finally radio frequencies (van Paradijs et al., 2000). Radio afterglows are seen anywhere from 0 to 80 days ⇠ post-burst (Chandra & Frail, 2012).

v. Unknown origin. A subset of radio transients have been detected only once, and no rigorous physical model exists for their progenitors. Of this small subset, the “Wow signal” is likely the most widely known. The signal was detected as a 72-second peak of radio emission at 1.4 GHz during a drift scan of the sky for signals as part of the Search for Extraterrestrial Intelligence (SETI) at Ohio State University (Kraus, 1979). A terrestrial origin was determined to be unlikely based on the detection in only one of two beams on the sky. A proposed explanation of the signal was artificial boosting of the source brightness due to interstellar scintillation, but a VLA search of the detection region revealed no sources at that frequency in a search with 100 times the sensitivity of the original detection receiver making scintillation an unlikely cause (Gray & Marvel, 2001). Ultimately no conclusive evidence has been put forward for the signal’s astrophysical or terrestrial origin and it remains an anomaly.

For each of these sources we can calculate a brightness temperature, or the temperature required to produce the observed radio intensity from a black body radiating in the classical Raleigh-Jeans part of the spectrum (Lorimer & Kramer, 2004, 3.4). For a source § at a distance D with a width (or duration) W and peak flux S emitted at a frequency ⌫ the brightness temperature will be

2 2 23 S.D ⌫.W TB & 10 K. (1.1) Jy.kpc2 GHz.ms ✓ ◆✓ ◆ It has been shown by Kellermann et al. (1969) that brightness above T B 1012 K cannot be produced by incoherent emission processes such as synchrotron radiation and must instead be produced through coherent emission processes, although coherent synchrotron emission may still be possible at these high brightness temperatures (Caroff & Scargle, 1970). Typical brightness temperatures for classes of short-duration radio transients are of the order of 108 1010 K for solar bursts, 7 109 K for pulses from brown dwarfs and 1014 ⇥ 1.2. Pulsars 5

K for the coherent radio flares from flare stars (Dulk, 1985; Hallinan et al., 2008; Osten, 2008). Single pulses from pulsars stand out from the transient classes listed above in their extremely short duration (second to nanosecond timescales) and relatively high brightness temperatures. The observed single pulses from pulsars span several orders of magnitudes ranging from T 1016 K for the least energetic observable pulses to 1035 1037 Kfor B ⇠ the nanosecond bursts of coherent emission from the (Hankins et al., 2003; Lorimer & Kramer, 2004, 3.4). § The radio transient parameter space is presented in Figure 1.1, from Macquart et al. (2015). The single pulses from pulsars occupy an extreme area of this parameter space high brightness temperatures and short transient timescales; however the pulses from the emerging population of sources called ‘Fast Radio Bursts’ (FRBs) are more extreme still. Since these pulses are thought to originate from outside the Galaxy (see Section 1.3) the inferred brightness temperatures and peak luminosities of the progenitors are required to be 10 orders of magnitude greater than the pulses from normal pulsars. These two classes of extreme phenomena, pulsars and fast radio bursts, are discussed in the following sections.

1.2 Pulsars

Pulsars were discovered through their periodic single pulses in 1967 (Hewish et al., 1968) and since then a population of over 2000 pulsars has been discovered1 (Manchester et al., 2005). Soon after their discovery the link was made between the observed periodic pulses and the theoretical objects called neutron stars (NSs) thought to be formed in the gravita- tional collapse of an intermediate mass star during a supernova (Baade & Zwicky, 1934). It is now widely accepted that the pulsar emission is from rapidly rotating, highly mag- netized neutron stars and is generated in the open magnetic field line region at the star’s magnetic poles. As the star spins, the highly focused beams of radiation from the poles sweep across the sky like a lighthouse and if one of these beams intersects the line of sight with Earth the star is seen as a pulsar. The pulsar emission mechanism is poorly understood (Melrose, 1992). The radio emis- sion is thought to be generated by charged particles accelerated along open magnetic field lines at the polar caps; however, the actual process by which this occurs is unknown. Pul- sar emission is also relatively broadband – most pulsars are observable from GHz down to 100s and even 10s of MHz in frequency with a spectral index of between 1.4 and 2 (i.e. 1For a complete list of known pulsars see the pulsar catalogue http://www.atnf.csiro.au/people/pulsar/psrcat/ 6 Chapter 1. Introduction

Figure 1.1 The parameter space of radio transients reproduced from Macquart et al. (2015) illustrating the wide range of transient radio phenomena. Several classes of transients are labeled including solar bursts, flare stars, single pulses from pulsars, and fast radio bursts. Lines of constant brightness temperature are shown diagonally. Sources in the blue triangle produce radio emission through incoherent emission processes, at T 1012 K and coherent B  emission processes occur above this boundary. The 1 kpc and 1 Gpc sensitivity curves are shown for Parkes (black), and two components of the (SKA): SKA1-LOW (pink), and SKA1-MID (grey). 1.2. Pulsars 7 pulsars are brighter at lower frequencies), although some exhibit spectral turnover and have lower flux densities than expected at lower frequencies (Karastergiou et al., 2014). More recently, gamma-ray emission has been detected from both young and old pulsars (Abdo et al., 2009). The gamma-ray emission peak is typically offset from the peak radio pulse suggesting that emission at high energies takes place in the pulsar magnetosphere, further from the surface than the narrowly-beamed radio emission (Weltevrede et al., 2010a).

The pulsar population can be divided into two general categories – the normal (or canonical) pulsars and the millisecond pulsars (MSPs). Canonical pulsars represent the majority of the pulsar population and have spin periods P 500 ms and period deriva- ⇠ tives P˙ 10 15 ss 1; in contrast millisecond pulsars have typical values of P 5 ms ⇠ ⇠ and P˙ 10 20 ss 1 (Lorimer & Kramer, 2004, 1.3). The evolutionary histories of these ⇠ § two classes are also very different. MSPs are formed when a canonical pulsar in a binary system accretes mass from a companion into a disk around the ; the accre- tion from this disk onto the NS surface is responsible for ‘spinning up’ the pulsar (Alpar et al., 1982). Approximately 80% of MSPs still reside within binary systems (Grégoire & Knödlseder, 2013). As pulsars are descendants of main sequence stars, their population is highly concentrated in the plane of the Galaxy (see Figure 1.2); however, many pulsars are moving at high velocities out of the plane, possibly due to natal kicks (van den Heuvel & van Paradijs, 1997). Consequently the older age, and by extension longer travel time, of the MSP population results in a more uniform distribution on the sky than the canonical pulsar population.

The incredibly stable periods of pulsars, especially of the MSPs, make pulsars excel- lent probes of gravity and extreme physics (Matsakis et al., 1997; Lattimer & Prakash, 2007; Hobbs et al., 2012). Through precision timing of pulsars the dynamics of the bi- nary systems, the parameters of their orbits, and the masses of their companions can be measured extremely accurately. For example, the energy lost in the decaying orbit of the NS-NS system observed via the pulsar PSR B1913+16 agrees with the expected emission of gravitational waves predicted by general relativity to within 0.3%: the first detection, although indirect, of gravitational radiation (Taylor et al., 1979; Weisberg et al., 2010). Similarly, precise determination of neutron star masses in the double pulsar system PSRs J0737 3039A&B allowed the most stringent test of general relativity in the strong-field regime (to 99.5% precision; Kramer et al., 2006) by comparing the observed properties of the relativistic orbit with the parameters predicted by general relativity.

Besides being excellent probes of gravity, pulsars are powerful tools for studying the interstellar medium (ISM) of our Galaxy. As radio pulses travel through the Galaxy they 8 Chapter 1. Introduction

Figure 1.2 The distribution of known pulsars in Galactic longitude and latitude in an Aitoff projection. The figure includes normal pulsars (black circles), MSPs (blue triangles), and RRATs (red stars). experience propagation effects caused by the ionized material in the ISM. The four main effects on a radio pulse traveling through the ISM are dispersion, scintillation, scattering, and Faraday rotation which we will discuss in the next four subsections. Dispersive effects on plane waves are most pronounced at low frequencies, thus astrophysical radio pulses are ideal for studies of the magnetoionic medium along the line of sight to the source.

1.2.1 Dispersion

Dispersion is observed as the frequency-dependent group velocity of radio waves. In the interstellar medium, radio signals experience a delay that is proportional to the integrated electron column density along the line of sight, such that the time delay between two frequencies ⌫high and ⌫low will be

d 2 2 ⌫low ⌫high 0 ned` t =4.148808 3 ms (1.2) GHz GHz ⇥ pc cm ⇣ ⌘ ⇣ ⌘ R d where the integral 0 ned` is the integrated electron density ne along the line of sight to a source at distance dR (Ekers & Moffet, 1968). This frequency-dependent delay can be seen in the frequency-time spectrum of a pulsar in Figure 1.3. The integral in Equation 1.2 is commonly defined as the dispersion measure (DM) which can be determined for a pulse 1.2. Pulsars 9

Figure 1.3 The frequency-time spectrum of the radio pulsar J1644 4599 over a range of 64 frequency channels between 710 and 760 MHz as a function of pulse phase. The pulse experiences a frequency dependent delay as it travels through the ionized interstellar medium causing the pulsar pulse to arrive later in lower frequency channels. The pulsar signal has been summed over several pulse periods to achieve a high signal to noise ratio. recorded over a finite bandwidth by measuring the time delay as a function of frequency. DM measurements for pulsars combined with electron density models of the Galaxy are used to estimate distances to Galactic pulsars. The most commonly used model of this type is NE2001, developed by Cordes & Lazio (2002), which models an elaborate, multi- component Galaxy accounting for spiral arms, thin and thick disks, and an outer halo. Distances estimated via DM are calibrated against parallax distances; however, these mea- surements are possible for only about 2% of the pulsar population (Brisken et al., 2002). As such this model is highly uncertain and errors of the order of a factor of two in the model-derived distances are to be expected, especially in regions such as the Galactic halo where the population of pulsars is sparse. 3 DM is measured in units of pc cm and typical values for Galactic pulsars range from 3 3 only 2 pc cm for the closest sources to >1700 pc cm for pulsars near the centre of our Galaxy (Eatough et al., 2013). For most purposes, a pulsar’s dispersion measure is treated as a constant, however some studies over multi-year timescales have detected dispersion measure variations due to changes in the amount of ionized material along the line of sight (Hamilton et al., 1985; Backer et al., 1993). Most variations reported in the literature have been attributed to supernova remnants or pulsar nebulae local to the pulsar; however, 10 Chapter 1. Introduction some studies have shown variations on timescales of a few years due to the structure of the ISM on larger scales (Keith et al., 2013). The interstellar medium is believed to contain turbulence within the larger smooth distribution of ionized material. The spectral energy density scale of interstellar turbulence is thought to obey a power law between large (1018 m) and small (106 m) spatial scales, q,suchthat

2 P (q) C q (1.3) 3N ⇡ N represents the power spectrum of the electron density P3N with a spectral index and a 2 structure coefficient CN (Armstrong et al., 1981). This spectrum is thought to be consistent with a Kolmogorov power law described by = 11/3, shown in Figure 1.4. Turbulent interstellar material is probed on various scales by different types of observable phenomena, from rotation measure variations at the largest scales, to dispersion measure variations on intermediate scales, to weak interstellar scintillation at the smallest scales (Armstrong et al., 1995). Variations in pulsar dispersion properties on day to decade timescales are able to probe various regimes of this power law. The turbulent plasma in the ISM is also the cause of the scattering and scintillation phenomena described in the following sections.

1.2.2 Scattering

Multi-path propagation through a turbulent ISM can produce an exponential scattering tail on the trailing edge of the pulsar pulse (Armstrong et al., 1995). The simplest model that can reproduce the observed results of scattering in the ISM is that of a single thin screen (Williamson, 1972). In this approximation, a plane wave experiences distortions due to propagation through a screen of inhomogeneous plasma. The scattering effect is maximized when the scattering screen is located halfway between the source and the observer. In this model the wave experiences a number of phase variations as it travels through the plasma; these arise from the deflection of light by an angle ✓0, which also produces a broadened image of the source with angular radius ✓d such that

2 e ne pd ✓d = ✓0/2= 2 (1.4) 2⇡me pa ⌫ where ne is the perturbation in electron density, a is the width of the screen, d is the distance to the source, and ⌫ is the frequency. The intensity distribution of the light coming from the pulsar through the screen has an angular dependence which also corresponds to a geometric time delay t. The intensity of the scattered pulse as a function of time is 1.2. Pulsars 11

Figure 1.4 The power law relation between spectral energy density and spatial scale for 1 turbulence in the ionized ISM in terms of spatial wavenumber, or (spatial scale) ,and spectral density. A line representing a spectral index of 4 (dot-dashed) and a Kolmogorov spectral index of 11/3 (dotted) are shown. Figure reproduced from Armstrong et al. (1995).

given by

2 t/⌧ I(t) exp( ct/✓ d) e d (1.5) / d ⌘ where ⌧d is defined as the scattering timescale. From Eq. 1.4:

2 4 2 ✓dd e ne 4 4 ⌧d = = 2 2 d⌫ d⌫ . (1.6) c 4⇡ me a /

From this equation we see that the scattering timescale is dependent on distance and heavily dependent on observing frequency. Thus scattering is stronger at larger distances and often correlates with DM. The relation in Eq. 1.6 is a theoretical approximation based upon a single thin screen. In reality the interstellar medium is made up of a large number of scattering screens; however, the observational data from pulsars agrees well with the thin screen model which, to close approximation, can reproduce the exponential scattering tails observed for many pulsars (Sutton, 1971; Williamson, 1972). The observed scattering relation for pulsars has been approximated by Bhat et al. (2004) by 12 Chapter 1. Introduction

Figure 1.5 A cartoon of the thin screen scattering model for radio pulses from a pulsar. A wave encountering a plasma screen half way between source and observer is distorted and deflected through the screen with propagation delays away from the direct line of sight. The result is a scatter-broadened image of the source which appears to have a radius ✓d. Figure reproduced from Lorimer & Kramer (2004).

log⌧ = a + b (logDM) + c (logDM)2 ↵ log⌫ (1.7) d where ⌧d is the scattering timescale in ms, ⌫ is the observing frequency in GHz, and DM is the dispersion measure. The coefficients a, b, c, as well as the scattering index ↵ were fit to the data for Galactic pulsars to obtain values of a = 6.46, b =0.154, c =1.07,and ↵ =3.86 0.16. This relation is useful when the scattering timescale cannot be directly ± measured observationally for a pulsar with a known DM; however, in observational data there are several orders of magnitude of scatter around this relation, especially at higher DMs (Bhat et al., 2004).

1.2.3 Interstellar scintillation

Multi-path propagation through turbulent regions leads to patterns of constructive and destructive interference observed as brightness variability that makes the pulsar appear to scintillate, or twinkle (Armstrong et al., 1995). Scintillation is seen as the observer travels through the interference pattern created by the thin screen. This interference pattern (shown on the right hand side of Figure 1.5) arises because the paths of light through the medium have a range of phases such that the characteristic phase difference is (Rickett, 1977)

2⇡⌫ ⌧ . (1.8) ⇠ d 1.2. Pulsars 13

This interference pattern depends on the movement through space of both the pulsar and the scattering screen, as well as the velocity of the observer. The timescale t on which the intensity fluctuates depends on all three velocity components. Interference in the plasma becomes decorrelated if the phase differences between scat- tered waves are more than approximately 1 radian; i.e. when

2⇡ ⌫⌧ 1 (1.9) d ⇠ where ⌫ is the decorrelation bandwidth, also called the ‘scintillation bandwidth’ and is the typical bandwidth of correlated intensity fluctuations for a source. From Eq. 1.9, the scintillation bandwidth scales as ⌫ 1/⌧ ⌫4. ⇠ d / Scintillation occurs over a range of scales and typically two scintillation regimes are de- fined: diffractive and refractive. Diffractive scintillation is caused by turbulence on smaller length scales (107 108 m) and results in intensity fluctuations on short timescales of the order seconds to minutes. Refractive scintillation is caused by turbulence on larger scales (1011 1012 m) and manifests itself as longterm variations from hours to days. Diffractive scintillation is primarily seen in observations of pulsars and refractive scintillation is most often observed in the brightness fluctuations of distant quasars at GHz frequencies (Kaspi & Stinebring, 1992; Burke & Graham-Smith, 2014). In pulsar observations scintillation can present itself in two forms: strong and weak scintillation. Strong scintillation occurs when both refractive and diffractive effects com- bine in the scintillation seen from the pulsar at various spatial scales. Weak scintillation occurs when variations in phase are small at the distance of the observer and diffractive and refractive effects are minimal. Pulsar observations are typically made in the strong scintillation regime unless observing sources that are very nearby or making observations at high frequencies (Rickett, 1977).

1.2.4 Faraday rotation

A radio pulse that is linearly polarized in a single plane can also be described as the sum of two oppositely handed circularly polarized waves. If the plasma through which a radio pulse travels is magnetized parallel to the direction of propagation, the two circularly polarized waves will propagate at different speeds relative to one another. This differential phase rotation, proportional to the magnetic field strength, is called Faraday rotation. The ionized plasma of the interstellar medium is permeated by the Galactic magnetic field and these phase rotations are observed in the form of Faraday rotation of linearly polarized pulses from pulsars. The differential phase rotation between left- and right- 14 Chapter 1. Introduction handed polarized waves propagating through the medium is given by

e3 d = neB d` (1.10) ⇡m2c2⌫2 k e Z0 where the integral is over the same distance as the DM from Eq. 1.2 and B is the parallel k component of the magnetic field along the line of sight. The differential phase rotation is periodic in phase on 2⇡; however, the physical manifestation of this phenomenon, the rotation of the polarization position angle (PPA) relates to the plane in which the linear polarization oscillates and is therefore periodic on ⇡ (Everett & Weisberg, 2001). Thus,

2 PPA = /2=RM (1.11) where RM is the rotation measure

e3 d RM = neB d` (1.12) 2⇡m2c4 k e Z0 2 and has units of rad m . The RM for a given observation is measured experimentally through the frequency-dependent phase rotation of the linear polarization Stokes parame- ters Q and U through a fit with respect to the phase rotation at the center frequency of the band such that ()= +RM 2. 0 ⇥ Through measurement of the RM and DM of a pulse it is possible to obtain the average magnetic field weighted by the local electron density along the line of sight (Smith, 1968; Han et al., 2006)

d 1 0 neB d` RM DM B = k =1.23µG . (1.13) h ki d rad m 2 pc cm 3 R 0 ned` ✓ ◆✓ ◆ While the combined measurementsR of RM and DM of polarized, dispersed pulses can be a powerful tool for measuring the magnetic field in the interstellar plasma there are a few cases in which this method is a poor measure of B . Firstly, when free electrons are h ki concentrated in a dense region along the line of sight, such as in an Hii region, the effect of the magnetic field in that region will dominate and the average magnetic field strength will be overestimated. Secondly, when sign reversals of the magnetic field occur along the line of sight, such as the field reversals observed between spiral arms of the Galaxy (Han et al., 2006), the average magnetic field strength will be underestimated. In these two cases, the value obtained from the calculation in Eq. 1.13 will not accurately reflect the large scale magnetic field along a line of sight. Studies of the effects of dispersion, scattering, scintillation, and Faraday rotation, com- 1.2. Pulsars 15 bined with accurate information about the location of a pulsar within the Galaxy produce a picture of the shape, size, density, and magnetic fields of the Galaxy in different regions (Rand & Kulkarni, 1989; Rand & Lyne, 1994; Han et al., 2006). In addition, variations of observed flux and dispersion over short (seconds to days) and long (months to years) timescales encode the structure of the turbulence in the ISM and the local environment of the pulsar (Keith et al., 2013). Thus the pulses from pulsars are some of the most powerful tools available for studying the makeup and distribution of the material in the Galaxy.

1.2.5 Pulsar searches and rotating radio transients

Numerous radio surveys designed to detect new pulsars have been undertaken since the very first pulsar discovery. Several bright pulsars were subsequently found through searches for individual pulses (Lyne & Rickett, 1968; Staelin & Reifenstein, 1968). However, as radio pulsar surveys began collecting more data, the primary method of pulsar searching shifted to periodicity searches using discrete Fourier transforms (Hulse & Taylor, 1974). These surveys were incredibly productive, with some effectively doubling the known population of pulsars with a single publication (Manchester et al., 1978). In just under 50 years the number of known radio pulsars has grown from 1 to over 2300 (Manchester et al., 2005) with even more new sources expected in upcoming surveys (Bates et al., 2014). As periodicity searches became the method de rigueur for finding large numbers of pulsars, single pulse searches were abandoned in the 1980s and 1990s. In 2006, however, McLaughlin & et al. (2006) published the results of a new single pulse search of the archival Parkes Multibeam Pulsar Survey (PMPS) data in which they detected eleven objects via single pulse emission. These objects, called Rotating Radio Transients (RRATs), are pul- sars that emit detectable radio pulses for only a fraction of the time, with intervals between pulses of minutes to hours. Most RRATs are difficult to detect in periodicity searches, and are best identified through single pulse processing. RRATs represent only about 3% of the known pulsar population, but an estimate of their overall numbers in the Galaxy based on their observed variability implies an enormous source population, larger than the estimated population of regularly-emitting radio pulsars (Burke-Spolaor & Bailes, 2010; Keane et al., 2011). If RRATs are a separate population they generate large discrepancies between the population size and the rate of core-collapse supernovae in the Galaxy, which are thought to be the progenitors of neutron stars (Keane & Kramer, 2008). Such a conflict is partially resolved if RRATs are part of the normal pulsar population and represent a particular phase in their evolution. The position occupied by RRATs in the pulsar population is still poorly understood 16 Chapter 1. Introduction and estimates of period, period derivative, and age have been made for only around 60 objects.2 A key objective for recent pulsar surveys has been the search for new RRATs through single pulse searches. The largest concerted effort in this area has been the High Time Resolution Universe survey (HTRU) undertaken with the Parkes radio telescope. The search techniques employed by this survey are described in Chapter 2.

1.3 Fast Radio Bursts

The interest in single pulse searches was revitalized by the discovery of RRATs and, begin- ning with their discovery in 2006, single pulse searches were carried out on both archival data and new survey observations. As a result of these searches, a single bright pulse was discovered by Lorimer et al. (2007) in archival data from 2001 (Figure 1.6). This pulse, now called the ‘Lorimer burst’ was detected with a peak flux density of 20 Jy and a DM ⇠ of 375 pc cm 3 at a high Galactic latitude (b = 41.8 ). The Galactic DM contribution 3 along this line of sight predicted by the NE2001 model is 25 pc cm ,onlyabout5% of the total DM. The excess dispersion is postulated to arise in the intergalactic medium (IGM), placing the source far outside our Galaxy. For some time the Lorimer burst was the only one of its kind. Only recently, similar bursts have been discovered in contemporary (Thornton et al., 2013; Ravi et al., 2015) and archival (Keane et al., 2012; Burke-Spolaor & Bannister, 2014; Spitler et al., 2014; Masui et al., 2015) radio pulsar surveys, revealing a population of highly dispersed pulses now known as ‘fast radio bursts’ (FRBs). The observational definition of an FRB is, at the time of writing, slightly tenuous. An FRB is generally defined as a bright (flux density, S & 0.5 Jy), narrow (width, W . 5 ms) single pulse with a DM greater than the modeled DM contribution of the Galaxy along a line of sight. For all but one of the 21 known FRBs 3 the ratio DMFRB/DMGalaxy > 2.5. The one exception to this is FRB 010621 which was discovered very close to the Galactic plane by Keane et al. (2012) with a DM of 746 1 pc ± 3 cm ; for this FRB the DM is only 40% greater than the expected contribution due to free electrons in the Galaxy. Recent investigation by Bannister & Madsen (2014) suggested an overdensity in the ionized material along this line of sight, which would make this burst Galactic, possibly due to an RRAT, a Galactic FRB progenitor, or another unknown source. Given the large uncertainties in the NE2001 model, a definition of an FRB based on a

2A full list of RRAT sources can be found at http://astro.phys.wvu.edu/rratalog/ 3The naming convention for FRBs is based on the date of detection, so an FRB detected on 2015 March 29 would be FRB 150329. 1.3. Fast Radio Bursts 17

Figure 1.6 The frequency-time spectrum of the primary discovery beam for FRB 010724 (the Lorimer Burst) discovered in 2007. The single bright pulse has a dispersion measure 3 of 375 pc cm of which only 5% is accounted by the model of free electrons in the ISM. The detection in the discovery beam (beam 6) was bright enough to saturate the automatic gain controller used at the time, causing the total intensity to drop below the baseline after the pulse. 18 Chapter 1. Introduction threshold ratio between the observed DM and that of the model is not robust, especially given that the model is poorly constrained in the halo (Cordes & Lazio, 2002) and the majority of FRBs to date have been found at high Galactic latitudes (Keane & Petroff, 2015). However, adopting an error on NE2001 of 50% to account for the unknown ⇠ parameters of the Galactic halo, all FRBs but FRB 010621 have DMs not easily accounted for purely by the cold plasma of the ISM.

1.3.1 Distances, energies, and brightness temperature

If the majority of the excess DM is due to the IGM, which is 1000 times less dense than ⇠ the ISM of the (Ioka, 2003), then the progenitors of FRBs would be located at cosmological distance. The excess DM for these objects might arise in a number of different regions along the line of sight, such that

DMFRB =DMGalaxy +DManomaly +DMIGM +DMhost +DMsource (1.14) where the subscripts indicate the DM contributions from the Galactic ISM, anomalous regions in the ISM such as local overdensities, the IGM, a host galaxy, and any contribution local to the source. If the interstellar medium in the host galaxy has properties similar to our own ISM and FRBs are not produced in overdense regions, then the host ISM would contribute little to the total DM except the fraction of the host galaxies where the host has a high inclination angle (i>85). Barring this orientation, assuming FRB host galaxies are randomly oriented with respect to the observer, the majority of the excess DM would be due to the IGM between the progenitor and the observer. No exact measurement relating DM to , z, is currently published and FRBs could be the first sources to make such a measurement possible if an independent distance measurement to the host can be made after an FRB detection (Macquart et al., 2015). However, models of the ionized IGM allow a rough estimate of FRB distances. The two models currently in use are those by Ioka (2003) and Inoue (2004), both of which propose a roughly linear relationship between DM and redshift out to z . 3 where helium reionization begins to strongly affect the electron density of the IGM (McQuinn et al., 2009). The conversion factor that will be used throughout this thesis is

(DMtot DMGalaxy) z 3 (1.15)  1200 pc cm where DMGalaxy is an estimate taken from the NE2001 model integrated out to the edge of the Galaxy. Eq. 1.15 becomes an equality only when DMhost =DMsource =0in Eq. 1.14 1.3. Fast Radio Bursts 19 and as such this relation provides only a very rough upper limit on the actual redshift of the source. Precise measurements of FRB could open the door to exciting new avenues of cosmological study only possible with these sources; these possibilities are discussed further in Chapter 8. For a radio pulse at some redshift the distance can be calculated by assuming a model for the expansion of the Universe. A flat Universe ⇤CDM is typically used (Wright, 2006) and two different distance parameters are obtained: the co-moving distance Dcomov and the luminosity distance DL. The co-moving distance takes the expansion of the Universe into account, provides a distance measure that does not change in time, and can be viewed as the physical distance to the source. The luminosity distance is essentially the distance derived from the observed flux using the inverse square law, or

2 DL = L/(4⇡S) (1.16) where L is the luminosity and S is the observed flux; the luminosity distance is the relevant quantity for calculating the energetics of an object at some redshift.

The energy of an FRB EFRB at a redshift z is calculated by

2 obs DL ⌫ 26 1 E = F 10 (1 + z) Joules (1.17) FRB Jy ms m Hz ✓ ◆ where = S W is the observed fluence, and ⌫ is the observing bandwidth (Thornton Fobs obs et al., 2013; Keane & Petroff, 2015). Using distances derived from the redshifts estimated using Eq. 1.15, the known FRBs have estimated energies between 1031 1033 Joules and are located at co-moving distances between 1 and 3 Gpc; however, these are upper limits given the large uncertainty on the true redshift (Keane & Petroff, 2015). An effective brightness temperature can also be calculated for each FRB given the flux, distance, width, and observing frequency, similar to Eq. 1.1; however, relevant equations require the luminosity distance and a corrective factor due to frequency dilation over the expanding Universe such that

2 2 4 Speak DL ⌫W (1 + z) T 1036K (1.18) B ' Jy Gpc GHz ms 2 ✓ ◆✓ ◆ ✓ ◆ where is the Lorentz factor used to account for relativistic beaming effects ( =1is used here as the beaming process is unknown for the FRB emission mechanism). The population of FRBs discovered to date have estimated brightness temperatures T 1035 B ⇠ K, which is orders of magnitude greater than any other transient population discussed in 20 Chapter 1. Introduction

Section 1.1 with the exception of the nanoshot pulses observed from the Crab pulsar. Such high brightness temperatures require coherent emission well above the 1012 K cutoff for ⇠ incoherent synchrotron emission.

1.3.2 FRB rates and progenitor theories

The true progenitor (or progenitors) of FRBs remains unknown although a large number of theories exist to explain their origin. Currently, the number of theories for their sources is greater than the total number of known FRBs and none of them have been conclusively proven correct. Indeed, some of the current theories may be difficult to verify or refute on timescales of human civilization. Theories for the progenitors of FRBs must not only explain the large DMs and seemingly high brightness temperatures, but they must also be consistent with the very high FRB event rates inferred from current observations. The current best rate estimate in the published literature comes from the detection of 4 FRBs in 24% of the HTRU high latitude survey by Thornton et al. (2013). From this survey the rate can be calculated as

+0.6 4 1 1 RFRB( 3Jyms) 1.0 0.5 10 sky day . (1.19) F⇠ ⇥ The inequality in the equation is due to a fluence incompleteness at broader pulse widths (Keane & Petroff, 2015). This incredibly high rate – several thousand FRBs per day – is approximately equal to the all-sky rate of core-collapse supernovae (Thornton et al., 2013), but places a difficult constraint on any progenitor theory. We provide an updated rate based on the full HTRU high latitude survey in Chapter 6. Here we discuss only a handful of the most promising theories based on their prevalence in the published literature and their testability. In the following paragraphs we will discuss theories for FRBs as hyperflares, , giant pulses from extragalactic pulsars, and pulsars in nearby galaxies.

i. Magnetar hyperflares. In this model first put forward in Popov & Postnov (2010) a short radio pulse can be generated during a highly energetic flare from a neutron star with a high magnetic field strength, a magnetar. Observed X-ray flares from are thought to be caused by reconnection of magnetic field lines on the surface of the star, and can be extremely violent events (Hurley et al., 2005). As the shock from a flare propagates outward it may create a highly relativistic forward shock. Lyubarsky (2014) proposes that a coherent synchrotron maser may be produced at the shock front which would be responsible for the millisecond duration radio emission seen as an FRB. 1.3. Fast Radio Bursts 21

This model can elegantly explain both the high brightness temperature and the high rate seen for FRBs, as the rate for such flares over cosmological volume should be similar to that derived for FRBs (Thornton et al., 2013; Kulkarni et al., 2014). Many observable magnetars in our Galaxy lie in overdense regions such as within supernova remnants or near the Galactic Center (Eatough et al., 2013). If the FRB progenitors

lay in similarly dense regions the contribution of DMsource to the total DM would be significant, greatly reducing the true redshift and distance.

ii. Blitzars. A is the theoretical emission from a neutron star as it collapses to form a black hole. Falcke & Rezzolla (2014) propose that a neutron star created above the theoretical mass limit, but able to maintain its state due to its high rotational velocity, eventually spins down over thousands to millions of years due to magnetic braking. At a certain point the star is no longer able to maintain itself against gravitational collapse and the neutron star forms a black hole at which point a strong shock will propagate outwards, disrupting the neutron star magnetosphere and producing bright radio emission. In this scenario the characteristic timescale for such an event would be of the order of the free-fall timescale of the collapsing material, estimated to be . 1 ms (Falcke & Rezzolla, 2014). The shock may manifest itself observationally as a sort of magnified pulsar emission, which would produce the required high brightness temperature and occur on the timescale of the collapse. Falcke & Rezzolla (2014) also argue that the rate is sufficiently high to explain the observed FRB rate as blitzars would need to occur as only a small subset ( 3%) of the core-collapse supernova ⇠ population out to z 1 to be consistent with FRB observations. The cataclysmic  of the blitzar progenitor implies a population of non-repeating sources. iii. Giant pulses from pulsars. Cordes & Wasserman (2015) have suggested that individ- ual energetic pulses from extragalactic pulsars may be responsible for the observed FRB population. The brightest pulse from the Crab pulsar over its lifetime would be visible above current detection thresholds within a distance of 300 Mpc and similar ⇠ abnormally bright pulses from a population of pulsars within this distance may be the source of FRBs. If each pulsar within a distance of 100 Mpc emits 2 105 such ⇥ pulses over their lifetime it would be enough to make this population consistent with the FRB rate. However, given that the giant pulses emitted from the Crab are not typical of the pulsar population as a whole, Cordes & Wasserman (2015) argue that a cosmological population is preferred in which pulsars out to z . 1 emit these giant pulses which might then be magnified through gravitational lensing of individual stars. 22 Chapter 1. Introduction

If such pulses are similar to those from the Crab it is more likely that an individual energetic pulse will be made up of a number of individual shot pulses that remain unresolved in current observations, such as the nanoshots seen from the Crab that are modulated with a Gaussian envelope (Hankins & Eilek, 2007). While the energet- ics and rate arguments align well with the properties of FRBs, Cordes & Wasserman (2015) acknowledge that even though the population itself is made of repeating sources the probability of seeing a repeat pulse from such an object is extremely low on human timescales.

iv. Pulsars in nearby galaxies. An extragalactic but non-cosmological origin for FRBs has recently been proposed in both Pen & Connor (2015) and Connor et al. (2015) suggesting that FRBs are bursts from local magnetars and young pulsars within 200 Mpc. In these models the FRB progenitor is an energetic and highly active neutron star that lies in a very dense region. Magnetars are observed to lie in overdense regions (Kulkarni et al., 2014) and young neutron stars are often still embedded within the left over from their birth. In both cases the removal of the z term in Eq. 1.18 makes it much easier to satisfy energetics requirements. In the case of a young energetic pulsar in a supernova remnant, if the number of young pulsars is proportional to the core-collapse supernova rate and each pulsar emits a giant pulse every 100 days, then the FRB rate is satisfied within the local volume out to about 200 Mpc (Connor et al., 2015). Given that magnetar giant flares are much less frequent than once every 100 days (Turolla et al., 2015), and that the population of magnetars is smaller than that of young pulsars, this model is difficult to reconcile with the large overall FRB rate; however, due to their proximity, much smaller flares could be detected than from magnetars at cosmological distances.

In each of the cases listed above one common thread emerges – neutron star emission or interaction with the circumstellar medium as the progenitor for FRBs. The large amount of energy released by neutron stars, combined with their ability to generate coherent emission from a small emission region, and their residence in extreme environments make them likely candidates for the type of short, extreme emission necessary to produce the observed pulses from FRBs. Many models of the FRB population assume for simplicity that each burst is a stan- dard candle, i.e. every FRB has the same intrinsic luminosity (Lorimer et al., 2013; Hassall et al., 2013; Macquart & Johnston, 2015). The interest in whether or not FRBs are stan- dard candles is considerable, as this would have profound implications for determining 1.3. Fast Radio Bursts 23

FRB distances without the need to identify a host galaxy. Additionally, a standard can- dle emission mechanism would make strong constraints on possible emission physics and progenitor theories. Currently, however, the population of known FRBs is not sufficient to make these measurements.

1.3.3 Pulse propagation

As FRB pulses travel through the ionized plasma in the host, IGM, and ISM they should experience the same effects of dispersion, scattering, Faraday rotation, and possibly scintil- lation, described in Sections 1.2.1 through 1.2.4. Dispersive effects in the IGM and the ISM are essentially the same as both are within the cold plasma regime. The only difference between the traditional DM measured for pulsars and the dispersion measured for FRBs is that, due to the expansion of the Universe, the frequency at which the FRB is observed is redshifted from the emitted frequency. At higher frequencies the dispersive effects are less severe and thus the measured DM for FRBs is slightly weighted towards the electrons in the local Universe, as those in the host galaxy and medium local to the progenitor have less effect. The redshift effect for dispersion holds true for Faraday rotation as well. The linearly polarized light from the source will experience less Faraday rotation at the higher emitted frequency than it will when traveling through the interstellar medium of our own Galaxy. This has been quantified by Hammond et al. (2012) as

zs ne(z)B (z) dl 2 RM(z )=0.81 k dz rad m (1.20) s (1 + z)2 dz Z0 for polarized, extragalactic radio sources, where zs is the redshift of the source and the functions ne(z) and B (z) are the electron density and the magnetic field strength as a k function of redshift along the line of sight. Even with an understanding of the weighting of the Faraday rotation along the line of sight, disentangling the effects of these different regimes, especially without an accurate distance to the source, is not currently possible. As with pulsars, maximal scattering of an FRB pulse should occur at a screen halfway between the source and the observer, however for FRBs this places the screen in the middle of the IGM, where turbulence is expected only on very small scales which are unlikely to produce scattering at an observable level for millisecond pulses (Luan & Goldreich, 2014). However, scattering is clearly seen for several FRB pulses (Lorimer et al., 2007; Thornton et al., 2013; Ravi et al., 2015; Chapter 6). The source of the scattering of FRB pulses is still a topic of debate with some arguing that scattering in the IGM is impossible and 24 Chapter 1. Introduction must occur in the host (Luan & Goldreich, 2014) while others argue that a host-based scattering origin is not possible and the IGM is a much more likely source of the scattering screen (Macquart & Koay, 2013). McQuinn (2014) has argued that the scattering seen for a subset of FRBs may be due to passing through the extended halo of an intervening galaxy which would act as an ideal scattering screen. This is a promising hypothesis given that it would explain the inconsistent presence of scattering in the FRB population. Ultimately FRBs may be powerful tools for studying extreme physics at cosmological distances, and may also be ideal probes of the ionized material in distant galaxies and in the IGM, both of which have been previously inaccessible. The total population of FRBs known to the author is only 21 sources. Far greater numbers are required to answer some of the questions presented here related to their progenitors, distances, luminosities, and emission mechanisms.

1.3.4 Perytons

Our understanding of the origin of FRBs has been challenged by a second poorly un- derstood class of objects called “perytons” identified in pulsar survey data (Burke-Spolaor et al., 2011a). Peryton signals mimic the sweep in frequency seen in dispersed astrophysical signals but occur simultaneously in all 13 beams of the multibeam receiver with a similar signal-to-noise ratio (S/N) in each beam. They also have peculiar frequency structure, with all perytons showing bright patches in specific regions of the observing band (Figure 1.7). They have never been successfully associated with any astrophysical source (Kocz et al., 2012). The spacing between beams of the receiver makes it such that an astrophysical pulse will usually occur in one beam, and at most around four beams if the source is so bright as to be visible in sidelobes of neighboring beams. Only one published FRB, the Lorimer burst, was detected in multiple beams, appearing in four adjacent beams of the receiver. However, the Lorimer burst was so exceptionally bright that it saturated the automatic gain controller in the primary beam (Lorimer et al., 2007). Nevertheless, the resemblance between perytons and localised astrophysical sources in their frequency-time spectrum is unsettling (see Figure 1.7 and compare with the FRB in Figure 1.6). However, perytons appear with non-uniform brightness across the observing band and are much broader than FRB pulses ( 30 ms for perytons and <5 ms for FRBs). ⇠ Until recently it was unknown what source of local radio frequency interference (RFI) might be capable of producing this frequency-dependent time delay. All known perytons were discovered in archival survey data from Parkes taken between the years 1998 and 2003 and all occurred in the middle of the day, which suggests an origin in human activity 1.4. Thesis outline 25

Figure 1.7 The frequency-time spectrum of a peryton detected at the Parkes radio telescope. The frequency-time behavior of the pulse mimics the dispersion seen for astrophysical sig- nals; however, the peryton spectrum is less broadband and more clumpy than the spectrum of a typical astrophysical pulse.

(Bagchi et al., 2012). Despite differences between FRBs and perytons with respect to many key properties, the “peryton problem” has led some to question the astrophysical nature of FRBs, implying that they could both be manifestations of the same local signal (Kulkarni et al., 2014). We address the peryton issue and identify their source in Chapter 5 concluding that FRBs and perytons are very unlikely to have the same progenitor and that the astrophysical origin of FRBs remains likely.

1.4 Thesis outline

This thesis focuses on the transient radio emission seen from pulsars and fast radio bursts using the single dish Parkes radio telescope. As shown above, short radio pulses provide the ability to study not only energetic and compact objects, but also the ISM (and even the IGM) in a way not possible otherwise. In Chapter 2 we present the technical systems used to detect radio transients at Parkes and how these systems have evolved over time. We introduce one of the most successful transient detection systems to date – the Berkeley Parkes Swinburne Recorder (BPSR) – and summarize the current real-time transient detection efforts underway. Chapter 3 details a long-term study of 168 young pulsars. The DMs of these pulsars were monitored over more than 6 years to look for dispersion measure variations and 26 Chapter 1. Introduction to probe the medium-scale turbulence in the interstellar medium. Of this sample four pulsars were found to have significant DM variations over this time period, likely due to a combination of ISM turbulence and turbulent material local to the source. We use these sources to explore the turbulent scales of the interstellar medium. In Chapter 4 we describe a search of the High Time Resolution Universe (HTRU) survey for fast radio bursts at intermediate and low Galactic latitudes. This search yielded no new FRBs, an unexpected result given the large amount of time on sky compared to previous surveys. This work reveals an absence of fast radio bursts at low Galactic latitudes, largely inconsistent with a homogeneous sky distribution and the current estimates of the FRB rate. This work also shows that Galactic obscuration cannot account for the non- detection of these sources and concludes that a larger population is needed to understand the latitude-dependence of FRB detectability. Chapter 5 reveals the source of the peryton population. Perytons are generated by premature opening of microwave oven doors during a heating cycle and are detected at Parkes when the telescope is at an appropriate angle relative to the oven. We determine that the microwave oven source of perytons cannot explain the observed properties of the FRB population and conclude that they have different origins. Chapter 6 details a renewed search of the HTRU high latitude survey and 5 additional bursts found in the remaining survey pointings and updated rate estimates for the FRB population. A follow-up campaign was conducted for 8 of the FRBs from the survey. In order to rule out sources that repeat on short timescales, detailed follow-up observations are needed. We detected no repeat emission from any FRB in the sample. We place limits on periodically repeating sources with periods less than a day, and completely rule out periodic repeaters with periods 8.6 hours.  Chapter 7 reports on a fast radio burst FRB 140514 discovered in real-time at the Parkes telescope. FRB 140514 was the first source detected with full-polarization and was found to be 21 7% circularly polarized, primarily on the leading edge of the pulse, ± with no detectable linear polarization above the & 10% level. Multi-wavelength follow-up observations were also made for this burst. No variable counterpart was detected in the observations; however, this resulted in the first X-ray, optical, and radio limits on any FRB afterglows. Finally, in Chapter 8 the results detailed in this thesis are reviewed. The future of transient radio astronomy is discussed, with emphasis on the bright prospects of next- generation telescopes such as the Square Kilometre Array. 2 Technological advances in transient radio astronomy

In this chapter we present the technical challenges to observing radio transients. Since the discovery of pulsars, specialized data acquisition hardware and software have been developed to maximize pulsar science. Here we describe some of the most useful advances in radio astronomy instrumentation for pulsar and radio transient observing modes, including the development of the real-time transient detection system in operation at the Parkes radio telescope.

2.1 Data acquisition

Single dish radio observations are typically made with orthogonally polarized receptors at the primary focus which couple the electromagnetic signal to two voltage signals. Bright pulses from pulsars and fast radio bursts increase the variance of the voltage output by each receptor. The power of pulses that have been subjected to dispersion is spread, or “smeared” out in time, making them difficult to detect without some correction for the frequency-dependent time of arrival (see Section 1.2.1). To correct for the effects of dispersion it is necessary to divide the signal into many frequency channels by passing it through a spectrometer. The current spectrometer of choice for pulsar observations is the polyphase filterbank (PFB), which separates the signal into a predetermined number of frequency channels across the bandwidth (Ferris & Saunders, 2004). The PFB is typically implemented using field programmable gate arrays (FPGAs) where the operations are encoded in firmware and highly parallelized (Manchester et al., 2013). Current systems at telescopes such as Parkes record 8-bit data and a PFB with frequency channel width b is able to sample with time resolution 1/b for complex samples

27 28 Chapter 2. Technological Introduction

(Nyquist, 1928). For a system with 1,024 frequency channels over 400 MHz of bandwidth this corresponds to 1,024 complex 8-bit samples for each polarization every 2.56 µs, a data rate of 1,600 Megabytes (MB) per second. Such a data rate is unsustainable for a large survey given current digital storage facilities. As such, the recorded data are often detected then integrated over multiple time bins to decrease the time resolution. The number of samples over which the data are integrated determines the resulting time resolution of the instrument. In most cases, after integration, the data are also reduced from 8-bit to 2-bit after summing the two polarizations and re-normalizing the data. Decimation in time and bits serves to reduce the data rate to a manageable level ( 1 10 MB/s). However, ⇠ it is worth noting that this data reduction strategy, performed before recording the data to disk, results in all polarization information being lost as the two independent signals from which polarization properties can be derived are summed, and only total intensity is recorded. While not ideal, this compromise allows the storage of high time resolution data over four times the amount of time on sky than would be possible otherwise. For the processing pipeline described here, the data are written to disk with the pro- grammed time resolution consisting of a time series for each of the individual frequency channels. This is known as the filterbank data format. Filterbank data are extremely versatile as they can be searched for both single pulses and periodic sources. All of this is possible in software using packages like the sigrpoc1, presto2, peasoup3,andheim- dall4 distributions. Searches for new pulsars and single pulses are carried out over a range of DMs (and periods for pulsars) as described in Section 2.2. Alternatively, if the period and DM of a pulsar are well determined, then the pulsar can be timed — the arrival time of the pulsar pulses can be compared with a model that predicts the phase of the pulsar’s periodic signal. For these observations the data can be added over many pulses at the period of the pulsar to obtain an ‘average’ or ‘integrated’ pulse profile. The time of arrival (TOA) of the integrated pulse profile in a single observation can be compared to the predicted arrival time extrapolated from previous timing observations. Deviations of the pulsar TOAs from the model are minimized to improve measurements of the pulsar parameters such as period, period derivative, and any orbital parameters for a pulsar in a binary system. The precise timing of pulsars can be useful for a variety of experiments, from gravitational wave detection using an array of precisely timed pulsars (Hellings & Downs, 1983; Foster & Backer, 1990; Manchester et al., 2013), to comparing the

1https://github.com/SixByNine/sigproc 2https://github.com/scottransom/presto 3https://github.com/ewanbarr/peasoup 4http://sourceforge.net/projects/heimdall-astro/ 2.1. Data acquisition 29

Figure 2.1 The configuration of the 13-beam Parkes 21-cm multibeam receiver. Individual beams have a full-width half maximum (FWHM) of 14.4 arcminutes. The distance between two beam centers in the x-direction as shown here is equal to the FWHM and the separation in the y-direction is p3 FWHM, or 25 arcminutes. The center beam is designated by beam 1, the inner ring consists of beams 2–7, and the outer ring consists of beams 8–13. The numbers labeled on this figure are used as the names of the beams throughout this thesis.

emission properties of young pulsars in radio and gamma-rays (Weltevrede et al., 2010b).

2.1.1 The Parkes radio telescope

Parkes survey data at 1.4 GHz are primarily acquired using the 21-cm multibeam receiver, which has been in operation at the telescope since 1997 (Staveley-Smith et al., 1996) and consists of a central beam surrounded by two concentric hexagonal rings (Figure 2.1). Each beam has a full-width half-maximum (FWHM) on the sky of approximately 14.4 arcminutes at 1 GHz. The gains and widths of the beams are slightly different for the inner and outer rings of the receiver and are listed in Table 2.1. In a single observation, or pointing, individual filterbank files are recorded for each beam and saved separately to disk. For recent surveys at Parkes this has been done using the Berkeley Parkes Swinburne Recorder (BPSR) instrument, which has an FPGA-based polyphase filterbank that passes 64-µs down-sampled data to 13 CPUs which perform the decimation in bits. The data are then recorded to disk, after which they are sent to the Swinburne gSTAR supercomputer facility via optical fibre for storage and a search can be performed off-line for any single pulses in the data. The advantage of this combined 30 Chapter 2. Technological Introduction

Table 2.1 Specifications for the 13 beams of the Parkes 21-cm multibeam receiver. Each beam has 2 orthogonal linear polarizations, and a system temperature of 23 K (Keith et al., 2010). The gain and beam width vary for the inner and outer rings and the central beam and are taken to be the values at the centre observing frequency of 1382 MHz. Specifications from Manchester et al. (2001).

Beam Center Inner Ring Outer Ring 1 Beam gain (G,KJy )0.7350.690 0.581 Beam width (arcmin) 14.0 14.1 14.5 hardware/software approach is the flexibility of the system as well as the ability to maintain the full-polarization 8-bit data in CPU memory for a certain amount of time, which is useful in the case of real-time searches (see Section 2.2.2). In the following sections we focus on the methods for finding single pulses in filterbank data including the algorithms used in these searches and the application of these techniques in recent surveys to search for fast radio bursts.

2.2 Detecting single pulses

A signal in filterbank data over a bandwidth ⌫ with a peak flux density Speak and width W will have a maximum signal-to-noise ratio

Speak S/N= np ⌫W, (2.1) Ssys p in a given telescope system where Ssys is the system equivalent flux density, and np is the number of polarizations summed to create the signal. The system equivalent flux density can be related to the configuration of the system by

Tsys S = , (2.2) sys G where Tsys is the system temperature, is a correction factor to account for small losses in the digitization process and G is the telescope gain, which mainly depends on the effective aperture of the telescope. These are a function of telescope parameters and receiver configuration. Using Equation 2.2 in Equation 2.1 gives

SpeakG S/N= np ⌫W. (2.3) Tsys p This equation represents the best-case detection S/N for a square pulse once the data have been de-dispersed to the exact DM of the pulse, summed in signal over the entire 2.2. Detecting single pulses 31 bandwidth, and integrated for a time t = W (i.e. a boxcar of width W has been used to smooth and decimate the time series). However, the detected width will be greater than the intrinsic width of the pulse due to the recording systems and the effects of the interstellar medium (as presented in Section 1.2). These effects sum in quadrature to yield the detected pulse width,

2 2 2 2 2 W = tsamp + Wint +tDM +tDMerr + ⌧d , (2.4) q where tsamp is the sampling time of the data, Wint is the intrinsic width of the pulse, tDM is the dispersion-induced time delay caused by intra-channel smearing of the pulse such 3 that tDM = 8300 DM b/⌫ for a frequency channel of width b at a frequency ⌫ (both in MHz), tDMerr is broadening of the pulse due to an error in DM of size DM such 3 that tDMerr = 8300 DM ⌫/⌫ over a bandwidth ⌫ where the fractional bandwidth is small, and ⌧d is the scattering timescale (Section 1.2.2; Cordes & McLaughlin, 2003). The goal of a survey is to search for pulses of any width contained in the data, which necessitates searching over a wide range of possible DMs and pulse widths for peaks in S/N as a function of time. Given the large parameter space to search, a fixed number of DM trials and boxcar filters over a range of possible pulse widths are used to detect peaks in S/N(t). Several search codes exist to do this, such as dedisperse_all5, destroy6, sigproc, presto,andheimdall. All these codes operate in essentially the same way in that they use a list of trial

DMs between some DMmin and DMmax. For each DM trial the data are de-dispersed, integrated in frequency, and searched for pulses over a range of widths. In the search codes mentioned above, the data are searched for pulses with widths W =2n time samples for n =0, 1, 2,...,12 by convolving the time series with a boxcar filter, essentially a square pulse. In the case of the 64-µs sampling discussed in Section 2.1 these trials correspond to pulse widths W =0.064, 0.128, 0.256,...,262 ms. The methods for searching for pulses with widths larger than a single sample are slightly different for the different search codes mentioned above. In some cases, such as for dedis- perse_all, the data are down-sampled by a factor of 2 for each trial by averaging every two samples (reducing the number of time samples by 2n for the n-th iteration) and then searched for peaks. If a pulse spans two adjacent decimated time samples then it will be detected with a less than optimal S/N (Figure 2.2). This down-sampling can decrease the S/N by up to 1/p2 and has been called the ‘root 2 problem’ (Keane & Petroff, 2015). The

5https://github.com/SixByNine/psrsoft 6https://github.com/evanocathain/destroy_gutted 32 Chapter 2. Technological Introduction

Figure 2.2 The recovered signal-to-noise ratio (S/N) of a pulse as a function of phase along the time series. In this case the pulse was injected with S/N = 16 and a width of 2 ms. Both heimdall and destroy, with their sliding boxcars, recover the maximum S/N throughout. dedisperse_all and seek both experience a decrease in S/N by up to 1/p2 when the boxcar and pulse are out of phase. seek peaks twice due to a 2-bit smoothing step performed prior to downsampling. From Keane & Petroff (2015) root 2 problem can be avoided if, instead of down-sampling, the data are convolved with a sliding boxcar of width W and the original temporal resolution is retained. Both destroy and heimdall make use of a sliding boxcar search and consistently recover the maximum signal-to-noise ratio. Using values from a typical single pulse search at 1.4 GHz — 1,749 DM trials optimally 3 3 spaced between DMmin = 0 pc cm and DMmax = 5,000 pc cm to account for the expected amount of pulse broadening between two adjacent trials (Levin, 2012), and 13 width trials from 20 to 212 samples — a single observation requires 22,737 separate searches. The large number of required operations is computationally expensive. By far the most time-consuming process in the search algorithm is the de-dispersion process, which takes approximately an order of magnitude more compute time than any other process including the matched filter searches and event detection (Barsdell, 2012). The de-dispersion speed constraint has been a serious impediment to faster search code in recent years making it difficult to achieve near real-time processing speeds (Magro et al., 2011; Barsdell et al., 2012). A recent solution to the de-dispersion speed problem was developed by Barsdell et al. (2012) who implemented the de-dispersion algorithm for computation on a graphics pro- 2.2. Detecting single pulses 33 cessing unit (GPU). Unlike a CPU which is designed to perform sequential operations on data, the GPU architecture is highly optimized to perform the same operation simultane- ously on multiple data using large numbers of parallel processors. Historically GPUs were developed for gaming which requires high-speed rendering of images; however, astronomers now apply this parallelized approach to astronomical problems such as N-body code accel- eration (Nitadori & Aarseth, 2012) and gravitational microlensing simulations (Vernardos & Fluke, 2013). The use of GPUs in the de-dispersion stage of single pulse data analysis has resulted in a processing speed nine times faster than what was previously possible on a CPU (Barsdell, 2012). The heimdall code written by Ben Barsdell and Andrew Jameson makes use of GPUs for data processing and is currently the fastest single pulse search software available. Due to its speed and optimized S/N search advantages over other software, heimdall is the primary code used for searches of Parkes filterbank data. Below we discuss how this software has been implemented to search for single pulses in generic Parkes survey data as well as how it has been modified and optimized for real-time searches.

2.2.1 The heimdall pipeline

The heimdall search code takes in a subset, or gulp,ofdatathatspanashorttime interval and processes it in its entirety over the DM–width parameter space in search of single pulses. The entire observation is processed gulp by gulp with overlaps between gulps to compensate for possible dispersed pulses at gulp edges. The results are output to a candidates file as an ASCII list of detected peaks with information on each candidate such as S/N, width, DM, and time into the observation. A single pulse will be detected at a range of DMs close to the true value and these events must be merged into one before returning the DM that gives the maximum S/N. Thus thousands of events can be identified prior to merging, which will go on to produce only a handful of independent candidates. After the individual beam data have been fully processed the candidate lists are combined to create a full candidate list for the observation along with information for each candidate such as the number of beams in which it was detected. Candidates detected simultaneously in multiple beams can be eliminated to reduce radio frequency interference (RFI) in the near-field. Candidate pulses can be visualized in a number of ways. The most direct way of validating a candidate is to look at a time-frequency plot of intensity for the source, as shown in Figure 1.3, to judge the bandwidth of the candidate and whether it truly appears to be a dispersed, broadband pulse. However, a single pointing can return hundreds of 34 Chapter 2. Technological Introduction false candidates and the quickest way to eliminate these is to look at an overview plot for the pointing (Figure 2.3). Both methods used together are useful for finding promising candidates in survey pointings. The pipeline for reduction of filterbank survey data is outlined in Figure 2.4 and roughly consists of processing individual beams of a pointing, merging the results from all beams and rejecting coincident detections that appear in multiple beams. Various cuts can then be applied to the data to search for candidates of interest for further study. Candidates with low DM (DM 2), and candidates that appear in fewer than 3 adjacent DM trials are  rejected immediately as these are typically caused by zero-DM RFI or noise in the data. The candidates remaining after these initial cuts then form the basis of our searches for rotating radio transients (RRATs) and FRBs within the survey. The single pulse searches described in this thesis are characterized by the following search parameters:

3 3 2pccm DM < 5000 pc cm  20 samples W 212 samples   N 4 (2.5) beams;adj  where the first two lines represent the DM and width ranges over which we search and

Nbeams;adj is the number of adjacent beams in which the signal is present. A signal can appear in up to 4 adjacent beams, beyond which it is deemed to be RFI. If the signal is coincident in multiple beams that are not adjacent it may pass this criterion and later be classified as RFI in the coincidence check. These are the basic criteria used to create the candidates files for all single pulses in a single survey pointing. Further refinements or cuts can be made on these candidates to select a sub-sample of interest, as described in Chapter 4.

2.2.2 Real-time searches

Real-time searches for single pulses are performed on-site at Parkes as the data are being taken. Instead of being sent to Swinburne, each beam of a pointing is processed on a separate GPU within the HI Pulsar signal processor (HIPSR) server system that hosts the BPSR backend. The processing time for a gulp of data varies depending on the gulp size. However, the performance of heimdall is optimized to make efficient use of GPU memory using a gulp size of 16.8 seconds which can be fully processed, on average, in under 10 seconds; these speeds make real-time searches possible for the first time. To maintain 2.2. Detecting single pulses 35 sintheDM-timeplot,andin 200 ⇠ t is visible at 3 search of a single Parkes pointing. The top left plot gives 940 pc cm heimdall the top right corner of the DM-S/N plot; this is FRB 110220 published by Thornton et al. (2013). a histogram of candidateseach over candidate the and DM itsthe range S/N. pulse In searched, width the one and bottom forcandidates S/N, panel each cut respectively, showing of with based time the the on and(orange). 13 primary candidate low A detection beams. DMs, DM single beam the bright The (cyan, labeled color pulse top bottom in and in right the of size beam scatter circle. of plot), 3 plot the All of or gives filled width other appearance the circles 16 symbols in DM represent ms represent too of and spurious DM ⇠ few DM trials and coincidence in multiple beams Figure 2.3 A sample overview plot of candidates generated in a 36 Chapter 2. Technological Introduction

Figure 2.4 The processing pipeline for filterbank data from Parkes using the heimdall single pulse search software. Operations performed on the FPGAs are shown in red, CPU operations are shown in blue, and GPU operations are shown in green. For offline process- ing the complete filterbank files for a pointing are transferred to the Swinburne gSTAR supercomputing facility and processed on the high performance computing nodes. For real- time processing the gulps of data are taken from the ring buffer and searched on HIPSR. Each beam of filterbank data is processed separately with heimdall and the candidates are cross-checked and concatenated into a single file. A variety of search thresholds can be applied to the full candidates list depending on the sources of interest. 2.2. Detecting single pulses 37 consistent processing speed, certain quick cuts must be made to ensure that no backlog of incoming data occurs when large numbers of false candidates are generated by the presence of strong RFI in the data. Therefore, an additional cut is made in the real-time heimdall that is not implemented in the off-line processing described in Section 2.2.1: if the number of ungrouped candidates identified in the gulp exceeds 100,000 the processing of the gulp stops and the processing moves on to the next gulp. The results of a single gulp from all 13 beams are run through the coincidence pipeline where strong cuts are made with the explicit goal to search for FRBs. These cuts are much more restrictive than what is applied off-line, but the aim of the search is to identify pulses which have strong FRB-like characteristics in the data as quickly as possible. Namely, pulses that have

DM 1.5 DM ⇥ Galaxy S/N 10 N 4 beams  W 8.192 ms  N (t 2s t +2s) 5 (2.6) events obs ! obs  where DMGalaxy is the DM contribution predicted by the NE2001 model (Cordes & Lazio, 2002), the width cut specifically focuses on events of short duration, and the final line of Equation 2.6 stipulates that there cannot be more than five other candidates in the pointing within a 4-second window centered around the candidate of interest. All archival FRBs found to date pass the thresholds set in this equation, although less stringent cuts are made in the off-line processing to ensure no potential ‘atypical’ FRBs are ignored (see Chapter 4 for details) and to allow RRATs through the first detection stage. The power of the real-time search system is that these operations are all performed while the 8-bit data and the full polarization information are still available. This is made possible by the use of a ring buffer incorporated into the BPSR system which holds 120 seconds of 8-bit data from all 13 beams while the real-time processing is underway. As long as the system keeps up with the incoming data any FRB in the pointing should be found before the 8-bit data are deleted from the buffer, making it possible to preserve the full polarization information for a FRB. Within the real-time heimdall pipeline, when a candidate matching all the criteria in Equation 2.6 is detected, the 8-bit data for the candidate are saved in the time window 38 Chapter 2. Technological Introduction

t t t t + 2t (2.7) 0   0 where t is the time into the observation, t0 is the time of the candidate event at the highest frequency in the observing band, and t is total dispersive delay between the highest and lowest frequencies in the observing band (given in Equation 1.2), essentially preserving the candidate and a buffer of data of length t on either side of the event. The pipeline was initially tested in early 2014 by eliminating the DM cutoff condition from Equation 2.6 and observing bright pulses from RRATs, which successfully triggered the real-time data acquisition under the anticipated conditions. The system was put into use as part of normal operations mode for BPSR in March 2014. The first real-time FRB discovery made by this system is presented in this thesis (see Chapter 7) and the system has primarily been used in transient surveys such as the High Time Resolution Universe survey (HTRU, real-time visualization only; Keith et al., 2010) and the ongoing Survey for Pulsars and Extragalactic Radio Bursts (SUPERB; Keane et al, in prep.).

2.3 The High Time Resolution Universe survey

The High Time Resolution Universe survey (hereafter, HTRU) began in 2008 as an ambi- tious effort to survey the entire radio sky with sub-ms time resolution to search for pulsars, RRATs, and other radio transients. The survey is conducted jointly by the Parkes radio telescope in the South (Keith et al., 2010), and the Efflesberg radio telescope in the North (Barr et al., 2013) both observing at a center frequency of 1.4 GHz. The HTRU survey consists of three regions or sub-surveys; these three elements of the HTRU South survey are presented in Table 2.2. Note that although the PFB has 1,024 frequency channels the usable Nchannels = 870. This is due to the presence of persistent strong RFI from transmit- ting satellites at the highest frequencies in the band, which reduce the available bandwidth that is useful in searches from 400 MHz to approximately 340 MHz. The Southern survey ended in February 2014 with 100% of the planned low and in- termediate latitude pointings and 98% of the high latitude survey completed over 6.5 ⇠ years, from 2008 to 2014. The pointing durations for each component are designed for different science goals in the different latitude regimes. The long pointings at low Galactic latitudes were designed to detect new pulsars too weak for detection in previous surveys. At intermediate latitudes the survey aim was to cover a large area of sky quickly to search for pulsars slightly out of the plane that may be useful for timing projects as well as to search for any new RRATs. The high latitude portion of the survey, which covers by far 2.3. The High Time Resolution Universe survey 39

Table 2.2 The parameters of the three components of the HTRU South survey conducted at the Parkes telescope. The survey bounds are listed in terms of Galactic longitude and latitude (`, b) and declination (). Table adapted from Keith et al. (2010).

Survey High Intermediate Low <+10 120 <` 80 <` Survey Bounds `<30 `<30 b < 15 b < 3.5 | | | | Pointing duration (s) 270 540 4300 Nbeams, completed 435 500 95 056 15 990 ⌧samp (µs) 64 64 64 ⌫usable (MHz) 340 340 340 ⌫channel (kHz) 390.625 390.625 390.625 Nchannels 870 870 870 Pointing length (samples) 222 223 226 ⇠ ⇠ ⇠ Data/beam (GB) 1.0 2.0 16.0 Data/total (TB) 435 190 250 the largest area of sky, was designed primarily to look for transient phenomena such as FRBs. Few new pulsars were expected in the high latitude survey. The intermediate latitude component was completed first with all pointings taken be- tween 2008 and the end of 2010. The low latitude survey completion was the second priority and the last months of the survey were entirely focused on the remaining high latitude pointings. Initial discoveries from the HTRU South survey have been incredibly fruitful. Highlights include the discovery and study of several new pulsars (Bates et al., 2011, 2012; Levin et al., 2013), the discovery of numerous millisecond pulsars (Keith et al., 2012; Burgay et al., 2013; Ng et al., 2014), studies of single pulse properties of pulsars (Burke-Spolaor et al., 2012), the discovery of new RRATs (Burke-Spolaor et al., 2011b), the discovery of a radio-loud magnetar (Levin et al., 2012), and the discovery of a millisec- ond pulsar with an ultra-compact planetary companion (Bailes et al., 2011). The citations above report on the processing and analysis of only a subset of the HTRU South data and further discoveries are expected. One of the most intriguing results from the high latitude survey was the discovery of four new fast radio bursts (Thornton et al., 2013) firmly establishing the FRB population and giving rise to a new wave of interest in the progenitors and sources of FRBs. These four sources were discovered in a single pulse search of only 24% of the full high latitude survey performed in late 2012, before the survey was complete. The numbers hinted at an enormous all-sky FRB rate ( 104 sky 1 day 1) making the detection of more bursts ⇠ in the full survey extremely likely. Ultimately this prompted development of the real- 40 Chapter 2. Technological Introduction time system currently in use at Parkes. Even before the real-time triggering mode was implemented, the data were visualized with heimdall overview plots such as the one in Figure 2.3 as observations were underway, effectively allowing for visual inspection in real- time beginning in May 2013. Three new FRBs were discovered in this observing mode in June and July 2013 in near-real-time; however, no polarization data were available at the time (Chapter 6). The HTRU North survey is not yet complete, and additional FRB discoveries are expected in coming years as part of the ongoing survey in the Northern Hemisphere. Analysis of the data from the HTRU South survey is still underway and work remains to be done on all components of the search: RRAT and pulsar searches in the high latitude survey; RRAT searches and improved acceleration searches in the intermediate latitude survey; and completion of ongoing RRAT, pulsar, and FRB searches of the full low latitude survey. Future searches are expected to yield additional interesting pulsars, RRATs, and FRBs. Searches of the intermediate latitude survey for FRBs and perytons and the high latitude survey for FRBs are presented in Chapters 4, 5, and 6, respectively. 3 Dispersion measure variations in a sample of 168 pulsars

In this chapter we present a study of a large population of young pulsars over a 6 year time period to search for variations in pulsar dispersion measure. Multi-year variations in DM can be used to probe intermediate scales of turbulence in the interstellar medium, a region inaccessible via other measurements. We find 5- variations over the span of the observations for only 4 pulsars in the sample, most of which is attributable to dense material local to the pulsar. These observations can be used to place limits on DM variations expected for pulsars and other radio sources at high DM.

3.1 Introduction

The emission from pulsars experiences a time delay as it passes through the interstellar medium (ISM) due to the dispersive effects of its plasma component. The group delay of this signal tg(⌫) depends on the observation frequency and the electron density along the line of sight as

d 0 ned` tg(⌫)= (3.1) hR K⌫2 i 4 2 3 1 where K is the dispersion constant with a value K 2.410 10 MHz pc cm s , ⌘ ⇥ ⌫ is the observing frequency in GHz, ne is electron density and d is the distance of the pulsar from the observer. The expression in brackets is refered to as the dispersion measure (DM) and describes the amount of ionised interstellar material between the observer and the pulsar (Chapter 1). 2 Dispersive effects are proportional to ⌫ and can be determined experimentally by

41 42 Chapter 3. Dispersion measure variations of young pulsars measuring delays in pulse arrival times for a pulsar across the bandwidth of an observation at a single frequency or fitting over a range of frequencies (Keith et al., 2013 and references therein). When a pulsar is first discovered, the spin period and DM are directly obtained as part of the search process. In most cases, a pulsar’s DM is treated as a constant but several long-term studies of DM have revealed temporal variations on timescales of months to years (Hamilton et al., 1985; Phillips & Wolszczan, 1991; You et al., 2007; Keith et al., 2013). Variations in dispersion have been used to study turbulent structure in the free electron density of the ISM. The spectral energy density scale of interstellar turbulence is thought to show power law statistics for interstellar material between large (1018 m) and small (106 m) spatial scales such that

2 P (q) C q (3.2) 3N ⇡ N represents the power spectrum of the electron density P3N with a spectral index and a 2 structure coefficient CN . It is estimated in Armstrong et al. (1995) that this spectrum is consistent with a Kolmogorov power law described by = 11/3. However, inhomogeneities in the form of highly anisotropic filaments are believed to exist in the ISM (Brisken et al., 2010) and may be responsible for so-called extreme scattering events (Fiedler et al., 1987; Romani et al., 1987). Turbulent interstellar material is probed on various scales by different types of observa- tions, from rotation measure variations at the largest scales to weak diffractive interstellar scattering at the smallest. Fluctuations in pulsar DMs provide the capability to probe the ISM in the middle of this spatial range between 1011 and 1012 m, where other techniques are incapable of detecting variations. Thus, DM variation measurements bridge a crucial gap in the ISM turbulence spectrum. Significant DM variations have previously been observed in studies of different classes of objects. Varying DM along the line of sight to the was first noted in Hamilton et al. (1985); DM was observed to decrease over the length of their study. They attributed these changes to a dense, magnetised filament within the (SNR) passing out of the line of sight over the 15 years of data (Hamilton et al., 1985). Similarly, 3 1 DM to the Crab pulsar has been observed to increase by 0.02 pc cm yr over 68 epochs between 1982 and 1988, attributed to variations within the turbulent environment of the local Crab SNR (Lyne et al., 1988). A study of seven pulsars over two years in Phillips & Wolszczan (1991) detected variations caused by interstellar turbulence with a spectral index 3.1. Introduction 43 of = 11/3, and DM variations have also been observed in high-precision observations of millisecond pulsars (e.g. You et al., 2007). These variations are generally consistent with levels of turbulence in the ionised ISM, though there is growing evidence that the exponent of the power-law noise process is steeper than expected from Kolmogorov turbulence for some lines of sight (Keith et al., 2013). Changes in DM over time provide a direct method of probing turbulence in the inter- stellar medium (Rickett, 1977). These changes relate to the DM structure function DDM as

1/2 dDM (DDM) = (3.3) dt ⌧ where dDM/dt is the absolute rate of change of the DM over time in pc cm 3 yr 1 and | | ⌧ is the span of the observations in years (Backer et al., 1993). The structure function, in turn, is related to the diffractive timescale, ⌧d, of the pulsar by You et al. (2007)

K⌫ 2 ⌧(s) ↵ D = . (3.4) DM 2⇡ ⌧ ✓ ◆ ✓ d ◆ K has the same value as in Equation 3.1, ⌧(s) is the time span of the observations in seconds, and ↵ = 2, where = 11/3 is taken to be the power-law exponent of a Kolmogorov spectrum (Armstrong et al., 1995). Here ⌧d is the diffractive timescale in seconds at the observing frequency ⌫. In this chapter we examine the DMs of more than 160 young, highly energtic pulsars monitored regularly over six years at the Parkes radio telescope as part of a radio coun- terpart study to one conducted with the Fermi gamma-ray telescope (Smith et al., 2008). The pulsars in our sample are distributed at a range of distances within the Galactic plane 3 3 with DMs between 2 pc cm and almost 1000 pc cm . The majority are at low Galactic latitudes and probe a diverse range of sight lines through the Galactic ISM. Our sample is additionally promising for DM variation studies as young pulsars are also more likely to be associated with supernova remnants remaining from their birth, providing the possibility of yet more ionised, turbulent local structure. Many previous studies mentioned here focus on millisecond pulsars, most with DM 3 < 100 pc cm . Canonical pulsars are detected up to much higher DMs and provide a complementary sample to the MSPs, allowing us to study turbulence over larger lines of sight. In Section 3.2 we describe our observations; in Section 3.3 we describe our data analysis procedures and our statistic for measuring DM variations in our pulsars. In Section 3.4 44 Chapter 3. Dispersion measure variations of young pulsars we outline our findings, with special attention paid to four pulsars of interest: PSRs J0835 4510, J0908 4913, J1824 1945, and J1833 0827, and we set upper limits for all others in Section 3.5; we conclude in Section 3.6.

3.2 Observations

Since early 2007, regular observations of a large sample of pulsars have been carried out with the 64- Parkes radio telescope in support of the Fermi gamma-ray mission. A total of 156 pulsars were drawn from the list of highly energetic pulsars in Smith et al. (2008) supplemented by a small number of other interesting southern sources. The initial description of the observational setup and early timing results from the Parkes dataset are described in Weltevrede et al. (2010b). For this work we used data taken using the Parkes telescope between February 2007 and October 2012. Observations were carried out on an approximately monthly basis with all 168 pulsars observed over a 24 hour period at a centre frequency near 1.4 GHz. At 6 month intervals, additional observations were obtained simultaneously at 3.1 and 0.7 GHz on the following day. During a single observation each pulsar was observed long enough to obtain a signal to noise ratio greater than 5, typically only a few minutes. Observations at 1.4 GHz with 256 MHz of bandwidth were taken using the centre beam of the Parkes multibeam receiver (Staveley-Smith et al., 1996). Observations at 3.1 and 0.7 GHz were done simultaneously using the 10/50 cm receiver installed at Parkes (Granet et al., 2005) and had 1,024 MHz and 64 MHz of bandwidth, respectively (Weltevrede et al., 2010b). The voltage signals from the orthogonal, linearly-polarised receptors were digitised and converted to a filterbank consisting of 1,024 frequency channels then folded using 1,024 phase bins across the pulse period of the pulsar. Incoming data were folded at the period of the pulsar in 30-second sub-integrations and then written to disk. Regular calibration was performed by sending an artificial calibration signal into the feed at a 45 angle from both linear probes to determine both their relative gain and the phase offset. The final calibrated data were saved to disk and used to create an average pulse profile over the full span of the observation.

3.3 Analysis

Initial data reduction was performed after each observation using the psrchive data anal- ysis package (Hotan et al., 2004). An integrated pulse profile for each pulsar observation 3.3. Analysis 45 was produced after excising time and frequency channels with significant radio frequency interference. The remaining channels were calibrated to produce a final pulse profile at each observing frequency which was compared with a standard profile for the pulsar, created by summing profiles from all previous observations (Weltevrede & Johnston, 2008). The time of arrival (TOA) for the integrated profile was converted to the solar sys- tem barycentre using the DE405 model (Standish, 1998) and compared with that of the timing model prediction using the tempo2 software package (Hobbs et al., 2006). The time difference between the observed TOA and the timing model gave a residual for each observation. Once several epochs of residuals were available, it became possible to remove common parameters such as effects from the pulsar spin frequency ⌫ and spin frequency derivative ⌫˙ through fits within tempo2. These fits removed linear and quadratic terms from the residuals, respectively. At this stage it was possible to fit for the DM contribution in the refined residuals. Previous studies of DM from pulsar timing residuals have identified problems of ob- taining reliable values and accurately correcting for DM effects in the timing solution. You et al. (2007) performed a linear fit by directly comparing the arrival times at two different frequencies. Since all epochs in their sample consisted of observations at 3 distinct centre frequencies they determined the best two with which to perform calculations on a case-by- case basis. This method was updated in Keith et al. (2013) by using all three observing frequencies simultaneously to determine the DM for each epoch. Keith et al. also incorpo- rated a smoothing function into the tempo2 fitting algorithm that was previously applied after fitting. We created two DM datasets using the algorithm described in Keith et al. (2013). Each pulsar has been observed at approximately 80 epochs at 1.4 GHz but at only 10 epochs at 0.7 and 3.1 GHz. Thus two separate time series were created, one for DM measured across the 20 cm band only, and one measured across all three frequencies, where available. In some cases the number of useful multi-frequency DM measurements is smaller than the number of epochs at which observations were taken because scattering effects preclude detection at 0.7 GHz. To calculate DM variations over the span of the dataset, we performed a weighted least-squares fit to measure the slope in DM over time. The result of this fitting procedure was a best-fit slope, dDM/dt, taken to be the variation in DM. Separate fits were done for 20 cm and multi-frequency DM measurements resulting in two separate dDM/dt values for each pulsar. Multi-frequency fits of dDM/dt for pulsars with only one or two usable multi-frequency epochs were not considered as the linear fit had no associated error. 46 Chapter 3. Dispersion measure variations of young pulsars

Table 3.1 Pulsars with DM variations over 6 years above 3 levels. Pulsars with detec- tions above 5 levels are listed in bold. Slopes for the multi-frequency (mf) and 20 cm 3 1 observations with errors in the last digit are in units of pc cm yr .

Name DM dDM/dtmf dDM/dt20cm J0835 4510 67.9 0.005(1) 0.0081(9) J0908 4913 180.4 0.038(4) 0.030(1) J1824 1945 224.4 0.011(2) 0.022(1) J1833 0827 411 0.13(2) 0.18(1) J0834 4159 240.4 0.020(5) 0.27(3) J1112 6103 599.0 0.11(4) 0.52(5) J1702 4128 366.7 0.12(3) 0.4(1) J1721 3532 497.01 0.047(9) 0.19(6) J1745 3040 88.112 0.018(2) 0.016(3) J1809 1917 196.9 0.07(2) 0.17(4) J1826 1334 230.8 0.14(4) 0.13(3)

We note that the weighted linear fit is not a perfect tool for the study of small-scale variations, and we may expect variations on timescales shorter than the many-year span of our dataset that do not fit this trend. Overarching linear changes in DM would be expected to arise in cases where a single large structure moves across the line of sight over the span of observations. Additionally, the steep power law of Equation 3.2 implies that the most power will be at the largest timescale, thus the largest DM variations. A linear fit may also encompass spatial density gradients but provides a good first order detection statistic.

3.4 Results - Detections

DM variations were detected in eleven pulsars from our sample over the 6 years of obser- vation. All are embedded in the Galactic plane with dispersion measures ranging from 67.9 to 599 pc cm 3. Only four pulsars in our sample, PSRs J0835 4510, J0908 4913, J1824 1945, and J1833 0827, had variations deemed highly significant. These pulsars wer identified based on the fact that they all had values of dDM/dt with agreement in sign between 20 cm and multi-frequency fits with an error in each fit 35%. Variations were  labeled as highly significant detections if errors were 20% in both fits. Fits for all other  pulsars in our sample failed to meet one or more of these criteria and the weighted linear fit was used to produce an upper limit on detectable variations. Marginal detections, pulsars with detections between 3 and 5, are largely consistent with variations predicted from an ISM dominated by Kolmogorov turbulence. 3.4. Results - Detections 47

Figure 3.1 The DM of PSR J0835 4510 over 2000 days for 18 epochs of multi-frequency measurements (circles) and 97 epochs of measurements across the band centred at 20 cm (squares) with weighted linear fits represented by the dashed and dot-dashed lines, respectively. The general trend is of increasing DM over the dataset with a notable reversal between MJDs 54600 and 55200.

Table 3.1 lists the pulsars with significant measurements of DM variations. Highly significant pulsars will be discussed individually below.

3.4.1 PSR J0835–4510

PSR J0835 4510 (B0833 45), located in the Vela supernova remnant, lies within the Gum Nebula (Large et al., 1968). It is one of the brightest pulsars in the sky, with a flux density at 1400 MHz of 1100 mJy (Backer & Fisher, 1974) and characteristic age ⌧c of 11 kyr. It is located at a distance from the Sun of approximately 300 pc (Dodson et al., 2003) at Galactic longitude and latitude (`, b) = (263.5 , 2.79 ). The pulsar has a very large DM 2 for its distance and using CN as a measure of turbulence (Cordes, 1986), its value is the highest measured of any pulsar (Johnston et al., 1998). The scintillation velocity at high observing frequencies has been measured by Johnston et al. (1998). Corresponding to a value of ⌧ 10 s at 1.4 GHz. We therefore expect DM d ⇠ 3 1 variations of order 0.01 pc cm yr based on Equation 3.3. Previous studies of changing DM along the line of sight to Vela (Hamilton et al., 1985) showed DM to be decreasing over 3 1 5 epochs between the years 1970 and 1985 at a rate of 0.040 pc cm yr , a factor of 4 greater than expected from turbulence. Hamilton et al. (1985) interpreted their measured 48 Chapter 3. Dispersion measure variations of young pulsars decrease in DM as the movement of a magnetised filament out of the line of sight within the SNR, meaning that changes in DM were due to the local environment of the pulsar rather than the turbulence in the greater ISM. As shown in Figure 3.1, single frequency DM measurements over the entire six years of 3 1 our dataset show DM increasing at a rate of 0.0081(9) pc cm yr .Similarly,dDM/dt 3 1 = 0.005(1) pc cm yr using multi-frequency fits. These variations are significantly smaller than those measured by Hamilton et al. (1985) but are more or less consistent with expected values given the measured scintillation parameters. The dramatic change in dDM/dt is highlighted in Figure 3.2 where we show the Hamilton et al. data over plotted with data from this study and analogue filterbank system archives at Parkes. In order to reduce the noise due to measurement uncertainty we plot weighted averages in 100-day bins. We caution that there may be a systematic offset between the DMs derived in the Hamilton et al. data and those derived from our work due to known intrinsic frequency evolution of the Vela profile (Keith et al., 2011). We interpret the dramatic change in dDM/dt to be evidence that the filament responsible for change seen by Hamilton et al. moved completely out of our line of sight some time near MJD 50000. We believe that the currently observable DM variations can be explained solely by the turbulent ISM. Although the overall trend in dDM/dt in the recent data is positive, variations are visible on shorter timescales within the span of our observations, most notably where DM appears to decrease between MJD 54600 and 55200. This is consistent with turbulence in the ISM. Contributions from the surrounding Gum Nebula are most likely minimal, however, as there are no detectable variations in other pulsars from the Gum in our sample –PSRsJ0738 4042, J0742 2822, J0745 5351, and J0905 5127.

3.4.2 PSR J0908–4913

The pulsar PSR J0908 4913 (B0906 49) has a spin period of 106 ms, and a DM derived distance of 6.7 kpc placing it well within the Galactic plane, at (`, b) = (270.27 , 1.02 ) (D’Amico et al., 1988). Gaensler et al. (1998) discovered a bow-shock in the immediate surroundings of the pulsar suggesting that it is travelling through the turbulent medium 1 of an associated (PWN) with velocity 60 km s at a position angle of 315 (north through east). Their derived velocity is consistent with the scintillation measurements of Johnston et al. (1998), who measured the diffractive timescale to be 4770 s at 1.5 GHz. This yields expected DM variations of 6.0 10 5 pc cm 3 yr 1. ⇥ No long term studies of DM along the line of sight to PSR J0908 4913 exist in the 3.4. Results - Detections 49

69 pc)

-3 68.5 DM (cm

68

67.5 40000 42000 44000 46000 48000 50000 52000 54000 MJD

Figure 3.2 DM measurements of PSR J0835 4510 since 1969 (MJD 40500). Square markers indicate single values taken from Hamilton et al. (1985). Other markers are weighted averages of the DM measured in 100 day bins of Parkes observations. Values before MJD 54200 are taken from archival Parkes data recorded using an analogue filterbank system. The lines indicate extrapolation of the trends from the Hamilton et al. dataset and from this work. literature, however previously published DM estimates for this pulsar do exist as far back as its discovery in 1988, when its DM was measured to be 192 12 pc cm 3 (D’Amico ± et al., 1988). At the beginning of the Fermi dataset in 2007 DM was measured to be 3 3 approximately 180.42 pc cm , but had dropped by 0.12 pc cm by the final epoch from 2012. Unfortunately the large uncertainty in the 1988 data and lack of data in the intervening years prevent us from drawing any conclusions as to the long term evolution of the DM.

The variation in DM along the line of sight to PSR J0908 4913 is one of the largest measured of any pulsar in our sample, with a best-fit linear slope of dDM/dt = 0.030(1) pc cm 3 yr 1 for the 20 cm DM measurements and dDM/dt = 0.038(4) pc cm 3 yr 1 over 12 multi-frequency epochs shown in Figure 3.3. Similar to PSR J0835 4510, fluctuations on shorter timescales are present in the 20 cm data, and are visible even in the more sparsely sampled multi-frequency dataset.

The DM variations along the line of sight are more than two orders of magnitude greater than expected from models. The high variations, then, seem likely to be caused by the immediate surroundings of the pulsar. PSR J0908 4913 has an unusual PWN and 3 appears to be moving slowly through a highly dense (ne > 2 cm ), and likely turbulent, medium (Gaensler et al., 1998). Thus for a region on the order of a in size, the 50 Chapter 3. Dispersion measure variations of young pulsars

Figure 3.3 The DM of PSR J0908 4913 over 12 multi-frequency epochs (circles) and 77 epochs measured across only the 20 cm band (squares) with weighted linear fits represented by the dashed and dot-dashed lines, respectively.

PWN contribution to total DM would be small but the nebula’s highly turbulent nature would be capable of much larger fractional contributions to dDM/dt as the pulsar moved through it. DM variations can then be attributed to a small percentage variation from within the local PWN.

3.4.3 PSR J1824–1945

PSR J1824 1945 (B1821 19) is a young pulsar with a period of 189 ms and a characteristic age ⌧ =5.7 105 yr (Manchester et al., 1978). It lies in the Galactic plane close to the c ⇥ centre of the Galaxy with Galactic coordinates ` = 12.28 and b = 3.11 at a DM-derived distance of approximately 5.2 kpc (Cordes & Lazio, 2002). Previous studies of PSR J1824 1945 were unable to make significant measurements of pulsar velocity or DM variations (Hobbs et al., 2004; Zou et al., 2005), thus neither signif- icant velocity nor diffractive timescale measurements exist for this pulsar in the literature. 1 Assuming a velocity of 300 km s we expect a diffractive timescale of order ⌧d =6s, 3 1 which in turn corresponds to DM variations of magnitude 0.02 pc cm yr using the Cordes & Lazio (2002) model. The general trend in our data is a decrease in DM over the six years of observations seen in Figure 3.4, although the best-fit rate of this variation differs between the 20 cm and the multi-frequency data with dDM/dt = 0.022(1) pc cm 3 yr 1 and dDM/dt = 20cm mf 0.011(2) pc cm 3 yr 1, respectively. These gradients differ by a factor of 2, with lower 3.4. Results - Detections 51

Figure 3.4 DM measurements for PSR J1824 1945 over 11 epochs of multi-frequency data (circles) and 75 epochs measured across the 20 cm band alone (squares) with weighted linear fits represented by the dashed and dot-dashed lines, respectively. error in the fit at 20 cm. The good fit to the data may arise because the trend over 2000 days is not a strictly linear one; additional fluctuations are visible on shorter timescales possibly due to turbulent structure on small scales passing through the line of sight. In the case of PSR J1824 1945 the DM variations we observe are consistent with values predicted for a turbulent ISM in this direction for our assumed velocity and DM-determined distance. This pulsar is not well-studied like PSRs J0835 4510 and J0908 4913, but there is no evidence for the presence of an associated SNR or PWN in the local neighbourhood. Therefore we attribute the variations in the DM towards this pulsar to the turbulent ISM alone.

3.4.4 PSR J1833–0827

PSR J1833 0827 (B0830 08) has a period of 85 ms and an approximate characteristic age of 150 kyr (Clifton & Lyne, 1986). At the time of its discovery it was identified as an isolated pulsar close to the supernova remnant W41. It has been argued that the pulsar’s high velocity away from SNR W41 indicates a past connection and that PSR J1833 0827 originated within the shell-like SNR (Hobbs et al., 2005). More recently, X-ray studies of the region using the XMM-Newton satellite discovered an X-ray pulsar wind nebula surrounding this pulsar in the form of a bow shock nebula (Esposito et al., 2011). PSR J1833 0827 is located towards the Galactic centre (`, b)=(23.39 , 0.06 )ata DM-derived distance of 5.6 kpc (Cordes & Lazio, 2002). This pulsar has been observed 52 Chapter 3. Dispersion measure variations of young pulsars

Figure 3.5 The DM of PSR J1833 0827 over the 2000 days of observations as measured over 11 epochs of multi-frequency data (circles) and 67 epochs measured across the 20 cm band alone (squares) with weighted linear fits represented by the dashed and dot-dashed lines, respectively. The general trend is that of decreasing DM with a sign reversal between MJD 54900 and 55300.

1 to move with a transverse velocity of approximately 740 km s , more than twice the rms velocity of non-millisecond pulsars (Nicastro & Johnston, 1995; Hobbs et al., 2005). The high transverse velocity and distance correspond to a scattering timescale at 1400 MHz of 3 1 5 s and expected DM variations of 0.02 pc cm yr . From our observations we find a general trend of decreasing DM along the line of sight to J1833 0827 with dDM/dt = 0.13(2) and dDM/dt = 0.18(1) in units mf 20cm 3 1 of pc cm yr , seen in Figure 3.5, for multi-frequency and 20 cm DMs, respectively. These variations are larger than those of any other pulsar in our sample by an order of magnitude. However, these fits include a change in sign of dDM/dt between MJD 54900 and MJD 55300 seen in both sets of DM measurements, most likely attributable to general turbulence; a piecewise fit to these data would yield an even larger value for dDM/dt over the regions of decreasing DM. PSR J1833 0827 also has the largest DM of any pulsar in which variations were detected. These extreme DM variations may seem more reasonable in light of the recent discovery of the pulsar’s X-ray PWN. The turbulent ISM along the line of sight to PSR J1833 0827 would be expected to contribute only about 10% of the observed variations and it is possible that observed behaviour is due entirely to the energetic bow shock nebula through which the pulsar is moving, including the brief passage of a dense filament through the line of sight corresponding to the temporary increase in DM midway through the dataset. 3.5. Results - Upper limits 53

10

) 1 -1

pc yr 0.1 -3

0.01 dDM/dt (cm

0.001

0.0001 1 10 100 1000 DM (cm-3 pc)

Figure 3.6 Upper limits on dDM/dt for pulsars in which no significant DM variations were detected. Overlayed lines display predicted variations detectable at a range of DMs for a 1 1 pulsar at ` = 330, b =0 with a velocity of 164 km s (dashed), 338 km s (dotted), 1 and 511 km s (dot-dashed), the median and 1 sigma velocities from the pulsar velocity distribution in Hobbs et al. (2005).

3.5 Results - Upper limits

Upper limits on dDM/dt for each of the pulsars in the Fermi project with no significant DM variations are listed in Table 3.2 along with the DM that best fits our data. All but 36 of our measured DM values have smaller uncertainties than previous best estimates. In Figure 3.6 we plot our upper limits as a function of DM. We compared the 25 pulsars common to our observations and the Hobbs et al. (2004) study and found our upper limits were consistent with but less constraining than theirs. Although our limits on dDM/dt may also contain contributions from spatial density gradients our data are not sufficiently sensitive to disentangle this contribution from that of variations due to turbulence. Because we are only setting upper limits on total varia- tions, we do not attempt to explicitly differentiate between the two. Ideally we want to compare our upper limits with theoretical predictions for dDM/dt based on Kolmogorov turbulence. To do this we would need to measure the diffractive timescale ⌧d and then apply Equations 3.3 and 3.4. Unfortunately, the high DMs of most of our sample preclude direct measurements of ⌧d as, in the majority of cases, they will be significantly smaller than our integration time. For any given line of sight we can however, estimate ⌧d using 54 Chapter 3. Dispersion measure variations of young pulsars the Cordes & Lazio (2002) model of the Galaxy, which estimates scintillation along the line of sight consistent with a Kolmogorov spectrum of turbulence. For illustrative purposes in Figure 3.6 we choose a representative line of sight at (`, b)=(330, 0)andusethe Cordes & Lazio (2002) model to step through DM and derive the scintillation bandwidth,

⌫. Conversion of ⌫ to ⌧d is obtained through

(D⌫)1/2 ⌧ = A , (3.5) d V ⌫v where D is the distance to the pulsar in kpc, ⌫ is the diffractive scintillation bandwidth 1 in MHz at the observing frequency ⌫ in GHz, v is the velocity of the pulsar in km s , and we adopt a value of A =3.85 104 from previous studies (Nicastro et al., 2001). V ⇥ Figure 3.6 shows the curves obtained for three different velocities, v=164,338,and511 1 km s , the median and one sigma 2D velocities obtained using the distribution in Hobbs et al. (2005). We note that a number of upper limits lie below the theoretical value expected from 1 the median two dimensional velocity of 338 km s (Hobbs et al., 2005). This result is in contrast to the results for the millisecond pulsars for which dDM/dt is higher than expected. By using the probability distribution function for pulsar velocities given by Hobbs et al. (2005) we estimate that we should have had 12 pulsars in our sample with measurable values of dDM/dt, in good agreement with the 11 measured. However, this velocity distribution may not be entirely accurate because it models a pulsar’s motion in 2-dimensional space. Detectable DM variations in some pulsars may be more realistically caused by irregularities along a single spatial axis in the interstellar turbulence (Brisken et al., 2010). If this is the case we only need concern ourselves with a one dimensional velocity as the pulsar moves through the ISM. The probability of detecting pulsars in our sample can be recalculated using a one dimensional Gaussian distribution. We find a much lower detection estimate of 6 pulsars, including our real detections. While this alternate model of ISM turbulence may be the cause of variations along some of our lines of sight, we expect this effect only at low DM as the effects of multiple irregularities would cancel out over large distances. We also note that our results appear to be at odds with Bhat et al. (2004) who conclude that NE2001 underpredicts scattering, particularly at high DM. However, Bhat et al. (2004) compare pulse-broadening times estimated from the Cordes & Lazio (2002) model and their own CLEAN-based deconvolution algorithm, which are directly sensitive to an inner scale. Measurements of DM variations are not sensitive to the ISM inner scale which may explain the difference between the findings of these studies. 3.5. Results - Upper limits 55

Table 3.2 All pulsars observed for the Fermi project are listed alongside DM fits from our 3 1 data with error in the last digit, and an upper limit on dDM/dt in pc cm yr .DM values with smaller uncertainties than previous publications are marked with an asterisk (*). Name DM dDM/dtlim PSR DM dDM/dtlimit J0108 1431 1(13) 0.18 J1320 5359 * 97.1(1) 0.08 J0401 7608* 21.7(1) 0.05 J1327 6400 * 679(1) 0.7 J0536 7543* 18.58(2) 0.2 J1341 6220* 719.65(5) 0.07 J0543+2329 77.556(5) 0.006 J1349 6130* 284.5(1) 0.1 J0614+2229* 96.91(4) 0.02 J1357 6429 126(1) 0.5 J0627+0705* 138.25(7) 0.05 J1359 6038* 293.736(3) 0.004 J0630 2834 34.4(1) 0.1 J1406 6121* 537.8(4) 0.5 J0659+1414* 13.94(9) 0.1 J1410 6132* 961.0(3) 0.548 J0729 1448* 91.7(2) 0.1 J1412 6145* 514.4(4) 0.4 J0738 4042* 160.896(3) 0.006 J1413 6141* 670.6(4) 0.4 J0742 2822* 73.728(1) 0.02 J1420 6048* 360.15(6) 0.1 J0745 5353* 121.38(2) 0.03 J1452 5851* 260.5(2) 0.3 J0821 3824 195.8(4) 0.3 J1452 6036* 349.54(2) 0.02 J0855 4644* 236.4(1) 0.02 J1453 6413* 71.248(2) 0.004 J0857 4424 183.4(1) 0.08 J1456 6843* 8.639(7) 0.008 J0901 4624* 199.3(2) 0.2 J1509 5850* 142.1(1) 0.2 J0905 5127 196.21(4) 0.06 J1512 5759* 627.47(1) 0.02 J0940 5428* 134.55(3) 0.04 J1513 5908 255.0(3) 0.46 J0954 5430* 201.57(5) 0.01 J1514 5925* 194.0(4) 0.3 J1003 4747* 98.49(8) 0.0001 J1515 5720* 480.6(4) 0.2 J1015 5719* 278.1(2) 0.1 J1524 5625* 152.2(1) 0.1 J1016 5819* 252.16(7) 0.08 J1524 5706* 832(1) 0.7 J1016 5857* 394.48(9) 0.1 J1530 5327* 49.6(1) 0.1 J1019 5749* 1040(1) 0.8 J1531 5610* 110.41(3) 0.05 J1020 6026* 441.5(4) 0.6 J1538 5551* 604.6(1) 0.3 J1028 5820 96.506(2) 0.006 J1539 5626* 175.85(3) 0.02 J1043 6116* 448.91(2) 0.02 J1541 5535* 426.1(1) 0.2 J1048 5832* 128.679(4) 0.006 J1543 5459* 345.9(3) 0.3 J1052 5954* 491.9(6) 0.5 J1548 5607* 314.66(7) 0.07 J1055 6032* 636.5(1) 0.07 J1549 4848 55.94(4) 0.04 J1057 5226* 29.69(1) 0.01 J1551 5310* 491.6(7) 0.7 J1105 6107* 271.24(1) 0.01 J1600 5044* 262.791(4) 0.006 J1115 6052* 226.92(5) 0.1 J1600 5751 176.4(1) 0.05 J1119 6127* 704.8(2) 0.4 J1601 5335* 195.2(6) 0.4 J1123 6259 223.4(1) 0.1 J1602 5100* 170.79(1) 0.007 J1138 6207* 520.4(4) 0.3 J1611 5209* 127.345(9) 0.01 J1156 5707* 243.2(1) 0.2 J1614 5048* 582.4(1) 0.09 J1216 6223* 790(1) 0.9 J1632 4757* 574.2(5) 0.8 J1224 6407* 97.686(4) 0.006 J1632 4818 758(5) 0.3 J1248 6344* 433.0(6) 0.4 J1637 4553 194.7(1) 0.09 J1301 6305* 374(1) 1.00 J1637 4642* 419.1(3) 0.5 J1302 6350* 146.73(1) 0.02 J1638 4417* 436.4(3) 0.2 J1305 6203* 471.0(1) 0.5 J1638 4608* 423.1(1) 0.2 56 Chapter 3. Dispersion measure variations of young pulsars

Name DM dDM/dtlimit PSR DM dDM/dtlimit J1640 4715* 586.32(6) 0.5 J1803 2137 234.01(5) 0.03 J1643 4505* 478.6(6) 0.3 J1806 2125* 747(1) 1 J1646 4346 499.2(3) 0.4 J1815 1738* 724.6(2) 0.3 J1648 4611* 392.3(3) 0.4 J1820 1529* 768.5(6) 0.7 J1649 4653* 331(1) 0.7 J1825 0935 18.9(2) 0.1 J1650 4502* 320.2(3) 0.2 J1825 1446* 352.23(4) 0.05 J1650 4921* 229.3(3) 0.2 J1828 1057* 249(2) 0.8 J1702 4305* 538.4(5) 0.2 J1828 1101* 605.0(1) 0.1 J1702 4310* 377.6(3) 0.2 J1830 1059* 159.70(1) 0.01 J1705 1906 22.94(2) 0.01 J1832 0827 300.84(2) 0.02 J1705 3950* 207.25(1) 0.05 J1834 0731* 294.0(9) 0.8 J1709 4429* 75.68(3) 0.02 J1835 0643* 467.9(4) 0.5 J1715 3903* 314.0(6) 0.4 J1835 0944* 276.2(1) 0.3 J1718 3825* 247.46(6) 0.04 J1835 1106 132.84(2) 0.04 J1722 3712* 99.49(3) 0.01 J1837 0559* 319.5(6) 0.5 J1723 3659* 254.4(1) 0.07 J1837 0604* 459.3(3) 0.4 J1726 3530* 718(4) 2 J1838 0453* 617.2(4) 0.4 J1730 3350* 261.29(4) 0.06 J1838 0549* 276.6(4) 0.5 J1731 4744* 123.056(4) 0.005 J1839 0321* 452.6(3) 0.6 J1733 3716* 153.18(8) 0.05 J1839 0905* 344.5(3) 0.7 J1735 3258* 758(3) 1 J1841 0425 325.13(3) 0.02 J1737 3137* 488.1(4) 0.3 J1841 0524* 284.5(3) 0.5 J1738 2955 222.5(6) 1 J1842 0905* 343.4(2) 0.1 J1739 2903* 138.55(2) 0.03 J1843 0355* 797.7(6) 0.7 J1739 3023* 170.5(1) 0.1 J1843 0702* 228.4(2) 0.2 J1740 3015* 151.96(1) 0.006 J1844 0256* 826(2) 3 J1745 3040 88.112(7) 0.007 J1844 0538* 411.71(4) 0.05 J1756 2225* 329(1) 7 J1845 0434* 230.8(2) 0.1 J1757 2421 179.38(2) 0.02 J1845 0743* 280.93(2) 0.01 J1801 2154* 386(1) 0.5 J1847 0402 140.8(1) 0.07 J1801 2304 1070(1) 0.7 J1853 0004* 437.5(1) 0.03 J1801 2451* 291.55(5) 0.07 J1853+0011 566(2) 2 3.6. Conclusions 57

We find the two dimensional scattering modelled in NE2001 to agree well with the findings of our study, even in the high DM regime where that model becomes discrepant with others such as Bhat et al. (2004).

3.6 Conclusions

We have analysed over five years of timing data for more than 160 young pulsars to search for any characteristic DM variations along several lines of sight. Only four pulsars in our sample PSRs J0835 4510, J0908 4913, J1824 1945, and J1833 0827 showed highly significant changes in DM over the span of the study with seven other pulsars identified as marginal detections. One pulsar, PSR J1824 1945, dis- played detectable DM variations at levels predicted by an interstellar medium dominated by Kolomogorov turbulence with no contribution from dense filaments local to the pulsar. DM variations for the Vela pulsar, PSR J0835 4510, were also consistent with a purely turbulent ISM, a dramatic change from measurements of large variations made 15 years ago attributed to the local SNR, indicating that perhaps the responsible filament is no longer moving through our line of sight. The other two detections, with variations well above those predicted by theory, are known to lie within turbulent local environments of supernova remnants or pulsar wind nebulae which make large contributions to observable turbulence along these particular lines of sight. No DM variations were observed along the lines of sight to most pulsars in our sample, but we were able to set upper limits on detectable variations. We compared these limits with DM variations predicted from models of Kolmogorov turbulence and found our limits to be within an order of magnitude of theoretical predictions. Comparisons with accepted two dimensional velocity distributions using NE2001 scattering models suggested our ex- periment should have detected DM variations to approximately 12 pulsars. Confining our models to one dimension to simulate scattering effects due to irregularities in the inter- stellar turbulence reduces this estimate to 6 pulsars. We find our results to be in good agreement with 2D scattering from the NE2001 model. The DM variations are a red process with a steep spectral exponent, therefore longer time baselines dramatically increase our sensitivity to DM variations. With more time the observation of young, high-DM pulsars will provide us with an excellent complement to the results obtained from millisecond pulsars at low DMs.

4 An Absence of Fast Radio Bursts at Intermediate Galactic Latitudes

In this chapter we detail a search for fast radio bursts in the HTRU intermediate latitude survey which yielded no detections. A significant population of FRBs was expected in this survey given the large total time on sky and the high FRB rate implied by Thornton et al. (2013). Modeled Galactic effects such as scattering and pulse broadening cannot fully explain the discrepancy between the high and intermediate latitude surveys. Ultimately, a larger total population is needed to understand the latitude distribution of these sources.

4.1 Introduction

Fast radio bursts (FRBs) are single, bright, highly dispersed radio pulses of millisecond duration. These bursts have fluences of 0.6 8.0 Jy ms, and dispersion measures (DMs) well in excess of the expected Galactic contribution along the line of sight. Given the high flux density and high DM-derived distances, they could arise from high luminosity events at z . 1. The first burst of extragalactic origin (Lorimer et al., 2007) was discovered using the Parkes multibeam receiver (Staveley-Smith et al., 1996) in a pulsar survey of the Magel- lanic Clouds at Galactic latitude b = 41 . Subsequently, the astrophysical origin of the Lorimer burst was called into question by the discovery of apparently dispersed sources in archival surveys called “perytons” (Burke-Spolaor et al., 2011a; Bagchi et al., 2012). Perytons exhibit dispersive properties that are similar, though not identical, to those of an astrophysical source; however, they are detected in all 13 beams of the Parkes 21-cm multibeam receiver, which is impossible for a distant point source. They have peculiar frequency structure, with all perytons showing bright ‘nodules’ in specific regions of the

59 60 Chapter 4. An Absence of Fast Radio Bursts at Intermediate Latitudes observing band. One of the scenarios considered in recent work by Kulkarni et al. (2014) is a terrestrial origin for FRBs, where they are interpreted as being similar to peryton events, but occuring at greater distances (> 40km) from the telescope, mimicking a source at infinity. In this model the Lorimer burst, which was detected in four adjacent beams of the multibeam receiver, occupies a place between traditional FRB events and traditional peryton events and is believed to have occured at some distance from the detector close to the Fresnel scale for Parkes, 20 km. Loeb et al. (2014) have proposed that FRBs originate in the envelope of main-sequence flare stars within our Galaxy, after finding a flare star within the full width at half maxi- mum of the detection beam of one FRB from the Thornton et al. (2013) sample. Recently, however, several authors have highlighted why such a model is physically untenable (Luan & Goldreich, 2014; Dennison, 2014). Several extragalactic progenitors (most at cosmolog- ical distances) have also been proposed. A list is presented in Section 1.3.2. Each of these mechanisms is theorized to be capable of producing coherent radio emission of millisecond duration. At present, archival and on-going radio pulsar surveys are the most immediate and obvious places to begin searching for more FRBs. Thornton et al. (2013) estimate an FRB +0.6 4 1 1 rate of RFRB 1.0 0.5 10 sky day for bursts with fluences 3 Jy ms, based ⇠ ⇥ F⇠ on 615 hours of observations. The HTRU intermediate latitude survey used an identical setup to Thornton et al. (2013) with 1,157 hours on-sky. Here we report on a search for FRBs in the HTRU intermediate latitude survey in a DM range from 100 5000 pc cm 3 which returned no new, highly-dispersed pulses. An introduction to the HTRU survey, the analysis tools, and results of the single pulse search are presented in Section 4.2. We present an in-depth discussion of our non-detection in Section 4.3.

4.2 Analysis and Results

The High Time Resolution Universe (HTRU) Survey was designed as a comprehensive sur- vey of the radio sky with 64-µs time resolution at 1.4 GHz to detect pulsars and other tran- sient radio phenomena. Observations for HTRU are divided into three observing regimes at low, intermediate, and high Galactic latitudes. This study focuses on the intermediate latitude component of the survey: 540-s pointings in the range 120 <`<30 and b < 15 with the 13-beam Parkes 21-cm multibeam receiver, which has a 0.5 deg2 field of | | view. 4.2. Analysis and Results 61

The survey data were searched for single pulses using techniques similar to those de- scribed in Burke-Spolaor et al. (2011b). The dedispersion and width filter matching were optimized for processing on a graphics processing unit (GPU) with the new software pack- age heimdall. heimdall produces a list of candidates for each beam. All 13 beams in a single pointing are then run through a coincidence detector. Candi- dates that occur in more than 4 beams, fewer than 3 adjacent DM trials, have a S/N < 6 3 or DM < 1.5 pc cm are rejected. With these cuts we are not sensitive to peryton-like events that occur in all 13 beams of the receiver. We will address the topic of perytons in the HTRU survey in Chapter 5. The candidates that remain after these cuts are conca- tenanted into a single candidate list for the entire pointing. Each pointing candidate file was searched for FRBs by looking for candidates matching the following criteria

S/N > 10 W 28 64 µs = 16.3ms  ⇥ DM/DMGalaxy > 0.9 (4.1) where the width W corresponds to 2n, where n is the trial (0, 1, 2,...8). The DM 2 threshold was intentionally set to include high-DM candidates from sources within the

Galaxy to ensure sensitivity near the theoretical DMGalaxy boundary along these lines of sight. DMGalaxy used in this analysis is obtained from NE2001, a model of the Galactic electron density (Cordes & Lazio, 2002). 3 All pointings were searched out to a DMmax of 5000 pc cm . Extragalactic sources will have a value of DM/DMGalaxy & 1 and pulsars will have DM/DMGalaxy . 1, if the estimated DMGalaxy from NE2001 is correct. For pointings in the intermediate latitude survey the ratio of DMmax to DMGalaxy ranges from 1.1 close to the Galactic centre to > 50 at higher latitudes; we were therefore sensitive to extragalactic sources along every line of sight observed during the survey. The spatial volume probed by this search is at least the same as that in the Thornton et al. (2013) analysis if not greater, depending on the maximum redshift to which FRBs are detectable. A total of 52 candidates were identified in the 1,157 hours of the intermediate latitude survey after applying these criteria. Of these, 29 were found to be zero-DM RFI, and 23 events were caused by narrow-band RFI. No new, highly-dispersed pulses were detected. Our pipeline was also run on a fraction of the HTRU high latitude beams to search for FRBs. All four previously published FRBs were recovered in this analysis. Had these 62 Chapter 4. An Absence of Fast Radio Bursts at Intermediate Latitudes events occured in the intermediate latitude dataset, they would have been detected.

4.3 Discussion

Although the HTRU intermediate latitude survey observed for almost twice as long as the survey reported by Thornton et al. (2013), no bursts were detected. The likelihood of finding N FRBs in our survey based on M FRBs detected at high latitudes, marginalized over the unknown FRB rate, is given by

1 P (N M) = P (N ↵⌘)P (⌘ M)d⌘ (4.2) | | | Z0 where ⌘ is the expected number of detections, and ↵ has a value of 1.88, the ratio of on-sky time of this survey compared to the Thornton survey. Here P (N ↵⌘),thedistributionof | the number of events given an expected number of detections, is Poissonian. Using Bayes’ Theorem and a flat prior for ⌘, the distribution for P (⌘ M) is also Poissonian. Equation 4.2 | thus reduces to: N (1+M+N) (M + N)! P (N M) = ↵ (1 + ↵) . (4.3) | M! N! A full derivation of this equation and why it is used is included in Appendix B. Based on a detection of 4 FRBs in 615 hours of high latitude data the probability of detecting 0 FRBs in the intermediate latitude survey is 0.5%, thus excluding the hypothesis that FRBs are uniformly distributed on the sky with 99.5% confidence. The low probability of this result poses a critical question regarding the spatial dis- tribution of these events. No systematic bias was introduced by the pointing position of the telescope, as both survey components covered similar distribution of telescope azimuth and elevation. The primary differences between the two HTRU survey components are Galactic effects introduced along sightlines at lower Galactic latitudes. Here we discuss four potential contributors to a decreased sensitivity to FRBs at b < | | 15 – dispersion in the interstellar medium (ISM), interstellar scattering, sky temperature, and scintillation effects.

4.3.1 Dispersion in the ISM

A broadband radio pulse traveling through the ISM experiences a dispersive delay pro- portional to the squared wavelength of the radiation and the magnitude of the delay is a measure of the electron column density along the line of sight, DM. All radio pulsars in the Galaxy have a measured DM that can be related to distance from Earth using Galactic 4.3. Discussion 63 electron density models such as NE2001, used in most pulsar DM/distance estimates. The NE2001 model has been calibrated against nearby pulsars for which distances are known. Using NE2001 as a model of the ionized Galaxy, we can estimate the maximum DM contribution DMGalaxy from the Milky Way along any line of sight. At high Galactic latitudes sightlines probe the diffuse halo of the Galaxy and DMGalaxy is typically less 3 than 50 pc cm . Within the region of the intermediate latitude survey, however, Galactic 3 dispersion contributes an average of 380 pc cm and has been measured to contribute as 3 much as 1778 pc cm near the Galactic center (Eatough et al., 2013).

After subtracting DMGalaxy as predicted by NE2001, the high-latitude FRBs have ex- cess DMs between 521 and 1072, which can be attributed to the IGM and any putative host galaxy. For an FRB pulse with a DM0 =DMhost +DMIGM entering the Galaxy along a sightline through the intermediate latitude region the total DM observed would, on average, be of order pc 1500 cm 3 for an FRB with DM similar to the maximum in ⇠ 0 the known sample. We would recover this pulse as it is still below the maximum DM trial 3 in our search (5000 pc cm ) in the absence of other pulse smearing effects.

4.3.2 Scattering in the ISM

Three possible scattering regimes can be considered for FRBs at cosmological distances, scattering due to the host galaxy, the IGM, and the ISM of the Milky Way. Previous studies concerned with the detectability of FRBs have assumed two extreme cases for the IGM: strong, ISM-like scattering, and no scattering (Hassall et al., 2013; Trott et al., 2013; Lorimer et al., 2013). If IGM scattering was as strong as that of the ISM, FRB pulses would be detected with much lower peak flux densities and broader pulses, leading to the conclusion that IGM scattering is likely weak or even un-observable (Macquart & Koay, 2013). Only one FRB in the HTRU sample, FRB110220, showed measurable scattering. Here we consider only the effects of Galactic scattering due to multipath propagation in the ISM, as IGM and host contributions appear to be minimal, and should have similar values for all FRBs irrespective of position. We scale the effects of Galactic scattering in the intermediate latitude survey region using the NE2001 model and the relationship between DM and scattering timescale from Bhat et al. (2004) (Equation 1.7). This relation calculates the expected total of pulse broadening effects along any line of sight for an input DM. We use the DMGalaxy obtained from NE2001 for our analysis in Section 4.3.1 as input to estimate a scattering timescale

⌧d. For the majority of survey pointings (>85%), we determine our measurements are still 64 Chapter 4. An Absence of Fast Radio Bursts at Intermediate Latitudes sensitive to FRB signals even in the presence of strong scattering in the ISM.

4.3.3 Sky Temperature

A third consideration at low Galactic latitudes is the decrease in sensitivity due to sky temperature (Tsky). At radio frequencies observations can be sensitivity limited if the standard deviation of noise fluctuations approaches the total power of the observed source

(Lorimer & Kramer, 2004). In the analysis that follows we use the 1.4 GHz Tsky map from de Oliveira-Costa et al. (2008) to estimate the sky temperature for the survey region.

Excluding pointings within a few degrees of the Galactic centre, Tsky lies between 3 K and 30 K over the intermediate latitude survey. For b > 2 , T is never more than 10 K, | | sky still well below our system temperature (Tsys = 23 K, Keith et al., 2010). The pointings for which Tsky is equal to or greater than Tsys are already sensitivity limited due to scattering and/or the total DM contribution along the line of sight as determined in Sections 4.3.1 and 4.3.2.

Comparing the mean value of Tsky for the intermediate latitude pointings with the mean value at high latitudes we find a difference of only 1 K between intermediate latitude survey pointings at b > 5 (4 K) and high latitude survey pointings (3 K). Sky temperature | | increases significantly only in the Galactic Plane where the mean temperature is closer to

10 K. In addition, Tatmosphere and Tspillover are negligible at Parkes at this frequency. Sky temperature is therefore not a significant factor in our comparison between the two survey components, and does not play a role in limiting our sensitivity, other than in regions where the survey is already limited by other Galactic factors.

4.3.4 Scintillation

The dearth of low latitude FRB events compared with high latitudes might also be ex- plained by scintillation at high Galactic latitudes where FRB pulses may be amplified by single wideband scintles. For the four known FRBs, the scintillation bandwidths ⌫ predicted using NE2001 for each line of sight are 4.8, 2.5, 5.8, and 6.1 MHz for FRB110220, FRB1106261, FRB110703, and FRB120127, respectively. All are around two orders of magnitude too small to produce significant amplification. However, these values are extrapolations from the Bhat et al. (2004) model fit, which has a high variance at the extremes, and may not be exact.

1The published name for this FRB in Thornton et al. (2013) (FRB110627) is incorrect based on the UTC naming convention. 4.3. Discussion 65

Amplifications of pulsar fluxes by orders of magnitude have been observed in the 47 Tucanae and SMC pulsars (Camilo et al., 2000). If some FRBs are similarly amplified it may be that a fraction of those observed would not have been detected unless favourable scintillation conditions existed at the time of their arrival. In the Galactic Plane we know that pulsar flux densities are relatively stable (Stinebring et al., 2000) and not amplified by diffractive scintillation to the same degree. So diffractive scintillation may help explain the larger number of FRBs we detect at high latitudes.

4.3.5 Sensitivity Map

The factors introduced in the previous sections can ultimately be combined to produce a map to determine the fraction of survey pointings in which the combination of these effects dramatically limits sensitivity. We created a simple mask to simulate the expected Galactic effect on a pulse traveling through the ISM. The effects of the temporal smearing of a pulse across the band due to dispersion, pulse broadening due to scattering, and sky temperature will combine to decrease the signal-to-noise detected for an FRB pulse from an ‘intrinsic’ value S/N to an effective S/N0 as

2 2 ⌧ 0 = ⌧d +tDM

q 2 2 W0 = ⌧ 0 +Wint (4.4) q Wint/W0 S/N0 =S/N ⇥ (1p + Tsky/23K) for a receiver with Tsys of 23 K. Here ⌧d is the pulse broadening time, and tDM is the pulse smearing time due to dispersion, which combine to give an effective scattering timescale ⌧ 0, and Wint and W0 are the intrinsic and effective widths, respectively. This set of equations allows us to estimate the detectability of an FRB pulse within the parameters of our survey. From the values in Equations 4.4 we can create a map of the intermediate latitude sur- vey and see where a simulated FRB pulse with properties similar to the FRBs in Thornton et al. (2013) falls below our signal-to-noise threshold of S/N = 10. In Figure 4.1 we simulate an FRB detection with S/N = 13, and a pulse width before traveling through the Galaxy,

Wint = 2 ms. The pulse falls below the detection threshold in 20% of all intermediate latitude pointings, primarily in high-DM regions where the dispersion smearing time and pulse broadening time are large. As expected the regions where we are least sensitive to FRBs are in the vicinity of the Galactic center, in the Galactic Plane, and through the Gum Nebula (at (`, b) ( 100 , 10 )). ⇠ 66 Chapter 4. An Absence of Fast Radio Bursts at Intermediate Latitudes

Figure 4.1 The effective signal-to-noise, S/N0, of an FRB observed along all sightlines of the intermediate latitude survey. An FRB event is simulated with S/N = 13, and width = 2-ms before entering the Galaxy and the effects of dispersive smearing, interstellar scattering, and Tsky are taken into account to estimate the effective signal-to-noise with which the pulse would be detected with the Parkes multibeam receiver.

The total time on sky of pointings not sensitive to FRBs by this metric is 231 hours of observing. This reduces the value of ↵ in Equation 4.3 from 1.88 to 1.51 and increases the probability to 1.0%.

4.3.6 Summary

Even when taking into account Galactic effects, the probabilty of non-detection at inter- mediate latitudes given the current rate estimate is extremely low. We note that the Bhat et al. (2004) model describes the scattering that occurs in our Galaxy; however, owing to the ‘lever arm’ effect (Hassall et al., 2013), the contribution of Galactic scattering to extra- galactic sources may be significantly reduced. In the ‘no scattering case’, the percentage of pointings no longer sensitive FRB pulses decreases to 14% (Figure 4.2) making our results and original predictions still more discrepant. Currently, the rate calculation is based on a handful of sources in only a fraction of the HTRU high latitude data. Further analysis is expected to reveal a substantial population of FRBs that will allow for a more precise rate calculation. An increased sample of FRBs is essential to resolving whether or not there is a true lack of FRBs detected through, or in the direction of, the Galactic plane. The search for additional FRBs at high latitudes is detailed in Chapter 6. 4.4. Conclusions 67

Figure 4.2 The effective signal-to-noise, S/N0 of the simulated FRB from Figure 4.1 with no ISM scattering effects applied. Pulse smearing due to dispersion in the ISM is still accounted for. In this case the simulated FRB would drop below our detection threshold in only 14% of survey pointings

4.4 Conclusions

We conducted a single pulse search of 1,157 hours of the High Time Resolution Universe in- termediate latitude survey of the Southern radio sky at 1.4 GHz with 64-µs time resolution. We searched the data using the new GPU-based single pulse processing tool heimdall over 3 a range of incoherent DM trials from 0 to 5000 pc cm for a range of pulse widths. We searched for pulses with FRB-like measurable pulse properties with a ratio DM/DMGalaxy > 0.9. We did not detect any FRBs in the intermediate latitude survey. We verified the search pipeline on the high latitude FRBs in a blind search and all were recovered. Based on the detections in 615 hours of observations by Thornton et al. (2013) the probability of a non-detection in the HTRU intermediate latitude survey was 0.50%; this low probability lead us to investigate possible causes of the discrepancy. The combined contribution along sightlines through the Galactic disk of dispersion smearing, interstellar scattering, and sky temperature account for a 20% decrease in FRB-sensitive pointings ⇠ for the survey, however this is still not enough to explain the null result and only increases the probability of a non-detection to 1.0%, thus excluding the hypothesis that FRBs are uniformly distributed on the sky with 99% confidence. We conclude that the low probability of agreement between results at high and intermediate latitudes reveals a disagreement between the rates calculated from these two surveys. A further analysis of the HTRU high latitude data will provide a more stringent limit on all-sky FRB rates. 68 Chapter 4. An Absence of Fast Radio Bursts at Intermediate Latitudes

We note that any Galactic model of FRB events must explain not only the dispersion and scattering seen in FRB pulses, but also the latitude distribution found in this work. 5 Identifying the source of perytons at the Parkes radio telescope

In this chapter we present recent searches of archival and contemporary Parkes data for perytons. New perytons, discovered within days of their occurrence combined with new data from on-site radio frequency interference monitoring led to the identification of the source. We were able to generate perytons through unconventional use of the microwave ovens at the Parkes site. We also discuss the implications of the man-made progenitor of perytons on the source of FRBs.

5.1 Introduction

‘Peryton’ is the moniker given to a group of radio signals which have been reported at the Parkes and Bleien Radio Observatories at observing frequencies 1.4GHz (Burke- ⇠ Spolaor et al., 2011a; Kocz et al., 2012; Bagchi et al., 2012; Saint-Hilaire et al., 2014). The signals are seen over a wide field-of-view suggesting that they are in the near-field rather than boresight astronomical sources (Kulkarni et al., 2014). They are transient, lasting 250 ms across the band, and the 25 perytons reported in the literature occurred only ⇠ during office hours and predominantly on weekdays. These characteristics suggest that the perytons are a form of human-generated radio frequency interference (RFI). In fact one of the perytons’ defining characteristics — their wide-field detectability — is routinely used to screen out local interference detections in pulsar searches (Keane et al., 2010; Kocz et al., 2012). Perytons’ most striking feature, which sets them apart from ‘standard’ interference signals, is that they are swept in frequency. The frequency dependent detection of the signal is sufficiently similar to the quadratic form of a bona fide astrophysical signal which has

69 70 Chapter 5. Identifying the source of perytons traversed the interstellar medium, that the origin of the first fast radio burst, FRB 010724 (Lorimer et al., 2007), was called into question by Burke-Spolaor et al. (2011a). This was mainly based upon the apparent clustering of peryton dispersion measures (DMs) around 400 pc cm 3, which is within 10% of FRB 010724’s DM. ⇠ ⇠ Ongoing searches are actively searching for FRBs and perytons and are capable of rapidly identifying detections. In this paper, we report on three new peryton discoveries from a single week in 2015 January made with the Parkes radio telescope. In addition to the rapid identification within the Parkes observing band, the RFI environment over a wider frequency range was monitored with dedicated equipment at both the and the Australia Telescope Compact Array (located 400 km north of Parkes). For one event, the Giant Metrewave Radio Telescope (GMRT), in India, was being used to observe the same field as Parkes. Below, in Section 5.2, we describe the observing setup and details of the on-site RFI monitors. In Section 5.3 we present the results of the analysis of our observations, and our successful recreation of peryton signals. Section 5.4 discusses, in more depth, the identified sources of the signals and we compare the perytons to the known FRB population in Section 5.5. We present our conclusions in Section 5.6.

5.2 Observations

As part of the SUrvey for Pulsars and Extragalactic Radio Bursts (SUPERB1; Keane et al., in prep.), at Parkes, real-time pulsar and transient searches are performed. The live transient searching system developed for SUPERB, which uses the heimdall single pulse search software package, is now routinely used by several projects. The survey data are taken using the Berkeley Parkes Swinburne Recorder (BPSR) with the observing setup described in Chapter 2. For each pointing 13 such data streams are recorded, one for each beam of the multibeam receiver (Staveley-Smith et al., 1996). The survey has been running since 2014 April to search for pulsars and FRBs. In 2014 December, an RFI monitoring system was installed on the Parkes site identical to ones which had been in operation at the Australia Telescope Compact Array (ATCA) since 2014 November. The RFI monitor itself is a Rhode & Schwarz EB500 Monitoring Receiver capable of detecting signals across a wide range of frequencies from 402 MHz to 3 GHz. The frequency and time resolution of the monitoring system are limited to 2 MHz and 10 s, respectively. The antenna is mounted on a rotator, which sweeps out 360 in azimuth every 12 min, then returns to an azimuth of 0 for another 8 min before repeating the cycle.

1https://sites.google.com/site/publicsuperb/ 5.2. Observations 71

A spectrum is produced every 10 s, which is obtained by stepping in 20 MHz steps across the full band. So each 10 s spectrum has only 0.1 s of data at any given frequency. The installation of the monitor gives an unprecedented view of the RFI ‘environment’ at the telescope at any given time and this setup is ideal for identifying very strong signals of RFI which may corrupt observations with the main dish at Parkes.

In 2015 January March, 319.2 h (13.3 d) of 13-beam BPSR data were recorded for the SUPERB survey alone to search for pulsars and FRBs. Total time in the BPSR observing mode in these months was 736.6 h over a range of observing projects aimed at detecting and studying fast transients. Ultimately 350.7 h of these observations were searched for perytons in the months of 2015 January March in this work. Three events were discovered, all occurring in the week starting 2015 January 19, on the 19th (Monday), 22nd (Thursday), and 23rd (Friday) in a search, the PULSE@Parkes outreach project (Hobbs et al., 2009) and SUPERB, respectively. For the event on January 23, simultaneous coverage with the GMRT was also available, which was shadowing Parkes as part of the SUPERB project’s effort to localize FRBs.

The peryton search for SUPERB and other BPSR data is performed after the Parkes data have been transferred to the gSTAR supercomputer facility at Swinburne University of Technology. The peryton search is performed by summing the frequency–time data of all 13-beams from BPSR and searching these summed data using heimdall for single pulses with a signal-to-noise ratio (S/N) 10 and DM 10 pc cm 3. This method ensures that dispersed pulses occurring in a majority of beams are efficiently detected even if they may be too weak to be detected in single-beam searches. For the perytons identified in 2015 January, once the date and UTC time were established the Parkes and ATCA RFI monitor data were checked around the times of the perytons for the presence of signals that might be correlated with the appearance of a peryton at 1.4 GHz. The same search technique was applied to search for perytons in the High Time Resolution Universe (HTRU) intermediate- and high-latitude surveys of Keith et al. (2010). The HTRU intermediate-latitude survey was conducted between 2008 and 2010 and the high latitude component was conducted between 2009 and 2014. The HTRU survey concluded in 2014 February and as such no RFI monitor data are available for events detected in these data nor for any peryton detected before those reported here. 72 Chapter 5. Identifying the source of perytons

Table 5.1 Properties of the perytons from 2015 January

Date Time DM DM S/N Width Telescope Telescope 3 (dd-mm-yy) (UTC) (pc cm ) error (beam 1) (ms) azimuth elevation (deg) (deg) 2015-01-19 00:39:05 386.6 1.7 24.8 18.5 10.7 75.3 2015-01-22 00:28:33 413.8 1.1 42.5 18.5 73.9 36.2 2015-01-23 03:48:31 407.4 1.4 10.6 18.5 323.2 40.2

5.3 Results

5.3.1 Three perytons

The properties of the three perytons discovered in 2015 January are noted in Table 5.1, and Figure 5.1 shows their time–frequency structure. These events are typical perytons in that they are bright and detectable in all beams of the multibeam receiver. They are also apparently dispersed or ‘chirped’ in frequency, but not strictly obeying the quadratic cold plasma dispersion law; signals from pulsars and FRBs are observed to obey this law precisely (Hassall et al., 2012; Thornton et al., 2013). They have a typical peryton spectrum, being broad-band, but brighter at higher frequencies. Conversely, an off-axis detection of an astronomical source (i.e. one effectively at infinity) would be suppressed at the highest frequencies, but the near-field beam pattern is radically different (see e.g. figure 10 in Kulkarni et al. 2014). The existence of a standard template for peryton spectra and similar DMs also suggests that the source, or sources, are at roughly constant distances and possibly consistently reproducible. These three perytons are the focus of our analysis as they were the first with simul- taneous coverage with additional instruments: the RFI monitors operating at both the Parkes and ATCA sites. For all three events, the Parkes RFI monitor detected emission in the frequency range 2.3 2.5 GHz consistent (to less than one time sample) with the time of the 1.4 GHz peryton event. This strongly suggests that the 1.4 GHz millisecond- duration burst is somehow associated with the episodes of 2.4 GHz emission, which last for some tens of seconds. The broad RFI spectra from the monitor at the times around the perytons are shown in Figure 5.2 with the bright emission shown as well as the time of the peryton. Simultaneous emission in the same frequency range was seen in the ATCA data at the time of the first peryton, but no such emission was seen for any other peryton detection, making it likely that this one event was a coincidence (see Figure 5.3). For the 5.3. Results 73

Figure 5.1 The time–frequency structure of the three January perytons (bottom of each panel) and the pulse shape after dedispersion to the optimal DM in Table 5.1 and summed in frequency across the band (top of each panel). In the case of events on 2015-01-19 and 2015-01-23, the summed 13-beam data are shown. For 2015-01-22 only beam 1 is plotted as the outer beam data were not recorded to disc. 74 Chapter 5. Identifying the source of perytons third peryton, simultaneous coverage with GMRT at 325 MHz observing in 2 s snapshots also produced no detection. The detection on only the Parkes site confines the source(s) of the peryton signals to a local origin. The 2.3 – 2.5 GHz range of the spectrum is allocated to ‘fixed’, ‘mobile’ and ‘broad- casting’ uses by the Australian Communications and Media Authority, and includes use by industrial, scientific and medical applications, which encompasses microwave ovens, wire- less internet, and other electrical items. This suggests that the perytons may be associated with equipment operating at 2.3 2.5 GHz, but that some intermittent event or malfunc- tioning, for example, from the equipment’s power supply, is resulting in sporadic emission at 1.4 GHz.

5.3.2 Prevalence of 2.3 2.5 GHz signals at Parkes As can be seen in Figure 2 there is at least one case where a single peryton is detected but there are multiple or ongoing detections at 2.3 2.5 GHz around the time of the peryton. This already indicates that while peryton detections at 1.4 GHz coincide with episodes of emission at higher frequency, the higher frequency emission can occur without generating a peryton. More detailed inspection of the archival RFI monitor data at Parkes gives an indication of the prevalence of these episodes at higher frequencies. In the months investigated several hundred spikes of emission were detected in the frequency range 2.3 2.5 GHz. These events cluster in time of day and are much more common during daytime (between the hours of 9am and 5pm local time). A time-of-day histogram of these spikes over the period of 2015 January 18 to March 12 is plotted in Figure 5.4. This is entirely consistent with the use of microwave ovens and other electrical equipment. Tests at Parkes confirmed that microwave ovens produced detectable levels of 2.4 GHz emission in the ⇠ RFI monitoring equipment independent of the azimuth of the rotator. Standard practice at ATNF observatories is not to allow the use of microwave ovens on site when observing in the 2.4 GHz band is taking place.

5.3.3 Archival perytons

Using the search technique described in Section 5.2, 15 perytons were found in the HTRU intermediate latitude survey and an additional 6 perytons were found in a search of 90% of the high latitude survey (the same survey region searched for FRBs in Chapter 6). While the RFI monitor had not yet been set up on site and the RFI environment is impossible to recover, we can use these perytons to study the ensemble properties. Combining the perytons from 2015 January, HTRU, Burke-Spolaor et al. (2011a), Kocz et al. (2012), and 5.3. Results 75

Figure 5.2 RFI monitor spectra from Parkes for the perytons in the week starting 2015 January 19. The time of peryton has been indicated around the 2.3 2.5 GHz range by black arrows. 76 Chapter 5. Identifying the source of perytons

Figure 5.3 RFI monitor data from Parkes and the ATCA between 2.30 and 2.50 GHz around the times of the three January perytons and one peryton from the Woolshed microwave oven tests (2015-03-17).

Figure 5.4 Number of narrow-emission spikes detected with the RFI monitor with S/N > 10 in a 60 MHz window around 2.466 GHz between 18 January and 12 March, 2015. 5.3. Results 77

Bagchi et al. (2012) the total number of perytons is 46. The properties of these sources, especially how they relate to the population properties of FRBs is discussed in more detail in Section 5.5.

5.3.4 Generating perytons

With the recognition that peryton signals are likely to be associated with equipment emit- ting at 2.3 2.5 GHz, an effort was made to try to identify such equipment on site, and attempt to ‘create’ a peryton. As microwave ovens are known to emit in this frequency range and could potentially produce short-lived emission the site microwave ovens were the focus of our initial tests for reproducing peryton signals. There are three microwave ovens on site in close proximity to the telescope that expe- rience frequent use located in the tower below the telescope, in the visitors centre and in the staff kitchen in the building traditionally referred to as the Woolshed. There are two additional microwave ovens at the observer’s quarters approximately 1 km from the main site. The first tests occurred on 2015 February 27 during scheduled maintenance while the telescope was stowed at zenith. The BPSR system was turned on for all 13 beams and the three microwave ovens on-site were run on high and low power for durations of 10 – 60 s. In each test the load in the microwave oven was a ceramic mug full of water. In the first set of tests, a single peryton was detected during tests of the tower microwave oven with 3 a DM of 345 pc cm . The detection of radiation from the tower microwave oven would be very surprising as the tower is shielded on the windows and in the walls and the dish surface blocks the line of sight to the receiver in the cabin at the prime focus. However it was later determined that the Woolshed microwave oven was also in use at the time, unrelated to these tests, and might potentially have been the source of the peryton. The second set of tests were conducted on 2015 March 12, this time pointing the telescope at azimuth and elevation combinations where we often see perytons. From the 21 perytons discovered in the HTRU survey and the known pointing locations a broad estimate of the peryton rate as a function of azimuth and elevation can be calculated. For the HTRU perytons, the rate is highest at an azimuth and elevation of ( 130 ,65)and ⇠ when pointing near zenith. An initial test was conducted with the microwave ovens while pointing the telescope at these locations and no perytons were seen. The decisive test occurred on 2015 March 17 when the tests were repeated with the same microwave oven setup but instead of waiting for the microwave oven cycle to finish the microwave oven was stopped by opening the door. This test produced three bright perytons from the staff kitchen microwave oven, all at the exact times of opening the microwave oven 78 Chapter 5. Identifying the source of perytons

Figure 5.5 One of the bright perytons generated during the test on 2015 March 17 with 3 DM = 410.3 pc cm . The plot elements are the same as those in Figure 5.1. RFI monitor data at the time of this peryton are shown in Figure 5.3.

3 door, with DMs of 410.3, 410.3, and 399.6 pc cm (the first of these generated perytons is used in Figures 5.3 and 5.5). With knowledge that this mode of operation of a microwave oven could produce perytons, we examined the range of azimuths and elevations at which there was direct line of sight from the microwave oven to the multibeam receiver (i.e., the underside of the focus cabin). As is apparent in Figure 5.6, almost all the perytons with DMs > 300 occurred when there was visibility of the focus cabin from the Woolshed microwave oven. This left the smaller sample of perytons with lower DMs, which were, however, consistent with an origin at the visitors centre or the Quarters. (This sample also included all five events which had been detected on the weekend, when there were generally no staff on-site and the Woolshed was not in use.) Similar tests were performed with a previously installed microwave oven in the visitors centre and six perytons were seen at the times corresponding to opening the door; however, these perytons had DMs of 206.7, 3 204.9, 217.0, 259.2, 189.8, and 195.2 pc cm . This process does not generate a peryton every time, however; in fact perytons appear to be generated with an 50% success rate. ⇠

A bimodal distribution of peryton DMs can be accounted for from at least two mi- crowave ovens on-site being used and stopped in this manner. The detectability of pery- tons with a given DM from a microwave oven stopped this way depends on the direction in which the telescope is pointing. The receiver is sensitive to perytons when the microwave oven producing the bursts has a direct line of sight to the focus cabin and receiver of the 5.3. Results 79

Figure 5.6 Azimuth and elevation combinations for which there is a direct line of sight from the microwave oven in the Woolshed to the multibeam receiver are broadly shown with circles. The pointing directions for the detected perytons with DMs 400 pc cm 3 ⇠ (crosses) and 200 pc cm 3 (pluses) are also shown. ⇠ telescope, i.e., a line of sight not blocked by the surface of the telescope, yet still seeing the underside of the focus cabin. As shown in Figure 5.6, for the Woolshed (located 100m from the Dish at an azimuth of 65), the broadest range of elevations providing a direct line of sight are offset by 80 in azimuth. ⇠

5.3.5 The Peryton Cluster of 1998 June 23

Of the 46 perytons detected at Parkes since 1998 some 16, more than a third of the total, occurred within a period of just seven minutes, on 1998 June 23. All have a DM consistent with an origin in the Woolshed. Kocz et al. (2012) noted that the interval between consecutive events is clustered around 22-s. In this more complete sample, we find that indeed eight of the 15 intervals between consecutive events fall within the range 22.0 0.3-s, which is exceedingly unlikely to have been produced by manually opening the ± oven. Rather, we believe that the operator had selected a power level of less than 100%, causing the magnetron power to cycle on and off on a 22-s cycle, the period specified in the manufacturer’s service manual and confirmed by measurement. It appears likely that over this 7-min period the oven produced a peryton on all or most completions of this 22-s 80 Chapter 5. Identifying the source of perytons cycle but that the operator stopped the oven manually several times by opening the door, each time restarting the 22-s cycle. Kocz et al. also noted a clustering of event times modulo 2-s (their figure 2). This can be explained if the 22-s cycle is derived from a stable quartz crystal oscillator, which is almost certainly the case as the oven has a digital clock display. However, we have been unable to repeat the production of perytons in this manner. The principal difficulty is to account for the peryton energy escaping the oven’s shielded enclosure without opening the door. A transitory fault condition seems an unlikely possi- bility, given the oven has continued to operate reliably for a further 17 years. We conjecture that on this occasion the operator inadvertently compromised the shielding by placing con- ducting material in the oven, perhaps aluminium cooking foil that became caught between the door and the body of the oven, creating a unintended antenna, but we have yet to devise an acceptable test of this scenario.

5.4 Discussion

The two ovens responsible for most or all of the observed perytons are from the same manufacturer (Matsushita/National) and are both in excess of 27 years of age though still working reliably. Our tests point clearly to the magnetron itself as the source of the perytons since these are not detected unless the oven door is opened. Further, our analysis of the peryton cluster of 1998 June 23 implies the perytons are a transient phenomenon that occurs only when the magnetron is switched off. That we have observed perytons from at least two ovens over 17 years suggests that they are not the product of an unusual failure or fault but are inherent to, and long-lived in, at least some common types of oven. The magnetron used in the Woolshed oven (type 2M210-M1) was used by Matsushita in new microwave ovens for at least a decade and remains readily available. However, the physical process that generates the swept or ‘chirped’ emission that defines these perytons is obscure. The duration of the perytons is also a puzzle. The Woolshed oven has a simple HV supply comprising a 2kVAC mains step-up transformer and Villard voltage doubler/rectifier, with no additional filtering. The magnetron supply voltage should decay rapidly after switch-off over a few mains cycles (of 20 ms) but the perytons have typical durations of 250 ms or more, decaying in power by only a factor of 3 or so over this time (e.g. figure 3 of Bagchi et al. 2012). By nature, magnetrons are highly non-linear devices and the mode competition occur- ring at the start-up and shut-down of the microwave oven can cause excitation within the magnetron. Magnetron cavities have several spacings through which electrons flow. Over 5.5. Relevance to FRBs 81 time the edges of these cavities may become worn down and arcing may occur across these cavities during start-up and shut-down. This arcing may produce a spark observable at other frequencies than those intended in the microwave oven specifications. The microwave oven itself should act as a Faraday cage and block these signals from exiting the microwave oven cavity. However, opening the door of the microwave oven during shut-down would allow for these signals to propagate externally. Escaping sparks at 1.4 GHz could be the perytons we see with the receiver (Anderson et al., 1979; Yamanaka & Shinozuka, 1995). Further tests on the site microwave ovens over a wider range of frequencies will be performed in near future and reported in an upcoming publication. More extensive testing is expected to provide greater insight into the emission mechanism of the sweep seen at Parkes observing frequencies as well as determining the conditions such as microwave oven power setting, contents, and door configuration that are responsible for the observed perytons.

5.5 Relevance to FRBs

5.5.1 Differences in observed properties

Having originally cast doubt on the first FRB discovered, FRB 010724, the origin of pery- tons has since cast a shadow on the interpretation of FRBs as genuine astrophysical pulses. We therefore wish to explicitly address whether perytons and FRBs could have a common origin. Even with the source of perytons identified as on-site RFI the question may remain as to whether the progenitors of FRBs and perytons are related or even the same event at different distances. Fundamental aspects of the FRB and peryton populations differ. The distribution of perytons in time-of-day occurrence and DM is highly clustered and very strongly indicative of a human-generated signal. The DM and time-of-day detections of perytons and FRBs are compared in Figure 5.7. In the case of the perytons, the clustering around the lunchtime hour becomes even more pronounced once an AEDT (Australian Eastern Daylight Time) correction is applied. The FRB distribution in time of day is consistent with a random distribution, which would be observed as essentially flat perhaps with a slight dip in number during office hours where occasional telescope maintenance is performed. Similarly, the bimodal DM distribution of the peryton population can be clearly seen in the larger peryton sample. No clear DM clustering can yet be identified for the FRBs although such a distribution may become clear with a population of thousands of sources if FRBs are cosmological (McQuinn, 2014; Macquart et al., 2015). Finally, a microwave 82 Chapter 5. Identifying the source of perytons oven origin is generally not well suited to explaining other observed properties of FRBs, such as the clear asymmetric scattering tails observed in some FRBs, the consistency with Komolgorov scattering (Thornton et al., 2013), and the apparent deficit of detections at low Galactic latitudes. These are major indicators of a genuine astrophysical population (Chapter 4; Burke-Spolaor & Bannister, 2014).

5.5.2 What is FRB 010724?

With an understanding of the conditions under which perytons are generated, we can reconsider the ‘Lorimer Burst’, FRB 010724 (Lorimer et al., 2007). As noted by Burke- 3 Spolaor et al. (2011a) and as is evident in Figure 5.7, the DM of 375 pc cm for this burst is entirely consistent with the DM 400 pc cm 3 events we now refer to as Woolshed ⇠ perytons. However, there are critical differences. The bright detection in three beams is indicative of a boresight detection. Furthermore, the event occurred with the telescope pointing almost due south, and the line of sight from the Woolshed microwave oven to the focus cabin is completely blocked by the telescope surface. While there is line of sight visibility from the visitors centre at this time, the DM is not consistent with the visitors centre microwave oven. Additionally, the event occurred at 19:50 UT, i.e., 5:50 am AEST (Australian Eastern Standard Time), when the visitors centre is closed and unstaffed. We conclude the evidence in favour of FRB 010724 being a genuine FRB is strong.

5.5.3 Deciphering new transient events

To discern between new millisecond transient detections, this work has demonstrated two critical discriminants that divide FRBs and perytons. A common, known RFI emission from microwave ovens—as detected concurrently to all perytons presented here—is at 2.3 2.5 GHz. Thus, an FRB detected with a non-detection of any 2.3 2.5 GHz, which we propose as a key characteristic of the Parkes perytons, would be another nail in the coffin for any association. It should be noted that while there are 2.3 2.5 GHz events with no L-band detection, there are not the converse, so there is some statistical probability that a 2.3 2.5 GHz spike occurs by chance around the same time as an FRB, particularly if it is detected during daytime (Figure 5.4). Second, as with FRB 010724, given that the telescope cannot point directly at a mi- crowave oven, fabricating a detection that does not appear in all beams, our results show that perytons can be discerned from FRBs by using a multibeam system to identify sky- localized events. For an event to appear point-like within the multibeam receiver’s beam pattern, as FRBs do, the target must be in the Fraunhofer regime. 5.5. Relevance to FRBs 83

25

20

15

Number 10

5

0 0 5 10 15 20 Time of day AEST hrs

25 H L 20

15 Number 10

5

0 0 5 10 15 20 Local Time hrs 25

H L 20

15

Number 10

5

0 500 1000 1500 2000 DM pc cm-3

Figure 5.7 The overlaid FRB (blue) and perytonH (red)L distributions as a function of time in AEST (top), the local time with AEDT accounted for (middle) and as a function of DM over the entire range searched (bottom). Clearly, the FRB distribution is uniform throughout the day, whereas the peryton signals peak strongly during office hours (particularly around lunch time). A random distribution would look approximately flat, with a slight dip during office hours where occasional maintenance is carried out. The bimodal peryton distribution with peaks at 200 and 400 pc cm 3 is evident. ⇠ ⇠ 84 Chapter 5. Identifying the source of perytons

5.6 Conclusions

Three peryton detections were made at the Parkes radio telescope on three separate days during the week of 2015 January 19. The installation of a new broad-band RFI moni- tor allowed for the first correlation between the peryton events and strong out of band emission at 2.3–2.5 GHz of local origin. Additional tests at Parkes revealed that peryton events can be generated under the right set of conditions with on-site microwave ovens and the behaviour of multiple microwave ovens on site can account for the bimodal DM distribution of the known perytons. Peryton searches in archival survey data also allowed for the detection of a further 21 bursts from the HTRU survey alone. A comparison of the population properties of FRBs and perytons revealed several critical conclusions as follows.

Perytons are strongly clustered in DM and time of day, strongly indicative of man- • made origins, whereas FRBs are not.

FRB detections to date faithfully follow cold plasma dispersion; some have shown • clear scattering tails whose frequency-dependent width follows a Kolmogorov spec- trum; FRBs appear to avoid the Galactic plane. Perytons do not exhibit these properties.

The peryton-causing ovens on the Parkes site could not have produced FRB 010724, • indicating that this burst is in fact an FRB rather than a peryton.

A direct test of ‘peryton versus FRB’ can be made via the detection or non-detection, • respectively, of concurrent 2.3–2.5 GHz emission.

We have thus demonstrated through strong evidence that perytons and FRBs arise from disparate origins. There is furthermore strong evidence that FRBs are in fact of astronom- ical origin. 6 Discovery and follow-up of FRBs in the HTRU high latitude survey

In this chapter we present 5 new FRBs discovered while reprocessing the HTRU high latitude survey with the heimdall search code. This search overlapped with the previously published search by Thornton et al. (2013) and found no new bursts in the previously processed region. The total of nine FRBs from the high latitude survey can be used to update the observed FRB rate and to infer further properties of the greater population. Follow-up observations of 8 of the 9 FRB fields was performed to search for repeating signals from FRB progenitors. No repetition was detected which provided strong limits on short-timescale periodic emission from a progenitor and weaker limits on progenitors repeating on longer timescales.

6.1 Introduction

The main objective of recent radio surveys has been to search for repeating transient phenomena such as pulsars and rotating radio transients (RRATs). The millisecond to sub-millisecond sampling times used for these surveys, combined with their large time on sky also made them excellent datasets in which to find FRBs. Prior to the work in this thesis almost all FRBs were found in high time resolution radio pulsar surveys that covered large swathes of sky in search of transient phenomena (Lorimer et al., 2007; Thornton et al., 2013; Spitler et al., 2014; Burke-Spolaor & Bannister, 2014). The most successful of these surveys in terms of FRB yield has been the high Galactic latitude component of the High Time Resolution Universe survey (HTRU; Keith et al. 2010), which found four FRBs in 24% of the survey (Thornton et al., 2013). More FRBs were expected in the remaining survey data. However, no FRBs were found in the interme-

85 86 Chapter 6. Discovery and follow-up of FRBs at high latitudes diate latitude portion of the survey (Chapter 4). Since most of the FRBs were discovered years after occurrence no systematic follow-up was undertaken as part of the HTRU sur- vey, which was completed in February 2014. This has changed with the advent of real-time detections such as that of FRB 131104, which was discovered in a targeted search of the dwarf spheroidal galaxy and was observed repeatedly for a total of 78 hours in the year following detection with no further FRBs detected (Ravi et al., 2015). Follow-up of FRBs on both short and long timescales is essential as it becomes increas- ingly important to solve the mystery of their origins. Although the true progenitors of FRBs are unknown, a cosmological origin is highly favoured (Chapter 1; Deng & Zhang, 2014; Luan & Goldreich, 2014) including magnetar flares, supergiant pulses from neutron stars, and pulsar-planet systems at cosmological distance (Mottez & Zarka, 2014). In the blitzar model an FRB is generated in a cataclysmic event and no repeat FRB emission is predicted. However, the magnetar flare, supergiant pulse, and pulsar-planet theories make specific predictions about FRBs as a repeating source on different timescales. In the pulsar-planet model, the beamed radio emission is produced in the Alfvén wings of a planet closely orbiting a pulsar. Such emission is constant but is observable along our line of sight for only 1 ms as the planet moves through its orbit. In such a case, the ⇠ FRB would be a repeating event, occurring once per orbit. Ultra-light companion systems, such as those with a planetary companion, detected in our own Galaxy through pulsar timing have periods ranging from 1.56 hours to over 70 days, with a median period of 4 hours (Manchester et al., 2005). In such a scenario, recurring FRB events would be best observed through continuous monitoring of the field of a known FRB. No such emission was detected in the over 40 hours of follow-up conducted by Lorimer et al. (2007) for FRB 010724 or in the extensive search by Ravi et al. (2015) for FRB 131104. Energetic magnetar flares are more commonly detected through their X-ray emission and there are 28 known magnetars in our own Galaxy, all of which have been found in X-ray searches except for a radio-loud magnetar that has not yet been associated with an X-ray outburst (Levin et al., 2012). Of these 28 sources (of which only 4 are visible at radio frequencies) approximately 23 have been seen to burst and 3 have documented giant flares1 (Olausen & Kaspi, 2014). If FRBs are produced in giant radio flares evenly distributed throughout the lifetime of sources from a similar population in distant galaxies such repe- tition might be untestable on human timescales. However, if the FRBs are associated with a period of heightened outbursting, long-term campaigns to re-observe the fields of known FRB fields over a period of several years or even several decades would be best-suited to

1From the online catalogue http://www.physics.mcgill.ca/ pulsar/magnetar/main.html ⇠ 6.2. Search of the HTRU high latitude survey 87

Table 6.1 Repeating progenitor models for FRBs and their timescales estimated by Mottez & Zarka (2014), Turolla et al. (2015), and Cordes & Wasserman (2015) for pulsar-planet, magnetar giant flare, and supergiant pulse progenitors, respectively. Timescales for mag- netar giant flares and supergiant pulsar pulses are given in terms of the number of events over the lifetime of a single source, and as the time between events assuming they are equally distributed throughout the progenitor’s life.

Progenitor Model Timescale Pulsar-planet Highly periodic FRB from Alfvén 1.5 hrs

6.2 Search of the HTRU high latitude survey

Following the promising results of the early searches of the HTRU high latitude survey (Thornton et al., 2013) a full FRB search of the dataset upon survey completion was expected to yield many more new bursts. We performed a thorough search of 90% of the high latitude survey — 30,000 of the approximately 33,500 pointings, each of length 88 Chapter 6. Discovery and follow-up of FRBs at high latitudes

270 seconds. All pointings not processed by Thornton et al. (2013) were included in this analysis. The offline heimdall pipeline search for FRBs described in Chapters 2 and 4 was applied to this dataset and each pointing was searched for FRBs over widths ranging 3 from 0.128 to 262 ms, and over 1749 DM trials from 0 to 5000 pc cm . A fraction of the pointings previously searched by Thornton et al. (2013) were also re-analysed using this method. The search thresholds for the FRB search were identical to those described in Equation 4.1 for the intermediate latitude survey. This was done to maintain consistency across the two components. This processing of the high latitude survey returned five fast radio burst discoveries not included in previous HTRU publication from 2009, 2012, and 2013 — FRBs 090625, 121002, 130626, 130628, and 130729. The discovery of FRB 121002 has been previously discussed in Thornton (2013). The four FRBs previously published by Thornton et al. (2013) were also recovered in this search, some with higher than previously reported signal-to-noise due to the correction in heimdall of the ‘root 2 problem’ (Section 2.2). No new discoveries were made in the previously searched survey region. The coordinates and observed properties of all bursts from the high latitude survey are given in Table 6.2 and the pulse profiles of the new bursts are presented in Figure 6.1.

6.2.1 Burst properties

From the observed properties listed in Table 6.2 and using the equations provided in Section 1.3 we can derive redshifts, distances, and energies for the new bursts which are presented in Table 6.3 alongside the previously published bursts from this survey. The expanded high latitude sample includes bursts with interesting new properties. 3 FRB 121002 has the highest DM, 1629(2) pc cm , of any FRB discovered to date and also exhibits a distinct double peak in the dedispersed profile. The redshift upper limit for this burst given its DM is z<1.3 meaning the observed emission at 1.4 GHz may have been emitted at a much higher frequency ( 3.2 GHz) relative to the other known bursts ⇠ and the double peaked structure may be a feature of the emission at these frequencies. Although the cause of the multi-component profile of FRB 121002 remains unknown, they may be indicative of internal structure in the emission region as is seen for long and short GRBs (Piran, 1999). Loeb et al. (2014) and Maoz et al. (2015) have argued that FRBs originate as coherent bursts of radio emission from Galactic flare stars meaning that the majority of ionisied material is local to the progenitor. In the case of FRB 121002, 1555 pc cm 3 worth of ⇠ ionised material would be required along the line of sight in the stellar corona to account 6.2. Search of the HTRU high latitude survey 89 .All +3 . 8 1 . 7 ⇤ +2 . 3 0 . 77 +1 . 4 0 . 48 +0 . 47 0 . 36 +1 . 5 0 . 97 +0 . 6 0 . 3 +2 . 4 1 . 7 +1 . 3 0 . 4 +2 . 3 1 . obs 3 9 8 8 F 79 30 32 22 36 7 . 0 . 1 . 0 . 3 . 2 . 1 . 1 . 2 . , obs +0 . 22 0 . 14 +1 . 12 0 . 10 +0 . 20 0 . 13 +0 . 21 0 . 10 +0 . 13 0 . 10 +0 . 08 0 . 06 +0 . 15 0 . 11 +0 . 30 0 . 22 +0 . 06 0 . 04 92 11 63 45 62 43 74 91 20 peak S 0 . 1 . 0 . 0 . 0 . 0 . 0 . 1 . 0 . +9 . 5 7 . 6 +0 . 04 0 . 01 +1 . 1 0 . 8 +2 . 3 2 . +0 . 9 1 . 5 +0 . 6 4 . 2 +0 . 5 0 . 3 +3 . 2 2 . 5 6 2(1) 5 6 7 6 4 5 joules) 18 0 . 1 . 3 . 3 . 2 . 9 . 0 . 14 . 32 0 . Energy < < < < < < < < 10 +9 . 9 6 . 3 < (2015). obs +0 . 13 0 . 13 +0 . 8 0 . 7 +1 . 3 1 . 0 +1 . 2 0 . 4 +2 . 2 1 . 9 +0 . 6 0 . 3 +3 . 5 1 . +1 . 2 0 . 44 8 4 9 2 6 3 5 8 W 64 1 . 3 . 1 . 6 . 5 . 2 . 1 . 18 . 0 . +0 . 2 0 . 1 +0 . 1 0 . 2 +0 . 1 0 . 2 +0 . 3 0 . 2 L 6(2) 8(1) 4(1) 2(2) 5(1) 2 8 3 8 D 4 . 5 . 2 . 4 . 4 . 9 . 4 . 3 . 1 . < < < < < < < < < )(ms)(Jy)(Jyms) 3 +0 . 06 0 . 03 +0 . 03 0 . 06 +0 . 07 0 . 06 +0 . 05 0 . 03 +0 . 07 0 . 03 +0 . 08 0 . 07 +0 . 03 0 . 09 cm 03(6) 65(8) 10 38 60 71 07 67 50 comov 4 . 2 . (Gpc) (Gpc) ( 2 . 1 . 2 . 2 . 3 . 1 . 2 . D < < < < < < < < < +0 . 01 0 . 02 +0 . 01 0 . 02 +0 . 02 0 . 01 +0 . 02 0 . 01 +0 . 02 0 . 01 56(2) 35(2) 30(3) 74(3) 2 76 89 43 69 model 0 . 0 . 1 . 0 . 0 . z 0 . 0 . 0 . 0 . < < < < < < < < < Galaxy 9.0 21.9 14.2 27.5 28.4 27.2 15.2 34.1 17.4 DM/DM ) 3 (hh:mm:ss) (hh:mm:ss) Beam ID (pc Galaxy cm 74 67 53 31 32 (pc ⇤ ⇤ ⇤ ⇤ ⇤ Date and Time RA Dec Parkes DM S/N 2012-10-02 13:08:362013-06-26 14:53:53 18:14:472013-06-28 03:56:53 16:27:06 -85:11:532013-07-29 09:01:05 09:03:02 -07:27:48 13:41:21 12 03:26:16 -05:59:43 1 1629(2) 5 10 16 952(2) 471(2) 20 853(2) 29 14 2009-06-25 21:50:31 03:07:47 -29:55:35 6 899(2) 28 FRB 121002 FRB 130626 FRB 130628 FRB 130729 EventFRB 090625 DM FRB 110220FRB 110626FRB 110703 35 FRB 120127 48 32 32 ⇤ ⇤ ⇤ ⇤ ⇤ FRB 110626FRB 110703 2011-06-26 21:31:05FRB 120127 2011-07-03 18:57:42 21:03:43FRB 121002 2012-01-27 08:08:41 23:30:51FRB -44:44:19 130626 23:15:06FRB -02:52:24 130628 FRB -18:25:37 12 130729 5 723.0(3) 4 1103.6(7) 12 17 553.3(3) 13 Event FRB 090625 FRB 110220 2011-02-20 01:52:19 22:34:38 -12:33:44 3 944.38(5) 54 . Values for the 4 published bursts from Thornton et al. (2013) are duplicated from Keane &ff Petro Table 6.3 The derived cosmological parameters for the 9 HTRU FRBs. Sources presented in this work are indicated with Table 6.2 Observed properties⇤ of the 9 FRBs from the HTRU high latitude survey. Sources presented in this work are indicated with values for redshift, comovingthat distance, all excess luminosity DM distance,(2015). is and attributed energy to presented the intergalactic here medium. are Values upper for limits the based published bursts on are the duplicated assumption from Keane &ff Petro 90 Chapter 6. Discovery and follow-up of FRBs at high latitudes

Figure 6.1 Pulse profiles for the 5 new bursts from the HTRU high latitude survey. FRB 121002 has a double peaked profile and two peaks may also be present in FRB 130729. The timescale at the top of the plot corresponds to the x-axis scaling for the first four bursts and the timescale at the bottom correponds to the final burst which, due to its large width, requires a different scale for the pulse shape to be seen properly. 6.2. Search of the HTRU high latitude survey 91 for the observed DM. However, Tuntsov (2014) argues that such dense environments would cause deviations in the cold plasma t ⌫ 2 relation and would be detectable by careful / fitting. Such variations are not seen for this burst despite the significant detection across the observing band. An additional curiosity is FRB 130729, which has a very wide pulse profile compared to the rest of the sample. This burst was detected only in the lower half of the observing band and in one of the outer beams of the receiver. It may have been detected far off axis or in one of the beam sidelobes as the beam sensitivity drops off much more quickly for higher frequencies away from beam center. This would make the true brightness much greater and the true properties of the burst (including its approximate position) very difficult to determine. This burst also appears to have a possible double pulse feature, however the lack of signal across the full band makes this difficult to determine reliably. The nine bursts from this survey represent the largest consistent sample currently 3 available. They range in DM from 471 to 1629 pc cm with widths ranging from 0.64 to 18.8 ms. Interestingly, only 4 of the 9 bursts (FRBs 090625, 110220, 121002, and 130626) have visible scattering tails. The presence of scattering does not appear to be correlated with DM, pulse width, S/N, or detection beam. No clear consensus has emerged as to the location of the scattering screen between the source and observer that could cause the observed scattering in the population of FRBs (Macquart & Koay, 2013; Luan & Goldreich, 2014). The random presence of scattering in the pulse profiles suggests that the scattering may be caused by propagation through an intervening galactic halo on the path between the host and the observer, which would explain why some bursts appear to be scattered while others remain unresolved. This property may be correlated with DM for a large enough sample of FRBs as more distant bursts would have a higher probability of passing through an intervening galaxy.

6.2.2 Updated FRB rates and Latitude Distribution

With the complete sample of FRBs from the high latitude survey we are able to provide an updated FRB rate with the largest sample of FRBs to date. The HTRU high latitude survey consisted of 1427 deg2 hrs of observations in which 9 FRBs were detected. This +4.2 3 1 1 results in a rate of RFRB( 0.6Jyms)> 5.7 2.7 10 (95%) FRBs sky day . The F ⇥ inequality in the rate estimate arises due to the fluence incompleteness of FRB searches at Parkes below =2Jy ms (Keane & Petroff, 2015). The rate above the level of fluence F +3.2 3 1 1 completeness is now RFRB( 2Jyms)=2.5 1.6 10 (95%) FRBs sky day . F ⇥ Although this value is lower than those previously reported in (Thornton et al., 2013) and 92 Chapter 6. Discovery and follow-up of FRBs at high latitudes

(Spitler et al., 2014), it is consistent with the 1- uncertainties given in those works. In Chapter 4 the reported rates at high latitudes and non-detection at intermediate latitudes were shown to be inconsistent with an isotropic FRB distribution with 99% con- fidence. The discrepancy between high and intermediate latitude results may be reconciled using a lower overall FRB rate or a latitude-dependent rate. Using our updated rate from the full high latitude survey the probability of finding 0 FRBs at intermediate latitudes assuming an isotropic distribution using Equation 4.3 becomes 2.5%. Thus the FRB dis- coveries of the HTRU high and intermediate latitude surveys are inconsistent with an isotropic FRB distribution/rate with 97.5% confidence. No sufficient explanations yet exist for why the detections of an extragalactic population of sources should be so highly dependent on Galactic latitude. Galactic effects such as scattering as currently modeled are insufficient to account for obscuration in the Galactic plane (Chapter 4). However, with the exception of FRB 121102 which was discovered near the Galactic anticentre no published FRBs have been found at a Galactic latitudes below b < 20 . Macquart & Johnston (2015) have proposed that the Galactic latitude | | dependence is caused by the amplification of FRB signals in the Galactic halo making the rate appear artificially high. A larger population is needed to verify this hypothesis.

6.3 Parkes follow-up

Eight of the nine bursts discovered in HTRU were were surveyed over the course of 110 hours between April and October 2014. FRB 130729 was excluded from follow-up ob- servations due to the highly uncertain position (Section 6.2.1). The observations that were undertaken are listed in Table 6.4. They were scheduled to allow for approximately monthly follow-up of each FRB field for 1 2 hours each. Each field was observed between 3 and 5 times in total over the 6 month period. The FRB positions were taken to be the coordinates of the beam centre from the discovery pointings, which are listed in Table 6.2, however the true coordinates of each FRB are unknown due to the large uncertainty in the location of the source within the Parkes beam. For this survey we performed observations in a 90 offset grid around the beam centre positions as outlined in Morris et al. (2002) with 15-minute duration pointings. In this way we sampled the entire FRB field in each observing session. In the second observing session for this project a new FRB was discovered in a grid pointing 90 offset from the field of FRB 110220 (FRB 140514, presented in Chapter 7). Systematic follow-up of this source was absorbed into the survey with minimal gridding and longer total observing times per session. 6.4. Data processing 93

Additionally, we performed a focused search of the field of FRB 090625 for short-term repeatability by observing for 2.5 8.6 hours per day over 5 days closely spaced. In these observations the pointing location was fixed at the beam centre position. FRB 090625 was observed on October 21, 23, 28, 29, and 30 for all available time while the source was above the horizon. These observation dates and total observing times are also listed in Table 6.4.

6.4 Data processing

The observing mode for this survey is described in detail in Chapter 7 where the new FRB discovery that resulted from this survey is presented. The observing system, based on the Berkeley Parkes Swinburne Recorder (BPSR) and the HI-Pulsar Signal Processor (HIPSR), incorporates two major upgrades that have become available since the HTRU sur- vey, namely the real-time processing system and the ability to record 8-bit full-polarisation data from the linear feeds in the event of an FRB discovery (Chapter 2). As outlined in Chapter 2 the real-time transient pipeline on BPSR uses the heimdall single pulse search software to search for burst-like signals in the data while they are stored in the 120-s ring buffer on HIPSR. The data are searched over a range of pulse widths (0.128 262 ms) and dedispersion trials (0 2000 pc cm 3) for candidates that satisfy the criteria in Equation 2.6. Once an FRB matching these criteria is identified, the 8-bit full-polarisation data around the time of the pulse are extracted from the buffer and saved to disk for all 13 beams. With this system it is now possible to record and recover the full-Stokes signal of a fast radio burst. The real-time burst search was performed at the time of observation for all data recorded during this survey. The real-time pipeline, which now operates on all data recorded with BPSR, is the first of two stages of processing to search the survey data for dispersed radio pulses. After the data are recorded at Parkes they are transfered to the Swinburne gSTAR supercomputing cluster via fibre link and stored on the supercomputer file system. The data are then processed again for potential pulses using a more thorough pipeline that does not run in real time. In this processing stage the data are searched using heimdall from 0.128 262 ms in pulse width, and from 0 5000 pc cm 3 over 9420 DM trials using a DM tolerance of 1.01 to avoid sensitivity loss due to poor trial spacing (Keane & Petroff, 2015). A more thorough cleaning process is also performed to remove radio frequency interference in both frequency and time (Kocz et al., 2012). Candidates satisfying the following criteria were flagged and inspected: 94 Chapter 6. Discovery and follow-up of FRBs at high latitudes

Table 6.4 The total hours observed for each FRB in this campaign, including the additional source FRB 140514, which was discovered 2 months into the survey and was then observed in place of FRB 110220 for the remainder of the survey. The total time spent at the beam centre position from Table 6.2 and in gridding around the field are also listed. FRB name Observing date Beam Centre Gridding Duration Total duration (UT) (hours) (hours) (hours) (hours) FRB 090625 2014-05-14 0 0.75 0.75 33.65 .2014-06-24/250.250.751 .2014-08-191.512.5 .2014-10-212.5502.55 .2014-10-235.8605.86 .2014-10-287.807.8 .2014-10-298.6608.66 .2014-10-304.504.5 FRB 110220 2014-05-14 0.75 1 1.75 1.75 FRB 110626 2014-04-21 0.5 2 2.5 11.25 .2014-05-140.7511.75 .2014-06-240.522.5 .2014-07-270.7511.75 .2014-08-190.7522.75 FRB 110703 2014-04-21 0.25 0.75 1 10.1 .2014-05-140.7511.75 .2014-07-271.512.5 .2014-08-191.512.5 .2014-09-300.51.251.75 .2014-10-300.600.6 FRB 120127 2014-05-14 0.75 1 1.75 5.5 .2014-06-240 0.50.5 .2014-08-190.7511.75 .2014-09-300.511.5 FRB 121002 2014-04-21 1 2 3 10.25 .2014-05-140.7511.75 .2014-06-240.7511.75 .2014-07-271 12 .2014-08-190.7511.75 FRB 130626 2014-04-21 1 2 3 9.5 .2014-05-140.7511.75 .2014-06-240.511.5 .2014-07-270.7511.75 .2014-08-190.511.5 FRB 130628 2014-04-21 0.75 2 2.75 9 .2014-05-140.7511.75 .2014-06-240.250.751 .2014-07-271 12 .2014-08-190.511.5 FRB 140514 2014-05-14 1.2 0 1.2 19.2 .2014-06-247.907.9 .2014-07-273.503.5 .2014-08-191 01 .2014-09-301.801.8 .2014-10-293.803.8 6.5. Follow-up Results 95

3 DM 5pccm S/N 8 (6.1) N 4 beams  t 8.192 ms.  Detection of a repeated FRB was defined as any single pulse identified in the field with a DM within 10% of the original FRB detection. Over timescales of several years, relative variations in dispersion measure greater than 10% would occur only if a large fraction of the ionised material was local to the progenitor; such variations are orders of magnitude greater than those observed from interstellar turbulence (Keith et al., 2013; Chapter 3). Physical constraints on dense environments around the FRB progenitor have been proposed by Luan & Goldreich (2014) that make such large variations in DM for FRB progenitors unlikely. However, the full range of Galactic and extragalactic DMs were searched to look not only for repeated FRB pulses, but also for any yet undiscovered pulsars or rotating radio transients that may lie in the survey fields.

6.5 Follow-up Results

Only one significant dispersed pulse was detected in the 110 hours of observations in the real-time pipeline: the new FRB 140514. This burst was detected in beam 1 of the telescope centred 9 arcmin from the beam centre position of FRB 110220 with RA 22:34:06 DEC -12:18:46, and was identified in the real-time pipeline with a S/N of 16. The DM of FRB 3 3 140514 was found to be 562.7(6) pc cm while that of FRB 110220 was 944.38(5) pc cm (Thornton et al., 2013). All other events could be classified as radiometer noise or RFI. The DM of FRB 140514 was 40% less than the DM of FRB 110220 and thus was judged to be a separate event based on the criteria in Section 6.4. However, this conclusion assumes that a single progenitor cannot produce two bursts of different DM (even when separated in time by several years). It is highly unlikely that the bulk of the line-of-sight electron column density could have changed so substantially (Luan & Goldreich, 2014; Tuntsov, 2014). It remains the case that the progenitor itself could be enshrouded in ionised material that varied significantly in density over time causing a large DM change. This would then place the source at a much smaller distance, perhaps in the local Universe (and with much lower total energies) as has been proposed by Pen & Connor (2015) and Connor et al. (2015). The discovery of FRB 140514 offered an unprecedented opportunity for immediate and sustained follow-up of an FRB field during the weeks and months after the event.The field 96 Chapter 6. Discovery and follow-up of FRBs at high latitudes of the new FRB was observed during all subsequent observing sessions including an 8-hour track on 24 June 2014 for the entire time the field was observable from Parkes. In total the field of FRB 140514 was observed for 19.2 hours with Parkes in the 5 months after the observed radio burst; the burst was also followed-up with telescopes at other (see Chapter 7 for details).

6.6 Discussion

In the following subsections we will discuss the implications of the non-detection on FRB progenitor models in the context of total observing time, the multi-day monitoring of FRB 090625, and the prompt follow-up in the months following FRB 140514.

6.6.1 Total time

The total observing time of 110 hours spaced roughly monthly over a 6-month period is insufficient to place substantial limits on infrequently occurring flares from bursting sources such as magnetars or supergiant pulses from extragalactic neutron stars, as noted in Cordes & Wasserman (2015). Our strongest limits on repetition from a single source during our observations comes from FRB 090625 which was observed for a total of 33.65 hours with no detected pulsed emission. To place stronger constraints on these types of events would require hundreds of hours of monitoring over multiple years. The anticipated timescale between magnetar giant flares (. 1 kyr; Olausen & Kaspi, 2014) and repetition timescales of . 1 Myr for supergiant pulses (Cordes & Wasserman, 2015) makes the probability of catching repeats extremely low. Ultimately stronger limits on long-term repeatability will come from wide-field radio telescopes capable of monitoring these fields as part of routine sky surveys. Dedicated time on telescopes with small field of view, such as Parkes, is difficult to justify given the amount of time needed and the inefficiency of any such search. Systematic follow- up of FRB discoveries made in future surveys and with future instruments will also be necessary to monitor these source fields in the months and years after a detection. Of course, detection of the afterglow from a cataclysmic event that produced an FRB would also resolve the question of FRB repeatability.

6.6.2 Multi-day observations of FRB 090625

Mottez & Zarka (2014) have predicted that a planet within the pulsar wind could produce a strictly periodic FRB signal that repeats as the period of the planetary orbit. To place 6.6. Discussion 97

Table 6.5 Summary of observations for a multi-day campaign in the field of FRB 090625

UTC start Tobs (hours) 2014-10-21 14:49:34 2.55 2014-10-23 10:49:04 5.86 2014-10-28 10:52:46 7.8 2014-10-29 10:06:00 8.6 2014-10-30 10:53:53 4.5 constraints on repetition on short timescales, we undertook a multi-day observing campaign for a single FRB. FRB 090625 was chosen for this additional observing as it was above the horizon for all time slots available to the project. The observations were conducted over five nights: 2014 October 21, 23, and 28 30. The total time spent on the source was 29.4 hours, and two exceptionally long tracks of 7.8 and 8.6 hours were recorded on October 28 and 29, respectively (Table 6.5). With these observations we can rule out a repeating progenitor system with a period P of less than 8.6 hours, the longest continuous observation in the campaign assuming that any repeat emission is above the flux limit of a Parkes beam, S & 0.5 Jy. Due to the spacing of our observations we can also rule out repeating progenitors with periods 8.6

21 our probability of detecting repeat emission, assuming the source emits a pulse on every rotation, decreases as 1/P with the exception of some poor sensitivity to certain periods due to observation spacing (Figure 6.2). Limits on a periodic repeating progenitor can be similarly placed for each source mon- itored in this campaign. The longest continuous observation tobs,max of a single source places a hard limit on repetition periods P t and a 90% confidence limit on peri-  obs,max ods P 2 t after which sensitivity decreases as approximately 1/P , as in the case . ⇥ obs,max of FRB 090625.

6.6.3 Follow-up of FRB 140514

Before the advent of real-time transient detection it was not possible to monitor the field of an FRB in the days and weeks after it occurred for pulses that might be associated with an active period of flaring or relaxation to a rest-state or for multi-wavelength emission associated with the radio transient. Such observations would give valuable clues about the events producing the bursts. The immediate discoveries of FRB 131104 (Ravi et al., 2015) and FRB 140514 enabled rapid follow-up on a timescale never before available. Observations of the field of FRB 140514 were performed 7 hours, 41 days, 74 days, 97 98 Chapter 6. Discovery and follow-up of FRBs at high latitudes

Figure 6.2 Probability of detection for repeating progenitors with a repetition period P in the 5 day campaign for FRB 090625. Sources with periods less than our longest observation (P<8.6 hrs, dot-dashed line) are ruled out. Periods P<21 hours are also ruled out with 90% confidence. At P = 21 hours the probability of detection drops off as 1/P (red dashed line). This limit assumes that FRBs are strictly periodic. days, 138 days, and 168 days after the event in which no repeat emission was detected. The longest observation conducted in this survey was undertaken 41 days after the event and consisted of a continuous 7.9 hour observation at the position of the discovery beam. The probability of detecting a new FRB in our observations, based on the total time on sky and given the Thornton et al. (2013) rate for an isotropic distribution of sources, is 33%. The revised, lower rate presented in Section 6.2.2 yields a 27% chance of detecting one new FRB in our survey, with a substantially higher probability (68%) of detecting no new bursts. Even with the lower rate, the probability of detecting a new event is still not negligible and we conclude that the detection of FRB 140514 in our survey is purely coincidental. Maoz et al. (2015) argue that FRB 140514 is a repeat burst from the progenitor of FRB 110220 with 99% confidence based upon the probability of detecting an FRB in the beam centred near the location of a prior event, which they determine to be 1%. The discrepancy between this probability and that derived above arises both from their use of only the central beam of the receiver in the time-on-sky calculation and from an additional correction factor to take into account that the burst was detected in the central beam and not detected in a beam not centred near the previous FRB. This correction factor of 0.85 6.7. Conclusions 99 is multiplied with the single beam probability. However, for the survey in which this burst was detected, all 13 beams were recording data and were searched for FRBs. Performing the same calculation with the updated rate presented in Section 6.2.2 and the probability function in Equation 4.3 with all 110 hours of the FRB follow-up survey we obtain a probability of detecting a new burst of 2.5%. Here we do not apply the 0.85 correction factor for two reasons: this term adds an additional bias on where one thinks an FRB should occur (a decision made after detection), and the authors do not explain how this factor of 0.85 was derived. Using this method, the probability that FRB 140514 originated from the same progenitor as FRB 110220 becomes 97.5%. However, ⇠ this method makes no use of the additional information obtained via the DM of the bursts. Based on the physical arguments outlined in Section 6.2.1 it is difficult to explain the large DM variations required for a single progenitor to produce both FRBs. Each of the probability calculations outlined in this section are correct for certain as- sumptions about the FRB progenitor(s); however, based on the large difference in DM between the two bursts we consider the probability of a new FRB somewhere in the multi- beam receiver during our survey, based on the all-sky FRB rate, to be a more plausible interpretation of the data.

6.7 Conclusions

We present the results of a search of the HTRU high latitude survey in which 5 new FRBs were discovered. From these bursts we are able to calculate an updated all-sky FRB rate assuming an isotropic distribution of bursts on the sky of R ( 0.6Jyms)> FRB F +4.2 3 1 1 +3.2 3 5.7 2.7 10 (95%) FRBs sky day ,orRFRB( 2Jyms)=2.5 1.6 10 (95%) FRBs ⇥ F ⇥ 1 1 sky day . This lower rate is still inconsistent with the results presented in Chapter 4 with 97.5% confidence. Two of the new FRBs show double peaked structure, possibly indicative of intrinsic structure in the FRB engine. These bursts were re-observed systematically to place limits on FRB repeatability. The total survey consisted of 110 hours over 6 months dedicated to re-observing the fields of 8 known sources. No repeat emission was detected from an FRB during this time placing weak limits on bursting or flaring sources; a more detailed and long-term study would be needed to rule out progenitors such as magnetar flares or supergiant pulses from extragalactic neutron stars. One component of this survey consisted of a multi-day campaign to observe a single FRB field and place limits on short-term repetition. From this sub-study we rule out repeating progenitors with periods less than 8.6 hours and place limits on repetition for periods between 8.6 and 21 hours at the 90% confidence level. 100 Chapter 6. Discovery and follow-up of FRBs at high latitudes

We are also able to constrain systems with greater orbital periods, making pulsar-planet systems unlikely progenitors for FRBs. In the course of this survey a new FRB was detected near the field of FRB 110220 and determined to be independent from the previous source due to difference in DM. Further effort is required to place strong limits on repetition of FRB sources. A dedicated monitoring campaign is not feasible using single dish telescopes (like Parkes, even with a feed) with a small field of view, and instead might be better-suited to wide field interferometric telescopes with high time resolution observing capabilities, such as UTMOST2, MeerKAT (Obrocka et al., 2015), or SKA1 (Macquart et al., 2015). Re- observation of an FRB in the days after a detection could provide valuable information about potential periods of high activity or relaxation experienced by the progenitor and would yield further insight into the origin of these bursts. Real-time detections of FRBs with future surveys should then be systematically followed up to search for such emission.

2http://astronomy.swin.edu.au/research/utmost 7 A real-time fast radio burst: polarization detection and multi-wavelength follow-up

In this chapter we present the real-time discovery of the fast radio burst FRB 140514 in the follow-up survey desribed in Chapter 6. Polarization information was recorded for FRB 140514 using the real-time detection system installed at Parkes. Multi-wavelength follow-up of the burst was also carried out at 12 telescopes to search for associated variable emission in the field. No counterpart was conclusively identified but observations placed the first limits on possible progenitors.

7.1 Introduction

A new class of objects called fast radio bursts (FRBs) have been discovered in radio pulsar surveys at Parkes and Arecibo within the last decade (Lorimer et al., 2007; Thornton et al., 2013; Spitler et al., 2014; Burke-Spolaor & Bannister, 2014). All FRBs discovered to date have been single radio events of millisecond duration. The electron column density, called the dispersion measure (DM), is also uncharacteristically high, leading to theories that they originate at cosmological distances (Thornton et al., 2013) and/or in extreme environments (Chapter 1). Recently, they have been the topic of considerable discussion, both as to their origins and their potential use as cosmological tools (Loeb et al., 2014; Kulkarni et al., 2014; Deng & Zhang, 2014; Gao et al., 2014). +0.6 4 1 Thornton et al. (2013) measure a rate of RFRB( 3Jyms) 1.0 0.5 10 sky F⇠ ⇠ ⇥ 1 day from 4 events found in a high Galactic latitude search, but the non-detection of FRBs in a survey twice as long suggests either a lower overall FRB rate or a latitude dependence, owing to a currently unknown obscuration effect at Galactic latitudes below b = 15 | | (Chapter 4). This result has recently been confirmed by Burke-Spolaor & Bannister (2014).

101 102 Chapter 7. A real-time fast radio burst

The true progenitors of FRBs remain unknown. All published FRBs were discovered in archival data years later and rapid follow-up of an FRB has never been possible. Recent efforts in time-domain radio astronomy have focused on real-time FRB detection with the promise of rapid follow-up of new events. Such capability was recently made possible with the development of a real-time transient pipeline at the Parkes telescope, discussed in Chapter 2. The new survey at Parkes described in Chapter 6 aimed to search the fields of previous FRB events for repeating bursts. The discovery of repeating FRB sources would strongly constrain emission mechanisms and possible progenitors. Here we report on the discovery of a new FRB in the field of FRB 110220 in this survey. In Section 7.2 we describe the real-time transient pipeline at Parkes using the multibeam receiver. In Section 7.3 we present the detection of FRB 140514 and, for the first time, the polarized radiation of an FRB. Section 7.4 details the follow-up efforts from X-ray to radio from 12 observatories. We summarize the results from these follow-ups in 7.5 and § discuss polarization in Section 7.5.1, connections between FRB 140514 and FRB 110220 in Section 7.5.2, and limits on an afterglow in Section 7.5.3. We provide a conclusion in Section 7.6.

7.2 Real-Time Transient Pipeline

Observations were conducted with the Berkeley Parkes Swinburne Recorder (BPSR) back- end for the 13-beam multibeam receiver (Staveley-Smith et al., 1996) at Parkes which covers 0.5 deg2 on the sky. We record 8-bit full-polarization data from two orthogonal linear feeds per beam, with 1024 frequency channels over 400 MHz of bandwidth, from 1182 1582 MHz, and 64-µs time resolution. Data are passed to the HI-Pulsar Signal Pro- cessor (HIPSR) where 120 s of observations are stored in a ring buffer using PSRDada1. The effective bandwidth for our data is 340 MHz from 1182 to 1522 MHz due to com- munications satellites operating in the 1525 to 1559 MHz band (Keith et al., 2010). The observing instrumentation is identical to that used for the High Time Resolution Universe (HTRU) survey and the FRB discoveries reported in Thornton et al. (2013). The real-time processing of the data for transient events is performed on the buffer using the heimdall single pulse processing software. The linear polarizations are summed into a single 8-bit data set and are passed to heimdall in 256 kilosample chunks (approximately 16.77 s). If a candidate is detected, the relevant 8-bit data are saved to disk. heimdall performs a search for pulses across a specified range of DMs and pulse widths and returns

1http://psrdada.sourceforge.net 7.3. Parkes real-time detection of FRB 140514 103 a list of candidates. The real-time transient pipeline searches 0 2000 cm 3 pc in DM and 0.128 262 ms in pulse width and identifies candidates that fit the criteria in Equation 2.6. For each candidate which meets all these criteria we also check that there is no known pulsar in the pulsar catalogue (Manchester et al., 2005) within a 5% DM range of the candidate for completeness, although this condition is typically precluded by the high-DM threshold. All known archival FRBs are identified using these criteria. When a candidate is detected the observable time span of the event is calculated using the total DM delay, t, across our observing bandwidth from Equation 1.2. The start time of the event is identified in the 120 s buffer and all samples in the range (tstart-t, tstart+2t) are saved to disk with full polarization 8-bit data. The BPSR real-time candidate detection and polarization triggering mode was com- missioned in March, 2014. Previously, it was impossible to obtain polarization data for FRBs at Parkes. Currently the triggers are configured to give a few false positives rather than miss a real event, and the real-time nature of the pipeline enables the trained observer to provide immediate feedback. The triggers have not yet been connected directly with other telescopes, but instead only initiate an email alert to observers related to the project when an event satisfying the above criteria is found.

7.3 Parkes real-time detection of FRB 140514

FRB 140514 was discovered on 14 May, 2014 at 17:14:11.06 UTC (15 May 03:14:11.06 local time) at 1.4 GHz in the centre beam (beam 1) of the multibeam receiver. It was identified 3 in the heimdall real-time transient pipeline with an S/N of 16, a DM of 562.7(6) pc cm , +3.5 and a pulse width of 2.8 0.7 ms. The pipeline identified the burst within 10 seconds and 2.22 seconds of data around the event were recorded to disk in 8-bit dual polarization for all 13 beams of the receiver. An FRB alert email was sent to project observers at 17:14:30 UTC. If the burst occurred at the beam-centre the detection corresponds to a peak flux +0.11 +2.3 density of 0.47 0.08 Jy and a fluence of 1.3 0.5 Jy ms. Further analysis of the FRB data resulted in a dispersion index ↵ = 2.000(4) such that t DM ⌫↵, in agreement with the / 2 ⌫ expected for cold plasma. The scattering time-scale was found to be ⌧1GHz =5.4(1) ms. There is a decreased uncertainty in the dispersion index and the scattering time-scale, as the scattering index was not a free parameter in the fit algorithm. While the scattering modelling done for these fits is consistent with a range of different pulse widths, leading to large error in width or W , the effect on the scattering tail, and thus ⌧1GHz, is negligible, giving a smaller error. 104 Chapter 7. A real-time fast radio burst

The DM, dispersion index and scattering time-scale were all fit for while the scattering index was held fixed at = 4. The limited S/N (16) of the pulse prohibited fitting for due to strong covariances between the four quantities. See Figure 7.1 and Table 7.1 for all observed FRB parameters, and Table 7.2 for derived cosmological parameters. All 13 beams of full-Stokes data were analysed in detail and the pulse was not detected in any other beam of the receiver. Since there was no coincident detection in other beams, we conclude that the event was not a sidelobe detection. Therefore we have used the coordinates from the beam center for the detection pointing with an error diameter of 14.40, the approximate full-width half-maximum (FWHM) of beam 1 at 1.4 GHz (Staveley-Smith et al., 1996).

FRB 140514 was discovered in a pointing centered just 90 away from the nominal position of known FRB 110220 during a standard gridding (Morris et al., 2002) of the region in our survey. The previous event nearby, FRB 110220, had DM = 944.38(5) pc 3 cm and a peak flux density of 1.3 Jy, if it occurred at beam-center. Models of the free electron content of the Milky Way predict that the ionized Galactic 3 interstellar medium contribution to the DM of FRB 140514 is only 35 pc cm (Cordes & Lazio, 2002), only 6% of the total, which sets an upper limit on redshift z<0.4(1) based on ionization models of the intergalactic medium (IGM), making no assumptions about a host contribution to the total DM, and assuming an upper limit on the Galactic 3 DM contribution of 70 pc cm (Ioka, 2003). This upper limit on redshift corresponds +0.04 to a co-moving distance of < 1.71(3) Gpc, a luminosity distance of < 2.46 0.06 Gpc, an +4.7 31 energy of < 3.7 2.0 10 Joules, and a distance modulus of < 42.2 mag (Wright, 2006). In ⇥ comparison, the upper limit on redshift for FRB 110220 was z<0.81, which corresponds to a co-moving distance of 2.8 Gpc, a luminosity distance of 5.1 Gpc, and a distance modulus of 43.5 mag (Thornton et al., 2013) if most of the excess DM is attributed to the IGM. A calibration observation was taken at the end of the observing session at 01:04:39 UTC on 15 May, 7h50m after FRB 140514, which was used to calibrate the polarized data. The feed was assumed to be ideal and the calibration was performed using the pac command in the PSRCHIVE software package2 (Hotan et al., 2004). We did not perform a Mueller matrix calculation as we cannot determine the exact location of the FRB within the Parkes beam. From the calibration of the orthogonal linear feeds we obtained all four Stokes parameters, plotted in Figure 7.2; this represents the first detection of polarized flux from an FRB. The emission of FRB 140514 was polarised with 21 7% (3-) circular polarisation ± 2http://psrchive.sourceforge.net/index.shtml 7.3. Parkes real-time detection of FRB 140514 105

Table 7.1 Observed properties of FRB 140514.

Event date UTC 14 May, 2014 Event time UTC, ⌫1.4GHz 17:14:11.06 Event time, ⌫ 17:14:09.83 1 Local date AEST 15 May, 2014 Local time AEST 03:14:11.06 RA 22:34:06.2 Dec 12:18:46.5 (`,b)(50.8, 54.6 ) Beam diameter 14.40 3 DMFRB (pc cm )562.7(6) 3 DMGalaxy (pc cm )34.9 Detection S/N 16(1) +3.5 Observed width, W (ms) 2.8 0.7 Scattering timescale, ⌧1GHz (ms) 5.4(1) Dispersion index, ↵ -2.000(4) +0.11 Peak flux density, S⌫,1400MHz (Jy) 0.47 0.08 +2 .3 Fluence, (Jy ms) 1.3 0.5 F

Table 7.2 Derived cosmological properties of FRB 140514. All properties are upper limits assuming no host contribution to the total DM.

z<0.44(1) Co-moving distance (Gpc) < 1.71(3) +0.04 Luminosity distance (Gpc) < 2.46 0.06 Energy (Joules) < 3.7+4.7 1031 2.0 ⇥ Distance modulus (mag) < 42.2 106 Chapter 7. A real-time fast radio burst

+3.5 Figure 7.1 The pulse profile and frequency-time plot of FRB 140514 with pulse width 2.8 0.7 3 ms, dedispersed to DM = 562.7 pc cm and summed to 8 frequency channels across the band. The total time plotted has been reduced to 400 ms for greater clarity. Frequency channels between 1520 to 1580 MHz are excised due to narrow-band radio interference from the Thuraya 3 satellite which transmits in this band. 7.3. Parkes real-time detection of FRB 140514 107

Figure 7.2 The full-Stokes parameters of FRB 140514 recorded in the centre beam of the multibeam receiver with BPSR. Total intensity, and Stokes Q, U,andV are represented in black, red, green, and blue, respectively. FRB 140514 has 21 7% (3-) circular ± polarisation averaged over the pulse, and a 1- upper limit on linear polarisation of L< 10%. On the leading edge of the pulse the circular polarisation is 42 9% (5-)ofthe ± total intensity. The data have been smoothed from an initial sampling of 64 µsusinga Gaussian filter of full-width half-maximum 90 µs. averaged over the whole pulse. On the leading edge of the pulse, however, the pulse is 42 9% circularly polarised, a 5- detection. No linear polarisation was detected and ± we place a 1- upper limit of 10% of the total intensity (Figure 7.2). We note that it would require a very rare and specific feed rotation to result in high fractional circular polarisation with no linear detection. Such a configuration would also result in a high correlation between Stokes V and I, and we do not observe the circular polarisation to tightly follow the total intensity. The measured circular polarisation is determined to be intrinsic to the observation, and not a calibration artefact. With a polarized signal it is possible to measure the Faraday rotation of the Stokes vectors as a function of frequency due to the magnetic field and electron column density along the line of sight. The amount of induced rotation is quantified by the rotation measure,

d RM neB dl, (7.1) / k Z0 where d is the distance to the source, ne is the electron column density, and B is the mag- k netic field parallel to the line of sight such that the rotation angle of the linear polarization = RM2 (see Section 1.2.4). An optimal rotation measure search was performed using 5 the rmfit code in the psrchive pulsar software package out to RMmax =1.18 10 | | ⇥ 108 Chapter 7. A real-time fast radio burst

2 rad m , the RM at which the signal is completely depolarized within a single frequency channel at our observing frequency. No linear polarization was evident at 3 significance.

7.4 FRB Follow-up at Other Telescopes

The real-time detection of FRB 140514 enabled an extensive coordination of telescopes and multi-wavelength observations. Three sources of interest were identified in the 14.40 diameter of the Parkes beam: two X-ray sources detected by Swift, referred to below as XRT1 and XRT2, and one from the Giant Metrewave Radio Telescope (GMRT), referred to as GMRT1. The search for a counterpart focused on identification of any variable slow transients in the field that either brightened or dimmed by > 2 mag between epochs. For objects near the magnitude limit of the observation (the magnitude of the dimmest detectable source in the observation) an appearance or disappearance between epochs was only considered significant if the source was 2 magnitudes or more above the limit in one observing epoch. Ultimately, no afterglow-like counterpart was identified at any wavelength involved in this effort. Here we report on the findings from twelve telescopes involved in the follow-up effort, listed in Table 7.3.

7.4.1 Parkes Radio Telescope

FRB 140514 was discovered in the first grid pointing around the position of FRB 110220 of the observing session, 90 away from the previous FRB discovery position, less than one beamwidth (Thornton et al., 2013). A grid of the field was observed for another 1.6 hours after detection as part of the scheduled observations, and then again at the end of the observing session, 7 hours after discovery, both with 145 mJy rms at 1.4 GHz. No further dispersed pulses at DM 5 pc cm 3 were detected in subsequent observations. No other FRB-like pulses were found throughout the observing session along any other sightlines, and there was no strong radio frequency interference (RFI). The field was re-observed for 8 consecutive hours on 24 June, 41 days after the FRB event and again for 3.5 hours on 27 July, 74 days after FRB 140514, and no new candidates were identified.

7.4.2 Australia Telescope Compact Array

The Australia Telescope Compact Array (ATCA) observed the FRB field as a target of opportunity (ToO, proposal CX293) starting at 00:10 UT on 2014 May 15, less than 7 hours after the FRB. The total observing time was 3 hours including calibration and overheads. Observations were made simultaneously from 4.5 6.5 GHz and 8 10 GHz for a total of 7.4. FRB Follow-up at Other Telescopes 109

60 min on source and from 1.1 3.1 GHz with 60 min on source. The rms of the images are approx 40 µJy at the higher frequencies and 60 µJy at 2 GHz. GMRT1 was identified in the ATCA image with a flux density of 1.5 mJy and XRT1 was also visible with a flux density of 3 mJy. While the ATCA was the first telescope other than Parkes to image the field, the lack of a second epoch days or weeks later hampered our ability to detect variable sources at these radio frequencies. No radio source could be targeted based on the ATCA observations.

7.4.3 Giant Metrewave Radio Telescope

The Giant Metrewave Radio Telescope (GMRT) began observing the FRB field as a target of opportunity (ToO, proposal DDT B124) 2 days after the event at 610 MHz at 01:30 UT on 16 May. The 3-hour observation (including overheads and calibrators) produced an image of the field with 123 µJy rms. We identified three sources within the field of view (J2000 coordinates): GMRT1 (RA = 22:34:08.493, Dec = 12:18:27.00), GMRT2 (RA = 22:34:19.003, Dec = 12:21:30.38), and GMRT3 (RA = 22:34:00.088, Dec = 12:14:50.00). The first, GMRT1, did not appear to correspond to any sources in the NRAO VLA Sky Survey catalog (NVSS, Condon et al. 1998), and was flagged for further follow-up by other telescopes as a potentially variable source given the temporal proximity of the GMRT observation and the FRB detection. The other two sources, GMRT2 and GMRT3, corre- lated well with positions for known radio sources in the NVSS catalog with consistent flux densities. Subsequent observations were taken through the GMRT ToO queue on 20 May, 3 June, and 8 June in the 325 MHz, 1390 MHz, and 610 MHz bands, respectively. The second epoch was largely unusable due to technical difficulties. The search for variablility focused on monitoring each source for flux variations across observing epochs. All sources from the first epoch appeared in the third and fourth epochs with no measureable change in flux densities.

7.4.4 Swift X-Ray Telescope

The first observation of the FRB 140514 field was taken using Swift XRT (Gehrels et al., 2004) only 8.5 hours after the FRB was discovered at Parkes. This was the fastest Swift follow-up ever undertaken for an FRB. 4 ks of XRT data were taken in the first epoch, and a further 2 ks of data were taken in a second epoch later that day, 23 hours after FRB 140514, to search for short term variability. A final epoch, 18 days later, was taken to search for long term variability. Two X-ray sources were identified in the first epoch of data within the 150 diameter of the Parkes beam. Both sources were consistent with 110 Chapter 7. A real-time fast radio burst sources in the USNO catalog (Monet et al., 2003). The first source (XRT1) is located at RA = 22:34:41.49, Dec = 12:21:39.8 with R = 17.5 and the second (XRT2) is USNO located at RA = 22:34:02.33 Dec = 12:08:48.2 with R = 19.7. Both XRT1 and USNO XRT2 appeared in all subsequent epochs with no observable variability on the level of 10% and 20% for XRT1 and XRT2, respectively, both calculated from photon counts from the XRT. Both sources were later found to be active galactic nuclei (AGN).

7.4.5 Gamma-Ray Burst Optical/Near-Infrared Detector

After 13 hours, a trigger was sent to the Gamma-Ray Burst Optical/Near-Infrared Detector (GROND) operating on the 2.2-m MPI/ESO telescope on La Silla in Chile (Greiner et al., 2008). GROND is able to observe simultaneously in J, H,andK near-infrared (NIR) bands with a 10 10 field of view (FOV) and the optical g , r , i ,andz bands with a 0 ⇥ 0 0 0 0 0 6 6 FOV. A 2 2 tiling observation was done, providing 61% (JHK)and22%(g r i z ) 0 ⇥ 0 ⇥ 0 0 0 0 coverage of the inner part of the FRB error circle. The first epoch began 16 hours after FRB 140514 with 460 second exposures, and a second epoch was taken 2.5 days after the

FRB with an identical observing setup and 690 s (g0r0i0z0)and720s(JHK) exposures, respectively. Limiting magnitudes for J, H,andK bands were 21.1, 20.4, and 18.4 in the first epoch and 21.1, 20.5, and 18.6 in the second epoch, respectively (all in the AB system). Of all the objects in the field, analysis identified three variable objects, all very close to the limiting magnitude and varying on scales of 0.2 - 0.8 mag in the NIR bands identified with difference imaging. Of the three objects one is a galaxy, another is likely to be an AGN, and the last is a main sequence star. Both XRT1 and GMRT1 sources were also detected in the GROND infrared imaging but were not observed to vary in the infrared bands on the timescales probed.

7.4.6 Swope Telescope

An optical image of the FRB field was taken 16h51m after the burst event with the 1-m Swope Telescope at Las Campanas. The field was re-imaged with the Swope Telescope on 17 May, 2 days after the FRB. No variable optical sources were identified in the observations field of view, 1.92 deg2,toalimitingmagnitudeofR = 16.

7.4.7 Palomar Transient Factory

The intermediate Palomar Transient Factory (iPTF) uses the 1.2-m Samuel Oschin Tele- scope at Palomar Observatory at R-band with 60 s exposure times to search for optical transients over 8.1 deg2 (Law et al., 2009; Rau et al., 2009). The iPTF was triggered within 7.4. FRB Follow-up at Other Telescopes 111

12 hours and observations began approximately 18 hours after FRB 140514 on the night of 15 May. Four epochs of suitable data were taken of the field on May 15, 16, 17, and 19 with an R-band limiting magnitude of 19.1, 19.3, 19.3, and 19.1, respectively. All data were reduced using the IPAC pipeline (Laher et al., 2014) with photometric calibration described in Ofek et al. (2012). Several stars and were identified in the field over the four epochs, however no variable or fading candidates were identified that might be associated with FRB 140514.

7.4.8 Magellan Telescope

Deep images of the FRB field were taken with the 6.5-m Baade telescope at Las Campanas in the R and I bands on 17 May, 3 days after FRB 140514, and again on 8 July, 55 days later. These data have provided the deepest optical images of the field, with a limiting magnitude, R, I = 22.5 in the first epoch and R, I = 24.5 in the second epoch with a field of view of 635 arcmin2. The first observation was 7 2-min exposures and the second consisted of 5 5-min exposures. These observations identified one extended object in the field that appeared with a magnitude of R =21.9 0.1 in the first epoch and was not detected in ± the second observation through point source searches. Due to its extended nature in the observation this source has been identified as a moving object such as a satellite or debris passing through the field and was not flagged as a potentially associated candidate.

7.4.9 SkyMapper

A ToO was sent to the 1.35-m SkyMapper telescope at Siding Spring in Australia. Ob- servations were taken on the night of the 16 May, 2 days after FRB 140514, and 23 May, 9 days after the event. The SkyMapper field of view is 5.7 deg2 and both images were centred on the FRB coordinates using the H↵ filter which was in place for those nights. No variable objects were seen across the two epochs of data through difference imaging.

7.4.10 Effelsberg Radio Telescope

The field was observed at 1.4 GHz (21 cm), 2.7 GHz (11 cm), and 4.85 GHz (6 cm) using the 100-m Effelsberg Radio Telescope in Germany five days after FRB 140514. A single object was detected in the field with S =447 30 mJy and a spectral index of 1.4GHz ± ↵ = 0.54 0.08. A source at this position was also visible in the NVSS with S = 493 15 ± ± mJy, which is consistent, indicating no change in brightness after FRB 140514. 112 Chapter 7. A real-time fast radio burst

7.4.11 Keck Spectroscopy

Spectroscopic follow-up of XRT1, XRT2, and GMRT1 was performed on 27 May, 13 days after FRB 140514, using the Low Resolution Imaging Spectrometer (LRIS) on the 10-m Keck I telescope (Oke et al., 1995). Based on their spectral properties, XRT1 and XRT2 were identified as AGN, and GMRT1 was identified as a starburst galaxy with a high star formation rate. The spectral features of GMRT1 were typical of a starburst galaxy and there were no strong or unexpected spectral line features that might hint at unusual activity.

7.4.12 Nordic Optical Telescope Spectroscopy

Additional spectroscopic observations were performed using the 2.5-m Nordic Optical Tele- scope (NOT) at La Palma for XRT1 and XRT2, 2.4 and 21.4 days after FRB 140514, re- spectively. Both were confirmed to be AGN. The brighter X-ray source, XRT1, is a Seyfert type 1.9 galaxy at z = 0.195, and the XRT2 is an AGN with z = 0.51. The spectra of both sources were not observed to evolve bewteen the two observations with NOT and Keck.

7.5 Interpretation and Discussion

Over the various epochs and wavelengths detailed in Section 7.4, no afterglow-like variable counterparts were detected that could be identified as a candidate host or progenitor asso- ciated with the radio observation of FRB 140514. Here we consider the possible sources and mechanisms that might produce the observed behavior and set limits on FRB detectability at other wavelengths on hour-to-day timescales.

7.5.1 Polarization

Any progenitor theory of FRBs must explain the observed polarization; several possibilities exist for this FRB. We consider three here:

Case 1: The emission is intrinsically only circularly polarized, as observed, perhaps • also with low linear polarization.

Case 2: The emission is intrinsically linearly and circularly polarized, but the linear • polarization was undetectable due to bandwidth depolarization by severe Faraday rotation.

Case 3: The emission is intrinsically unpolarized and is scintillation- • induced. 7.5. Interpretation and Discussion 113

Table 7.3 Follow-up observations conducted at 12 telescopes. Limits presented are the minimum detectable magnitude or flux of each epoch. All dates are for the year 2014.

Telescope Date Start time T+ Limits UTC Parkes May 14 17:14:12 1 s 1.4 GHz - 145 mJy ATCA May 15 00:10:00 7 h 5.5 GHz - 40 µJy 2GHz-60µJy Parkes May 15 23:57:38 6 h 52 m 1.4 GHz - 145 mJy Swift May 15 01:44:43 8 h 30 m 8.2 10 15 erg cm 2 s 1 ⇥ GROND May 15 08:49:30 16 h J -21.1,H -20.4, K -18.4 Swope May 15 09:57:13 16 h 51 m R -16 iPTF May 15 11:16:03 18 h 11 m R -19.1 Swift May 15 16:08:44 23 h 18 m 3.9 10 15 erg cm 2 s 1 ⇥ GMRT May 16 01:30:00 1.3 d 610 MHz - 125 µJy Effelsberg May 16 06:50:00 1.4 d 4.8 GHz - 2.5 mJy iPTF May 16 11:18:21 1.7 d R -19.3 SkyMapper May 16 17:57:24 2 d H↵ -17 NOT May 17 04:48:46 2.4 d 370 730 nm GROND May 17 09:04:13 2.6 d J -21.1,H -20.5, K -18.6 Swope May 17 09:50:00 2.6 d R -16 Magellan May 17 10:11:19 2.6 d R -22.5,I -22.5 iPTF May 17 11:15:33 2.7 d R -19.3 Effelsberg May 18 03:50:00 3.4 d 2.7 GHz - 1.2 mJy iPTF May 19 11:23:52 4.7 d R -19.1 Effelsberg May 21 05:35:00 7.5 d 1.4 GHz - 1.2 mJy SkyMapper May 23 17:45:48 9 d H↵ -17 Keck May 27 14:06:22 12.8 d 30 1000 nm Swift June 02 00:06:02 18.3 d 6.35 10 15 erg cm 2 s 1 ⇥ GMRT June 03 00:20:00 19.3 d 1390 MHz - 61 µJy NOT June 05 03:51:09 21.4 d 370 730 nm GMRT June 08 20:30:00 24.1 d 610 MHz - 150 µJy Parkes June 24 14:36:40 41 d 1.4 GHz - 145 mJy Magellan July 8 07:34:44 55 d R -24.5,I -24.5 Parkes July 27 12:14:00 74 d 145 mJy 114 Chapter 7. A real-time fast radio burst

Case 1. Few sources observed at radio frequencies produce only circularly polarized emission. The flare star AD Leonis has been observed to produce 90 100% circularly polarized radio emission coincident with optical flares (Osten & Bastian, 2006, 2008); the brown dwarf TVLM 513 46546 and the Sun have also been observed to emit 100% circularly polarized bursts at GHz frequencies, both attributed to cyclotron maser emission (Hallinan et al., 2007; Melrose & Dulk, 1982). These radio bursts typically last seconds 2 to minutes with no frequency-dependent time delays comparable to the ⌫ dispersive sweep seen for FRBs. The level of circular polarization (CP) in FRB 140514 ( 21%)is ⇠ also much lower than that observed in other cases. Some AGN have been observed with more circular polarization than linear (Homan & Lister, 2006), however the overall levels of CP were much lower (typically 0.3%) and would not have been detected here. Some single pulses from pulsars have been observed with high fractional circular polarisation and a small linear component and this FRB may represent such a state (Levin et al., 2012; Osłowski et al., 2014).

Case 2. The maximum RM for the search in Section 7.3 describes the complete depo- larization of a 100% linearly polarized source. A weaker level of linear polarization might have been depolarized at these frequencies by RMs of order 104 rad m 2 and greater, ⇠ however such values are still several orders of magnitude greater than theorized for Faraday rotation in the IGM (Akahori & Ryu, 2010). Additionally, if the source originated at some redshift and the emission observed at 1.4 GHz was redshifted into our observing band, the Faraday rotation at the source would need to be much higher given the frequency of emis- sion. The electron density component of the RM is constrained by the source DM, thus high magnetic fields are required to produce the necessary rotation of the plane of linear polarisation. Such high rotation measures are incredibly rare, but have been observed in the magnetar PSR J1745-2900 near the Galactic Centre (Eatough et al., 2013; Shannon & Johnston, 2013). The required path-averaged magnetic field to produce these rotation measures would be 250 µG for FRB 140514. A source located within 1 pc of the center 3 of its host galaxy, with the host contributing 100 pc cm to the total DM, could produce path-averaged magnetic field strengths of 10 100 µG. FRB 140514 would need to have originated in a region of similar magnetic field strength to produce the necessary Faraday rotation. It is also worth noting that observations of a radio-loud magnetar have shown the occurence of an infrequent state in which the emission is highly circularly polarized with a lower-than-average linear component (Levin et al., 2012). With a sufficiently strong integrated magnetic field along the line of sight, the linear component could be completely depolarized for a magnetar flare. Such a flare would still be in good agreement with the 7.5. Interpretation and Discussion 115 models and conditions put forward in Kulkarni et al. (2014). Case 3. It has been theorized that scintillation-induced CP may arise in an intrinsically unpolarized source (Macquart & Melrose, 2000). In the diffractive regime - for pulsars and other Galactic sources - CP up to 15% may be induced by a birefringent medium at low ⇠ frequencies, and in the refractive regime 0.1% may be induced for extragalactic sources ⇠ observed at higher radio frequencies. Such scintillation requires the presence of an RM gradient across the turbulent region, which we cannot constrain with available data. The 15% CP in the Macquart & Melrose (2000) simulation was derived from the Vela pulsar, ⇠ an extreme example within the pulsar population, being bright, young, and surrounded by a turbulent and high-velocity medium (Hamilton et al., 1985; Chapter 3). Such an object would very likely be detected in future follow-up of the detection position. Of the three posibilities presented here, Case 3 requires very specialized Galactic con- ditions to produce CP close to the level observed in this work and is not the best model to describe the observed FRB circular polarization. Both Case 1 and Case 2 require CP intrinsic to the source, and vary only in on the predicted level of linear polarization. Recent theoretical work has speculated that mechanisms to produce sufficiently high brightness temperatures for observed FRBs require beaming of coherent emission (Katz, 2014) which would produce intrinsic linear polarization, making Case 2 more appealing.

The required brightness temperature TB for FRB 140514, ignoring relativistic effects, can be calculated using Equation 1.18. For FRB 140514 we estimate a brightness temperature T =5.3 1035 K, not including relativistic effects ( =1). Such a high brightness B ⇥ temperature is beyond the regime of pulses from typical pulsars but approaches what is seen in the brightest nanoshot pulses from the Crab pulsar (1041 K; Hankins & Eilek, 2007). Temperatures in this regime preclude synchrotron emission, thus making it likely that the pulse emission is coherent (Readhead, 1994). However, no linear polarization was detected, an unexpected result given previous observations of coherent emission mechanisms at such high brightness temperatures (Hankins et al., 2003). An additional theory of FRB origins put forward by Mottez & Zarka (2014) suggests that FRBs are created in the Alfvén wings of a planet orbiting a neutron star within the pulsar wind via the electron cyclotron maser (ECM) instability at Gpc distances. This theory predicts stong circular polarization such as the 100% circular emission seen in other objects that emit via ECM such as the M dwarves observed by Hallinan et al. (2008). While this is the only current theory that explicitly predicts circularly polarised emission from FRBs, a significant fraction of the intrinsic circular polarisation (>50%) would have been lost by some unknown mechanism for FRB 140514 to explain the observed polarised 116 Chapter 7. A real-time fast radio burst profile. We then conclude that Case 1 or Case 2 may be the best explanation of the observed polarization for FRB 140514, although, the non-detection of linear polarization at 1.4 GHz might require extremely high magnetic fields to produce the necessary Faraday rotation, possibly near a galactic center. In the future, more sensitive measurements of FRB polarization may be possible with coherent baseband capture buffers such as the CASPER Parkes Swinburne Recorder (CASPSR) installed on the center beam of the multibeam receiver at Parkes, and those being designed for the Square Kilometre Array (SKA).

7.5.2 Possible connection with FRB 110220

FRB 140514 was discovered in radio follow-up observations of a previous FRB event, FRB 110220, published in Thornton et al. (2013). FRB 110220 was the most extensively ana- lyzed FRB in this sample as it was the brightest, detected with a S/N of 49, an extremely 3 2.003 0.006 high DM of 944 pc cm ,a⌫ ± dispersion relation, and a significant scattering tail. FRB 140514 was discovered in a grid pointing around the position of FRB 110220: the centre beam of the receiver was centered 90 away from the detection beam position for FRB

110220. Given the overlap of the 14.40 beam on-sky between the two FRBs it is tempting to make an association. A few considerations must be made before attributing both events to the same source - the probability of detecting a new source given the FRB rate and the time on-sky for these observations, the physical mechanism necessary to produce a large change in dispersion measure over the time between detections, and the time of day and position of the telescope at the time of each detection. The probability of detecting a new FRB in our observations can be calculated using the formula N (1+M+N) (M + N)! P (N M) = ↵ (1 + ↵) , (7.2) | M! N! derived in Chapter 4 to find the likelihood of detecting N FRBs in our survey based on M detections in a previous survey with a ratio between their cumulative time on sky of ↵. This FRB campaign (with extra time granted by the scheduler) has spent 110 hours on-sky over the duration of the survey. The probability of finding a new FRB in these data, given the FRB rate from Thornton et al. (2013) is 33%, and 26% using the new rate derived ⇠ in Section 6.2.2, if FRBs are non-repeating. This probability has been used instead of one estimating only the probability in the centre beam for reasons discussed in Section 6.6.3. Recent results from Burke-Spolaor & Bannister (2014) and Chapter 4 have both ad- 7.5. Interpretation and Discussion 117 dressed the possibility of a lower rate based on results from searches at lower Galactic latitudes, possibly due to foreground effects. However, Burke-Spolaor & Bannister (2014) note that the number of detectable FRBs may be heavily latitude dependent, with different rates applying for surveys at high and low Galactic latitude. In the absence of an expres- sion for this dependence we will continue to use the probability derived from the results in Chapter 6 as the HTRU survey explicity sampled the region around FRB 140514. Based on comparison with previous rate estimates it is therefore not unexpected that we would find a new FRB in this project. 3 FRB 110220 was observed with a DM of 944.7 pc cm , compared to a DM of 562 pc 3 3 cm for FRB 140514, a difference of 380 pc cm for events separated by 1179 days. This is a temporal variation in the DM which is four orders of magnitude greater than what is seen for pulsars in the most turbulent environments (e.g. dDM/dt =0.18 1 pc cm 3 | | ± 1 yr for PSR J1833-0827, Chapter 3). In order for such large DM variations to be observed for these two FRBs, most of the dispersive medium for FRB 110220 would be local to the source and thus extremely dense. Dennison (2014) and Tuntsov (2014) have both argued that if the majority of the DM were produced in a dense plasma around the emission region, such as in a stellar corona, the 2 observed dispersion relation would be poorly fit by a ⌫ relation, which would have been easily detected in the analysis conducted by Thornton et al. (2013) for FRB 110220 and the analysis conducted here for FRB 140514. Additionally, FRB 140514 was discovered at

03:14 AEST local time at a telescope orientation in azimuth and elevation of (84.9,30.3) while FRB 110220 occurred at 11:52 AEST local time at a telescope orientation of (29.9,

66.3). We conclude that FRB 110220 and FRB 140514 are different sources, and their prox- imity is purely due to sampling bias in our choice of observing location. This proximity does not affect the proposed cosmological origin of FRBs.

7.5.3 Limits on a varying counterpart

In explosions such as supernovae (SNe) and superluminous supernovae (SLSNe), or in long gamma-ray bursts (GRBs), a counterpart is detectable as an object of varying brightness in subsequent observations. Variations in brightness would be observable on timescales of hours (for a long GRB), days (for typical SNe) or weeks (for a SLSN). Such variations were not seen in our data in X-ray, near-infrared, optical, or radio regimes. Additionally, no gamma-ray emission from the field was observed by the interplanetary network (IPN) in either the hard or soft gamma-ray energy bands, ruling out all but soft, short gamma- 118 Chapter 7. A real-time fast radio burst ray bursts (V. Pal’shin, priv. comm.). Therefore, no variable counterpart or related transient emission was observed in association with FRB 140514. From the extensive dataset collected in this analysis we can then place limits on the magnitude of any potential afterglow for FRB 140514. We place the limits on related transient emission at 20.0 in J band (1.65 µm), 19.2 in H band (1.2 µm), 18.6 in K band (2.2 µm), 24.5 in R band, 24.5 in I band, 40 µJy at 4.8 GHz, 60 µJy at 1 3 GHz, and 125 µJy at 610 MHz. We compare these limits to light curves of known variable supernovae and gamma-ray bursts (Figure 7.3) and find that many nearby sources would have been detected in this analysis (Evans et al., 2007; Rest et al., 2011; Kulkarni et al., 1998; Galama et al., 1998). We rule out local superluminous supernovae and nearby (z<0.3) type Ia supernovae, as well as slow transients with variations greater than 2 mag AB between our epochs of observation. We also rule out an FRB association with long GRBs. Such constraints make associations between FRB 140514 and some SNe or long GRBs highly unlikely. Any theoretical description of FRB emission must adhere to these con- straints.

7.5.4 Host galaxies

Given the large error radius of the Parkes beam (7.20) an identification of a host galaxy in the original radio detection was not possible and no other association was detected at other wavelengths as no variable sources were identified in the field. In our follow-up observations three sources of interest were detected in the field of FRB 140514 by Swift and the GMRT: two X-ray luminous AGN and one radio-loud starburst galaxy. The identification of AGN within a 150 beam is not unexpected; approximately 3 AGN are expected to occur within the area of the Parkes beam out to z =2based on cosmological distribution alone (Fiore et al., 2003). Finding a starburst galaxy in our field is also likely based on galaxy distribution studies (Hirashita et al., 1999). Therefore, we are unable to make any robust connection between the FRB and any other objects identified in the field on probability arguments alone.

7.6 Conclusions

We report here on an FRB discovered in real-time at 1.4 GHz at the Parkes radio tele- scope. FRB 140514 was the first real-time FRB detection and the first with polarization information. The pulse was observed to be 21 7% (3-) circularly polarized with a 1- ± 7.6. Conclusions 119

Figure 7.3 The limits for optical in apparent magnitude (green), radio flux density in mJy 2 1 (red), and X-ray flux in erg cm s (blue) of our observations of the field of FRB 140514 from 8 telescopes that fully sampled the Parkes beam. Colors of data points refer to the axis scale of the same color. Light curves from GRB140512A (z = 0.725), 1.4 GHz radio data and R band optical data for supernova SN1998bw (z 0.008), R band data for ⇠ SN2003ma (z =0.289)andanR band light curve for a typical type-Ia SN (z = 0.5) have been included for reference (Evans et al., 2007; Rest et al., 2011; Kulkarni et al., 2014; Galama et al., 1998). 120 Chapter 7. A real-time fast radio burst upper limit of 10% on linear polarization which was not detected. We coordinated the fastest and largest follow-up effort ever undertaken for an FRB, with data from 12 tele- scopes. No associated slow transient, progenitor, or host galaxy was identified, effectively ruling out any association between this FRB and a concurrent supernova at z<0.3 or long gamma-ray burst. A tighter constraint on FRB origins in the future will require not only robust and immediate triggering or commensal observing at multiple observatories, but also improved sky localisation of radio pulses within FRB and pulsar surveys. We note that any theory of FRB origin must satisfy the polarization, brightness temperature, and afterglow limits put forward in this analysis.

Public Data Release

The data presented in this paper are made available through Research Data Australia3 and can be processed using the publicly available heimdall single pulse processing software and the psrchive software package.

3https://researchdata.ands.org.au/fast-radio-burst-frb-140514/468269 8 Conclusions and Future Directions

8.1 Major Findings of the Thesis

The work presented in the preceding chapters was focused on detecting and understanding high time resolution phenomena in radio astronomy. We presented novel results from the Parkes radio telescope over a range of topics related to pulsars and radio transients including studies of the interstellar medium, the peryton phenomenon, and fast radio bursts. The general conclusions in these areas are presented below.

1. Variations in pulsar dispersion measures. This thesis presented new investiga- tion into the phenomenon of dispersion measure (DM) variations due to turbulence in the interstellar medium and local circumstellar medium. In Chapter 3 we pre- sented an analysis of dispersion variations for a sample of over 160 young pulsars to complement the studies undertaken with millisecond pulsars at low DM (Keith et al., 2013). The available data were not able to probe the turbulent structure of the interstellar medium on larger scales but were able to probe variations in pulsar DMs due to ionized material local to the source, such as in local supernova remnants or pulsar wind nebulae. In turn, these findings are useful for constraining changes in DM local to any young energetic source, such as possible DM changes over several years along lines of sight to fast radio bursts (FRBs) if they are repeating sources, discussed in Chapters 6 and 7. The local DM contribution to the total FRB DMs remains an open question and this work helps quantify the magnitude of variations that might be expected for both pulsars and radio transients.

2. Radio transient searches. Searches for fast radio burst and peryton pulses are presented in this thesis from several available datasets. New search techniques such as the heimdall single pulse processing software and real-time FRB pipeline were

121 122 Chapter 8. Conclusions

described in Chapter 2, and the peryton search pipeline was described in Chapter 5. These pipelines were applied to the High Time Resolution Universe (HTRU) survey at intermediate and high latitudes (Chapters 4 and 6) and the follow-up survey to search for FRB repetition (Chapter 6). The searches yielded many thousands of candidates, out of which 5 new FRBs and 35 perytons emerged. Prior to this ⇠ thesis, comprehensive and complete searches of massive survey datasets were difficult to achieve due to the large amount of processing time required. This thesis presented the first searches with near real-time search software of over 500 TB of data.

3. The source of perytons. The discovery of a large number of perytons in both the archival intermediate and high latitude HTRU surveys and ongoing observing projects resulted in the ability to study the population in a more robust and mean- ingful way than was previously possible with only 15 sources from Burke-Spolaor ⇠ et al. (2011a), Kocz et al. (2012), and Bagchi et al. (2012). The installation of a broad-spectrum radio frequency interference monitor at the Parkes site also provided valuable information as to the source of the perytons and the microwave ovens on site were ultimately implicated (Chapter 5). The similarities between perytons and FRBs has caused concern since the discovery of perytons in 2011 (Burke-Spolaor et al., 2011a; Kulkarni et al., 2014). Once the source of the perytons was finally discovered it became possible to delineate peryton and FRB populations and demon- strate that the microwave ovens on site are unlikely to be related to the observed FRBs, including FRB 010724 (the Lorimer burst).

4. Discovery of fast radio bursts. This thesis also presented the discovery of 5 FRBs using the heimdall search pipeline. Four bursts were found in the HTRU high lat- itude survey (Chapter 6) and a discovery was made using the real-time pipeline described in Chapter 2 for which full polarization information was recorded (Chap- ter 7). These bursts have all been found at high Galactic latitudes and the rate at b > 20 continues to be discrepant with the possible latitude dependence presented | | in Chapter 4. FRB 140514 was found to be 21 7% circularly polarized with no de- ± tected linear polarization, which may indicate an emission mechanism that produces only circular polarization or that the linear polarization was lost due to a strong local magnetic field. More polarized FRB pulses are needed to place this result in the context of the greater population. These advances have greatly expanded the FRB population and our knowledge of FRB properties.

5. Multi-wavelength follow-up of a radio transient. Upon the discovery of FRB 8.2. Future Directions 123

140514, multi-wavelength follow-up was carried out by twelve telescopes operating from X-ray to decimeter wavelengths. No variable source was detected in the FRB field in the weeks after the FRB event (Chapter 7). The follow-up conducted as part of this thesis represented the first attempt to carry out multi-wavelength follow-up of an FRB detected by a radio telescope. These observations, while ruling out sources such as long gamma-ray bursts and Type Ia supernovae out to z . 0.5,didnot exclude the favored progenitors (described in Section 1.3) or repeating progenitors (Chapter 6).

8.2 Future Directions

The results presented in this thesis provided a new study of the variable radio sky and an exploration of the radio transient parameter space. However, many questions related to these topics remain unsolved, particularly the questions surrounding the origin of the FRB pulses detected at radio telescopes. The true distances to their progenitors and their usefulness as probes of the intervening ionized material remain to be determined. The first critical aspect of the emerging field of FRB science is that we must know what causes FRBs and from where they originate. To do this we need a larger source population: we need to find more FRBs. The current distribution of known FRBs is insufficient to perform population studies (Figure 8.1). Given the relatively high all-sky rate inferred in Chapter 6 a large number of bursts should be straightforward to find with a large field of view, a large amount of time on sky, or both. However these searches will be most effective with high sensitivity instruments which present a complex technical challenge. As more telescopes begin searching for bursts, in both new and archival data, a population (or pop- ulations) with distinct properties will begin to emerge, which will enable us to place further constraints on progenitor models. Ultimately better understanding of FRBs requires in- strumentation with more precise localization and better observing tools, preferably with commensal multi-wavelength observing. A critical tool for the search is a reliable triggering mechanism both for finding FRB signals in the data and relaying these detections to other telescopes. Due to the presence of terrestrial radio frequency interference (RFI) the number of false positive FRB detections can be large in a period of strong RFI (10s to 100s of triggers in a day). Human verification of a trigger is still currently required for searches at Parkes; however, such a human-based approach has obvious pitfalls, i.e. lag time between detection and trigger, human error, and habituation to false alerts. New verification steps can be built into search code for each type of RFI encountered at the telescope to filter out as many local sources as possible, but 124 Chapter 8. Conclusions

Figure 8.1 The all-sky distribution of the known FRBs. Sources found in real-time at Parkes are shown in red and all other Parkes FRBs are shown in yellow. The burst detected at the is shown in green, and the burst detected at the is shown in blue. The six additional real-time FRBs will be presented in future publications (SUPERB collaboration, in prep.). these may never be enough. Ultimately, the most promising option for robust triggering may be machine learning. A machine learning algorithm trained to reject RFI based on a training set of known RFI sources will be able to learn and extrapolate to extract other forms of RFI as well and isolate the uniquely FRB-like signals. Machine learning has already been used successfully for pulsar searches (Law et al., 2009; Eatough et al., 2010; Lyon et al., 2013; Morello et al., 2014) and could be a powerful tool in the search for FRB signals as well. As the field currently stands the most important facets of understanding the sources of fast radio bursts are localizing the progenitors, finding any multi-wavelength counter- parts associated with the bursts, and determining the distance to the sources. The first, localization, may be the easiest to accomplish within the next few years as powerful new radio interferometers come on-line. Precursor telescopes for the Square Kilometre Ar- ray (SKA) such the Australia SKA Pathfinder (ASKAP), MeerKAT, the Molonglo radio telescope (UTMOST), the Aperture Tile In Focus (Apertif) upgrade to the Westerbork Synthesis Radio Telescope, as well as existing telescopes such as the Karl G. Jansky (VLA; Law et al., 2015), the Owens Valley (LWA; Lazio et al., 2010), and the Canadian Hydrogen Intensity Mapping Experiment telescope (CHIME; Newburgh et al., 2014) will begin transient searches in the near future. All will 8.2. Future Directions 125 use transient search pipelines on beamformed data to search for FRBs and other radio transients. The advantage of an interferometric approach to FRB searches is the potential for accurate localization of the radio signal across a long baseline (Obrocka et al., 2015). If FRBs originate at a local source this will be easily determined based on the detection time delay between detectors. However, if the source of FRBs is indeed astrophysical an inter- ferometer will be able to pinpoint a bright source within the beam pattern to arcsecond positional uncertainty.

Accurate localization of an FRB within the beam pattern is necessary for determining the observational properties of the bursts. With multibeam detectors such as the current 13 beam multibeam receiver on Parkes, the separation of the beams on the sky and the large beam size make it difficult to determine where in the beam the burst was detected. This uncertainty also means that measured values such as flux, fluence, and spectral index cannot be precisely determined as they are degenerate with the intrinsic beam response to an unknown extent (Keane & Petroff, 2015). More precise localization within the beam pattern of a multi-element interferometer would greatly reduce this uncertainty and allow for more robust measurement of burst properties which, in turn, may lead to greater understanding of their progenitor population. On a single dish telescope localization may be greatly improved with the use of a phased array feed (PAF) capable of synthesizing many beams on the sky using a digital beamformer. The advantage of the PAF is that the many beams on the sky are capable of Nyquist sampling the sky and potentially offering sub- arcminute precision for a bright transient. Installing and calibrating such an instrument on a single dish presents less of a technical challenge than the equivalent operation on an interferometer and may provide accurate localization sooner than any of the upcoming interferometric projects.

The detection of multi-wavelength counterparts to fast radio burst events would provide some of the most valuable information towards not only identifying the progenitor but also the emission mechanism of FRBs. The discovery of a Type Ib supernova associated with a gamma-ray burst provided the first definitive link between the gamma-ray burst phenomenon and the core collapse supernova population (Galama et al., 1998; Woosley & Bloom, 2006). A similar associative discovery would galvanize the field of FRB science and perhaps definitively identify a progenitor. Multi-wavelength emission from an FRB would also provide another window from which to observe the possible emission mechanism. All FRB emission to date has been discovered at radio frequencies (at or around 1 GHz) and while the signal is broadband over the spectrum to which our survey instrumentation is sensitive nothing is known about the properties of the emission at lower or higher radio 126 Chapter 8. Conclusions frequencies or in optical or X-ray bands. Detections outside the observing band of our radio telescope would give clues as to the type of emission mechanism producing the radiation and its beaming properties.

Multi-wavelength search and follow-up for FRBs remain important topics going for- ward. The non-detection of counterpart emission reported in Chapter 7 suggests that a more rapid response is needed upon detection of a burst, particularly at X-ray wavelengths where short-lived emission might rapidly decay such as for a GRB. Additionally, follow-up efforts need to probe deeper into the field of the burst, to search out to higher redshifts and to lower magnitude limits. The error radius for an FRB currently detected at Parkes

(14.40) is prohibitively large for many powerful optical telescopes and the number of sources in the search area makes statistically robust analysis extremely difficult. The more precise localization promised by the interferometers discussed above will be a boon to the multi- wavelength transient community participating in such follow-up. Arcsecond localization will allow for unambiguous identification of a progenitor or host galaxy and facilitate much more targeted follow-up searches.

The type of multi-wavelength emission expected from an FRB (if any) is poorly con- strained. Optical and X-ray counterparts for FRBs have been considered under a few progenitor models (Yi et al., 2014); however, there is a large degree of uncertainty on what might be expected. Despite the relatively high frequency of occurrence of FRBs, if their optical counterparts fade on short timescales (. 10 minutes) they will not be detected in most optical surveys unless they are extremely bright. This is because most optical sur- veys typically integrate over a period of minutes or hours on a particular region of the sky and such short transients would quickly become undetectable due to the addition of many minutes of noise on top of the signal. To probe these timescales, commensal observing is required. Optical telescopes taking short exposures and shadowing a radio telescope like Parkes could effectively detect this type of emission if it exists. Therefore, in addition to multi-wavelength follow-up after an FRB is detected, commensal observing also has the potential to yield exciting new results in the field of FRBs and should be considered as a valuable project in the coming years.

Shadowing and commensal observing offer the greatest potential gains in the search for an FRB progenitor. Telescopes both at other radio frequencies and in other regions of the electromagnetic spectrum will provide the strongest indicators of the progenitor emission. Shadowing efforts using lower frequency radio telescopes such as the Giant Metrewave Radio Telescope (GMRT) at 600 MHz and the Murchison Widefield Array (MWA) at 200 MHz are currently implemented as part of the SUPERB survey with these telescopes 8.2. Future Directions 127 shadowing the normal observations of the Parkes telescope in high time resolution ob- serving modes. Additionally, an ongoing project using the Cerro Tololo Inter-American Observatory 4-m telescope and the associated Dark Energy Camera (DECam) imaging instrument has been developed at Swinburne to search for short (< 10-s) duration tran- sients that would be missed by conventional optical surveys. In this project the DECam instrument takes 10-s exposures of the optical sky while shadowing the Parkes telescope FRB survey observations to search for transient optical emission in the event of an FRB detection. The DECam instrument is uniquely suited to these types of observations as its enormous field of view (2 deg2) and high sensitivity ensure full coverage of the multibeam receiver beam pattern down to a low magnitude limit. Commensal observing could provide valuable answers to questions of multi-wavelength FRB emission, however telescope time is a precious resource and such projects are unlikely to be feasible in the long-term unless equally valuable secondary science can be achieved, or telescopes are built exclusively for shadowing purposes.

Precise localization and detection of any multi-wavelength counterpart of an FRB may make it possible to estimate an accurate distance to the FRB source. If the source is local, either in the Galaxy or in the local Universe, the location within its host galaxy can be pre- cisely determined through radio and optical imaging and an estimate of the contributions to DM given in Equation 1.14 can be measured observationally using DM modeling and the morphology of the galaxy. If the source is at cosmological distance and a host galaxy can be unambiguously identified, then the redshift can be precisely measured spectroscopically with a powerful optical telescope. For a source at a significant redshift, disentangling the contributions from the many components of Equation 1.14 becomes extremely difficult as the local environment of the progenitor cannot be directly observed. However, the effects of the circumstellar medium surrounding the progenitor may be measured via the burst rotation measure. If FRBs are intrinsically linearly polarized the magnetic field local to the burst may be probed through the frequency-dependent rotation of the linear polarization vector in the local field. One proposed explanation for the polarization of FRB 140514 in Chapter 7 is the loss of the linear polarization in a strong magnetic field local to the source. The detection of other polarized FRBs and estimates of their rotation measures will provide valuable information in this regard.

On larger scales, accurate measurements of redshifts for a large population of FRBs will open the door for proposed cosmological studies such as probing galactic halos, measuring the missing baryons (McQuinn, 2014), and measuring the dark energy equation of state (Zhou et al., 2014; Macquart et al., 2015). These studies, of course, require a cosmological 128 Chapter 8. Conclusions population of FRBs with some distribution in redshift space. The distribution of, and even the distance to, FRB progenitors remains unknown and none of the current population of known FRBs will be useful in these proposed studies as their redshifts cannot be accurately determined with available data. More FRBs must be found, localized, and followed up to truly understand the sources and causes of the signals and prepare the way for FRB science with future telescopes capable of detecting a large number of radio transients per year. The SKA telescope will be the most powerful radio telescope in the world, and is expected to find a population of thousands of FRBs over the years of its operation. The two main components of the telescope, SKA-MID and SKA-LOW, while operating over different frequency ranges, will both be suitable for transient studies and will operate with unprecedented sensitivity. In the first stage of the SKA, the MID array will consist of 200 ⇠ individual dishes capable of forming around 1,500 tied array beams on the sky operating around 1 GHz. SKA-LOW will consist of over 100,000 dipole antennas operating between approximately 50 and 300 MHz. If FRBs are visible at low frequencies SKA-LOW may detect an FRB every few days, and SKA-MID may expect a source every week (Macquart et al., 2015). These numbers may be optimistic given the computational challenges recently encountered while searching for fast transients in beamformed interferometric data (Law et al., 2015). Work with upcoming interferometric arrays mentioned above may resolve some of these issues in the coming years and facilitate more efficient transient searches with the SKA. In the time between now and when the SKA comes on-line it will be crucial to answer as many of the fundamental FRB questions as possible so that robust searching and triggering methods are in place to amass a large dataset of FRB detections. Once the origins of FRBs are better understood, informed decisions can be made as to the types of experiments that can be done with the large number of sources expected from the SKA. The study of fast radio bursts is just beginning. In recent years, a population has emerged which implies a high all-sky rate, an extragalactic and possibly cosmological origin, and a potential use as probes of the ionized content of the Universe. Many challenges remain. The source of FRBs is currently unknown (although a large number of progenitor models exist) and an accurate distance has yet to be measured to a single source. The next decade of study may offer answers to some of these questions, particularly with the use of new and refurbished radio interferometers to precisely localize and measure transient radio signals. The greatest gains will be achieved when the largest number of telescopes are looking for FRBs, preferably in real-time, over the largest amount of sky possible. Bibliography

Abdo, A. A., Ackermann, M., Ajello, M., et al. 2009, Science, 325, 848

Akahori, T., & Ryu, D. 2010, ApJ, 723, 476

Alpar, M. A., Cheng, A. F., Ruderman, M. A., & Shaham, J. 1982, Nature, 300, 728

Anderson, B., Pritchard, R., & Rowson, B. 1979, Nature, 282, 594

Antoniadis, J., Freire, P. C. C., Wex, N., et al. 2013, Science, 340, 448

Appleton, E. V. 1945, Nature, 156, 534

Armstrong, J. W., Cordes, J. M., & Rickett, B. J. 1981, Nature, 291, 561

Armstrong, J. W., Rickett, B. J., & Spangler, S. R. 1995, ApJ, 443, 209

Baade, W., & Zwicky, F. 1934, Proceedings of the National Academy of Science, 20, 254

Backer, D. C., & Fisher, J. R. 1974, ApJ, 189, 137

Backer, D. C., Hama, S., van Hook, S., & Foster, R. S. 1993, ApJ, 404, 636

Bagchi, M., Nieves, A. C., & McLaughlin, M. 2012, MNRAS, 425, 2501

Bailes, M., Bates, S. D., Bhalerao, V., et al. 2011, Science, 333, 1717

Bannister, K. W., & Madsen, G. J. 2014, MNRAS, 440, 353

Barr, E. D., Champion, D. J., Kramer, M., et al. 2013, MNRAS, 435, 2234

Barsdell, B. R. 2012, PhD thesis, Swinburne University of Technology

Barsdell, B. R., Bailes, M., Barnes, D. G., & Fluke, C. J. 2012, MNRAS, 422, 379

Bates, S. D., Lorimer, D. R., Rane, A., & Swiggum, J. 2014, MNRAS, 439, 2893

Bates, S. D., Bailes, M., Bhat, N. D. R., et al. 2011, MNRAS, 416, 2455

Bates, S. D., Bailes, M., Barsdell, B. R., et al. 2012, MNRAS, 427, 1052

Bhat, N. D. R., Cordes, J. M., Camilo, F., Nice, D. J., & Lorimer, D. R. 2004, ApJ, 605, 759

Brisken, W. F., Benson, J. M., Goss, W. M., & Thorsett, S. E. 2002, ApJ, 571, 906

Brisken, W. F., Macquart, J.-P., Gao, J. J., et al. 2010, ApJ, 708, 232

129 130 Bibliography

Burgay, M., Bailes, M., Bates, S. D., et al. 2013, MNRAS, 433, 259

Burke, B. F., & Graham-Smith, F. 2014, An Introduction to Radio Astronomy

Burke-Spolaor, S., & Bailes, M. 2010, MNRAS, 402, 855

Burke-Spolaor, S., Bailes, M., Ekers, R., Macquart, J.-P., & Crawford, III, F. 2011a, ApJ, 727, 18

Burke-Spolaor, S., & Bannister, K. W. 2014, ApJ, 792, 19

Burke-Spolaor, S., Bailes, M., Johnston, S., et al. 2011b, MNRAS, 416, 2465

Burke-Spolaor, S., Johnston, S., Bailes, M., et al. 2012, MNRAS, 423, 1351

Camilo, F., Lorimer, D. R., Freire, P., Lyne, A. G., & Manchester, R. N. 2000, ApJ, 535, 975

Caroff, L. J., & Scargle, J. D. 1970, Nature, 225, 168

Chandra, P., & Frail, D. A. 2012, ApJ, 746, 156

Clifton, T. R., & Lyne, A. G. 1986, Nature, 320, 43

Condon, J. J., Cotton, W. D., Greisen, E. W., et al. 1998, AJ, 115, 1693

Connor, L., Sievers, J., & Pen, U.-L. 2015, arXiv:1505.05535

Cordes, J. M. 1986, ApJ, 311, 183

Cordes, J. M., & Lazio, T. J. W. 2002, arXiv:0207156

Cordes, J. M., & McLaughlin, M. A. 2003, ApJ, 596, 1142

Cordes, J. M., & Wasserman, I. 2015, arXiv:1501.00753

D’Amico, N., Manchester, R. N., Durdin, J. M., et al. 1988, MNRAS, 234, 437 de Oliveira-Costa, A., Tegmark, M., Gaensler, B. M., et al. 2008, MNRAS, 388, 247

Deng, W., & Zhang, B. 2014, ApJ, 783, L35

Dennison, B. 2014, MNRAS, 443, L11

Dodson, R., Legge, D., Reynolds, J. E., & McCulloch, P. M. 2003, ApJ, 596, 1137

Dulk, G. A. 1985, ARAA, 23, 169 Bibliography 131

Eatough, R. P., Molkenthin, N., Kramer, M., et al. 2010, MNRAS, 407, 2443

Eatough, R. P., Falcke, H., Karuppusamy, R., et al. 2013, Nature, 501, 391

Ekers, R. D., & Moffet, A. T. 1968, Nature, 220, 756

Esposito, P., Israel, G. L., Turolla, R., et al. 2011, MNRAS, 416, 205

Evans, P. A., Beardmore, A. P., Page, K. L., et al. 2007, A&A, 469, 379

Everett, J. E., & Weisberg, J. M. 2001, ApJ, 553, 341

Falcke, H., & Rezzolla, L. 2014, A&A, 562, A137

Ferris, R. H., & Saunders, S. J. 2004, Experimental Astronomy, 17, 269

Fiedler, R. L., Dennison, B., Johnston, K. J., & Hewish, A. 1987, Nature, 326, 675

Fiore, F., Brusa, M., Cocchia, F., et al. 2003, A&A, 409, 79

Foster, R. S., & Backer, D. C. 1990, ApJ, 361, 300

Gaensler, B. M., Stappers, B. W., Frail, D. A., & Johnston, S. 1998, ApJ Letters, 499, L69

Galama, T. J., Vreeswijk, P. M., van Paradijs, J., et al. 1998, Nature, 395, 670

Gao, H., Li, Z., & Zhang, B. 2014, ApJ, 788, 189

Gehrels, N., Chincarini, G., Giommi, P., et al. 2004, ApJ, 611, 1005

Granet, C., Zhang, H. Z., Forsyth, A. R., et al. 2005, IEEE Antennas Propagation Maga- zine, 47, 13

Gray, R. H., & Marvel, K. B. 2001, ApJ, 546, 1171

Grégoire, T., & Knödlseder, J. 2013, A&A, 554, A62

Greiner, J., Bornemann, W., Clemens, C., et al. 2008, PASP, 120, 405

Hallinan, G., Antonova, A., Doyle, J. G., et al. 2008, ApJ, 684, 644

Hallinan, G., Bourke, S., Lane, C., et al. 2007, ApJ, 663, L25

Hamilton, P. A., Hall, P. J., & Costa, M. E. 1985, MNRAS, 214, 5P

Hammond, A. M., Robishaw, T., & Gaensler, B. M. 2012, arXiv:1209.1438 132 Bibliography

Han, J. L., Manchester, R. N., Lyne, A. G., Qiao, G. J., & van Straten, W. 2006, ApJ, 642, 868

Hankins, T. H., & Eilek, J. A. 2007, ApJ, 670, 693

Hankins, T. H., Kern, J. S., Weatherall, J. C., & Eilek, J. A. 2003, Nature, 422, 141

Hassall, T. E., Keane, E. F., & Fender, R. P. 2013, MNRAS

Hassall, T. E., Stappers, B. W., Hessels, J. W. T., et al. 2012, AAP, 543, A66

Hellings, R. W., & Downs, G. S. 1983, ApJ, 265, L39

Hewish, A. 1955, Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences, 228, pp. 238

Hewish, A., Bell, S. J., Pilkington, J. D. H., Scott, P. F., & Collins, R. A. 1968, Nature, 217, 709

Hey, J. S., Parsons, S. J., & Phillips, J. W. 1946, Nature, 158, 234

Hirashita, H., Takeuchi, T. T., Shibai, H., & Ohta, K. 1999, PASJ, 51, 81

Hobbs, G., Lorimer, D. R., Lyne, A. G., & Kramer, M. 2005, MNRAS, 360, 974

Hobbs, G., Lyne, A. G., Kramer, M., Martin, C. E., & Jordan, C. 2004, MNRAS, 353, 1311

Hobbs, G., Hollow, R., Champion, D., et al. 2009, PASA, 26, 468

Hobbs, G., Coles, W., Manchester, R. N., et al. 2012, MNRAS, 427, 2780

Hobbs, G. B., Edwards, R. T., & Manchester, R. N. 2006, MNRAS, 369, 655

Homan, D. C., & Lister, M. L. 2006, ApJ, 131, 1262

Hotan, A. W., van Straten, W., & Manchester, R. N. 2004, PASA, 21, 302

Hulse, R. A., & Taylor, J. H. 1974, ApJ Letters, 191, L59

Hulse, R. A., & Taylor, J. H. 1975, ApJ, 195, L51

Hurley, K., Boggs, S. E., Smith, D. M., et al. 2005, Nature, 434, 1098

Inoue, S. 2004, MNRAS, 348, 999 Bibliography 133

Ioka, K. 2003, ApJ Letters, 598, L79

Jansky, K. G. 1933a, Proc. IRE, 21, 1387

Jansky, K. G. 1933b, Nature, 132, 66

Johnston, S., Nicastro, L., & Koribalski, B. 1998, MNRAS, 297, 108

Karastergiou, A., Johnston, S., Andersson, N., et al. 2014, Advancing Astrophysics with the Square Kilometre Array, 38

Kaspi, V. M., & Stinebring, D. R. 1992, ApJ, 392, 530

Katz, J. I. 2014, Phys Rev D, 89, 103009

Keane, E. F., & Kramer, M. 2008, MNRAS, 391, 2009

Keane, E. F., Kramer, M., Lyne, A. G., Stappers, B. W., & McLaughlin, M. A. 2011, MNRAS, 415, 3065

Keane, E. F., Ludovici, D. A., Eatough, R. P., et al. 2010, 401, 1057

Keane, E. F., & Petroff, E. 2015, MNRAS, 447, 2852

Keane, E. F., Stappers, B. W., Kramer, M., & Lyne, A. G. 2012, MNRAS, 425, L71

Keith, M. J., Johnston, S., Levin, L., & Bailes, M. 2011, MNRAS, 416, 346

Keith, M. J., Jameson, A., van Straten, W., et al. 2010, MNRAS, 409, 619

Keith, M. J., Johnston, S., Bailes, M., et al. 2012, MNRAS, 419, 1752

Keith, M. J., Coles, W., Shannon, R. M., et al. 2013, MNRAS, 429, 2161

Kellermann, K. I., Pauliny-Toth, I. I. K., & Williams, P. J. S. 1969, ApJ, 157, 1

Kocz, J., Bailes, M., Barnes, D., Burke-Spolaor, S., & Levin, L. 2012, MNRAS, 420, 271

Kramer, M., Stairs, I. H., Manchester, R. N., et al. 2006, Science, 314, 97

Kraus, J. 1979, Cosmic Search, 1, 31

Kulkarni, S. R., Ofek, E. O., Neill, J. D., Zheng, Z., & Juric, M. 2014, ApJ, 797, 70

Kulkarni, S. R., Frail, D. A., Wieringa, M. H., et al. 1998, Nature, 395, 663

Laher, R. R., Surace, J., Grillmair, C. J., et al. 2014, PASP, 126, 674 134 Bibliography

Large, M. I., Vaughan, A. E., & Mills, B. Y. 1968, Nature, 220, 340

Lattimer, J. M., & Prakash, M. 2007, Phys. Rep., 442, 109

Law, C. J., Bower, G. C., Burke-Spolaor, S., et al. 2015, ApJ, 807, 16

Law, N. M., Kulkarni, S. R., Dekany, R. G., et al. 2009, PASP, 121, 1395

Lazio, T. J. W., Clarke, T. E., Lane, W. M., et al. 2010, AJ, 140, 1995

Levin, L. 2012, PhD thesis, Swinburne University of Technology

Levin, L., Bailes, M., Bates, S. D., et al. 2012, MNRAS, 422, 2489

Levin, L., Bailes, M., Barsdell, B. R., et al. 2013, MNRAS, 434, 1387

Loeb, A., Shvartzvald, Y., & Maoz, D. 2014, MNRAS, 439, L46

Lorimer, D. R., Bailes, M., McLaughlin, M. A., Narkevic, D. J., & Crawford, F. 2007, Science, 318, 777

Lorimer, D. R., Karastergiou, A., McLaughlin, M. A., & Johnston, S. 2013, MNRAS

Lorimer, D. R., & Kramer, M. 2004, Handbook of Pulsar Astronomy, Vol. 4 ( University Press)

Lovell, B. 1963, Nature, 198, 228

Luan, J., & Goldreich, P. 2014, ApJ Letters, 785, L26

Lyne, A. G., Pritchard, R. S., & Smith, F. G. 1988, MNRAS, 233, 667

Lyne, A. G., & Rickett, B. J. 1968, Nature, 219, 1339

Lyon, R. J., Brooke, J. M., Knowles, J. D., & Stappers, B. W. 2013, arXiv:1307.8012

Lyubarsky, Y. 2014, MNRAS, 442, L9

Macquart, J.-P., & Johnston, S. 2015, MNRAS, 451, 3278

Macquart, J.-P., & Koay, J. Y. 2013, ApJ, 776, 125

Macquart, J.-P., & Melrose, D. B. 2000, ApJ, 545, 798

Macquart, J. P., Keane, E., Grainge, K., et al. 2015, Advancing Astrophysics with the Square Kilometre Array (AASKA14), 55 Bibliography 135

Magro, A., Karastergiou, A., Salvini, S., et al. 2011, MNRAS, 417, 2642

Manchester, R. N., Hobbs, G. B., Teoh, A., & Hobbs, M. 2005, AJ, 129, 1993

Manchester, R. N., Lyne, A. G., Taylor, J. H., et al. 1978, MNRAS, 185, 409

Manchester, R. N., Lyne, A. G., Camilo, F., et al. 2001, MNRAS, 328, 17

Manchester, R. N., Hobbs, G., Bailes, M., et al. 2013, PASA, 30, 17

Maoz, D., Loeb, A., Shvartzvald, Y., et al. 2015, MNRAS, 454, 2183

Masui, K., Lin, H.-H., Sievers, J., et al. 2015, ArXiv e-prints

Matsakis, D. N., Taylor, J. H., & Eubanks, T. M. 1997, A&A, 326, 924

Mayall, N. U., & Oort, J. H. 1942, PASP, 54, 95

McLaughlin, M. A., & et al. 2006, Nature, 439, 817

McQuinn, M. 2014, ApJ, 780, L33

McQuinn, M., Lidz, A., Zaldarriaga, M., et al. 2009, ApJ, 694, 842

Melrose, D. B. 1980, Solar Physics, 67, 357

Melrose, D. B. 1992, Royal Society of London Philosophical Transactions Series A, 341, 105

Melrose, D. B., & Dulk, G. A. 1982, ApJ, 259, 844

Monet, D. G., Levine, S. E., Canzian, B., et al. 2003, AJ, 125, 984

Morello, V., Barr, E. D., Bailes, M., et al. 2014, MNRAS, 443, 1651

Morris, D. J., Hobbs, G., Lyne, A. G., et al. 2002, MNRAS, 335, 275

Mottez, F., & Zarka, P. 2014, A&A, 569, A86

Newburgh, L. B., Addison, G. E., Amiri, M., et al. 2014, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 9145, 4

Ng, C., Bailes, M., Bates, S. D., et al. 2014, MNRAS, 439, 1865

Nicastro, L., & Johnston, S. 1995, MNRAS, 273, 122

Nicastro, L., Nigro, F., D’Amico, N., Lumiella, V., & Johnston, S. 2001, AAP, 368, 1055 136 Bibliography

Nitadori, K., & Aarseth, S. J. 2012, MNRAS, 424, 545

Nyquist, H. 1928, Physical Review, 32, 110

Obrocka, M., Stappers, B., & Wilkinson, P. 2015, A&A, 579, A69

Ofek, E. O., Laher, R., Law, N., et al. 2012, PASP, 124, 62

Oke, J. B., Cohen, J. G., Carr, M., et al. 1995, PASP, 107, 375

Olausen, S. A., & Kaspi, V. M. 2014, ApJS, 212, 6

Osłowski, S., van Straten, W., Bailes, M., Jameson, A., & Hobbs, G. 2014, MNRAS, 441, 3148

Osten, R. A. 2008, arXiv:0801.2573

Osten, R. A., & Bastian, T. S. 2006, ApJ, 637, 1016

Osten, R. A., & Bastian, T. S. 2008, ApJ, 674, 1078

Osten, R. A., Phan-Bao, N., Hawley, S. L., Reid, I. N., & Ojha, R. 2009, ApJ, 700, 1750

Pen, U.-L., & Connor, L. 2015, ApJ, 807, 179

Phillips, J. A., & Wolszczan, A. 1991, ApJ Letters, 382, L27

Piran, T. 1999, Phys. Rep., 314, 575

Popov, S. B., & Postnov, K. A. 2010, in Evolution of Cosmic Objects through their Physical Activity, ed. H. A. Harutyunian, A. M. Mickaelian, & Y. Terzian, 129–132

Rand, R. J., & Kulkarni, S. R. 1989, ApJ, 343, 760

Rand, R. J., & Lyne, A. G. 1994, MNRAS, 268, 497

Rau, A., Kulkarni, S. R., Law, N. M., et al. 2009, PASP, 121, 1334

Ravi, V., Shannon, R. M., & Jameson, A. 2015, ApJ, 799, L5

Readhead, A. C. S. 1994, ApJ, 426, 51

Rest, A., Foley, R. J., Gezari, S., et al. 2011, ApJ, 729, 88

Rickett, B. J. 1977, ARA&A, 15, 479

Romani, R. W., Blandford, R. D., & Cordes, J. M. 1987, Nature, 328, 324 Bibliography 137

Route, M., & Wolszczan, A. 2013, ApJ, 773, 18

Saint-Hilaire, P., Benz, A. O., & Monstein, C. 2014, ApJ, 795, 19

Shannon, R. M., & Johnston, S. 2013, MNRAS, 435, L29

Smith, D. A., Guillemot, L., Camilo, F., et al. 2008, AAP, 492, 923

Smith, F. G. 1968, Nature, 218, 325

Spitler, L. G., Cordes, J. M., Hessels, J. W. T., et al. 2014, ApJ, 790, 101

Sramek, R. A., Panagia, N., & Weiler, K. W. 1984, ApJ, 285, L59

Staelin, D. H., & Reifenstein, III, E. C. 1968, Science, 162, 1481

Standish, E. M. 1998, AAP, 336, 381

Staveley-Smith, L., Wilson, W. E., Bird, T. S., et al. 1996, PASA, 13, 243

Stinebring, D. R., Smirnova, T. V., Hankins, T. H., et al. 2000, ApJ, 539, 300

Sutton, J. M. 1971, MNRAS, 155, 51

Taylor, J. H., Fowler, L. A., & McCulloch, P. M. 1979, Nature, 277, 437

Thornton, D. 2013, PhD thesis,

Thornton, D., Stappers, B., Bailes, M., et al. 2013, Science, 341, 53

Trott, C. M., Tingay, S. J., & Wayth, R. B. 2013, ApJ Letters, 776, L16

Tuntsov, A. V. 2014, MNRAS, 441, L26

Turolla, R., Zane, S., & Watts, A. 2015, arXiv:1507.02924 van den Heuvel, E. P. J., & van Paradijs, J. 1997, ApJ, 483, 399 van Paradijs, J., Kouveliotou, C., & Wijers, R. A. M. J. 2000, ARA&A, 38, 379

Vernardos, G., & Fluke, C. J. 2013, MNRAS, 434, 832

Weiler, K. W., Panagia, N., Montes, M. J., & Sramek, R. A. 2002, ARA&A, 40, 387

Weiler, K. W., Sramek, R. A., van der Hulst, J. M., & Panagia, N. 1982, in NATO Advanced Science Institutes (ASI) Series C, Vol. 90, NATO Advanced Science Institutes (ASI) Series C, ed. M. J. Rees & R. J. Stoneham, 281–291 138 Bibliography

Weisberg, J. M., Nice, D. J., & Taylor, J. H. 2010, ApJ, 722, 1030

Weltevrede, P., & Johnston, S. 2008, MNRAS, 391, 1210

Weltevrede, P., Abdo, A. A., Ackermann, M., et al. 2010a, ApJ, 708, 1426

Weltevrede, P., Johnston, S., Manchester, R. N., et al. 2010b, PASA, 27, 64

Williamson, I. P. 1972, MNRAS, 157, 55

Woosley, S. E., & Bloom, J. S. 2006, ARA&A, 44, 507

Wright, E. L. 2006, PASP, 118, 1711

Yamanaka, Y., & Shinozuka, T. 1995, International Symposium on Electromagnetic Com- patibility, 57

Yi, S.-X., Gao, H., & Zhang, B. 2014, ApJ, 792, L21

You, X. P., Hobbs, G., Coles, W. A., et al. 2007, MNRAS, 378, 493

Zhou, B., Li, X., Wang, T., Fan, Y.-Z., & Wei, D.-M. 2014, Phys. Rev. D, 89, 107303

Zou, W. Z., Hobbs, G., Wang, N., et al. 2005, MNRAS, 362, 1189 A Glossary

↵ Dispersion index

Lorentz factor, 1/ 1 v2/c2 p Observing wavelength, m

⌫ Observing frequency, GHz or MHz

⌧c Characteristic age, yrs

⌧d Scattering timescale, s

B Magnetic field parallel to the line of sight, G k b Frequency channel width, MHz c Speed of light, 3 108 ms 1 ⇥ D Distance, kpc or Mpc

DL Luminosity distance, Gpc

Dcomov Comoving distance, Gpc

3 DM Dispersion measure, pc cm

3 DMGalaxy Modeled DM contribution from the Milky Way, from NE2001, pc cm e Elementary charge, 1.60 1019 C ⇥

EFRB Energy of fast radio burst, Joules

Fluence, Jy ms F 139 140 Appendix A. Glossary

1 G Telescope gain, K Jy

(`, b) Galactic longitude and latitude, deg

m Mass of the electron, 9.11 10 31 kg e ⇥ 3 ne Electron density, cm

P Pulsar period, s or ms

1 P˙ Pulsar period derivative, s s

2 RM Rotation measure, rad m

S Flux density, Jy

S/N Signal-to-noise ratio

t Dispersive time delay, ms

tDM Dispersion-induced time delay from intra-channel smearing, s

TB Brightness temperature, K

Tsys System temperature, K

Tsky Sky temperature, K

W Pulse width, ms

z Redshift B Derivation of Bayesian probabilities

This appendix presents a full derivation of Equation 4.3 as well as the basic statisti- cal grounding for why such an equation is used in the analysis of the FRB distribution throughout the thesis.

B.1 Derivation

Given an expected number of events ⌘1 for a specific survey and area of sky we can estimate the number of events expected over a different area of sky and survey design ⌘2. The two rates are linearly related such that

T2⌦2 ⌘2 = ⌘1  = ⌘1↵1 2 (B.1) T1⌦1 ! where T and ⌦ are the time on sky and the solid angle observed for each system and  is a coefficient used to describe any other necessary corrections. All these factors can be brought together as the constant ↵ that describes all corrections and conversions between the two surveys. Given the expected number of events for a survey, the probability of observing N events is given by

N ⌘ ⌘ e P (N ⌘)= (B.2) | (N + 1) where (N +1) = N!. The probability distribution described in Equation B.2 is Poissonian, where the number of trials is large, but the number of detected events can be small. This nicely describes the predicted FRB distribution and is suitable for this example. We can then use Bayes’ theorem to describe the distribution of the expected number of events as a function of the number of detected events as

141 142 Appendix B. Derivation of Bayesian probabilities

P (N ⌘)P (⌘) P (⌘ N)= | . (B.3) | P (N) Normalization of Equation B.3 with a flat prior for P (⌘) results in

N ⌘ ⌘ e P (⌘ N)= . (B.4) | (N + 1)

For a second set of observations with ⌘2 = ↵⌘, the probability distribution for the number of detected events N2 will be (per Equation B.2)

N2 ↵⌘ (↵⌘) e P (N2 ↵⌘)= . (B.5) | (N2 + 1)

Combining Equations B.4 and B.5 we can derive the distribution of N2 given the number of events detected in the first survey of N as

1 P (N2 N)= P (N2 ↵⌘)P (⌘ N)d⌘ | 0 | | Z N ↵⌘ 1 (↵⌘) 2 e P (N ⌘)P (⌘) = | (N1 + 1) P (N) Z0 (B.6) (↵⌘)N2 e ↵⌘ ⌘N e ⌘ = 1 (N + 1) (N + 1) Z0 1 N2 (1+N+N2) (1 + N + N2) = ↵ (1 + ↵) (N + 1)(N2 + 1) where the final equation represents the probability of detecting N2 events in a given survey based on a detection of N events by a different survey. In the case of the Parkes FRB observations the conversion factor ↵ is easily calculated as the observing system and observed area are equal for all observations made with BPSR and the Parkes multibeam receiver. In the case of Parkes surveys ↵ is the ratio of time on sky between surveys ↵ = T2/T1. This is the equation used to calculate probabilities of FRB detections in various surveys throughout this thesis.

B.2 Justification

As it currently stands, the population of FRBs is in the regime of small-number statistics. Only 21 bursts are known, but of those 17 have been found with the BPSR system at Parkes and thus are contained within a self-consistent sample. Deriving the probability of detecting new bursts in ongoing surveys becomes extremely important for telescope B.2. Justification 143 proposals as well as follow-up coordination. We believe the method outlined in the previous section is the best way to calculate these probabilities. Given the small population it may be tempting to calculate probabilities using a bi- nomial distribution, i.e. the probability of detecting N bursts in one survey rather than another is 1/2N . However, the binomial approach poorly reflects the observations taken to find the existing FRBs as it uses each burst as a ‘trial’ rather than taking into account the different times spent on sky for different surveys. The true number of ‘trials’ is very high. Each minute on sky (or 64 µs sample) might be considered a trial, in which case the components of the HTRU survey have thousands of trials each, but the probability of finding a burst within those trials is low. In such a case the Poisson distribution and the Bayesian approach provide the best reflection of the survey datasets.