Detection and Observable Behaviours of High Redshift Superluminous Supernovae

Christopher M. Curtin

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

August 28, 2019

Faculty of Science and Engineering Technology Swinburne University

i Abstract

We apply Lyman Break Galaxy (LBG) Selection and Monitoring to photometry from the ongoing Dark Energy Survey (DES), Survey Using DECam for Superluminous Super- novae (SUDSS) and Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP) to more efficiently identify high redshift (z & 2) superluminous supernovae (SLSNe) and enable spectroscopic follow-up near peak. SLSNe are a luminosity class of supernovae ∼10–100× more luminous than an average type-Ia supernova. We construct deep griz stacks with cadenced data from DES and the SUDSS, as well as uncadenced DECam archival data. Deep image stacks are required for effective selection of LBGs. We achieve gri(z) depths of ∼27.5 (27) mag on the two DES deep fields (6 deg2) and ∼26.5 (26) mag on two SUDSS fields (6 deg2). These stacks can be used to select LBGs at z ∼ 4 with no additional photometry. We initiated u0SUDSS, the DECam u0-band extension to SUDSS, to provide u0-band deep stack photometry (∼26.2 mag) on select fields. Deep u-band (3000–4000Å) photometry extends LBG selection to lower redshifts (z ∼ 2 and z ∼ 3) and increases selection sizes by up to a factor of 30. We construct deep u0-band stacks from proprietary and archival photometry on three DES/SUDSS fields (9 deg2) with a fourth field under way. This depth is sufficient to increase the LBG sample size by a factor of ∼15. We develop LBG colour selection criteria for DECam using composite LBG spectra and synthetic colours of model star-forming galaxy spectral templates measured from 0 < z < 6. We then refine the criteria with spectroscopic redshift confirmations of ∼100 sources observed with the Keck Low Resolution Imaging Spectrometer (LRIS) and several hundred archival spectroscopic redshifts. Efficiency estimates of the refined criteria range from 70–85% and completeness estimates range from 40–50%. The most developed catalogue on the DES C3 deep field (3 deg2) includes ∼150,000 sources. Extensive z ∼ 2 LBG selections (∼60,000 sources) are catalogued on two other fields (6 deg2) with deep u0-band photometry. We monitor LBGs in the DES C3 field for transient activity during the fourth and fifth years of the survey using the DES transient catalogue. We inspect the light curve behaviour of each event to rule out active galactic nuclei (AGN) and low redshift SNe- Ia. We identify 25 photometric candidates of interest over both seasons. High redshift SLSN candidates are considered for spectroscopic follow-up based on their confidence and brightness. Additional targets are provided by the Subaru HIgh-Z sUpernovae CAmpaign (SHIZUCA). ii

We perform four spectroscopic follow-up campaigns using Keck LRIS. Poor observing conditions and light cirrus during each campaign restricted the selection and number of targets and reduced the quality of collected spectra, with one campaign being completely weathered out. Spectra of three DES targets are successfully collected and reduced. Of these two events are considered to be likely low redshift supernovae and one is an AGN at z = 1.697. Spectra of ﬁve SHIZUCA photometric candidates are also collected and reduced. Two are low redshift supernovae and three are determined to be SLSNe at z = 1.851, 1.965 and 2.399, the latter being the highest redshift supernova observed spectroscopically near peak to date. The spectroscopic subtype of each event is indeterminate. With this thesis we demonstrate that deep follow-up spectroscopy of z ∼ 2–4 SLSNe near peak is achievable with modern surveys and 8m-class telescopes, even when classically scheduled. The three spectra of z ∼ 2 supernovae near maximum light obtained during this work nearly double the number of such spectra collected to date, and there are many more photometric candidates than are able to be targeted with the follow-up time awarded. Accumulating more spectra of SLSNe at redshifts of z ∼ 2–6 is crucial for recognizing and understanding their far-UV behaviours. This is a pre-requisite for identifying SLSNe at z ∼ 6–20 as future large telescopes and JWST will only have access to the restframe far-UV at these redshifts. Future work includes analysing all photometric SLSN candidates and obtaining spectroscopic redshifts of their hosts to constrain rates, identify general far-UV light curve behaviours and optimise the eﬃciency of LBG S&M for future surveys. iii iv Acknowledgements

I’d like to express my deepest gratitude to my advisory team, Professors Jeff Cooke, Emma Ryan-Weber and Jeremy Mould, for their steady support and guidance throughout my project. Thanks to Australia, Swinburne and CAS for the opportunity and resources to conduct my research and write my thesis. To my fellow team members, Igor, Uros, Garry, Trooper, Sara, Stephi, it was a pleasure working with you and thanks for all your help. To the science fiction book club, Colin, Rob, Stefan, our wildly speculative conversa- tions were both a scientific motivation and a much needed distraction for me near the end. Thanks for the ride. To those who wouldn’t leave me alone near the end, Yeshe, Rossana, Wael, and to those who kept tabs, of whom there are too many to list, I truly appreciate it. You held me accountable, as good friends do. To my family, you know yourselves, thanks for making me know that I was loved and believed in at every step along the way here. I love you, too. To the next gen, David Isaac, Samantha, Dain, Topher, you can do anything you put your mind to. And know that whatever that is, I’ll be cheering you on. A special thanks to the counsellors at Swinhealth who kept me sane at no charge. And I thank my God, for everything, always. v vi Declaration

The work presented in this thesis has been carried out in the Centre for Astrophysics & Supercomputing at Swinburne University of Technology between 2014 and 2019. 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 5 has been published as “First Release of High-redshift Superluminous Super- novae from the Subaru HIgh-Z sUpernova CAmpaign (SHIZUCA). II. Spectroscopic Properties” in The Astrophysical Journal Supplement Series, Volume 241, Issue 2, article id. 17, 14 pp. (2019). My contribution to this paper was the collection of the spectra, data reduction and analysis, and the writing of the text. My co-authors contributed in discussions about and revisions to early manuscript drafts, and certain specialised software tools used in the analysis of the reduced data.

Christopher M. Curtin Melbourne, Victoria, Australia August 28, 2019 vii viii

Dedicated to the memory of my dear Aunt Martha, 1951–2019, who loved me to pieces. See you among your stars Mart. 0.1. SURVEYS ix Acronymus, Abbreviations and Conventions

0.1 Surveys

• ASAS-SN: All-Sky Automated Survey for Supernovae

• CRTS: Catalina Real-time Transient Survey

• CFHTLS: Canada-France-Hawaii Telescope Legacy Survey

• DES: Dark Energy Survey

• GOODS: Great Observatories Origins Deep Survey

• HSC-SSP: Hyper-SuprimeCam Subaru Strategic Program

• HZT: High-Z Supernova Search Team

• PS1: Panoramic Survey Telescope And Rapid Response System 1

• PTF: Palomar Transient Factory

• SCP: Supernova Cosmology Project

• SDSS: Sloan Digital Sky Survey

• SHIZUCA: Subaru HIgh-Z sUpernovae CAmpaign

• SNLS: Supernova Legacy Survey

• SUDSS: Survey Using DECam for Superluminous Supernovae

• TSS: Texas Supernova Search

• USNO-B1: United States Naval Observatory B1.0 Catalogue

• u0SUDSS: The DECam u0-band SUDSS Extension

• VIDEO: VISTA Deep Extragalactic Observations Survey

• ZFOURGE: FourStar Galaxy Evolution Survey

• ZTF: Zwicky Transient Facility x

0.2 Observatories and Instruments

• AAO: Australian Astronomical Observatory

• ACS: Advanced Camera for Surveys

• CFHT: Canada-France-Hawaii Telescope

• COS: Cosmic Origin Spectrograph

• CTIO: Cerro Tololo Inter-american Observatory

• DECam: Dark Energy Camera

• HSC: Hyper-SuprimeCam

• HST: Hubble Space Telescope

• JWST: James Webb Space Telescope

• KDUST: Kunlun Dark Universe Survey Telescope

• LRIS: Low Resolution Imaging Spectrometer

• LSST: Large Synoptic Survey Telescope

• NOAO: National Optical Astronomical Observatories

• STIS: Space Telescope Imaging Spectrograph

• TAO: University of Tokyo Atacama Observatory

• VIRCAM: VISTA InfraRed CAMera

• VISTA: Visible and Infrared Survey Telescope for Astronomy

• VLT: Very Large Telescope Contents

Abstract i

Acknowledgements iii

Declaration v 0.1 Surveys ...... ix 0.2 Observatories and Instruments ...... x

Acronymus, Abbreviations and Conventions ix

List of Figures xiii

List of Tables xvii

1 Introduction 1 1.1 Background ...... 1 1.2 SLSN Subtypes and Mechanics ...... 3 1.2.1 SLSNe-I ...... 4 1.2.2 SLSNe-II ...... 6 1.2.3 PISNe ...... 9 1.2.4 P-PISN ...... 11 1.3 SLSNe at High Redshift ...... 12 1.4 Lyman Break Galaxy Selection & Monitoring ...... 15 1.5 Surveys and Fields ...... 16 1.6 Purpose of the Thesis ...... 20

2 Deep Stacking 21 2.1 Introduction ...... 21 2.2 Maximizing Depth ...... 23 2.3 Image Sets ...... 24 2.4 Software ...... 26 2.5 Astrometrics ...... 26 2.6 Image Set Calibration ...... 27 2.7 Image Stacking and Source Extraction ...... 30 2.8 Photometric Calibration ...... 33 2.8.1 Zeropoints ...... 36

xi xii CONTENTS

2.8.2 u0-band Rectiﬁcation ...... 37 2.8.3 Photometric Uncertainties ...... 39 2.9 Filter Crossmatching ...... 39 2.10 Depth Estimates and Conclusion ...... 41

3 DECam LBGS 45 3.1 Introduction ...... 45 3.2 Filter Sets ...... 48 3.3 LBG Composite Spectra ...... 48 3.4 Model Spectral Templates ...... 50 3.5 Model-Based Colour Selection Criteria ...... 53 3.6 Spectroscopic Colour Selection Criteria ...... 59 3.7 Photometric Eﬃciency Analysis and Conclusion ...... 63

4 DES Spectroscopy 71 4.1 Introduction ...... 71 4.2 LBG Monitoring ...... 73 4.2.1 Seasonal Stacks ...... 74 4.3 Photometric Candidate Evaluation ...... 74 4.4 Keck Spectroscopic Follow-up ...... 79 4.4.1 DES16C3bac ...... 79 4.4.2 DES16C3bn ...... 80 4.4.3 DES16C3cv ...... 83 4.4.4 Other Targets ...... 86 4.5 Conclusion ...... 86

5 SHIZUCA Spectroscopy 91 5.1 ABSTRACT ...... 91 5.2 INTRODUCTION ...... 92 5.3 OBSERVATIONS ...... 93 5.4 DATA REDUCTION AND ANALYSIS ...... 95 5.4.1 HSC16adga ...... 95 5.4.2 HSC17auzg ...... 101 5.4.3 HSC17dbpf ...... 102 5.5 DISCUSSION ...... 105 5.5.1 Nature ...... 105 CONTENTS xiii

5.5.2 Host Subtraction ...... 107 5.5.3 SLSNe-II ...... 108 5.5.4 SLSNe-I ...... 109 5.6 CONCLUSION ...... 111

6 Conclusion 115 6.1 Summary ...... 115 6.2 Future Work ...... 118

List of Figures

1.1 Photometric Deﬁnitions of SLSNe ...... 4 1.2 SLSNe-I ...... 5 1.3 SLSNe-II ...... 7 1.4 SLSNe-R ...... 10 1.5 SEDs of SLSNe and SLSNe-I as Standard Candles ...... 13 1.6 Redshift Dependent SLSN Rates and LBG Luminosity Functions ...... 14

2.1 Loss of Conservation of Source Flux in Image Stacks using High Resolution, Background-Derived Weightmaps ...... 29 2.2 Degenerecy between FWHM and Non-Photometric Extinction ...... 31 2.3 Optimization of Detection Thresholds ...... 34 2.4 MAG_AUTO vs. MAG_ISO ...... 35 2.5 Filter Throughput Response Functions of the DECam u0-band, the SDSS u-band and the MegaCam u*-band ...... 38 2.6 SExtractor Noise-Derived Magnitude Uncertainties Compared with Em- pirically Measured Values ...... 40 2.7 Source Recovery and Ambiguous Associations as a Function of Association Radius ...... 42

3.1 Filter Throughput Response Functions for the Four Filter Sets Discussed in Section 3.2 ...... 49 3.2 The Four z ∼ 3 LBG Composite Spectra from Shapley et al. (2003) . . . . . 51 3.3 Spectral Templates of Star-Forming Model Galaxies which Resemble LBGs in the Far-UV as Observed at z ∼ 3 ...... 52 3.4 Initial Redshift Bin Assignment for LBGS with DECam ...... 54 3.5 The Six Most Relevant Colour Planes for Establishing the Initial DECam LBGS Criteria ...... 55 3.5 Continued ...... 56 3.5 Continued ...... 57 3.6 The Six Most Relevant Colour Planes for Constraining the Initial Model- Based Selection Criteria using Observed Spectra ...... 64 3.6 Continued ...... 65 3.6 Continued ...... 66

xv xvi LIST OF FIGURES

4.1 ATC Five Season Light Curves of Three Transients Targeted in the 2016 Keck Spectroscopic Follow-up Campaign Overlaid with Seasonal and Detec- tion Stack Photometry ...... 75 4.2 DECam griz Light Curves of DES16C3bac Using DES Single Epoch Pho- tometry ...... 81 4.3 2500 × 2500 i-band Stamps Centred on the Host of DES16C3bac ...... 81 4.4 Flux-Calibrated Observer-Frame 1-D Spectrum of DES16C3bac ...... 82 4.5 The Spectrum of DES16C3bac Smoothed and Redshifted to z = 1.697 . . . 82 4.6 DECam griz Light Curves of DES16C3bn Using DES Single Epoch Pho- tometry ...... 84 4.7 2500 × 2500 i-band Stamps Centred on the Host of DES16C3bn ...... 84 4.8 Flux-Calibrated Observer-Frame 1-D Spectrum of DES16C3bn ...... 85 4.9 The Spectrum of DES16C3bn Smoothed and Redshifted to z = 0.601 . . . . 85 4.10 DECam griz Light Curves of DES16C3cv Using DES Single Epoch Pho- tometry ...... 87 4.11 2500 × 2500 i-band Stamps Centred on the Host of DES16C3cv ...... 87 4.12 Flux-calibrated observer-frame 1-D spectrum of DES16C3cv ...... 88 4.13 The Spectrum of DES16C3cv Smoothed and Redshifted to z = 3.747 . . . . 88 4.14 Two Important Colour Planes for Assessing the Conﬁdence of HR SLSN Candidates Based on the Colours of Their Hosts ...... 90

5.1 The HSC-grizy Light Curves of HSC16adga, HSC17auzg and HSC17dbpf . 96 5.2 Maps of the Spectral Slits Used for the Observations of HSC16adga, HSC17auzg and HSC17dbpf ...... 97 5.3 Flux-Calibrated Observer-Frame 1-D Spectrum of HSC16adga and the Same Spectrum Zoomed-in, Smoothed and Redshifted to z = 2.399 ...... 99 5.4 MIZUKI Narrow-Band Redshift Probability Distributions for Each Event . . 101 5.5 Flux-Calibrated Observer-Frame 1-D Spectrum of HSC17auzg and the Same Spectrum Zoomed-in, Smoothed and Redshifted to z = 1.965 ...... 103 5.6 Flux-Calibrated Observer-Frame 1-D Spectrum of HSC17dbpf and the Same Spectrum Zoomed-in, Smoothed and Redshifted to z = 1.851 ...... 104 5.7 All Three Supernova Spectra Including Their Host Galaxy Contributions, and a QSO Composite and the TDE, ASASSN-14li ...... 106 5.8 The Three Pseudo-Host-Subtracted Supernova Spectra, and the SLSN-II, LSQ15abl, and 2 UV-Bright SNe-IIn ...... 110 LIST OF FIGURES xvii

5.9 The Three Pseudo-Host-Subtracted Supernova Spectra, and the High Red- shift SLSN-I, DES15E2mlf, and the Two Low Redshift SLSNe-I ...... 112

List of Tables

1.1 Survey speciﬁcs of the photometry utilised for this thesis ...... 18 1.2 Sky ﬁelds covered by the u0SUDSS program ...... 19

2.1 Astrometric precision of DECam C3 sources ...... 28 2.2 Source recovery in photometric imagery compared to extincted images and the eﬀects of including extincted images in stacks without ﬂux scales . . . . 32

2.3 Median values over the 62 CCD DECam mosaic for the mˆ0 terms of Equa- tion 2.6 ...... 37 2.4 DECam u0griz 5σ limiting magnitudes of C3 deep stacks ...... 43

3.1 LBGS criteria of the Steidel UnGRi filter set ...... 47 3.2 z ∼ 2 LBGS criteria of selected instruments ...... 60 3.3 z ∼ 3 LBGS criteria of selected instruments ...... 61 3.4 z ∼ 4 LBGS criteria of selected instruments ...... 62 3.5 Keck LRIS spectroscopic follow-up campaigns ...... 62 3.6 DECam LBGS efficiencies of Keck LRIS and ZFOURGE spectra ...... 67 3.7 DECam LBGS efficiencies based on ZFOURGE photometric redshifts of the C3 deep stack source catalogue ...... 68 3.8 Memberships and densities of DECam LBGS catalogues from the C3 field . 69

4.1 Per-epoch and seasonal stack depths in C3 over the ﬁrst three seasons of DES 74 4.2 HR SLSN photometric candidates identiﬁed in C3 with ATC using LBG S&M 78 4.3 Keck spectroscopic follow-up of DES LBG S&M photometric candidate HR SLSNe ...... 80

5.1 SHIZUCA spectroscopic follow-up targets ...... 98

xix

1 Introduction

1.1 Background

Supernovae are some of the brightest optical transient phenomena known to occur in nature. They signal the sudden explosion of stellar masses worth of material. Almost all observed supernovae are thought to arise through one of two explosion mechanisms, core-collapse (CC-SNe) or thermonuclear runaway (SNe-Ia; Filippenko, 1997).

For zero-age main sequence (ZAMS) masses greater than ∼8M , the cores of stars collapse into compact remnants, either neutron stars or black holes, after exhausting their nuclear fuel supply. For a wide range of masses and circumstances the infall and rebound of the outer layers is explosive, synthesizing large masses of radioactive 56Ni which is expelled along with a massive amount stellar debris in the form of a hot, rapidly expanding ejecta shock wave, a CC-SN. As the hot ejecta expand the radius grows, increasing the luminosity of the supernova photosphere. The increase in luminosity is restricted by the adiabatic nature of the ejecta expansion, exchanging heat for volume. Photospheric temperatures drop as the radius increases and the supernova evolves in colour from blue to red. The rate of heat loss is itself slowed by the mixed-in 56Ni as it decays into 56Co, providing the ejecta with an additional, regulated energy source suﬃcient to power the event for days to years (Barbon et al., 1984). There is a wide variety in the appearance of CC-SNe due to the number of components at work (see Smartt, 2009 for a review of the mechanics of the diﬀerent types of CC-SNe). Some of the more common types are those with an intact hydrogen envelope (SNe-II) and those which have had their envelopes stripped to varying degrees (SNe-Ibc). Among these latter stripped-envelope supernovae are a class of events called broad-lined SNe-Ic which are associated with long gamma ray bursts (LGRBs) and higher than average ejecta velocities (Galama et al., 1998). Among H-rich supernovae are those which dim at a linear

1 2 CHAPTER 1. INTRODUCTION pace consistent with the decay of radioactive 56Ni freshly synthesized in the explosion itself (SNe-IIL), and those which plateau for an extended period due to a sudden increase in opacity as hydrogen in the outer photosphere is ionised (SNe-IIP). Another relatively rare class of SNe-II exhibits multi-component hydrogen emission features, including narrow features (SNe-IIn), indicating that circumstellar medium interaction (CSMI) is taking place between the ejecta and a cold, slowly moving CSM. CSMI converts the kinetic energy of supernova ejecta into heat and electromagnetic radiation with an eﬃciency dependent on the properties of the system components. The wide variety of possible conﬁgurations makes the SN-IIn subtype very diverse, including some of the brightest and hottest colour- temperature supernovae observed among the ordinary classes (Smith, 2017).

Unlike CC-SNe, SNe-Ia occur when a mass-accreting white dwarf (WD) reaches the point of carbon ignition. The ignition adds heat unchecked by thermal expansion (a thermonuclear runaway event) until the degeneracy pressure on the WD is overcome and the entire remnant is unbound in a brilliant and sudden explosion. There are two competing theories as to the circumstances of the mass accretion. In the single-degenerate scenario, mass accretion proceeds steadily onto the WD from the envelope of a nearby main sequence companion overﬂowing its Roche lobe (Whelan & Iben, 1973). In the double-degenerate scenario, the tight orbit of a pair of WDs decays via gravitational radiation until the two coalesce (Iben & Tutukov, 1984; Webbink, 1984). SNe-Ia resemble some CC-SNe photometrically, but they are readily distinguishable by their spectra which lack both hydrogen and helium and exhibit broad silicon absorption dips (Filippenko, 1997). The light curves of SNe-Ia are remarkably uniform and there is a tight relationship between the peak magnitude and rate of decline (the Phillips relation, Phillips, 1993), making them ideal for use as standard candles. The potential for using SNe-Ia to measure cosmological distances motivated surveys dedicated to their detection, most notably the Supernova Cosmology Project (Perlmutter et al., 1997) and the High-Z Supernova Search Team (Schmidt et al., 1998), which culminated in the discovery and ﬁrst measurements of the acceleration of the Universe (Riess et al., 1998; Perlmutter et al., 1999).

Over the past two decades, due in large part to the success of the pioneering SCP and HZT surveys in constraining the Hubble constant, and also to the steady advancement of observational technology, both the number of large survey collaborations and the total surveyed sky volume have increased steadily. The optical transient discovery space has grown immensely, and identified and categorized supernovae now number in the tens of thousands thanks to completed and ongoing wide-area surveys such as (see Acronymus) SNLS (Astier et al., 2006), TSS (Quimby, 2006), CRTS (Drake et al., 2009), PTF (Law 1.2. SLSN SUBTYPES AND MECHANICS 3 et al., 2009), PS1 (Kaiser et al., 2010), ASASSN (Shappee et al., 2014), DES (Dark Energy Survey Collaboration et al., 2016), ZTF (Bellm & Kulkarni, 2017) and HSC-SSP (Aihara et al., 2018). The expansion of the field has been rapid and profound, advancing our physical understanding of supernovae but also revealing a much richer diversity within the class than previously realized, spawning blended subtypes, hybrid subtypes and entirely new subtypes. Superluminous supernovae (SLSNe) are a luminosity class of supernovae ∼10–100× brighter at peak than an average SN-Ia. The prototypical events which motivated the creation of the SLSN class were discovered just over a decade ago (Smith et al., 2007; Quimby et al., 2007; Gal-Yam et al., 2009). These early SLSNe had a high impact on the supernova community as it was difficult to reconcile the observed energy output with contemporary explosion models. Continued discoveries of similar bright events prompted an early review (Gal-Yam, 2012, G12 hereafter) which provided the first definition of SLSNe as supernovae that exceed an absolute magnitude of M = −21 in any band. G12 populated the new class with 18 events, including a rate estimate of ∼0.001× the global rate of CC-SNe (Quimby et al., 2013) and a host analysis suggesting a strong preference for low-metallicity dwarf galaxies with high specific star formation rates (sSFRs; Neill et al., 2011). G12 divided the class into 3 subtypes based on differences in observables and the implications for distinct explosion scenarios (see Section 1.2). More recent reviews provide an advanced but still incomplete picture of the true nature of SLSNe (Moriya et al., 2018; Howell, 2017).

1.2 SLSN Subtypes and Mechanics

As a luminosity class, SLSNe are described in terms of their photometry. Spectroscopic follow-up is then used to assign a subtype. The G12 definition of SLSNe (M ≥ −21) is set to 1.5 mag brighter than the peak B-band luminosity of SNe-Ia (Figure 1.1, left). This peak represents a sharp cut-off in the luminosity distribution of ordinary supernovae, with brighter events considered over-luminous (Richardson et al., 2002). The G12 definition assumes an underpopulated luminosity gap between ordinary supernovae and SLSNe and is set to distinguish unusual, over-luminous supernovae from the more uniform SLSN subtypes. But developments in the field of SLSNe suggest the G12 limit is too strict, excluding a significant number of related but slightly dimmer events (Nicholl et al., 2015). Furthermore the limit applies to any band, but was drawn from predominantly optical magnitudes and is less reliable outside of this wavelength range. Still this early definition served to populate and develop the fledgling SLSN class, 4 CHAPTER 1. INTRODUCTION

Figure 1.1 Left: The G12 definition of SLSNe. Below the threshold are representative light curves of ordinary supernova subtypes. The SLSNe are separated into 3 subtypes (see Section 1.2). Right: A photometric description of SLSNe-I from Nicholl et al. (2015). A continuum of optical light curves of SLSNe-I are used to form a light curve region representative of the class, and a similar region is shown for SNe-Ibc (Taddia et al., 2015). After scaling, the regions are markedly similar. and continues to offer a benchmark for describing observed supernovae as superluminous with only minimal photometric information. The G12 definition of SLSNe is not the only one that has been proposed. Quimby et al. (2013) suggest defining SLSNe with a softer minimum magnitude of MR = −20.5, or ∼ 30× brighter than an average SN-IIn (Li et al., 2011). This definition is more inclusive of events spectroscopically related to SLSNe, but is specific to the R-band though it applies reasonably well throughout the optical. Nicholl et al. (2015) offer a more developed photometric description of H-poor SLSNe, given as supernovae whose light curves resemble those of SNe-Ic brightened by a factor of ∼25 and time-extended by a factor of ∼3 (Figure 1.1, right). While this description only applies to a particular SLSN subtype, it provides more natural, wavelength-dependent photometric criteria for inclusion. Advances in SLSN spectroscopy have made possible spectroscopic identification of some SLSNe independent of photometry. This has sparked some debate in the community over the specific usage of the SLSN label (Howell, 2017). In general though, discussions of SLSNe are referring to one or more subtypes of supernovae that are on average significantly over-luminous.

1.2.1 SLSNe-I

The SLSN-I subtype refers to H-poor SLSNe just as the SN-I subtype refers to H-poor SNe. The ﬁrst well-known case of a SLSN-I is that of SN2005ap (Quimby et al., 2007). 1.2. SLSN SUBTYPES AND MECHANICS 5

Observations of similar events over the next several years were eventually grouped into a new class (Barbary et al., 2009; Pastorello et al., 2010; Quimby et al., 2011). The events are united by their similar and peculiar spectra which all have several broad absorption dips in common around restframe ∼4300Å attributed to O-II (Quimby et al., 2011; Fig- ure 1.2, right). Many more absorption dips appear with varying regularity and strength throughout the optical and ultraviolet (UV; 1000-4000Å) of SLSN-I spectra, and the work of determining the cause and signiﬁcance of each dip is ongoing (Dessart et al., 2012; Maz- zali et al., 2016; Yan et al., 2017a, 2018). At present the SLSN-I classiﬁcation is most often assigned from spectra alone and without the enforcement of a magnitude limit. The late-time spectra of SLSNe-I resemble the spectra of stripped-envelope SNe-Ic near peak (Pastorello et al., 2010), so much so that the class is often referred to as SLSNe-Ic throughout the literature (e.g., Inserra & Smartt, 2014; Nicholl & Smartt, 2016; Moriya et al., 2018; Quimby et al., 2018; Dessart, 2019). The average peak bolometric magnitude of SLSNe-I is an over-luminous –20.7 mag with a narrow 0.5 mag dispersion (Nicholl et al., 2015; Fig- ure 1.2, left). In general SLSNe-I are found to be over-luminous in both the optical and UV alike (Quimby et al., 2011; Yan et al., 2017a, 2018; Quimby et al., 2018).

Currently the strongest hypothesis for the explosion mechanism of SLSNe-I is the central engine (Woosley, 2010; Kasen & Bildsten, 2010). In the central engine scenario the compact remnant of a CC-SN imparts energy into the expanding supernova ejecta, slowing the loss of heat during adiabatic expansion and extending the rise to peak into superluminosity. The nature of the engine is suspected to be that of a spinning down, nascent magnetar. Magnetars are rapidly spinning neutron stars with strong magnetic ﬁelds (see Kaspi & Beloborodov, 2017 for a review). The spin period of a neutron star increases over

Figure 1.2 Left: A continuum of light curves of bright SLSNe-I from Moriya et al. (2018). Right: The spectra of SLSNe-I exhibit multiple broad absorption dips. The O-II absorption series in particular is regarded as a defining feature of the class (Quimby et al., 2011). 6 CHAPTER 1. INTRODUCTION time as magnetic dipole radiation saps the energy well of angular momentum, a process called spin-down. This process can be accelerated through interaction with a dense medium such as supernova ejecta, efficiently converting magnetic dipole radiation into kinetic and radiative energy. For massive magnetars with short periods an enormous amount of spin- down energy is available of the order hypothetically necessary to energize supernova ejecta to superluminous proportions. Photometric models of magnetar-driven supernovae have been markedly successful at fitting a wide variety of SLSN-I multi-band light curves by tuning a small number of free parameters over a realistic range. By varying the spin period, magnetic field strength and energy conversion efficiency of the magnetar and matching the observed kinetic energy and mass of the ejecta, the light curves of many spectroscopically confirmed SLSNe-I are confidently recovered (Nicholl et al., 2015, 2017). Spectroscopic modelling of magnetar-driven supernovae has also been highly produc- tive. Models using simple assumptions about the rate of spin-down energy infusion and the composition of the ejecta can reproduce many of the absorption dips observed in the spectra of SLSNe-I, providing the means to identify these features based on theory and not solely wavelength (Dessart et al., 2012; Mazzali et al., 2016; Dessart, 2019). The rates and hosts of SLSNe-I are also consistent with the magnetar scenario. SLSNe- I show a strong preference for low-metallicity dwarf galaxy hosts with high sSFRs, similar to the hosts of LGRBs (Lunnan et al., 2014; Angus et al., 2016; Perley et al., 2016). It has long been suspected that LGRBs are produced by magnetars (Duncan & Thompson, 1992), and it has been suggested that SLSNe-I are a different manifestation of the same phenomenon based on the resemblance of their late-time nebular spectra to those of supernovae associated with LGRBs (Nicholl et al., 2016; Jerkstrand et al., 2016). This picture is supported by observations of a possible bridging event, an over-luminous SN-Ic, SN2011kl, associated with an ultra-long GRB (Greiner et al., 2015). SLSNe-I occur 1000–20000× less frequently than ordinary CC-SNe (Quimby et al., 2013), and to first order this rate is consistent with expectations on the rate of magnetar formation (Nicholl et al., 2017).

1.2.2 SLSNe-II

SLSNe-II are described in G12 as H-rich SLSNe. Most SLSNe-II exhibit strong, narrow hydrogen emission and are UV-luminous, similar to SNe-IIn (Ofek et al., 2007; Smith et al., 2007; Drake et al., 2010; Chatzopoulos et al., 2011; Rest et al., 2011). These events are often referred to as SLSNe-IIn in the literature (Nicholl et al., 2015; Howell, 2017; Moriya et al., 2018). There are examples of SLSNe-II that do not exhibit strong narrow hydrogen 1.2. SLSN SUBTYPES AND MECHANICS 7

Figure 1.3 Left: A continuum of optical light curves of SLSNe-II from Moriya et al. (2018). Right: A time-series of spectroscopic observations of the most well-studied SLSN-II, SN2006gy, from Smith et al. (2010). The spectra exhibit an evolving, multi-component H-α emission feature suggesting CSMI is producing the excessive luminosity similar to over-luminous SNe-IIn. emission, but rather are spectroscopically similar to SLSNe-I with broad hydrogen emission (Miller et al., 2009; Gezari et al., 2009; Benetti et al., 2014; Inserra et al., 2018b). These are often grouped with SLSNe-I in population studies based on the assumption that the mechanism which causes SLSNe-I need not be subject to completely H-free environments (e.g., Inserra et al., 2013; Perley et al., 2016). This is also supported by the observation of late-time hydrogen emission in a signiﬁcant fraction of SLSNe-I (Yan et al., 2017b).

The ﬁrst and still best observed SLSN-II is SN2006gy (Ofek et al., 2007; Smith et al., 2007, 2008b, 2010). At the time of discovery it was the most luminous supernova ever reported (Smith et al., 2007). The presence of strong, narrow hydrogen emission immediately suggests that the event is powered by CSMI just as other SNe-IIn (see Figure 1.3, right). However when modelled it is found that the CSMI indicated by the strength of the narrow hydrogen emission lines in SN2006gy cannot reproduce the observed luminosity. This led to the development of a new type of CSMI model called shell-shocked diﬀusion (SSD; Smith & McCray, 2007; see also Chevalier & Irwin, 2011).

In the SSD model, a CC-SN occurs within an opaque, unbound shell of material on the 15 order of ∼10M at a large radius of ∼10 cm. As the supernova collides with the shell, the kinetic energy of the ejecta is efficiently reprocessed into electromagnetic radiation effectively transforming the shell into the photosphere of the supernova. Because of the large initial radius of the shell and high energy of interaction, the supernova attains a much 8 CHAPTER 1. INTRODUCTION greater UV-optical luminosity (Smith et al., 2008b). Evidence for SSD in SN2006gy is drawn from the observed hydrogen emission, which includes broad, intermediate and narrow components. Treating the broad and intermediate components as the primary CSMI zone, the strength of these components combined with a large radius in a shell geometry can successfully reproduce the observed luminosity. In this interpretation the narrow component of the hydrogen emission in SN2006gy arises as neutral hydrogen external to the shell is ionised when the shell transforms into the supernova photosphere (Smith et al., 2010). The shell configuration is not essential because the same CSM mass distributed uni- formly in a spherical geometry would produce the same effect (Ginzburg & Balberg, 2012). However the shell configuration is suggested because the production of such shells has a natural precedent in the extreme mass loss events of luminous blue variables (Humphreys, 1999; Tominaga et al., 2008), whereas to account for a spherical geometry the necessary steady stellar wind intensity is difficult to reconcile with observations. Whether assuming a shell or a sphere, the multi-band light curves of many SLSNe-II are reproducible with models based on the same principle as SSD (Chevalier & Irwin, 2011; Ginzburg & Balberg, 2012; Chatzopoulos et al., 2013). The peak magnitudes and light curve shapes of SLSNe-II are more diverse than SLSNe- I, similar to the diversity observed in SNe-IIn (Figure 1.3, left). There are likely many varied geometric configurations of CSM that will result in SLSNe-II (Moriya et al., 2018). It is suggested that SLSNe-II represent the brightest SNe-IIn in a normal distribution of peak magnitudes (Howell, 2017) but this is not necessarily true (Richardson et al., 2014). In SNe-IIn the CSMI producing narrow hydrogen emission is also a main source of luminosity. However in the SSD scenario of SLSNe-II, the CSMI producing the narrow hydrogen emission provides only a minor contribution to the total observed luminosity. It is the CSMI responsible for the broad and intermediate components of hydrogen emission that powers the display, perhaps highlighting a fundamental difference in explosion mechanisms. This difference may serve as a means of distinguishing SNe-IIn and SLSNe-II more precisely than can be accomplished using only peak magnitudes. A bridging event has been observed in the SN-IIn, SN2006tf, which when fit by models is found to be powered by similar contributions from both CSMI energy reservoirs (Smith et al., 2008a). The rate of SLSNe-II is poorly constrained because there are fewer published events to draw from. An early estimate using a minimum absolute peak magnitude limit of -20.5 mag finds the rate of SLSNe-II is ∼5× the rate of SLSNe-I, or 200–4000× rarer than ordinary CC-SNe (Quimby et al., 2013). One possible reason fewer events have been observed than 1.2. SLSN SUBTYPES AND MECHANICS 9

SLSNe-I (aside from the possibility of a stronger interest in SLSNe-I within the community) is that conﬁrming SLSNe-II from the detection of narrow H-α in optical spectra is limited to z . 0.5. If SLSNe-II do in fact represent the bright end of a normal distribution of SNe-IIn, the rate is a function of the luminosity distribution of SNe-IIn and the applied SLSN magnitude limit. But if there is a fundamental diﬀerence between SLSNe-II and SNe-IIn the rates are largely independent of one another. Regardless of the distinction between SLSNe-II and SNe-IIn, the early SLSN-II rate is consistent with model indications of extremely massive progenitors (Chatzopoulos et al., 2013). The hosts of SLSNe-II also corroborate this picture, exhibiting high sSFRs like those of SLSNe-I, though without as pronounced a dependence on metallicity (Neill et al., 2011; Perley et al., 2016).

1.2.3 PISNe

Pair-instability supernovae (PISNe) are a type of supernova known to exist only theoretically (Barkat et al., 1967; Rakavy & Shaviv, 1967; Rakavy et al., 1967). As stars with

ZAMS masses of ∼140–260M reach the end of their fuel supply, the extreme temperatures and pressures in the core soften the equation of state to the point that the production of positron-electron pairs becomes unstable (Heger & Woosley, 2002). After, the fusion energy output into pressure-supporting photons is reduced and no longer suﬃcient to prevent gravitational collapse. As the collapse proceeds, the core temperature rises until the ignition of oxygen burning at which point the collapse is violently reversed into a thermonuclear runaway. The circumstances leading up to the runaway provide it with the energy necessary to completely unbind the star, leaving no remnant (though see Section 1.2.4). 56 The resultant supernova ejecta is imbued with up to 40M of Ni and presents as a long, bright SLSN with ∼100× the energy of the average SN-Ia, ∼10× the peak luminosity, and ∼10× the duration. (Kasen et al., 2011).

Whether a star can maintain a core consistent with a ZAMS mass of 140M or more is an open question, and to date no supernova has been observed with undisputed PISN origins. The high initial mass requirement implies that if PISNe do occur in nature, they are exceptionally rare outside of environments with top-heavy initial mass functions (IMFs). In addition most PISNe are expected to resemble average CC-SNe and it is diﬃcult to distinguish any PISN minority from the larger general CC-SN population (Dessart et al., 2012; Kozyreva et al., 2018). Only the most massive (and presumably rarest) examples are expected to clearly distinguish themselves from CC-SNe as both superluminous and excessively long-lived. The long duration of PISNe, particularly the very gradual rise, distinguishes them from other SLSNe-I and SLSNe-II (Heger & Woosley, 2002). Also 10 CHAPTER 1. INTRODUCTION

Figure 1.4 Left: The light curve of SN2007bi compared to model light curves of PISNe of diﬀerent ZAMS masses from Gal-Yam et al. (2009) and Heger & Woosley (2002). Right: Light curves of PS1-11ap and PTF12dam, two SLSNe-I with declines very similar to SN2007bi but detected earlier in their evolution. The observed, rapid rise times are inconsistent with PISN models (Nicholl et al., 2013). unlike the characteristically blue SLSN-I and SLSN-II classes, PISNe are expected to be exceptionally red due to strong line blanketing from massive amounts of synthesized Fe, low photospheric temperatures and slow expansion velocities (Kasen et al., 2011).

The advent of SLSNe revived the concept of PISNe, which was immediately considered a prime candidate explosion mechanism. Both SN2005ap and SN2006gy evolved too quickly to have been caused by PISNe according to the best models available at the time (Heger & Woosley, 2002), but PISN variant scenarios were suggested for each (Woosley et al., 2007; Quimby et al., 2007). Then the SLSN, SN2007bi was observed with a very gradual decline precisely consistent with the latest PISN model predictions (Gal-Yam et al., 2009; Heger & Woosley, 2002; Figure 1.4, left). The case was so convincing that exclusive from SLSNe-I and SLSNe-II, a third subclass of SLSNe-R was proposed in G12. SLSNe-R (Radioactive) are defined as SLSNe that evolve on timescales consistent with the radioactive decay of massive amounts of 56Ni that presumably can only be produced in PISNe. Objects similar to SN2007bi have been observed and 5 SLSN-R candidates are discussed in G12. Since serious discussions of SLSNe-R began, the theoretical field of PISNe has un- dergone dramatic development with several teams exploring different models (Kasen et al., 2011; Chatzopoulos & Wheeler, 2012; Kozyreva et al., 2014; Chen et al., 2014; Woosley, 2017).

Critics of the PISN explanation of SN2007bi point out that the colour of SN2007bi is 1.2. SLSN SUBTYPES AND MECHANICS 11 theoretically inconsistent with a PISN and that spectra of SN2007bi, while not inconsistent with PISN models, are conspicuously similar to the unique spectra of SLSNe-I (Dessart et al., 2012). Now that the magnetar model of SLSNe-I is more developed, it is shown to be versatile enough to mimic the theoretical decline of PISNe and also fits observations of SN2007bi and other SLSNe-R, including the bluer colours (Inserra et al., 2013; Nicholl et al., 2015). The magnetar model predicts another strong photometric distinction to PISNe in the rise time. While the magnetar model can produce a very gradually declining SLSN-I, it always predicts a significantly shorter rise time than PISN models. The rise of SN2007bi is not sufficiently constrained to distinguish the models. However, observations of the SLSNe-I, PS1-11ap and PTF12dam impressively demonstrate magnetar model predictions of slowly declining SLSNe-I with rapid rises and convincingly argue that SLSNe-I and SLSNe-R share a common explosion mechanism (Nicholl et al., 2013; Figure 1.4, right). Over time the SLSN-R classification has fallen into disuse, and formerly classified SLSNe-R have been reclaimed as SLSNe-I in recent population studies (Lunnan et al., 2014; Nicholl et al., 2017; Howell, 2017). Though no undisputed PISN has yet been observed, this does not imply that they do not occur. The extreme ZAMS masses required to produce the brightest possible PISNe exceed the theoretical upper mass limits of all but the most metal poor stars (Banerjee et al., 2012). Modelers have long suggested the environments of metal-free Population III stars (Pop III) are the most conducive to PISN production (Heger & Woosley, 2002). With a not unlikely top-heavy Pop III IMF (Marks et al., 2012), PISNe may be fairly common in the very early Universe. The observability of Pop III PISNe by JWST and other future facilities is a topic of analysis (Scannapieco et al., 2005; de Souza et al., 2013; Whalen et al., 2013).

1.2.4 P-PISN

Leading up to the 140M minimum ZAMS mass of PISNe, from ∼80–140M stars en- counter the pair-instability process but do not completely unbind. Instead they experience energetic pulsations which can cause extreme mass loss events (Barkat et al., 1967). The pulsations themselves resemble dim supernovae and can be singular or reoccurring (Heger & Woosley, 2002). In the latter case, separate massive unbound shells which collide can produce supernova-like transients of a bright and long-lasting nature (Woosley, 2017). These pulsations and pulsational interactions are referred to as pulsational PISNe (P-PISNe). P-PISNe do not appear to be a viable explanation for SLSNe. The most energetic simulated pulsational interactions generate less than half the total energy of the least bright 12 CHAPTER 1. INTRODUCTION

SLSNe Woosley (2017). However, P-PISNe do present a natural means of producing dense shells of CSM such as are invoked in the SSD model of SLSNe-II. Models suggest that the remnants of P-PISNe almost certainly collapses to black holes and releases no additional energy beyond what is imparted to the shells (Heger & Woosley, 2002; Kasen et al., 2011). But it is not yet clear from models whether this fate is unavoidable or if certain conditions such as rapidly rotating progenitors allow for an energetic core-collapse. By artiﬁcially inducing an energetic CC-SN at the centre of a massive P-PISN shell at an optimal radius, simulations of the resultant interaction between ejecta and shell are as energetic as SLSNe (Woosley, 2017).

1.3 SLSNe at High Redshift

The extreme luminosity of SLSNe has obvious implications on their observability at high redshift (HR; z & 2). Massive progenitors invoked in models of SLSNe suggest no redshift cut-oﬀ for detection back to the beginning of star formation at z ∼ 20–30. This is in contrast to the expected SNe-Ia redshift delay until z . 2 due to the time necessary to evolve the required white dwarf progenitor systems (Madau et al., 1998, but see also Maoz, 2010). The bright UV luminosity of SLSNe (Figure 1.5, left) makes them detectable in optical photometry to z ∼ 6 and photometric candidates approaching this limit have been drawn from modern surveys (Cooke et al., 2012; Mould et al., 2017; Moriya et al., 2019). Deep infrared (IR; 1–10 micron) surveys can push the detection limit out to z ∼ 20 and the beginning of star formation, providing a promising avenue for the direct observation of metal-free Pop III environments. Observations of SLSNe at such high redshifts will provide empirical data on the SFRs and IMFs of primordial galaxies along with diagnostic information via temporary illumination of the local interstellar medium. Such observation are also useful for probing the substantial intergalactic medium along the line-of-sight (Berger et al., 2012). In addition SLSNe-I show potential for use as standard candles (Figure 1.5, right), and may enable extended measurements of the Hubble constant well into the epoch of deceleration (Inserra & Smartt, 2014; Inserra et al., 2018a). There are rate-related advantages to surveying for SLSNe at high redshift. The rarity of SLSNe is due in large part to the high expected ZAMS mass requirements on the progenitors. The rates of SLSNe then depend on the rate of star formation (Figure 1.6, left). The peak of star formation is observed at z ∼ 2 (Madau & Dickinson, 2014) implying that the volumetric rates of SLSNe are higher at z ∼ 2 than at lower redshifts. This conclusion is consistent with measured rates of SLSNe (Quimby et al., 2013; Prajs et al., 2017; Moriya et al., 2019). The preference of SLSNe to low metallicity hosts suggests that 1.3. SLSNE AT HIGH REDSHIFT 13

Figure 1.5 Left: A comparison of ordinary supernova spectra to spectra of SLSNe from Quimby et al. (2018). Note that the spectra of SLSNe are signiﬁcantly steeper towards shorter wavelengths than ordinary supernovae, indicating bluer colours and greater UV-luminosity. SNe-Ia also have steep blue SEDs at very early times, but exhibit a sharp drop shortward of ∼4000Å from heavy line-blanketing (Foley et al., 2012). Right: From Inserra et al. (2018a) a relation between the peak absolute magnitude and the rate of decline of SLSNe-I similar to the Phillips relation for SNe-Ia is considered. The relation suggests the use of SLSNe-I as standard candles for measuring distances far in excess of what is possible using SNe-Ia. 14 CHAPTER 1. INTRODUCTION

Figure 1.6 Left: From Moriya et al. (2019), measurements of the rates of SLSNe at different redshifts from multiple studies. Changes in the rates with redshift are consistent with the changing rate of star formation. Right: LBG luminosity functions at z ∼ 2, z ∼ 3 and z ∼ 4 from Parsa et al. (2016). We use these functions to estimate the expected density of LBGs at different redshifts on sky based on the limiting magnitude of the photometry (see Chapter 3). the massive progenitors of SLSNe may also be subject to a metallicity threshold. In this case the rates of SLSNe may continue to increase beyond the peak of star formation. High redshift rates based on photometric examples of SLSNe support this possibility (Cooke et al., 2012). PISN rates are also likely heavily dependent on metallicity (Limongi & Chieffi, 2018). If PISNe are exclusively limited to metal-free Pop III environments they may never be detected in the local Universe. However, pockets of pristine, metal-free gas are observed at redshifts as low as z ∼ 3 (Fumagalli et al., 2011). Models of PISNe predict very red colours, but significant flux is still expected at wavelengths as short as ∼2000Å (Kasen et al., 2011; Cooke et al., 2012). At z ∼ 3, PISNe from metal-free progenitors are potentially detectable as transients that only brighten in redder optical bands.

The over-luminous aspects of SLSNe in the UV prompt observations in this wavelength range. Far-UV spectra of SLSNe-I reveal absorption dips that can be used for classiﬁcation in addition to the characteristic optical features (Yan et al., 2017a, 2018). Far-UV studies of SNe-IIn demonstrate that narrow hydrogen emission from CSMI includes narrow Ly-α emission (Fransson et al., 2002, 2005, 2014). This may also be the case for SLSNe-II, though to date this is only supported by the single SLSN-II far-UV spectrum, that of LSQ15abl (Quimby et al., 2018). Far-UV spectroscopy of very low redshift SLSNe must be collected from space-based observatories, mainly HST, and is further limited by the low rate of SLSNe. However at high redshift this information is pushed into the optical, 1.4. LYMAN BREAK GALAXY SELECTION & MONITORING 15 enabling far-UV spectroscopy of SLSNe from the ground. At present there are more far-UV spectra of SLSNe at high redshift than at low redshift (Berger et al., 2012; Howell et al., 2013; Pan et al., 2017; Smith et al., 2018; Curtin et al., 2019).

1.4 Lyman Break Galaxy Selection & Monitoring

The standard technique of supernova detection involves observing a specific area of sky over many epochs and subtracting a significantly time-separated template image in order to identify non-zero flux sources extracted from the differenced images. This method has proven remarkably effective at identifying supernovae and other transient phenomena, and generally constitutes the basis of the primary method of supernova detection in wide-area surveys. The key strength of difference imaging, that of being sensitive to all types of transient phenomena, is ironically detrimental to the task of detecting HR SLSNe. Ordinary supernovae outnumber SLSNe by ∼1000× at low redshift, but low redshift SLSNe are detected with reasonable efficiency because of their conspicuous apparent magnitudes. At high redshift the volumetric rate of SLSNe increases, but the range of apparent magnitudes becomes unremarkable compared to ordinary low redshift supernovae. Thus a survey powerful enough to detect significant numbers of HR SLSNe is ill-equipped to distinguish them from the much larger ordinary supernova population at low redshift using difference imaging alone. Lyman Break Galaxy Selection and Monitoring (LBG S&M; Cooke, 2008) is a transient detection method designed to improve the efficiency of identifying HR events. Template images are constructed from survey data as it is collected, deepening over time. LBG selection is performed on the templates and HR LBG catalogues are compiled. The selected LBGs are monitored in subsequent seasons for HR transient activity, being effectively isolated from the much more numerous low redshift events. The new data is then included into the stacked template images to increase their depth and the number of selected LBGs for monitoring as the survey continues. LBGs are star-forming galaxies with remarkably flat frequency-dependent far-UV spectral energy distributions (SEDs) which exhibit flux breaks at the Lyman-limit (912Å) from internal absorption and in the Ly-α forest (912–1216Å) from intergalactic absorption (Stei- del et al., 1996). At z . 1.6 the Ly-α forest effect is minimal, while at z & 6 the effect eliminates almost all flux (Fan et al., 2006; Bañados et al., 2016). These SED step-downs produce conspicuous colour breaks in broad-band photometry, allowing the use of colour cuts to select LBGs in limited redshift ranges with high efficiency (Steidel et al., 2003, 16 CHAPTER 1. INTRODUCTION

2004; Adelberger et al., 2004; Cooke et al., 2006). LBG selection is useful only for selecting this particular HR galaxy type, but can be performed effectively with as few as three filters. Additional filters increase the efficiency and extend the redshift range of the selections. More inclusive photometric redshift techniques can identify other galaxy types, but require many more filters and a more complex selection methodology. Spectroscopic redshifts are more precise but far less efficient than photometric methods given the more stringent depth limitations and smaller fields-of-view (FOVs) of modern multi-object spectrographs. Furthermore, LBGs constitute a significant percentage of galaxies at high redshift (Marchesini et al., 2007), and the majority of star-forming galaxies unobscured by dust within which SLSNe are expected to be observed. The effectiveness of LBG S&M has been demonstrated and is credited with some of the highest redshift supernova discoveries to date (Cooke et al., 2009, 2012). These discoveries were made archivally in the CFHTLS, preventing spectroscopic follow-up of the events while in progress that may have provided spectroscopic subtypes and valuable far-UV information. However, in principle by applying the method to an active survey, spectroscopic follow-up of identified ongoing HR SLSNe is readily achievable.

1.5 Surveys and Fields

Implementing LBG S&M effectively requires deep, wide-area image templates in order to adequately sample the luminosity functions of LBGs at high redshift and yield a sizeable selection. Depth refers to the limiting magnitude of an image, with deeper images being sensitive to dimmer sources. One way to obtain deep templates is to combine, or stack, repeated observations of an area of sky collected as part of a years-long survey. These stacked images produce the effect of increased exposure time and result in increased depth. Often large surveys include deep fields in their survey area, referring to particular pointings which are subject to longer cumulative exposure times than the wider area. Enlarging the étendu of the survey facility is another means of increasing depth. Étendu is an indicator of the magnitude-limited volume of space probed by an observatory in a given amount of time. It is the product of the primary mirror area and the imager FOV. This thesis works with the following contemporary, high étendu surveys, applying custom image stacking procedures to accessible, large data sets to construct deep stacks and perform LBG S&M with a high likelihood of detecting HR SLSNe. DES is a five year (2013A–2018A) optical (griz) survey of 5000 deg2 of the Southern Sky, restricted to the southern autumn (the A semester). It operated from CTIO with the Blanco 4m telescope and the DECam imager. The DECam is a 3 deg2 FOV camera 1.5. SURVEYS AND FIELDS 17 equipped with a SDSS-like ugriz filter set (Fukugita et al., 1996). We refer to the DECam u-band throughout this work as u0 to distinguish it from the much bluer SDSS u-band, but do not use alternate labels for the other, more similar griz bands. The DECam Y -band is mostly overlapping with the z-band and is not utilized in DES or this project. DES also provides useful DECam pre-commissioning data collected in 2012. DES includes two designated deep fields, C3 and X3, which were imaged with a longer exposure time format and a denser cadence. SUDSS is a 3 year (2013A–2016A) DES boutique survey of 24 deg2 (8 DECam pointings) imaged in griz with a ∼30 day cadence. SUDSS maintained all-year observations, surveying 4 fields overlapping with DES, including both DES deep fields, in the A semester and 4 independent fields in southern spring (the B semester). The A semester SUDSS observations overlapping with DES are accessible by DES for extended coverage of these fields, and corresponding DES observations are available for analysis by SUDSS as external collaborators. We initiated a u0-band extension to SUDSS, dubbed u0SUDSS to facilitate the more effective implementation of LBG S&M. With griz observations alone, HR LBG selection is only effective from z ∼ 4. By including u0-band observations we are able to lower the initial selection redshift to z ∼ 1.5, exploiting the much brighter apparent luminosity distributions of these less distant LBGs (see Figure 1.6) and increasing the selection size by up to ∼30× (Steidel et al., 2004). Observations through u0 are not included in the DES/SUDSS survey strategies because much longer exposure times are required to match the target griz depths per epoch due to increased atmospheric extinction in this filter. However u0 observations need not be cadenced for the effective use of LBG selection. The purpose of u0SUDSS is to collect deep DECam u0 observations of SUDSS fields. We were awarded two dark-time periods through a combination of NOAO1 and AAO2 time allocations for this work, and also made use of data from the NOAO Science Archive. Over the past 4 years u0SUDSS has reached partial to full depth (target limiting magnitude mu0 = 26.5) over a sky area of 12 deg2 (4 DECam pointings), including both DES deep fields. HSC-SSP is a deep optical survey (2014–Present) of 1400 deg2 of the Equatorial Sky using the Subaru 8.2m telescope and HSC, a 1.8 deg2 FOV imager equipped with a SDSS- like griz filter set plus a sensitive y-band filter at for the reddest optical wavelengths (9000-11000Å). The extended red coverage and higher étendu compared to DES from the larger primary mirror makes HSC-SSP sensitive to higher redshift transients. SHIZUCA is a HSC-SSP boutique survey using photometric redshifts to identify and monitor HR

1NOAO Prop. IDs: N4042, N0260 ; PI: C. Curtin 2AAO Prop. IDs: C/2016A/03, C/2016B/05 ; PI: C. Curtin 18 CHAPTER 1. INTRODUCTION

Table 1.1 Survey speciﬁcs of the photometry utilised for this thesis

Surveya Areab Seasonsc Cadenced Epochse Filter Format CETf (Deg2) (Days) (DECam) DESg 6 4 6 120 g 3×200s 72,000s r 3×400s 144,000s i 5×360s 216,000s z 11×330s 435,600s SUDSS 12 4 30 24 g 3×180s 12,960s r 3×250s 18,000s i 4×280s 26,880s z 3×300s 21,600s u0SUDSS 12 2+h - 1-2 u0 60×400s 24,000s+ SHIZUCA 1.8 4+ 20 1i u0 60×1200s 72,000s g 50×300s 15,000s r 21×600s 12,600s i 53×600s 31,800s z 38×500s 19,000s a See Acronymus. b Total sky area overlapping fields with deep u0-band coverage. c An observational season is the ∼6 month window each year during which a field is visible through a reasonably low airmass from a ground-based observatory at a suitable latitude. d The average in-season time interval between observational epochs. e The approximate total number of observational epochs (typically individual nights) included in deep stacks. f Cumulative Exposure Time (approximate). g Survey specifics given are for DES deep fields only. h u0SUDSS is uncadenced and not restricted to continuous seasonal coverage, however the program has been operating since 2015. i Deep stacks of the u0SUDSS area overlapping with SHIZUCA were constructed from archival DECam pre-commissioning observations. galaxies for supernovae. One SHIZUCA field overlaps with u0SUDSS, and a collaboration exists to allow SHIZUCA to make use of u0-enabled LBG S&M in the common field. We also proposed for and were awarded Keck time for SHIZUCA spectroscopic follow-up through the Swinburne Keck time access program (Moriya et al., 2019; Curtin et al., 2019). The u0SUDSS footprint consists of four fields with medium to deep u0-band coverage. The fields are distributed throughout the Southern and Equatorial Skies to allow for all- year spectroscopic follow-up opportunities from both southern and equatorial facilities. Chandra Deep Field South (CDFS; Giacconi et al., 2001) is a pointing in the South- ern/Equatorial Sky often included in surveys for its low galactic extinction. DES includes 3 CDFS pointings labeled C1, C2, C3, with C3 being a DES deep field. C3 is the pri- 1.5. SURVEYS AND FIELDS 19

Table 1.2 Sky ﬁelds covered by the u0SUDSS program

Label Association RAa Decb Skyc Semesterd Epochse CDFS,C3 DES,SUDSS 03:30:36 -28◦0600000 S/E A 2 NSF2 SUDSS 21:28:00 -66◦4800000 S B-A 2 COSMOS SHIZUCA 10:00:28 +02◦1203600 E A-B 1 XMM,X3 DES,SUDSS 02:25:48 -04◦3600000 E A 2 a J2000 Right Ascension of field centre. b J2000 Declination of field centre. c Southern (S), Equatorial (E), Northern (N) or combination. d Academic designation of observable semester or semester transition. e Number of u0SUDSS observing epochs including archival epochs. mary u0SUDSS field for spectroscopic follow-up of HR SLSN photometric candidates. The denser cadence and longer exposure time format on this field by DES provides superior griz image sets for stacking and monitoring for transient activity. Deep u0-band observations by u0SUDSS enable LBG selection to be performed to z ∼ 1.5. C3 is also observable from Keck, allowing more blue-sensitive spectroscopic follow-up than is available from more southern observatories.

XMM is a pointing in the Equatorial Sky named for the X-ray Multi-Mirror Mission (Jansen et al., 2001). DES includes 3 XMM pointing labeled X1, X2, X3, with X3 being a DES deep field. X3 is visible from Keck during the same period as C3, but the field lacks deep u0-band observations at present making LBG S&M less effective at z ∼ 2 and z ∼ 3. The field is used as a secondary source for targets during spectroscopic follow-up campaigns on C3.

New South Field 2 (NSF2) is one of two SUDSS pointings in the Southern Sky not previously surveyed. The ﬁeld includes deep u0-band observations by u0SUDSS, but SUDSS griz image sets are less extensive than on the DES deep ﬁelds, leading to shallower stacks in these bands and ultimately fewer LBGs selected at z ∼ 3 and z ∼ 4. NSF2 is too far south to be observed from Keck, preventing deep blue spectroscopic follow-up. However the generated LBG catalogues remain valid for future surveys such as LSST.

COSMOS is a pointing in the Equatorial Sky named for the Cosmological Evolution Survey (Scoville et al., 2007). SUDSS includes two COSMOS pointings but the cadence on these fields is very poor. Substantial DECam pre-commissioning u0griz observations were collected on one of these fields which overlaps with the SHIZUCA COSMOS field. Deep stacks constructed from this archival data are used to perform LBG selection on the field to compliment SHIZUCA photometric redshift catalogues. 20 CHAPTER 1. INTRODUCTION

1.6 Purpose of the Thesis

SLSNe are a recent discovery which are 10–100× brighter than ordinary supernova types and represent an opportunity to push the practical observational redshift limits of optical transient astronomy. With the possibility of observing stellar explosions at cosmological redshifts beyond even the most distant observable galaxies, the potential applications are profound. Over the past few decades the field of optical transient astronomy has seen major technological advancements leading to improvements in sky monitoring and candidate detection. However, the benefits pursued have mainly been in regard to the regional z < 2 Universe. SLSNe are very rare in any volume, but efficiently picking out these events at high redshift from much more numerous low redshift transients requires a different approach to candidate detection. LBG S&M is specifically designed to be highly sensitive to HR SLSNe and UV-bright SNe while remaining insensitive to most low redshift transients. We have applied this technique to ongoing deep-and-wide area surveys for the first time in order to efficiently detect and collect spectroscopic follow-up of HR SLSNe near peak. In doing so we demonstrate and develop this alternative method of transient detection, shifting the focus to more distant horizons. In Chapter 2 we describe the methodology developed and applied to DES, SUDSS and u0SUDSS image stacking and discuss the depths achieved. In Chapter 3 we establish colour cut criteria for LBG selection with DECam. For this we use model spectral templates, composite spectra, LBG selection catalogue source spectra collected with Keck LRIS, and other available archive spectra. We discuss the efficiencies and completenesses of the selections. In Chapter 4 we perform photometric selection of DES HR SLSN candidates and spectroscopic follow-up campaigns with Keck LRIS. The spectra of candidates are presented, and strategies for increasing the confidence of HR SLSN candidates from their photometry are discussed. In Chapter 5 we present spectra of confirmed HR SLSNe from SHIZUCA, including the highest redshift supernova ever captured spectroscopically near peak. We compare the host-subtracted supernova spectra to far-UV spectra of low redshift SLSNe. We present our conclusions in Chapter 6. Throughout this thesis we −1 −1 assume a ΛCDM cosmology with H0 = 70km s Mpc , ΩM = 0.3 and ΩΛ = 0.7, and photometric magnitudes are calculated in the AB system unless otherwise specified. 2 Deep Stacking

As we are using Lyman Break Galaxy Selection & Monitoring (LBG S&M) to search for high redshift (HR; z & 2) superluminous supernovae (SLSNe), we will be performing LBG Selection (LBGS) on the u0SUDSS search fields (see Table 1.2, Chapter 3). LBGS is a method by which HR sources in a field can be efficiently identified from broadband multi- filter optical photometry alone. Efficiency refers to a high level of selection completeness and a low percentage of low redshift interlopers (LRIs; z . 1). In order to significantly sample the luminosity function of HR LBGs and generate host galaxy catalogues large enough to effectively monitor for transient activity (hundreds of thousands of sources), we require deep, multi-filter template images with large cumulative exposure times. We construct these deep stacks using large DECam image sets (see Table 1.1). Three of four u0SUDSS fields (C3, NSF2, COSMOS; 9 deg2) reach the LBGS target limiting magnitude 0 0 of u5σ ∼26.2 with matching or deeper griz stacks. DECam u observations of the fourth 2 0 field (X3; 3 deg ) currently reach depth u5σ∼25.5 with deeper griz stacks. The deepest 2 griz stacks (C3, X3; 6 deg ) achieve limiting magnitudes of griz5σ = 27.3, 27.4, 27.2, 26.9. We have shared our procedures and data products with DES for further application.

2.1 Introduction

One of the ﬁrst and best demonstrations of the scientiﬁc potential of truly deep stacking is the Hubble Deep Field (HDF; Williams et al., 1996). A custom HST image reduction pipeline including a new ‘drizzle’ stacking routine (Fruchter & Hook, 2002) was written to construct the deep stacks, mainly accounting for cosmic rays and hot pixels. The stable pointing and lack of variability in observing conditions simplify the stacking process. HDF covers an area of ∼6 arcmin2, uncadenced, in 4 bands spanning the observer-frame near-

UV to far-optical to depths of m5σ = 27.8 − 29. The stacks revealed an unexpected

21 22 CHAPTER 2. DEEP STACKING density of HR galaxies out to z ∼ 6 and expanded the range of cosmological analysis. The success of HDF quickly prompted a second iteration, HDF-South (Casertano et al., 2000). Similarities between the two opposing ﬁelds provided additional evidence for the cosmological principle, and the observations have been applied to a variety of additional cosmological studies.

The success of HDF and HDFS inspired the conception of more ambitious deep stacking programs. GOODS produced a shallower but ∼30× wider pair of north and south deep fields which spanned from X-ray to IR wavelengths using multiple instruments (Giavalisco et al., 2004). The Hubble Ultra-Deep Field (Beckwith et al., 2006) pushed ∼1.5 mag deeper than HDF and extended wavelength coverage into the near-IR. The extended coverage enables the detection of sources which have been redshifted out of the optical entirely. A fraction of this field was deepened into the eXtreme Deep Field with limiting magnitudes of m5σ = 29.1 − 30.3 in 9 filters which, when combined provides a limiting magnitude of ∼31.2, the deepest image ever constructed (Illingworth et al., 2013). These stacks have been invaluable to the advancement of cosmology, but being over a narrow area and without a cadence strategy they do not contribute significantly to the study of HR transient phenomena.

Similar deep fields projects have been undertaken from large 8–10m ground-based telescopes. While atmospheric turbulence prevents ground-based observations from reaching resolution comparable to that achievable from space-based observatories, deep observations from the ground have the advantage of ease of access to more current and more massive instrumentation. This typically permits larger instrumental FOVs and allows greater areal coverage of deep fields. The Subaru Deep Field (Kashikawa et al., 2004) reaches depths of 2 m5σ = 26.3 − 28.3 in 5 optical broadband filters in a sky area of 918 arcmin , over 150× the area of HDF. The Keck Deep Fields (Sawicki & Thompson, 2005) reach Rlim∼27 and similar depths in 3 other broadband filters on 3 widely separated fields constituting a total sky area of 169 arcmin2. While ground-based deep observations increase the area of deep fields, performing such observations in a reasonably short period requires large telescopes with limited availability.

Smaller telescopes can reach similarly deep observations but only by combining many epochs of observation. Such image sets are available as a by-product of wide-area sky surveys. The CFHTLS deep fields (Astier et al., 2006) are made up of densely cadenced survey data collected in the years 2003–2009 with the CFHT 4m telescope. The depths of the ∗ 2 2 stacks are m5σ∼27.5 (26.5) in MegaCam u gri(z) over an area of 4 deg (14,400 arcmin ). Constructing deep stacks in this way requires additional procedures to compensate for 2.2. MAXIMIZING DEPTH 23 greater changes in atmospheric conditions, detector sensitivity and survey operations. Many software tools have been developed for image stacking and photometry to accom- modate large surveys like the SNLS, but deep stacking procedures are complex and vary depending on image set specifics and the desired data products. Significant programming and development are typically required. Here we develop stacking procedures to produce deep stacks using DES, SUDSS and u0SUDSS photometry. We discuss the specifications of each image set and the software used to process the images. We apply astrometric corrections before stacking and photometric calibration of the stacked images. We test the depths of the resultant stacks to ensure their integrity.

2.2 Maximizing Depth

The depth of an image or its limiting magnitude is the faintest magnitude at which a source can be reliably measured. We adopt in this work the usual 5σ limiting magnitude, or the magnitude at which a source has a signal-to-noise ratio (S/N) of 5.

The 1σ noise of an image, σi, is the standard deviation of the background sky ﬂux measured in counts per pixel. In ground-based observations, the point spread function (PSF) of a point-source is degraded by the atmosphere and the detected ﬂux is spread out over an area of pixels. In such images without sub-pixel scale seeing, the noise over an area of n pixels is approximately

σ · n √ σ = √i = σ · n (2.1) n n i

It is useful to define n in terms of the full width at half-maximum (FWHM) seeing. The FWHM of an image is the full width of the PSF of a point-source at half of the maximum pixel count value. The flux of a point-source is measured within a circular aperture with radius p×FWHM, where p is a multiplicative factor selected so that the aperture is inclusive of a specific percentage of the total flux from the source independent of seeing. Using this radius we have

√ p 2 σn = σi × π · (p · FWHM) = σi × ( π · p · FWHM) (2.2)

Images can be stacked to reduce noise and increase depth. A set of N images with the same exposure time and taken under the same conditions combined together, produces a stacked image with N× the variance in the background and N× the signal from sources. The noise of the stacked image is the standard deviation of the background, or the square 24 CHAPTER 2. DEEP STACKING root of the variance. Thus the S/N of sources in the stacked image increases by a factor √ of N.

√ S/Nstack = N · S/Nimage (2.3)

The 5σ limiting magnitude is measured relative to a zeropoint. The true zeropoint of an image, m0, is the apparent magnitude of a source with a flux of 1 count in the specified aperture (some definitions call for a source with a flux of 1 count/s which we refer to when necessary as the unit zeropoint or mˆ0). In this work we define zeropoints using the AB-magnitude system.

Finally we deﬁne the depth of an image, m5σ, as

√ m5σ = −2.5 · log10[5 · (σi × ( π · p · FWHM))] + m0 (2.4)

For p we assign a value of 1.02, consistent with referenced SDSS catalogues and inclusive of more than 98% of the total ﬂux from an ideal point-source. Thus from Equation 2.4, maximizing depth amounts to to minimizing both seeing and noise.

2.3 Image Sets

The complexities of image stacking are greatly alleviated by using a uniform image set in which exposure times and seeing are similar throughout, and in which the basic individual image reduction processes are nearly identical. Non-uniform image sets can be stacked with the prior application of a normalization procedure. However the normalization of a non-uniform image set is not trivial, and if mishandled can create inconsistencies in the data that are difficult to detect from the final product. We construct our deep stacks from the largest uniform image sets available on u0SUDSS fields, but exclude images in different formats to avoid complications. The numbers of excluded images are small, and their exclusion does not have a significant effect on the final depths. We source our images from two image archives, the NOAO Science Archive (NOAO- SA) and the DES Database (DES-DB). Access to DES-DB is granted by our external collaborator membership through SUDSS. Images collected by SUDSS or u0SUDSS are available to survey members through the NOAO-SA. Archival images are also available freely through the NOAO-SA after a two year proprietary period. NOAO-SA images are delivered after bias subtraction and flat-fielding, and include associated data quality masks (DQMs) and weightmaps. The weightmap products from the NOAO-SA are derived from the reciprocals of the images themselves. Such weightmaps 2.3. IMAGE SETS 25 are sufficient for precision photometry, but are detrimental to stacking large image sets (see Section 2.6). DES-DB images are accessed through a secure SQL database. These images are in a variety of reduction stages as the survey continues to develop its data products to suit the needs of its members. There is a good deal of unpublished documentation on the state of the various image reductions, but it does not lend itself to a clear understanding of all the processing involved. The most appropriate data products must be identified by the user and may require additional reduction. We take images at an early stage of reduction to ensure our dataset has the uniformity needed for deep stacking. The images are bias subtracted and flat-fielded, and include DQMs and weightmaps. The weightmaps available from DES-DB are derived from a series of flats and are more appropriate for stacking than the image reciprocals produced by NOAO-SA. One format we exclude from our image sets when possible is images largely dithered to cover chip gaps. Because the centre of each field remains nearly constant throughout each survey, the overlap between the CCD mosaics of individual images is almost complete except for a few pixels around the edge of each CCD. Including dithered images in a stack of undithered images produces a stack with a variable depth between the main CCD areas and the chip gaps, which increases the photometric uncertainty of sources measured in these areas. LBGs and transients detected near CCD edges are considered unreliable and not included in our analysis, and including dithered images in the stack only complicates the process of identifying these sources while not contributing significantly to depth. An exception to this policy exists in the u0-band deep stacks (note the use of u0 to distinguish the DECam u-band from the bluer SDSS u-band). Image dithering is included in u0SUDSS strategy to maximise the utility of the of the u0-band image sets to the community, so the largest u0-band image sets available on non-archival u0SUDSS fields are dithered image sets. Because the entire set is dithered in a regular pattern, variations in depth are minimal. However, there is an added complication to measuring the colors of sources using a mix of dithered and undithered deep stacks. To overcome this we construct our u0-band deep stacks using the entire 62-CCD DECam mosaic as an image template. When stacking griz images, the consistent DECam pointing allows the mosaic to be conveniently disassembled into individual CCDs which are each stacked separately, reducing demand on computer memory. To stack the mosaic as a single unit requires a considerable amount of computer memory, and we assign this job to the Swinburne supercomputer, G2. Once the set of dithered mosaics is combined, we disassemble the deep stack using the CCD boundaries of the griz deep stacks. These u0- 26 CHAPTER 2. DEEP STACKING band deep stack CCD cut-outs can then be compared directly to the same regions in the other deep stacks directly.

2.4 Software

Some of the fundamental steps within the stacking pipeline are accomplished with repeated use of software from the AstrOmatic suite of wide-field image reduction and analysis applications. Software for Calibrating AstroMetry and Photometry (SCAMP) assigns and adjusts the world coordinate system (WCS) astrometry of images into agreement with a given reference catalogue. SWarp (SWarp, Bertin, 2006) is an image combination tool. It also performs image reprojection and background-derived weightmap generation. Source Extractor (SExtractor) detects sources of flux and generates and cross- matched catalogues of source positions, flux measurements and morphological information.

2.5 Astrometrics

To maximise depth, precise relative astrometry among images is required. The astrometric uncertainty must be significantly less than the FWHM of the stack. This ensures that the FWHM of a deep stack is a function of the FWHMs of the individual images and is not significantly affected by differences in the WCS solutions within the image set. Calibrating the astrometry over an image set consists of reprojecting the images to correct the pixel scale using SWarp, extracting robust point-sources from the images with SExtractor and feeding the point-source catalogues into SCAMP to assign a uniform WCS. SWarp includes a reprojection routine to make adjustments to the pixel scale of very wide-field images. In wide-field images a pincushion effect arises as a consequence of the optics when projecting the sky onto a flat focal plane. Reprojection is the process of correcting these distortions with the assignment of a position-dependent pixel scale function. The DECam optics largely correct for this effect, but as SCAMP is prone to failures in large image sets even a small adjustment to the reprojected images can increase the success rate. SCAMP requires input catalogues with an initial WCS in order to perform astrometric correction. We generate these catalogues with SExtractor using the WCS delivered with the image. SCAMP can process large image sets in unison to improve the internal consistency. From a set of image catalogues SCAMP will identify a set of sources common 2.6. IMAGE SET CALIBRATION 27 among the catalogues with which to calibrate the WCS solutions. This ensures that the WCS of each image is solved from a similar set of reference sources. SCAMP is prone to encountering a percentage of convergence failures when working with image sets. To minimise this failure rate we use a conservative configuration of SExtractor which extracts only high confidence, unsaturated point-sources. We input the reprojected image point-source catalogues in SCAMP for WCS correction. We use the optical USNO-B1 catalogue (Monet et al., 2003) as our astrometric reference, which yields internally precise astrometric solutions on all the u0SUDSS fields and filters. SCAMP outputs headers for the input images with the necessary WCS corrections which are automatically applied by SWarp and SExtractor in future iterations. Unfortunately SCAMP is not well equipped to handle convergence failures and has no verbosity to identify failures easily. In addition large image sets with a high degree of variability in depth and seeing make it unlikely that a single configuration of SCAMP can obtain universal convergence. A convergence failure that results in a large astrometric offset is easily detected in a deep stack as a separated, shallow projection of the aligned mosaic. Small offsets on the order of the FWHM of the images are more problematic as shifts of this degree in the central pointing of the images within a set are common. Slightly offset images are effectively averaged out of a weighted mean deep stack. Still they contribute source flux to sky and sky flux to sources, simultaneously increasing background noise and decreasing signal. Convergence failures must be identified and the associated images recalibrated to maximise depth. To do this we include a short Python script in the stacking pipeline which generates a secondary SExtractor source catalogue for each image after SCAMP astrometric correction, and then measures the internal astrometric accuracy of each image set. We configure SExtractor to crossmatch each catalogue with an arbitrary master catalogue and measure the average offset and 1σ uncertainty in both RA and Dec. Prop- erly aligned images produce average offsets consistent with zero within a small uncertainty (see Table 2.1). Misaligned images in RA, Dec or both are identified by their non-zero average offsets while rotation angle errors are identified by their large uncertainties. These images are then recalibrated with SCAMP using an adjusted SExtractor point-source catalogue and retested until a uniform WCS is achieved.

2.6 Image Set Calibration

Deep stacks are combinations of large image sets. To maximise depth the images within each set are assigned conﬁdence weights, ﬁltered for data quality and scaled by exposure 28 CHAPTER 2. DEEP STACKING

Table 2.1 Astrometric precision of DECam C3 sources

a b a b b c Filter ∆α σα ∆δ σδ σαδ FWHM g 0.011 0.441 0.013 0.448 0.627 4.45 r 0.011 0.410 0.042 0.456 0.616 4.18 i 0.015 0.331 0.004 0.388 0.509 4.03 z 0.011 0.456 0.027 0.460 0.650 3.88

All values are measured in pixels where the central pixel scale of the DECam is 0.2600 pixel−1. a Average RA (α) and Dec (δ) differences in source locations between images. b Root-mean-square uncertainties in RA, Dec and both added in quadrature. σαδ is the most appropriate measure of astrometric precision of individual sources. c The average FWHMs are determined using the SExtractor CLASS value to identify point- sources and the SExtractor SEEING value to measure the FWHM of each point-source. time. A weightmap, also called an inverse variance map, quantifies the individual pixel response uncertainties for a CCD. The map gives the inverse variance of counts measured by a pixel exposed repeatedly to a source of constant flux, such as a flatlamp. Sky background can also be used as a source of approximately constant flux, and weight maps are often background-derived. This offers additional information about background noise levels which vary with seeing and sky brightness. For image sets using a common imager, background variations are much more severe than differences in individual pixel response uncertainties. Weightmaps of the background-derived type are more useful for image stacking as the assigned weights reflect the quality of the observing conditions. Before stacking we require a uniform method for deriving weightmaps. DES-DB images include flat-field-derived weightmaps which are scaled by background noise. NOAO-SA images include weightmaps derived from the image reciprocal. This latter method automatically accounts for background variations and works well for calculating the uncertainty of flux measurements, but for stacking purposes low resolution background maps are more reliable. By maintaining native resolution in the image reciprocals, pixels with high counts corresponding to actual sources are assigned very low weights. In poor seeing images with high FWHMs the weights assigned to pixels at source flux centroids are higher than average because the wider PSFs lead to lower count peaks. Thus in stacks of large image sets taken in a variety of seeing conditions, images with the highest FWHMs have the highest weighted pixels at source flux centroids. This reduces the overall signal and S/N of sources in the stack, and lowers depth (see Figure 2.1). Because some of our stacks include a mix of images from DES-DB and NOAO-SA, we produce our own weightmaps for all images using a common method. The dominant source 2.6. IMAGE SET CALIBRATION 29

Figure 2.1 Loss of conservation of source flux in image stacks using high resolution, background- derived weightmaps. Two stacks of images taken over a 4 year period are constructed. In one stack the NOAO-SA weightmaps are assigned. These are constructed using high resolution image reciprocals. The other stack uses our weightmaps derived from low resolution background maps. The relative magnitudes of sources as measured in the latter stack, shown along the x-axis, are linearly consistent with SDSS catalogues. The difference between this magnitude measurement and the magnitude measured in the former stack for each source is shown on the y-axis. The use of the NOAO-SA weightmaps produces a non-linear, magnitude-dependent deviation in the photometry which increases with the size of the image set. of noise in each image is sky and a low resolution background map produced by SWarp is a valid approximation of a weightmap. This is sensitive to seeing and sky brightness, as well as regions of higher than average noise from nearby bright sources. By reducing the background map resolution, pixels associated with source flux are assigned weights based on the surrounding background noise levels. This method assumes small sensitivity variations at the pixel scale compared to larger scales, and this is true except for faulty pixels which are corrected separately (see below). Faulty pixels and other effects such as cosmic rays and saturating sources are accounted for with the use of data quality masks (DQMs). DQMs are used to assign pixels with data quality flags. Functioning pixels are flagged as 1, while different codes are used for pixels with different types of malfunctions. Similar versions of DQMs are provided by both NOAO-SA and DES-DB. Because our image sets are so large, we simply set all malfunction-flagged pixels to zero weight and leave functioning pixels unchanged. Finally we assign each image a flux scale factor. The flux scale factor is a measure of the effective exposure time (EET) of an image. The EET of an image is the exposure time required to reach the depth of the image with the same instrument and filter under 30 CHAPTER 2. DEEP STACKING photometric conditions and low airmass. This is not the same as the actual exposure time for images that are taken on non-photometric nights which include atmospheric extinction in excess of the measured airmass. Signal and variance increase linearly with exposure time, and noise and S/N increase as the square root of exposure time (Equation 2.3). As exposure time increases, S/N and depth increase but weight as inverse variance decreases. However, deeper images should receive greater weight. To adjust for this we simply scale the weight of each image in each filter by the EET. To determine the EET of an image we measure the difference in the relative zeropoint of the image compared to the entire set of images through the relevant filter. A relative zeropoint is one which is arbitrary but, if consistently assigned to a set of images, accurately reflects relative differences in the true zeropoints. With SExtractor we extract a common set of sources from each image using a constant relative zeropoint and the built-in association function. We then measure the average difference of the magnitudes of common sources in each image compared to the set. In Figure 2.2 we plot the average zeropoint offset per image for a particular image set. Most of the catalogues provide very similar magnitudes for common sources, as is apparent from the clustering of the data points at the 0 mag additional extinction level. These correspond to images taken under photometric conditions. Catalogues from images with additional atmospheric extinction exhibit significant average differences in common source magnitudes relative to the photometric image catalogues. The differences are used to calculate the EETs of these non-photometric images using

[−E/2.5] EET = texp · 10 (2.5) where E is the additional extinction given by the average diﬀerence in magnitudes of common sources, and texp is the actual exposure time. Signal and variance scale linearly with EET just as with actual exposure time. Thus without the application of ﬂux scales, background-derived weightmaps produced by SWarp assign much greater weight to images with low EETs and low variance when in fact the low signal suggests reducing the weight by the same factor. This is extremely detrimental to the depth of the resultant stack (see Table 2.2).

2.7 Image Stacking and Source Extraction

We use SWarp to combine the reprojected images into deep stacks. SCAMP output headers are appended to each image with calibrated astrometry. Each deep stack is a 2.7. IMAGE STACKING AND SOURCE EXTRACTION 31

Figure 2.2 Degeneracy between FWHM and non-photometric extinction in terms of image weights. Sources are extracted from z-band exposures taken during the third year of DES on C3. The panels share a common x-axis identifying the relative date the exposures were taken. Epochs are collected every ∼7 days and each epoch consists of 11×330s exposures as per the DES survey strategy. The top panel shows the average FWHM of each exposure. The average weight of a background- derived weightmap is sensitive to changes in seeing, resulting in lower weights for images with wider FWHMs. The bottom panel shows the average magnitude offsets of the image source catalogues using a constant zeropoint and set to a minimum of zero. On photometric nights there is no significant difference between the measured magnitudes. However on non-photometric nights the difference to photometric nights can be quite large. Background-derived weightmaps are insensitive to zeropoint changes due to extinction, and the plot shows there is no correlation between images with greater than typical atmospheric extinction and seeing. For accurate weighting these images must be identified independently and scaled to the correct EET. A procedure for this is clear from the plot, using the measured magnitude offset to calculate the flux scale using Equation 2.5. Without this scaling the images with the most extinction are assigned the highest weight and significantly reduce the depth of the stack (see Table 2.2). 32 CHAPTER 2. DEEP STACKING

Table 2.2 Source recovery in photometric imagery compared to extincted images and the eﬀects of including extincted images in stacks without ﬂux scales

Image Type CETa Noiseb Detectionsc (seconds) (counts/pixel) Extincted 330 9.2 66 Extincted Stack 3630 2.56 434 Photometric 330 35.1 942 Photometric Stack 3630 9.61 2722 Mixed Stackd 7260 2.51 1338 a Cumulative Exposure Time, or actual time spent on exposures. b In the absence of ﬂux factors the weight assigned by SWarp is proportional to the squared inverse of the measured noise. c The number of detections is typically 2–3× greater for every additional magnitude of depth. In the absence of extinction, a 1 mag increase in depth corresponds to a factor of 2.5 in exposure time. d By mixing extincted and photometric images without ﬂux scales, the weights of the extincted images are exaggerated to the detriment of the depth of the resultant stack. local-background-subtracted weighted average of the input image set using our background- derived weightmaps convolved with a corresponding DQM and scaled by the EET. We set the gain and saturation levels to the header values delivered with the images, though in practice saturated pixels are masked by the DQM, and the estimated gains do not vary over time.

We perform source extraction on the deep stacks with SExtractor. The default configuration is left largely unchanged, but some customization is required. We convolve the images with a 7×7 element Gaussian kernel suitable for FWHMs of ∼4 pixels to improve detection efficiency. Many HR LBGs are near-point-sources at these FWHMs, and they are often observed in interacting pairs. To better identify these cases, the deblending parameters are set to the maximum 64 deblending sub-thresholds and the minimum 0.0001 contrast. Cleaning is turned off, meaning detected sources that drop below the detection threshold after the subtraction of neighbouring source PSFs are not removed from the output catalogues. Flux measurements of these sources are very uncertain, but in the event of a transient the increased S/N can be used to re-evaluate the necessary colour information. We leave the deblending calibrator in the default mode, which separates the flux measured from blended sources into components based on internally measured source profiles. Flux error estimates are calculated with the aid of the output weightmap of the SWarp stacking procedure. This is an internally calibrated combination of the input weightmaps, and provides SExtractor with weights for the pixels used to measure 2.8. PHOTOMETRIC CALIBRATION 33 background noise. This ensures that the flux errors of sources are derived from nearby background noise measurements weighted toward more confident pixel values. For sources near a CCD chip gap this is particularly relevant, and errors are based primarily on noise measurements made away from the gap. SExtractor takes an input detection threshold which sets the minimum required flux of every pixel over a minimum detection area to constitute a source. We use the default 5 pixel minimum detection area to avoid extraction of statistical noise. This corresponds to a circular aperture with radius 1.3 pixels or 0.3400 (note this aperture is used for detection, not flux measurement). 5σ detections are the standard when defining depth, however sources below this threshold are still confident if detected in multiple bands. We estimate the minimum reasonable detection threshold for each deep stack empirically by extracting sources over a range of thresholds in both the stack and the negative inverse and comparing the recovery ratios. We find detection thresholds of ∼0.8σ introduces only 5–10% false positives (see Figure 2.3), which falls to ∼0% after multi-band crossmatching.

2.8 Photometric Calibration

We use two SExtractor internal methods for making flux measurements of extracted sources. These are recorded in output catalogues under the labels of MAG_AUTO and MAG_ISO. Both methods are similarly accurate for dim sources and provide independent colour information for each source. The MAG_AUTO measurement is generally recommended in the SExtractor user manual 1. From the flux distribution of each source the Kron radius is calculated and an elliptical aperture is generated (Kron, 1980). The source flux measured within the aperture is then extrapolated to an estimate of the total. The apertures are expanded to twice the Kron radius as recommended for measuring dim extended sources, and the minimum aperture radius is set to the average FWHM of 4 pixels similar to the aperture size used to measure depth. The MAG_ISO measurement is more suited to asymmetrical, clumpy sources which are difficult to approximate with an ellipse. The flux of a source is measured from only those pixels which exceed the detection threshold, tracing the morphology of sources precisely. Without extrapolation this method neglects a fraction of flux from each source, but because we are using a low detection threshold the difference between the MAG_AUTO and MAG_ISO measurements for most sources is actually quite small (see Figure 2.4).

1The latest version of the SExtractor user manual can be found at https://sextractor.readthedocs.io/en/latest/index.html 34 CHAPTER 2. DEEP STACKING

Figure 2.3 To optimise the detection threshold we compare the number of sources recovered in an image to the number recovered in the negative inverse of the image. The orange line represents image sources minus inverse sources as a percentage of the maximum of this difference as a function of detection threshold. In this case the difference is maximised with a detection threshold of 0.75σ. The blue line represents inverse recoveries as a percentage of image recoveries, an estimate of the rate of false positives as a function of detection threshold. Detection thresholds of 0.72σ and 0.80σ correspond to false positive rates of 10% and 5% respectively (horizontal dashed red lines). The green line represents the product of the orange line and the true positive rate (one minus the blue line), in effect quantifying the cost of real source recoveries in terms of associated false positives. In this case the cost is minimised at a detection threshold of 0.83σ. A detection threshold of ∼0.8σ is generally appropriate for recovering the majority of real sources at a low rate of false positives. 2.8. PHOTOMETRIC CALIBRATION 35

Figure 2.4 DECam i − r MAG_AUTO colour compared to MAG_ISO colour. MAG_AUTO is designed to measure the same percentage of flux from every source and extrapolate a total. MAG_ISO in principle measures most of the flux from each source, but this percentage varies with the brightness of the source and there is no extrapolation. The consistency between the measures is more apparent in terms of colour than individual magnitudes. The colour difference between the measures is small and approximately normally distributed with a 1σ uncertainty everywhere less than the 0.2 mag level indicated by the dashed blue lines. This uncertainty is a benchmark for high confidence detections (see Section 2.10). 36 CHAPTER 2. DEEP STACKING

2.8.1 Zeropoints

Accurate zeropoints are assigned to the images and stacks to ensure accurate photometry and colour information used in LBGS (zeropoints are defined in Section 2.2). Normally this is done using standard stars in the observed field whose apparent AB magnitudes are known. For SDSS-like griz images, SDSS catalogue stars can be used as standards. However some u0SUDSS fields are outside the SDSS footprint, and the DECam u0-band is not as similar to the SDSS u-band as the griz filters are. It is not DES procedure to image standard fields for photometric calibration during every observing epoch. Instead the zeropoints are calibrated periodically and assumed to be stable in the interim. Adjust- ments to zeropoints on non-photometric nights are made by comparing the photometry to photometric nights (similar to our procedure for measuring extinction illustrated in Figure 2.2). The stable photometric unit zeropoints for each CCD of the 62 CCD DECam mosaic in each filter are publicly available 2. The equation provided for calculating unit zeropoints, mˆ0, for DECam images is given as

mˆ0 = ZP − AMF · (airmass) − CCT · (colourDECam − colourSDSS) (2.6) where ZP is the unit zeropoint without atmosphere, AMF is a filter-dependent airmass extinction coefficient for the site, CCT is a colour correction term for reproducing SDSS catalogue values and ‘colour’ refers to the reference colour to use with the CCT term for each filter (see Table 2.3). The zeropoint for a given exposure time is simply

m0 =m ˆ0 + 2.5 · log10(exposure time) (2.7)

We ignore the CCT in this work because the adjustment is small compared to the ∼0.2 mag photometric uncertainty tolerance applied during LBGS. Because the CCTs are small as are the listed colour differences given the similarity between the DECam and SDSS filter sets, SDSS catalogues when available are useful for confirming the publicly available zeropoints. We test the zeropoints on the u0SUDSS COSMOS field (see Table 1.2) which lies within the SDSS footprint. Without the use of the CCTs the agreement between the catalogues is noisy but confident in all of the griz filters. However there is a significant discrepancy between the DECam u0-band and the SDSS u-band catalogues which must be resolved.

2The DECam photometric calibration page can be found at http://www.ctio.noao.edu/noao/content/decam-photometric-calibration 2.8. PHOTOMETRIC CALIBRATION 37

Table 2.3 Median values over the 62 CCD DECam mosaic for the mˆ0 terms of Equation 2.6

a a b Filter ZP σZP AMF σAMF CCT σCCT Colour u0 23.62 0.16 0.44 0.03 0.06 0.04 u0 − g g 25.42 0.08 0.20 0.02 -0.11 0.01 g − r r 25.48 0.06 0.10 0.02 -0.10 0.01 g − r i 25.35 0.05 0.06 0.01 -0.30 0.02 i − z z 25.07 0.04 0.07 0.02 -0.09 0.02 i − z a The standard deviation of the 62 separate CCD values. b The reference colour for use with the CCT term.

2.8.2 u0-band Rectiﬁcation

To rectify the discrepancy with our u0-band catalogues, first we perform a comparison of the relevant filter throughput response functions (see Figure 2.5). The u0-band is significantly redder than the SDSS u-band, though the CCT provided to relate these filters is very small (see Table 2.3). The MegaCam3 u*-band is also significantly redder than u, and samples redder wavelengths than u0 although the effective wavelengths of the two filters are similar. The transformation from u* to u calls for a significant CCT 0.2324. If the public u0 unit zeropoint is set to minimise the average difference to SDSS catalogues, this could explain the lower than expected CCT and suggests a lower ZP for u0. Vega- like sources are on average 0.6 mag brighter in u* than in u, and from Figure 2.5 this difference is approximately split by u0, suggesting that a ZP decrease of 0.2–0.4 mag may be appropriate. Next we attempt to reproduce the provided ZPs using the full set of DECam filter throughput response functions (see Figure 3.1). To do this we pass an artificial source of constant flux per unit frequency through the filter set and assign the i-band magnitude as the given i-band ZP. In principle the measured magnitude of the source in each of the other filters corresponds to the ZP of the filter relative to the assigned i-band ZP. Each of the griz ZPs is confirmed this way as expected from the agreement of the griz catalogues to SDSS catalogues. However, the implied u0-band ZP is 23.33, almost 0.3 mag brighter than the median value provided for the CCD mosaic. Finally we seek a reference star catalogue to measure the u0 ZP directly. DES provides links to overlapping SDSS catalogues for manual photometric calibration, however SDSS u-

3Speciﬁcations of the CFHT MegaCam can be found at https://www.cfht.hawaii.edu/Instruments/Imaging/Megacam/ 4Photometric transformations between MegaCam and SDSS are discussed at http://cfht.hawaii.edu/Instruments/Imaging/MegaPrime/generalinformation.html 38 CHAPTER 2. DEEP STACKING

Figure 2.5 The filter throughput response functions of the DECam u0-band, the SDSS u-band and the MegaCam u*-band. The dashed black lines indicate the effective wavelengths of each filter. This is the wavelength at which the integrated area under the curve of the associated filter is divided into equal halves. The u*-band extends to much redder wavelengths than the u-band, and conversion between the two includes a CCT of 0.232. The red edge of the u0-band falls midway between the red edges of the u-band and the u*-band, suggesting conversion between u0 and u requires a CCT midway between 0 and 0.232, in contradiction to the provided value (see Table 2.3). 2.9. FILTER CROSSMATCHING 39 band catalogues are not suitable for this measurement because of the difference between u and u0. The Southern u0g0r0i0z0 Standards catalogue (Smith et al., 2003) includes DECam u0 standards in the C3 field. Only 8 standards are recovered in our catalogues, but a weighted least squares fit yields a u0 ZP estimate of 23.28. From this and the above estimates we adopt for this work a u0 ZP of 23.3 mag in place of the 23.62 mag ZP provided publicly. Our adopted ZP yields colour information for catalogue sources consistent with catalogue colours in the literature. Unfortunately the lower ZP implies that the u0 deep stacks are shallower than expected and the resultant LBGS catalogue sizes are reduced.

2.8.3 Photometric Uncertainties

SExtractor outputs a 1σ flux uncertainty based on the background noise in a neighbour- subtracted annulus around each source. Such values are required for determining the significance of observed brightening events and the accuracy of colour information. SEx- tractor flux uncertainties are accurate as a measure of observed signal to observed noise, but do not account for systematics introduced during image stacking procedures. We derive a second error estimation using an independent method to identify any systematics. We calibrate some of the single epochs which go into each stack and produce multiple catalogues of common sources. For non-variable sources the measured flux in a series of catalogues varies according to the single epoch uncertainty. The stack uncertainty is the measured single epoch uncertainty divided by the root of the number of epochs less 1 (the sample standard deviation). The common source stack uncertainties are then compared to the corresponding SExtractor uncertainties. If there are substantial differences the stacking procedure is examined for the introduction of systematics. If no systematics are identified the greater of the two uncertainties is used (see Figure 2.6).

2.9 Filter Crossmatching

The required colour information for LBGS is defined by the magnitude differences of sources between filters. We crossmatch sources with two SExtractor routines for redundancy. These are referred to as ASSOC_MODE and DUAL_MODE. The preferred method is ASSOC_MODE. In ASSOC_MODE extracted sources are associated with sources in a reference catalogue according to proximity. By ignoring detections not in the proximity of a reference source, ASSOC_MODE allows for arbitrarily low detection thresholds with an almost non-existent rate of location-coincident spurious detections. This is useful for images with similar depth, but otherwise allows the shallowest images to 40 CHAPTER 2. DEEP STACKING

Figure 2.6 SExtractor noise-derived magnitude uncertainties compared with empirically measured values. We use a test image set of 11 z-band images collected as a single epoch under photometric conditions and through low airmass from which a stack is also generated. The zeropoints of the images and the resultant stack are similar, and are calibrated relative to one another. Red points correspond to sources from the stack with SExtractor measured magnitudes and noise-derived magnitude uncertainties. Blue points represent the average SExtractor measured magnitudes of sources in common among the set of 11 images and in the stack (x-axis), and the sample standard deviation about the average (y-axis). Note that variable sources, saturating sources, and sources near CCD chip gaps produce much higher empirical uncertainties than can be accounted for by noise alone. The spread of the empirical uncertainties is slightly broader than the noise-derived uncertainties, but the trends are nearly identical. This indicates SExtrac- tor noise-derived uncertainties in image stacks are generally reliable provided no systematics are introduced. 2.10. DEPTH ESTIMATES AND CONCLUSION 41 determine the depth of the resultant catalogue. Another advantage of ASSOC_MODE is the dynamic adjustment of apertures. Using either MAG_AUTO or MAG_ISO the region defined to measure the flux of a source is provided by the source itself in every instance. In this way adjustments among the different filter images for seeing and morphology are internal. This is problematic in the rare case when a clumpy source or interacting pair is extracted as a single source in one filter but multiple sources in another. Such a discrepancy is obvious upon visual inspection of apertures and DUAL_MODE provides more accurate colour information in these cases. In using ASSOC_MODE an association radius must be defined which is large enough to associate most real sources with their catalogue counterparts but small enough to minimise instances of sources being associated with multiple catalogue entries. Our 3σ astrometric precision is ∼2 pixels (∼0.500) while the average FWHM is ∼4 pixels (∼100; see Table 2.1). In Figure 2.7 we determine an appropriate association radius with a method similar to that used to determine appropriate detection thresholds (Figure 2.3). We find using wide association radii of 3–4 pixels introduces only tens of blended sources to inspect (∼0.01%). In DUAL_MODE images are specified for aperture generation and photometry separately. In this way a deep image can be used to generate apertures for photometry in a shallower image, allowing for flux measurements of sources below the detection threshold. Such measurements are highly uncertain but are still useful for constraining colour information and quantifying the significance of any observed brightening. The apertures used in DUAL_MODE are tied to the seeing and morphology of the aperture image and are not adaptive as with ASSOC_MODE. This is generally disadvantageous but produces more accurate colour information in the rare case of a blended source which appears singular in one image and as a multiple source in another. The choice of detection threshold is more pertinent in DUAL_MODE as spurious detections are not eliminated from the aperture image. We primarily rely on colour information provided by ASSOC_MODE for LBGS and use DUAL_MODE colours for blended sources and seasonal and nightly monitoring of HR hosts (see Chapter 3).

2.10 Depth Estimates and Conclusion

From the extracted photometry of the stacked images we make two estimates of the depth of each stack and compare them with the DECam exposure time calculator5 (ETC) using the sum of the input image EETs (see Table 2.4).

5The DECam exposure time calculator, written by Daren DePoy in 2008 and last updated in 2016, is freely available as part of the DECam user guide at http://www.ctio.noao.edu/noao/node/5826 42 CHAPTER 2. DEEP STACKING

Figure 2.7 Source recovery and ambiguous associations as a function of association radius. The blue line represents sources recovered in a g-band deep stack relative to total sources in the r-band deep stack association catalogue as a function of the ASSOC_MODE association radius. The number of sources recovered in the g-band deep stack that are associated with multiple r-band deep stack catalogue sources is represented by the red line. The number of ambiguous associations is kept low but not minimised, as some of these detections may be source pairs of interest, deblended in one ﬁlter but not in the other.

For one estimate we identify the magnitude at which the average magnitude uncertainty is 0.198mag, which is the uncertainty of a source with S/N = 5. This is seen by solving for the 1σ ∆m uncertainty of m5σ in Equation 2.4 using a ﬂux ratio of 6/5, the ﬂux of a 5σ detection plus the 1σ noise:

(m5σ − ∆m) − m5σ = −∆m = −2.5 · log10(6/5) = −0.198 (2.8)

We also calculate m5σ directly using Equation 2.4. Values of m0 are set relative to single epoch catalogues calibrated as discussed in Section 2.8.1. The FWHMs are taken as averages of SExtractor-measured FWHMs of SExtractor-identified point-sources. The 1σ background noise levels are as output by SExtractor. The similarity of the estimated depths suggests we have accounted for most systematics. Also the estimated depths approach the ETC projections, indicating the stacking methodology is sound. The depths are substantial, enabling LBGS on significant portions of the luminosity functions of z ∼ 2–4 LBGs (see Figure 1.6). The implied selection size ensures sufficient hosts to monitor for HR SLSNe. 2.10. DEPTH ESTIMATES AND CONCLUSION 43

Table 2.4 DECam u0griz 5σ limiting magnitudes of C3 deep stacks

a b c d e Filter m0 FWHM Noise CET Error5σ FWHM/Noise5σ ETC5σ pixels counts/pixel sec mag mag mag u0 28.95 3.90 0.36 27200 25.84 26.21 26.62 g 30.81 4.20 0.68 39400 27.28 27.30 27.31 r 31.76 4.02 1.43 76000 27.41 27.48 27.43 i 31.63 3.82 1.62 106920 27.11 27.28 27.28 z 31.22 4.16 1.42 225390 26.89 26.92 26.98 a The central pixel scale of the DECam is 0.2600/pixel. b Cumulative Exposure Time, or actual time spent on exposures. c 5σ limiting magnitude from Equation 2.8. d 5σ limiting magnitude from Equation 2.4. e 5σ limiting magnitude estimate from the DECam Exposure Time Calculator.

3 Lyman Break Galaxy Selection with DECam

With the deep stacks designed in Chapter 2, we proceed with Lyman Break Galaxy Selec- tion (LBGS) on u0SUDSS fields. LBGS produces catalogues of high redshift (HR; z & 2) LBGs with a low percentage of low redshift interlopers (LRIs) which can then be monitored as potential hosts of HR superluminous supernovae (SLSNe). LBGS is a colour selection method which must be calibrated to the filter set of the imager in use. We perform an initial calibration on the DECam using model star-forming galaxy spectral templates and composite spectra of z ∼ 3 LBGs. The criteria are refined using sources with spectroscopically confirmed redshifts. We estimate selection efficiencies using the refined criteria of 70–85% and completeness measures of 40–50%. The DECam LBGS catalogues of DES C3 include ∼150,000 sources, while other u0SUDSS fields have smaller but comparable catalogue sizes.

3.1 Introduction

For the redshift range in which Lyman-α (1216Å) is visible to optical detectors (1.5 . z . 7.2 from 3000–10000Å) the SEDs of galaxies exhibit a distinctive break in the Ly-α forest (912Å–1216Å). This is a consequence of repeated Ly-α absorption by intervening, optically thick neutral hydrogen systems (McDonald et al., 2006). Shortward of the Lyman-limit (912Å) there is a second break due to the near complete absorption of ionising photons by intrinsic neutral hydrogen. The Ly-α forest eﬀect intensiﬁes with redshift and also results in near complete absorption by z ∼ 6 (Fan et al., 2006; Bañados et al., 2016). Thus from 2.3 . z . 6 there are two optical breaks in galaxy SEDs. However these breaks are only detectable in our data for galaxies which are luminous in the restframe far-UV. Galaxy populations at 2.3 . z . 7.2 can be divided into the three general categories of quiescent, dusty star-forming and blue star-forming (Spitler et al., 2014). The last of

45 46 CHAPTER 3. DECAM LBGS these are dominated by LBGs as these galaxies are bright enough in the restframe for the breaks discussed above to be apparent in our data. These breaks give LBGs a distinctive appearance in colour space, allowing them to be eﬃciently selected with the use of colour cuts to broadband photometry (i.e., LBGS). LBGs constitute up to 65% of z ∼ 2 galaxies with a large uncertainty (Adelberger et al., 2004), and LBGS is a convenient method for producing large catalogues of HR sources.

LBGS was first developed in the late 1990s as a means of accelerating the study of the HR Universe (Madau, 1995; Madau et al., 1996; Steidel et al., 1996, 1999; Lowenthal et al., 1997). The advent of 10m-class telescopes had made practical the spectroscopic observation of HR objects, but the identification of such objects could only be made through blind spectroscopic surveys or with the use of photometric redshifts. The time required to accumulate a sizeable sample of HR objects with the former method is prohibitively long, while the latter method requires coordinated photometry with multiple instruments which must also go deep enough to adequately sample HR galaxy populations. LBGS rivals the selection efficiency of any current photometric redshift method, but can be performed with a single instrument with as few as three filters.

A survey for z ∼ 3 galaxies using LBGS with photometry from a set of custom filters (Steidel UnGRi hereafter; Steidel & Hamilton, 1993) found a selection efficiency (the percentage of selected objects which are HR LBGs) of ∼95% and ∼30× the spectroscopic efficiency of a blind spectroscopic survey for targeting HR sources (Steidel et al., 2003). The collected spectra enabled one of the first studies of the far-UV characteristics of HR LBGs, revealing common nebular emission features and high- and low-ionisation interstellar absorption features (Shapley et al., 2003). The identification and alignment of these features is still the most reliable method for determining the precise redshifts of HR LBGs from spectra (Pettini et al., 2001; Steidel et al., 2003, 2004; Shapley et al., 2003; Adelberger et al., 2004; Cooke et al., 2006; Jones et al., 2012).

A number of similar surveys with deeper and wider photometry have since been con- ducted (e.g., Paltani et al., 2007; Yoshida et al., 2008). LBGS was performed on the Keck Deep Fields (KDF), imaged with Keck LRIS. The fields were imaged in a set of filters identical to the Steidel UnGRi set, allowing the use of the same selection criteria. But the KDF observations are ∼1.5 mag deeper than the photometry of earlier work, probing further into the LBG luminosity functions at different redshifts (Sawicki & Thompson, 2005, 2006a,b; Savoy et al., 2011). LBGS was also performed on the SNLS Deep Fields, imaged with the CFHT MegaCam (Astier et al., 2006). The MegaCam has a set of SDSS-like griz filters, and a u∗-band that is broader and redder than the SDSS u-band. An independent 3.1. INTRODUCTION 47

Table 3.1 LBGS criteria of the Steidel UnGRi ﬁlter set

Publication Redshift Bin Selection Criteria

Adelberger et al. (2004) 1.5 . z . 2.0 (Un − G) ≤ (G − R) + 0.2

(Un − G) ≥ (G − R) − 0.1

(Un − G) ≥ 5 × (G − R) − 2.0 (G − R) ≥ −0.2

Adelberger et al. (2004) 2.0 . z . 2.5 (Un − G) ≤ (G − R) + 1.0

(Un − G) ≥ (G − R) + 0.2

(Un − G) ≥ 5 × (G − R) − 2.0 (G − R) ≥ −0.2

Steidel et al. (2003) 2.7 . z . 3.3 (Un − G) > (G − R) + 1.0 (G − R) ≤ 1.2 Steidel et al. (1999) 3.8 . z . 4.5 (G − R) ≥ 2.0 (G − R) ≥ 2.0 × (R − i) + 1.5 (R − i) ≤ 0.6

set of LBGS criteria were developed for this ﬁlter set (Cooke et al., 2013). The photometry is of similar depth to KDF but covers nearly 100× the sky area. LBGS catalogue sources from this work number in the hundreds of thousands, allowing large population studies of HR LBGs (Cooke et al., 2014), including searches for rare UV-bright supernovae (Cooke, 2008).

Here we calibrate the DECam for LBGS. The constructed deep stacks discussed in Chapter 2 provide large catalogues of sources with colour information. We establish an initial set of colour selection criteria based on synthesized colours of model star-forming galaxy spectral templates. SLSN candidates are identified by monitoring these initial selections, and follow-up spectroscopy is performed with Keck LRIS (see Chapters 4 and 5). Keck LRIS is a multi-object spectrograph, and slit masks are designed for follow-up which allow the collection of spectra of nearby field LBGS catalogue objects. These spectra are examined and redshifts confirmed, providing real source colour information to refine the colour selection criteria and measure efficiency. Additional spectra are drawn from ZFOURGE catalogues to aid in this refinement, and we use the ZFOURGE photometric redshift catalogue to analyze the DECam LBGS efficiencies. 48 CHAPTER 3. DECAM LBGS

3.2 Filter Sets

We compare several ﬁlter sets to identify diﬀerences relevant for calibrating DECam for

LBGS. In Figure 3.1 we plot the ﬁlter throughput response functions of the Steidel UnGRi (Steidel & Hamilton, 1993), the MegaCam1 u∗griz and the DECam2 u0griz ﬁlter sets. Also plotted is the SDSS3 standard ugriz set for reference.

Overall the DECam filter set is more similar to MegaCam than Steidel UnGRi and the colour selection criteria are also more similar (see Tables 3.2, 3.3, 3.4). The Steidel R-band more closely resembles the DECam i-band, and this is reflected in source colour planes (see Section 3.5). The DECam u0-band is narrower than the MegaCam u*-band and the Steidel Un-band (see Section 2.8.2), which narrows the redshift range over which z ∼ 3 LBGs exhibit two distinct flux breaks and lowers selection efficiency in this bin. The asymmetrical shape of the u0-band response function also lowers the selection efficiency of LBGs at the low redshift end of the z ∼ 2 bin. This is not ideal but there is no better alternative for deep-and-wide u-band imaging of the Southern Sky at present.

3.3 LBG Composite Spectra

Spectra of LBGs and model star-forming galaxies are used to set up the initial LBGS criteria for a new imager (Steidel et al., 2003; Cooke et al., 2013). The spectra are passed through the filter throughput response functions to estimate the DECam observed colours of different types of objects at different redshifts. The LBG composite spectra of Shapley et al. (2003) (the Shapley composite spectra hereafter) are extremely useful in this regard. They divide 794 z ∼ 3 LBG spectra into quartiles based on the equivalent widths of the Ly-α features and combine the divided spectra into four high S/N composites (see Figure 3.2). The composites are an accurate representation of the typical spectroscopic appearance of z ∼ 3 LBGs and provide an inclusive range of colour expectations in this redshift regime. There is a correlation between Ly-α emission strength and continuum blueness, making the composite of strong emitters much bluer than the composite of absorbers in observed optical colours corresponding to the restframe far-UV. There is also a correlation between Ly-α equivalent width and

1Specifications of the CFHT MegaCam can be found at https://www.cfht.hawaii.edu/Instruments/Imaging/Megacam/ 2DECam filter throughput information can be found at http://www.ctio.noao.edu/noao/content/DECam-filter-information 3SDSS filter throughput information can be found at http://skyserver.sdss.org/dr2/en/proj/advanced/color/sdssfilters.asp 3.3. LBG COMPOSITE SPECTRA 49

Figure 3.1 Filter throughput response functions for the four filter sets discussed in Section 3.2. The shapes of the function profiles are preserved but the efficiency has been scaled arbitrarily to separate the filter sets for clearer presentation. The actual throughput efficiencies are similar and have a much smaller effect on depth than telescope primary mirror size and cumulative exposure time. Initial DECam LBGS redshift bin endpoints are determined by setting filter sensitivity cut-off wavelengths to Ly-α and the Ly-limit (see Figure 3.4). 50 CHAPTER 3. DECAM LBGS intrinsic dimness such that brighter LBGS objects are redder and dimmer LBGS objects are bluer. The spectra extend to restframe 2000Å, corresponding to 8000Å in the observer-frame for z ∼ 3 LBGs. In order to garner colour information over the full wavelength range of the DECam u0griz filter set, we extend the red tails of the composite spectra with StarBurst (SB) model spectral templates (Calzetti et al., 1994) through the z-band (∼11000Å, see Figure 3.3). The ‘SB1’ template (E(B-V)<0.10) is well fit to the bluer continuum of the composite of strong emitters and the next strongest Ly-α equivalent width quartile, and the dustier ‘SB2’ template (0.10

3.4 Model Spectral Templates

We pass model spectral templates of star-forming galaxies through the ﬁlter response functions to generate redshift tracks in colour-colour space which sketch out the regions of interest. The Shapley composite spectra are useful for marking the colour ranges expected of LBGs at z ∼ 3, but projection to lower and higher redshifts requires adjustment to the Ly-α forest and extension of the spectra to longer wavelengths. Rather than introduce uncertainties into the composite spectra, we use the spectra of models of similar objects with more extensive wavelength coverage. A variety of star-forming models are used to account for diﬀerences within the LBG population apparent from the Shapley composite spectra in Figure 3.3. We use the bluest star-forming galaxy models from Coleman et al. (1980), labeled ‘Sbc’ and ‘Scd’, along with an irregular galaxy model undergoing enhanced star formation (‘Irr’). We also include the bluest StarBurst models from Calzetti et al. (1994), SB1 and SB2, used to extend the red tails of the Shapley composite spectra. LBGs with active galactic nuclei (AGN or QSOs) follow a similar track as non-active LBGs in colour-colour space, and we also include a QSO composite spectrum from Lanzetta et al. (1993). Two known types of low redshift interlopers (LRIs) that tend to contaminate LBGS are lenticular galaxies and old elliptical galaxies (Adelberger et al., 2004; Cooke et al., 2013). The SEDs of these LRIs resemble HR LBGs, with the Balmer break mimicking the Ly-limit and the red continua mimicking the Ly-α forest. We use an Elliptical/S0 (‘ES0’) model from Coleman et al. (1980) to estimate the colours of lenticulars and a 2Gyr elliptical model (‘E2G’) from (Bruzual & Charlot, 2003) to estimate the colours of old ellipticals. The star-forming spectral templates in Figure 3.3 are shown as they would appear if observed at z ∼ 3, along with the discussed LRIs at z = 0.3. Shortward of the Ly-limit 3.4. MODEL SPECTRAL TEMPLATES 51

Figure 3.2 The four z ∼ 3 LBG composite spectra from Shapley et al. (2003), normalized at 2000Å and smoothed with a first order Gaussian convolution. The composite spectrum of each quartile is shown in ascending order of Ly-α equivalent width strength with red representing the absorbers and blue representing the strong emitters. The spectra are offset slightly for clarity, but the apparent trend of bluer colour for stronger Ly-α equivalent width is real. The z ∼ 3 Ly-α forest is apparent from Ly-α to the Ly-limit, both marked by vertical dashed black lines. Longward of Ly-α the spectra are remarkably flat, producing characteristically neutral colours over this wavelength range. Shortward of the Ly-limit the flux is essentially zero. 52 CHAPTER 3. DECAM LBGS

Figure 3.3 Spectral templates of star-forming model galaxies which resemble LBGs in the far-UV as observed at z ∼ 3. The template labels are identified in the legend (see text). The DECam u0griz response functions are shown as coloured dashed lines along the x-axis. Colour estimates of star-forming galaxies at different redshifts are made by redshifting the templates and passing them through the response functions. Spectral templates of expected LRIs are also shown as thick dashed lines as indicated in the legend (see text). These are displayed as observed in optical bands at z = 0.3. we assume zero flux. To calculate the fractional amount of Ly-α forest absorption, DA, we use the equation for the redshift evolution of DA from (Kirkman et al., 2005):

DA = 0.0062 × (1 + z)2.75 (3.1)

This formula quantifies the decrement from hydrogen only, but additional absorption from metals is insignificant in comparison. There is consistency with the observed Ly-α forest effect within the Shapley composite spectra, allowing extension to lower and higher redshifts. 3.5. MODEL-BASED COLOUR SELECTION CRITERIA 53

3.5 Model-Based Colour Selection Criteria

We estimate the colours of HR LBGs and LRIs by passing the composite and template spectra through the DECam filter throughput response functions. This process is illustrated in Figure 3.4. From this we assign our initial LBG redshift selection bins as 1.9 < z < 2.6, 2.6 < z < 3.4 and 3.4 < z < 4.9 which are referred to as the z ∼ 2, z ∼ 3 and z ∼ 4 bins respectively. The bins can be contrasted against the bins defined by Steidel et al. (2003); Adelberger et al. (2004), namely 1.4 < z < 2.1, 1.9 < z < 2.7, 2.7 < z < 3.4 and 3.8 < z < 4.5. Our lowest redshift is set where Ly-α enters the sensitive region of the DECam u0-band. At this redshift LBGs are selected by their characteristically flat SEDs (in flux per unit frequency, fν), and this flatness persists to lower redshifts (1.2 < z < 1.9) at which Ly-α is not visible in the optical. Many sources in this redshift range meet the criteria of the z ∼ 2 bin, and we test pushing the redshift limit of the bin lower to increase the selection size without decreasing efficiency (see Section 3.6). However we are cautious of going so low as to leave the HR regime of greatest interest (z & 2). We have defined our bins exclusively, though significant overlap is expected given the range of colours within the HR LBG population itself. Overlap between bins is not a large concern as the HR label is still valid and the overlap is localized about the borders of the colour cuts. Defining the bins exclusively provides more useful bin populations for efficiency analysis (see Section 3.7). Figure 3.5 consists of the six most relevant colour planes for the selection of z ∼ 2, z ∼ 3 and z ∼ 4 LBGs. Within each figure are plotted, as solid lines, the redshift tracks of the star-forming galaxy spectral templates and, as dashed lines, the redshift tracks of the LRI spectral templates, all discussed in Section 3.4. The colour of each track is matched to the respective template in Figure 3.3. Along each star-forming template track are marked, as squares, the beginning of the track (z = 0) and the boundaries of each visible redshift bin. The beginning and selected end (z = 2) of the LRI template tracks are marked with triangles. The colours of the four Shapley composite spectra are marked as stars, and the colour of each star is matched to the respective spectrum in Figure 3.2. The translucent contours are of source colour isodensity with a logarithmic colour scale for ∼700,000 sources extracted from the C3 deep stacks. Finally the initial colour cuts are given by the shaded regions, with the z ∼ 2, z ∼ 3 and z ∼ 4 bins represented in blue, green and red respectively. The contours in the (u0 − g) vs. (g − i) plane are indicative of the principle of LBGS. The galactic locus spans from 0.0 . (u0 − g) . 0.4 and 0.0 . (g − i) . 1.0. Elliptical galaxies and dim M-type stars are located in the upper right of the plane, producing the 54 CHAPTER 3. DECAM LBGS

Figure 3.4 Initial redshift bin assignment for LBGS with DECam. The Shapley composite spectrum of weak Ly-α emitters is shown as observed at particular redshifts. The Ly-α forest is attenuated using Equation 3.1 and the SB2 template is appended longward of 2000Å, restframe. 0 The DECam u griz ﬁlters are shown as dashed lines along with the Steidel UnGRi ﬁlters as solid lines. The front of the z ∼ 2 bin (blue spectrum) is set as Ly-α enters the sensitive portion of the u0-band (see text). The front of the z ∼ 3 bin (orange spectrum) is set as the Ly-limit enters the u0-band and the back of the bin (green spectrum) is set as Ly-α exits the g-band, ensuring that all objects in the z ∼ 3 bin exhibit two distinct colour breaks from the Ly-α forest and the Ly-limit. The back of the z ∼ 4 bin (red spectrum) is set as Ly-α exits the r-band. 3.5. MODEL-BASED COLOUR SELECTION CRITERIA 55

Figure 3.5 The six most relevant colour planes for establishing the initial DECam LBGS criteria. Each given legend applies to all panels of the ﬁgure. The colour schemes of the Shapley composite spectra and the model redshift tracks are coordinated with Figures 3.2 and 3.3. See text for further explanation of the signiﬁcance of particular features. 56 CHAPTER 3. DECAM LBGS

Figure 3.5 Continued 3.5. MODEL-BASED COLOUR SELECTION CRITERIA 57

Figure 3.5 Continued 58 CHAPTER 3. DECAM LBGS apparent overdensity of sources in this area. The model star-forming galaxy redshift tracks indicate that as redshift increases, these types of sources move nearly straight up in the plane. The colours of the Shapley composite spectra suggest that the model tracks skirt the right edge of the area of the plane where HR LBGs reside. A second overdensity indicated by the contours in the upper left of the plane corroborates this interpretation. The z ∼ 3 colour regions are designed to encompass this overdensity suspected of being populated predominantly by HR LBGs.

The colour selection areas are designed to encompass the high redshift portions of the model tracks while excluding the low redshift portions and avoiding the model elliptical redshift tracks. There is substantial overlap from the very low redshift portions of the model tracks in the z ∼ 2 area before the tracks move out of the bin to the right at intermediate redshifts. The physical proximity of this sub-population within our selection ensures that the apparent magnitude distribution of the members is skewed to brighter values. To reduce the level of this type of LRI contamination in the z ∼ 2 bin, we enforce a minimum magnitude limit for inclusion. The luminosity functions of HR LBGs from Parsa et al. (2016) suggest that the number density of z ∼ 2 LBGs with apparent magnitudes brighter than i = 22 is insigniﬁcant in our volumes (see Figure 1.6). In the z ∼ 3 and z ∼ 4 areas there is some overlap by the elliptical redshift tracks. By making use of multiple colour planes the ES0 track is eﬀectively avoided, but the model star-forming tracks cannot be fully disentangled from the E2G track.

The formal criteria are given in Tables 3.2, 3.3, and 3.4 alongside the LBGS criteria for the Steidel UnGRi filter set (Steidel et al., 1999, 2003; Adelberger et al., 2004), the CFHT MegaCam filter set (Cooke et al., 2013) and spectroscopically refined criteria (see Section 3.6). In the z ∼ 2 bin we use a constant upper limit in (u0 − g) and in (g − i) for simplicity until these limits can be refined with observed spectra. In the z ∼ 3 bin our criteria are similar to the MegaCam criteria, but we use the (u0 −g) vs. (g −z) colour plane instead of the (u0 − g) vs. (g − r) colour plane. Criteria for the (u0 − g) vs. (g − r) colour plane are added in the spectroscopic refinement. In the z ∼ 4 bin our criteria are similar to the Steidel criteria with the addition of the (g − r) vs. (r − z) colour plane available with deep z-band photometry.

Additional criteria are enforced to reduce LRI contamination within the selections.

Only conﬁdent 5σ detections (σi ≤ 0.2 mag) are considered. Also a 5σ detection is required in the Ly-α forest ﬁlter of each selection (u0, g, r for z ∼ 2, z ∼ 3, z ∼ 4 respectively). Each of the z ∼ 2, z ∼ 3 and z ∼ 4 bins have minimum magnitude restrictions of i ≥ 22, 23 and 24, respectively, based on the luminosity functions of LBGs at these redshifts 3.6. SPECTROSCOPIC COLOUR SELECTION CRITERIA 59

(Parsa et al., 2016). Finally we require the SExtractor CLASS value be less than 0.98 to exclude stars from the selections.

3.6 Spectroscopic Colour Selection Criteria

The initial model-based colour selection criteria produce catalogues of LBGs which are monitored for transient activity in the ongoing DES and SUDSS surveys. HR SLSN candidates are identified in the DES C3 field and spectroscopic follow-up is performed with Keck LRIS (see Chapter 4). Additional candidates in the COSMOS field are provided by SHIZUCA for follow-up (see Chapter 5). We use the multi-object capability of LRIS to collect 30–40 spectra of LBGS field sources in the vicinity of each HR SLSN candidate. Spectroscopic measurements of their redshifts are made and the additional data is used to refine the model-based DECam LBGS criteria. Four spectroscopic follow-up campaigns have been performed with Keck LRIS using time awarded through the Swinburne Keck time access program (see Table 3.5). The depth of observations is severely reduced by light cirrus during each campaign, as is evident from the low ratio of masks imaged to masks designed and of redshifts identified to objects observed. The spectra are reduced manually with IRAF4 and each slit is visually inspected for a dispersion line. Where dispersion lines are detected, redshifts are derived by identifying characteristic features such as the Ly-α forest decrement and Ly-α emission for high redshift, or common transitions of low redshift galaxies (i.e., [Oii] λλ3727, H-β, [Oiii] λ5007). Each measured redshift is given a confidence score out of 4 depending on the significance and alignment of the observed features (see Figure 3.6). The ZFOURGE spectroscopic library provides a second source of spectroscopic redshift information for DECam LBGS catalogue objects. ZFOURGE is a deep, near-IR survey with supplementary optical and mid-IR photometry and narrow-band photometry designed to enable precision photometric redshifting over a wide range in deep imaging (Straatman et al., 2016). ZFOURGE test their photometric redshifts with spectroscopic samplings and these spectroscopic redshifts are included in the ZFOURGE data release5. One of the ZFOURGE fields overlaps with the DES C3 field, providing several hundred additional spectroscopic redshifts for sources in the C3 deep stack catalogue. Figure 3.6 is similar to Figure 3.5 with the addition of individual sources with spectroscopic redshifts. The (u0 − r) vs. (r − z) plane is replaced with the (u0 − g) vs. (g − r)

4IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. 5ZFOURGE data products are available at http://zfourge.tamu.edu/ 60 CHAPTER 3. DECAM LBGS

Table 3.2 z ∼ 2 LBGS criteria of selected instruments

Filter Set Redshift Bin Selection Criteria

Steidel UnGRi 1.5 . z . 2.0 (Un − G) ≤ (G − R) + 0.2

(Adelberger et al., 2004) (Un − G) ≥ (G − R) − 0.1

(Un − G) ≥ 5 × (G − R) − 2.0 (G − R) ≥ −0.2

Steidel UnGRi 2.0 . z . 2.5 (Un − G) ≤ (G − R) + 1.0

(Adelberger et al., 2004) (Un − G) ≥ (G − R) + 0.2

(Un − G) ≥ 5 × (G − R) − 2.0 (G − R) ≥ −0.2 DECam u0griz: 1.9 . z . 2.6 −0.2 < (u0 − g) ≤ 0.7 model-based (g − i) ≤ 0.8 (this work) (r − z) ≤ 0.6 (g − z) ≤ 1.2 DECam u0griz: 1.7 . z . 2.6 (u0 − g) ≥ −0.2 spectroscopic (u0 − g) ≤ 0.32 × (g − i) + 1.03 (this work) (u0 − g) ≥ 0.76 × (g − i) − 0.24 (u0 − g) ≥ 8.09 × (g − i) − 4.67 (g − i) ≥ −0.6 (u0 − g) ≥ 2.17 × (g − z) − 1.36 (g − z) ≤ 1.04 (g − z) ≥ −0.6 (u0 − g) ≥ 2.319 × (g − r) − 0.71 (g − r) ≥ 0.0 (u0 − r) ≥ 0.83 × (r − z) − 0.09 3.6. SPECTROSCOPIC COLOUR SELECTION CRITERIA 61

Table 3.3 z ∼ 3 LBGS criteria of selected instruments

Source Redshifts Selection Criteria

Steidel UnGRi 2.7 . z . 3.3 (Un − G) > (G − R) + 1.0 (Steidel et al., 2003) (G − R) ≤ 1.2 MegaCam u∗griz 2.7 . z . 3.3 (u∗ − g) > 0.7 (Cooke et al., 2013) (u∗ − g) > 1.2 × (g − r) + 0.9 −1.0 < (g − r) < 1.0 (u∗ − g) > (g − i) + 0.7 −1.0 < (g − i) < 1.3 (r − i) < 0.4 DECam u0griz: 2.6 . z . 3.4 (u0 − g) > 0.7 model-based (u0 − g) ≥ 3 × (g − i) − 1.7 (this work) (g − i) ≤ 1.3 (u0 − g) ≥ 3 × (g − z) − 2.3 (g − z) ≤ 1.3 (r − i) ≤ 0.3 (r − z) ≤ 0.6 DECam u0griz: 2.6 . z . 3.5 (u0 − g) > 0.32 × (g − i) + 1.03 spectroscopic (u0 − g) ≥ 2.64 × (g − i) − 1.34 (this work) (g − i) ≥ −0.6 (u0 − g) ≥ 6.0 × (g − z) − 5.76 (g − z) ≥ −0.6 (u0 − g) ≥ 3.41 × (g − r) − 1.60 (g − r) ≥ −0.2 (g − r) ≤ 1.2 (r − i) ≥ −0.4 (r − i) ≤ 0.42 (r − z) ≥ −1.0 (r − z) ≤ 0.5 (g − i) ≤ −3.62 × (i − z) + 2.34 (i − z) ≥ −0.7 62 CHAPTER 3. DECAM LBGS

Table 3.4 z ∼ 4 LBGS criteria of selected instruments

Source Redshifts Selection Criteria

Steidel UnGRi 3.8 . z . 4.5 (G − R) ≥ 2.0 (Steidel et al., 1999) (G − R) ≥ 2.0 × (R − i) + 1.5 (R − i) ≤ 0.6 DECam u0griz: 3.4 . z . 4.9 (g − r) ≥ 0.8 model-based (g − r) ≥ 1.33 × (r − i) + 0.73 (this work) (r − i) ≤ 1.8 (g − r) ≥ 1.67 × (r − z) + 0.333 (r − z) ≤ 1.6 (g − i) ≥ 1.0 (i − z) ≤ 0.25 DECam u0griz: 3.5 . z . 4.9 (g − r) > 1.2 spectroscopic (g − r) ≥ 1.81 × (r − i) + 0.72 (this work) (r − i) ≥ −0.4 (g − r) ≥ 2.0 × (r − z) + 0.2 (r − z) ≥ −1.0 (g − i) ≥ 0.8 −0.7 ≤ (i − z) ≤ 0.4

Table 3.5 Keck LRIS spectroscopic follow-up campaigns

Campaign Designsa Imaged Observedb HR LBGs LRIs No ID 2016DEC26-27 6 2 55 12 2 41 2017MAR23 6 3 88 35 3 50 2017DEC14-15 12 4 107 25 1 81 2018APR11 1 0 0 0 0 0 TOTALS 25 9 250 72 6 172 a Individual multi-object slitmasks designed prior to follow-up and imaged on the night. b The number of objects exposed on, including non-detections. 3.7. PHOTOMETRIC EFFICIENCY ANALYSIS AND CONCLUSION 63 plane which is more informative after including the spectroscopic data. The points are colour and symbol coded by spectroscopic redshift and source as indicated in the legends. The sizes of the crosses representing the Keck LRIS targets increase with the confidence score. The selection criteria on ZFOURGE objects is relaxed by 0.2 mag on all sides so that the redshifts of sources that fall just outside the colour regions are also visible. All ZFOURGE objects with z > 1.7 not identified by DECam LBGS are also shown in the planes. The spectral model redshift tracks are shown with the same colour scheme as in Figure 3.5, with the star-forming model tracks starting at z = 1.7 for clarity. The refined z ∼ 2, z ∼ 3 and z ∼ 4 colour regions are represented in blue, green and red respectively. The refined colour selection criteria are formalized in Tables 3.2, 3.3 and 3.4. The cuts are similar to the model-based criteria but with differences in the slopes of the boundaries to more cleanly separate the different galaxy populations. Realistic limits are included on the blue sides of the colour regions along the x-axis beyond which no HR objects are observed spectroscopically. The redshift bin limits have been adjusted to more discretely separate the spectroscopic source populations. A simple efficiency analysis comparing the spectroscopically refined colour selection criteria to the model-based criteria is given in Table 3.6. LBGS efficiency is a measure of the ratio of HR objects to total objects within the colour regions of a redshift bin. This measure is typically made using a subset of LBGS catalogue objects with spectroscopically verified redshifts (Steidel et al., 2003). Also included is a measure of completeness using the ratio of total ZFOURGE sources with spectroscopic redshifts of z ≥ 1.7 to those within the LBGS colour regions. The efficiency measurement of the z ∼ 2 bin is based on a redshift lower limit of z = 1.2. At the low redshift end of the z ∼ 2 bin the Ly-α forest decrement is quite mild

(see Figure 3.4), and LBGs are selected by the characteristic ﬂatness of their SEDs (in fν). This continues to lower redshifts where Ly-α is not visible in the optical. Many LBGs at these intermediate redshifts (1.2 . z . 1.7) meet the colour selection criteria of the z ∼ 2 bin. We do not treat these objects as LRIs, but rather as the low redshift tail of the z ∼ 2 distribution.

3.7 Photometric Eﬃciency Analysis and Conclusion

The Keck LRIS spectroscopic efficiencies in Table 3.6 are measured in the same manner as the efficiency values from Steidel et al. (2003), and the high efficiencies of both the model-based and spectroscopically refined colour cut criteria are in line with the 95% benchmark. However, the measurement method used assumes the redshift distribution of 64 CHAPTER 3. DECAM LBGS

Figure 3.6 The six most relevant colour planes for constraining the initial model-based selection criteria using observed spectra. Each given legend applies to all panels of the ﬁgure. Upward facing arrows represent lower limits. The model redshift tracks are included, colour-coded as in Figure 3.5, however the star-forming model redshift tracks commence at z = 1.7 for clarity. See text for further explanation of the signiﬁcance of particular features. 3.7. PHOTOMETRIC EFFICIENCY ANALYSIS AND CONCLUSION 65

Figure 3.6 Continued 66 CHAPTER 3. DECAM LBGS

Figure 3.6 Continued 3.7. PHOTOMETRIC EFFICIENCY ANALYSIS AND CONCLUSION 67

Table 3.6 DECam LBGS eﬃciencies of Keck LRIS and ZFOURGE spectra

Bin Keck LRIS ZFOURGE Model S.E.a Spec. S.E.b Spec. S.E.c Completenessd z ∼ 2 88% 100% 89%e 60% z ∼ 3 92% 100% 98% 61% z ∼ 4 93% 100% 100% 58% a The selection efficiency of the model-based criteria relative to Keck LRIS spectra from both C3 and COSMOS. b The selection efficiency of the spectroscopically refined criteria relative to Keck LRIS spectra from both C3 and COSMOS. c The selection efficiency of the spectroscopically refined criteria relative to ZFOURGE spectra from CDFS. d The percentage of all ZFOURGE CDFS spectra in the redshift bin which meet some set of DECam LBGS criteria. e Based on a redshift lower limit of z = 1.2. the spectra of sources which cannot be verified is similar to those that are verified. We also assume that our low confidence spectroscopic redshifts are completely accurate, which is unrealistic. The ZFOURGE spectroscopic redshifts are all of high confidence, but these are taken from the literature to test their photometric redshift accuracy. Such spectra are not a random sampling of sources in the field, but rather are drawn from a variety of targeted spectroscopic surveys. As such the ZFOURGE spectroscopic sample is not necessarily representative of the on-sky distribution of galaxies by type and redshift, which may effect both the efficiency and completeness measures in Table 3.6. We perform an independent efficiency analysis using randomly distributed ZFOURGE photometric redshifts. Each source in the ZFOURGE CDFS catalogue (limiting magnitude Ks5σ = 25.5) includes a photometric redshift and a probability. Crossmatching this catalogue with the DECam LBGS catalogues yields thousands of sources with colour information from deep stacks and confident (probability≥90%) ZFOURGE photometric redshifts. The results of the analysis are given in Table 3.7. ZFOURGE uses the near-IR Ks-band (2 micron) filter for source detection, making it less sensitive than u0SUDSS to z ∼ 2–4 LBGs which are brighter in the observer-frame optical. The lower limits of the spectroscopic efficiency ranges in Table 3.7 are calculated using only sources which are detected in both the ZFOURGE and u0SUDSS deep stack catalogues. The upper limits assume all u0SUDSS deep stack catalogue sources in the ZFOURGE footprint that are selected into a redshift bin but are not recovered in the ZFOURGE catalogue are also HR LBGs. The efficiencies measured with photometric redshifts are lower than in the spectroscopic 68 CHAPTER 3. DECAM LBGS

Table 3.7 DECam LBGS eﬃciencies based on ZFOURGE photometric redshifts of the C3 deep stack source catalogue

Bin Mediana Dispersionb S.E.c Completenessd z ∼ 2 2.12 0.25 84–86%e 38% z ∼ 3 2.97 0.25 78–81% 52% z ∼ 4 3.71 0.20 66–72% 50% a The median redshift of correctly selected sources. b The standard deviation about the median of correctly selected sources. c The selection efficiency of the spectroscopic colour cuts based on ZFOURGE photometric redshifts. d The percentage of all ZFOURGE CDFS sources in the redshift bin which meet some set of DE- Cam LBGS criteria. e Based on a redshift lower limit of z = 1.2. efficiency analysis, but the z ∼ 2 and z ∼ 4 bin efficiencies are similar to those found by Steidel et al. (2004) and Steidel et al. (1999) (83–94% and 66–80% respectively). The z ∼ 3 bin efficiency is quite high but falls significantly short of the 95% benchmark from Steidel et al. (2003). This may be due to the narrower sensitivity of the u0-band to the Ly-limit break at this redshift compared to the Steidel U-band (see Figure 3.4), or a consequence of using photometric redshifts instead of spectroscopic redshifts. As pointed out in Adelberger et al. (2004); Cooke et al. (2013), the selection efficiency can be increased by using more conservative colour cut criteria at the expense of completeness.

Completeness measures are also decreased from the spectroscopic analysis, but not as significantly as the efficiencies. The difference in completeness between the two analyses reflects the severity of the selection bias in the ZFOURGE spectroscopic sample. Estimates on the fraction of HR galaxies that qualify as LBGs range from 14% (Spitler et al., 2014) to 80% (Marchesini et al., 2007), and our estimates fall comfortably between these two extremes. The completeness measure in the z ∼ 4 bin represents a lower limit as the sampling of the LBG luminosity function at this redshift is much shallower.

The DECam LBGS catalogue sizes of the C3 deep stacks are listed in Table 3.8, as well as the density of sources on sky. These are similar to LBGS source densities measured in KDF in the z ∼ 3 and z ∼ 4 bins (Sawicki & Thompson, 2005). In the z ∼ 2 bin, agreement is reached by scaling the KDF density to the limiting magnitude of the u0-band deep stack using the luminosity function of z ∼ 2 LBGs (Parsa et al., 2016; see Figure 1.6). These catalogue sizes ensure that numerous HR transients are likely to be detected in the monitoring stage of LBG S&M. Shallower griz deep stacks on the COSMOS and NSF2 ﬁelds produce less numerous LBGS catalogues in the z ∼ 3 and z ∼ 4 bins, while a shallower 3.7. PHOTOMETRIC EFFICIENCY ANALYSIS AND CONCLUSION 69

Table 3.8 Memberships and densities of DECam LBGS catalogues from the C3 ﬁeld

Bin Membership Density KDFa Bin KDF Density per arcmin2 per arcmin2 z ∼ 2 63,486 7.0 z ∼ 2.2 6.1b z ∼ 3 69,979 7.7 z ∼ 3 8.8 z ∼ 4 18,307 2.0 z ∼ 4 2.5 a From Sawicki & Thompson (2005). b Scaled to the u0-band limiting magnitude of 26 mag. u0 deep stack on X3 reduces the number of LBGS objects in the z ∼ 2 bin and lowers the eﬃciency of the z ∼ 3 bin.

4 Candidate Selection and Spectroscopic Follow-up

The transient detection stage of the Lyman Break Galaxy Selection & Monitoring (LBG S&M) consists of monitoring the DECam LBGS catalogues developed in Chapter 3. Mon- itoring only the sources in these catalogs for transient activity ﬁlters out the majority of low redshift transients, which greatly outnumber the high redshift (HR; z & 2) events.

In addition, the ∆z 0.5 redshift preselection significantly increases the efficiency with which HR superluminous supernova (SLSN) candidates can be identified and aids in the wavelength set-up and feature expectation of spectroscopic follow-up. We focus here on the DES C3 field for candidate identification and Keck spectroscopic follow-up as it is the only u0SUDSS field accessible from Keck for which deep, long-baseline photometry is available and LBGS catalogs are fully developed (see Chapter 5 for SHIZUCA candidates in the COSMOS field). We monitor the DES transient database during seasons 4 and 5 of the survey for brightening in DECam LBGS catalog sources exclusively. We then evaluate the confidence of each transient in the context of HR SLSN origins, recovering 25 candidates. The projected apparent r-band magnitude of a candidate on the date of a classically scheduled spectroscopic follow-up campaign (limiting magnitude mr = 25.5) determines the targeting eligibility. We execute two spectroscopic follow-up campaigns on DES C3 photometric candidates with Keck LRIS, targeting six objects. Dispersion lines for three objects are recovered. Of these, two are likely at low redshift, and the third is identified as an AGN at z = 1.697.

4.1 Introduction

Supernovae are very useful as distance indicators, and there are constantly eﬀorts to increase their observational distance limits. During the ﬁrst surveys for SNe-Ia to measure the rate of expansion of the Universe, the observational distance limit was z ∼ 0.5 (Lei-

71 72 CHAPTER 4. DES SPECTROSCOPY bundgut et al., 1996). Technological improvements in instrumentation along with more expansive and efficient survey collaborations continually push this limit to greater distances and earlier times. Nature provides an additional push with SLSNe. Because an observational distance limit depends on the brightness of the object being observed, the limit on SLSNe is greater than that of ordinary supernovae. SLSNe have been detected in ground-based photometry to z ∼ 2−4 (Smith et al., 2018; Moriya et al., 2019) and are theoretically detectable to z ∼ 6 (Mould et al., 2017). At this redshift the Ly-α forest has arrived at the blue edge of the optical i-band and most of the flux from any source including SLSNe is absorbed at shorter wavelengths. The apparent magnitude of a source with a k-corrected absolute magnitude in the far-UV of M = −21 is mizY ∼ 25.75. Few transient surveys to date achieve this depth per epoch, but this can be overcome by dynamically binning epochs over a period consistent with the redshift-dilated rise time of SLSNe in the far-UV (Cooke, 2008; Cooke et al., 2012). The standard method of detecting transients in a survey is using difference imaging. High quality template images of a field are collected at the beginning of or previous to the survey, and these are subtracted from each subsequent epoch. Sources are then extracted from the differenced image, representing objects which either brightened or were not detected in the template image. This method has been used to detect SLSNe to z ∼ 2 (Berger et al., 2012; Howell et al., 2013; Pan et al., 2017; Smith et al., 2018). The efficiency of detecting HR SLSNe with difference imaging is low. To measure the redshift of a transient detected using this method, spectroscopic follow-up is required. Such follow-up time is limited and most photometric transient candidates go unidentified. In addition, HR SLSNe are greatly outnumbered by low redshift transients with similar apparent magnitudes. Thus the number of HR SLSNe successfully identified this way is likely only a small fraction of the total number of actual photometrically detected HR events. LBG S&M is a more efficient method for the detection of HR SLSNe and UV-bright SNe. Optically detected HR supernovae are observed in the restframe far-UV and so must be UV-bright or superluminous in general. By monitoring catalogues of HR LBGs for transient activity, a much larger fraction of detections are HR transients than with difference imaging. In addition, LBGs likely constitute the majority of HR galaxies within which SLSNe are observable, fitting the characteristics of low redshift SLSN hosts (Neill et al., 2011) and lacking obscuring dust. Low redshift transients (LRTs) are still detected in LBGS low redshift interloper (LRI) contaminants, but the majority of low redshift hosts are excluded from the search. HR superluminous transients other than SLSNe, i.e. active 4.2. LBG MONITORING 73 galactic nuclei (AGN) and tidal disruption events (TDEs), are also detected, but these are highly centralised events and can often be distinguished from SLSNe by their periodicity or colour evolution. LBG S&M was successfully performed in CFHTLS archival data, identifying over 30 HR SN candidates in the photometry. The high redshifts of the hosts of 15 of these candidates were confirmed with late-time spectroscopic follow-up (Cooke et al., in prep.). However, spectra of the candidates near peak could not be collected as they were identified archivally. Such spectra are needed to establish the efficiency of LBG S&M at identifying HR SNe over other types of transients. Here we present spectroscopic follow-up of HR SLSN photometric candidates detected using LBG S&M in DES. We monitor the DES transient catalogue for events observed in members of the DECam LBGS catalogues. SHIZUCA produces photometric candidates independently (see Chapter 5). We evaluate the confidence of detected candidates as HR SLSNe and use projected apparent magnitudes to establish spectroscopic follow-up priority. The reduced spectra are presented and target selection protocol is discussed.

4.2 LBG Monitoring

The transient detection stage of LBG S&M consists of photometrically monitoring colour selected HR LBGs for any increase in luminosity indicative of an ongoing supernova. Be- cause we are attempting spectroscopic follow-up with Keck, C3 is the primary field for monitoring (see Table 1.2). We use the DES web-based transient database (referred to as air traffic control or ATC) which performs transient detection using difference imaging after each epoch of DES photometry on a supernova search field has been collected. Pho- tometric candidates are identified by crossmatching ATC detections with DECam LBGS catalogues. The LBG S&M method as described in Cooke (2008) increases photometric transient detection sensitivity by using seasonal stacks rather than individual epochs (see below). We likewise construct seasonal stacks for the first three seasons of DES C3 photometry for reference and candidate evaluation. We also construct a detection stack with photometry from the first half of season 4 to test against ATC. Each of the transients targeted in the 2016 Keck spectroscopic follow-up campaign are recovered in the detection stack (see Figure 4.1). However, for this work we do not rely on seasonal stacks for transient detection because the DES photometric sensitivity per epoch for ATC detection is approximately the same as the limiting magnitude for spectroscopic follow-up (mr∼25.5). 74 CHAPTER 4. DES SPECTROSCOPY

Table 4.1 Per-epoch and seasonal stack depths in C3 over the ﬁrst three seasons of DES

DECam Filter Epocha S1b S2b S3b g 25.0 27.06 27.02 27.19 r 25.2 26.92 27.09 27.20 i 25.1 26.58 26.79 26.79 z 25.1 26.49 26.47 26.54 a 5σ limiting magnitude based on the DECam exposure time calculator. b 5σ limiting magnitude based on Equation 2.4.

4.2.1 Seasonal Stacks

Because time-dilated HR SLSNe tend to evolve over a period similar to the length of an observing season, seasonal stacks can be used to increase photometric transient detection sensitivity. Such photometry is also useful for measuring the stability of LBGS catalogue objects in quiescence, helping to identify sporadically outbursting AGN. A seasonal stack is constructed in the same way as a deep stack (see Chapter 2), but only over an observational season. An observational season is the ∼6 month window each year during which a field is visible through a reasonably low airmass from a ground-based observatory at a suitable latitude. Seasonal stack photometry is performed with SExtractor in DUAL_MODE using the deep stack of each filter for aperture creation. Seasonal stacks are shallower than deep stacks constructed over multiple seasons, and many objects in the LBGS catalogues are not detectable at high confidence in the seasonal stacks. Using the deep stacks for aperture creation in DUAL_MODE ensures LBGS catalogue objects that are not detected in a seasonal stack are still monitored for significant brightening. Seasonally averaged brightenings in LBGS catalogue sources above the limiting magnitude of the seasonal stack are detected irrespective of the brightness of the source in quiescence.

4.3 Photometric Candidate Evaluation

The photometric candidates drawn from ATC are evaluated for HR SLSN characteristics to filter out other types of transients such as LRTs, AGN and TDEs based on their light curve profiles, host morphology and ATC low redshift supernova template fits. Candidates considered potential HR SLSNe are assigned a confidence score from 1 for low confidence to 3 for highest confidence, and an apparent r-band magnitude projection is made for the date of spectroscopic follow-up. Candidates with projected apparent magnitudes brighter 4.3. PHOTOMETRIC CANDIDATE EVALUATION 75

Figure 4.1 ATC five season light curves of three transients targeted in the 2016 Keck spectroscopic follow-up campaign overlaid with seasonal and detection stack photometry. ATC forced photometry on differenced images of the transient alone is shown in squares and downward triangles (upper limits). Seasonal stack photometry of the transient host and detection stack photometry of the transient and host combined is shown as diamonds. 76 CHAPTER 4. DES SPECTROSCOPY than the limiting magnitude of spectroscopic follow-up (mr ∼ 25.5) are potential targets. LRTs arising in low redshift elliptical and star-forming galaxies which mimic HR LBGs are the most common type of false detection when searching for HR SLSNe. Because of their lower luminosity distance, supernovae in these hosts are detectable to much dimmer absolute magnitudes. These greatly outnumber SLSNe and UV-bright SNe, thus the ∼20% LRI contamination rate of DECam LBGS (see Table 3.7) is expected to be higher when considering only the hosts of detected transients. However, other host and transient observables can be used to identify LRTs and reduce contamination. The areas in the colour planes where a transient host falls indicate the probability and type of a LRI host. The colour selection regions (see Figure 3.5) can be subdivided into regions with a high instance of low redshift star-forming galaxies (the red side along the x-axis of the z ∼ 2 regions), a high instance of elliptical galaxies (the red side along the x-axis of the z ∼ 3 and z ∼ 4 regions), and a low instance of LRIs (the blue side along the x-axis of all regions). We also use the host morphology and k-corrected absolute magnitude range of each transient as indicators. The morphological structure of a LRI is often too complex and extended to be confused with a HR LBG. And the brightness of a LRT is often unrealistic (MFUV < −23) when projected to z & 2. Elliptical LRIs within the DECam LBGS catalogues are expected to host only SNe-Ia. This is useful as SNe-Ia can be confidently identified in photometry alone from their highly regular light curve profiles. The ATC includes a SN-Ia probability fit to every detected transient. Sources selected in the colour region with a high instance of ellipticals that host transients with a SN-Ia probability greater than 80% are regarded as confident LRTs. These objects are not considered for spectroscopic follow-up, but we maintain a catalogue of such events for population comparisons (see Section 4.5). Star-forming LRIs in the DECam LBGS catalogues host all types of supernovae, and CC-SNe cannot be as reliably identified from photometry as SNe-Ia. Only the z ∼ 2 bin is expected to include a significant number of star-forming LRIs. At this redshift an absolute magnitude of MFUV = −21 corresponds to an apparent magnitude of moptical = 23.5. HR SLSNe brighter than this in the far-UV typically last 50–150 days in the observer-frame optical (Cooke et al., 2009; Nicholl et al., 2017) above the ATC limiting magnitude of m5σ ∼ 25.5. Events that evolve significantly faster or slower than this are more likely to have originated in a low redshift host (but see Sections 1.2.2 and 1.2.3 and references therein for scenarios concerning more slowly evolving events and their expected observable behaviours). Detected LRTs are most likely to arise in LRI hosts meeting the criteria for LBGS, but 4.4. KECK SPECTROSCOPIC FOLLOW-UP 77 transient host confusion in dense fields is also a possibility. We examine all ATC transients that fall within 200 of a DECam LBGS catalogue object (∼16kpc projected radius at z ∼ 2). This search radius is chosen with the aim of being sufficiently small to prevent host confusion, but is still inclusive of most of the light from the DECam LBGS catalogue sources where the transient frequency is highest. For candidates exhibiting LRT behaviour the risk of host confusion is assessed by inspecting the photometry for other potential hosts nearby. The two non-SN types of HR transients, AGN and TDEs, are both highly centralised events within their hosts. AGN outburst sporadically, and one discriminant of these events is outburst detection in one or more previous seasons. AGN account for ∼3% of LBGS catalogue objects (Steidel et al., 2003). Obvious AGN activity is not considered for follow- up, but we maintain a catalogue for such objects. A HR SLSN detected in a known AGN can be identified as such if it is significantly offset (&0.500) from the flux centroid of the host. TDEs are highly centralised, UV-bright transient events observed less frequently than SLSNe (Donley et al., 2002). TDEs can be distinguished from supernovae by the power law decay of their light curves as distinct from the linear decay of supernova light curves (van Velzen et al., 2011). There are 25 candidates not ruled out as potential HR SLSNe, and these are given a confidence score based on several photometric factors. For superluminosity we consider a peak absolute magnitude limit of MFUV < −20 (Moriya et al., 2019), which corresponds to moptical < 24.5 at z ∼ 2 and all detections to the ATC limiting magnitude (m5σ ∼ 25.5) at z ∼ 3 and z ∼ 4. More events are detected near the ATC limiting magnitude where the different transient luminosity functions are more deeply sampled. For events where only the peak is observed, little is revealed by the behaviour of the light curve. In these cases a confidence of 1 (low) or 2 is assigned depending on the colour and morphology of the host and any observed offset. The confidence of brighter events is further assessed based on colour evolution. At z ∼ 2 HR SLSNe tend to exhibit faster evolution in the bluest optical bands, sampling blueward of the SLSN SED peak in the restframe far-UV (Pan et al., 2017; Smith et al., 2018), and at higher redshift this effect is even more pronounced (Cooke et al., 2012).

4.4 Keck Spectroscopic Follow-up

One of the goals of this thesis is to use LBG S&M on an ongoing survey to collect spectra of HR SLSNe near peak. To accomplish this we perform four spectroscopic follow-up campaigns using Keck LRIS with time awarded through the Swinburne Keck time access 78 CHAPTER 4. DES SPECTROSCOPY

Table 4.2 HR SLSN photometric candidates identiﬁed in C3 with ATC using LBG S&M

a b c DES-ID MJD Bin Conﬁdence MFUV mr Season 4 DES16C3cv 57615 z ∼ 3 1 -22.9 23.7 DES16C3px 57616 z ∼ 2 1 -20.3 Nond DES16C3bac 57629 z ∼ 2 1 -20.9 24.6 DES16C3ddo 57686 z ∼ 3 1 -21.2 25.5 DES16C3dee 57686 z ∼ 3 1 -21.2 Nond DES16C3ekq 57714 z ∼ 3 1 -20.9 24.2 DES16C3fhj 57739 z ∼ 3 3 -21.8 Poste DES16C3fuh 57739 z ∼ 2 1 -20.2 Poste DES16C3ggx 57746 z ∼ 2 2 -20.6 Poste DES16C3gzn 57778 z ∼ 2 2 -21.6 Poste DES16C3hab 57770 z ∼ 3 1 -21.2 Poste Season 5 DES16C3eco 57714 z ∼ 2 1 -21.8 24.2 DES17C3et 57986 z ∼ 4 2 -22.1 Nond DES17C3eu 57986 z ∼ 2 1 -21.6 25.5 DES17C3ew 57986 z ∼ 2 2 -21.3 Nond DES17C3px 57987 z ∼ 3 3 -21.1 24.6 DES17C3bsd 57991 z ∼ 3 2 -20.8 Nond DES17C3aae 57999 z ∼ 3 2 -21.4 Nond DES17C3avi 58019 z ∼ 2 2 -20.5 Nond DES17C3dsi 58036 z ∼ 2 2 -21.2 24.4 DES17C3bxx 58040 z ∼ 2 1 -20.0 Nond DES17C3czn 58052 z ∼ 4 2 -21.7 25.5 DES17C3egx 58064 z ∼ 3 1 -21.4 24.7 DES17C3fgb 58076 z ∼ 3 2 -20.9 24.5 DES17C3ivf 58122 z ∼ 3 1 -21.3 Poste a The Mean Julian Date (less 2400000) of initial detection. The spectroscopic follow-up campaign dates are 57749 (Season 4) and 58101 (Season 5). b The k-corrected absolute magnitude estimate using the median redshift of the bin. c Estimated apparent r-band magnitude projected to the date of spectroscopic follow-up. d Projected non-detection. e Detected post-follow-up. 4.4. KECK SPECTROSCOPIC FOLLOW-UP 79 program (see Table 3.5). HR SLSN photometric candidates are dim by the nature of their supposed distance. A source with MFUV = −21 at z ∼ 2 has a k-corrected apparent magnitude of mr ∼ 23.5. The best redshift indicator at z ∼ 2 is Ly-α which is observed at ∼3600Å, considered extremely blue for most ground-based observatories. Thus the combination of the large 10m primary mirror of Keck and the blue sensitivity of LRIS make for the most capable facility for spectroscopic follow-up of HR SLSNe at present. The spectroscopic follow-up campaigns are divided between DES and SHIZUCA photometric candidates. The latter are presented in Chapter 5.

Photometric candidates with mr < 25.5 at the pre-scheduled timing of an awarded spectroscopic follow-up campaign are targeted based on conﬁdence and brightness. The Keck LRIS spectroscopic observations are outlined in Table 4.3 and discussed in the following sections. Spectroscopic follow-up observations of SHIZUCA photometric candates are outlined in Table 5.1. The spectra are reduced with standard IRAF1 routines.

Table 4.3 Keck spectroscopic follow-up of DES LBG S&M photometric candidate HR SLSNe

a DES-ID z-bin Conﬁdence mr Exp Time Spec-z Result 2016 DEC DES16C3bac z ∼ 2 1 24.6 10800s 1.697 AGN DES16C3bn z ∼ 4 0b 23.4 2400s 0.601c LRT DES16C3cv z ∼ 3 1 23.2 10800s 0.727d SLSN-I 2017 DEC DES17C3eco z ∼ 2 1 24.2 3600s – Male DES17C3eu z ∼ 2 1 Nonf 7200s – Nonf DES17C3fgb z ∼ 3 2 25.5 7200s – Nonf a Total LRIS blue-arm exposure time. The red-arm exposure time is set slightly shorter to account for a longer readout time. b Considered a potential HR TDE from the light curve. c Preliminary measure. d Conﬁrmed externally with DES spectroscopic follow-up (Angus et al., 2018). e Irreducible spectrum due to internal LRIS malfunction. f Non-detection (photometric or spectroscopic as indicated).

4.4.1 DES16C3bac

DES16C3bac was detected by ATC in a host in the DECam LBGS z ∼ 2 bin on 29 August 2016, the beginning of the fourth season of DES and four months before our ﬁrst Keck

1IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. 80 CHAPTER 4. DES SPECTROSCOPY spectroscopic follow-up campaign. The event is highly centralised (see Figure 4.3) and the light curve does not exhibit a clear rise, peak or decline (see Figure 4.2), supporting an AGN rather than a SN interpretation. However, the host shows little previous activity (see Figure 4.1, top), and the apparent light curve profile is compromised by the fact that the event is detected near the ATC limiting magnitude. In addition, DES16C3bac is in relatively close proximity to a higher priority candidate, DES16C3cv (see Section 4.4.3), and both objects are able to be spectroscopically imaged on a single LRIS slitmask. We followed-up DES16C3bac with Keck LRIS on 27 December 2016, and the reduced 1-D spectrum is shown in Figure 4.4. We interpret the strong emission feature observed at ∼4200Å as Civ, suggesting DES16C3bac is an AGN at z = 1.697 0.002 (Figure 4.5). Emission features at wavelengths shortward of Civ where the flux uncertainty is high are less confident. The Mgii λ2804 feature is not clearly observed, but there is a strong telluric absorption line at this precise wavelength in the observer-frame that may be masking this feature. Alternatively, identifying the strong emission as Ly-α gives a redshift of z = 2.43. However, in this case an AGN is ruled out by the lack of any corresponding Civ emission, and a SLSN-II interpretation is similarly unlikely as the feature is broader than Ly-α observed in UV-bright SNe-IIn (Fransson et al., 2002, 2005, 2014).

4.4.2 DES16C3bn

DES16C3bn was detected in ATC on 15 August 2016 in a host meeting the DECam LBGS z ∼ 4 criteria. The event is much brighter than expected for a SLSN at this redshift, with an estimated peak absolute magnitude range of −23.4 > MFUV > −24. However, the main LRI contaminants in the z ∼ 4 bin are elliptical galaxies that are only expected to host SNe-Ia, and DES16C3bn does not resemble a SN-Ia. The greater than 500 day observed duration of the transient (see Figure 4.1, middle) and the decremented g-band luminosity (see Figure 4.6) are consistent with a z ∼ 4 interpretation as a result of time dilation and the Ly-α forest. The event is not discernibly offset from the flux centroid of its host galaxy (see Figure 4.7) and does not evolve significantly in colour as it fades, leading to the suspicion that DES16C3bn may be an extremely luminous TDE similar to ASASSN-15lh (Leloudas et al., 2016). We followed-up DES16C3bn with Keck LRIS on 27 December 2016, however we noted strong, steady continuum flux to the blue sensitivity limit of LRIS in the first 2-D spectrum exposure. At z ∼ 4 DES16C3bn must exhibit a Ly-α forest decrement and, more importantly, a strong flux break shortward of the Ly-limit similar to LBGs as these effects arise from intergalactic and interstellar absorption. Having ruled this out from the 2-D 4.4. KECK SPECTROSCOPIC FOLLOW-UP 81

Figure 4.2 DECam griz light curves of DES16C3bac using DES single epoch photometry. The date of Keck spectroscopic follow-up is indicated with a vertical dashed yellow line.

Figure 4.3 2500 × 2500 i-band stamps centred on the host of DES16C3bac. Shown are the deep stack image of the host in quiescence (left), a ∼3 month stack from the ﬁrst half of DES season 4 of the host and transient (middle), and the transient alone in the diﬀerenced image (right). The green circular regions are centred on the transient. 82 CHAPTER 4. DES SPECTROSCOPY

Figure 4.4 Flux-calibrated observer-frame 1-D spectrum of DES16C3bac, shown in gray. A smoothed spectrum is overlaid in black. The 1σ pixel−1 error is shown in red. The green dashed line marks the dichroic separation at 5600Å. The shape of the spectrum becomes less reliable shortward of 3600Å (1050Å restframe) due to systematics in ﬂat-ﬁelding. Longward of 7000Å (2040Å restframe), the spectrum becomes more sky-dominated and the smoothed spectrum has been manually clipped.

Figure 4.5 The spectrum of DES16C3bac smoothed and redshifted to z = 1.697. A quasi-stellar object (QSO, equivalent to AGN) composite spectrum (Vanden Berk et al., 2001) at the same redshift is overlaid in blue. The labeled vertical dashed red lines are transitions commonly observed in AGN. 4.4. KECK SPECTROSCOPIC FOLLOW-UP 83 spectrum, we cut short observations of the event to allow more time for targets more likely to be at high redshift. The reduced 1-D spectrum of DES16C3bn is shown in Figure 4.8. The S/N of the spectrum is low due to the shortened exposure time, but the emission feature observed at ∼6000Å is discernible in the 2-D spectrum. The z ∼ 4 estimate of DES16C3bn is made by attributing a colour break at about this wavelength in the SED of the host to the Ly-α forest. However, this eﬀect can be mimicked by a Balmer break in a low redshift elliptical (see Figure 3.3). Balmer breaks are often accompanied by the emission of [Oii] λλ3727, and associating this emission with the feature observed in the spectrum of DES16C3bn at ∼6000Å yields a redshift of z = 0.601. This redshift solution is one possible ﬁt, but when applied to the spectrum (Figure 4.9) it seems to recover several other common features. Regardless of the accuracy of this redshift solution, we strongly suspect DES16C3bn of being at low redshift. Still, given the long duration of the event and the suspected elliptical host type, DES16C3bn remains an interesting unexplained transient worth further investigation.

4.4.3 DES16C3cv

DES16C3cv was first observed by ATC on 15 August 2016 with no detectable host within 500 in the DES template images. However there is a host coincident with the event detected in our deep stacks which falls into the DECam LBGS z ∼ 3 bin (see Figure 4.11). A DES spectrum collected with VLT X-shooter on 25 September 2016 was used to derive a preliminary redshift of z = 0.727 from emission lines associated with [Oiii] in the undetected host, but from the deep stack photometry two objects are apparent in the immediate vicinity of DES16C3cv and it is unclear which is the host and which is producing the [Oiii] emission. In addition to this uncertainty, at z = 0.727 DES16C3cv constitutes a SLSN, making the strong decrement observed in the g-band and the slow evolution highly unusual (see Figure 4.10). At z ∼ 3 the decrement can be explained by a strong Ly-α forest effect and time dilation accounts for the long apparent timescale. We followed-up DES16C3cv with Keck LRIS on 27 December 2016 in an attempt to clarify the redshift of the event. Unfortunately suboptimal seeing and light cirrus during observations reduce the S/N and our ability to resolve details any finer than those detected in the X-shooter spectrum. The reduced 1-D spectrum is shown in Figure 4.12. The spectrum is dominated by the transient but must still exhibit a strong Ly-α forest effect and Ly-limit break if at z ∼ 3. We compare the spectrum to a HR LBG composite spectrum from (Shapley et al., 2003) which includes these same features, and find a reasonable 84 CHAPTER 4. DES SPECTROSCOPY

Figure 4.6 DECam griz light curves of DES16C3bn using DES single epoch photometry. The date of Keck spectroscopic follow-up is indicated with a vertical dashed yellow line.

Figure 4.7 2500 × 2500 i-band stamps centred on the host of DES16C3bn. Shown are the deep stack image of the host in quiescence (left), a ∼3 month stack from the ﬁrst half of DES season 4 of the host and transient (middle), and the transient alone in the diﬀerenced image (right). The green circular regions are centred on the transient. 4.4. KECK SPECTROSCOPIC FOLLOW-UP 85

Figure 4.8 Flux-calibrated observer-frame 1-D spectrum of DES16C3bn, shown in gray. A smoothed spectrum is overlaid in black. The 1σ pixel−1 error is shown in red. The green dashed line marks the dichroic separation at 5600Å.

Figure 4.9 The spectrum of DES16C3bn smoothed and redshifted to z = 0.601. The emission feature labeled as [Oii] is confident and discernible in the 2-D spectrum. Several other common features are seemingly aligned to this redshift solution, but these are less confident. 86 CHAPTER 4. DES SPECTROSCOPY continuum fit and possibly strong Ly-α absorption at a redshift of z = 3.747 (Figure 4.13), but this is not definitive. The spectrum is also fit reasonably well as a SLSN-I at z = 0.727 (Angus et al., 2018). While doubts about the redshift of DES16C3cv remain, we consider the z = 0.727 value most likely. Future spectra could reveal late-time circumstellar emission features as seen in some SLSNe (Fox et al., 2015; Yan et al., 2017b), enabling a more secure redshift measure.

4.4.4 Other Targets

Three additional DES ATC transients identified with LBG S&M as candidate HR SLSNe were followed-up in a second Keck campaign in December 2017 (see Table 4.3). Extinction from cirrus cloud cover significantly reduced the depth of observations. Two targets resulted in non-detections or either the transient or host. DES17C3eu had faded to non-detection in ATC by the time of the campaign, but the bright host was targeted in an effort to obtain redshift confirmation. DES17C3fgb began to rise less than 30 days prior to the campaign, but subsequent photometry revealed that the event was already fading rather than brightening at the time of spectroscopic follow-up. The spectrum of the third target, DES17C3eco, was irreducible due to an internal malfunction of LRIS.

4.5 Conclusion

We were unsuccessful in collecting spectra of HR SLSNe near peak from follow-up of DES photometric candidates identified with LBG S&M. But the results are still useful for assessing the confidence of the photometric candidates. Poor weather during both Keck follow-up campaigns reduced the number of candidates able to be targeted and limited targeting to the brightest candidate-host combinations rather than the most confident events. Two of three informative spectra are confirmed LRTs, but the LRT contamination rate among only the brightest targets is likely much higher than the overall rate. The LBG S&M candidate identification rate is high enough for classically scheduled follow-up, but the program would benefit from the timing flexibility and observing condition specificity provided with queue scheduling. However this would require the use of a facility other than Keck LRIS, resulting in a loss of light gathering power and blue sensitivity. The colour distributions of the hosts of the 25 LBG S&M identified HR SLSN photometric candidates are illustrated in Figure 4.14. The confidence of each candidate from 1 (low) to 3 (high) is indicated by the size of the representative marker. Also plotted are the colours of the hosts of identified AGN and low redshift SNe-Ia. The colour uncertainties 4.5. CONCLUSION 87

Figure 4.10 DECam griz light curves of DES16C3cv using DES single epoch photometry. The date of Keck spectroscopic follow-up is indicated with a vertical dashed yellow line.

Figure 4.11 2500 × 2500 i-band stamps centred on the host of DES16C3cv. Shown are the deep stack image of the host in quiescence (left), a ∼3 month stack from the ﬁrst half of DES season 4 of the host and transient (middle), and the transient alone in the diﬀerenced image (right). The green circular regions are centred on the transient. The host of DES16C3cv is not visible in the DES template image. 88 CHAPTER 4. DES SPECTROSCOPY

Figure 4.12 Flux-calibrated observer-frame 1-D spectrum of DES16C3cv, shown in gray. A smoothed spectrum is overlaid in black. The 1σ pixel−1 error is shown in red. The green dashed line marks the dichroic separation at 5600Å.

Figure 4.13 The spectrum of DES16C3cv smoothed and redshifted to z = 3.747. A Ly-α absorber LBG composite spectrum (Shapley et al., 2003) at the same redshift is overlaid in blue. Though it appears that the transient spectrum exhibits a strong Ly-α forest effect consistent with the high redshift, narrow [Oiii] emission features identified in an alternate DES spectrum place the host at z = 0.727 which also yields a fit to the transient as a SLSN-I. 4.5. CONCLUSION 89 are measured as the sum of the uncertainties in the component photometric magnitudes. z ∼ 2 and z ∼ 3 candidates are included in the (u0 − g) vs. (g − i) colour plane, and z ∼ 3 and z ∼ 4 candidates are included in the (g − r) vs. (r − i) colour plane. The figure illustrates how the colour of the host of a HR SLSN candidate is useful when assessing the confidence of the event. 90 CHAPTER 4. DES SPECTROSCOPY

Figure 4.14 Two important colour planes for assessing the confidence of HR SLSN candidates based on the colour of their hosts. The solid lines represent redshift tracks of star-forming galaxies with square markers placed at redshifts of z = 0, 1.7, 2.6, 3.5, as in Figure 3.5. The dashed line is the redshift track of an elliptical template. The blue, green and red shaded regions represent the spectroscopically refined colour selection criteria of z ∼ 2, 3, 4 LBGs respectively (see Tables 3.2, 3.3, 3.4). Upward facing triangles represent lower limits and are colour-coded to the objects identified in the legend. 5 FIRST RELEASE OF HIGH REDSHIFT SUPERLUMINOUS SUPERNOVAE FROM THE SUBARU HIGH-Z SUPERNOVA CAMPAIGN (SHIZUCA).II. SPECTROSCOPIC PROPERTIES.

5.1 ABSTRACT

We present Keck spectroscopic observations of three probable high redshift superluminous supernovae (SLSNe) from the Subaru HIgh-Z sUpernova CAmpaign (SHIZUCA), confirming redshifts of 1.851, 1.965 and 2.399. The host galaxies were selected for transient monitoring from multi-band photometric redshifts. The supernovae are detected during their rise, and the classically scheduled spectra are collected near maximum light. The restframe far-ultraviolet (FUV; ∼1000Å–2500Å) spectra include a significant host galaxy flux contribution and we compare our host galaxy subtracted spectra to UV-luminous SNe from the literature. While the signal-to-noise ratios of the spectra presented here are sufficient for redshift confirmation, supernova spectroscopic type confirmation remains inconclusive. The success of the first SHIZUCA Keck spectroscopic follow-up program demonstrates that campaigns such as SHIZUCA are capable of identifying high redshift SLSNe with sufficient accuracy, speed and depth for rapid, well-cadenced and informative follow-up.

91 92 CHAPTER 5. SHIZUCA SPECTROSCOPY

5.2 INTRODUCTION

It is now known that some supernovae exceed an absolute magnitude of M ' −21, giv- ing rise to a luminosity class of supernovae termed superluminous supernovae (SLSNe; Quimby et al., 2011; Gal-Yam, 2012; Nicholl et al., 2015; Moriya et al., 2018). SLSNe are also more luminous in the far-ultraviolet (FUV; ∼1000Å–2500Å) relative to other supernovae (Quimby et al., 2011; Cooke et al., 2012; Howell et al., 2013; Yan et al., 2017a, 2018). SLSNe are exceedingly rare, occurring at ∼ 0.001× the frequency of general core-collapse supernovae (Quimby et al., 2013; Prajs et al., 2017). But these transients promise to be powerful probes of the early universe as they are already visible in ground based observations to redshifts of z = 4 and greater (Cooke et al., 2012; Mould et al., 2017; Moriya et al., 2019). Recent improvements to survey astronomy, such as large-format CCD mosaics, are enabling more wide-area surveys to operate, such as the Hyper Suprime-Cam Subaru Strate- gic Program (HSC-SSP; Aihara et al., 2018), the Dark Energy Survey (DES; Dark Energy Survey Collaboration et al., 2016), the Zwicky Transient Facility (Bellm & Kulkarni, 2017) and Pan-STARRS (Kaiser et al., 2010). These surveys monitor the vast amounts of sky required to detect SLSNe with a reasonable frequency. In addition these surveys are broad enough to allow boutique surveys such as the Subaru HIgh-Z sUpernova CAmpaign (SHIZUCA; Moriya et al., 2019) and the Survey Using DECam for Superluminous Super- novae (SUDSS; Smith et al., 2016) to exercise more focused monitoring techniques and detect SLSNe with higher efficiency. There are advantages to surveying for SLSNe at high redshift (z & 2). Massive progenitors are invoked in all the explosion mechanisms used to explain SLSNe (Gal-Yam, 2012; Moriya et al., 2018), suggesting that the rate of SLSNe per unit volume varies with redshift as the cosmic star formation rate, rising and perhaps peaking at z ∼ 2 (Madau & Dickin- son, 2014). This is consistent with rates measured up to this redshift (Quimby et al., 2013; Prajs et al., 2017). Some if not all types of SLSNe show a preference for low-metallicity hosts (Lunnan et al., 2014; Leloudas et al., 2015; Angus et al., 2016; Schulze et al., 2018), suggesting that SLSNe may require low-metallicity progenitors and that the volumetric rate of SLSNe may continue to increase with redshift beyond the peak of star formation into epochs of universally low metallicity. The measured rate at z ∼ 2–4 is consistent with a continuing increase beyond z ∼ 2 (Cooke et al., 2012), though this measurement is based on two events. Another advantage of looking to high redshift is the ability to sample the restframe-FUV of SLSNe with optical facilities. FUV spectra of low redshift supernovae are rare because 5.3. OBSERVATIONS 93 they can only be collected from space-based telescopes. Studies of such spectra have been done (Fransson et al., 2002, 2005, 2014; Panagia, 2007; Bufano et al., 2009), including SLSNe (Yan et al., 2017a, 2018; Quimby et al., 2018). Due in part to their excessive UV luminosity and to model-predicted UV indicators of certain explosion mechanisms (Mazzali et al., 2016), there are programs focused on collecting FUV spectra of SLSNe in particular (Quimby, 2014, 2016). Still, given the low rate of SLSNe, few such spectra have been collected to date. At high redshift this critical UV information is pushed into the optical bands, enabling this wavelength region to be explored with ground-based observations. A number of SLSN FUV analyses of have been accomplished in this way (Berger et al., 2012; Howell et al., 2013; Pan et al., 2017; Smith et al., 2018). High éntendu surveys like SHIZUCA are already capable of identifying supernovae to z > 4 (Moriya et al., 2019; Mould et al., 2017). Spectroscopic follow-up of high redshift supernovae must be carried out on 8-meter class telescopes due to their intrinsically faint apparent magnitudes (often mr & 24 for z & 2). Here we present spectra obtained with Keck of three probable high redshift SLSNe. This paper summarizes the first spectroscopic follow-up program of SHIZUCA. It has been written in parallel with Moriya et al. (2019), the first photometric analysis of the SHIZUCA program, hereafter M19. We summarize the observations in Section 2. In Section 3 we present the spectra and their redshift measurements. In Section 4 we discuss our analysis of the observations and we present our conclusions in Section 5. All calculations in this −1 −1 paper assume a ΛCDM cosmology with H0 = 70km s Mpc , ΩM = 0.3 and ΩΛ = 0.7. All magnitudes are AB and all wavelengths are quoted in restframe Ångströms unless otherwise specified.

5.3 OBSERVATIONS

The COSMOS field (Scoville et al., 2007) of the HSC-SSP1 supplies the photometry for this work and consists of a 1.8 deg2 field-of-view Hyper Suprime-Cam pointing imaged in five filters (grizy). The first SHIZUCA season was active on the COSMOS field from 2016 November to 2017 May. SHIZUCA uses COSMOS2015 photometric redshifts (Laigle et al., 2016) and MIZUKI redshifts (Tanaka, 2015) to identify potential hosts of high redshift transients from HSC- SSP photometry. The per-epoch depth of observation from the HSC-SSP using the 8.2m

Subaru telescope (mi . 26.5) enables SHIZUCA to monitor the ﬂuxes of high redshift

1General information for the HSC-SSP, such as location, cadence, and data products can be found at: http://hsc.mtk.nao.ac.jp/ssp/ and the associated links. 94 CHAPTER 5. SHIZUCA SPECTROSCOPY sources in individual observing epochs. Any flux variations observed in these sources are then analyzed in the context of the photometric redshift of the host. Flux variations are identified as SLSN candidates using criteria such as non-recurrence, duration, light curve shape, peak magnitude and color evolution. Details of the SHIZUCA photometric analysis and relevant MIZUKI redshift probability distributions are presented and discussed in M19. Follow-up spectroscopy of select SHIZUCA photometric high redshift SLSN candidates was acquired on 2016 December 28 and 2017 March 22–23 using the Low Resolution Imag- ing Spectrometer (LRIS; Oke et al., 1995; Steidel et al., 2004) on the Keck-I telescope. These data were obtained using the 400/3400 grism on the blue arm and the 400/8500 grating on the red arm separated at ∼5600Å using the D560 dichroic. The CCDs were read out using 2×2 binning with a spectral resolution of ∼500km s−1. The full width at half-maximum (FWHM) seeing ranged from 000. 9–100. 1 and normal atmospheric extinction increased from light cirrus. During the March observations the blue shutter of LRIS failed and was fixed to remain open, and the trapdoor was used in its place. The exposures were set at 1200s on the blue side and 1179s on the red side with the difference chosen to allow the CCDs to finish reading out at the same time. For follow-up we targeted SHIZUCA photometric candidates near maximum light projected to be mr . 25.5 during observation. This magnitude limit aims to achieve combined transient and host continuum signal-to-noise ratios (S/N) near restframe 1500Å of S/N ∼ 3–5 pixel−1 for integrations of ∼2 hours. SLSNe usually arise in star-forming galaxies such as Lyman break galaxies (LBGs), and these S/N are sufficient to reliably measure redshifts via identification of strong UV absorption features from the interstellar medium along with Lyman-α (Ly-α) emission when present and the Ly-α forest when visible (Steidel et al., 1998). However the low S/N combined with systematics in the internal flats increase flux uncertainties at wavelengths shortward of 3600Å, observer-frame. On 2016 December 28, 3 SHIZUCA photometric candidates were eligible for follow-up of which 2 were targeted in the available time, HSC16adga and HSC16aaqc. On 2017 March 22–23 another 3 SHIZUCA photometric candidates made up the primary target list: HSC17auzg, HSC17dbpf and HSC17cywe. We followed-up each of these targets with sufficient time remaining to follow-up one backup target, HSC17davs, below our specified magnitude limit. A spectrum of such low S/N can still yield a measurable redshift and even supernova type information if it includes identifiable strong emission features. The SHIZUCA photometric candidates are transient phenomena detected photometrically in or near host galaxies with high photometric redshifts (see M19). There are cases of photometric redshift confusion in which separate, discrete photometric redshift probabili- 5.4. DATA REDUCTION AND ANALYSIS 95 ties persist for a single galaxy. For example, the decrement in the restframe UV continuum of a z ∼ 3 LBG caused by the Ly-α forest can be photometrically mimicked by the Balmer- break in the restframe optical continuum of a z ∼ 0.3 galaxy (Steidel et al., 2003; Cooke et al., 2006). Because the number of targets available for follow-up in a narrow observation window is limited, candidates observed in hosts exhibiting photometric redshift confusion with a reasonable probability of being high redshift are not disqualified from our target lists. However to minimize the loss of time observing low redshift transients, we inspect the raw 2-D spectra of each target as they are read out for any obvious emission features. Often one or two emission features are sufficient to differentiate between high and low redshift probabilities. If a target is determined to be low redshift during observation, the target is aborted and the next target is acquired. From the 6 SHIZUCA photometric candidates followed up, 3 arise in z > 1.8 hosts which can be reliably matched with a LBG reference spectrum (see Table 5.1 and Sec- tion 5.4). Two candidates, HSC16aaqc and HSC17cywe, were found to be low redshift through the positive association of emission features from the raw 2-D spectra to commonly observed transitions in low redshift galaxies (e.g., [Oii] λλ3727, H-β and [Oiii] λ5007). The spectrum of one target, HSC17davs and host, recorded no detectable signal or emissions. None of the spectra exhibited obvious signs of being other types of high redshift transients, namely active galactic nuclei (AGN) or tidal disruption events (TDEs). Among the spectra we use as templates for comparison are those of the SLSN-II, LSQ15abl (Quimby et al., 2015). The formal presentation and analysis of these spectra is part of a work in progress. They were collected with the Hubble Space Telescope (HST) using the Cosmic Origins Spectrograph (COS; Shull, 2009) and the Space Telescope Imaging Spectrograph (STIS; Woodgate et al., 1997). The reduced spectra are from the Mikulski Archive for Space Telescopes (MAST) and have been adjusted for scale, noise and redshift.

5.4 DATA REDUCTION AND ANALYSIS

5.4.1 HSC16adga

HSC16adga (SN-2016jhm) was first detected by SHIZUCA on MJD 57715 at coordinates (RA, Dec) = (10:02:20.12, +02:48:43.4). The transient is detected in a star-forming host +0.25 with a COSMOS photo-z of 2.26−0.30 and a MIZUKI photo-z peaking at 2.19. It is observed offset by 000. 36000. 104 (∼3kpc) from the host galaxy flux centroid, helping to rule out an AGN or a TDE. The light curve of HSC16adga is shown in Figure 5.1. 96 CHAPTER 5. SHIZUCA SPECTROSCOPY

Restframe Phase Restframe Phase 0 10 20 30 0 10 20 30 40 24.0 23.5 HSC16adga HSC17auzg 24.0 24.5

24.5 25.0

25.0 25.5 25.5 AB mag AB mag 26.0 1400Å(g) 1600Å(g) 1800Å(r) 26.0 2100Å(r) 2300Å(i) 2600Å(i) 26.5 2600Å(z) 26.5 3000Å(z) 2900Å(y) 3400Å(y) 27.0 27.0 720 740 760 780 800 820 840 760 780 800 820 840 860 MJD-2457000 MJD-2457000 Restframe Phase 5 0 5 10 15 20 23.0 HSC17dbpf 23.5

24.0

24.5

25.0

25.5 AB mag 26.0 1700Å(g) 2200Å(r) 26.5 2700Å(i) 3100Å(z) 27.0 3500Å(y) 27.5 810 820 830 840 850 860 870 MJD-2457000

Figure 5.1 The HSC-grizy light curves of HSC16adga (top), HSC17auzg (middle) and HSC17dbpf (bottom). Triangles indicate upper limits and errors are 1σ. The restframe timescales are relative to the dates of detection and the effective restframe wavelengths of the filters are given in the legends using z = 2.399, 1.965 and 1.851 respectively. Dates that spectra were acquired are identified by dashed black lines. K-corrected absolute magnitudes are discussed in M19. 5.4. DATA REDUCTION AND ANALYSIS 97

Figure 5.2 Maps of the spectral slits used for the observations of HSC16adga (top), HSC17auzg (middle) and HSC17dbpf (bottom). The slits are projected onto publicly available HST images of the hosts in quiescence. The open circles mark the 1σ regions of the transient locations as reported by SHIZUCA (see M19). 98 CHAPTER 5. SHIZUCA SPECTROSCOPY

Table 5.1 SHIZUCA spectroscopic follow-up targets

SN-ID Exp Time Photo-za Spec-z λ-Rangeb blue ; red (s) (Å) 2016 DEC +0.25 HSC16adga 7200;7074 2.26−0.30 2.399 942–2765 +1.20 c HSC16aaqc 2400;2358 1.25−0.03 ∼0.36 – 2017 MAR +0.06 HSC17auzg 8400;8253 1.65−0.08 1.965 1080–3170 +0.08 HSC17dbpf 6000;5895 2.25−0.53 1.851 1123–3297 HSC17cywe 2400;2358 0.45,3.26d ∼0.46c – +1.27 e HSC17davs 4800;4716 3.34−2.75 – – a COSMOS2015 unless otherwise speciﬁed b Restframe, derived from Spec-z cDerived using estimated wavelengths of emission features in raw 2-D spectra dTwo alternative MIZUKI photo-z peaks. COSMOS2015 photo-z not available eNo signal detected in the reduced imagery

We collected the Keck LRIS spectrum of HSC16adga on MJD 57751.09. The positioning of the spectral slit is mapped in Figure 5.2. From the light curve in Figure 5.1 the spectrum appears to have been taken near peak in the restframe-UV. The spectrum, composed of 6 exposures (7200s blue, 7074s red), was reduced using standard IRAF2 procedures. The combined blue and red-side 1-D spectrum is shown in Figure 5.3. The spectrum includes a significant host galaxy flux contribution from which we derive a spectroscopic redshift. The photometry of the host is suggestive of a high redshift starforming galaxy such as a LBG. The accepted method of spectroscopic redshift determination of high redshift LBGs is to infer an initial coarse redshift from visual confirmation of Ly-α, the Ly-α forest decrement and strong interstellar medium (ISM) absorption lines, and to then constrain the redshift by comparison to LBG composite spectra with full FUV line lists and consideration of the behavior of the lines as a result of outflows (i.e., the ubiquitous presence of ∼100–200km s−1 blueshifted ISM absorption line profiles and the P-Cygni-like profile of the Ly-α line; Steidel et al., 1996, 1998, 2003, 2004; Pettini et al., 2000, 2001; Shapley et al., 2003; Adelberger et al., 2003; Cooke et al., 2006; Jones et al., 2012). In the reduced spectrum of HSC16adga we identify a strong absorption feature at ∼4100Å, observer-frame, with a slight flux break in the continuum at shorter wavelengths,

2IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. 5.4. DATA REDUCTION AND ANALYSIS 99

HSC16adga 2 ) 1 Å 2 cm 1 (erg s 18 Flux × 10

0 4000 6000 8000 Observer Frame Wavelength (Å)

2 OI CII SiII SiII AlII CIII CIV FeII HeII Ly- Ly- /OVI SiIV/OIV] Relative Flux (F )

0 1200 1400 1600 Restframe Wavelength (Å)

Figure 5.3 Top: Flux-calibrated observer-frame 1-D spectrum of HSC16adga, shown in gray. A smoothed spectrum is overlaid in black. The 1σ pixel−1 error is shown in red. The green dashed line marks the dichroic separation at 5600Å. The shape of the spectrum becomes less reliable shortward of 3600Å (1060Å restframe) due to systematics in flat-fielding. Longward of 7000Å (2060Å restframe), the spectrum becomes more sky-dominated and the smoothed spectrum has been manually clipped. The yellow dashed box is the zoomed-in region shown in the lower plot. Bottom: The same spectrum zoomed-in, smoothed and redshifted to z = 2.399. A weak Ly-α absorber LBG composite spectrum (see text) is overlaid in blue and scaled arbitrarily to emphasize the alignment of features. A subset of the features used to constrain the redshift is illustrated with dashed blue lines and labeled. 100 CHAPTER 5. SHIZUCA SPECTROSCOPY which we attribute to Ly-α absorption and the Ly-α forest, respectively. For comparison throughout we use the LBG composite spectra of Shapley et al. (2003). From Figure 5.3 several features are apparent (e.g. Siii λ1260, Oi λ1302, Cii λ1335, Siiv λ1394, Oiv] λ1403), indicating a redshift of z = 2.399 0.004, consistent with both the COSMOS2015 photometric redshift estimate and MIZUKI redshift probability distribution. The error is given in terms of the spectral resolution of LRIS, doubled to account for a 2 pixel smoothing algorithm applied to clarify absorption features. At present spectroscopic cross-correlation is ineffective for the positive determination of precise redshifts of high redshift LBGs. One reason for this is because none of the identifiable FUV features of LBGs are at rest with respect to the redshift of the LBG system itself (Steidel et al., 2003). Still, cross-correlation can be used to quickly rule out the possibility of an alternative low redshift fit. We perform a simple cross-correlation of the spectrum with the Manual and Automatic Redshifting Software (MARZ; Hinton et al., 2016). MARZ cross-correlates input spectra with a library of galaxy templates and outputs the five most likely fits. Each fit includes a likelihood score from 1–4 representing respectively a likelihood of less than 50%, 50%– 90%, 90%–99% and greater than 99%. The MARZ top five redshift fits to the spectrum of HSC16adga range from 0.2845 < z < 2.42 with each fit receiving a score of 1, less than 50% likely. As an independent check for the visual redshift estimates, we perform a coarse fit on the observed spectra with a library of spectral templates to infer their redshifts. The strong night sky lines are not always subtracted perfectly in spectroscopic observations, and the sky residuals are the main source of systematic uncertainties in observed spectra, which are not captured by noise spectra. To suppress such systematics, we bin the spectra in 200Åwindows (although in practice the inferred probability distributions are not very sensitive to the choice of bin size) by first clipping the top and bottom 15% of the flux distribution in each bin and then taking the weighted average of the remaining data points. The clipping here is intended to largely eliminate the systematics. We then feed the binned fluxes to the photo-z code, MIZUKI, to infer narrow-band redshift probability distributions to compliment the photometric redshifts derived from broad-band photometry (see M19). Only galaxy templates are used in the fitting, and the supernova contribution to the spectrum is a source of uncertainty. The MIZUKI narrow-band redshift estimate of the spectrum of HSC16adga is shown in Figure 5.4. Compared to the broad-band estimate, the probability of the spectroscopic redshift is higher because several alternative high redshift probabilities have been eliminated 5.4. DATA REDUCTION AND ANALYSIS 101

Figure 5.4 MIZUKI narrow-band redshift probability distributions for each event. The MIZUKI broad-band photometric redshift probability distributions are presented in identical fashion in M19.

(see M19).

5.4.2 HSC17auzg

HSC17auzg (SN-2016jhn) was first detected by SHIZUCA on MJD 57745 at coordinates (RA, Dec) = (09:59:00.42, +02:14:20.8). The transient is detected in a star-forming host +0.06 with a COSMOS photo-z of 1.65−0.08 and a MIZUKI photo-z peaking at 1.78. It is observed offset by 000. 78000. 051 (∼6kpc) from the host galaxy flux centroid, helping to rule out an AGN or a TDE. The light curve of HSC17auzg is shown in Figure 5.1. We collected the Keck LRIS spectrum of HSC17auzg on MJD 57835.80. The posi- 102 CHAPTER 5. SHIZUCA SPECTROSCOPY tioning of the spectral slit is mapped in Figure 5.2. From the light curve in Figure 5.1 we estimate that the spectrum was taken some days after the restframe-UV peak. The spectrum, composed of 7 exposures (8400s blue, 8253s red), was reduced using standard IRAF procedures. The 1-D spectrum is shown in Figure 5.5. We identify a strong absorption feature at ∼3500Å, observer-frame, which we attribute to Ly-α. From this we make a first approximation of the redshift whereby several key ISM absorption features of LBGs become apparent (e.g. Oi λ1302, Siiv λ1394, Oiv] λ1403, Siii λ1527, Civ λλ1548, 1551, Alii λ1671), indicating a redshift of z = 1.965 0.004 (see Figure 5.5). This is significantly higher than the COSMOS2015 photometric redshift estimate but is represented in the MIZUKI redshift probability distribution. The MARZ five most likely cross-correlation redshift fits to the spectrum of HSC17auzg range from 0.428 < z < 3.95 with each fit receiving a score of 1, less than 50% likely. We produce a MIZUKI narrow-band redshift estimate for HSC17auzg (see Figure 5.4). Compared to the broad-band estimate (see M19) the probability of the spectroscopic redshift is similar but its proximity to a probability peak is significantly improved. There is a more dominant probability peak, however spectroscopic feature alignment occurs only at the redshift of the secondary probability peak.

5.4.3 HSC17dbpf

HSC17dbpf (SN-2017fei) was first detected by SHIZUCA on MJD 57816 at coordinates (RA, Dec) = (09:58:33.42, +01:59:29.7). The transient is detected in a star-forming host +0.08 with a COSMOS photo-z of 2.25−0.53 and a MIZUKI photo-z peaking at 1.58. It is observed offset by 000. 58000. 052 (∼5kpc) from the host galaxy flux centroid, helping to rule out an AGN or a TDE. The light curve of HSC17dbpf is shown in Figure 5.1. We collected two Keck LRIS spectra of HSC17dbpf on MJD 57835.93 and MJD 57836.84. The positioning of the spectral slit is mapped in Figure 5.2. The light curve in Figure 5.1 indicates that the event evolved very quickly, making the timing of the restframe-UV peak difficult to estimate. But it appears that the spectra were taken within a few days of this peak. The spectra (see Figure 5.6) were collected over 2 consecutive nights which we treat as a single epoch composed of 5 exposures (6000s blue, 5895s red). We attribute the strong absorption feature at ∼3400Å, observer-frame, to Ly-α. From this we make a first approximation of the redshift whereby several key ISM absorption features of LBGs become apparent (e.g. Cii λ1335, Siiv λ1394, Oiv] λ1403, Siii λ1527, Civ λλ1548, 1551), indicating a redshift of z = 1.851 0.004 (see Figure 5.6). This is consistent with the COSMOS2015 photometric redshift estimate but higher than predicted 5.4. DATA REDUCTION AND ANALYSIS 103

HSC17auzg

4 ) 1 Å 2 cm 1 (erg s

18 2 Flux × 10

0 4000 6000 8000 Observer Frame Wavelength (Å) OI CII SiII SiII AlII CIII CIV FeII HeII Ly- 4 SiIV/OIV] Relative Flux (F ) 2

1200 1400 1600 Restframe Wavelength (Å)

Figure 5.5 Top: Flux-calibrated observer-frame 1-D spectrum of HSC17auzg, similar to Figure 5.3. The shape of the spectrum becomes less reliable shortward of 3600Å (1215Å restframe), due to systematics in ﬂat-ﬁelding. Longward of 7000Å (2360Å restframe) the spectrum becomes more sky-dominated. Bottom: The same spectrum zoomed-in, smoothed and redshifted to z = 1.965, similar to Figure 5.3. A strong Ly-α absorber LBG composite spectrum (see text) is overlaid in blue and scaled arbitrarily to emphasize the alignment of features. A subset of the features used to constrain the redshift is illustrated. 104 CHAPTER 5. SHIZUCA SPECTROSCOPY

HSC17dbpf ) 1 Å 2 2 cm 1 (erg s 18 Flux × 10

0 4000 6000 8000 Observer Frame Wavelength (Å) OI CII SiII SiII AlII CIII CIV FeII HeII Ly- SiIV/OIV]

2 Relative Flux (F )

1200 1400 1600 Restframe Wavelength (Å)

Figure 5.6 Top: Flux-calibrated observer-frame 1-D spectrum of HSC17dbpf, similar to Figure 5.3. The shape of the spectrum becomes less reliable shortward of 3600Å (1260Å restframe) due to systematics in flat-fielding. Longward of 7000Å (2455Å restframe), the spectrum become more sky-dominated. Bottom: The same spectrum zoomed-in, smoothed and redshifted to z = 1.851, similar to Figure 5.3. A weak Ly-α absorber LBG composite spectrum (see text) is overlaid in blue and scaled arbitrarily to emphasize the alignment of features. A subset of the features used to constrain the redshift is illustrated. 5.5. DISCUSSION 105 by the MIZUKI redshift probability distribution. The MARZ five most likely cross-correlation redshift fits to the spectrum of HSC17dbpf range from 0.0155 < z < 4.67 with each fit receiving a score of 1, less than 50% likely. We produce a MIZUKI narrow-band redshift estimate for HSC17dbpf (see Figure 5.4). Compared to the broad-band estimate which is inconsistent with the spectroscopic redshift (see M19), the probable redshift range is similar but encompasses higher values including the spectroscopic redshift.

5.5 DISCUSSION

5.5.1 Nature

The high redshift supernova nature of each event is well-supported. The spectroscopic redshifts are precisely constrained and consistent with the MIZUKI narrow-band redshifts which offer no low redshift probability, and cross-correlation with MARZ produces no reliable low redshift alternative. Furthermore, when the measured redshifts are applied to the transient photometry, M19 find low redshift supernovae with comparable UV light curves in each case. Although TDEs can resemble supernovae in their light curve behavior, these are highly centralized events and our candidates are all significantly offset from their host galaxy flux centroids (see Figure 5.2). Thus it is unlikely that the candidates are of this type. In Figure 5.7 we compare the observed spectra to the FUV spectrum of the TDE, ASASSN- 14li (Cenko et al., 2016). We include a Sloan Digital Sky Survey (SDSS) quasi-stellar object (QSO, equivalent to AGN) composite spectrum (Vanden Berk et al., 2001) to which the TDE is similar. The lack of any strong Nv emission or any other nitrogen emission in the observed spectra is a notable difference to that expected of a TDE. The SLSN class of supernovae was first defined as a luminosity class that includes all supernovae exceeding an absolute magnitude of M = −21 in any band (Gal-Yam, 2012). However, currently SLSNe are identified and grouped spectroscopically and comprise event with a continuum of peak magnitudes which extends below the M = −21 threshold (Nicholl et al., 2017; Moriya et al., 2018; Quimby et al., 2018). The k-corrected peak absolute magnitudes of the presented SHIZUCA photometric candidates (see M19) approach but do not meet the strict photometric criteria of SLSNe set by Gal-Yam (2012). However the over-luminous aspect of each of the candidates in the FUV, while difficult to reconcile with ordinary supernovae, is representative of SLSNe (Quimby et al., 2011; Yan et al., 2017a, 2018). The intermediate FWHM seeing and light 106 CHAPTER 5. SHIZUCA SPECTROSCOPY

25 NV CIV SiIV HeII CIII] NIII] NIV] Ly-

SDSS QSO Composite

15 ASASSN-14li

HSC16adga

Relative Flux 10

5 HSC17auzg

HSC17dbpf 0 1500 2000 Restframe (Å)

Figure 5.7 All three supernova spectra including their host galaxy contributions. Also presented are a QSO composite and the TDE, ASASSN-14li. We have increased the noise in the spectrum of the TDE to approximate that of our observations, and have applied a Ly-α deforestation factor to the TDE and the QSO at wavelengths shortward of 1216Å consistent with a redshift of z ∼ 2. The labeled transitions are emission features observed in ASASSN-14li by Cenko et al. (2016), most of which are also commonly seen in AGN (as is evident from the QSO composite). 5.5. DISCUSSION 107 cirrus during spectroscopic follow-up resulted in the candidate spectra having lower S/N than anticipated and prevents positive supernova type-determinations. Thus as implied throughout we identify these transients as probable SLSNe. Late-time follow-up of the candidates could enable supernova type-determinations by providing more precise host galaxy subtraction templates and revealing any late-time circumstellar emission features as seen in some SLSNe (Fox et al., 2015; Yan et al., 2017b).

5.5.2 Host Subtraction

We compare our host-subtracted spectra to some of the few FUV spectra of early-time SLSNe and UV-bright SNe in an effort to discern the spectroscopic type of each supernova. While unsuccessful in this regard, the comparisons still offer useful information. Because we do not have spectra of the host galaxies alone, we perform a “pseudo-host” subtraction on each spectrum to enable our analysis. We assume a LBG-like host for each supernova, consistent with the templates fit with MIZUKI to the broad-band photometry (see M19) and with the hosts of SLSNe in general. Analyses of a large sample of LBGs (Shapley et al., 2003; Cooke et al., 2009) have shown that the characteristics and colors of a LBG spectrum are strongly correlated with the equivalent width of the Ly-α feature. Shapley et al. (2003) divided their LBG sample into quartiles based on this equivalent width and subsequently constructed composite spectra for each. The quartiles are referred to here as strong Ly-α absorbers, weak Ly-α absorbers, weak Ly-α emitters and strong Ly-α emitters. We include these composites in our pseudo-host constructions using the estimated Ly-α absorption or emission strengths of the hosts to select the specific composite for use in each subtraction. The wavelength ranges of the spectra include the HSC-gri bands. To construct the pseudo-hosts over this whole range we use the LBG composite spectra from 912Å–2000 Å (their full extent) and starburst (SB) templates from Calzetti et al. (1994) at wavelengths beyond 2000Å. This extension is valid because LBGs are highly star-forming galaxies. Ly-α absorbers have redder UV continua compared to Ly-α emitters (Shapley et al., 2003; Cooke et al., 2009). We therefore use the SB2 template (0.10 < E(B-V) < 0.21) to complete a pseudo-host constructed with a Ly-α absorber LBG composite, and we use the SB1 template (E(B-V) < 0.10) to complete a pseudo-host constructed with a Ly-α emitter LBG composite. To accurately scale the pseudo-hosts we calculate the ratio of host-flux to supernova- flux in each spectrum using the associated HSC photometry as measured most closely in time to our observations (see M19). The high redshift hosts are near-point sources and 108 CHAPTER 5. SHIZUCA SPECTROSCOPY the bulk of the flux enters the spectroscopic slit in each case. But each slit is centered on the supernova, thus some of the host flux may not be included (see Figure 5.2). When necessary we apply a geometrical correction to account for any portion of the host not within the spectroscopic slit.

5.5.3 SLSNe-II

Type-II supernovae (SNe-II) are defined as those that exhibit hydrogen lines in spectra, typically H-α (Filippenko, 1997), though they can also be confirmed with Ly-α emission in FUV spectra (Fransson et al., 2002, 2005, 2014). LBGs, which are highly star-forming galaxies, are a common host type for high redshift supernovae (Cooke et al., 2009, 2012). But they are also often strong Ly-α absorbers or emitters, which complicates the task of identifying supernova Ly-α emission. Observed Ly-α emission in a high redshift supernova/host spectrum can still be used as a positive identifier of a SNe-II if it can be confidently attributed to the supernova. In cases where the host and supernova are spatially separated on the sky such that their profiles are distinguishable, the source or sources of Ly-α emission can be observed directly. If this is not the case the source of observed Ly-α emission can still be inferred from the characteristics of the feature. Ly-α emission from a supernova originates in the ejecta or a circumstellar medium (CSM) or both. Ly-α emission from ejecta is broadened according to the ejecta velocity. Ly-α from a cold, slow-moving CSM is observed as a narrow feature with a blue velocity offset from the redshift of the host. In cases of LBGs with Ly-α in emission, the feature arises as Ly-α emission from the LBG is back-scattered off receding gas outflows and passes back through the LBG off-resonance avoiding absorption. Conse- quently the (narrow) feature is typically redshifted from the host by up to 1000km s−1 or more (Shapley et al., 2003), thus distinguishing it from any relatively blue-shifted supernova Ly-α emission. High redshift supernova Ly-α emission can also be largely or completely absorbed by neutral hydrogen (Hi) in the host interstellar medium, its circumgalactic medium, and the intervening intergalactic medium. So the absence of observable supernova Ly-α emission does not necessarily imply the absence of hydrogen integral to the explosion scenario. In cases where the lack of Ly-α emission is ambiguous, the presence of certain features unattributable to the host alone may be another reliable way to identify a SLSN-II from its FUV spectrum. For example, most SLSNe-II are spectroscopically classified as SNe- IIn based on the presence of narrow hydrogen lines in emission (Ofek et al., 2007; Smith et al., 2007, 2008a; Drake et al., 2010; Chatzopoulos et al., 2011). Studies by Fransson 5.5. DISCUSSION 109 et al. (2005, 2014) have shown that SN-IIn FUV spectra can have signature strong emission features such as Oi λ1300, Niv λ1486 and Mgii λ2800. These features could conceivably be used to confirm a SLSN-II classification (Cooke et al., 2009). Magnesium emission would be a particularly useful identifying feature of SLSNe-II as it has been observed to persist strongly to very late times in SNe-IIn (Fransson et al., 2005). In Figure 5.8 we compare our pseudo-host-subtracted spectra to the early-time FUV spectrum of the low redshift SLSN-II, LSQ15abl (Quimby et al., 2015), and the earliest FUV spectra of the UV-luminous SNe-IIn, SN1998S (Fransson et al., 2005) and SN2010jl (Fransson et al., 2014). None of our supernovae exhibit obvious Ly-α emission. It is tempting to attribute the blunted emission-like feature at ∼1190Å in the spectrum of HSC17dbpf to a component of slightly blue-shifted Ly-α emission, but this feature arises at a wavelength at which the shape of the spectrum is unreliable due to systematics in flat- fielding. It is worth noting that in the noise-added spectra being used for comparison, Ly-α emission is likewise not obvious. We know that these events do exhibit Ly-α emission, so this comparison illustrates the baseline S/N required of a high redshift supernova spectrum in order to rule out at least any unmolested Ly-α emission. The coarse blackbody shapes of our spectra are bluer than the comparison spectra, but this is not surprising given their earlier collection times. Beyond these observations, the spectra are simply of too low S/N to say anything about the more subtle emission features anticipated in the FUV of SLSNe-II.

5.5.4 SLSNe-I

The spectra of SLSNe-I are uniquely conspicuous, displaying a number of characteristic broad hump-like features throughout the optical and UV (Quimby et al., 2011; Howell et al., 2013; Nicholl et al., 2015; Yan et al., 2017a, 2018). The number of spectroscopically observed SLSNe-I is increasing rapidly and the nature of these features is being clariﬁed (Mazzali et al., 2016; Quimby et al., 2018; Lunnan et al., 2018). Our understanding of the UV features of SLSNe-I is also improving thanks in large part to spectroscopic studies of high redshift examples (Berger et al., 2012; Howell et al., 2013; Pan et al., 2017; Smith et al., 2018). From what has been observed to date, SLSNe-I typically exhibit broad absorption troughs centered around ∼1930Å, 2200Å, 2400Å and 2660Å (Smith et al., 2018). Spectral synthesis models have attributed these features to blended transitions of iron, cobalt, carbon, titanium, silicon and magnesium (Mazzali et al., 2016). Prominent features at still shorter wavelengths (e.g., 1400Å, 1700Å) have been observed in spectra of high redshift SLSNe-I for which UV coverage extends suﬃciently far 110 CHAPTER 5. SHIZUCA SPECTROSCOPY CIV NIII] [CII] MgII NIV] [NII] OI/CII

30 [NeIV] Ly- /NV HeII/OIII] SiIV/OIV] SiIII]/CIII]

HSC16adga 25 -2d@2600Å

HSC17dbpf +0d@3100Å 20 HSC17auzg +6d@3000Å

15 LSQ15abl Relative Flux +6d@4000Å

10 SN1998S +14d@4000Å

SN2010jl +44d@4000Å 0 1500 2000 2500 3000 Restframe (Å)

Figure 5.8 The three pseudo-host-subtracted supernova spectra (see text) along with the SLSN-II, LSQ15abl, and 2 UV-bright SNe-IIn. The spectra are labeled along with their approximate phase in days since peak at the associated restframe wavelength. We have increased the noise in the comparison spectra to approximate that of our observations, and have applied a Ly-α forest eﬀect for a redshift of z ∼ 2 at wavelengths shortward of 1216Å. The labeled transitions are emission features identiﬁed in SNe-IIn by Fransson et al. (2005, 2014). 5.6. CONCLUSION 111

(Pan et al., 2017), as well as in low redshift SLSNe-I with HST FUV spectral coverage (Yan et al., 2017a, 2018). But the consistency of these features among such events is insufficient to draw conclusions about their normalcy at this time. In Figure 5.9 we compare our pseudo-host-subtracted spectra to an extensive FUV spectrum of the high redshift SLSN-I, DES15E2mlf (Pan et al., 2017), and the two low redshift SLSNe-I with FUV spectral coverage, Gaia16apd (Yan et al., 2017a) and SN2017egm (Yan et al., 2018). In the noise-added comparison spectra the absorption troughs are significantly dampened, but most are still discernible. Differences in the strengths, widths and central wavelengths of individual troughs are also apparent. The longer wavelength features are not observed to any significance in our high redshift sample, though at these wavelengths our spectra are all sky-dominated. There seem to be some similarities between the comparison spectra and the spectra of HSC17auzg and HSC17dbpf at short wavelengths (e.g. the absorption dip at ∼ 1400Å, the general trends of the spectra from ∼ 1700 − 2000Å). Positive SLSN-I classifications can be made with optical spectra by performing a χ2 mini- mization of model spectra created by scaling, adding, and then redshifting supernova and host galaxy templates (Howell et al., 2005; Quimby et al., 2018). We attempt to quantify the significance of any similarities between our spectra and the FUV spectra of SLSNe-I by the same method. However with so few FUV spectra of SLSNe-I available for comparison the significance of the presence or lack of features in the FUV spectra of our targets cannot be quantified and our results are inconclusive.

5.6 CONCLUSION

With this SHIZUCA pilot spectroscopic program we have demonstrated the feasibility of spectroscopic follow-up of high redshift SLSNe, testing the current limits of high redshift transient spectral analysis. The high éntendu and dense cadence of the HSC-SSP combined with the detection efficiency of SHIZUCA can generate high redshift SLSN candidates at a sufficient rate to statistically enable follow-up of several targets near maximum light at any time during an active observation campaign. Keck LRIS provides the best facility for spectroscopic follow-up of z ∼ 2 supernovae in terms of light-gathering power combined with blue sensitivity. We have reported on FUV spectroscopic follow-up of three probable SLSNe at z = 1.851, 1.965 and 2.399. The spectra are extracted from ∼2 hour exposures taken under suboptimal seeing and weather conditions. The S/N of the reduced spectra are relatively low, and flux measurements become unreliable at wavelengths shortward of 3600Å, observer- frame. Thus in LRIS spectra of similar S/N, moderate-to-weak Ly-α features, perhaps 112 CHAPTER 5. SHIZUCA SPECTROSCOPY SiII,CII CIII CIII,CII TiIII CII,TiIII SiIII

30 CIV,SiII TiIII MgII,CII FeIII,CoIII

DES15E2mlf 25 -4d@2710Å

HSC16adga 20 -2d@2600Å

Gaia16apd +0d@3500Å 15 Relative Flux

HSC17dbpf 10 +0d@3100Å

HSC17auzg +6d@3000Å 5

SN2017egm +7d@2000Å 0 1500 2000 2500 3000 Restframe (Å)

Figure 5.9 The three pseudo-host-subtracted supernova spectra (see text) along with the high redshift SLSN-I, DES15E2mlf, and the two low redshift SLSNe-I. The spectra are labeled along with their approximate phase in days since peak at the associated restframe wavelength. We have increased the noise of the comparison spectra to approximate that of our observations, and have applied a Ly-α forest effect for a redshift of z ∼ 2 at wavelengths shortward of 1216Å. Transitions reproduced in the spectral models of Mazzali et al. (2016) are marked with black dotted lines, while those marked with red dotted lines are derived by Yan et al. (2018) with a model fitting method. 5.6. CONCLUSION 113 dampened by Hi absorption, cannot be confirmed or ruled out at z . 2. Such analyses can be extended to shorter wavelengths and lower redshifts using spectra with higher S/N or exhibiting strong Ly-α emission. Our analysis of the pseudo-host-subtracted supernova spectra is inconclusive regarding the spectroscopic type of each event. We estimate that doubling the S/N of the spectra presented here would enable supernova type-confirmations of future events by substantiat- ing any strong emission features of SLSNe-II or the broad absorption troughs of SLSNe-I. By observing under optimal seeing conditions, employing longer exposures, and perhaps taking advantage of brighter than average targets, this S/N is achievable. Future follow-up is merited to ascertain if any of the supernovae exhibit late-time strong emission (e.g., Ly-α, Mgii) from which type-information can be inferred. In addition, late- time spectra of the hosts with little-to-no supernova contributions can be used to produce more precise host-subtracted supernova spectra and better constrain their UV continuum slopes, enabling further analysis. An approximate rate calculation of z ∼ 2 SLSNe based on the SHIZUCA photometric candidate catalogue is discussed in M19. The accuracy of this rate is directly dependent on the efficiency of SHIZUCA in selecting high redshift transients and reducing photometric redshift confusion. More spectra are needed to establish and improve this efficiency and separate high redshift supernovae from other high redshift transient phenomena such as AGN and TDEs. As a first approximation, 3 of 5 SHIZUCA photometric candidates with spectroscopically confirmed redshifts (excluding the non-detected HSC17davs) are at high redshift, which gives a current redshift efficiency of 60%. SHIZUCA is capable of detecting supernovae to z > 6, and with the right observing strategy and clear skies, spectroscopic follow-up of these events out to the edge of the epoch of reionization may be achievable. And the potential of the James Webb Space Telescope will be greater still, capable of acquiring spectra of SLSNe as far as z = 20. At such high redshifts only deep, wide-field infrared surveys will be capable of producing targets. Such surveys are already being considered in future facilities like the University of Tokyo Atacama Observatory (TAO) and the Kunlun Dark Universe Survey Telescope (KDUST). By exploring the current limits of high redshift transient astronomy, SHIZUCA and similar programs are setting the stage for observing the explosions of the very first stars.

6 Summary and Future Work

6.1 Summary

In this thesis we use Lyman Break Galaxy Selection & Montoring (LBG S&M; Cooke, 2008) to efficiently identify high redshift (HR; z & 2) superluminous supernova (SLSN) candidates for spectroscopic follow-up near peak. SLSNe are a recently discovered stellar phenomenon (Gal-Yam, 2012) which theoretically will be observable with near-future technology to the very beginning of star formation in the Universe. Currently SLSNe are the most distantly observed type of supernova at z ∼ 2–6. However many such events go unidentified as such because they are indistinguishable from low redshift ordinary supernovae using photometry and difference imaging, and are not bright enough to be prioritized for spectroscopic follow-up, even with 8m class telescopes. LBG S&M is designed to enable more efficient identification of HR SLSNe and make the population more available for photometric and spectroscopic analysis. In addition the method provides fainter host galaxy detection and colour information. But it can only be effectively performed with data from a deep-and-wide multi-filter survey cadenced over several years. Prior to this work LBG S&M had only been performed on CFHTLS archival data (Cooke et al., 2009, 2012), and photometric candidates were identified very late in their evolution and after the survey had ended. By employing LBG S&M during ongoing surveys, we are able to identify HR SLSN candidates early in their evolution and perform spectroscopic follow-up on the brighter events during their more dynamic phases. Performing LBG S&M at z ∼ 2–4 requires deep ugriz photometry, and deep stacks are constructed from existing DES and other DECam data to satisfy this requirement. Because DES does not utilize the DECam u0-band, we initiated the u0SUDSS program to collect deep u0-band data on select fields and facilitate the selection of z ∼ 2 and z ∼ 3 LBGs. The colours of sources detected in the deep photometry are plotted in various colour-colour

115 116 CHAPTER 6. CONCLUSION planes, and redshift colour-tracks of star-forming galaxy model spectral templates are used to trace out the areas of interest. For z > 1.7 the ﬂux breaks for which LBGs are named are redshifted into the observer-frame optical, distinguishing LBGs by colour from the bulk of low redshift objects. The selected LBGs are monitored exclusively in ongoing, cadenced survey data for transient activity. Detected transients are evaluated through photometric observables on their likelihood of being HR SLSNe, generating a catalogue of photometric candidates. The phase and apparent magnitude of high-conﬁdence candidates at the timing of spectroscopic follow-up is used to establish the targeting priority.

In Chapter 2 we outline our procedure for constructing deep stacks from large sets of DECam images (see Table 1.1), focusing on the four u0SUDSS ﬁelds. The images are astrometrically aligned, ﬂux scaled and quality weighted to maximise stack depths. Photometric calibrations are performed using publicly available DECam u0griz zeropoints, and the photometry is tested internally. During testing we identify a systematic ∼0.3 mag error in the DECam provided u0-band zeropoint, lowering our time-based estimates on the depths of the u0 stacks.

0 0 2 We achieve full depth (u5σ ∼ 26.2) on 3 u SUDSS fields (9 deg ) and medium depth on a fourth. The 5σ limiting magnitudes of the DES deep field griz stacks (6 deg2) are ∼27.3, 27.4, 27.2 and 26.9 mag respectively, rivalling the SNLS deep stacks (4 deg2; Astier et al., 2006) and even surpassing them in the z-band. Our deep stacks are deeper than any product yet made available by DES, and have been used on multiple occasions to identify DES transient hosts undetectable in the difference imaging templates. In addition, the u0SUDSS program has provided most of the u0-band photometry being used by DES to extend their deep field wavelength coverage.

In Chapter 3 we calibrate the DECam for LBG selection (Steidel et al., 2003) and perform selection of z ∼ 2, z ∼ 3 and z ∼ 4 LBGs on DECam deep stack photometry. We assign initial colour selection criteria based on synthetic colours of model star-forming galaxy spectral templates measured through the redshift range of each bin, and real spectra of z ∼ 3 LBG composites from (Shapley et al., 2003). We consider the colours of the star- forming templates at low redshift and the colours of model elliptical galaxies to constrain the criteria for increased selection efficiency. We collect spectra of sources selected with the initial criteria during Keck follow-up of HR SLSN candidates using the multi-object capability of LRIS. We spectroscopically confirm the redshifts of 72 HR LBGs and 6 LRIs and use their colours measured in the DECam deep stacks to refine the LBG selection criteria. Several hundred additional ZFOURGE (Straatman et al., 2016) spectroscopic redshifts of sources with DECam C3 deep stack colours are also considered. 6.1. SUMMARY 117

We crossmatch the refined DECam LBG selection catalogues with the ZFOURGE photometric redshift catalogue in DES C3 to estimate the efficiency of DECam LBGS. From this procedure we estimate accuracies of ∼80% and HR object completeness levels of ∼50% in each redshift bin. The LBG selection size of the combined redshift catalogues includes ∼150,000 sources in DES C3 alone (3 deg2), a sufficient quantity for effectively performing LBG S&M. Having calibrated the DECam for LBG selection the procedure can now be performed on any archival or future DECam photometry of sufficient depth with the provided criteria, and our existing HR LBG catalogues are a useful resource for a variety of applications.

In Chapter 4 we collect Keck LRIS spectra of DES HR SLSN photometric candidates identified with LBG S&M. We identify the candidates by monitoring our HR LBG selections for brightening using the DES difference imaging transient catalogue. The candidates are drawn from the C3 field which has deep u0-band coverage, is observed frequently, and is observable from Keck. We use photometric observables to evaluate the confidence that a candidate is a HR SLSN, filtering out unwanted transients such as AGN and low redshift SNe-Ia.

In a 3 deg2 search area over two DES observing seasons we identify 25 HR SLSN photometric candidates to our spectroscopic follow-up limiting magnitude (mr ∼ 25.5). We performed two spectroscopic follow-up campaigns with Keck LRIS, targeting six candidates. Three targets were not recovered in the reduced spectra, two targets are suspected low redshift supernovae, and one is a HR AGN. Unusually poor weather conditions at Keck over the two years of the program were a significant factor in our lack of success in collecting spectra of HR SLSNe, limiting targeting to fewer and brighter events than anticipated and lowering the S/N of the reduced spectra. Still, we demonstrate the effectiveness of LBG S&M at identifying HR SLSN photometric candidates for Keck classically scheduled spectroscopic follow-up targeting and that effective follow-up is achievable under normal observing conditions.

In Chapter 5 we present Keck spectroscopic follow-up of SHIZUCA HR SLSNe (Curtin et al., 2019). The targets are identified by SHIZUCA using a photometric redshift method similar to LBG S&M but not limited to LBGs specifically. We also provide complimen- tary z ∼ 2 and z ∼ 3 LBG catalogues on the field for reference by SHIZUCA. Spectra of five candidates are successfully collected and reduced. Two of these are misidentified low redshift supernovae, and three are determined to be HR SLSNe. We compare the spectra to the few available far-UV spectra of SLSNe-I and SLSNe-II to identify any similarities. However the S/N of the spectra are low due to extinction from light cirrus during observa- 118 CHAPTER 6. CONCLUSION tions, and while the redshifts of the events are confident their spectroscopic subtypes are indeterminate. The result is significant. To date only two supernovae at z > 1.8 have been spectroscopically observed near peak and subtyped, the SLSN-I DES15E2mlf at z = 1.8607 (Pan et al., 2017) and the SLSN-I DES16C2nm at z = 1.9982 (Smith et al., 2018). These were drawn from the 30 deg2 DES supernova survey area in the third and fourth photometric seasons respectively. The three SHIZUCA events at z > 1.8 are identified in a single season of HSC-SSP photometry in a 1.8 deg2 survey area, and each event peaks above the DES single-epoch limiting magnitude of mr ∼ 25.5 (see Figure 5.1). The large implied difference in the discovery rates between the surveys is due to the use of photometric redshifts by SHIZUCA to estimate the redshift of each detected transient prior to spectroscopic follow-up.

6.2 Future Work

Improvements to the LBG selection catalogues on u0SUDSS fields can be made with supplementary DECam observations. The LBG catalogues on the DES C3 field are the largest produced with ∼150,000 sources in total. The DES X3 deep field has the potential for similarly sized LBG catalogues, but requires a small investment of 6–12 dark time hours of 0 0 0 observation to reach full depth in the u -band (u5σ ∼ 26.2). Without this deep u -band photometry the DES X3 LBG catalogues consist of only ∼40,000 sources. The COSMOS and NSF2 fields include full depth u0-band photometry, but much shallower griz photometry. Matching the depth of the griz photometry on the DES C3 and X3 fields is impractical, considering that this depth was achieved with regular observations over 3 years. However, matching the depths in each filter to the depth of the u0-band photometry would enable z ∼ 2 LBG selection on these fields similar to that on DES C3 (∼70,000 sources) with only ∼24 hours of additional observations per field. Only ∼12 hours are required if the z-band is neglected, which does not significantly increase the efficiency of LBG selection at z ∼ 2. The increased LBG catalogue sizes would provide more source for immediate monitoring at the commencement of any future surveys on u0SUDSS fields. There is a substantial population of photometric HR SLSN candidates to be drawn from the now archival DES and SUDSS data. In addition to ATC transients coincident with LBG catalogue sources, candidates beyond the limiting magnitude of ATC can be detected by monitoring LBGs in dynamic stacks such as seasonal stacks to increase the detection sensitivity. Spectroscopic follow-up of candidates near peak is unavailable, but late-time follow-up that reveals identifiable emission features can be used to confirm HR 6.2. FUTURE WORK 119 supernova classification for several years in the observer-frame (Fransson et al., 2014; Yan et al., 2017b). Spectra of hosts alone can also be used to confirm redshifts and quantify the efficiency of LBG S&M. This efficiency combined with other observable behaviours of a population of photometric candidates can be used to constrain the rates of HR SLSNe and could potentially enable SLSN type differentiation from far-UV photometry alone.

There are variations on LBG S&M that can be tested with DES archival photometry. The LBG catalogues on DES fields developed for this work are compiled using colour information from deep stacks constructed with the first three seasons of DES survey data. LBGs hosting SLSNe in one of these seasons may not meet the criteria for selection because of a temporary change in colour due to the presence of the transient, though this effect is usually small. This can be avoided by excluding the photometry of the season being monitored from the deep stack used for LBG selection. Aside from LBGs, dusty star- forming galaxies (DSFGs) are also likely to host HR SLSNe. The density of DSFGs on sky may match or even exceed LBGs (Spitler et al., 2014) but they are practically invisible in optical photometry. HR SLSNe occurring in DSFGs that are not significantly obscured by their dusty hosts are detected in the optical as hostless transients. HR SLSNe occurring in faint LBGs, especially at z ∼ 4, are also detected as hostless. There are also scenarios that can lead to hostless transients at low redshift. By monitoring deep stack photometry for hostless transients it may be possible to develop a method of distinguishing their redshift as high or low from observables such as their colour and colour evolution (Tanaka et al., 2012).

The upcoming LSST represents an excellent opportunity for LBG S&M, and the necessary preparations are outlined by this work. The technique can be effectively applied on the proposed deep drilling fields in a similar fashion to how it was applied on u0SUDSS fields. Furthermore, three of these fields overlap with u0SUDSS fields, and large LBG catalogues will be available for monitoring by LSST from first light. LSST LBG selection will require calibration with spectroscopic redshift measurements of initially selected sources. The initial selection can be made with model-based criteria as in this work. However, using the DECam LBG catalogues as an initial selection allows spectroscopic redshift measurements to begin sooner. LSST will match our depths on these fields in ∼1 season of observations, and over a wider area (nearly 40 deg2). The bluer LSST u-band will enable more efficient and more numerous selection of z ∼ 2 LBGs, and deep Y -band photometry can be used to select and monitor LBGs at z ∼ 6, the end of reionization.

JWST will be capable of spectroscopic follow-up of SLSNe to z ∼ 20, the era of the first stars, and LBG S&M applied in sufficiently deep-and-wide IR surveys can provide these 120 CHAPTER 6. CONCLUSION targets. SLSNe with −20 > MFUV > −22.5 have a k-corrected apparent magnitude range of 24.75 < mIR < 27.25 at z ∼ 6 and 26 < mIR < 28.5 at z ∼ 20. If the volumetric rate of SLSNe mirrors the cosmic star formation rate, events at these redshifts may be rarer than at z ∼ 2–4 (see Figure 1.6), requiring a minimum survey area of several square degrees. VIDEO is an ongoing deep-and-wide IR survey over 12 deg2 with the VISTA 4m telescope and VIRCAM 2 deg2 FOV imager (Jarvis et al., 2013). The per-epoch depth (23.5 < mIR < 24.5) is insufficient to detect SLSNe at z & 6, but performing detection on seasonal stacks (24.25 < mIR < 25.25) probes the peak of the luminosity distribution of SLSNe at z ∼ 6–10. However the estimated depths of multi-season deep stacks are insufficient for deeply sampling the z ∼ 6 LBG luminosity function and not enough LBGs can be selected by colour for monitoring to be effective. Deeper photometry can be compiled in less time using a larger IR-optimised telescope such as the recently commissioned 6.5m TAO (Yoshii et al., 2014), but this currently lacks a wide FOV imager which again leads to small numbers of sources for monitoring. KDUST, the proposed IR complement to LSST based in Antarctica (Zhao et al., 2011) could provide the necessary combination of depth and width to enable effective LBG S&M of z ∼ 6–20 SLSNe, but construction has yet to begin. Other options should be explored soon to ensure that we do not miss the opportunity with JWST to witness the most distant observable events in the Universe, the explosions of the very first stars. Bibliography

Adelberger, K. L., Steidel, C. C., Shapley, A. E., et al. 2004, ApJ, 607, 226

Adelberger, K. L., Steidel, C. C., Shapley, A. E., & Pettini, M. 2003, ApJ, 584, 45

Aihara, H., Armstrong, R., Bickerton, S., et al. 2018, Publications of the Astronomical Society of Japan, 70, S8

Angus, C. R., Levan, A. J., Perley, D. A., et al. 2016, MNRAS, 458, 84

Angus, C. R., Smith, M., Sullivan, M., et al. 2018, arXiv e-prints, arXiv:1812.04071

Astier, P., Guy, J., Regnault, N., et al. 2006, A&A, 447, 31

Bañados, E., Venemans, B. P., Decarli, R., et al. 2016, The Astrophysical Journal Supple- ment Series, 227, 11

Banerjee, S., Kroupa, P., & Oh, S. 2012, ApJ, 746, 15

Barbary, K., Dawson, K. S., Tokita, K., et al. 2009, ApJ, 690, 1358

Barbon, R., Cappellaro, E., & Turatto, M. 1984, A&A, 135, 27

Barkat, Z., Rakavy, G., & Sack, N. 1967, Phys. Rev. Lett., 18, 379

Beckwith, S. V. W., Stiavelli, M., Koekemoer, A. M., et al. 2006, AJ, 132, 1729

Bellm, E., & Kulkarni, S. 2017, Nature Astronomy, 1, 0071

Benetti, S., Nicholl, M., Cappellaro, E., et al. 2014, MNRAS, 441, 289

Berger, E., Chornock, R., Lunnan, R., et al. 2012, ApJ, 755, L29

Bertin, E. 2006, in Astronomical Society of the Paciﬁc Conference Series, Vol. 351, Astro- nomical Data Analysis Software and Systems XV, ed. C. Gabriel, C. Arviset, D. Ponz, & S. Enrique, 112

Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000

Bufano, F., Immler, S., Turatto, M., et al. 2009, ApJ, 700, 1456

Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1994, ApJ, 429, 582

Casertano, S., de Mello, D., Dickinson, M., et al. 2000, AJ, 120, 2747

Cenko, S. B., Cucchiara, A., Roth, N., et al. 2016, ApJ, 818, L32

121 122 BIBLIOGRAPHY

Chatzopoulos, E., & Wheeler, J. C. 2012, ApJ, 748, 42

Chatzopoulos, E., Wheeler, J. C., Vinko, J., Horvath, Z. L., & Nagy, A. 2013, ApJ, 773, 76

Chatzopoulos, E., Wheeler, J. C., Vinko, J., et al. 2011, ApJ, 729, 143

Chen, K.-J., Woosley, S., Heger, A., Almgren, A., & Whalen, D. J. 2014, ApJ, 792, 28

Chevalier, R. A., & Irwin, C. M. 2011, ApJ, 729, L6

Coleman, G. D., Wu, C.-C., & Weedman, D. W. 1980, ApJS, 43, 393

Cooke, J. 2008, ApJ, 677, 137

Cooke, J., Omori, Y., & Ryan-Weber, E. V. 2013, MNRAS, 433, 2122

Cooke, J., Ryan-Weber, E. V., Garel, T., & Díaz, C. G. 2014, MNRAS, 441, 837

Cooke, J., Sullivan, M., Barton, E. J., et al. 2009, Nature, 460, 237

Cooke, J., Wolfe, A. M., Gawiser, E., & Prochaska, J. X. 2006, ApJ, 652, 994

Cooke, J., Sullivan, M., Gal-Yam, A., et al. 2012, Nature, 491, 228

Curtin, C., Cooke, J., Moriya, T. J., et al. 2019, ApJS, 241, 17

Dark Energy Survey Collaboration, Abbott, T., Abdalla, F. B., et al. 2016, MNRAS, 460, 1270 de Souza, R. S., Ishida, E. E. O., Johnson, J. L., Whalen, D. J., & Mesinger, A. 2013, MNRAS, 436, 1555

Dessart, L. 2019, A&A, 621, A141

Dessart, L., Hillier, D. J., Waldman, R., Livne, E., & Blondin, S. 2012, MNRAS, 426, L76

Donley, J. L., Brandt, W. N., Eracleous, M., & Boller, T. 2002, AJ, 124, 1308

Drake, A. J., Djorgovski, S. G., Mahabal, A., et al. 2009, ApJ, 696, 870

Drake, A. J., Djorgovski, S. G., Prieto, J. L., et al. 2010, ApJ, 718, L127

Duncan, R. C., & Thompson, C. 1992, ApJ, 392, L9 BIBLIOGRAPHY 123

Fan, X., Carilli, C. L., & Keating, B. 2006, Annual Review of Astronomy and Astrophysics, 44, 415

Filippenko, A. V. 1997, ARA&A, 35, 309

Foley, R. J., Kromer, M., Howie Marion, G., et al. 2012, ApJ, 753, L5

Fox, O. D., Smith, N., Ammons, S. M., et al. 2015, MNRAS, 454, 4366

Fransson, C., Chevalier, R. A., Filippenko, A. V., et al. 2002, ApJ, 572, 350

Fransson, C., Challis, P. M., Chevalier, R. A., et al. 2005, ApJ, 622, 991

Fransson, C., Ergon, M., Challis, P. J., et al. 2014, ApJ, 797, 118

Fruchter, A. S., & Hook, R. N. 2002, Publications of the Astronomical Society of the Paciﬁc, 114, 144

Fukugita, M., Ichikawa, T., Gunn, J. E., et al. 1996, AJ, 111, 1748

Fumagalli, M., O’Meara, J. M., & Prochaska, J. X. 2011, Science, 334, 1245

Gal-Yam, A. 2012, Science, 337, 927

Gal-Yam, A., Mazzali, P., Ofek, E. O., et al. 2009, Nature, 462, 624

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

Gezari, S., Halpern, J. P., Grupe, D., et al. 2009, ApJ, 690, 1313

Giacconi, R., Rosati, P., Tozzi, P., et al. 2001, ApJ, 551, 624

Giavalisco, M., Ferguson, H. C., Koekemoer, A. M., et al. 2004, ApJ, 600, L93

Ginzburg, S., & Balberg, S. 2012, ApJ, 757, 178

Greiner, J., Mazzali, P. A., Kann, D. A., et al. 2015, Nature, 523, 189

Heger, A., & Woosley, S. E. 2002, ApJ, 567, 532

Hinton, S. R., Davis, T. M., Lidman, C., Glazebrook, K., & Lewis, G. F. 2016, Astronomy and Computing, 15, 61

Howell, D. A. 2017, Superluminous Supernovae, 431

Howell, D. A., Sullivan, M., Perrett, K., et al. 2005, ApJ, 634, 1190 124 BIBLIOGRAPHY

Howell, D. A., Kasen, D., Lidman, C., et al. 2013, ApJ, 779, 98

Humphreys, R. M. 1999, The Long-Term Variability of Luminous Blue Variables, ed. B. Wolf, O. Stahl, & A. W. Fullerton, 243

Iben, I., J., & Tutukov, A. V. 1984, ApJS, 54, 335

Illingworth, G. D., Magee, D., Oesch, P. A., et al. 2013, The Astrophysical Journal Sup- plement Series, 209, 6

Inserra, C., Prajs, S., Gutierrez, C. P., et al. 2018a, ApJ, 854, 175

Inserra, C., & Smartt, S. J. 2014, ApJ, 796, 87

Inserra, C., Smartt, S. J., Jerkstrand, A., et al. 2013, ApJ, 770, 128

Inserra, C., Smartt, S. J., Gall, E. E. E., et al. 2018b, MNRAS, 475, 1046

Jansen, F., Lumb, D., Altieri, B., et al. 2001, A&A, 365, L1

Jarvis, M. J., Bonﬁeld, D. G., Bruce, V. A., et al. 2013, MNRAS, 428, 1281

Jerkstrand, A., Smartt, S. J., & Heger, A. 2016, MNRAS, 455, 3207

Jones, T., Stark, D. P., & Ellis, R. S. 2012, ApJ, 751, 51

Kaiser, N., Burgett, W., Chambers, K., et al. 2010, in Proc. SPIE, Vol. 7733, Ground-based and Airborne Telescopes III, 77330E

Kasen, D., & Bildsten, L. 2010, ApJ, 717, 245

Kasen, D., Woosley, S. E., & Heger, A. 2011, ApJ, 734, 102

Kashikawa, N., Shimasaku, K., Yasuda, N., et al. 2004, Publications of the Astronomical Society of Japan, 56, 1011

Kaspi, V. M., & Beloborodov, A. M. 2017, Annual Review of Astronomy and Astrophysics, 55, 261

Kirkman, D., Tytler, D., Suzuki, N., et al. 2005, MNRAS, 360, 1373

Kozyreva, A., Blinnikov, S., Langer, N., & Yoon, S. C. 2014, A&A, 565, A70

Kozyreva, A., Kromer, M., Noebauer, U. M., & Hirschi, R. 2018, MNRAS, 479, 3106

Kron, R. G. 1980, The Astrophysical Journal Supplement Series, 43, 305 BIBLIOGRAPHY 125

Laigle, C., McCracken, H. J., Ilbert, O., et al. 2016, ApJS, 224, 24

Lanzetta, K. M., Turnshek, D. A., & Sandoval, J. 1993, The Astrophysical Journal Sup- plement Series, 84, 109

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

Leibundgut, B., Schommer, R., Phillips, M., et al. 1996, ApJ, 466, L21

Leloudas, G., Schulze, S., Krühler, T., et al. 2015, MNRAS, 449, 917

Leloudas, G., Fraser, M., Stone, N. C., et al. 2016, Nature Astronomy, 1, 0002

Li, W., Leaman, J., Chornock, R., et al. 2011, MNRAS, 412, 1441

Limongi, M., & Chieﬃ, A. 2018, The Astrophysical Journal Supplement Series, 237, 13

Lowenthal, J. D., Koo, D. C., Guzmán, R., et al. 1997, ApJ, 481, 673

Lunnan, R., Chornock, R., Berger, E., et al. 2014, ApJ, 787, 138

Lunnan, R., Chornock, R., Berger, E., et al. 2018, ApJ, 852, 81

Madau, P. 1995, ApJ, 441, 18

Madau, P., della Valle, M., & Panagia, N. 1998, MNRAS, 297, L17

Madau, P., & Dickinson, M. 2014, ARA&A, 52, 415

Madau, P., Ferguson, H. C., Dickinson, M. E., et al. 1996, MNRAS, 283, 1388

Maoz, D. 2010, in American Institute of Physics Conference Series, Vol. 1314, American Institute of Physics Conference Series, ed. V. Kalogera & M. van der Sluys, 223–232

Marchesini, D., van Dokkum, P., Quadri, R., et al. 2007, ApJ, 656, 42

Marks, M., Kroupa, P., Dabringhausen, J., & Pawlowski, M. S. 2012, MNRAS, 422, 2246

Mazzali, P. A., Sullivan, M., Pian, E., Greiner, J., & Kann, D. A. 2016, MNRAS, 458, 3455

McDonald, P., Seljak, U., Burles, S., et al. 2006, The Astrophysical Journal Supplement Series, 163, 80

Miller, A. A., Chornock, R., Perley, D. A., et al. 2009, ApJ, 690, 1303 126 BIBLIOGRAPHY

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

Moriya, T. J., Sorokina, E. I., & Chevalier, R. A. 2018, Space Sci. Rev., 214, 59

Moriya, T. J., Tanaka, M., Yasuda, N., et al. 2019, ApJS, 241, 16

Mould, J., Abbott, T., Cooke, J., et al. 2017, Science Bulletin, Volume 62, Issue 10, pp. 675-678, 62, 675

Neill, J. D., Sullivan, M., Gal-Yam, A., et al. 2011, ApJ, 727, 15

Nicholl, M., Guillochon, J., & Berger, E. 2017, ApJ, 850, 55

Nicholl, M., & Smartt, S. J. 2016, MNRAS, 457, L79

Nicholl, M., Smartt, S. J., Jerkstrand, A., et al. 2013, Nature, 502, 346

Nicholl, M., Smartt, S. J., Jerkstrand, A., et al. 2015, MNRAS, 452, 3869

Nicholl, M., Berger, E., Margutti, R., et al. 2016, ApJ, 828, L18

Ofek, E. O., Cameron, P. B., Kasliwal, M. M., et al. 2007, ApJ, 659, L13

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

Paltani, S., Le Fèvre, O., Ilbert, O., et al. 2007, A&A, 463, 873

Pan, Y.-C., Foley, R. J., Smith, M., et al. 2017, MNRAS, 470, 4241

Panagia, N. 2007, in American Institute of Physics Conference Series, Vol. 937, Supernova 1987A: 20 Years After: Supernovae and Gamma-Ray Bursters, ed. S. Immler, K. Weiler, & R. McCray, 236–245

Parsa, S., Dunlop, J. S., McLure, R. J., & Mortlock, A. 2016, MNRAS, 456, 3194

Pastorello, A., Smartt, S. J., Botticella, M. T., et al. 2010, ApJ, 724, L16

Perley, D. A., Quimby, R. M., Yan, L., et al. 2016, ApJ, 830, 13

Perlmutter, S., Gabi, S., Goldhaber, G., et al. 1997, ApJ, 483, 565

Perlmutter, S., Aldering, G., Goldhaber, G., et al. 1999, ApJ, 517, 565

Pettini, M., Shapley, A. E., Steidel, C. C., et al. 2001, ApJ, 554, 981 BIBLIOGRAPHY 127

Pettini, M., Steidel, C. C., Adelberger, K. L., Dickinson, M., & Giavalisco, M. 2000, ApJ, 528, 96

Phillips, M. M. 1993, ApJ, 413, L105

Prajs, S., Sullivan, M., Smith, M., et al. 2017, MNRAS, 464, 3568

Quimby, R. 2014, The First UV Spectra of a Hydrogen-Rich Superluminous Supernova, HST Proposal

Quimby, R. 2016, Far UV Spectroscopy of Superluminous Supernovae, HST Proposal

Quimby, R. M. 2006, PhD thesis, The University of Texas at Austin

Quimby, R. M., Aldering, G., Wheeler, J. C., et al. 2007, ApJ, 668, L99

Quimby, R. M., Cooke, J., & Pritchard, T. 2015, The Astronomer’s Telegram, 7439

Quimby, R. M., Yuan, F., Akerlof, C., & Wheeler, J. C. 2013, MNRAS, 431, 912

Quimby, R. M., Kulkarni, S. R., Kasliwal, M. M., et al. 2011, Nature, 474, 487

Quimby, R. M., De Cia, A., Gal-Yam, A., et al. 2018, ApJ, 855, 2

Rakavy, G., & Shaviv, G. 1967, ApJ, 148, 803

Rakavy, G., Shaviv, G., & Zinamon, Z. 1967, ApJ, 150, 131

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

Richardson, D., Branch, D., Casebeer, D., et al. 2002, AJ, 123, 745

Richardson, D., Jenkins, Robert L., I., Wright, J., & Maddox, L. 2014, AJ, 147, 118

Riess, A. G., Filippenko, A. V., Challis, P., et al. 1998, AJ, 116, 1009

Savoy, J., Sawicki, M., Thompson, D., & Sato, T. 2011, ApJ, 737, 92

Sawicki, M., & Thompson, D. 2005, ApJ, 635, 100

Sawicki, M., & Thompson, D. 2006a, ApJ, 642, 653

Sawicki, M., & Thompson, D. 2006b, ApJ, 648, 299

Scannapieco, E., Madau, P., Woosley, S., Heger, A., & Ferrara, A. 2005, ApJ, 633, 1031

Schmidt, B. P., Suntzeﬀ, N. B., Phillips, M. M., et al. 1998, ApJ, 507, 46 128 BIBLIOGRAPHY

Schulze, S., Krühler, T., Leloudas, G., et al. 2018, MNRAS, 473, 1258

Scoville, N., Aussel, H., Brusa, M., et al. 2007, The Astrophysical Journal Supplement Series, 172, 1

Shapley, A. E., Steidel, C. C., Pettini, M., & Adelberger, K. L. 2003, ApJ, 588, 65

Shappee, B. J., Prieto, J. L., Grupe, D., et al. 2014, ApJ, 788, 48

Shull, J. M. 2009, in American Institute of Physics Conference Series, Vol. 1135, American Institute of Physics Conference Series, ed. M. E. van Steenberg, G. Sonneborn, H. W. Moos, & W. P. Blair, 301–308

Smartt, S. J. 2009, Annual Review of Astronomy and Astrophysics, 47, 63

Smith, J. A., Tucker, D. L., Allam, S. S., & Rodgers, C. T. 2003, AJ, 126, 2037

Smith, M., Sullivan, M., D’Andrea, C. B., et al. 2016, ApJ, 818, L8

Smith, M., Sullivan, M., Nichol, R. C., et al. 2018, ApJ, 854, 37

Smith, N. 2017, Interacting Supernovae: Types IIn and Ibn, 403

Smith, N., Chornock, R., Li, W., et al. 2008a, ApJ, 686, 467

Smith, N., Chornock, R., Silverman, J. M., Filippenko, A. V., & Foley, R. J. 2010, ApJ, 709, 856

Smith, N., & McCray, R. 2007, ApJ, 671, L17

Smith, N., Li, W., Foley, R. J., et al. 2007, ApJ, 666, 1116

Smith, N., Foley, R. J., Bloom, J. S., et al. 2008b, ApJ, 686, 485

Spitler, L. R., Straatman, C. M. S., Labbé, I., et al. 2014, ApJ, 787, L36

Steidel, C. C., Adelberger, K. L., Dickinson, M., et al. 1998, ApJ, 492, 428

Steidel, C. C., Adelberger, K. L., Giavalisco, M., Dickinson, M., & Pettini, M. 1999, ApJ, 519, 1

Steidel, C. C., Adelberger, K. L., Shapley, A. E., et al. 2003, ApJ, 592, 728

Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K. L. 1996, ApJ, 462, L17 BIBLIOGRAPHY 129

Steidel, C. C., & Hamilton, D. 1993, AJ, 105, 2017

Steidel, C. C., Shapley, A. E., Pettini, M., et al. 2004, ApJ, 604, 534

Straatman, C. M. S., Spitler, L. R., Quadri, R. F., et al. 2016, ApJ, 830, 51

Taddia, F., Sollerman, J., Leloudas, G., et al. 2015, A&A, 574, A60

Tanaka, M. 2015, ApJ, 801, 20

Tanaka, M., Moriya, T. J., Yoshida, N., & Nomoto, K. 2012, MNRAS, 422, 2675

Tominaga, N., Limongi, M., Suzuki, T., et al. 2008, ApJ, 687, 1208 van Velzen, S., Farrar, G. R., Gezari, S., et al. 2011, ApJ, 741, 73

Vanden Berk, D. E., Richards, G. T., Bauer, A., et al. 2001, AJ, 122, 549

Webbink, R. F. 1984, ApJ, 277, 355

Whalen, D. J., Fryer, C. L., Holz, D. E., et al. 2013, ApJ, 762, L6

Whelan, J., & Iben, Icko, J. 1973, ApJ, 186, 1007

Williams, R. E., Blacker, B., Dickinson, M., et al. 1996, AJ, 112, 1335

Woodgate, B. E., Kimble, R. A., Bowers, C. W., et al. 1997, in Proc. SPIE, Vol. 3118, Imaging Spectrometry III, ed. M. R. Descour & S. S. Shen, 2–12

Woosley, S. E. 2010, ApJ, 719, L204

Woosley, S. E. 2017, ApJ, 836, 244

Woosley, S. E., Blinnikov, S., & Heger, A. 2007, Nature, 450, 390

Yan, L., Perley, D. A., De Cia, A., et al. 2018, ApJ, 858, 91

Yan, L., Quimby, R., Gal-Yam, A., et al. 2017a, ApJ, 840, 57

Yan, L., Lunnan, R., Perley, D. A., et al. 2017b, ApJ, 848, 6

Yoshida, M., Shimasaku, K., Ouchi, M., et al. 2008, ApJ, 679, 269

Yoshii, Y., Doi, M., Kohno, K., et al. 2014, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 9145, Ground-based and Airborne Telescopes V, 914507

Zhao, G.-B., Zhan, H., Wang, L., Fan, Z., & Zhang, X. 2011, PASP, 123, 725