UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The role of submillimetre galaxies in galaxy evolution Pope, Erin Alexandra 2007

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
831-ubc_2007-319086.pdf [ 22.76MB ]
Metadata
JSON: 831-1.0085807.json
JSON-LD: 831-1.0085807-ld.json
RDF/XML (Pretty): 831-1.0085807-rdf.xml
RDF/JSON: 831-1.0085807-rdf.json
Turtle: 831-1.0085807-turtle.txt
N-Triples: 831-1.0085807-rdf-ntriples.txt
Original Record: 831-1.0085807-source.json
Full Text
831-1.0085807-fulltext.txt
Citation
831-1.0085807.ris

Full Text

The Role of Submillimetre Galaxies in Galaxy Evolution by ' Erin Alexandra Pope B.Sc. (Physics) The University of Lethbridge, 2002 M.Sc. (Astronomy) The University of British Columbia, 2004 A THESIS SUBMITTED IN PARTIAL F U L F I L M E N T OF T H E REQUIREMENTS FOR T H E D E G R E E OF DOCTOR OF PHILOSOPHY in T H E F A C U L T Y OF G R A D U A T E STUDIES (Astronomy) T H E UNIVERSITY OF BRITISH C O L U M B I A October 2007 © Erin Alexandra Pope, 2007 Abstract ii ABSTRACT This thesis presents a comprehensive study of high redshift submillimetre galaxies (SMGs) using the deepest multi-wavelength observations. The submm sample consists of galaxies de-tected at 850 with the Submillimetre Common User Bolometer Array (SCUBA) in the Great Observatories Origins Deep Survey-North region. Using the deep Spitzer Space Tele-scope images and new data and reductions of the VeryLarge Array radio data, I find statisti-cally secure counterparts for 60% of the submm sample, and identify tentative counterparts for most of the remaining objects. This is the largest sample of submm galaxies with statistically secure counterparts detected in the radio and with Spitzer. This thesis presents spectral energy distributions (SEDs), Spitzer colours, and infrared (IR) luminosities for the SMGs. A com-posite rest-frame SED shows that the submm sources peak at longer wavelengths than those of local ultraluminous IR galaxies (ULIRGs), i.e. they appear to be cooler than local ULIRGs of the same luminosity. This demonstrates the strong selection effects, both locally and at high redshift, which may lead to an incomplete census of the ULIRG population. The SEDs of submm galaxies are also different from those of their high redshift neighbours, the near-IR selected BzK galaxies, whose mid-IR to radio SEDs are more like those of local ULIRGs. I fit templates that span the mid-IR through radio to derive the integrated IR luminosities of the submm galaxies and find a median value of L T R (8 — 1000/xm) — 6.0 x 1O 1 2L 0 . I also find that submm flux densities by themselves systematically overpredict L I R when using templates which obey the local ULIRG temperature-luminosity relation. The SED fits show that SMGs are consistent with the correlation between radio and IR luminosity observed in local galaxies. Because the shorter Spitzer wavelengths sample the stellar bump at the redshifts of the submm sources, one can obtain a model independent estimate of the redshift, a(Az/(l + z)) = 0.07. The median redshift of the secure submm counterparts is 2.0. Using X-ray and mid-IR imaging data, only 5% of the secure counterparts show strong evidence for an active galactic nucleus Abstract iii (AGN) dominating the IR luminosity. This thesis also presents deep Spitzer mid-IR spectroscopy of 13 of these SMGs in order to determine the contribution from AGN and starburst emission to the IR luminosity., I find strong polycyclic aromatic hydrocarbon (PAH) emission features in all of the targets, while only 2/13 SMGs have a significant mid-IR rising power-law component which would indicate an AGN. In the high signal-to-noise ratio composite spectrum of the SMGs I find that the AGN component contributes at most 30% of the mid-IR luminosity, implying that the total L i R in SMGs is dominated by star formation and not AGN emission. I also find that the SMGs lie on the relation between the luminosity of the main PAH features and L I R established for local starburst galaxies, confirming that the PAH luminosity can be used as a proxy for the star formation rate. Interestingly, local ULIRGs, which are often thought to be the low redshift analogues of SMGs, lie off these relations, as they appear deficient in PAH luminosity for a given L i R . In terms of an evolutionary scenario for IR luminous galaxies, SMGs are consistent with being an earlier phase in the massive merger (compared with other local or high redshift ULIRGs) in which the AGN has not yet become strong enough to heat the dust and dilute the PAH emission. I further investigate the overlap between high redshift infrared and submm populations using a statistical stacking analysis to measure the contribution of near- and mid-IR galaxy populations to the 850 submm background. For the first time, it is found that the 850 /im background can be completely resolved into individual galaxies and the bulk of these galaxies lie at z < 3. j. .; ; : , " Additionally I present a detailed study of the most distant SMG discovered to date, which I call GN20. This unusually bright source led to the discovery of a high redshift galaxy cluster, which is likely to be lensing the SMG. I discuss the potential for using bright SMGs in future submm surveys to identify high redshift clusters. Finally, for this complete sample of SMGs, I present the cumulative flux distribution at X-ray, optical, IR and radio wavelengths and I determine the depths at which one can expect to detect the majority of submm galaxies in future mm/submm surveys, such as with SCUBA-2, the successor to SCUBA. Contents iv CONTENTS Abstract ii Contents iv List of Tables ix List of Figures x List of Acronyms xiii Acknowledgements xvi 1 Introduction 1 1.1 Galaxy formation and evolution 1 1.2 Extragalactic submillimetre astronomy 2 1.3 Spectral energy distribution . 6 1.4 Infrared emission and star formation rates 9 1.5 Great Observatories Origins Deep Survey 10 1.6 A guide to this thesis 11 2 Data 12 2.1 Submillimetre data 12 2.1.1 Submillimetre Common User Bolometer Array 12 2.1.2 Data analysis 14 2.1.3 850 (im source list 17 2.2 GOODS-N multi-wavelength data 22 2.2.1 Spitzer imaging observations and data 23 Contents v 2.2.2 Spectroscopic and photometric redshifts 25 2.2.3 New radio data reduction 25 3 Spectral energy distributions of submillimetre galaxies 27 3.1 Multi-wavelength identifications 27 3.1.1 Redshifts 35 3.1.2 New radio detections 37 3.1.3 Double radio sources 39 3.1.4 MIPS 24 fim detections 41 3.2 SED templates 44 3.3 Infrared properties 46 3.3.1 Probing dust emission . .. • • •. 46 3.3.2 Probing stellar emission 49 3.3.3 Spitzer photometric redshifts 52 3.4 Spectral energy distributions of SMGs 54 3.5 Infrared luminosities and star formation rates 60 3.5.1 Estimating LIR and SFR 60 3.5.2 Evidence for evolution? 64 3.6 AGN contribution 66 3.7 Summary 69 4 Mid-infrared spectroscopy of submillimetre galaxies 72 4.1 Introduction 72 4.1.1 PAH emission lines 74 4.1.2 The Infrared Spectrometer 76 4.2 IRS target sample 78 4.3 Observations 79 4.4 Data analysis .• 79 4.4.1 Spectral line measurements 88 4.5 Results 90 Contents vi 4.5.1 Redshifts 94 4.5.2 SMG composite spectrum 97 4.5.3 Mid-IR spectral decomposition of individual galaxies 103 4.5.4 AGN classification 104 4.5.5 The GN39 double system 110 4.6 Full SED fits 110 4.6.1 Radio-IR correlation 117 4.7 PAH luminosities 120 4.8 Discussion 128 4.9 Conclusions 131 5 The story of GN20 133 5.1 Follow-up observations of GN20 134 5.1.1 IRAM continuum imaging 134 5.1.2 Submillimeter Array observations 136 5.1.3 SHARC-II imaging 136 5.1.4 Millimetre imaging 138 5.1.5 GMOS optical spectroscopy 138 5.1.6 Keck optical spectroscopy 139 5.2 Counterpart of GN20 139 5.3 Redshift of GN20 143 5.3.1 Photometric redshift 143 5.3.2 Spectroscopic redshift 143 5.4 The SED of GN20 145 5.4.1 Dust properties 148 5.4.2 Stellar properties 149 5.5 Companion source: GN20.2 153 5.6 Discovery of a high redshift cluster 155 5.6.1 Future potential 160 Contents vii 6 Resolving the cosmic infrared background 161 6.1 Stacking method 162 6.2 Stacking IRAC galaxies 164 6.3 Stacking MIPS 24 ^ m galaxies 167 6.4 Comparison to other studies 172 7 How bright are submillimetre galaxies? 174 7.1 Multiwavelength flux distributions 174 7.1.1 Submillimetre 176 7.1.2 Radio 176 7.1.3 Mid-infrared 178 7.1.4 Near-infrared 182 7.1.5 Optical 182 7.1.6 X-ray 185 7.2 Discussion 185 8 Summary and future work 190 8.1 Summary 190 8.1.1 Star formation rate density 191 8.2 Future telescopes and instruments 192 Bibliography • 198 A Relevant formulae 211 A. l Observational cosmology 211 A.2 Poisson probabilities 212 A. 3 Magnitudes 213 B Notes on individual sources 214 B. l Secure counterparts 215 B.2 Tentative counterparts 221 Contents B.3 Multiple counterparts B.4 Possible spurious sources List of Tables ix LIST OF TABLES 2.1 Submm sources in GOODS-North: positions and submm flux densities 21 2.2 Submm sources rejected due to flux boosting 22 2.3 New submm sources from 2006 super-map 23 2.4 Number density of sources in GOODS-N multi-wavelength catalogues 26 3.1 Multi-wavelength photometry of secure submm counterparts 33 3.2 Multi-wavelength photometry of tentative submm counterparts 34 3.3 RMS scatter in the SEDs of SMGs 59 3.4 Candidate AGN-dominated submm sources 67 4.1 Main PAH emission lines 75 4.2 IRS low resolution module properties 76 4.3 IRS observations 82 4.4 IRSredshifts 95 4.5 Classification of mid-IR spectra 108 4.6 PAH and IR luminosities 109 5.1 Positions and fluxes of counterparts to GN20 141 5.2 Summary of physical parameters for GN20 153 5.3 Positions and fluxes of galaxies in cluster SMM1 159 6.1 Contributions from Spitzer galaxies to the CIB 171 7.1 Multiwavelength depths needed to detect 850 yum SCUBA galaxies 187 List of Figures x LIST OF FIGURES 1.1 The extragalactic background 3 1.2 Evolution of the star formation rate density 5 1.3 ULIRG spectral energy distribution 7 1.4 K-correction in the infrared and submm 8 2.1 Atmospheric transmission above Mauna Kea in the submm window 13 2.2 850/xm signal-to-noise super-map 18 3.1 'Postage stamp'images of submm counterparts 31 3.1 (continued) 32 3.2 Redshift distribution of SMGs 36 3.3 Submm-radio positional uncertainty in GOODS-N 38 3.4 Multiple MIPS 24 /xm sources within one SCUBA 850 /xm beam 42 3.5 S24/S850 r a t * ° a s a f u n c t i ° n °f redshift for submm sources 47 3.6 IRAC-based colour-magnitude diagram for GOODS-N 50 3.7 IR photometric redshift accuracy 51 3.8 Composite rest-frame SEDs for submm sources 55 3.9 5g5o/£I.4GHZ r a t i ° a s a function of redshift for submm sources 57 3.10 IR luminosities and SFRs for submm sources 61 3.11 IR luminosity as a function of submm flux density and redshift 65 4.1 Selection of IRS targets 77 4.2 Postage stamp images of IRS targets 80 4.2 (continued) 81 4.3 Calibration of IRS spectra compared to the 16 /xm peak-up photometry 85 4.4 Calibration of IRS spectra compared to the 24 /xm MIPS photometry 86 List of Figures xi 4.5 RMS as a function of integration time for the IRS observations 87 4.6 IRS spectra of SMGs. . . 91 4.6 (continued) . . 92 4.6 (continued) 93 4.7 Comparison of IRS spectroscopic redshifts and previous optical redshifts. . . . 96 4.8 Composite IRS spectrum of SMGs 99 4.9 SMG composite spectrum compared to those of other ULIRGs 102 4.10 Spitzer colour-colour diagram 106 4.11 IR SED of source C2 I l l 4.12 Mid-IR to radio SEDs of SMGs 114 4.12 (continued) 115 4.12 (continued) 116 4.13 Radio IR correlation 118 4.14 Comparison of PAH luminosities 121 4.15 Rest-frame equivalent widths of the main PAH features 123 4.16 Correlations between L I R and PAH luminosities 125 5.1 IRAM PdB 1.3 mm image of GN20 135 5.2 SMA image of GN20 137 5.3 z-band and IRAC 3.6 iim images of GN20 140 5.4 GMOS spectrum of GN20 144 5.5 Keck spectrum of GN20 146 5.6 IR-radio SED of GN20 147 5.7 Stellar population fits to the optical plus IRAC photometry 150 5.8 i-band image of the counterpart to GN20.2 154 5.9 Biz HST colour composite of the central field surrounding GN20 156 6.1 Average submm flux density of Spitzer galaxies 164 6.2 Contribution of IRAC galaxies to the CIB 166 6.3 Distribution of submm SNR for the positions of 24/xm-selected galaxies . . . . 169 List of Figures xii 6.4 Contribution of 24/um galaxies to CIB 170 7.1 Cumulative distribution of deboosted 850 flux density 175 7.2 Cumulative distribution of 1.4 GHz radio flux density 177 7.3 Cumulative distribution of 24/xm flux density 179 7.4 Cumulative distributions of IRAC flux 181 7.5 Cumulative distribution of Ks magnitude 183 7.6 Cumulative distribution of ii75 and 5 4 3 5 magnitudes 184 7.7 Cumulative distribution of X-ray flux 186 8.1 Star formation rate density 193 8.2 Evolution of the S^o/SVo ratio with redshift 196 List of Acronyms xiii LIST OF A C R O N Y M S ACS - Advanced Camera for Surveys AGN - Active Galactic Nucleus AIPS - Astronomical Imaging Processing System ALMA - Atacama Large Millimetre Array AOR - Astronomical Observation Request APEX - Atacama Pathfinder Experiment AzTEC - Astronomical Thermal Emission Camera BGS - Bright Galaxy Sample (IRAS) BLAST - Balloon-borne Large Aperture Submillimeter Telescope CADC - Canadian Astronomy Data Centre CDF - Chandra Deep Field CIB - Cosmic Infrared Background CMB - Cosmic Microwave Background COBE - COsmic Background Explorer COSMOS - Cosmic Evolution Survey CSO - Caltech Submillimeter Observatory CUDSS - Canada-UK Deep Submillimetre Survey DEC - DEClination DEIMOS - DEep Imaging Multi-Object Spectrograph ERO - Early Release Observation EW-Equivalent Width FCF - Flux Conversion Factor FIDEL - Far-Infrared Deep Extragalactic Legacy survey FIR - Far-InfraRed FIRAS - Far- InfraRed Absolute Spectrophotometer FLSV - First Look Survey Verification FWHM - Full Width Half Maximum GOODS - Great Observatories Origins Deep Survey GTO - Guaranteed Time Observations HDF - Hubble Deep Field HST - Hubble Space Telescope ID - Identification IDL - Interactive Data Language List of Acronyms xiv IR - InfraRed IRAC - InfraRed Array Camera IRAF - Image Reduction and Analysis Facility IRAM - Institut de Radio Astronomie Millimetrique IRAS - InfraRed Astronomical Satellite IRS - InfraRed Spectrograph ISO - Infrared Space Observatory JCMT - James Clerk Maxwell Telescope JWST - James Webb Space Telescope KPNO - Kitt Peak National Observatory LABOCA - Large APEX BOlometer CAmera LBG - Lyman Break Galaxy LIR - InfraRed Luminosity LIRG - Luminous InfraRed Galaxy LL - Long Low (IRS module) LMT - Large Millimeter Telescope LRIS - Low Resolution Imaging Spectrometer MAMBO - MAx-Planck Millimetre BOlometer MERLIN - Multi-Element Radio-Linked Interferometer Network MIPS - Multiband Imaging Photometer for Spitzer MNRAS - Monthly Notices of the Royal Astronomical Society NED - NASA/IPAC Extragalactic Database PACS - Photodetector Array Camera and Spectrometer PAH - Polycyclic Aromatic Hydrocarbon PdB - Plateau de Bure PSF - Point Spread Function QSO - Quasi-Stellar Object RA - Right Ascension RMS - Root Mean Square SB - StarBurst SCUBA - Submillimetre Common-User Bolometer Array SED - Spectral Energy Distribution SFR - Star Formation Rate SFRD - Star Formation Rate Density SHADES - SCUBA HAlf Degree Extragalactic Survey SHARC-II - Submillimeter High Angular Resolution Camera SINGS - Spitzer Infrared Nearby Galaxies Survey SL - Short Low (IRS module) SLUGS - SCUBA Local Universe Galaxy Survey List of Acronyms SMA - SubMillimeter Array SMART - Spectroscopy Modeling Analysis and Reduction Tool SMG - Submillimetre Galaxy SNR - Signal to Noise Ratio SPICE - SPitzer IRS Custom Extraction SPIRE - Spectral and Photometric Imaging REceiver SSC - Spitzer Science Center SURF - SCUBA User Reduction Facility SWIRE - Spitzer Wide-area InfraRed Extragalactic survey TKRS - Team Keck Redshift Survey UKIRT - United Kingdom InfaRed Telescope ULIRG - Ultra Luminous InfraRed Galaxy UV - Ultraviolet VLA - Very Large Array WFCAM - Wide-Field CAMera WIRCAM - Wide-field InfraRed CAMera WSRT - Westerbork Synthesis Radio Telescope WVM - Water Vapour Monitor 2D - 2-Dimensional 2MASS - Two Micron All Sky Survey xvi Acknowledgements I would like to thank my supervisor, Douglas Scott, for his ideas, encouragement and sup-port. I have appreciated the opportunities to travel to conferences and telescopes and the en-couragement of collaborating with many experts in the field. I am also grateful to the rest of my PhD committee: Mark Halpern, Vesna Sossi, Ludovic Van Waerbeke and Jasper Wall for stimulating discussions and detailed comments on this thesis. I am grateful to all the members of the GOODS team for useful discussions and for help-ing to make the observations possible. I would like to thank all of my collaborators from the GOODS team, in particular Mark Dickinson and Ranga Chary, whose expertise in galaxy evolution have greatly improved the quality of my research. I am grateful to Colin Borys for his pioneering work with the GOODS-N SCUBA data and for providing his SCUBA data reduction software. I thank Anna Sajina for helpful discussions and for providing the composite IRS spectra of their high redshift Spfeer-selected ULIRGs. I am very grateful to Bernhard Brandl for providing the IRS spectra of local starburst galaxies. This work was supported by the Natural Sciences and Engineering Research Council of Canada and the Canadian Space Agency. Observing trips to Hawaii were funded by the Na-tional Research Council of Canada. The James Clerk Maxwell Telescope is operated by The Joint Astronomy Center on behalf of the Particle Physics and Astronomy Research Council of the United Kingdom, the Netherlands Organisation for Scientific Research, and the National Research Council of Canada. Some of the data used for the analysis was obtained via the Canadian Astronomy Data Centre, which is operated by the Herzberg Institute of Astrophysics, National Research Council of Canada and also supported by the Canadian Space Agency. This work is based on observations made with the Spitzer Space Telescope, which is operated by XVII the Jet Propulsion Laboratory, California Institute of Technology under a contract with the National Aeronautics and Space Administration (NASA). Support for this work was provided by.NASA through an award issued by the Jet Propulsion Laboratory, Caltech. The IRS was a collaborative venture between Cornell University and Ball Aerospace Corporation funded by NASA through the Jet Propulsion Laboratory and Ames Research Center. This research has made use of the NASA/IPAC Extragalactic Database (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA. Last, but not least, I would like to thank my. family and friends for their continuous love and support. In particular, I am grateful to my supportive husband, Robert Kain, for always being my number 1 fan and keeping me in touch with reality. 1 C H A P T E R 1 INTRODUCTION 1.1 Galaxy formation and evolution It came as a surprise when the COsmic Background Explorer (COBE) found that the extra-galactic background light at infrared (IR) wavelengths was roughly equal to that at optical wavelengths (e.g. Puget et al. 1996; Hauser & Dwek 2001; Fig. 1.1). In the local Universe, the volume density is such that there are not enough dusty, IR-luminous galaxies to account for this background, therefore the density of such systems must evolve strongly with redshift (z) and must be important to the build-up of galaxies in the Universe (e.g. see Lagache, Puget, & Dole 2005 for a review). In fact, imaging observations with the Spitzer Space Telescope (hereafter Spitzer) have shown that about 70% of the comoving star formation rate density (SFRD) at 0.5 < z < 3 is obscured by dust (LeFloc'h et al. 2005; Chary 2006). In the local Universe, these dusty galaxies are referred to as luminous infrared galaxies (LIRGs, 11 < log ( L M [LQ]) < 12) and ultraluminous infrared galaxies (ULIRGs, log (Z/TR [L0]) > 12, see Sanders & Mirabel 1996 for a review). However, at high redshift, different selection functions have led to a somewhat disjoint IR luminous galaxy population. Dusty galaxies are detected in surveys at different wavelengths and there exists only marginal overlap between the samples (Daddi et al. 2005; Lutz et al. 2005b; Yan et al. 2005; Pope et al. 2006a). Clearly, with the much larger samples of IR luminous galaxies at high redshift, we are learning that this is a diverse population. Various pieces of observational evidence suggest that galaxy formation follows a 'down-sizing' scheme, where the most massive, most luminous galaxies are forming at early times (e.g. Cowie et al. 1996). However, this seems counterintuitive in the context of the hierarchical theory for the build-up of dark matter haloes (Lacey & Cole 1993). The complexity comes from trying to understand how the gas forms galaxies within these haloes. Two important out-1.2. EXTRAGALACTIC SUBMLLLIMETRE ASTRONOMY 2 standing questions in galaxy evolution are: 1) How do massive systems form at such early times in the Universe?; and 2) How is the build-up of stars, dust and gas related to the growth of black holes in massive galaxies? An evolutionary scenario for massive galaxies was proposed by Sanders et al. (1988a) in which massive galaxies undergo several stages en route to becoming a massive elliptical galaxy. The process starts with an IR luminous phase, most likely triggered by a massive merger. During this stage of intense star formation, the active galactic nucleus (AGN) is also growing. As the AGN becomes larger it begins to feed back on the galaxy, eventually quenching the star formation completely by blowing off all the remaining dust and gas. Thus begins the QSO phase, where the AGN is free to dominate the emission. After exhausting it fuel, eventually the QSO settles down and we end up with a quiescent massive elliptical galaxy. While this scenario is generally accepted, the details such as timescales are not completely undersood. In particular, how long is the IR luminous phase, how long does it take the AGN to develop and can there be multiple episodes of ULIRG and QSO activity? 1.2 Extragalactic submillimetre astronomy The most enigmatic members of the high redshift ULIRG population are the submillimetre (submm) galaxies (SMGs). These galaxies were discovered in abundance starting in the late 1990s with the Submillimetre Common User Bolometer Array (SCUBA, Holland et al. 1999) on the James Clerk Maxwell Telescope (JCMT) and then later with the MAx-Planck Millimetre BOlometer (MAMBO) camera on the Institut de Radio Astronomie Millimetrique (IRAM) 30-m telescope (see Blain et al. 2002 for a review). Due to the poor spatial resolution of current submm telescopes and their faintness at all other wavelengths (e.g. Ivison et al. 2002; Pope et al. 2005), these dusty galaxies have proven to be extremely difficult to study. However, since the dust enshrouded star formation phase is the dominant source of star formation at high redshift, the importance of SMGs to galaxy evolution has motivated a great deal of challenging follow-up work. It has been found that SMGs are very massive systems (Swinbank et al. 2004; Borys et al. 2005; Greve et al. 2005; Tacconi et al. 2006) at z ~ 2 (Chapman et al. 2005, 1.2. EXTRAGALACTIC SUBMILLIMETRE ASTRONOMY 3 Wavelength X (jam) 1000 100 10 1 0.1 Frequency v (GHz) Figure 1.1: The extragalactic background demonstrating that the total power output in the Uni-verse at infrared wavelengths is comparable to the output at optical wavelengths. The grey curve shows the spectrum for one representative z ~ 0 starburst galaxy, M82; the background is the emission-weighted average of all galaxies at all red-shifts. This figure is taken from Lagache et al. (2005). Reprinted, with permission, from the Annual Reviews of Astonomy and Astrophysics, Volume 43 ©2005 by Annual Reviews; http : //www. annual-reviews . org . 1.2. EXTRAGALACTIC SUBMILLIMETRE ASTRONOMY 4 hereafter C05; Pope et al. 2005; Aretxaga et al. 2007) with disturbed morphologies (Chapman et al. 2003b; Conselice et al. 2003; Pope et al. 2005) and number densities comparable to local massive elliptical galaxies. Fig. 1.2 shows the contribution from a sub-sample of SMGs to the global SFRD compared to that of ultraviolet (UV)-selected star-forming galaxies. The co-moving space density of UV-selected galaxies at z ~ 2 is roughly 300 times that of bright SMGs (Adelberger et al. 2005; Chapman et al. 2003a). Even with this incomplete sample of SMGs, it is clear that, while they are much less numerous than UV-selected galaxies, they are just as important to the SFRD. For these reasons, SMGs are thought to be connected to the most massive present day galaxies via an evolutionary sequence (e.g. Lilly et al. 1999). Whether the SMGs and other high redshift populations are entirely different galaxy sys-tems or simply galaxies seen at different phases in their evolution requires a complete multi-wavelength picture. High redshift galaxies selected at optical or near-IR wavelengths such as Lyman Break Galaxies (LBGs, Steidel et al. 1998) and massive near-IR-selected galaxies (BzK galaxies, Daddi et al. 2004) are relatively straightforward to'study at other wavelengths since their positions are accurately known. The follow-up study of galaxies detected with SCUBA suffers from the large submm beam (~15 arcseconds at 850 /mi). The density of optical galax-ies in deep Hubble Space Telescope (HST) surveys is such that there are often ~10 optical galaxies in every SCUBA beam, making identification difficult. Follow-up studies of SMGs present two main challenges: 1) obtaining deep enough data at other wavelengths to have confidence that the plausible counterpart(s) has been detected; and 2) deciding among possible counterparts in order to make the correct identification. The first of these requires considerable amounts of telescope time, while the second can be par-ticularly difficult given the large beam size. Deep radio observations have proven to be the best way to localize the submm emission coming from high redshift 850 /mi selected galaxies (e.g. Ivison et al. 2000; Smail et al. 2000; Barger, Cowie, & Richards 2000). Sources detected in deep 1.4 GHz Very Large Array (VLA) observations are rare enough (n (S\A > 20/xJy) ~ 2 arcmin-2) that the probability that they are randomly associated with an SMG is quite low (Ivison et al. 2002; Borys et al. 2004b). However, even in the deepest radio maps, there is still a substantial fraction (at least 1/4) of SMGs which remain undetected, and 10% have more 1.2. EXTRAGALACTIC SUBMILLIMETRE ASTRONOMY 5 ~\—I—i—i—I—i—I—i—i—i—i—I—i—i—i—r 1 2 3 4 Redshift I i i i i I i i i i L Figure 1.2: Evolution of the star formation rate density (SFRD) in the Universe (Chap-man et al. 2005). The large solid squares show a sub-sample of radio-detected spectroscopically-identified SMGs and the large open square is their estimate of where the radio-undetected SMGs lie - the solid curve is a Gaussian fit to the 4 large squares. The smaller squares are the same SMG points corrected for com-pleteness down to 1 mJy. The smaller open symbols are from UV (high-z circles, triangles), optical (stars, hexagon), mid-IR (low-z circles) and radio (solid circle) surveys (see Chapman et al. 2005 for a full list of references). Symbols which are smaller and shifted above slightly larger symbols have been corrected for dust ex-tinction. The dashed line is the prediction from the model in Blain et al. (2002). Even with this incomplete sample of SMGs, it is clear that, while they are much less numerous than UV-selected galaxies, they are just as important to the SFRD. This figure is reproduced by permission of the American Astronomical Society. 1.3. SPECTRAL ENERGY DISTRIBUTION 6 than one radio source to choose from (Ivison et al. 2002; Chapman et al. 2005). Our current knowledge of SMGs is thus biased toward the radio-detected sub-sample. As a complement, or alternative, to the radio, Multiband Imaging Photometer for Spitzer (MIPS; Rieke et al. 2004) observations with the Spitzer Space Telescope can probe the rest-frame mid-IR emission in SMGs. In addition to being much closer in wavelength to the submm than the optical, the surface density of 24 /xm detected sources (n (S24 > 20/xJy) ~ 13 arcmin-2) is such that the probability of randomly finding a source within a SCUBA search radius, while a little higher than for the radio data, is still much less than in the optical. I use deep MIPS 24 /xm observations to study SMGs in Chapter 3. 1.3 Spectral energy distribution A measure of the energy output from a galaxy as a function of wavelength is called the spectral energy distribution (SED). A fully sampled SED tells you many fundamental properties about a galaxy, such as the redshift, stellar mass, dust mass, dust temperature and star formation rate (SFR). Fig. 1.3 shows a typical ULIRG SED from the Lagache et al. (2003) models, as it would appear at a redshift of 0.02 (left panel) and 2.5 (right panel). The asterisks indicate the two submm channels which are observed with SCUBA (and in the future SCUBA-2, Holland et al. 2006). We are seeing several different types of emission in Fig. 1.3. The smaller peak near 1.6 /xm rest-frame is due to optical emission from stars, while the large peak near 100 /xm rest-frame is from the re-radiation of starlight by dust. SMGs are selected on the basis of this far-IR dust peak which is often modeled at longer wavelengths (> 100 /xm) as a modified blackbody spectrum of the form v$Bv in which Bv is the Planck function (e.g. Halpern et al. 1988; Blain et al. 2002). In this simple model the free parameters are the normalization, the average dust temperature and the dust emissivity, f3. The complex spectrum seen between 3-20 /xm is a combination of warm dust continuum, spectral features and absorption (see Chapter 4 for more details). As an object of a given luminosity is moved to greater distances in the Universe, its light 1.3. SPECTRAL ENERGY DISTRIBUTION 7 100.0 1 10 100 1000 10000 10 100 1000 10000 Wavelength ((im) Wavelength (u.m) Figure 1.3: ULIRG spectral energy distribution from the optical to the radio. The plot on the left is a local ULIRG (z ~ 0.02) taken from Lagache, Dole & Puget (2003), nor-malized to 1000 mJy at 850 /xm, which is a comparable flux to the local ULIRG Arp 220. The right panel plots the same SED shape, shown as it would appear at z ~ 2.5, close to the median redshift for extragalactic submm sources. The plot on the right is normalized to an 850 /xm flux of 10 mJy, which is a typical flux for high redshift SCUBA galaxies. SCUBA observes at 450 and 850 /xm, denoted by asterisks on the SEDs. 1.3. SPECTRAL ENERGY DISTRIBUTION 8 0 1 2 3 4 5 Redshift Figure 1.4: K-correction in the infrared and submm. The red and blue curves show the flux of a galaxy of a given luminosity at 850 and 70 jim, respectively, as function of redshift. Beyond a redshift of 1, the submm flux is essentially constant, whereas the infrared flux drops off rapidly. 1.4. INFRARED EMISSION AND STAR FORMATION RATES 9 is redshifted to longer wavelengths and its flux decreases as the inverse of the distance squared (see Appendix A.l). 'K-correction' refers to the correction applied to an object's flux to ac-count for the fact that the light is observed at a different wavelength than where it was emitted (Oke & Sandage 1968). The correction depends on the shape of the SED at the wavelength of the observations. As seen in Fig. 1.3, the submm channels sample higher up on the far-IR dust peak as one moves to higher redshifts and this is referred to as a 'negative K-correction' (e.g. Blain et al. 2002). This correction compensates for the effects of distance, so that the expected flux of ULIRGs at submm wavelengths does not depend on redshift from z ~ 1-5. Fig. 1.4 shows the submm and infrared flux of a galaxy of a given luminosity as a function of redshift. Beyond z ~ 1, the submm flux is essentially constant, whereas the infrared flux drops off quickly. This allows us to observe galaxies in the submm just as easily at z ~ 5 than at z ~ 1. 1.4 Infrared emission and star formation rates The road to determining the contribution of SMGs to the global star formation budget at high redshift requires a detour to accurately estimate the IR luminosity for individual galaxies. The IR wavelength regime covers over 2 orders of magnitude in wavelength. However, our access to these wavelengths is severely limited by background emission from the atmosphere and the sky. This results in large gaps in the spectral coverage which forces us to extrapolate the flux at near-IR, mid-IR and/or submm wavelengths. Another avenue to get to the total infrared luminosity is through the radio flux and the known radio-infrared correlation (see Condon 1992; Carilli & Yun 1999). The non-thermal radio emission in galaxies, or synchrotron radiation, is caused by the acceleration of relativistic electrons originating in supernova remnants. In the infrared, the galaxy spectrum is dominated by blackbody radiation from dust. Dust grains absorb optical light from stars, and this is re-emitted in the infrared - the key connection between these two processes is the young massive stars. The radio-IR correlation is well established locally but has not been fully tested at high redshift (e.g. Condon 1992 and references therein; Appleton et al. 2004). These approaches for obtaining the infrared luminosity are indirect and require 1.5. GREAT OBSERVATORIES ORIGINS DEEP SURVEY 10 assumptions about the shape of the SED which is known to differ for different galaxy types (e.g. Dale et al. 2005). Before Spitzer, the region of the IR SED shortward of 850 /xm was unexplored in high redshift SMGs (Blain et al. 2002). To obtain the most accurate measure of the IR luminosity requires at least several data points from 8-1000 /xm restframe, and we can now begin to do this with Spitzer. It will not be until the Herschel Space Observatory (Pilbratt 2001) launches in 2008 that we will be able to fill in the most crucial gap in the far-IR SED with deep, high resolution photometry at ~ 200 /xm. The IR luminosity is then used to estimate the star forma-tion rate in individual galaxies after making several assumptions, the most important being that there is little or no contribution from an AGN to the IR luminosity. The combination of uncer-tainties in the shape of the IR spectral energy distribution (SED) and an unknown contribution from an AGN to the IR emission contributes to an incomplete picture of the star formation history of the Universe. 1.5 Great Observatories Origins Deep Survey The Great Observatories Origins Deep Survey (GOODS, Giavalisco et al. 2004) is a huge multi-wavelength campaign to study galaxy evolution in the early Universe. The survey covers two fields with three of NASA's great observatories - HST, Spitzer and Chandra - obtaining some of the deepest images of the Universe ever taken at these wavelengths. The GOODS North field is approximately 10 arcminx 16.5 arcmin, centred on the Hubble Deep Field (HDF) North at 1 2 h 3 6 m 5 5 s , +62°14 , 15" (Giavalisco et al. 2004). The GOODS-N field is one of the most extensively studied regions of the sky, with deep data existing across all wavebands, and therefore it is an ideal data-set for studying SMGs. A full description of the GOODS multi-wavelength data-set is given in Chapter 2. 1.6. A GUIDE TO THIS THESIS 11 1.6 A guide to this thesis The focus of this thesis is to understand the nature of high redshift submm selected galaxies and their role in galaxy evolution using a multi-wavelength approach. The idea is to combine data across the electromagnetic spectrum to make secure identifications of submm counterparts and to study their SEDs to probe many fundamental properties. I use mid-IR spectroscopy to place constraints on the role of AGN in these systems and compare them to other IR luminous galaxies at low and high redshift. The format of this thesis is as follows. Chapter 2 summarizes the submm sample and all other multi-wavelength data used in this thesis. In Chapter 3, I follow a rigorous procedure to identify the counterparts to SMGs and then explore the broad-band SEDs of SMGs to put constraints on their redshifts, dust properties and star formation rates. I present Spitzer mid-IR spectroscopy of SMGs in Chapter 4 which allows me to determine the contribution from an AGN to the luminosity in these systems. In Chapter 5 I present a detailed study of the most distant SMG discovered and present a new method for discovering high redshift galaxy clusters. In Chapter 6 I use statistical stacking methods on Spider-selected galaxies to resolve the cosmic infrared background at submm wavelengths. Chapter 7 provides multi-wavelength flux distributions for this nearly complete sample of SMGs and makes predictions for the depths needed to follow up SMGs in future submm surveys. Finally in Chapter 8, I place this work within the context the field of submm astronomy and galaxy evolution as a whole. I also discuss future telescopes and instalments which will facilitate the next major advancements in the field. In the Appendix, I provide some of the relevant formulas used in this thesis and provide notes on the individual SMGs in the GOODS-N sample. All magnitudes in this thesis use the AB system unless otherwise noted. I use a standard cosmology with H0 = 73kms_ 1 Mpc - 1 , Q M = 0.3 and ftA = 0.7 (e.g. Spergel et al. 2007). C H A P T E R 2 12 DATA This chapter contains a description of the data used in this thesis. Portions of this chapter have been published in the following papers: Pope A., et al., 2005, The Hubble Deep Field North SCUBA Super-map - III. Optical and near-infrared properties of submillimetre galaxies, MNRAS, 358, 149; Pope A., et al., 2006, The Hubble Deep Field-North SCUBA Super-map - IV. Characterizing submillimetre galaxies using deep Spitzer imaging, MNRAS, 370, 1185, and reprinted in this chapter with permission from Blackwell Publishing. 2.1 Submillimetre data The submm sample studied in this thesis is from a SCUBA survey of the GOODS-N field. In this section, I discuss the submm data reduction and source list. 2.1.1 Submillimetre Common User Bolometer Array SCUBA is mounted on the Nasmyth focus of the JCMT on Mauna Kea in Hawaii. The JCMT has a 15-metre primary mirror, making it the largest submm telescope in the world. Ground-based submm observations are hindered primarily by atmospheric absorption by water vapour. Therefore the observatory, at 13,400 feet, is sited above the majority of the water. SCUBA simultaneously observes the sky through two filters, centred on 450 and 850 /mi, where there is significant transmission at this altitude (Fig. 2.1). SCUBA has 3 observing modes: photometry; jiggle-mapping; and scan-mapping. All modes include chopping to remove the effects of sky emission. The GOODS-North SCUBA data-set is composed of observations taken in all 3 SCUBA modes with many different chop patterns, and mostly using in-field chopping (i.e. with a chop throw that is less than the array size). 2.1. SUBMILLIMETRE DATA 13 Wavelength (urn) Figure 2.1: Atmospheric transmission above Mauna Kea in the submm window. The solid curve is for 0.5 mm precipitable water vapour (very dry conditions). The dot-dash and dashed lines are the SCUBA 450 and 850 /mi filters, respectively. Note that while the beam size is smaller at the shorter wavelengths, the transmission is much less than at 850 /im, and this becomes considerably worse when the weather con-ditions are poor. This is why useful 450 /im data are obtained only in the very best weather. Data for this plot were provided by D. Naylor. 2.1. SUBMILLIMETRE DATA 14 2.1.2 Data analysis The submm observations used in this thesis contain all the data in GOODS-N taken by the UBC group (up to the end of 2004) and all other data publicly available in the Canadian Astronomy Data Centre (CADC) JCMT archive at that time (Hughes et al. 1998; Barger, Cowie & Richards 2000; Serjeant et al. 2003). This unbiased survey combines all SCUBA observations into one map. For these reasons, when I construct the combined signal-to-noise SCUBA map, I refer to it as the 'super-map', to emphasize that it is really providing the best estimate for the signal-to-noise ratio for a point source centred on each pixel1. The submm observations and data analysis for this sample are described in detail in Borys et al. (2003, 2004b) and Pope et al. (2005). For simplicity I will refer to these three papers as Paper I, Paper II and Paper III, respectively. I follow the data reduction and source extraction procedure as described in Paper I with a few minor improvements. Starting in 2003, the JCMT Water Vapor Monitor (WVM, B. We-ferling, private communication) became the primary device used for atmospheric corrections, since it provides much more frequent measurements using the same submm beam. In the new observations, I use the WVM to correct for sky extinction (Archibald et al. 2002) whenever those data are available. I exclude data files affected by a known error in the tracking model (Tilanus 2004) and also those files affected by the noise spike that was found in the power spectrum of some of the bolometers from December 2002 to June 2003 (see Paper III for more details). I have also updated the calibration factors used for all observation files since 1997. I calculate the flux conversion factors (FCFs) for a given observation from the average of the calibrations from the same night taken in the same SCUBA mode. These calculated values are used for calibration unless they show a large discrepancy, in which case I use the new standard FCFs published on the JCMT web-page2, which are tabulated approximately monthly. Following the preliminary data analysis steps, which combine the individual nods, cor-1 Similarly, I create a signal 'super-map' where the value in each pixel gives the best estimate for point source centred on that pixel. I use the term 'super-map' to refer to either the signal or signal-to-noise ratio map. 2 h t t p : / /www. jach .hawaii . edu/JACpublic/JCMT/Continuum_observing/SCUBA/astronomy-/ c a l i b r a t i o n / g a i n s . h t m l 2.1. SUBMILLIMETRE DATA 15 rect for the gains in the arrays, and correct for the airmass and sky opacity of observations, the SCUBA data are rebinned into a final map. The preliminary data analysis is performed using SCUBA User Reduction Facility (SURF, Jenness & Lightfoot 2000) and the maps are made using software developed by C. Borys (see Borys 2002 and Borys et al. 2003). I chose 3 arcsecx3 arcsec pixels for the maps, oriented in the Right Ascension (RA) and Declination (DEC) directions. The pixels are small enough that there is no significant loss of resolution at 850 /mi (where the standard deviation of the Gaussian beam is crDeam = 6.3 arcsec). While the pixels are not really small enough at 450 /im (where a b e a m = 3.2 arcsec), these data are noisy enough that there is little advantage in going to smaller pixels. I now describe the mathematical procedure for making the 'super-map'. Let n be the num-ber of integrations, and m the number of bolometers. Then each piece of data can be labelled by t, where t goes from 1 to n, and b, where b goes from 1 to m, and each represent the small-est unit of data after the preliminary data steps, containing roughly 2 seconds of data. I loop through the timestream of data (t = 1, ..n) for each bolometer (6=1, ..m), and add a contribu-tion to the flux (sftt), weighted by the variance3 in the bolometers (oft), to the appropriate pixel i. I follow the same procedure for the photometry, jiggle map and scan map data. After normalizing, the final signal (S), noise (AT), and signal-to-noise ratio (SNR) maps are given by Em ir^n c 6 = 1 Lt=l 6 Sbt °bt Em sr^n r — 6 = 1 2^t=l6(7bt q _ ^ 6 = 1 Z ^ t = l " J 0 t ^ b t n n 2 ' 0.5 N* = E E ' K T ' , (2-2) \ 6 = 1 t= l / and SNR, = (2-3) Flux is only added to a pixel if the position of the bolometer at each integration, P(bt), is 3Because of the Gaussian nature of the noise I use a variance-weighted mean which is equivalent to the maximum likelihood estimator of the mean of the probability distributions. 2.1. SUBMILLIMETRE DATA 16 equal to the position of the pixel, P(i). 5 = 5p(pt)P{i) is the Kronecker delta function, which is 1 when P(bt) = P(i), and 0 otherwise. To aid in the detection of sources, I also fold in the flux from the off-beams. For each sample, in addition to adding the flux to the target pixel, I also add its negative flux at the positions of the off-beams (see Borys et al. 2003 for more details). I can then use a simple Gaussian to fit for sources. The method for detecting sources in these maps requires that I smooth the raw maps with a Gaussian with a full width at half maximum of 14.7 arcsec (shape of the JCMT beam at 850 yum), and then pick off the peaks in the smoothed signal-to-noise ratio map. I perform the smoothing as a convolution weighted by N~2. This is mathematically equivalent to a reduced x 2 minimization of the data with the point spread function (PSF). Since sources are much more apparent by eye in the smoothed maps, it is the smoothed maps that are plotted in this thesis (e.g. Fig. 2.2). The signal super-map gives the best estimate for the flux of a point source centred on each pixel - the convolution of the signal-to-noise ratio super-map with the PSF gives the maximum likelihood of an isolated point source centred on each pixel. When using the scan-map mode on the JCMT, pointing errors on the order of 3 arcsec are a concern since this is the distance the telescope moves on the sky in one sampling time. In order to correct the astrometry of the scan-map data, I did a least squares comparison between overlapping jiggle map and scan map data4. I found that a shift of 1 pixel (3 arcsec) in each direction improves the scan-map astrometry and applied this shift to all the scan-map data that went into the final super-map (see also Borys et al. 2003). Once the astrometry within the submm data are consistent, I check the correlation between the super-map and the 1.4 GHz VLA radio map. I also check for evidence of any overall shift between the positions of the radio counterparts and the submm sources. Both methods find an additional shift of 1.5 arcsec. This shift is less than the size of the pixels in the submm maps and may have little effect, nev-ertheless I apply this correction to the submm positions when searching for multi-wavelength counterparts to achieve the best possible astrometry in the super-map. The noise in the super-map is very non-uniform, due to combining all different modes of SCUBA observations (scan-mapping, jiggle-mapping and photometry), as well as different 4The astrometry of SCUBA jiggle map observations is much better than that of scan maps. 2.1. SUBMILLIMETRE DATA 17 exposure times and different chopping configurations. The new super-map covers a total area of approximately 200 square arcminutes with an average la Root Mean Square (RMS) of 3.4 mJy. However, half of the super-map is much deeper with a la RMS of < 2.5 mJy, and 70% of the sources come from this deeper region. Fig. 2.2 shows the SCUBA signal-to-noise ratio super-map along with the noise map which indicates the locations of the sources. 2.1.3 850 /jm source list The 2005 SCUBA 'super-map' of GOODS-N contains 40 submm sources detected above 3.5a at 850 /im (Paper III). Table 2.1 lists the complete sample of GOODS-N submm mapping sources, along with the submm positions, the 850 fim flux densities and the coordinates of the identified galaxy counterparts (see Chapter 3). Since all mapping observations taken in GOODS-N are combined into the super-map, the depths achieved in this map are very non-uniform. The raw noise levels of the detected sources range from 0.3^ 1.1 mJy, as seen in Table 2.1. I use the naming convention with the prefix 'GN' when referring to individual sources in this thesis and list the the full submm names in Table 2.1. Since I have chosen a 3.5a detection threshold, I have to be aware of the possibility of spurious sources in the sample. Due to the effects of confusion noise and the steeply-falling source submm counts, sources detected in SCUBA maps at low significance have typically had their fluxes boosted by some factor which depends on both the source flux and local noise level (see Coppin et al. 2005 and references therein). Several ways of dealing with this bias have been used in other submm studies; the fundamental issues one needs to deal with are that low signal-to-noise sources are, on average, at somewhat lower flux than they appear in maps, and that sources with higher noise (at a given signal-to-noise level) are more likely to be spurious. In order to obtain the most robust sample, I choose to apply the flux deboosting prescription of Coppin et al. (2005) which is a Bayesian method dependent on prior knowledge of the source count model5. This method has been tested extensively for the SHADES survey (see Coppin 5I used a simplified approach of adopting a prior distribution coming from a single chop function. This does not strictly apply to the entire super-map (where several chop strategies have been used), but I verified that this makes little difference to the results (see Coppin et al. 2006). 2.1. SUBMILLIMETRE DATA 18 • * •9 / 1 - . m 2 -4o Figure 2.2: The 850 signal-to-noise super-map of the HDF-N region. The rectangle shows the boundary of the GOODS-North region. This map has been cleaned to remove the negative beams of the sources (see Pope 2004 and Pope et al. 2005). The noise map is shown in the top right corner, where white corresponds to lower noise level. The crosses show the positions of the 40 candidate sources. 2.1. SUBMILLIMETRE DATA 19 et al. 2006) and simplifies to a relation involving only the raw map flux (sraw) and noise (nraw) of each source; s r a w > -1 + 3 x n r a w -f 0.2 x n^aw. (2.4) If a source does not meet this criteria then it has a non-negligible probability (> 5 %) of deboosting to zero flux. Applying this prescription, I find that five of the 40 sources have a non-negligible probability (> 5 %) of deboosting to zero flux. The threshold chosen is based on having a very low false detection rate in the final catalogue. As expected (Ivison et al. 2002; Coppin et al. 2005), these five sources, GN27, GN29, GN33, GN36 and GN38, have the highest noise levels (n r a w > 5 mJy). Given that I expect some small fraction of sources in the sample to be spurious, I excluded these five from the list in order to make a more robust sample. If I ignore regions of the SCUBA map with n r a w > 5 mJy then the sample is complete down to 3.5a. The submm sample in GOODS-N contains 35 850 jim sources. Note that the main results of this thesis are not strongly affected by details of how I construct the final sample. Three of the five sources which are excluded from the sample based on flux deboosting (GN27, GN29 and GN38) appear to have radio and/or Spitzer identifications and are probably genuine submm sources. I list these additional sources in Table 2.2. However, I stress that they are not in the main catalogue of 35 submm sources, the selection of which is based solely on the SCUBA data, uninfluenced by whether there appears to be a counterpart at other wavelengths. The flux deboosting algorithm also provides corrected values of the flux and noise for all remaining sources, based on the probability distribution of flux. Obtaining the most accurate submm photometry is important in order to correctly model the SED and investigate variations in properties with submm flux. Fitting the results from the SHADES sample (see Coppin et al. 2006), I derive the following relations for the deboosted flux (saeb) and deboosted noise ( « d e b ) ; s d e b = 3.67 x s™ 7 - 75.36 x n™ 5 + 69.20, (2.5) 2.1. SUBMILLIMETRE DATA 20 n d e b = -14.25 x s™ 5 + 6.86 x n°T^ + 9.38. (2.6) For the 35 sources in the sample I list the raw and deboosted fluxes in Table 2.1. It is clear that the low signal-to-noise ratio and high noise sources are most affected by flux-boosting. In response to recent claims in the literature that radio-undetected submm sources may be spurious (Ivison et al. 2002; Greve et al. 2004), I have performed additional statistical tests on the data in both the spatial and temporal domains. These tests, and the results, are discussed in detail in Paper III and Pope (2004). In summary, I find no indication in the statistics of the data that the radio-undetected sources are any less secure than the radio-detected sources and none of the sources show significant inconsistencies in the raw temporal data. While the focus is normally on the 850 /tin observations in extragalactic submm surveys, SCUBA simultaneously collects data at both 450 /xm and 850 /xm. The atmosphere in the shorter waveband is much worse than at 850 /xm and therefore observing conditions that are adequate for 850 /xm observations are not sufficient to provide high quality 450 /xm data. Nev-ertheless, I have reduced the 450/xm data. None of these 35 sources is individually de-tected above 3a in the SCUBA 450 yum map. However, the sample is statistically detected at 6.3 ± 2.2 mJy. This value is obtained by measuring the 450 /xm flux at the positions of the 850 /xm detections and calculating a noise-weighted average. I expect that this measurement will be biased low due to positional uncertainties (from low signal-to-noise ratios) in both the 850 /xm and 450 /im maps. Nevertheless this value is low compared to what is predicted from SED templates (roughly by a factor of 3). In addition to the positional offsets, the 450 /xm calibration is very uncertain and the noise properties in deep 450 /xm SCUBA maps are known to be complicated. Hence stacking flux down to these levels is essentially untested at 450 /xm. For this reason (and the reasons discussed in Paper III, where I abandoned the attempt to detect individual sources) I do not consider 450 /xm data further. SUBMILLIMETRE DATA Table 2.1: Submm sources in GOODS-North. SMMID SMM Name RA IRAC position Dec. Raw Ss50Mm (mJy) Corrected Sgm^m (mJy) GN01 SMMJ123606 7+621556 12 36 06.70 62 15 50 43 7.3 ± 1.5 6.2 ± 1.6 GN02 SMMJ 123607 7+621147 12 36 08.81 62 11 43 57 16.2 ± 4 . 1 12.1 ± 4 . 3 GN03 SMMJ 123608 9+621253 12 36 08.65 62 12 50 81 16.8 ± 4 . 0 12.8 ± 4 . 2 GN04 a SMMJ123616 6+621520 12 36 16.11 62 15 13 53 5.1 ± 1 . 0 4.9 ± 0 . 7 12 36 15.82 62 15 15 34 GN05 SMMJ123618 8+621008 12 36 19.13 62 10 04 32 6.7 ± 1.6 5.2 ± 1.8 GN06 SMMJ123618 7+621553 12 36 18.33 62 15 50 40 7.5 ± 0 . 9 7 . 5 ± 0 . 9 b GN07 a SMMJ 123621 3+621711 12 36 21.27 62 17 08 16 8.9 ± 1.5 8.9 ± 1.5b 12 36 20.98 62 17 09 55 GN08 SMMJ123622 2+621256 12.5 ± 2 . 7 10.0 ± 3 . 0 GN09 SMMJ 123622 6+621617 12 36 22.07 62 16 15 81 8.9 ± 1.0 8.9 ± 1.0b GN10 SMMJ123633 8+621408 12 36 33.40 62 14 08 72 11.3 ± 1.6 11.3 ± 1.6b G N U SMMJ 123637 2+621156 12 36 37.51 62 11 56 52 7.0 ± 0 . 9 7 . 0 ± 0 . 9 b GN12 SMMJ 123645 8+621450 12 36 46.07 62 14 48 76 8.6 ± 1 . 4 8.6 ± 1.4b GN13 SMMJ 123650 5+621317 12 36 49.72 62 13 12 89 1.9 ± 0 . 4 1 . 9 ± 0 . 4 b GN14 SMMJ123652 2+621226 12 36 52.08 62 12 26 21 5.9 ± 0 . 3 5 . 9 ± 0 . 3 b GN15 SMMJ 123656 5+621202 12 36 55.82 62 12 01 13 3.7 ± 0 . 4 3 . 7 ± 0 . 4 b GN16 SMMJ 123700 4+620911 12 37 00.26 62 09 09 77 9.0 ± 2 . 1 6.9 ± 2 . 5 GN17 SMMJ 123701 2+621147 12 37 01.59 62 11 46 25 3.9 ± 0 . 7 3 . 9 ± 0 . 7 b GN18 SMMJ 123703 0+621302 12 37 02.55 62 13 02 22 3.2 ± 0 . 6 3 . 2 ± 0 . 6 b GN19 a SMMJ 123707 7+621411 12 37 07.19 62 14 07 97 10.7 ± 2 . 7 8.0 ± 3 . 1 12 37 07.58 62 14 09 50 GN20 SMMJ123711 7+622212 12 37 11.88 62 22 12 11 20.3 ± 2 . 1 20.3 ± 2 . l b GN21 SMMJ123713 3+621202 12 37 14.06 62 11 56 75 5.7 ± 1.2 4.9 ± 1.2 GN22 SMMJ123607 3+621020 12 36 06.84 62 10 21 36 14.4 ± 3 . 9 10.5 ± 4 . 3 GN23 SMMJ 123608 4+621429 12 36 08.60 62 14 35 30 7.0 ± 1.9 4.9 ± 2 . 3 GN24 SMMJ123612 4+621217 12 36 12.00 62 12 21 99 13.7 ± 3 . 6 10.0 ± 4 . 1 GN25 SMMJ123628 7+621047 12 36 29.12 62 10 45 92 4 . 6 ± 1.3 3.2 ± 1.4 GN26 SMMJ123635 5+621238 12 36 34.51 62 12 40 93 3.0 ± 0 . 8 2.2 ± 0 . 8 GN28 SMMJ123645 0+621147 12 36 44.51 62 11 41 92 1.7 ± 0 . 4 1 . 7 ± 0 . 4 b GN30 SMMJ123652 7+621353 12 36 52.75 62 13 54 59 1.8 ± 0 . 5 1 . 8 ± 0 . 5 b GN31 SMMJ 123653 1+621120 12 36 53.22 62 11 16 69 2.8 ± 0 . 8 2.1 ± 0 . 6 GN32 SMMJ123659 1+621453 12 36 58.74 62 14 58 92 5.3 ± 1.4 3.8 ± 1.6 GN34 SMMJ 123706 5+622112 12 37 06.22 62 21 11 57 5.6 ± 1 . 6 3.8 ± 1 . 9 GN35 SMMJ 123730 8+621056 14.3 ± 3 . 9 10.4 ± 4 . 3 GN37 SMMJ 123739 1+621736 12 37 38.26 62 17 36 38 6.8 ± 1.9 4.7 ± 2 . 3 GN04.2 SMMJ123619 2+621459 12 36 18.67 62 15 03 09 3.6 ± 1.0 2.7 ± 1 . 0 GN20.2 SMMJ123709 5+622206 12 37 08.77 62 22 01 78 11.7 ± 2 . 2 9.9 ± 2 . 3 a These submm sources appear to have two radio/ERAC counterparts contributing to the submm flux. b No correction is applied to these sources, since they are either high signal-to-noise or have very low noise levels. 2.2. GOODS-N MULTI-WAVELENGTH DATA 22 Table 2.2: Submm sources rejected due to flux boosting. These five sources are rejected from the sample based on the fact that they have a non-negligible probability of deboost-ing to zero flux density. Nevertheless, I include them here since the probability is high that most of them are real. S M M I D S M M Name IRAC position Raw 5 8 5 0 M m Corrected Ssso R A Dec. (mJy) (mJy) GN27 S M M J 123636.9+620659 12:36:36.00 62:07:00.51 24.0 ±6.1 17.6 ±5.8 GN29 SMMJ123648.3+621841 12:36:48.17 62:18:42.93 20.4 ±5.7 14.7 ±5.6 GN33 SMMJ123706.9+621850 21.8 ±5.8 15.9 ±5.6 GN36 S M M J 123731.0+621856 24.8 ±7.0 17.7 ±6.5 G N 3 8 a SMMJ123741.6+621226 12:37:41.64 62:12:23.61 24.9 ±6.5 18.1 ±6.1 12:37:41.16 62:12:20.89 a This submm sources appear to have two radio/TRAC counterparts contributing to the submm flux. 2006 super-map sources In 2006, I re-reduced all of the super-map data along with some additional photometry data. These photometry data were initially left out of the super-map due to their incompatibility with the reduction software (see Paper I for more details). This problem with the data files was remedied in 2006 by M . Crowe. The inclusion of these data in the super-map results in 3 new sources which I named GN39, .GN40 and GN41 following the previous naming convention. A l l three of these sources survived the flux deboosting criteria outlined in the previous section. Table 2.3 lists the positions and fluxes of these 3 sources. Due to their late arrival in the sample, they are not included in all the analysis in this thesis. However, I list their counterparts for completeness. 2.2 GOODS-N multi-wavelength data The GOODS Legacy survey consists of three major components: Chandra 2 Msec X-ray ob-servations (Alexander et al. 2003b); deep HST optical imaging using the Advanced Camera for Surveys (ACS) in four bands, - B 4 3 5 , Vm6, i775 and z 8 5 0 (Giavalisco et al. 2004); and deep 2.2. GOODS-N MULTI-WAVELENGTH DATA 23 Table 2.3: New submm sources from 2006 super-map. These three sources were discovered in the new 2006 super-map after adding in additional photometry observations. SMMID SMM Name IRAC position Raw Ss50 M m Corrected SssoMm RA Dec. (mJy) (mJy) GN39 a SMMJ123711.1+621325 12:37:11.33 62:13:31.02 7.4 ± 1.9 5.2 ± 2 . 4 12:37:11.97 62:13:25.77 GN40 SMMJ123713.7+621822 12:37:13.86 62:18:26.24 13.1 ± 2 . 7 10.7 ± 2 . 9 GN41 SMMJ 123639.4+620752 11.9 ± 3 . 1 8.8 ± 3 . 5 This submm source appears to have two radio/IRAC counterparts contributing to the submm flux. Spitzer imaging at five infrared wavelengths (Dickinson et al., in preparation6). In addition to the extensive space-based imaging campaign, several ground-based pro-grammes are also targeting the GOODS fields for imaging and spectroscopy. J and iTs-band imaging has been taken using Flamingos at the Kitt Peak National Observatory (KPNO, Elston et al. 2003). While the background noise in the near-IR images is fairly uniform, a variable PSF may lead to further variations in the sensitivity.'All the submm sources are in regions which reach Ks = 22.5. A l l optical and near-IR photometry is carried out using SExtractor (Bertin & Arnouts 1996) and I use the MAG_AUTO values in this thesis. Optical counterparts for a large fraction of the submm sample using the HST images of GOODS-N (Giavalisco et al. 2004) are presented in Paper III. The Spitzer imaging of GOODS-N was completed in November 2004 and since it is the focus of much of the analysis in this thesis, I discuss the observations and data analysis in more detail now. 2.2.1 Spitzer imaging observations and data Spitzer imaging observations of the field were obtained as part of the GOODS Legacy pro-gramme (Dickinson et al., in preparation). The entire GOODS-N field was imaged at 3.6, 4.5, 5.8 and 8.0/jm with InfraRed Array Camera (IRAC; Fazio et al. 2004) and at 24 /jm with 6While the GOODS-N Spitzer data have not been published yet, this is a legacy program and science quality images and catalogues based on extensive testing have been made publicly available. Many papers have been published using the GOODS Spitzer data. 2.2. GOODS-N MULTI-WAVELENGTH DATA 24 MIPS. These observations are currently the deepest Spitzer images. The Spitzer/GOODS data reduction will be presented in Dickinson et al. (in preparation). The IRAC data have a resolution of ~2 arcseconds. It was found that using a 'Mexican hat' kernel for IRAC source detection improved deblending in crowded regions. This occasionally results in improved IRAC source positions and fluxes for the submm counterparts. The IRAC photometry was measured using matched apertures in SExtractor (Bertin & Arnouts 1996). For the present analysis, I use a 4 arcsecond diameter aperture and apply aperture corrections as determined through simulations of the GOODS IRAC data. The uncertainties in the IRAC photometry were also estimated through these simulations, since the SExtractor values were not accurate. The formal la point source sensitivities of the IRAC data for isolated sources are 0.026, 0.044, 0.290 and 0.321 /xJy at 3.6, 4.5, 5.8 and 8.0/xm, respectively. Based on simulations, the 50% detection completeness limits for the SExtractor IRAC catalogue used in this paper are 0.4 /xJy at 3.6 and 4.5 /xm and 0.9 /Jy at 5.8 and 8.0 /xm. The IRAC and ACS (vl.0) images are aligned to an overall RMS of 0.25 arcsec and I have corrected for the known offset of -0.38 arcsec in declination between the ACS and radio frames7. All positions in this paper are relative to the radio frame. The MIPS 24 /xm data were reduced using an Interactive Data Language (IDL) pipeline, and photometry was obtained by fitting the Point Spread Function (PSF). MIPS source detection was carried out using the IRAC source positions as a prior, which facilitates the comparison between the IRAC and MIPS catalogues. 24 /xm sources are selected to be above 3<r. Since the 24 /xm catalogue is fundamentally defined by the existence of IRAC priors, the low signal-to-noise ratio 24 /xm sources are much more likely to be real. Final astrometric accuracy is better than 0.2 arcseconds. More details on the MIPS source catalogues will be given in Chary et al. (in preparation). The final 24 /im image reaches a la depth of about 5 /xJy, with an 84% completeness limit of 24 /xJy. The number densities of sources in the Spitzer catalogues are listed in Table 2.4. 7http://data.spitzer.caltech.edu/popular/goods/Documents/goods-drl. html 2.2. GOODS-N MULTI- WAVELENGTH DATA 25 2.2.2 Spectroscopic and photometric redshifts Photometric redshifts have been estimated for a large sample of sources in GOODS-North us-ing all available optical/near-IR data (U KPNO, B, V,R,I,z SUBARU, Capak et al. 2004, J , Ks KPNO, £ 4 3 5 , U 6 0 6 , ^775 , ^850 ACS HST, Giavalisco et al. 2004). These extensive photometric data have been fit to a suite of SED templates to estimated redshifts (see Mobasher et al. 2004 for more details). In Paper III, we looked in detail at the photometry and redshift probability distribution for each submm counterpart to check for noisy or inconsistent values and only reported photometric redshifts which were well constrained. In addition to photometric redshifts for a large fraction of optical galaxies, there are roughly 1500 publicly available spectroscopic redshifts over the 160 square arcminute GOODS-N field (Cohen et al. 2000; Cowie et al. 2004; Wirth et al. 2004; Chapman et al. 2005). This means that several counterpart candidates already have spectroscopic redshifts (Section 3.1.1). 2.2.3 New radio data reduction The GOODS-N region was initially imaged with the V L A at 8.5 and 1.4 GHz (Richards et al. 1998; Richards 2000). The original 1.4 GHz catalogue contained 371 objects (n ~ 0.8 arcmin -and the RMS in the centre of the image was around 7.5 /dy. The radio data used in this pa-per are a combination of the reprocessed V L A A-array HDF data plus new V L A B-array data (~28 hr, Morrison et al. in preparation). The reprocessed A-array data lead to a 25% reduction in the phase centre noise, and the A+B array data lead to a 30% reduction in the phase centre noise, a ~ 5.3 /dy. In addition, the local RMS is significantly improved at greater distances from the centre of the map. Catalogues were made down to 5a, 4a and 3a, and the number density of sources is given in Table 2.4. The radio positional error is 0.16 and 0.26 arcsec for a 5a and 3a detection, respectively. These new data and reductions of the V L A - H D F radio data will be presented in Morrison et al. (in preparation). In order to estimate the radio fluxes, I used radio maps at three different resolutions: the full resolution (1.6 arcsec) map; a 3 arcsec convolved map; and a 6 arcsec convolved map. The lower resolution maps are useful for searching for lower surface brightness sources and 2.2. GOODS-N MULTI-WAVELENGTH DATA 26 Table 2.4: Number density of sources in GOODS-N multi-wavelength catalogues. Uncertain-ties given are the la Poisson errors from the number of galaxies in each cata-logue. The submm number densities are from the number counts given in Borys et al. (2003) for the GOODS-N SCUBA data. Catalogue n (arcmin 2) Submm S > 6 mJy 0.14 ± 0 . 0 4 Submm S > 2 mJy 0.53 ± 0 . 1 2 Radio 4a 1.80 ± 0 . 1 Radio 3a 11.30 ± 0 . 1 MIPS 3a 12.5 ± 0 . 2 IRAC 3a 88.9 ± 0 . 7 Radio and MIPS 3a 2.2 ± 0 . 1 for accurately measuring fluxes of the resolved sources (Owen et al. 2005). Of these three measurements, the full resolution map yielded the best signal-to-noise ratio in most cases. For sources that have a nearby radio source, I only used the full resolution map. The Astronomical Imaging Processing System (AIPS) J M F I T task was used to measure the flux density of the radio sources, which are corrected for primary beam attenuation and bandwidth smearing ef-fects. Sources which were found by J M F I T to have zero as the minimum size of the major axis were assumed to be unresolved. For such unresolved sources the fitted peak flux density for the Gaussian functions fitted with J M F I T is the best estimate for the total flux density (Owen et al. 2005). Errors in the flux density and position were calculated using the method discussed in Condon (1997). 27 CHAPTER 3 SPECTRAL ENERGY DISTRIBUTIONS OF SUBMILLIMETRE GALAXIES In this chapter, I use the deep Spitzer images from the GOODS Legacy programme and a new reduction of the 1 . 4 GHz VLA radio data (Morrison et al. in preparation) to study the sample of bright (S 8 5 0 > 2 mJy) SMGs introduced in Chapter 2. I follow a strict procedure to identify counterparts to the submm emission to produce a robust sample of submm counterparts. Once the counterparts are known, I use the available optical spectroscopic and photometric data to obtain a redshift estimate for each source. I present SEDs, Spitzer colours, and IR luminosities for these galaxies. I explore how well different points on the IR SED probe the total IR lumi-nosity. I use the rest-frame near-IR SED to Investigate the presence of AGN in these systems. The appendix contains notes on individual objects and postage stamp figures at optical, mid-IR and radio wavelengths. The results presented in this chapter are published in the following paper: Pope A., et al., 2006, The Hubble Deep Field-North SCUBA Super-map - IV. Characterizing submillimetre galaxies using deep Spitzer imaging, MNRAS, 370, 1185. The contents of this paper has been reprinted in this chapter with permission from Blackwell Publishing. This chapter does not contain references to all other papers in the literature published after this paper was published. 3.1 Multi-wavelength identifications To identify counterparts to the sample, I have used all available multi-wavelength data, in particular the new Spitzer images and the new reduction of the VLA 1.4 GHz radio data. I employ a search radius of 8 arcseconds, which I derived by minimizing the probability that K or more SMGs (out of M) have at least one radio source within the search region at random (see Paper II for more details). Note that I have slightly expanded the search radius from the 3.1. MULTI- WAVELENGTH IDENTIFICATIONS 28 7 arcseconds used in Paper II and Paper III in light of the new radio reduction; when carrying out this calculation with the new radio catalogue, there is a clear minimum at 8 arcseconds. The counterpart identification of SMGs comes down to a set of statistical estimates. For all possible counterparts within the search region I need to determine the reliability of the associ-ation. In the submm, this is usually done using Poisson statistics to calculate the probability of finding the object at random at that position (e.g. Lilly et al. 1999; Ivison et al. 2002). The probability that a source of a given flux density, S, is randomly found within 6 of a submm source is given by ps = l- e x p ( - 7 r n ( > s ) 6 | 2 ) , (3.1) where n>5 is the surface density of sources above flux density level S per unit solid angle, and 9 is the search radius. The final probability of random association usually includes a correction factor to take into account the fact that one is looking for counterparts within a given search region using catalogues which have a specific depth (Downes et al. 1986; Dunlop et al. 1989; see Appendix A.2 of this thesis). I denote the corrected Poisson probability (using the method of Downes et al. 1986) as P. I want to use as much information as possible when assessing the reliability of the associations and I expect a large fraction of the SMGs to be detected in the radio and mid-IR. I therefore made a joint catalogue by matching radio and MIPS sources within 1 arcsecond. Recall that since IRAC priors were used in making the MIPS catalogue, I already have a joint MIPS and IRAC catalogue. For all counterparts discussed in this thesis, I have calculated P using sev-eral catalogues and criteria: 1) a > 4a radio catalogue (R); 2) a > 3a MIPS 24/xm catalogue (M); 3) a joint radio and 24 /xm catalogue (R/M); and 4) a joint 24 /xm and red IRAC catalogue (M/l). In the latter, I use the MIPS/IRAC catalogue and make cuts on colour, rather than mag-nitude or flux, when calculating n in Equation 3.1. Considering all four catalogues, I choose the counterpart with the lowest probability of random association. I have assigned counterparts in a systematic way, starting with the catalogue with the smallest number density (> 4a radio catalogue) and progressing to other catalogues if potential counterparts are not found. Table 2.4 lists the number densities for each of these catalogues at the limit of the survey. Note that I 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 29 am confident in using these low signal-to-noise ratio catalogues since all of the potential coun-terparts are also detected at much higher significance in all four IRAC channels, and therefore it is unlikely that they are random noise peaks in the MIPS or radio images. A counterpart is considered secure if it has P < 0.05. This is the same threshold used by Ivison et al. (2002; 2005; 2007) and ensures that the fraction of incorrect counterparts remains low. I also list tentative counterparts for sources which have 0.05 < P <0.20. Statistically I expect only ~ 10% of these to be just random associations, but clearly I cannot tell which ones are genuine. Therefore I have taken the conservative approach to only use the secure counterparts in the analysis for this chapter, but note there are no changes in the results if I included the tentative counterparts. As an additional test into the reality of the secure counterparts, I have simulated random positions in GOODS-N and applied the above procedure to look for counterparts with P < 0.05. The results of these Monte Carlo simulations is that I find a counterpart with P < 0.05 about 5% of the time, and therefore I expect that only one of the secure counterparts is a random association. Using the above procedure, I have found secure counterparts for 21/35 (23/38 including the 2006 super-map sources) SMGs and an additional 12 tentative counterparts. Table 3.1 and Table 3.2 provide the optical-through-radio photometry for the secure and tentative identified SMGs, respectively. Sources are ordered by redshift where bold-faced redshifts are spectro-scopic, otherwise they are photometric. Sources with no optical redshift information have tentative redshifts determined from IRAC colours listed in brackets (see Section 3.3.3). The second to last column in these tables lists the probability that the counterpart is a random asso-ciation. The letter after the probability indicates which catalogue the random probabilities are based on: R (> 4a radio), M (> 3a MIPS 24 /mi), R/M (> 3a radio and > 3a MIPS 24/im), M/I (MIPS and red IRAC detection). In the last column I list the IR luminosity (see Section 3.5) and I only list the infrared luminosity (LIR) if the source is detected at 1.4 GHz and 24 jim and has a reliable redshift estimate. All but one of the secure counterparts are detected in the radio, at 24 /im, and in all four IRAC channels, but this is partially an artifact of the identifica-tion procedure, since I require P < 0.05 in one of these multi-wavelength catalogues in order 3.1. MULTI- WAVELENGTH IDENTIFICATIONS 30 for a idenfication to be classified as secure. However, the presence of bright 24 /j,m or red IRAC counterparts enhances my confidence in low signal-to-noise ratio radio detections as submm counterparts. There are three sources for which I am unable to assign a unique identification due to the fact that they have multiple counterparts which appear equally likely. The last 2 sources in Table 3.1 are from the 2006 super-map and are not included in any of the analysis in this chapter. The list of securely identified sources in GOODS-N includes GN14, also known as HDF850.1 (Hughes et al. 1998), which is discussed in detail in Dunlop et al. (2004, see also Appendix A l for more details). The submm position is known from deep Multi-Element Radio-Linked Interferometer Network (MERLIN) and VLA data. However, because the submm source lies behind a bright elliptical galaxy, I cannot separate the Spitzer flux density of the submm counterpart from that of the elliptical. Lensing from the foreground elliptical galaxy may also complicate this submm system. Since I can only give upper limits to the flux density at mid-IR wavelengths, I exclude it from the rest of the analysis in this chapter. Ours is not the first study to combine SCUBA and Spitzer data and I now summarize what has previously been discovered. Egami et al. (2004) looked at the coincidence of IRAC and MIPS detections with 10 submm sources from the 8-mJy SCUBA Survey (Scott et al. 2002; Fox et al. 2002) using the Lockman Hole East Spitzer images. Five out of 10 submm sources are considered to have a secure radio detection (la RMS of 4.8 /dybeam - 1, Ivison et al. 2002) and of these all are detected with IRAC and in the MIPS 24 /mi data (3a depth of 120 /dy, Egami et al. 2004). Using the same Spitzer data in the Lockman Hole East, Ivison et al. (2004) examined the Spitzer properties of MAMBO-selected galaxies and found counterparts for all 9 > 3a MAMBO galaxies within the Spitzer field. Frayer et al. (2004) obtained targeted SCUBA photometry of optically faint radio sources in the First Look Survey Verification (FLSV) field. Seven out of 28 sources are detected above 3a at 850 /mi and all of these 7 are detected at 24/xm (although note that one source is only detected at 2.7cr). In this thesis I present the largest sample of SMGs with statistically secure Spitzer counterparts. In star forming galaxies, X-rays can come from star formation, weak AGN activity (usually soft X-rays) and/or strong AGN activity (usually hard X-rays). I have used the main and 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 31 Figure 3.1: 'Postage stamp' images of submm counterparts, from left to right: ACS z850; IRAC 3.6/xm; MIPS 24/xm; and VLA 1.4 GHz. All images are 20 x 20 arcsec and the black circle is centred on the SCUBA position with a 8 arcsec radius. The smaller square, or circle, indicates the secure, or tentative, counterpart, respectively. 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 32 Figure 3 .1: (continued). Table 3.1: Multi-wavelength photometry of secure submm counterparts. SMMID z i 7 7 5 mag 53.6 54.5 5s.8 58.0 524 5l.4 P LIR (AB) (MJy) (MJy) (MJy) (^ Jy) (/Jy) (MJy) ( 1 01 2 L o ) GN13C 0.475 21.6 39.43 ± 1.24 39.94 ± 0 . 9 9 36.53 ± 1.49 100.28 ± 1.74 371.0 ± 10.4 45.4 ± 5 . 4 0.042 R/M 0.13 GN31 0.935 21.8 60.64 ± 1 . 2 4 44.83 ± 0.99 39.91 ± 1.49 34.64 ± 1.74 367.0 ± 6 . 4 16.3 ± 5 . 4 0.034 M 0.31 GN25C 1.013 22.8 92.74 ± 1.24 85.64 ± 0.99 75.42 ± 1.49 73.02 ± 1.74 724.0 ± 12.0 93.8 ± 12.9 0.020 R 1.1 GN26° 1.219 22.7 63.11 ± 1 . 2 4 72.06 ± 0.99 59.55 ± 1.49 75.32 ± 1.74 446.0 ± 5 . 1 194.3 ± 10.4 0.027 R 1.6 GN30 1.355 22.7 14.29 ± 0 . 8 9 14.61 ± 0.79 10.92 ± 1.28 12.80 ± 1.48 100.0 ± 5.9 19.9 ± 5 . 4 0.024 R/M 0.36 GN16 1.68 26.7 1 7 . 0 4 ' ± 0 . 8 9 25.40 ± 0.99 32.64 ± 1.49 26.25 ± 1.48 267.0 ± 7.7 361.9 ± 6 . 0 0.00093 R 5.2 GN17 1.72 27.7 53.87 ± 1.24 70:13 ± 0 . 9 9 71.54 ± 1.49 57.95 ± 1.74 710.0 ± 7 . 8 110.0 ± 11.0 0.0072 R 5.5 GN06 1.865 27.4 14.54 ± 0 . 8 9 19.41 ± 0.79 26.94 ± 1.49 20.07 ± 1.48 330.0 ± 7.6 178.9 ± 6 . 4 0.0077 R 7.2 GN07 a 27.8 14.24 ± 0 . 8 9 19.37 ± 0 . 7 9 26.05 ± 1.49 19.18 ± 1.48 20.0 ± 6 . 2 155.7 ± 6 . 3 0.0055 R 10.9 1.998 23.9 14.82 ± 0 . 8 9 18.67 ± 0 . 7 9 25.10 ± 1.49 18.58 ± 1.48 346.9 ± 8 . 8 55.10 ± 12.6 0.016 R GN03 (2.2) > 28 5.29 ± 0 . 5 5 8.27 ± 0 . 7 9 12.69 ± 1.28 14.56 ± 1.48 172.0 ± 6 . 8 29.1 ± 6 . 0 0.022 R/M GN10 (2.2) > 28 1.21 ± 0 . 3 9 1.96 ± 0 . 3 6 2.72 ± 0 . 8 8 5.11 ± 1.14 30.7 ± 5 . 4 33.1 ± 10.3 0.027 R G N U (2.3) > 28 9.71 ± 0 . 8 9 13.89 ± 0 . 7 9 19.56 ± 1.28 19.39 ± 1.48 145.0 ± 4 . 2 25.6 ± 5 . 4 0.016 R GN19 a 2.484 25.4 17.97 ± 0 . 8 9 23.30 ± 0 . 9 9 31.76 ± 1.49 28.95 ± 1.74 34.7 ± 1 0 . 8 39.0 ± 5 . 5 0.037 R 9.6 2.484 > 28 9.83 ± 0 . 8 9 14.51 ± 0 . 7 9 22.23 ± 1.28 22.51 ± 1.48 242.7 ± 1 0 . 8 38.0 ± 9 . 7 0.0071 R GN18 (2.5) > 28 2.53 ± 0 . 3 9 4.12 ± 0 . 5 1 6.66 ± 0.98 7.63 ± 1 . 1 4 79.9 ± 6.7 21.2 ± 5 . 5 0.050 R/M GN22 2.509 24.6 27.28 ± 1.24 30.70 ± 0.99 35.80 ± 1.49 27.50 ± 1.74 70.1 ± 5 . 2 65.4 ± 1 1 . 8 0.020 R 3.6 GN04 a 2.578 26.2 12. '32± 0.89 18.07 ± 0.79 29.49 ± 1.49 43.36 ± 1.74 302.8 ± 6 . 5 d 58.0 ± 6 . 3 0.049 R 8.0 > 28 14,92 ± 0.89 19.47 ± 0 . 7 9 27.93 ± 1.49 27.09 ± 1.74 31.5 ± 6 . 3 0.075 R GN05 2.60 24.9 12.63 ± 0 . 8 9 16.58 ± 0 . 7 9 21.79 ± 1.28 18.26 ± 1.48 215.0 ± 6 . 0 50.3 ± 15.4 0.034 R/M 6.5 GN12C 3.10 26.2 5.75 ± 0 . 5 5 7.89 ± 0.79 11.22 ± 1.28 13.84 ± 1.48 67.0 ± 9 . 0 103.2 ± 5 . 5 0.0056 R 19.5 GN20.2 3.90 24.7 3.87 ± 0 . 5 5 3.89 ± 0 . 5 1 6.06 ± 0.98 9.36 ± 1.14 30.2 ± 5.6 180.7 ± 8 . 4 0.028 R 14.0 GN20 3.955 24.4 6.79 ± 0 . 5 5 9.23 ± 0.79 15.93 ± 1.28 25.28 ± 1.48 68.9 ± 4 . 8 70.0 ± 16.3 0.0092 R 14.0 GN14 b 4.1 > 24.6 < 19.5 < 17.6 < 13.4 < 16.2 < 26.1 < 15.9 N/A GN39 a 1.996 23.4 37.94 ± 1.24 44.97 ± 0.99 53.33 ± 1.49 37.82 ± 1.74 537.0 ± 9 . 0 133.2 ± 13.0 0.024 R 5.6 1.992 24.9 9.16 ± 0 . 8 9 11.44 ± 0 . 7 9 16.06 ± 1.28 12.28 ± 1.48 225.0 ± 7 . 0 61.0 ± 9 . 6 0.036 R 2.4 GN40 (2.6) > 28 3.80 ± 0 . 5 5 6.23 ± 0 . 7 9 9.42 ± 1 . 2 8 15.86 ± 1.48 53.8 ± 6 . 0 615.0 ± 11.3 0.003 R a These submm sources have two radio/IRAC counterparts very close together, and therefore I list the multi-wavelength flux densities for both components. The counterpart is hidden behind a elliptical galaxy, and 1 quote the upper limits to the IRAC and MIPS flux densities - photometric redshift is from Dunlop et al. (2004). c These sources are also detected in 16 fj,m imaging of GOODS-N with the Spitzer IRS (Teplitz et al. 2005). d This MIPS source appears to be located between 2 IRAC sources. Table 3.2: Multi-wavelength photometry of tentative submm counterparts. SMM ID z. 1775 mag 53.6 54.5 55.8 58.0 524 5l.4 P LIR (AB) (MJy). (^y) (MJy) (MJy) (MJy) ( 1 01 2 L Q ) GN04.2 0.851 22.6 12.41 ± 0.89 9.28 ± 0.79 6.21 ± 0.98 4.76 ± 1.14 25.4 ± 5 . 2 17.8 ± 5 . 9 0.14R/M GN28 1.02 23.0 9.24 ± 0 . 8 9 6.54 ± 0 . 5 1 5.68 ± 0.98 4.35 ± 1.14 20.9 ± 5 . 3 90.3 ± 24.4 0.080 R 0.016 ,GN02C 1.32 23.8 65.42 ± 1.24 69.17 ± 0 . 9 9 54.51 ± 1.49 57.08 ± 1.74 265.0 ± 6 . 2 43.4 ± 6 . 1 0.11 R 1.4 • GN34 1.36 22.9 18.4 ± 0 . 8 9 18.7 ± 0 . 7 9 14.0 ± 1.28 18.3 ± 1/48 81.7 ± 4 . 1 < 23.6 0.065 M GN32 (.1.9) 27.8 6.95 ± 0.55 9.74 ± 0.79 10.78 ± 1.28 9.22 ± 1.14 128.0 ± 6 . 2 < 16.8 0.15M/I GN01 2.415 23.3 9.50 ± 0.89 12.97 ± 0 . 7 9 19.66 ± 1.28 26.88 ± 1.74 119.0 ± 6 . 0 26.7 ± 1 2 . 1 0.065 R/M 4.9 GN23 (2.6) > 28.0 6.45 ± 0.55 9.51 ± 0 . 7 9 13.44 ± 1 . 2 8 18.31 ± 1.48 55.9 ± 6 . 7 34.9 ± 6 . 1 0.10 R/M GN15 2.743 24.3 15.68 ± 0.89 18.74 ± 0.79 22.04 ± 1.28 18.95 ± 1.48 200.0 ± 6,0 < 16.2 0.10 M/I GN21 (2.8) > 28 3.95 ± 0.55 5.64 ± 0 . 5 1 9.29 ± 1.28 10.05 ± 1.14 46.3 ± 5 . 4 33.5 ± 5 . 6 0.11 R/M GN09 (2.9) .' > 28 3.97 ± 0.55 6.60 ± 0 . 5 1 . 10.56 ± 1.28 14.18 ± 1.48 49.8 ± 6 . 1 < 19.5 0.17 M/I ' GN24 2.91 24.7 5.94 ± 0.55 7.15 ± 0 . 5 1 10.50 ± 1.28 7.37 ± 1.14 75.5 ± 6 . 2 < 18.0 0.12 M/I GN37 3.190 23.1 6.20 ± 0 . 5 5 6.58 ± 0 . 5 1 8.45 ± 0 . 9 8 9.22 ± 1:14 30.5 ± 4 . 6 < 21.0 0.12 M/I c This source is detected in 16 imaging of GOODS-N with the Spitzer IRS (Teplitz et al. 2005). 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 35 supplementary catalogues from Alexander et al. (2003b) to look for any X-ray emission from the submm counterparts and attempt to determine whether the X-ray emission is due to AGN activity. I use the simple criterion that any hard X-ray detected counterpart is considered an X-ray AGN. Although this criteria is not as rigorous as the classification used in Alexander et al. (2003a) and Alexander et al. (2005b), in practice all of the AGNs in those studies were detected in the hard band. 11/21 (52%) of the secure counterparts are X-ray detected in any band and 4/21 (19%) are classified as X-ray AGNs. This AGN fraction is somewhat lower than that given in Alexander et al. (2005a, 38t\l%), although not inconsistent, given the small number statistics. The AGN fraction could also be affected by variations in the X-ray and submm sensitivities across the field and/or differences in the derived AGN fraction from a purely submm-selected sample to that extrapolated from spectroscopically identified SMGs. Both results are consistent with the majority of SMGs having their IR luminosity predominantly powered by star formation and not AGN activity. Throughout the figures in this chapter, I denote submm counterparts which are classified as X-ray AGN (i.e. hard X-ray detected) with a cross symbol. I discuss the AGN signatures more in Section 3.6 and Chapter 4. 3.1.1 Redshifts Table 3.1 and Table 3.2 also provides redshift estimates for each source. When available I list the spectroscopic redshift, otherwise I list the photometric redshift. Details on the spec-troscopic and photometric redshift estimates were given in Section 2.2.2. The spectroscopic redshifts were obtained from publicly available redshift surveys (see Section 2.2.2). Specific details on the UV and optical spectra of SMGs and their diagnostic power for interpreting the nature SMGs can be found in Chapman et al. (2005) and Swinbank et al. (2004). Fig. 3.2 shows the redshift distribution for all SMGs in my sample with counterparts (including those without a bright radio counterpart). The three SMGs without a unique counterpart (see Section B.3) are not included in this plot. The redshift distribution of this sample is consistent with the pre-vious redshift distribution presented in Paper III and also with the sample of 73 radio-detected, 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 36 1 0 -0 1 2 3 Redshift Figure 3.2: Redshift distribution of SMGs. The solid histogram shows the redshift distribution (including photometric and spectroscopic redshifts) of all submm counterparts in GOODS-N (including those without a bright radio counterpart) while the dashed histrogram shows the redshift distribution from Chapman et al. (2005) for a sub-sample of SMGs with radio detections and optical spectroscopic redshifts. 3.1. MULTI- WAVELENGTH IDENTIFICATIONS 37 spectroscopically-confirmed SMGs in Chapman et al. (2005). Note that 7/21 of the secure counterparts are also in the Chapman et al. 2005 sample and have spectroscopic redshifts. The median redshift for the secure counterparts with optical spectroscopic or photometric redshifts (21/35 sources) is 2.0 (interquartile range 1.3-2.7), while the median redshift for all counter-parts, including those with only IR photometric redshifts (33/35 sources), is 2.2 (interquartile range 1.4-2.6). The large fraction of submm sources with counterparts below z = 4 in this sample, which contains sources covering S850 = 2-20 mJy, seems to imply that there is not a large population of 850 /im-selected galaxies at z > 4. Even if I conservatively assume that one of the secure counterparts and two of the tentative counterpart are incorrect identifications and that they, and the two sources with multiple counterparts, are really at high redshift, this still constrains the fraction of submm sources at z > 4 to < 14% (< 5/35). There is evidence which suggests that galaxy samples selected at longer wavelengths (~ 1 mm) have a higher fraction of sources at z > 4 (Eales et al. 2003). 3.1.2 New radio detections Using the published Richards (2000) 5<r 1.4 GHz catalogue, I found radio counterparts for 37% of this sample (see Paper II and III). I find that all of the VLA 1.4 GHz radio counterparts pre-sented in Paper II and Paper III are confirmed in the new radio reduction and at similar flux levels. In addition, I find 18 new radio sources above 3a in the new radio reduction within the search radius of a submm source, all of which are detected with IRAC and/or MIPS. The sep-aration between the radio and IRAC positions of the submm counterparts is < 0.5 arcseconds. All of the secure counterparts are radio-detected and, including the tentative identifications, 74% (26/35) of submm sources in GOODS-N have a radio counterpart using the new VLA image. Note that 10% of this improvement is simply due to lowering the signal-to-noise ratio threshold to 3a. Taking the 20 secure radio counterparts (excluding GN14), 8/20 are only present in the new radio reduction. Radio flux densities for all of the sources detected above 3a are listed in Table 3.1 and Table 3.2, and in the absence of a detection I list the 3a upper limit. Figure 3.3: Submm-radio positional uncertainty in GOODS-N. The solid histogram is the cu-mulative distribution of radial offsets between the submm and radio positions for the 21 secure radio counterparts in GOODS-N (see Table 3.1). The dashed curve is the best-fit radial offset distribution assuming a Gaussian with ar — a x ( F W H M / S N R ) s u b m m , with a = 1.0. I have used the average submm SNR which has been corrected for flux deboosting. 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 39 Our confidence in these new radio detections is considerably bolstered by the fact that several of the radio sources correspond to the optical galaxies suggested as counterparts in Paper III, based only on optical/near-IR colours. The new radio counterparts that lie just above the detection threshold in this improved VLA reduction are consistent with the results of Paper III, which indicated that the radio-detected and radio-undetected submm sources are not drawn from very different populations. Extrapolating from this, I suggest that the remaining submm sources which are not currently detected in the radio are likely to lie just below the detection threshold. This is supported by the fact that several of them appear as ~ 2a features in the radio image. Taking the 21 secure radio counterparts, I can investigate the distribution of offsets be-tween the radio and submm position. Ivison et al. (2005) found that the distribution of offsets for robust submm sources detected at both 850 am and 1.2 mm was approximately Gaussian with a ~ F W H M / S N R , where FWHM is the Full Width at Half Maximum. In Fig. 3.3, I plot the cumulative distribution of radial offsets between the radio and submm positions for the 21 secure counterparts. The dashed curve is the expected radial offset distribution assum-ing a Gaussian with ar = a x ( F W H M / S N R ) S U B M M , with a = 1.0. I have used the average submm SNR which has been corrected for flux deboosting. Despite clear systematic differ-ences, a KS test fails to find a significant difference between these two distributions. The value of a found here is somewhat larger than what might be expected if centroiding errors dominate (e.g. Hughes et al. 1998). However, I expect astrometry shifts to be important in the hetero-geneous GOODS-N super-map. A uniform submm survey where all data were taken with the same chopping pattern, such as the SCUBA HAlf Degree Extragalactic Survey (SHADES, Mortier et al. 2005), would be expected to provide a better test of the above relation (see Ivison et al. 2007). 3.1.3 Double radio sources When deep radio observations are present in a submm field, around 10% of the of the SMGs have two radio sources within the search radius (Ivison et al. 2002; Chapman et al. 2005). I 3.1. MULTI- WAVELENGTH IDENTIFICATIONS 40 find a consistent rate in the GOODS-N field. With the new radio reduction of the 1.4 GHz VLA data, I find that 4/35 submm sources have two radio counterparts in the search radius. Note that these additional radio sources were not present in the original Richards (2000) radio catalogue. The probability of having two radio sources of any brightness within 8 arcseconds of a submm source at random is less than 1%, and therefore it is likely that the radio sources are associated with each other. Three of these double radio pairs (GN04, GN07 and GN19) are very close (< 3 arcseconds) and two have an optical counterpart for only one of the radio sources, meaning that I am unable to get a redshift estimate of the second radio source (although all of the additional radio sources have a counterpart detected in IRAC and/or MIPS). As I discuss in the Appendix, if I assume both radio components are at the same redshift, then the separation between them is ~ 20kpc, indicating that these could be interacting or multi-component systems. The IRAC data for these objects also support the notion that the radio sources are both at the same redshift and contributing to the total submm flux (see Section 3.3.3). Therefore I conclude that the double radio sources associated with GN04, GN07 and GN19 represent interacting systems, and hence I use the sum of the radio and mid-IR flux from the two components when determining the global multi-wavelength properties of the submm source. For the other double radio source in this sample, GN17, both radio components have an infrared and optical counterpart and so I can estimate photometric redshifts, which I find to be different (z p h 0 t =1.7 and 1.2). In the next section, I describe how I fit the MIPS 24 fim flux and redshift to a suite of SED templates to determine the likely contributions to the 850 fxm flux from each radio source. I found that, for GN17, the second radio/24 pirn source is likely to be subdominant for the submm emission. GN38, which does not make the final submm catalogue due to the probability of flux boost-ing, has three radio sources within the search radius, two of which look to be associated. All of these double radio sources in GOODS-N are discussed individually in the Appendix, along with notes for the other submm sources. GN39 from the 2006 super-map also has two radio counterparts, both confirmed to lie at the same redshift (see Appendix B). 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 41 3.1.4 MIPS 24 /im detections The FWHM of the MIPS PSF is 5.7 arcseconds, which is significantly better than the 14.7 arcsecond resolution obtained with SCUBA. In addition, detection of sources in the MIPS 24 um image was carried out using the IRAC source positions as a prior, which facilitates matching between IRAC and MIPS images. There are a few cases where blending in the MIPS image is an issue, and I have dealt with these on a case by case basis to make sure I am getting the best estimate of the 24 //m flux at the position of the counterpart. As a result, all of the candidate 24 /tm counterparts within 8 arcseconds of the submm sources have a single IRAC identification. The measured FWHM of the IRAC PSF is < 2 arcseconds (Fazio et al. 2004), which is sufficient to identify unique optical counterparts to IRAC sources in most cases. Using no prior knowledge at any wavelength, I have explored submm counterpart identi-fication using only the MIPS 24/mi image. I find 10/21 secure counterparts have P < 0.05 using only the MIPS 24 /xm catalogue. These secure MIPS counterparts are all brighter than 70/dy (> 14CT) at 24/tm. While this is only 30% (10/35) of the whole submm sample, it demonstrates that relatively shallow MIPS observations (5a ~ 70 /dy) can be useful for iden-tifying secure submm counterparts in the absence of radio data. Deep radio observations may be difficult to obtain in some fields, due to their location on the sky or the presence of bright quasars in the field which drive the dynamic range, and MIPS observations can be used as an alternative. In situations where one is fortunate enough to have deep radio and infrared observations, the high overlap between 24 /tm and submm sources provides an independent confirmation of the robustness of the radio counterparts. Roughly half (17/35) of the submm sources in GOODS-N have more than one > 3a 24 /tm source within the search radius. SCUBA observations are limited by confusion noise; there are many faint submm sources ( < 1 mJy) which get blended in the large JCMT beam. Hence it is possible that when there are multiple 24 /tm sources, they are all contributing to the submm flux1. The important question is whether more than one of the mid-IR sources is making a 'It is also possible that none of the 24 um source are associated with the SMG. This possibility is dealt with in the counterpart identification procedure described in Section 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 42 0.5 n i r~ i 1 r n 1 r 0.4 Fraction with no MIPS in search radius: Submm positions =0.03 Random positions =0.5 0 .3 o 0.2 0.1 max S24/total S24 0.2 0.0 ure3.4: Multiple MIPS 24 pm sources within one SCUBA 850 pm beam. I have calcu-lated the ratio of the flux density of the brightest 24 pm source to the total flux of all 24 pm sources within a search radius of 8 arcseconds for the submm sources (shaded region) and for 1000 random positions in GOODS-N (dark line, slightly offset in the -^direction for clarity). These distributions have been normalized by the total number of sources in each sample (including sources with no MIPS in search radius). 3.1. MULTI-WAVELENGTH IDENTIFICATIONS 43 significant contribution to the submm flux? I investigate this in two different ways. In the first test, I assume that the brightest 24 /xm source within the SCUBA 850 /xm beam will be associated with the submm source. While this may not always be the case, it is statisti-cally the most likely. Note that I are not saying that I require the brightest 24 /xm source to be the correct submm counterpart. In Fig. 3.4,1 plot the distribution of the ratio of the total flux density of the brightest 24 /xm source to the total flux of all 24 /xm sources within a SCUBA beam (search radius of 8 arcseconds, see also Sajina et al. 2006). This ratio provides a measure of the importance, in terms of 24 /xm flux, of any additional sources in the search radius. A ra-tio close to one indicates that the submm emission is more likely dominated by a single 24 /xm source (under the assumption that 24 /xm and 850 /xm fluxes of sources are related). The results for the submm sources in GOODS-N are shown as the shaded region in Fig. 3.4 and, for com-parison, the dark curve is the distribution for 1000 random positions in GOODS-N. Roughly 50% of the random positions do not have any 24 /xm sources within the search radius, whereas fewer than 3% of the submm positions have no 24 /xm sources within the search radius. Find-ing that 50% of random positions contain at least one 24 /xm source does not mean that half of the associations around SCUBA sources are false associations since 60% of SCUBA sources have more than one 24 /xm source within the search radius. I have calculated the probability of finding a correct 24 /xm counterpart around a SCUBA source given the probability of randomly finding a 24 /xm source in the search radius. I conclude that 94% of the SCUBA sources are correctly associated with a MIPS 24 /xm counterpart within the search radius. In other words, only 3% of the SCUBA sources are incorrectly associated with a 24 am source2. Furthermore, of those positions which have at least one 24 /xm source in the search radius, both distributions in Fig. 3.4 are consistent with being dominated (> 50% of flux) by a single 24 um source the majority of the time. Only about 10% of the submm sources have more than one 24 /xm galaxy with the additional source(s) contributing up to ~ 50% of the total MIPS flux in the SCUBA beam. Whether or not this additional source (or sources) contributes significantly to the submm flux depends not only on the MIPS flux, but also the SED and redshift, which I investigate next. 2Note that this spurious association rate is based solely on the MIPS 24 um counterparts; when I consider the radio data, the robustness of the associations becomes even stronger (see Section 3.1). 3.2. SED TEMPLATES 44 A second estimate of how much the multiple 24 /xm sources within a SCUBA beam are contributing to the total submm flux, I have fit the 24 /xm flux of both (or three in some cases) sources at their redshifts (spectroscopic or photometric, see Section 3.1.1) using a range of SEDs to determine the range of 850 /xm flux expected from each 24 /xm source. Templates used include Chary & Elbaz (2001, hereafter CEOl) and Dale & Helou (2002) star forming galaxy SED models, covering a wide range of luminosities, as well as an AGN template, Mrk231 (mid-IR spectrum from Rigopoulou et al. 1999, spliced with a fit to the Infrared Astronomical Satellite, IRAS, photometry). I am then able to assess what combinations of sources are likely to have produced the observed 850 /xm flux of each submm system. Note that in this procedure I am not assuming that the brightest 24 /tm sources within the beam is the submm counterpart, since the fit will depend on the 24 /xm flux and the redshift. Of course this procedure can only be a general indication, since I have not yet measured the entire SED shape of SMGs. However, I have been quite liberal in using a fairly wide range of templates, and it would take quite unusual SEDs to significantly change the results. The conclusion is that, for the majority of the cases, one of the 24 /xm sources is the dominant submm emitter (i.e. more than half of the submm flux comes from it) and most often it turns out to be the brightest 24 /xm source in the search radius. In most cases, the second 24 /xm source in the SCUBA beam is usually faint and at low redshift, and therefore not capable (assuming the wide range of SED templates) of producing the observed submm flux. See the Appendix B for detailed comments on individual sources. 3.2 SED templates Our goal is to use the multi-wavelength identifications of SMGs to study their SEDs. In this section I discuss several SED models which I use in interpreting the observed SEDs of SMGs. The mid-IR portion of the SED of galaxies is sensitive to thermal dust emission, AGN power-law emission (Clavel et al. 2000) and polyaromatic hydrocarbon (PAH) emission fea-tures (Hudgins & Allamandola 2004). The MIPS 24 /xm filter is narrow enough that strong spectral features have a significant effect on the photometry. In particular, the 9.7 /xm silicate 3.2. SED TEMPLATES 45 absorption feature has been found to be very strong in IR-luminous galaxies (Dudley & Wynn-Williams 1997; Spoon et al. 2002, 2004a; Armus et al. 2004). The strength of this, and other, mid-IR spectral features in different galaxy types is still not completely understood, but deep InfraRed Spectrograph (IRS) observations of galaxies at low and high redshift is beginning to provide a more complete picture (Armus et al. 2004; Yan et al. 2005). Most models of the entire IR SED prior to Spitzer did not properly characterize the mid-IR spectrum of galaxies, in particular the effects on the broadband fluxes of varying PAH emission features and strong silicon absorption. CE01 provide SED templates for galaxies as a function of infrared luminosity which reproduced the IR and submm observations of nearby galaxies at that time. More recently, Draine (2003 and references therein) has created a carbonaceous-silicate grain model which includes absorption features, such as the 9.7 /im silicate feature. This model provides wavelength-dependent extinction, which reproduces the observed obscu-ration of starlight and infrared emission. In order to account for possible effects on the mid-IR photometry from strong mid-IR spectral features, I have created a combined model which in-cludes the CE01 SED templates and additional extinction using the Draine (2003) grain models and absorption cross-sections, allowing for the amplitude of the additional extinction to vary. Hereafter I refer to these models as modified CE01 templates. At submm wavelengths, the SED is often simplified to a greybody, whose shape is charac-terized by the temperature and ft (e.g. Halpern et al. 1988; Blain et al. 2002), which are related to the wavelength of the far-IR peak and the slope of the SED, respectively. The dust model for CE01 was created from four components, at 18 K, 40 K, 300 K plus the PAH emission, and it is the relative contribution of each which varies as a function of luminosity. For the pur-poses of this paper, a cool CE01 model has a higher contribution from the lower temperature components than the warm CE01 model, and therefore peaks at longer wavelengths. A rough estimate of the dominant temperature of the CE01 templates can be determined by considering the wavelength of the far-IR peak. In CE01, the temperature increases slowly with luminosity, due to the local observed luminosity-temperature relation (see CE01 and references therein). However, at high redshift, little is known about the temperatures of IR-luminous galaxies, and so I have fit for luminosity (amplitude of SED) and temperature (shape of SED) separately in 3.3. INFRARED PROPERTIES 46 this chapter. The radio portion of the CE01 models is estimated by assuming the local radio-IR correla-tion. Therefore if the models fit my data I can say that the correlation holds for this sample. Arp220 is the best studied local ULIRG (Gonzalez-Alfonso et al. 2004; Spoon et al. 2004b) and is often compared to high redshift SMGs (e.g. Chapman et al. 2005). However, it is not representative of the entire sample of local ULIRGs, as its IR SED is cool compared to most IRAS-bright ULIRGs (Farrah et al. 2003), and has one of the most extreme mid- to far-IR slopes. I compare the submm results to the observed SED of Arp220 interpolated from pho-tometry listed in the NASA/IPAC Extragalactic Database. As I show in Section 3.4, I find evidence that many high redshift submm sources have dust temperatures as cool as or cooler than Arp220. For completeness, I also compare to the observed SED of the luminous AGN Mrk231, where the mid-IR spectra from Rigopoulou et al. (1999) has been spliced with a fit to the IRAS photometry. 3.3 Infrared properties With multi-wavelength identifications and redshift estimates for the majority of this submm sample, I now explore the infrared properties of SMGs including their mid-IR-to-submm flux ratios, their mid-IR colours, and, finally, their full SED from near-IR to radio wavelengths. 3.3.1 Probing dust emission In high redshift SMGs, the 850 /im window is sensitive to cool dust close to the peak of the far-IR dust emission. On the other side of the far-IR peak, the mid-IR regime is also sensitive to thermal dust emission. However, as discussed in Section 3.2, there are possible additional contributions, including AGN power-law emission and PAH emission features. As one moves to higher redshift, the 850 pm filter climbs the smooth grey-body dust peak, while the 24 /im filter passes through various spectral features and also changes depending on the intensity of the AGN emission. For this reason, it would be surprising if there was a tight correlation 3.3. INFRARED PROPERTIES 47 Redshift Figure 3.5: The S ^ / S ^ o r a t ' ° a s a m n c t I 0 n °f redshift for submm sources in GOODS-N. The open circles show the average values of the submm sources in four redshift bins, with roughly equal numbers of sources in each. The dotted curve is a cool SED template from CE01 models, the solid curve is the same CE01 template with ad-ditional extinction from Draine (2003), the dashed curve is the observed SED of Arp220 and the dash-dot curve is Mrk231, an IR-luminous AGN. The shaded re-gion represents the redshifts where the 9.7 /xm silicate feature passes through the 24 /xm passband. The Su/S85o ratios for the submm sources suggest that they have higher levels of extinction, especially in the mid-IR, than the local galaxies which had been used to construct the CE01 models. 3.3. INFRARED PROPERTIES 48 between S24 and Sgso as a function of redshift, although there should be a general trend. In Fig. 3.5 the SW/SW) ratio is plotted as a function of redshift for the sample of secure submm counterparts in GOODS-N. As listed in Table 3.1, 17/21 secure counterparts have reliable redshifts, with more than half of these being spectroscopic. Due to the many factors influencing the 24/im photometry for individual sources (as discussed in Section 3.2), I also plot the average ratio in 4 redshift bins as the open circles. For comparison, I overplot several galaxy SED models: Arp220; Mrk231; a cool CE01 template; and the same CE01 template with additional extinction (see Section 3.2). The shaded region in Fig. 3.5 denotes the area where the 9.7 pm silicate feature passes through the 24 pm passband. The cool CE01 model (dotted curve) is a poor fit to the data and overpredicts the SW/Ssso ratio at all redshifts. The solid curve in Fig. 3.5 is a modified CE01 template (see Section 3.2), where the additional extinction is of the order Av ~8 magnitudes (in Fig. 3.8, I will show that the total mid-IR extinction in this modified CE01 model is roughly equal to the extinction observed in Arp220 and therefore this model is entirely reasonable). Overall, the observed S24/S850 ratio of the submm sources as a function of redshift appears consistent with the modified CE01 model. It also agrees reasonably well with the ratios expected for a redshifted Arp220 SED, particularily at higher redshifts. Submm sources classified as having an AGN due to the presence of hard X-rays do not have substantially higher S^/Sgso ratios, as they would if the mid-IR emission was dominated by the AGN. Overall, Fig. 3.5 suggests that the submm sources have higher levels of extinction, especially in the mid-IR, than the local ULIRGs which had been used to constrain the CE01 models. In particular, the silicate absorption feature at 9.7 pm may be attenuating the 24 pm flux. Based on 16/24 pm colours, Kasliwal et al. (2005) predict that roughly half of ULIRGs at z ~ 1-2 might be missed at 24 pm due to silicate absorption. As I discuss in Section 3.4, 1 require a cooler template when fitting the 5*850/>Si.4GHz ratio as a function of redshift. However, decreasing the temperature of the template is not enough to account for the low S24/SS50 ratios seen in Fig. 3.5, and therefore I require the additional extinction in the mid-IR. Another class of star forming galaxy at similar redshifts to the SMGs which has been well studied in the IR are the BzK galaxies (Daddi et al. 2004). These galaxies are selected on 3.3. INFRARED PROPERTIES 49 the basis of their stellar SEDs (through if-band fluxes and optical-IR colours), rather than based on longer-wavelength (mid-IR, submm, radio) properties that trace dust heated by star formation, and therefore they are subject to different selection effects than SMGs. While SMGs at 1.4 < z < 2.5 with a measurable K magnitude generally do satisfy the BzK criterion, the majority of submm counterparts at these redshifts have K > 21.9, and therefore there is very little overlap between galaxies selected at 850 pm and bright BzK galaxies. Only one (out of 21) of the secure submm counterparts qualifies as a K < 21.9 BzK galaxy, but this increases to four if I also consider consider K > 21.9 BzK galaxies. Daddi et al. (2005) found that 82% of the GOODS-N BzK galaxies with K < 21.9 (AB magnitudes) are detected at 24 pm, with an average flux density of about 125 pJy. At z ~ 2, they have typical SW/Ssso ratios of 0.16 (Daddi et al. 2005), which is larger (by a factor of 4) than that of most of the submm sources at z ~ 2. This might indicate that their SEDs peak at shorter wavelengths. However, a difference in AGN contribution and PAH features in the IR luminosity could also cause this effect. BzK galaxies (with K < 21.9) are ~ 5 times less luminous in the infrared than typical SMGs (see Table 3.1 and Chapman et al. 2005). The higher luminosities of 850 pm selected galaxies (and their lower surface densities) means that, at best, submm sources represent the most extreme BzK galaxies (Daddi et al. 2005; Dannerbauer et al. 2006). 3.3.2 Probing stellar emission At the redshifts of these SCUBA sources, the IRAC channels are sampling the rest-frame near-IR, which is sensitive to direct emission from stars. If these galaxies are not dominated by an AGN at these wavelengths then the IRAC colours should show an indication of the 1.6 pm local peak of the fv stellar SED. In Fig. 3.6,1 plot an IRAC-based colour-magnitude diagram, specifically S 5 . 8 / S 3 . 6 as a function of S5.& for the submm sources and the GOODS-N field galaxies. The redshifts of the submm sources are indicated by the symbol (black for z < 1.5, grey for z > 1.5, open for unknown redshift). It can been seen that z ~ 1.5 provides a clear transition in this IRAC colour for the submm sources, due to the filters moving over the stellar peak. Since the 3.6 pm channel samples the peak of the stellar light at z ~ 1.3, a galaxy at 3.3. INFRARED PROPERTIES 50 CO x CO "So o 0.6 0.4 0.2 0.0 -0.2 -0.4 0.0 0.5 1.0 1.5 l0g(S5.8pm(riTy)) 2.0 Figure 3.6: IRAC-based colour-magnitude diagram for GOODS-N. Larger symbols are the submm sources, while the small black dots are field galaxies which are detected in both IRAC bands. Filled circles and squares denote submm sources with spectro-scopic and photometric redshifts, respectively. Crosses are sources which contain an AGN, as indicated by the presence of hard X-rays. The black solid symbols are at low redshift (z < 1.5), the grey solid symbols are at high redshift (z > 1.5) and the black open symbols have unknown redshift. The dashed lines indicate the re-gion where S5.s > 6 and S5.8/S3.6 > 1.3, within which 11/11 of the radio-detected submm counterparts are confirmed to be at z > 1.5 - the low density of sources in this region of the diagram makes this colour-magnitude cut a useful tool in assign-ing counterparts to SMGs. 3.3. INFRARED PROPERTIES 51 4 £ - 3 CO ~o CD i o to 2 E o o £ 1 o 1 1 I 1 t I I 1 I I I I I 1 I I I I 1 I I I I 1 I I I I I | 1 1 1 1 1 1 1 1 1^  / / / / / / / / / & / / / \ / / / / / / — • / / / > i i 1 i i i i / / / r , & -/ / ' i r 1 i 1 1 0 1 2 3 Optical spectroscopic redshift Figure 3.7: IR photometric redshift accuracy. I plot the (model independent) IR photometric redshift, as determined only from IRAC and MIPS photometry, versus the known optica] spectroscopic redshift for 10 secure submm counterparts in GOODS-N. Crosses indicate counterparts which are classified as X-ray AGN due to the pres-ence of hard X-rays. Spitzer photometry can be used to constrain the redshift for sources which are faint or undetected in the optical. While the IR photometric red-shifts are not as accurate as the optical photometric redshifts (see Pope et al. 2005), they are good to Az < 0.4 for 8/10 sources. 3.3. INFRARED PROPERTIES 52 redshifts greater than this would move upwards in this diagram until z ~ 2.6 at which time the 5.8 pm channel passes over the peak the same galaxy would move towards the left of this diagram. The sources which lack redshift information are all consistent with the colours of the higher redshift submm sources. These sources are all faint in the optical, which indicates that they are either at high redshift or have a very high dust extinction (or both). The dashed lines indicate the region where S5.8 > 6 and S5.8/S3.e > 1.3, within which 11/11 of the radio-detected submm counterparts are confirmed to be at z > 1.5. This IRAC colour-magnitude cut can be used to help identify submm counterparts, since there is a low probability of landing in this region of the diagram at random. The SMGs are bright at 5.8 pm for a given colour compared to the field galaxies. Since 5.8 pm samples near the peak of the stellar SED in SMGs, this could mean that the SMGs are on average more massive (e.g. Borys et al. 2005). The presence of an AGN could complicate matters, and so I need to check for any bias. Sources which are detected in the hard-band X-ray image are indicated in Fig. 3.6 by a cross, and there seems to be no strong colour distinction between the sources with and without a hard X-ray detection, although the former are generally brighter at 5.8 pm. BzK galaxies (with K < 21.9) are also bright at 5.8 pm, indicating that they are massive. However, their S5.8/S3.Q colour is slightly bluer than that of the SMGs of similar redshifts. This may indicate that, while this IRAC colour is primarily sensitive to redshift, the Ss.g/Sss colour of SMGs may also be affected by an AGN or dust obcuration. 3.3.3 Spitzer p h o t o m e t r i c r e d s h i f t s Expanding on the separation of IRAC colour with redshift seen in Fig. 3.6,1 have attemped to estimate IR photometric redshifts for this sample using only the IRAC and MIPS photometry. The idea is to solve for the redshift as a linear combination of the logarithm of the flux density at the mid-IR wavelengths (see Equation 3.2), in order to provide a completely model inde-pendent estimate of the redshift (Connolly et al. 1995; Sajina 2006). I use the secure submm counterparts with known optical spectroscopic redshifts to solve for the coefficients and then apply the formula to sources with unknown redshifts. This is particularily useful for the submm 3.3. INFRARED PROPERTIES 53 counterparts which are faint or undetected in the optical and thus lack an optical photometric redshift estimate. The redshift is calculated using the following formula: zm = a + b- log(53.6) + c • log(55.8) + rf.log(<S8.o) + e - l o g(5 2 4 ) . (3.2) Using the 10 secure submm counterparts with spectroscopic redshifts, I fit for the five coefficients which give the lowest RMS when compared to the spectroscopic redshifts. I have performed this fit for various combinations of these flux densities, (including functions without the logarithm), and found this combination to give the lowest RMS for this sub-sample of galaxies (although it is by no means unique, and different coefficients could also give low RMS). Using the 4.5 /xm flux density instead of (or as well as) the 24 /mi flux density resulted in an RMS which was higher, and a coefficient for the 5*4.5 term which is close to zero. This is not surprising because 4.5 /xm samples the peak of the stellar light distribution at z ~ 1.8, and since I am not fitting for any sources near this redshift, the inclusion of this parameter does not significantly improve the redshift estimate. Fig. 3.7 shows the accuracy of the IR photometric redshift for these 10 submm sources. The RMS dispersion of the redshift errors for these SMGs is a(Az/(l + z)) = 0.07, using values of 3.3, -2.5, 4.6, -1.1 and -1.4 for a, b, c, d and e, respectively. For the submm counterparts which lack an optical photometric redshift estimate I list the zm in Table 3.1 and Table 3.2 in brackets. While these IR-derived redshifts are not as accurate as optical photometric redshifts for SMGs (Paper III), they do provide an accuracy of Az < 0.4 for 8/10 secure counterparts with spectroscopic redshifts. This is not accurate enough for exact determination of physical parameters, but it can be useful for separating the high redshift and low redshift sub-populations of SMGs, particularly for sources where the optical counterpart is invisible or ambiguous. This technique only works for particular sub-samples of galaxies (i.e. Equation 3.2 does not apply to other galaxy samples), since I am essentially assuming that the SEDs have a similar shape at these wavelengths (i.e. fitting for the coefficients using all spectroscopically confirmed IRAC sources in GOODS-N would give less accurate IR pho-3.4. SPECTRAL ENERGY DISTRLBUTIONS OF SMGS 54 tometric redshifts for the submm sources). However, this technique will likely improve with larger submm samples with known redshifts. As we learn more about the range of SEDs for SMGs, then photometric redshift estimators using the entire wavelength range should be able to achieve much better results. 3.4 Spectral energy distributions of SMGs The addition of the IRAC and MIPS photometry to the multi-wavelength dataset for SMGs provides powerful constraints on the shape of the SED, the nature of the power source, and the source redshift. In Fig. 3.8,1 plot composite rest-frame SEDs. All photometry points from the near-IR through to the radio are plotted, with each colour indicating a different passband. Error bars on the photometry are typically much smaller than the scatter among the sources, and have been omitted for clarity. Note that I have only included the submm sources with spectroscopic redshifts, since I would like to isolate the scatter due to different SEDs and/or luminosities; if I also include the sources which have photometric redshifts, it does not qualitatively change these figures, but the scatter is increased. By minimizing the scatter (RMS of the logarithm of the 24 /xm, 850 /xm and 1.4 GHz flux densities relative to the model), I find the best-fit modified CE01 template (see Section 3.2). This model (solid curve in Fig. 3.8) has an IR luminosity of 3 x 10 1 2LQ. The additional extinction supplied by the Draine (2003) models amounts to Ay ~ 8 magnitudes. In the top panel, I plot the rest-frame composite SED in physical units. In the middle, and bottom, panels, the SED is normalized to the submm and radio flux densities, respectively, to match the modified CE01 template. In the top panel, the dash-dot curve is a CE01 model with the same luminosity but a shorter peak wavelength. The dashed curve in the middle and bottom panels is the observed SED of Arp220. The three panels of Fig. 3.8 can be used to assess how well this model fits the submm sources using these different normalization approaches, as I discuss below. In this plot it can been seen that high redshift SMGs appear to have different SED shapes from those observed locally for galaxies of similar luminosities. A major contribution to this difference could be the temperature of the dust, but it is important to realize that this just means 3.4. SPECTRAL ENERGY DISTRLBUTIONS OF SMGS 55 22 3 X 3 —I u N 15 c 10 1.0000 0.1000 0.0100 0.0010 0.0001 0.01 I i I inij—I I 11 I I I I | — I I I mil—i i i nm—i i 11 > 1111 I I I —I I I —I I I mil|—I I I — I I I 10° 101 102 103 1 04 10s Rest wavelength (um) Figure 3.8: Composite rest-frame SEDs for the GOODS-N submm sources. See the text for complete description of this figure. 3.4. SPECTRAL ENERGY DISTRR3UTIONS OF SMGS 56 an SED peaking at longer wavelengths. The best-fit submm SED (solid curve in Fig. 3.8) peaks at ~ 100 u,m (T ~ 30 K), while a typical local ULIRG template from CE01 (dash-dot curve in top panel of Fig. 3.8) of the same luminosity peaks at ~ 85 (T ~ 34 K). The same point is illustrated in Fig. 3.9, where I plot the SSSO/SI^GHZ ratio as a function of redshift. Again, the submm counterparts are inconsistent with the warm local ULIRG template and fit better to the cooler template3. A cooler average temperature for the SMGs as compared to local ULIRGs of the same luminosity is consistent with predictions from hierarchical galaxy formation models (Kaviani, Haehnelt, & Kauffmann 2003). Similar results are also seen in Blain et al. (2004) and Chapman et al. (2005) for a large sample of 73 bright radio-detected SMGs. Using the redshift, radio and submm flux, Chapman et al. (2005) found that the median dust temperature of their sample is lower than that estimated for local IRAS 60 u.m galaxies of the same luminosity. The SCUBA Local Universe Galaxy Survey (SLUGS, Dunne et al. 2000; Dunne & Eales 2001) measured the submm fluxes of galaxies from the 7/MS-bright galaxy sample. They find an average dust temperature of 35.6 ± 4.9 K for f3 — 1.3 ± 0.2. However, their sample is less luminous than the blank sky SMGs, with IR luminosities in the range 10 1 0 L o to 3 x 1 0 n L o . Farrah et al. (2003) fit for the dust temperatures for a sample of local (U)LIRGs including faint IRAS galaxies, and find a median temperature of 32 K for sources with an average IR luminosity of ~ 10 1 2 L©. This is a similar dust temperature found for SMGs however these local galaxies are 6 times less luminous than SMGs on average. My results imply that the local luminosity-temperature relation does not appear to apply to high redshift SMGs. This result emphasizes the existence of strong selection effects, both locally and at high red-shift. Our knowledge of local ULIRGs is not immune to selection effects, since it is dominated by results from IRAS which may preferentially select galaxies with warmer SEDs (peaking at shorter wavelengths). Similarily at high redshift, 850 um imaging preferentially selects galax-ies which peak at longer wavelengths, and therefore comparing the two becomes problematic. 3 An additional follow-up study of this sample of SMGs using deep Spitzer 70 am imaging, which samples the short wavelength side of the far-IR dust peak, has confirmed the need for cooler dust temperature in SMGs (see Huynh et al. 2007), 3.4. SPECTRAL ENERGY DISTRIBUTIONS OF SMGS 57 T — i — . — r Redshift Figure 3.9: The £85o/<Si.4GHz ratio as a function of redshift for submm sources in GOODS-N. Filled circles and squares denote submm sources with spectroscopic and photomet-ric redshifts, respectively. The crosses denote sources which contain an AGN, as indicated by the presence of hard X-rays. The open circles are the average val-ues of the submm sources in four redshift bins with roughly equal numbers of sources in each. The solid curve is a cool SED template from CE01 models, while the dashed curve is a warmer CE01 template and the dash-dot curve is Mrk231, an IR-luminous AGN. The average S850/SIAGHZ r a t * ° ^ o r t n e S U D m m sources suggests that they have SEDs which are cooler, particularily those sources at lower redshifts. 3.4. SPECTRAL ENERGY DISTRIBUTIONS OF SMGS 58 My average value of 30 K for SMGs is slightly lower than the 36 K estimated in Chapman et al. (2005) for brighter SMGs, although this is probably due to the fact that my sample contains more low redshift, less luminous objects, where the 850 u,m selection effects are even stronger (see Fig. 3.9). These selection effects also mean that we may be missing a population of hotter ULIRGs which could contribute significantly to the star formation at high redshift (e.g. Chap-man et al. 2004b). We require more objects selected at shorter submm wavelengths and longer IR wavelengths to test this hypothesis and determine if there are actual differences in the SED shapes of ULIRGs at high and low redshift, or if we are just missing a large fraction of ULIRGs at high redshift. If high redshift star-forming galaxies really do have cooler temperatures then this may indicate that their emission is more extended than that of local ULIRGs, in which the majority of the IR emission comes from within the central kpc (Charmandaris et al. 2002). In support of this, Chapman et al. (2004a) found that 8/12 SMGs were spatially extended (on scales of ~ 10 kpc) in their radio emission (see also Ivison et al. 2002 and Muxlow et al. 2005), although much larger samples of radio-detected SMGs are needed in order to come to any definitive conclusions. In the middle and bottom panels of Fig. 3.8 I normalize the composite SED to the submm and radio flux, respectively. I expect Ss5Q^m to be a proxy for the far-IR luminosity, since the negative K-correction effectively cancels distance dimming over the typical z ~ 1-4 redshift range (see e.g. Blain et al. 2002). But if this is strictly true then the SEDs normalized at 850 jum will be independent of luminosity and should show variations in the SED shape only. In Table 3.3, I quantify the scatter (RMS of the logarithm of the flux density) at 24 /mi, 850 /um and 1.4 GHz for all three panels in Fig. 3.8, relative to the modified CE01 model. I have determined that the scatter due to photometric errors is negligible at this level, and therefore the scatter in the panels of Fig. 3.8 represents ranges in IR luminosity and SED shape. The scatter at 24/tm and 1.4 GHz is lower for the submm normalized SED than for the physical SED, which implies that on average 850 /im is a good proxy for luminosity. In addition, the fact that there is still scatter after normalizing to either submm or radio flux (i.e. after removing the scatter due to luminosity) implies that the variations in SED shape are contributing significantly 3.4. SPECTRAL ENERGY DISTRIBUTIONS OF SMGS 59 Table 3.3: RMS scatter in the SEDs of SMGs relative to the modified CE01 model. Fig. 3.8 shows the composite SEDs for spectroscopically identified SMGs in GOODS-N. In this table, I quantify the scatter (RMS of the logarithm of the flux densities) seen in the data points with respect to the best-fit model (cool CE01 model with additional extinction from the Draine 2003 models). The rows list the scatter (in dex) as a function of observed wavelength. The columns list the scatter when the SED is in physical units, normalized to the submm, and radio, respectively (corresponding to the three panels in Fig. 3.8). A RMS scatter (dex) Physical Submm norm Radio norm 24/mi 0.5 0.4 0.4 850/um 0.4 - 0.4 1.4 GHz 0.9 0.4 3.5. INFRARED LUMINOSITIES AND STAR FORMATION RATES 60 to the overall scatter seen in the top panel of Fig. 3.8. While I have not used the IRAC points to fit to the model SED, I notice that the intrinsic scatter of the IRAC points is lower when normalized to submm flux. Since IRAC is sensitive to stellar mass and 850 pm is sensitive to dust mass at the redshifts of these sources, this might indicate a correlation between stellar mass and the dust mass. Testing this further will require individual optical-radio SEDs fits for a large sample of SMGs and this work is reserved for a future paper. 3.5 Infrared luminosities and star formation rates 3.5.1 Estimating LIR and SFR Given that the average SMG seems to be fit well by the modified CE01 templates when both temperature and luminosity are allowed to vary, I have estimated the IR luminosity of each submm source by fitting its 24 pm, 850 pm and 1.4 GHz flux density to a suite of modified CE01 templates. In particular, I am interested in determining which of the three wavelengths is a better indicator of luminosity and thus star formation rate (SFR) in the SMGs using cur-rent templates. I have fit for IR luminosity (LIR) using the redshifted flux density at each wavelength separately (giving LIR 2 4, LIR 8 5 0 and LIR 1 4), and using all three together (giv-ing LIRALL)- I use the flux uncertainties as inverse weights when fitting for the best template and check to make sure that all data points (including the uncertainty) are consistent with the best-fit template. When I am fitting for LIR 2 4, LIR 8 5 0 or LIR 1 4 , I only have one data point, so I cannot constrain luminosity (normalization of the SED) and temperature (shape of the SED), and therefore I must assume that the CEO 1 templates accurately represent the luminosity-temperature relationship and fit for just the luminosity. However, when I am fit-ting all three data points to get LIR ALL> I can allow both temperature and luminosity to vary independently. I consider L I R A L L to be the best luminosity estimator, since it uses the most information, has the widest range of templates, and fits best (lowest reduced x2) to all three data points. I define IR luminosity as the integral under the SED from 8-1000 pm. I have fit for the IR 3.5. INFRARED LUMINOSITLES AND STAR FORMATION RATES 61 100 ™ ° ' 1000 SFR„(M Gyr') 100 1000 SFR M (M s yr ' ) 100 1000 L I R „ ( L B ) 10 , ! 10" LIR 1 4(Lil Figure 3.10: Infrared luminosities and SFRs for submm sources detected at both 24 /im and 1.4 GHz. Luminosities are estimated by integrating from 8-1000/im under the best-fit modified CE01 template. The bottom axes show the luminosities esti-mated using only one wavelength and the redshift. The left-hand axis is the esti-mated IR luminosity using data at all three wavelengths. Filled circles and squares denote submm sources with spectroscopic and photometric redshifts, respectively. The crosses denote sources which contain an AGN, as indicated by the presence of hard X-rays. The top and right-hand axes show the corresponding SFR using the expression in Kennicutt (1998). When only one data point is available to constrain the SED then the IR luminosity is overpredicted by 850 /im, since the temperature is not allowed to vary and it is forced to fit to a warmer template. However, when the temperature and luminosity are allowed to vary, 850 /mi, 24 pm and 1.4 GHz flux densities together give a good fit to the luminosity. 24 pm and 1.4 GHz data are not as sensitive to temperature and therefore they are able to predict an IR luminosity which is consistent with the result obtained when fitting all three and allowing both temperature and luminosity to vary. 3.5. INFRARED LUMLNOSITLES AND STAR FORMATION RATES 62 luminosity of all submm sources which have a redshift estimate (either photometric or spectro-scopic) and the results are shown in Fig. 3.10. Using templates based on local ULIRGs with a fixed luminosity-temperature relation (CE01), the submm flux will typically overestimate the luminosity. This effect is most dramatic for lower luminosity submm objects which, as I will show in the next section, tend to lie at lower redshifts where the 850 /xm window samples even further from the IR SED peak. However, when the temperature and luminosity are allowed to vary, 850 /xm, 24 /xm and 1.4 GHz flux density together give a good fit to the luminosity. Both 24 /xm and radio flux appear to be good indicators of the IR luminosity on their own, regard-less of the 850 /im flux. This is because the rest-frame near-IR and radio are less sensitive to temperature variations. Therefore, SMGs appear to follow the local mid-IR radio correlation but not the local far-IR (200-400 /im) radio correlation, since they appear to have cooler tem-peratures than local ULIRGs (as discussd in Section 3.4). Reddy et al. (2006) also find that the infrared luminosity, as estimated from the submm flux alone is higher than that estimated using only 24 /xm flux, for a sample of 9 radio-detected SMGs with spectroscopic redshifts. Dale et al. (2005) predict a very large scatter in the IR luminosity as estimated at 24 /im (rest-frame 8 /im for the average submm source) in high redshift galaxies, based on obser-vations of relatively local galaxies in the Spitzer Infrared Nearby Galaxies Survey (SINGS, Kennicutt et al. 2003). These galaxies are much less luminous than submm-selected galaxies and are perhaps forming stars under different physical conditions than starbursting galaxies at high redshift. In my sample of 850 /xm selected galaxies, I do not see the large scatter in the IR luminosity predicted by Dale et al. (2005). However, it could be that I am preferentially only selecting the cool ULIRGs at high redshift. The mid-IR spectral features, including the 9.7 /xm absorption feature, do not seem to have a large effect on the estimate of the luminosity. Also the presence of an X-ray counterpart does not appear to change the IR luminosity estimated by the 24 /xm, 850 /xm and 1.4 GHz flux, as might be expected if the AGN was strongly influencing the mid-IR spectra. Daddi et al. (2005) found that the mean 24 /xm, 850 /xm and 1.4 GHz fluxes for the K < 20 BzK sample agree within a factor of 2 with those predicted by an unmodified CE01 template with IR luminosity of l-2x 10 1 2LG. While their redshift ranges overlap significantly, SMGs in 3.5. INFRARED LUMINOSITIES AND STAR FORMATION RATES 63 this sample have average 24 /im, 850 /mi and 1.4 GHz fluxes which are a factor of 2, 7 and 2 times larger, respectively, than those for the BzK galaxies in Daddi et al. (2005). BzK galaxies seem to have similar SED shapes (temperatures) to local ULIRGs, while the SEDs of SMGs peak at longer wavelengths. SMGs appear to be quite atypical (being both more luminous and cooler) when compared to local ULIRGs and high redshift BzKs. Vega et al. (2005) have recently suggested that enshrouded star forming galaxies undergo four distinct phases charac-terized by different mid- and far-IR colours. Under their scheme, my results suggest that the SMGs, with cooler temperatures, may be an earlier phase in the star formation process and the BzK galaxies are a later, more evolved, phase (see also Dannerbauer et al. 2006). Individual values for the IR luminosity of the submm sources with spectroscopic redshifts are listed in Table 3.1 and Table 3.2. These are the luminosity estimates from the best-fit template using all three flux densities and the redshift. The mean and median IR luminosity is 6.7 and 6.0 x l 0 1 2 L o . These values are consistent with the luminosity values derived in Chapman et al. (2005) by fitting the 5 8 5 0/S 1 A for 73 radio-detected SMGs to the Dale & Helou (2002) templates (although their definition of IR luminosity is slightly different, extending to 1100/im). In Fig. 3.10 the top and right-hand axes give the corresponding value of the SFR, follow-ing the relationship between SFR and IR luminosity for starburst galaxies given in Kennicutt (1998): SFR(M 0 yr- 1 ) = 1.8 x 10~10 L s - i o o o ^ L © ) . (3.3) This relation assumes a Salpeter (1955) intial mass function and applies to starbursts with ages less than 100 Myr. It also assumes little or no AGN contribution to the IR luminosity (see Section 3.6 and Chapter 4). This estimate of the SFR is the rate of production of all stars in the galaxy from 0.1-100 M 0 . The mean and median SFR for this sub-sample of submm sources in GOODS-N is 1200 and 1100 M Q y r _ 1 , respectively. This is consistent with previous studies, which means that SMGs are significant contributors to the global star formation at high redshift (e.g. Hughes et al. 1998; Lilly et al. 1999; Barger et al. 2000). However, this is the first time that such values have been estimated using the mid-IR as well as the radio and submm regions 3.5. INFRARED LUMLNOSITLES AND STAR FORMATION RATES 64 of the SED. 3.5.2 Evidence for evolution? Although the IR luminosity derived from the submm flux (left panel of Fig. 3.10) is higher than that derived from all three wavelengths, there is still a correlation between the LIR 8 5 0 and L IRALL- TO investigate this, in Fig. 3.11 I plot the luminosity (LIRALL) separately as a function of the 850 flux density and the redshift. Note that plots of the 3.6, 4.5, 5.8, 8.0 and 24 /xm flux density as a function of 850 jttm flux density do not show any significant trends and are dominated by the scatter. The correlations seen in Fig. 3.11 were already suggested in Paper III, where I found a lack of submm-faint objects at high redshift (see Fig. 3 of Pope et al. 2005). I concluded that there might be several factors affecting this trend, including the strong evolution of the number density of ULIRGs with redshift. Ivison et al. (2002) also pointed out the tendency for brighter submm sources to lie at higher redshifts and the recent results of Wang, Cowie & Barger (2006), which suggest that a large fraction of the 850 /mi backgroud light comes from sources at low redshift (z = 0-1.5), are also consistent with this. The right panel of Fig. 3.11, which shows that the IR luminosity, as determined by the mid-IR, submm and radio flux ( L I R A L L X is also correlated with redshift, strengthens the case that we are seeing the effects of evolution. At higher redshifts, the lack of lower luminosity 850 /xm selected sources is surprising, since the increase in volume implies that one should see roughly twice as many as those seen at lower redshifts. While it is contributing, the effect of volume cannot completely account for the trends seen in Fig. 3.11. The fact that the SMGs are characterized by cooler SEDs than local ULIRGs of the same luminosity implies that the typical SED temperature of ULIRGs may change with redshift, which could contribute to the trend in the left panel of Fig. 3.11. Lower luminosity SMGs at z > 2 would have to be characterized by an SED that peaks at even longer wavelengths in order to be detected at 850 /xm. It is worth noting that the sample of 73 radio-detected SMGs in Chapman et al. (2005) also shows a lack of low luminosity objects (LIR < 2 x 10 1 2L o) at high redshift (z > 1.5). To fully test the conclusions hinted at by my sample and that of Chapman 3.5. LNFRARED LUMINOSITIES AND STAR FORMATION RATES 65 SgsoCmJy) Redshift Figure 3.11: IR luminosity as a function of submm flux density and redshift. The quantity on the y-axis is the 8-1000 pm IR luminosity derived from fitting the 24 pm, 850 pm and 1.4 GHz flux densities to the CE01 templates. Filled circles and squares de-note submm sources with spectroscopic and photometric redshifts, respectively. Note that only submm sources with secure counterparts are plotted in this figure. The crosses denote sources which contain an AGN, as indicated by the presence of hard X-rays. In the left panel, I plot lines of constant LIR (1O12L0)/S ,85o = 0.34, 0.86, and 1.3, which correspond to the lower quartile, median and upper quartile for this sub-sample of submm sources. While there is a linear correla-tion in these data, the scatter (roughly two orders of magnitude) indicates that the submm sources are characterized by a wide range of SED shapes (since the neg-ative K-correction largely removes the effects of redshift). The right panel shows that the IR luminosity of submm counterparts is a strong function of the 850 /im flux. A plot of redshift as a function of 850 /tm flux for this sample is presented in Pope et al. (2005) and also shows a correlation. These plots are difficult to understand without invoking strong evolution of ULIRGs with redshift. 3.6. AGN CONTRIBUTION 66 et al. (2005) will require thousands of submm selected galaxies. Surveys with SCUBA-2 will produce these large samples and, along with extensive multi-wavelength data, will allow for a detailed study of the luminosity function of 850 pm selected galaxies. 3.6 AGN contribution In the previous section, I assumed that IR luminosity and SFR are directly related. But are the very large IR luminosities of SMGs powered by starbursts, active nuclei, or both? Many SMGs show evidence of an AGN, as determined either through X-ray observations (Alexander et al. 2003b, 2005b), or with optical spectra (Ivison et al. 1998; Swinbank et al. 2004; Chapman et al. 2005). However, the most important question from the point of view of understanding the galaxy properties is not whether there are detectable AGN signatures, but whether or not the AGN is a significant contributor to the bolometric luminosity of the galaxy. Although, this question is difficult to answer definitively, data in the mid-IR can help determine what powers SMGs. With deep IRS spectra, Lutz et al. (2005) have determined the relative AGN contribution for two well-studied SMGs. One galaxy in the Lutz et al. (2005) sample shows that the AGN is a minor contributor to the luminosity, while the other seems to be powered equally by a starburst and an AGN. This small sample was limited to spectroscopically identified SMGs which are bright at 24 pm, and it seems reasonable to expect that the whole submm population will show an even wider range of AGN properties. Both Egami et al. (2004) and Ivison et al. (2004) have used the mid-IR photometry as a diagnostic tool to put limits on the contribution made to the IR luminosity by an AGN and agree that this contribution is < 25% for the SCUBA/MAMBO sources. Alexander et al. (2005b) put constraints on the AGN contribution to the energetics of a larger sample of radio-detected SMGs on the basis of their absorption-corrected X-ray luminosities. They found that, on average, the AGN contribution was likely to be negligible (~ 10%) unless SMGs have a substantially larger dust-covering factor than optically selected quasars. The data from mid-IR and X-ray observations are complementary and both are re-quired to place the tightest constraints on the power source of SMGs. In this Chapter I explore 3.6. AGN CONTRIBUTION 67 Table 3.4: Candidate AGN-dominated submm sources in GOODS-N from the secure counter-part list, either hard X-ray detected (Alexander et al. 2005b), or with a power-law shape in their mid-IR SEDs. The second column indicates if the source is detected in any X-ray band, while the third column indicates if it is hard X-ray detected. For the final column I assume that the mid-IR SED has the form Lv oc v~a and fit the IRAC data points to this model for each submm source. If the fit is good (low x2) then I list the best-fit value for a in the last column. A missing value in this column indicates that a power-law is a poor fit to the data and that the spectrum is more com-plicated than what is seen in simple AGN. Based only on these two AGN indicators, hard X-rays and mid-IR SED, I find that there is only one submm source in this list, GN04, which is likely to be dominated by an AGN in the mid-IR. SMMID X-ray X-ray IR SED AGN a GN04 Y Y 1.2 GN06 Y N GN10 Y N GN12 Y N 1.1 GN13 Y N GN17 Y N GN19 Y Y GN20 N N 1.7 GN20.2 N N 1.2 GN22 Y Y GN25 Y Y GN26 Y N GN30 Y N 3.6. AGN CONTRIBUTION 68 the AGN contribution to SMGs using mid-IR photometry and in Chapter 4 I present mid-IR spectroscopy for a subset of this sample. As shown in Fig. 3.5, the AGN Mrk231 has a higher S^/SW) ratio, due to additional contribution to the mid-IR flux from the warm dust which is heated by the AGN. The low S2A/SS5O ratios for the SMGs argue that an AGN does not dominate the bulk of the IR emission in the submm systems. I can use the optical through mid-IR photometry to look for a significant AGN power-law component. In addition, I can use hard X-rays to indicate the presence of an X-ray AGN, although a lack of hard X-rays does not rule out the possibility of an AGN being present. Note that it is important to look at both indicators of AGN presence, since locally there are powerful AGN which look like starbursts in the infrared while their X-ray properties reveal the AGN (e.g. NGC6240, Vignati et al. 1999; Lutz et al. 2003). On the other hand, there are also cases where the mid-IR SED does a better job of identifying the AGN and the X-rays are inconclusive (e.g. NGC1068, Jaffe et al. 2004). In Table 3.4 I summarize the candidate AGN-dominated submm sources in my sample based on X-ray and mid-IR photometry. All sources listed here are either X-ray detected or have a power-law spectrum in the optical/IR regime. I find that 3/21 secure submm counter-parts (14%) emit hard X-rays, indicative of an AGN. For 4/21 sources (19%), the colours in the IRAC bands can be well fit by a single power law Lv oc v~a, whose spectral index a I report in Table 3.4. A missing value in this column of the table indicates that a power-law is a poor fit to the data and that the spectrum is more complicated than what is seen in simple AGN. Even in the cases where the mid-IR SED appears to follow a power law, this does not guarantee that the IR luminosity is dominated by the AGN, since there could be non-AGN components contributing to the mid-IR SED. I find that, using both X-ray and mid-IR data, there is only 1/21 (5%) source in this sample which has the potential for having its IR lumi-nosity dominated by an AGN: GN04. Note that this is only an indication of the fraction of AGN-dominated sources, since I am not probing the far-IR luminosity directly. The combina-tion of mid-IR spectroscopy and X-ray information will improve the census of AGN activity in these sources and will provide the most definitive evidence on what is powering the energetics in SMGs, as well as helping to unravel whether there is an evolutionary connection between 3.7. SUMMARY 69 luminous SMGs and quasars (see Chapter 4). 3.7 Summary I present multi-wavelength identification of 33 (/35) SMGs in GOODS-N using the Spitzer GOODS Legacy data and a new reduction of the VLA 1.4 GHz radio data. 21 of these iden-tifications have a low probability of being random associations. This is the largest sample of statistically secure Spitzer counterparts to SMGs. 12 additional sources have a likely, although less secure, counterpart, and the remaining two sources have multiple possible IRAC identifi-cations, which would require somewhat deeper mid-IR or radio data to disentangle. The median redshift for the secure counterparts with optical spectroscopic or photometric redshifts (21/35 sources) is 2.0, while the median redshift for all counterparts, including those with only IR photometric redshifts (33/35 sources), is 2.2. With conservative assumptions I place a limit of < 14% on the fraction of 850 pm selected galaxies lying at z > 4. This is a lower fraction than is suggested for galaxies selected at ~ 1 mm (Eales et al. 2003). Having a large identified submm sample has allowed me to investigate properties of the bright (585o > 2 mJy) submm population. In most cases, the submm flux is clearly dominated by a single 24 yum source (20-700 pJy) with a wide range of 24 pm to 850 flux ratios as a function of redshift, suggesting the presence of spectral features and extinction in the mid-IR. I will explore this further in Chapter 4. I find significant scatter in the SEDs of SMGs, due to both a range of luminosities and SED shapes (temperatures). A composite rest-frame SED shows that the submm sources have SEDs which peak at longer wavelengths than those of local ULIRGs. Using a basic greybody model, 850 pm selected galaxies appear to be cooler (~ 30 K) than local ULIRGs of the same luminosity. These results indicate that the luminosity-temperature relation observed in local galaxies does not apply to high redshift SMGs. This may imply that the emission coming from high redshift SMGs is more extended than that in local ULIRGs, although it could also be due to selection effects, both locally and at high redshift. I also investigate the IR luminosities of the submm sources in GOODS-N, as estimated 3.7. SUMMARY 70 through their mid-IR, submm and radio flux densities. When all three measures are combined I find values for LIR which are consistent with the values derived from the mid-IR and radio data separately. Using templates based on local ULIRGs with a fixed luminosity-temperature relation, the submm flux will typically overestimate the luminosity. However, when the tem-perature and luminosity are allowed to vary, 850 yum, 24 yum and 1.4 GHz flux density together give a good fit to the luminosity. The median IR luminosity (using radio, submm and in-frared flux) for this sample is L (8 - 1000 yum) = 6.0 x 1O 1 2 L 0 , which corresponds to a SFR of 1100 M 0 y r - 1 . This is consistent with the values derived in Chapman et al. (2005) for 73 spec-troscopically identified radio-detected SMGs using only the radio and submm flux densities. I compare the infrared properties of these SMGs to a large sample of BzK galaxies in GOODS-N (Daddi et al. 2005). The median 24 yum to 850 /um flux density ratio of SMGs is a factor of ~ 4 lower than that of their high redshift neighbours, the BzK galaxies, indicating that these galaxy populations are characterized by different SED shapes. This difference may be understood in terms of the evolutionary scenario advocated by Vega et al. (2005), in which the SMGs, with cooler temperatures, may be an earlier phase in the star formation than the BzK galaxies. But a simple comparison of surface densities and luminosities also shows that the SMGs are more extreme star-formers than the average BzK galaxy. The IR luminosity (derived from all three wavelength flux densities) correlates with both the submm flux density and the redshift. While selection effects will affect this to some level, it is hard to understand this correlation without invoking strong evolution of ULIRGs with redshift. At shorter wavelengths, the mid-IR SED shows evidence for the presence of the stellar 1.6 /xm 'bump'. The Spitzer photometry alone can be used to constrain the redshift with rea-sonable accuracy, a(Az/(l + z)) — 0.07. Roughly 2/3 of the secure submm counterparts satisfy the criteria S5.8 > 6 pm and S5.8/S3.Q > 1.3, which has a low probability of occurring at random within the 8 arcsec search radius in GOODS-N. Hence a simple cut on this IRAC colour-magnitude plane is an efficient way of finding counterparts to SMGs. Several sources in this sample show a power-law shape at mid-IR wavelengths, which may indicate the presence of a powerful AGN. However, when combined with deep X-ray imaging, 3.7. SUMMARY 71 I find that only 1/21 (5%) of the secure sources in this sample are likely to be dominated by an AGN in the mid-IR emission. For the bulk of the sources the IR luminosity appears to be dominated by star formation rather than nuclear activity. Spitzer has opened up a new wavelength range through which one can study SMGs. Not only has it allowed me to identify many more submm counterparts, but I can begin to explore some of their physical properties. As a future project, the HST and Spitzer photometry can be fit to stellar population models to estimate the stellar masses and hence the star formation rates per unit mass of SMGs. Deep IRS spectroscopy can provide a more detailed look at the power behind the intense star formation in SMGs and disentangle many of the issues discussed in this chapter. This is the topic of the next chapter. 72 CHAPTER 4 MID- INFRARED SPECTROSCOPY OF SUBMILLIMETRE GALAXIES 4.1 Introduction In this chapter1,1 present mid-IR spectroscopy of a sample of SMGs in order to determine the contribution from AGN and starburst emission to the IR luminosity. There is very little overlap between samples of SMGs and other high redshift ULIRG samples detected with Spitzer (see Chapter 1). One of the main factors which could cause there to be a distinction between SMGs and other populations of ULIRGs at high redshift is the role of the active galactic nuclei (AGN). There has been a detailed investigation into the power behind the intense luminosities in local ULIRGs (e.g. Genzel et al. 1998; Lutz et al. 1999; Rigopoulou et al. 1999; Sturm et al. 2000; Tran et al. 2001; Armus et al. 2007). The mid-IR spectral regime is a particularly good probe of what is producing L I R since it consists of components from both the starburst (SB) and the AGN. The rest-frame 5-15 pm mid-IR spectral energy distribution (SED) can be decomposed into 3 main components (e.g. Sajina et al. 2007, hereafter S07): (1) emission features from what are believed to be polycyclic aromatic hydrocarbon (PAH) molecules (Puget & Leger 1989; Allamandola et al. 1999); (2) power-law (or warm blackbody) emission; and (3) extinction characterized by prominent Si absorption features (e.g. Draine 2003). The first component is believed to be powered entirely by star formation while the second is predominantly emission from the AGN (Genzel et al. 1998). The second can also contain a contribution from hot dust present in the most energetic HII regions (Tran et al. 2001). The third component can be found in both SB and AGN. Determining the contribution from ULIRGs to the global star formation rate requires dis-'The results in this chapter have been submitted for publication in the Astrophysical Journal: Pope A., et al., 2007, Mid-Infrared Spectral Diagnosis of Submillimeter Galaxies. 4.1. INTRODUCTION 73 secting their IR emission into contributions from starbursts (SB) and active galactic nuclei (AGN). With the arrival of the Infrared Space Observatory (ISO), mid-IR spectroscopy made it possible to decompose the various contributors in local ULIRGs with the conclusion that they are predominantly SB powered, although they also show signs of AGN activity (e.g. Genzel et al. 1998; Lutz et al. 1999; Rigopoulou et al. 1999; Sturm et al. 2000; Tran et al. 2001). In par-ticular, ISO studies of local ULIRGs showed that at low luminosities, the ULIRGs are powered mainly by a SB, but at the very highest luminosities the dominant emission mechanisms switch to the AGN (Lutz et al. 1998; Tran et al. 2001). From this one might expect that SMGs would be primarily AGN-dominated. The Spitzer InfraRed Spectrograph (IRS, Houck et al. 2004), with its improved sensitivity and spectral resolution, can extend this effort outside the local Universe, allowing much larger samples to be observed to much fainter flux limits. The IRS can detect emission features and continuum in ULIRGs out to z ~ 4 (Valiante et al. 2007). The 3 components listed above have been fit to the Spitzer IRS spectra of ULIRGs at high and low redshift to determine the level of AGN contributing to the luminosity (e.g. S07). This technique has been very successful, and it can be argued that this is the best way to probe the power coming from the AGN, since AGN indicators at UV, optical and X-ray wavelengths can be very obscured (Swinbank et al. 2004; Alexander et al. 2005b). Furthermore, the ultimate goal is to determine the contribution of the AGN to the total IR luminosity, and mid-IR spec-troscopy provides the best probe since it is sensitive to emission from the dust that is producing the far-IR peak. Two previous studies have presented IRS spectroscopy of SMGs. Menendez-Delmestre et al. (2007) presented the IRS spectra of 5 radio-detected, spectroscopically confirmed SMGs. They found that four SMGs with lower redshifts (z < 1.5) showed PAH emission and a com-posite spectrum of these four sources fit well to a template of M82 plus a power-law compo-nent. They concluded that these systems host both star formation and AGN activity. Their fifth source was found to have a more AGN-type spectrum. Valiante et al. (2007), the first two spectra also being presented in Lutz et al. (2005a) discussed IRS spectra of 13 bright SMGs, 9 of which are detected with the IRS. This sample was more representative of the redshift dis-tribution of SMGs. They also found that SMGs are likely to be mainly SB powered. S07 used 4.1. INTRODUCTION 74 the IRS to study a sample of Spitzer 24 /im-selected galaxies at z ~ 2. Follow-up studies of these galaxies have shown that they are not generally submm sources (Lutz et al. 2005b). 75% of this sample consists of continuum-, or AGN-, dominated sources, with some weak PAH emission indicative of star formation activity. They conclude that these sources, while an order of magnitude more luminous than local ULIRGs, have mid-IR spectra which are very similar to those of local ULIRGs. I present Spitzer IRS spectra of a sample of 13 SMGs which have extensive multi-wavelength information. The only bias in my sample is the 24 pm flux limit, and here I go substantially deeper than previous studies (5*24 > 200 yidy). The main goal of this program is to decompose the mid-IR emission into SB and AGN components, in order to infer how much of the total IR luminosity is coming from each. From the mid-IR spectroscopy I am also able to obtain spectroscopic redshifts and measure the luminosities and equivalent widths (EWs) of individ-ual PAH lines to investigate various line diagnostics. I compare the measurements of SMGs to those from other star-forming galaxies in order to put constraints on the evolution of massive galaxies and the relationship between star formation and black hole growth. 4.1.1 PAH emission lines Small dust particles are present in the interstellar medium throughout the Universe. In addition to absorbing starlight and re-radiating it as continuum emission, this dust also emits spectral features. As mentioned, the smallest grains are known as PAHs and emit a series of broad emission lines from ~ 3-20 pm (Puget & Leger 1989). These transitions arise due to C-C and C-H stretching/bending vibrational modes excited by a UV or optical photon, and they should be good tracers of the on-going star formation (e.g. Peeters et al. 2004). In Table 4.1, I list the major PAH emission lines between 5.5-13 pm which will be of interest in this thesis. In addition to the main 6.2, 7.7, 8.6, 11.3 and 12.6 pm lines which are often discussed, there are also fainter features at 5.7, 6.7, 8.3 and 10.7 p,m. 4.1. INTRODUCTION 75 Table 4.1: Main PAH emission lines from 5.5-13 /um (rest-frame). Central wavelengths and widths of the lines come from a combination of observational and laboratory results. The emission lines are much stronger in ionized PAHs as opposed to neutral PAHs, except for the features near 11.3 and 12.6 um which are equally strong in ionized and neutral PAHs (Allamandola, Hudgins & Sandford 1999; Draine & Li 2007). All information in this table comes from Draine & Li (2007) and Smith et al. (2007). Wavelength FWHM Tentative emission mode (/mi) (/xm) 5.700 0.200 C-H bend + C-H stretching 6.220 0.187 Aromatic C-C stretch in plane 6.690 0.468 Uncertain 7.417a 0.935 Aromatic C-C stretch 7.598a 0.334 Aromatic C-C stretch 7.850° 0.416 C-C stretch + C-H bending 8.330 0.417 C-C stretch + C-H bending 8.610 0.336 C-H in plane bending 10.68 0.214 C-H out-of-plane bending 11.236 0.135 C-H out-of-plane bending 11.336 0.363 C-H out-of-plane bending 12.62c 0.530 C-H out-of-plane bending 12.69c 0.165 C-H out-of-plane bending a7.7 yum complex. b 11.3 /um complex. c 12.7 /um complex. 4.2. IRS TARGET SAMPLE 76 Table 4.2: IRS low resolution module properties Module Order Wavelength range (pm) Slit length (arcsec) Slit width (arcsec) Pixel scale (arcsec) R- = A /AA Short-low (SL) 1 7.4-14.5 57 3.6 1.8 80--128 2 5.2-7.7 57 3.6 1.8 64--128 bonus 7.3-8.7 57 3.6 1.8 64--128 Long-low (LL) 1 19.5-38.0 168 10.7 5.1 80--128 2 14.0-21.7 168 10.7 5.1 64--128 bonus 19.4-21.7 168 10.7 5.1 64--128 4.1.2 The Infrared Spectrometer The IRS is one of the three instruments on the Spitzer Space Telescope and the only one capable of spectroscopy. The IRS is composed of four modules, two for low resolution spectroscopy (R ~ 60 — 127) from 5.2-38 p,m and two for high resolution spectroscopy (R ~ 600) from 9.9-37.2 fim. For faint high redshift targets, observations are made in low resolution. The two low resolution modules are broken down further into two orders plus one bonus order. Table 4.2 gives the properties of the low resolution modules. The IRS also has two 'peak-up' imaging arrays used for target acquisition. There are three observing modes with the IRS: staring; spectral mapping; and peak-up imaging. The first is most often used for individual fixed targets, although it is also possible to observe single targets in spectral mapping mode. During staring observations, the target is observed at two nod positions in the slit (1/3 and 2/3 of the slit length). The off-nod observations can then be used in the data analysis procedure to remove the sky signal. 4.2. IRS TARGET SAMPLE 77 Figure 4.1: Selection of IRS targets. The solid black dots are all SMGs with secure counter-parts from the SCUBA super-map (see Chapter 3). The horizontal line indicates the 24 /.tm flux cut of 200 /zJy for IRS targets. The red open circles indicate the 13 IRS targets; 10 of which are from the SCUBA super-map and 3 of which are from the C05 sample. 4.2. IRS TARGET SAMPLE 78 4.2 IRS target sample The sample of SMGs for IRS spectroscopy was primarily chosen from the sources with z > 0.5 secure counterparts in the GOODS-N SCUBA super-map (see Chapter 2 and Chapter 3). Since these Spitzer IRS observations take a large amount of telescope time, I chose only secure counterparts to be sure that each source was genuine submm galaxy. In Table 3.1 there are 23 SMGs with z > 0.5 secure counterparts when I include the 2006 'super-map' sources2. In order to be sure that the targets are detected with the IRS in a reasonable amount of time (< 5 hrs per IRS order), I imposed a 24 pm flux cut of S24 > 200 ply. 12/23 secure counterparts meet this 24 pm flux cut. Two of these sources, GN16 and GN25, were already observed with the IRS as part of other programmes which leaves 10 IRS targets. To expand the sample slightly, I also included three SCUBA photometry sources from the C05 sample. These three sources are not > 3.5a in the 'super-map' but they are detected in SCUBA photometry observations at a lower significance. The IRS target sample contains 13 SMGs. This sample will be limited to SMGs with z < 2.6 since the average SMG above this redshift is fainter than 200 ply (see Fig. 3.5 and Table 3.1) due to distance dimming and K-correction. Since SMGs at z > 2.6 are not signifi-cantly different from SMGs at z < 2.6 there is no reason to expect that the IRS sample is not representative of the submm population with similar mid-IR and submm fluxes. For z < 2.6 the majority of the SMGs in my sample have 624 > 200 /Jy therefore the IRS sample is not biased towards sources which have bright PAH features which are enhancing the 24 pm flux. Fig. 4.1 shows that distribution of 5 8 5 0 and S , 2 4 for the IRS sample compared to the distribution for all secure counterparts in the SCUBA super-map. The 24 pm flux densities of the sample range from 200-1200 pjy. 11/13 of these SMGs had previous optical spectroscopic redshifts while the other two had photometric redshifts; the redshift distribution of this sample is con-sistent with that of all SMGs (see Chapter 3; C05). Fig. 4.2 shows postage stamp images of the targets, roughly the size of the IRS aperture. In some cases there are several optical or even 2GN39 is considered as two separate sources since its two counterparts are far enough apart to be individually resolved by the IRS at these wavelengths. 4.3. OBSERVATIONS 79 IRAC sources in the aperture. For GN04, GN07 and GN19, the two IRAC sources are both expected to be associated with the submm emission, since both components are at the same redshift (see Appendix B). The IRS spectra of these 3 SMGs will contain contributions from both components. Some of the flux from GN39a is going to fall into the slit for GN39b. How-ever, a comparison of the 24 p,m flux from the IRS spectra to that from the MIPS 24 pm image confirms that only 1/3 of the flux in the IRS spectrum of GN39b is from GN39a. For the re-maining sources, the 8, 16 and 24 pm images show that only the primary targets are producing the mid-IR emission which I will detect with the IRS. 4.3 Observations Observations presented here are part of the Spitzer GO-20456 program (PI R. Chary). The full sample consists of high redshift SMGs, AGNs and optically faint MIPS sources. Here I present the results for the SMGs. The remaining sources will be discussed in future papers. I chose to observe in spectral staring mode which included observations of the target at two nod positions in the slit. I observed in low resolution (R = A/A A ~100) using the Short-Low 1 (SL1), Long-Low 2 (LL2) and/or Long Low 1 (LL1) orders, depending on the redshift of the source. I chose the wavelength coverage in order to include one or more of the main PAH features at 6.2, 7.7 and 11.3 pm. Integration times varied depending on the flux of the target at 24 pm. I used blue peak-up acquisition on nearby isolated bright Two Micron All Sky Survey (2MASS) stars. The observations were taken in April and May 2006. Table 4.3 shows the list of targets and integration times. 4.4 Data analysis The main challenge in observing in the mid-IR is emission from the sky. At these wavelengths, even from space, the dominant source of noise comes from such emission in the form of zo-diacal light. Since I am dealing with long integrations of very faint targets I need to perform additional data reduction steps in order to properly remove all the sky signal. 4.4. DATA ANALYSIS 80 Figure 4.2: Postage stamp images of IRS targets, in order of increasing redshift. Images are 10" x 10" (roughly the size of the aperture of the IRS observations) at B, i, 3.6, 8.0, 16 and 24 /im from left to right. Data are not available in some of these wavebands for source C3, since it is outside the HST GOODS-N coverage. Figure 4.2: (continued). Table 4.3: Observations: in order of increasing integration time. All observations were carried out in spectral staring mode with two nod positions. S M M IDs starting with 'GN' are from the SCUBA super-map. Sources with SMM IDs starting with ' C are from G05. Sources with integration time in square brackets are serendipitous sources observed in the slits of the primary targets. SMM Name . SMM ID RA DEC 5l6 524 Integration time _ SMMJ... J2000' J2000 (mJy) (mJy) SLl(x240s) LL2(xl20s) LLl(xl20s) Total (hr) 123600.2+621047 CI 12:36:00.16 62 10:47.3 0.478 ± 0 . 0 1 1 1.220 ± 0.008 6x2 10x2 1.1 123653.1+621120 . GN31 12:36:53.22 62 11:16.7 0.301 ± 0.007 0.367 ± 0.006 35x2 2.3 123622.6+621629 C2 12:36:22.66 62 16:29.5 0.098 ± 0.006 0.414 ± 0 . 0 0 7 45x2 3.0 123711.1+621325 • GN39a 12:37:11.37 62 13:31.1 0.097 ± 0.005 0.537 ± 0.009 20x2 28x2 3.2 123711.9+621331 ,'GN39b 12:37:11.97 62 13:25.8 0.051 ± 0.005 0.225 ± 0.007 [20x2] [28x2] [3.2] 123707.7+621411 GN19 12:37:07.19 62 14:08.0 < 0.02 0.280 ± 0 . 0 1 5 [20x2] [28x2] [3.2] 123701.2+621147 < V.GN17 12:37:01.59 62 11:46.2 0.209 ± 0.006 0.710 ± 0 . 0 0 8 20x2 12x2 3.5 123618.8+621008 .. GN05 12:36:19.13 62 10:04.3 0.044 ± 0.007 0.215 ± 0 . 0 0 6 55x2 3.7 123635.5+621238 GN26 12:36:34.51 62 12:40.9 0.992 ± 0 . 0 1 2 0.446 ± 0.005 15x2 20x2 [12x2] 3.3 [4.1] 123555.1+620901 , C3 12:35:55.13 62 09:01.6 n/a 0.374 ± 0.009 31x2 40x2 4.7 123616.6+621520 -GN04 12:36:16.11 62 15:13.5 0.060 ± 0.006 0.303 ± 0.007 70x2 4.7 123621.3+621711 GN07 12:36:21.27 62 17:08.1 0.048 ± 0.007 0.370 ± 0.011 35x2 45x2 5.3 123618.7+621553 GN06 12:36:18.33 62 15:50.4 0.036 ± 0.006 0.330 ± 0.008 35x2 55x2 6.0 4.4. DATA ANALYSIS 83 Initial reduction of the data was done using the S 14.0.0 IRS pipeline3 which corrects for most of the instrumental effects. The pipeline produces 2-dimensional (2D) Basic Calibrated Data (BCD) files which I performed further reduction on. Note that the targets are not resolved with Spitzer at these wavelengths and therefore they can be treated as point sources. The first step is to identify and clean the rogue pixels. There are a number of hot pixels in the IRS arrays (~ 15%). Rogue pixel masks were created from the campaign masks and from examining each 2D spectral image individually. I used IRSCLEAN4 to replace the values of the hot pixels by extrapolating from the surrounding pixels. Note that IRSCLEAN fails when there is a group of several hot pixels together, so I made sure to exclude these from the spectral extraction. The fraction of pixels in the 2D files which remain unusable after IRSCLEAN is only ~ 2%. The next step is to remove the latent charge build-up on the array. For long integrations with the IRS, it has been found that latent charge builds up with time, despite being reset at the end of each integration. This build-up is different depending on the wavelength (y position on the 2D spectral image). For more details see the relevant Spitzer Science Center (SSC) memo5. I found build-up on the arrays for integrations of more than 1 hour in all LL observations. For the SL1 observations, I only saw the latent build-up for a few targets, even though all of these observations were greater than 1 hour. Once the 2D files are clear from the sources of systematic noise produced by the detector, I can remove the sky noise. I explore several methods for removing the sky signal from these long integration files including: a) subtracting the coadded spectrum from each nod; b) sub-tracting each nod file and then coadding; c) subtracting a 'supersky' created from all files from that Astronomical Observation Request (AOR); and d) subtracting a supersky created from all files from a given campaign. In all methods, I mask out all other sources which fall within the IRS slit. The supersky methods were carried out with and without normalization. While the lat-ter method may seem like the best option, since it uses the greatest amount of data to estimate 3 h t t p : / / s s c . s p i t z e r . c a l t e c h . e d u / i r s / d h / 4 h t t p : / / s s c . s p i t z e r . c a l t e c h . e d u / a r c h a n a l y / c o n t r i b u t e d / i r s c l e a n /IRSCLEANJVIASK. html 5 h t t p : / / s s c . s p i t z e r . c a l t e c h . edu/irs/documents/irs_ultradeep_memo .pdf 4.4. DATA ANALYSIS 84 the sky, I found that because of changes in the detector responsivity throughout a campaign (because of observations of bright targets, etc.) and the time variation in the zodiacal spectrum itself, this provided a very poor estimate of the sky background in individual observations. The first two methods gave mediocre results for the faintest targets, since they did not account for changes in the sky signal over the duration of the observations. Furthermore it was impossible to use these methods in several cases, since there were other sources in the off-nod positions. In the end the third method of subtracting a normalized supersky created from all files from that AOR gave the lowest residual sky noise (and thus highest SNR) in the final spectrum. After the sky was removed from the individual data files, I then coadded them using a clipped median combine. The coadded spectrum was checked again for persisting hot pixels and cleaned once more using IRSCLEAN. These steps produced two 2D spectra for each target, one for each nod position. I used the SPitzer IRS Custom Extraction (SPICE 6) to extract the ID spectra at each nod position. I used a narrow extraction window of 2 pixels (~10 arcseconds for L L observations) fixed as a function of wavelength. I found this produced higher SNR spectra than using an extraction window of increasing width, starting with 2 pixels at the blue end. Using the optimal extraction in SPICE, I found comparable results to the 2 pixel fixed extraction. For each science spectrum, I also extracted a residual sky spectrum, offset from any sources, to give the final uncertainties in the spectrum as a function of wavelength. For each source, the ID spectra at each nod position were then averaged together. In order to calibrate the final spectra, I reduced standard star calibrator observations from the same campaign as my observations, in exactly the same way that I reduced the science observations. I then extracted the spectrum for the calibrator using the 2 pixel width and the default SPICE width in order to determine the calibration factor as a function of wavelength. This calibration was verified against the broad-band photometry at 8, 16 and 24 ^m, and I found very good agreement (see Fig. 4.4). Fig. 4.5 shows the final RMS of the spectra as a function of the integration time and spectral 6 h t t p : / / s s c . s p i t z e r . c a l t e c h . e d u / p o s t b c d / s p i c e . h t m l 4.4. DATA ANALYSIS 85 ~i 1 — i — i — r 1.0 in o.i V \ i i i i i i r 1 ?! 00 / 1 _ l I I I I 1_ 0.1 Peak-up S1 6 (mJy) 1.0 Figure 4.3: Calibration of IRS spectra compared to the 16 izm peak-up photometry (Teplitz et al. 2005). Most of the 16 /im IRS fluxes are systematically higher because roughly half of the 16 yum filter falls outside of the L L 2 IRS coverage and so the IRS flux is overestimated, since the spectrum actually drops off at lower wavelengths. These sources are indicated as the open circles. For sources with SL1 and L L 2 observa-tions, there should be no such discrepancy, which I confirm. 4.4. DATA ANALYSIS 86 MIPS S2 4 (mJy) Figure 4.4: Calibration of IRS spectra compared to the 24 pm MIPS photometry (Chary et al., in preparation). Again, open circles indicate sources where the 24 pm filter falls outside of the IRS coverage. 4.4. DATA ANALYSIS 87 0.25 0.20 0.15 CO 1 °-10 p i I i I i i r i • t 0.05 -0.00 SL1 • L L 2 • L L 1 • J I I l_ J I I I-5 10 15 Integration time (ks) Figure 4.5: Observed la RMS as a function of integration time for the IRS staring mode ob-servations. All observations from the GO2-20456 program are shown here. The dashed curves show the predicted values from SPEC-PET for each order. I confirm that the sensitivity goes as a a t~ 1 / / 2 even for very long integrations. 4.4. DATA ANALYSIS 88 order. Values from the Spitzer Science Center provided sensitivity calculator; SPEC-PET7 are given for comparison as the dashed curves. I confirm that the sensitivity goes as a(mJy) — a x t(s) - 1 / 2 even for very long integrations. The scatter in the measured values is probably dominated by the accuracy of the sky subtraction in these crowded fields. Given the scatter, the observed sensitivities are consistent with those from SPEC-PET. The decrease in sensitivity in the SL1 order could be related to the issues to do with the droop correction found for SL1 observations with 240 second ramps in fields with backgrounds > 20 MJy sr _ 1 at 24 u,m (see the Spitzer Observer's Manual8). 4.4.1 Spectral line measurements Probably the biggest uncertainty in comparing results from different surveys comes from dif-ferences in how the spectral lines are measured, more specifically, how the continuum is esti-mated for each line. This can lead to variations in the line fluxes and equivalent widths of up to a factor of 4 (see S07; Smith et al. 2007). Some groups simultaneously fit individual profiles for each line plus a continuum and an extinction term (e.g. PAHFIT9, Smith et al. 2007), while others remove the continuum by find-ing best-fit templates (S07) or cubic splines (Brandl et al. 2006, hereafter B06) and then fit the individual lines. I experimented with using a simplified PAHFIT on my data, but found that the narrow spectral coverage of my data and the low SNR meant that I did not have enough wave-length range or degrees of freedom to simultaneously fit all the individual components. I did employ a template fitting procedure to the spectra to determine the level of AGN contributing to the luminosity at these wavelengths (see Sections 4.5.2 and 4.5.3), but used a more simple approach to measure individual line strengths. In order to measure the PAH line fluxes I used the IDEA (ISAP™) environment in the Spectroscopy Modeling Analysis and Reduction Tool 7 h t t p : / / s s c . s p i t z e r . c a l t e c h . e d u / t o o l s / s p e c p e t / 8http.- //ssc . s p i t z e r . c a l t e c h . edu/documents/SOM/ 9 h t t p : / / t u r t l e . a s . a r i z o n a . e d u / j dsmith/pahfit.php 10The ISO Spectral Analysis Package (ISAP) is a joint development by the LWS and SWS Instrument Teams and Data Centers. Contributing institutes are CESR, IAS, IPAC, MPE, RAL and SRON. 4.4. DATA ANALYSIS 89 (SMART 1 1 ) software package (Higdon et al. 2004). For each of the 4 main PAH features (6.2, 7.7, 8.6 and 11.3 /mi), I isolated a region around each line and simultaneously fit the line and continuum. Both the centre of the line and the width were allowed to vary. The measured line widths were consistent within their errors with the widths measured in IRS spectra of local star forming galaxies (Smith et al. 2007). My method of measuring the line strengths gives very similar results to the cubic spline continuum fits performed in B06. This method will underes-timate the flux of the PAH lines (particularly the 7.7 am line) compared to methods which fit all components simultaneously (S07; Smith et al. 2007), since the continuum I adopt is higher. Therefore, I took care to measure the lines in the comparison samples in the same way as I did for the S M G spectra. Throughout this chapter, I compare the SMG results to those of other galaxy populations, specifically local SB galaxies (B06) and local ULIRGs (Armus et al. 2007). B06 presented IRS spectra of a sample of 22 'classical' SB galaxies in the local Universe with infrared lumi-nosities ranging from 5 x 10 9 -5 x 1 0 u L o . These galaxies are known to have little or no A G N contributing to their infrared luminosity. Local ULIRGs are known to show a wide range of properties, in particular, their infrared luminosities are fueled by a combination of A G N and SB activity (e.g. Tran et al. 2001). Armus et al. (2007) presented IRS spectroscopy of 10 ULIRGs from the Bright Galaxy Sample (BGS, 5 6 0 > 5.4mJy, Soifer et al. 1987). 8/10 ULIRGs in this sample are classified as AGN-dominated sources based on 6.2 /mi equivalent width. A larger sample of ULIRGs will show a wider range of properties in the mid-IR with more SB dominated sources (Desai et al. 2007). Because the choice for estimating the continuum under the PAH lines makes a huge difference to the results, I have measured the lines for the ULIRGs in exactly the same way as I did for the SMGs. A l l line measurements plotted in this chapter for the local ULIRGs were made by me and may be different from those published in Armus et al. (2007). I also check by measuring the lines in 10 of the B06 sources that my method is consistent with their measurements. On average the luminosities of the 6.2 and 11.3 am lines I measure are consistent with the values listed in B06, but the 7.7 /mi line luminosity is 1.5 times "SMART was developed by the IRS Team at Cornell University and is available through the Spitzer Science Center at Caltech. 4.5 . RESULTS 90 larger on average for my measurements. This is probably because I am using a straight line for the continuum as opposed to a spline. For comparison, S07 found that their measurements of the B06 7.7 /xm lines were 4 times larger. I have corrected the measurements of the 7.7 /mi line luminosity for the B06 SBs for this factor of 1.5 in all analysis in this chapter. For the B06 SBs I also applied a correction to the line measurements to account for the fractional flux found within the IRS apertures (see Table 2 of B06). The error on the line fluxes for the B06 SBs is dominated by the calibration error which is on the order of 10% (less than the size of the symbols in most of the plots in this chapter). Note the the errors I calculated from the residual sky were propagated in the fitting of individual lines. All SMG line measurements used in this chapter are listed in Table 4.6. 4.5 Results Fig. 4.6 shows the rest-frame (unsmoothed) IRS spectra for the 13 SMGs. PAH emission lines are present in all sources and the underlying continuum typically seems very flat. Even with these low SNR spectra, all SMGs seem to show very similar mid-IR spectral shapes with the exception of CI. While this particular SMG clearly shows PAH emission, it is superimposed on a steeply rising continuum. I will discuss this source further when I carry out spectral decomposition in Section 4.5.3. In this section I use the mid-IR spectra to explore the nature of SMGs. I start by using the identification of PAH lines to estimate redshifts. I then proceed to decompose the mid-IR spectra into SB and AGN components; starting first with a composite SMG spectrum since the SNR is higher and then proceeding to decompose each individual SMG spectrum. I then use the information from the spectral decomposition to classify the level of AGN contributing to the mid-IR luminosity and compare this with other AGN indicators such as optical spectroscopy and X-ray emission. In the next section, I fit the full infrared SED of each galaxy to put constraints on the contribution of an AGN to the bolometric luminosity. 4.5. RESULTS 91 6 8 10 12 Rest wavelength (um) 14 16 Figure 4.6: IRS spectra of SMGs. The solid black curves are the raw (unsmoothed) IRS spec-tra, while the shaded region is the associated la noise from residual sky emission. The red dashed curves show the best-fit SED which is made up of an extincted power-law component (blue dashed, consistent with zero in most cases) and an extincted starburst component (green dashed). Recall that the extinction curve is not just monotonic in wavelength and contains silicate absorption feature, the most notable being at 9.7 /im. 4.5. RESULTS Figure 4 .6: (continued). 4.5. RESULTS 94 4.5.1 Redshifts Given the presence of PAH features in all of the SMGs I am able to extract redshifts from the IRS spectra. In order to calculate the redshift and its associated error from the IRS data I follow the steps in Yan et al. (2007). Rest wavelengths for the PAH features are taken from Draine & Li (2007). For each PAH line, I find the centre of the line from SMART and calculate the redshift. The redshift for each source is calculated by averaging the redshift estimates from each PAH line. The total error in the redshift is the quadratic sum of the deviation due to the spread in the redshift from the different PAH lines and the centroid uncertainty. For sources whose spectra only contain 1 PAH line, the error is just the centroid uncertainty of that line. The deviation in the redshift from each line is calculated as la = y/J2i(zt — (z))2/\/n — 1- I calculate the centroid uncertainties by performing Monte Carlo simulations. I take a Gaussian of the same height and width as each PAH feature and I run 1000 trials where I add noise which has the same standard deviation as the noise in the real data. For each trial, I fit the data to get the line centre and keep track of the difference between the input line centre and the measured line centre. The centroid error is estimated from the standard deviation of this difference for all 1000 trials. I repeat this for each source. With the exception of GN26 whose spectrum has a relatively high SNR, the centroid uncertainties range from 0.08-0.25 /im, which corresponds to 0.5-1.5 pixels. This is roughly an order of magnitude higher than what one would expect from this instrument in the case of no noise. Table 4.4 lists the new IRS redshifts and the previous optical, spectroscopic or photomet-ric, redshifts. In general, the IRS derived redshifts are completely consistent with the optical spectroscopic redshifts, within the errors (see Fig. 4.7). The exception to this is that two of the sources seem to have very different redshifts as determined from IRS and optical spectroscopy (C2 and GN06). There are a number of reasons why the optical and mid-IR spectroscopic redshift might be different. The two wavebands could be picking up different galaxies, since the IRS aperture is so large. Alternatively, either the optical redshift or the IRS redshift could be misidentified. Since the PAH features are so broad and leave such a unique signature in the mid-IR spectrum, it seems unlikely that I have misidentified the redshift based on the IRS 4.5. RESULTS 95 Table 4.4: Redshifts: in order of increading redshift. All redshifts here are spectroscopic, and i magnitudes are from the GOODS ACS observations, unless otherwise noted. C3 is outside the GOODS-N HST region. SMGs with two values for i mag have two counterparts within the IRS aperture. GN39 is also a double SMG but the IRS observations resolve the two components. SMMID i mag (AB) -^optical ZIES ± Comment GN31 21.8 0.935d 0.93 ± 0.03 GN26 22.7 1.219a 1.23 ± 0.01 GN17 27.7 1.72c 1.73 ±0 .01 new specz C2 25.5 2.466°'6 1.79 ± 0.04 inconsistent C3 23.5e 1.875° 1.88 ± 0.02 GN39a 25.9 1.9966 1.98 ±0 .01 GN39b 26.4 1.992° 1.99 ± 0 . 0 4 GN07 27.8/23.9 1.988°, 1.9926 1.99 ± 0 . 0 2 GN06 27.8 1.865° 2.00 ± 0.03 inconsistent CI 24.8 1.994a, 2.0026 2.01 ± 0.05 GN05 24.9 2.60c 2.21 ± 0.03 new specz GN19 25.4/>28 2.484°, 2.4906 2.48 ± 0.03 GN04 26.2/>28 2.578° 2.55 ± 0.01 aChapmanetal. (2005). b Swinbank et al. (2004). c Chapter 3, photometric redshift. d Cowie et al. (2004). e From Capak et al. (2004) Subaru data. 4.5. RESULTS 96 2.5 2.0 c/: 1.5 -1.0 1.0 n i i r # 4 A ' _i i i i i ' -I I I I I L_ 1.5 2.0 2.5 "optical Figure 4.7: Comparison of IRS spectroscopic redshifts and previous optical redshifts. All op-tical redshifts here are spectroscopic except for the two sources indicated by open symbols which are photometric redshifts. There appear to be 2 optical spectro-scopic redshifts which disagree with the new IRS redshifts. In these cases I believe that the IRS redshift is correct for the SMG, since submm emission is more likely to trace the mid-IR emission than the optical/UV emission. 4.5. RESULTS 97 spectra. For GN06, the IRS redshift is based on 3 PAH features, 6.2, 7.7 and 11.3 pm, and the optical spectroscopic redshift of 1.865 is inconsistent with the positions of these emission features, even within the errors. The optical spectroscopic redshift of GN06 comes from C05 where they note that the optical galaxy was offset from the radio counterpart by about an arc-second. From the postage stamp images of GN06 in Fig. 4.2, it is clear that there are two optical knots, both of which are offset from the counterpart which is detected in the radio and mid-IR. C2 has the largest discrepancy between the IRS and optical spectroscopic redshifts. The IRS redshift for this source is based on the 7.7 and 11.3 pm PAH features and there is also a hint of the 8.6 ^ra line. The optical spectroscopic redshift would have placed the 7.7 pm PAH line right in the middle of this spectrum which it is inconsistent with the IRS spectrum. It seems likely that either the optical spectroscopic redshift is coming from a different galaxy or that the estimation procedure is picking up the incorrect redshift. Looking at.Fig. 4.2 it is clear that while there are several optical galaxies within the IRS aperture, there is clearly only one dominating the emission from 8 pm onwards. Since the mid-IR is much more likely to trace the submm emission than the UV/optical emission, I conclude that the IRS redshifts presented here are correct for the SMGs. While many of these sources are'very obscured in the optical due to the massive amounts of dust, the mid-IR is sensitive to dust emission and therefore it provides an unobscured estimate of the redshift. In the absence of optical redshift estimates, this IRS program would have provided secure (> 1 PAH line) redshifts for 70% (9/13) of the sources. The IRS has proven to be a valuable instrument for obtaining redshifts for SMGs. 4.5.2 SMG composite spectrum Since individually, the SMG spectra are low SNR, I can improve this by combining them to make an average mid-IR spectrum for SMGs. In doing this I can choose to normalize the spec-tra to the mid-IR luminosity (e.g. MD07) or the far-IR luminosity (Valiante et al. 2007) or not at all (this assumes that the dynamic range is low and all SMGs will have essentially the same level of luminosity in the mid-IR). As I show in Table 4.6, there is not a broad range of PAH 4.5. RESULTS 98 luminosities in the sample of SMGs, with the exception of GN31, which is roughly an order of magnitude fainter. Furthermore, Fig. 4.16 shows that there is significant scatter between L I R and L 7 7 which suggests that normalizing to L T R is not correct. I experimented with normaliz-ing the mid-IR spectra to various mid-IR wavelengths, but I found that this introduced biases in the composite, since normalizing to one of the features will decrease the relative strengths of the other features due to increased scatter. Furthermore, I do not have sufficient SNR in the continuum of the IRS spectra in order to normalize to it. Therefore, I choose not to normalize the individual SMG spectra (with the exception of GN31) in making the composite spectrum. I have excluded CI from the composition spectrum, since it shows a very different spectral shape from all the other SMGs (see next section). I can then calculate the mean, median or noise-weighted mean composite. Since the spectrum of GN26 is such high SNR, the latter results in a composite spectrum which is dominated by GN26, so I choose not to perform a noise weighted mean. I found that the mean and the median produced the same result within the errors. The error on the composite spectrum is calculated from combining the errors on the individual flux measurements. I restrict the final SMG composite spectrum to areas where there are more than 3 data files present at that wavelength, which results in a wavelength coverage of roughly 5-12 pm in the rest frame. Fig. 4.8 shows the composite spectrum of 12 SMGs (excluding CI). As I saw in the indi-vidual SMG spectra (albeit at lower SNR), there is strong PAH emission in these galaxies. In particular, I detect the 6.2, 7.7 and 11.3 pm emission lines, with a hint of the 8.6 pm emission line. The composite spectrum also shows a very flat underlying continuum. In order to quan-tify the emission from SB vs AGN at these wavelengths, I fit the composite spectrum with a model containing the 3 components discussed in Section 4.1. The PAH emission is fit using two different templates: 1) the mid-IR spectrum of M82 (Forster Schreiber et al. 2003); and 2) the SB composite template from B06. M82 is a prototypical star forming galaxy and its mid-IR spectrum is expected to arise predominantly from SB. The SB template from B06 is a composite of 13 local SB galaxies with high fluxes and without a strong AGN component. The AGN emission is characterized by a power law with both the normalization and slope as free parameters. For the extinction, I obtain rv from the Draine (2003) extinction curves and 4.5. RESULTS 99 Figure 4.8: Composite IRS spectrum of SMGs (excluding CI). This SMG composite spectrum (black curve) has been smoothed by a Gaussian with a width of 0.12 pm which is the average instrument resolution in the shortest order. The shaded region shows the la uncertainty from combining the errors from the individual spectra. In this figure, I fit the spectrum using two different approaches. Left: I fit the composite using a scaled PAH template (M82 or the SB composite from B06) plus a contin-uum. The red curve is the best-fit model, which is composed of a PAH template (green curve) and a continuum component (blue curve). I allowed both compo-nents to have extinction applied. Right: I fit the composite using individual PAH line profiles plus a continuum with extinction. The PAH lines are modeled as Drude profiles (e.g. Draine & Li 2007). I include the PAH lines at 5.70, 6.22, 7.42, 7.60, 7.85, 8.33, 8.61, 11.23 and 11.33 pm. In both panels I find that the PAH emission dominates the mid-IR luminosity. 4.5. RESULTS 100 applied it to both the PAH and continuum components separately. The extinction curve is not just monotonic in wavelength and contains silicate absorption features, the most notable being at 9.7 /im. The model, Fv, can be expressed as Fu = ci v~C2 e~C3 T- + c4 U " 0 5 Tv (4.1) where fv is the PAH template. I performed a x2 minimization fit for each of the two PAH templates and solved simultaneously for ci, c2, c3, c4 and c5. The best fit model, Fu, is shown in the left panel of Fig. 4,8 as the red dashed curve. The blue dash-dot and green dotted curves show the individual contributions from the continuum and PAH components, respectively. The best-fit PAH template was the starburst composite from Brandl et al. (2006) although M82 gave very similar results. With this PAH template, the extinction needed is r9.7 ~ 1 , most of which is applied to the PAH template as opposed to the continuum. The slope of the power law is very shallow (C2 ~ 0.2). Menendez-Delmestre et al. (2007) found a slope of 2.9 in their continuum component in a fit to a composite of 4 SMGs. However, they did not include extinction in their fit which removes some of the need for a rising continuum. Furthermore, these 4 SMGs were all at low redshift, allowing for their composite spectrum to extend out to 14 /im rest-frame. Given a longer baseline, the sample might show more of a rising continuum. I experimented with replacing the power-law component with a warm (T — 300 K) blackbody, which is perhaps more physically motivated. This gave very similar results, but produced higher reduced x2 values. The continuum component accounts for 30% of the luminosity at these wavelengths. The SMG composite shows that SMGs are SB dominated systems with an AGN that contributes up to 30% of the mid-IR luminosity. One downside of this approach is that it assumes that the relative strengths of the PAH do not vary in SMGs. The SB composite spectrum is shown in B06 to be different from the M82 spectrum. The latter shows stronger Si absorption and the relative line strengths are different. I found that the M82 spectrum and SB composite gave almost identical results in terms of the AGN contribution. This shows that, at the SNR of the data, the results are insensitive to the choice of the PAH template. I further tested this by fitting the composite to individual PAH line profiles together with a continuum with extinction. I include the PAH lines at 5.70, 6.22, 4.5. RESULTS 101 7.42, 7.60, 7.85, 8.33, 8.61, 11.23 and 11.33 /xm modeled as Drude profiles (e.g. Draine & Li 2007). A Drude profile is similar to a Lorentzian profile and has more power in its wings than a Gaussian profile (see right panel of Fig. 4.8). The resulting best-fit model shown in the right panel of Fig. 4.8 is very similar to the best-fit model from the PAH template and continuum fit (left panel). Furthermore, I again find that the PAH emission dominates the mid-IR luminosity. The PAH profiles model fits the SMG composite slightly better, specifically around 5.7 and 8.0/xm. This results from including the 5.70 and 8.33 /xm PAH features, which do not appear to be prominent in M82 or the SB composite. While these effects are near the level of the noise, it is worth noting that this might indicate that the dust composition in high redshift SMGs might may be slightly different than that in local SBs. Overall the scaled M82 and average SB templates provide a good fit to the data. Menendez-Delmestre et al. (2007.) and Valiante et al. (2007) also found thai M82 plus a power-law pro-vided a good fit to the SMGs in their samples. This is interesting, since M82 is 2 orders of magnitude less luminous in the infrared than a typical SMG. Apr220, which is a local ULIRG and more similar to SMGs in many other aspects, does not provide a good fit to the SMG com-posite spectrum. Arp220 shows very deep Si absorption and suppressed 6.2, 8.6 and 11.3 /xm PAH emission, relative to its 7.7 /xm PAH feature. It appears as if the average high redshift SMG shows similar mid-IR spectral features to lower luminosity SB galaxies than to its local high luminosity counterparts. I will explore this further when I look at individual PAH line luminosities in Section 4.7. In Fig. 4.9,1 show the SMG IRS composite spectrum compared to the IRS spectra of z ~ 2 Spitzer 24 /xm-selected ULIRGs from Sajina et al. (2007) and the local ULIRG Arp220. While Arp220 is often considered the typical local analogue to high redshift SMGs, it appears to have more silicate absorption and less 6.2 and 11.3 /xm PAH emission than the average SMG. Recall that the sample in Sajina et al. (2007) consists of red 24/xm-selected galaxies with S24 > 0.9 mJy, which is much brighter than most of the SMG population. The majority of the high redshift ULIRGs in the Spitzer-selected sample have weak PAH emission and are AGN-dominated; the green and red curves in Fig. 4.9 are composite spectra of the weak-PAH (i.e. AGN-dominated) Spitzer 24 /xm-selected ULIRGs with TQ.7 > landr 9. 7 < 1, respectively. 4.5. RESULTS 102 1111111111111111111111 M 11 M 11111111111111 M 11111 11111 SMG composite (12) S07 weak-PAH x 9 7 >l composite (14) S07 weak-PAH T<,7<1 composite (17) S07 strong-PAH composite (4) Arp220 5 6 7 8 9 10 11 Rest wavelength (urn) Figure 4.9: SMG composite spectrum compared to those of other ULIRGs. The SMG com-posite spectrum is the black curve and the shaded region shows the la error. The coloured histograms are high redshift Sp/Yz^ r-selected ULIRGs observed with the IRS from S07; the green and red curves are composite spectra of the weak-PAH sources for r 9 7 < 1 and T 9 . 7 > 1, respectively, and the dark blue curve is the composite spectrum for the strong-PAH sources. The light blue curve is the local ULIRG Arp220 (Charmandaris et al. 1999; Fischer et al. 1999). All curves have been normalized at ~ 7 /xm. The numbers in the legend in brackets indicate the number of sources in each composite spectrum. 4.5. RESULTS 103 The dark blue curve is the composite spectrum for the strong-PAH Spitzer 24 i^n-selected ULIRGs and is very similar to the SMG composite. In particular the relative line strengths are consistent. The strong-PAH Spitzer-selected sources represent only a small fraction of the Sajina et al. (2007) sample and these sources tend to have a higher detection rate in the (sub)mm than the rest of their sample (Sajina et al. in preparation). From Yan et al. (2007) and Sajina et al. (2007), the surface density of the z ~ 2 Spitzer 24 yum-selected ULIRGs is 10.4 deg - 2 and the strong-PAH sub-sample of these sources has a surface density of 2.6 deg - 2. The surface density of SMGs with 5850 > 4mJy (i.e. Lm > 5 x 1O 1 2 L 0 , Pope et al. 2006) is 844 deg - 2 (Coppin et al. 2006). While most submm-selected galaxies are much fainter at 24 pm, the strong-PAH Spitzer 24//m-selected ULIRGs appear to be a very small sub-sample of the SMG population. 4.5.3 Mid-IR spectral decomposition of individual galaxies Given the good agreement between M82 (or the SB composite spectrum) and the SMG com-posite, I proceed to fit the individual SMG mid-IR spectra to the model presented in the previ-ous section. Fig. 4.6 shows the best-fit model as the red dashed curve. Again the individual SB and AGN components are shown as the green and blue curves. In some cases, a significant ex-tinction component is needed (e.g. C3, GN04), although in most cases the fit without extinction provides an equally good fit. The wavelength range is very narrow and more observations of lower redshift SMGs (like GN26) which extend out to 15 pm in the rest-frame could help put tighter constraints on the amount of mid-IR extinction in these systems. Most SMGs fit well to this simple model, with the exception of GN26 which seems to deviate from the model at longer wavelengths (A > 13 pm). This is the only source in the sample whose mid-IR spectral coverage extends,past 13 pm, so that I cannot explore a more complicated model. I note that the spectrum of GN26 is even flatter than the model at long wavelengths which confirms that the AGN component must be negligible in this source. From the model fits, I derive the fraction of the mid-IR luminosity which comes from the continuum component. The continuum component provides an upper limit to the AGN 4.5. RESULTS 104 contribution, since hot dust present in the most energetic HII regions can also contribute to the continuum. When the best-fit model does not include a continuum component, I place an upper limit by forcing a continuum component. I use this fraction to classify the mid-IR spectrum and the results are listed in Table 4.5. 8/13 SMGs are clearly SB-dominated sources (continuum component < 50% of the mid-IR luminosity) and only 2/13 SMGs are AGN dominated (continuum component > 50% of the mid-IR luminosity) in the mid-IR. The remaining sources have a combination of PAH and continuum emission in the mid-IR. These results for the individual galaxies are in agreement with the composite spectrum which showed 30% AGN contribution on average. >' Note that all of this is based on the level of contribution from the AGN to the mid-IR luminosity. In Section 4.6.1 will discuss the contribution of the AGN to the total IR luminosity. 4.5.4 A G N classification Alexander et al. (2005b) studied the X-ray spectral properties of SMGs to probe the nature of the energy source. They conclude that the majority of SMGs host an AGN, but that the bolo-metric output from SMGs is dominated by SB, because of the low ratio of X-ray luminosity to IR luminosity. They also stress that some fraction of SMGs will be AGN dominated. Fol-lowing the same method as Alexander et al. (2005b) I list the X-ray classification in Table 4.5 where a hard X-ray detection is required to classify as an AGN dominated SMG. I find that the X-ray AGN classification gives a higher fraction of AGN dominated sources than the mid-IR spectra. There are 6 AGN classified sources from the X-rays and 4 of these are classified as SB from the mid-IR spectra. For the X-ray classified AGN-dominated sources, the X-ray emission expected from SF (e.g. Bauer et al. 2002) is less than that observed indicating that there is an AGN present. These IRS observations show that, while these systems may contain an AGN, it is not important to the bolometric luminosity. CI is classified as SB from the X-ray data, but shows a very strong AGN in the mid-IR. This SMG is likely to host a very obscured AGN, the kind of system that can help explain the unresolved portion of the hard X-ray background (e.g. Brandt & Hasinger 2005). This interesting source will be discussed in 4.5. RESULTS 105 more detail in a future paper. These results show that the fraction of SMGs which are AGN dominated is probably less than that suggested in Alexander et al. (2005b). On the other hand, the UV/optical spectra more often indicate that SMGs are SB dominated (Swinbank et al. 2004; Chapman et al. 2005) and are therefore more consistent with the IRS classification. In Chapter 3,1 found that most SMGs showed evidence for a stellar bump in the 4 IRAC channels. I found only 4 sources which had a good fit to a simple power-law from 3.6-24 pm, however only one of these is in the IRS sample: GN04. Comparing this to the IRS and X-ray classifications, I find that all 3 indicators converge on the conclusion that there is an energet-ically important AGN in GN04. The UV/optical spectra of this SMG did not elude to the presence of the AGN. Another approach for separating sources which contain a mid-IR AGN component is through the use of Spitzer colour-colour diagrams. Ivison ct al. (2004) proposed that the Sg/S^ vs S24/S8 colour plane could be used to weed out AGN dominated SMGs. The IRS has allowed me to directly measure the SB and AGN components in SMGs, and so I am in a position to test this plot as a diagnostic. In Fig. 4.10, I plot this Spitzer colour-colour diagram. The orange dashed and blue dotted curves show the positions of Mrk231 (mid-IR spectrum from Rigopoulou et al. 1999, spliced with a fit to the IRAS photometry) and M82 (Forster Schreiber et al. 2003), respectively, as a function of redshift (redshift is indicated by the numbers along the track). Based on these broad-band colours, AGN dominated sources are expected to lie along the Mrk231 tracks, while SB dominated systems should have similar colours to M82. Sajina, Lacy & Scott (2005) showed that while AGN dominated sources do lie on this AGN sequence, there are also just as many high extinction PAH dominated sources there. The large red dots are the SMGs from this study for which I have IRS spectra. Interestingly, both the SMGs (CI and GN04) which have a > 50% contribution from continuum (AGN) emission to the mid-IR luminosity based on the IRS spectra (see Table 4.6) lie off the SB sequence (and although GN04 is vaguely consistent with Mrk231, CI has quite extreme colours). Based on the IRS results, SMGs which lie in the light shaded region of Fig. 4.10 are SB dominated in the mid-IR. SMGs outside the shaded region likely have a combination of SB and AGN emission in the mid-IR or lie at the highest redshifts (z > 4). However, while these sources may have 4.5. RESULTS 106 I ! I I I I I I I I I I I I I I I I I I I 1 ! I I I I I I I I I I I I I I I I I I I I 1 I I I I 20 15 o DC ™ 10 SB SMGs A A A A A A A A A 1 • • • •. < A A )^GN04 • \ r '•A M82 Mrk231 •CI A * _ * 4 0 11 I I I I I I I I I 1 I I I 1 I I I I I I I I I 0 2 3 $8.0^4.5 4 Figure 4.10: Spitzer colour-colour diagram. The orange dashed and blue dotted curves show the positions of Mrk231 and M82, respectively, as a function of redshift (redshift is indicated by the numbers along the track). The large red dots are the SMGs from this study for which I have IRS spectra, where those with crosses are classified as X-ray AGN (Alexander et al. 2005b). Based on the IRS results, SMGs which lie in the light shaded region are SB dominated in the mid-IR. The smaller dots are the rest of the SMGs from Chapter 3. The small open triangles are the Spitzer 24 /im-selected ULIRGs from Sajina et al. (2007) with z > 1. 4.5. RESULTS 107 an AGN which is significant at mid-IR wavelengths, it becomes sub-dominant once extrapo-lated to total IR luminosity (see Section 4.6). This division between SB and AGN dominated sources based on IRS spectra agrees with the broad-band colours. The crosses in Fig. 4.10 indicate sources which are classified as X-ray AGN (Alexander et al. 2005b). Many of these lie along the SB sequence and have IRS spectra which are SB dominated. While the X-ray emission may indicate that an AGN is present in the system, it does not seem to be a very good diagnostic of whether the AGN contributes to the bolometric luminosity. Sources which have AGN dominated mid-IR spectra, like CI, clearly show very extreme colours in this diagram, although there are not enough of these sources to characterize their allowable positions in this colour-colour space. The smaller dots are the rest of the SMGs from the GOODS-N super-map. Extrapolating the IRS results, the fraction of all SMGs which are SB dominated in the mid-IR is > 80%. The small open triangles are the z > 1 bright 24/tm ULIRGs from Sajina et al. (2007); very few of these sources lie within the starburst region defined for SMGs indicating that they have a wider range of mid-IR spectra. Recently, Daddi et al. (2007) found that some massive z ~ 2 near-IR selected galaxies showed excess emission in the mid-IR above what is expected from star formation alone. They find this excess emission most prominent in the brightest mid-IR galaxies, which suggests that the most luminous 24 /xm galaxies at z ~ 2 contain a higher AGN contribution than those with fainter 24 /xm fluxes. CI is brighter than the rest of the SMG sample by a factor of almost 2 and shows the highest AGN contribution in the mid-IR. Furthermore, the brightest SMG in the Menendez-Delmestre et al. (2007) sample, SMMJ6350, also shows the highest AGN contri-bution within their sample. While the data do not prove that the brightest 24 /xm galaxies at z ~ 2 contain the highest AGN contribution, my. results are certainly consistent with this idea. This idea is also consistent with the brighter 24 /xm ULIRGs in the S07 sample, which con-tains a larger fraction of AGN-dominated sources;than SMG samples. Menendez-Delmestre et al. (2007) also suggest that since SMMJ6350 is their only high redshift SMG, there might be increasing AGN activity in SMGs at higher redshifts. However, I find no correlation between AGN contribution and redshift in my sample. 4.5. RESULTS 108 Table 4.5: Classification of mid-IR spectra. I compare my classification of the mid-IR spectra to classifications based on X-ray imaging and UV/optical spectroscopy. Sources are listed in order of increasing redshift. SMGs are classified as X-ray AGN if they are detected in the hard band (e.g. see Alexander et al. 2005b). Classification of UV and optical spectra are from Chapman et al. (2005) and/or Swinbank et al. (2004). Using the mid-IR IRS spectra, I classify each SMG based on the spectral decomposition fits shown in Fig. 4.6. I integrate the best-fit template to get the contribution from the continuum component to the mid-IR luminosity. When the best-fit model does not include a continuum component, I,place an upper limit by forcing a continuum component. SMMID Classification0 Mid-IR X-ray UV/Op Mid-IR continuum%6 GN31 U n/a SB < 45 GN26 SB SB SB < 15 GN17 SB n/a SB < 29 C2 AGN SB SB < 34 C3 AGN SB SB+AGN 48 GN39a oAGN SB SB < 35 GN39b oAGN SB SB < 10 GN07 SB SB SB+weak AGN 18 GN06 SB . SB SB+AGN 47 CI SB • SB AGN+weak SB 82 GN05 U n/a SB < 7 GN19 oAGN SB SB < 14 GN04 AGN SB AGN+SB 61 a oAGN = obscured AGN, SB=starburst, U=undetected. b Percentage of mid-IR flux that comes from continuum emission. 4.5. RESULTS 109 Table 4.6: PAH and IR luminosities. Sources are listed in order of increasing redshift. Recall that not all sources have enough wavelength coverage to include all 3 of the main PAH lines listed here. The equivalent widths are all in the rest-frame. The last 3 columns list the IR luminosity from the SB and AGN components and the total L I R . SMMID PAH luminosity (1O 9 L 0 ) PAH equivalent width (fim) ' L I R ( 1 O 1 2 L 0 ) 6.2 /jm 7.7 /nm 11.3 fim 6.2 fj,m 7.7 jum 11.3 /xm SB AGN Total GN31 3.4 ± 1.2 0.32 ± 0 . 1 1 > 0.20 < 0.11 0.31 GN26 12.2 ± 1 . 2 42.9 ± 5 . 1 8.1 ± 1 . 4 0.38 ± 0 . 0 4 1.00 ± 0 . 1 3 1 . 1 8 ± 0 . 5 4 2.8 < 0.01 2.8 GN17 13.0 ± 4 . 1 0.22 ± 0 . 0 7 3.4 < 0.07 3.4 C2 31.1 ± 6.3 7.0 ± 1 . 8 1.08 ± 0 . 3 0 1.00 ± 0 . 2 9 3.2 <0.01 . 3.2 C3 9.2 ± 2 . 7 7.9 ± 2 . 7 0.25 ± 0 . 0 8 0.45 ± 0 . 1 6 11.3 0.05 11.3 GN39a 13.9 ± 4 . 9 39.2 ± 1 2 . 9 0.66 ± 0 . 2 6 0.86 ± 0 . 3 0 > 5.3 < 0.3 5.6 GN39b 12.3 ± 5 . 6 23.1 ± 5 . 1 1.09 ± 0 . 6 5 0.59 ± 0 . 1 4 2.4 < 0.02 2.4 GN07 12.5 ± 2 . 6 17.6 ± 3 . 2 9.9 ± 1 . 8 1.10 ± 0 . 3 8 0.41 ± 0 . 0 8 1.58 ± 0 . 4 7 10.4 0.01 10.4 GN06 18.0 ± 1.5 26.3 ± 4.4 10.5 ± 2 . 2 0.83 ± 0 . 1 0 0.93 ± 0 . 1 6 2.95 ± 1 . 3 8 10.6 0.05 10.6 CI 6.3 ± 1.5 59.1 ± 12.6 0.05 ± 0 . 0 1 0.54 ± 0 . 1 4 6.4 4.3 10.7 GN05 12.9 ± 4 . 3 0.44 ± 0 . 1 5 4.1 < 0.01 4.1 GN19 12.5 ± 3.0 32.5 ± 9 . 0 0.59 ± 0 . 2 0 0.77 ± 0 . 2 2 24.9 < 0.02 24.9 GN04 43.3 ± 9 . 9 0.63 ± 0 . 1 5 19.2 1.4 20.6 4.6. FULL SED FITS 110 4.5.5 The GN39 double system One further SMG deserves special discussion and that is GN39. This SMG has two radio/MIPS counterparts separated by 8 arcseconds, which I refer to as GN39a and GN39b. Both are confirmed to lie at z ~ 2 (Swinbank et al. 2004; C05), and therefore it is likely that they are both contributing to the submm emission. The separation of 8 arcseconds corresponds to 70 kpc at this redshift. The IRS observations picked up both counterparts and, with a narrow extraction window, I am able to extract a separate spectrum of each component. Some of the flux from GN39a is going to fall onto the slit of GN39b, although a comparison of the 24 /mi flux from the IRS spectra to that from the MIPS 24 /xm image confirms that only 1/3 of the flux in the IRS spectrum of GN39b is from GN39a. The extracted spectra confirm that these two sources are indeed at the same redshift and both spectra look very similar to each other and to the rest of the SMGs, supporting the idea that they are both submm emitters. Furthermore, both components are X-ray detected and classified as obscured AGN by Alexander et al. (2005b), although the two IRS spectra show a completely SB dominated system at mid-IR wavelengths. Since these two SMGs are at the same redshift, I know that they must be associated. Given the physical distance between them, we could be seeing the merger of two massive galaxies (Springel, Di Matteo & Ffernquist 2005). The apparent separation is quite large, compared with what is seen in most local ULIRGs (Murphy et al. 1996), implying that it might be a very early stage of the merger. Note that GN04, GN07 and GN19 also have double radio counterparts, which are thought to be at the same redshift. However, in these sources, the two components are separated by only 3 arcseconds and therefore they do not resolve into two separate 24 /xm sources. Thus the IRS spectra for these sources contain contributions from both components. 4.6 Full SED fits In Chapter 3,1 fit the 24 /xm, 850 /im and radio photometry to a suite of models to determine the total infrared luminosity and dust temperatures. 1 used the Chary & Elbaz (2001, hereafter 4.6. FULL SED FITS 111 Rest wavelength (nm) Figure 4.11: IR SED of source C2. This figure demonstrates the template fitting to the full SED from mid-IR to radio wavelengths. The green dotted curve is simply the CE01 template which best matches to the submm flux, while the dot-dash dark blue curve is the the CE01 template which best matches the 16 and 24 /.im fluxes. For these two models I emphasize that the CE01 templates have not been allowed to scale in luminosity. The light blue dash-dot-dot curve is the best fit to all the data points > 7 /mi in the rest frame, where I have allowed the CE01 templates to scale. The red dashed curve is a scaled CE01 template with additional extinc-tion from the Draine (2003) extinction curve, and this provides the best-fit to the data. I require both a scaling of local templates (in this case CE01) and additional extinction to fit the SED of high redshift SMGs. 4.6. FULL SED FITS 112 CE01) templates and modified them by adding additional extinction from the Draine (2003) models. The CE01 templates were developed to be representative of local galaxies, and they con-tain an inherent luminosity-temperature relation. When fitting, I allowed the models to have freely-floating normalization so that I could solve for both the luminosity and the tempera-ture independently (i.e. a model with high luminosity but cool average dust temperature is allowed). I found that SMGs fit best.to scaled-up versions of lower luminosity templates which had cooler average dust temperatures than local ULIRGs. In these fits the short wavelength part of the spectrum was constrained only by the 24 pm flux. From the present study it is apparent that the 24 pm flux of SMGs will vary considerably with redshift, due to the presence of strong PAH emission lines. Therefore, I have repeated this full SED fitting now that I have the IRS spectroscopy. Fig. 4.11 demonstrates why I need to allow the CE01 templates to scale (i.e. having lumi-nosity and temperature both as variables) and the additional extinction from the Draine (2003) models. Here I plot the mid-IR to radio SED for one of the SMG sources, C2. First I simply plot the CE01 template which best matches the submm flux as the green dotted curve and the CE01 template which best matches the 16 and 24 pm fluxes as the dot-dash dark blue curve. For these two models 1 emphasize that the CE01 templates have not been allowed to scale, and therefore they follow the local luminosity-temperature relation. Because SMGs are so lumi-nous at mid-IR and submm wavelengths, then the green and dark blue models have a warm average dust temperature (~ 50K). The light blue dash-dot-dot curve is the best fit to all the data points > 7 pm rest frame where I have allowed the CE01 templates to scale. While this provides a good fit to the mid-IR spectrum and the submm flux, the radio flux and L I R are underestimated. I know that the CE01 templates need to be scaled up to match the photometry of SMGs, but as I scale them up I am introducing a greater dust mass which must then cause more extinction. The red dashed curve is a scaled CE01 template with additional extinction from the Draine (2003) extinction curve which provides the best-fit to the data. I require both a scaling of local templates (in this case CE01) and additional extinction to fit the SED of high redshift SMGs. 4.6. FULL SED FITS 113 I repeat this procedure for all SMGs, fitting the IRS spectrum, and the submm and radio photometry to the CE01 templates plus Draine (2003) extinction, allowing luminosity and average dust temperature to vary. Fig. 4.12 shows the full SED fits. For the GN39 double system, I split the total submm flux according to the radio flux of each component (note that this is the same as splitting by 24 pm). In general, the resulting fits are very good. The dark curve and dots are the points used for the fitting procudure, while the open symbols show the other photometry points (mostly 16 and 24 pm) which were not used in the fits but agree well with the best-fit SED. The vertical dotted lines indicate the region within which I measured the total IR luminosity. I integrated the full SEDs in Fig. 4.12 from 8-1000 pm to get the total infrared luminosity. Values of X I R are listed in Table 4.6 and they are in good agreement with the values from Chapter 3 For sources which I found to contain an AGN component in the mid-IR, I need to calculate the contribution to L I R from the AGN. Using ISO spectra, Tran et al. (2001) found that the ratio of i s - i o ^ m / ^ i R was about 2.5 times greater for AGN than for SBs. In other words, as one extrapolates from mid-IR to far-IR wavelengths, AGN become less important bolometrically than SBs. In order to quantify this for my sample, I scale the Mrk231 (an AGN whose L I R is known to be dominated by AGN emission and not star formation, Armus et al. 2007) template to the AGN fraction of the mid-IR luminosity (listed in Table 4.6) and then integrate the SED from 8-1000 pm to estimate the contribution to L i R from the AGN component. For sources which have an upper limit to the AGN contribution, I plot the scaled Mrk231 in Fig. 4.12 at this limit. The values of I/fjpN and L f R = L T R — Z/^fN from integrating under the curves are given in Table 4.6. As can be seen in Fig. 4.12, only CI (and possibly GN31, although the AGN fraction for that source is an upper limit) has enough AGN contribution in the mid-IR to make a significant contribution to L I R . Therefore SMGs really do have intense IR luminosities powered almost entirely by star formation. In Section 4.8,1 discuss how this luminosity can be converted in to an estimate of the star formation rate (SFR). To get a measure of how important the mid-IR spectrum is to the total L T R , I calcuate L 8 _2o by integrating the best-fit model from 8-20 /.im (there is no point including shorter wavelengths here since they do not go into the L i R quantity). On average LS-2o is roughly 6% (4-10%) of 4.6. FULL SED FITS 114 N X > — GN31 i GN26 GN17 C2 IO1 IO2 IO3 104 Rest wavelength (um) IO5 Figure 4.12: Mid-IR to radio SEDs of SMGs. The IRS spectrum (dark curve) plus submm and radio photometry (solid circles) are used in the fit. The open squares indicate the photometry at 16 and 24 pm (and 70 jim in the case of GN26, Huynh et al. 2007). The red dashed line is the best fit modified CE01 template, while the blue curve is Mrk231 normalized to the AGN fraction of the mid-IR given in Table 4.6. The horizontal dotted lines indicate the region within which L i R is calculated. 4.6. FULL SED FITS 115 lO27 1026 iO25 [024 IO23 [O22 IO26 IO25 C3 GN39a o24 \J0: [O23 [022 IO27 IO26 IO25 [024 [O23 [O22 [O27 IO26 [O25 IO24 IO 2 3 [ 0 2 2 GN39b GN07 101 102 IO3 104 Rest wavelength (\xm) IO5 Figure 4.12: (continued). 4.6. FULL SED FITS to27i LO26 IO25 LO241 23 22 0 10 IO27 IO26 IO25 IO24 IO23 IO22 IO27 [ 0 26 | IO25 I O 2 4 1 [023 IO22 IO27 IO26 iO25 IO24 IO23 [O22 IO27 IO 2 6 iO25 IO24 IO23 IO22 GN06 H Lrf* N N CI GN05 GN19 H GN04 H IO1 IO2 IO3 104 105 Rest wavelength (\im) Figure 4.12: (continued). 4.6. FULL SED FITS 111 LIR. This is less than for normal star forming galaxies (Smith et al. 2007). As might be expected, there is a tendency for the most luminous SMGs to have the least contribution from the PAHs to L I R . Therefore, while knowing the exact shape of the mid-IR spectrum does not have a huge effect on deriving the total LTR, it is nevertheless important to have an anchor point in the mid-IR to constrain the fits. Furthermore, the shape of the mid-IR spectrum is crucial for constraining the contribution from an AGN to L I R and thus calculating Lf^. It is interesting to ask if the additional extinction term from the Draine (2003) models is required in order to obtain a good fit. The answer is yes. Since SMGs appear well fit to scaled up versions of lower luminosity templates, they do not contain sufficient extinction, which I would expect to be present in galaxies with ULIRG-like luminosities. In order to quantify the effect of the extinction term on the models, I have calculated LIR without any additional extinction. The median amount that the extinction changes LTR by is 13% (0-20%), and so models which do not include the extinction term will incorrectly determine LIR by around 13%. I find an average value of r 9 . 7 ~ 1, and I do not see any correlation between the extinction and LIR. By simply scaling to the wavelength where the SED is a maximum as an estimate of the average dust temperature, I obtain a median of 32 K for these 13 SMGs. This is consistent with previous estimates of dust temperatures in SMGs (see Chapter 3; C05; Kovacs et al. 2006; Coppin et al. 2007). I also found a similar trend between luminosity and temperature as found in C05 which is offset to cooler temperatures from the local correlation. As mentioned in C05 the lack of warm objects with lower luminosities is due to the selection at 850 /um. ULIRGs selected at 24 and 70 /xm, like those in the Yan et al. (2007) sample, are likely to fill in this region of the L-T plane. I can also estimate the dust mass from the far-IR SED - following the relation given in Dunne et al. (2000), the median dust mass for this sample is 2.4 x 1O 8M 0. 4.6.1 Radio-IR correlation Locally, a tight correlation has been established between the radio and IR emission in galaxies (e.g. Condon 1992). The non-thermal radio emission, or synchrotron radiation, is caused by 4.6. FULL SED FITS 118 10 25 N X 10 24 N X o -i-10 2 3 10 11 j i i 1 1 1 1 — I — F T J I I I I I L J _ 10 12 10 13 LFIR(L©) Figure 4.13: Radio IR correlation: 1.4 GHz radio luminosity density as a function of the far-IR luminosity for the 13 SMGs with IRS data. I assume a power-law synchrotron spectrum with a slope of 0.7 to calculate LI.4GHZ, while lJjPIR is defined as the total luminosity after removing the AGN components from 4 0 - 5 0 0 fxm. There is a tight correlation over several orders of magnitude. The dashed line shows the best-fit linear relation. 4.6. FULL SED FITS 119 the acceleration of relativistic electrons originating in supernova remnants. In the infrared, the spectrum is dominated by blackbody radiation from dust. Dust grains absorb optical light from stars, and this is re-emitted in the infrared. The key connection between these two processes is the young massive stars - any contribution from AGN to these luminosities will disrupt the correlation. I found that high redshift SMGs can be well tit by modified CE01 templates, and these contain an inherent radio-IR correlation based on local galaxies. Furthermore, I found little contribution from an AGN to the L T R S , so that SMGs should also follow to the radio-IR corre-lation. In Fig. 4.13 I plot the 1.4 GHz radio luminosity as a function of the far-IR luminosity. In order to calculate the rest-frame 1.4 GHz luminosity I assumed a power-law synchrotron spectrum with a slope of 0.7 (Condon 1992). The far-IR luminosity here, L | f R , is the total luminosity from 40-500 /xm (Sanders & Mirabel 1996). I have removed the contribution from the AGN to the L | f R . Indeed, there is a tight correlation over several orders of magnitude in I/|fR. The dashed line shows the best-fit line, which can be described as log ( ^ j g z r ) = (9-4 ± 0.9) + (1.18 ± 0.07) log (^j . (4.2) The radio-IR correlation is often described in terms of the q parameter defined by Helou, Soifer & Rowan-Robinson (1985) as " = ' ° g (375 x 10" Wm-») - '°g Um-'lfa- ) - < 4 3 > Using this definition, I calculate an average q = 2.16 with aq = 0.16 for these 13 SMGs. While slightly lower, this is consistent with values derived for local IRAS galaxies (q ~ 2.3, Condon et al. 1991, Yun et al. 2001) and high redshift Spitzer galaxies (Appleton et al. 2004). Using follow-up observations of SMGs at 350 yum with SHARC-II on the CSO to constrain the far-IR SED, Kovacs et al. (2006) found a slightly lower value of q = 2.07 with aq = 0.21, although this is consistent witlvmy estimate within the errors. Based.on this analysis, I find that the radio-IR correlation in high redshift SMGs is similar to the correlation established locally. 4.7. PAH LUMINOSITIES 4.7 PAH luminosities 120 The estimated PAH line luminosities and equivalent widths for the sources are listed in Table 4.6. Line luminosity here means the best-fit Gaussian, after continuum removal, integrated and then converted to the rest-frame using the luminosity distance (see Appendix A.l). The equivalent width is an alternate measure of the strength of a spectral line. For emission lines, it is defined as the width of a rectangular region between zero and the continuum level which has the same area as the spectral line. It is apparent that L 7 . 7 for the SMGs varies by- over-an order of magnitude, while LQ,2 and Xii .3 have much smaller dynamic ranges of a factor of a few. While this might seem odd, I note that only a fraction (40-60%) of the sample have 6.2 and 11.3 pm spectral coverage whereas 92% have coverage of the 7.7 /im emission line. Furthermore, the redshift range for the sources with 6.2 and 11.3 /im spectral coverage is very small and therefore I expect a small dynamic range in PAH luminosity. The AGN-dominated sources, as found from the IRS spectroscopy (CI and GN04), all show suppressed 6.2 pm emission relative to their 7.7 pm emission. Fig. 4.1.4 shows the relative strengths of PAH lines compared to local SBs (B06) and ULIRGs (Armus et al. 2007). The two panels showL7.7 as a function of L 6 2 (left) and Ln.3 (right). The outlying source in the left panel of Fig. 4.14 is CI, the most AGN dominated SMG. The sampled rest-frame for CI does not extend out to 11.3 pm, and therefore I cannot plot this source on the right panel. While the SMGs do not show enough dynamic range in PAH luminosity to pick out any strong correlations within the population, they do fall on the relation between various PAH lines established for local SB galaxies. The lines in Fig. 4.14 are the best-fits to the local SB (blue points). I find essentially the same relation if I fit the local SBs and the SMGs, indicating that the PAH luminosities in SMGs are simply scaled up (albeit by several orders of magnitude) from less luminous SB galaxies. The best-fit relations from fitting both the SMGs and the SBs are log(L7.7) = (1.1 ± 0.2) + (0.92 ± 0.02) x log(L6.2) 4.7. PAH LUMINOSITIES 121 107 1 0 8 1 0 9 1 0 1 0 107 108 IO9 IO10 1-6.2 (Us) L l t_ 3 (L®> Figure 4.14: Comparison of PAH luminosities. Here I compare the SMGs presented in this thesis (red circles) to local SBs (blue squares, B06) and local ULIRGs (green crosses, Armus et al. 2007) which have been analysed in a consistent way. la error bars are shown on all points (the 10% errors on the SBs are similar in size to the symbols). The outlying SMG point in the left panel is CI which shows a strongly rising continuum in the mid-IR, unlike the rest of the SMGs. The solid line shows the best fit to the SBs. The SMGs clearly lie on the relation established for local SB galaxies, while the average ULIRG and the SMG CI are inconsistent with the solid line (less 6.2fim PAH luminosity for a given L 7 7 ) and appear to fall on a different relation. 4.7. PAH LUMINOSITIES 122 log(L7.7) = (0.8 ± 0.3) + (0.96 ± 0.04) x log(Ln.3). These are very close to linear relations, i.e. L 6 . 2 oc L 7 . 7 oc Ln . 3 . If I assume a slope of 1 then I find L 7 . 7 ~ (3.0 ± 0.7) x L 6 2 for the SBs and SMGs. The average ULIRG appears to lie off this relation with an excess of 7.7 pm PAH emission for a given L 6 2 or Ln .3. Recall that the ULIRG line measurements were obtained in exactly the same way as for the SMGs and I also checked that the SB line luminosities from B06 are consistent with my estimates. The ULIRGs, along with the SMG CI, appear to fall on a separate relation L 7 . 7 ^ 8 x L 6 2 . The two ULIRGs closest to the line are the two sources classified as SB-dominated in Armus et al. (2007). Lutz et al. (1998) also noted that the ratio of the 6.2/7.7 pm features in ISO spectra of ULIRGs were slightly lower than those seen in starbursts. This indicates that, on average, the local ULIRGs contain a more dominant AGN component or greater extinction than both high redshift SMGs and local SBs. While SMGs likely contain an AGN, it is not contributing significantly to their mid-IR luminosity, just as it is not significant for local SB galaxies. Fig. 4.15 shows the equivalent widths of the 6.2 and 7.7 pm PAH lines. The vertical dashed line at EW6,2 — 0.2 is often used to separate SB dominated and AGN dominated sources (Armus et al. 2007; S07). With the exception of CI, all of the sources which have measurable 6.2 pm PAH emission have EW6.2 > 0.2, which indicates that they are SB dominated. I have again plotted the local ULIRGs and SBs for comparison. Recall that all measurements have been made in a consistent way (see Section 4.4.1) to allow this comparison to be made. I found that the values in B06 for EW 7 . 7 were on average 1.6 times lower than the measurements of the SB EW 7 . 7 , and so I have corrected all the B06 points in this plot. Most of the ULIRGs have very small £ W 6 2 indicating that they are AGN-dominated (see Armus et al. 2007). The SBs cover a small part of this plot, while the SMGs are spread over a larger region, although also with much larger error bars. The current IRS data do not have sufficient SNR to determine if there is any genuine spread or correlation among the equivalent widths. However, the error bars are sufficiently small to exclude equivalent widths of less than 0.3 in either PAH line. The SMGs show similar equivalent widths to the strong-PAH ULIRGs from the Spitzer selected 4.7. PAH LUMINOSITIES 123 i 1 1 r i 1 1 r T a 1.4 1.2 1.0 0.8 W 0.6 Tn SMGs» Local ULIRGs x Local SBs • 0.4 0.2 0.0 _i i i_ J i i i _ 0.0 0.5 1.0 E W 6 2 ((im) 1.5 Figure 4.15: Rest-frame equivalent widths of main PAH features. Again the SMGs are denoted by the red circles, the local SB (from B06) by the blue squares and the local ULIRGs (Armus et al. 2007) by the green crosses. All measurements have been made in a consistent way (see Section 4.4.1). Measurements of both the 6.2 and 7.7 yum PAH lines are not available for all SMGs. CI stands out again as having a very small 6.2 /xm EW, below the value of EW 6 . 2 = 0.2 (vertical dashed line) which is often used to separate AGN dominated and SB dominated systems. 4.7. PAH LUMINOSITIES 124 sample in S07, if I decrease their values of £ W 7 . 7 by a factor of 4 to account for the different measurement methods. With EW6.2 values of > 0.5 and r 9 . 7 ~ 1, high redshift SMGs lie in region '2C' of the diagnostic plot in Spoon et al. (2007). Sources in this region are classified as PAH-dominated and, not surprisingly, M82 also falls here. Utilizing ISO spectra of SBs, ULIRGs and AGN, Genzel et al. (1998) used the ratio of the 7.7 yum PAH line to the continuum, l/c, as a diagnostic for AGN contribution to the mid-IR emission. In general, galaxies with l/c > 1 are classified as SB and those with l/c < 1 are AGN-dominated. As I stressed in Section 4.4.1, this type of analysis is strongly dependent on the specific choice made for the continuum fitting procedure. Nevertheless, I calculate l/c ~ 2.3 for the composite SMG and I note that all SMGs in the sample satisfy the l/c > 1 criterion for SB dominated sources. This is true even for CI, which clearly shows a strongly rising continuum indicative of the presence of an AGN. I note from Fig. 4.15, however, that the 6.2 jim PAH line may be more useful in weeding out AGN-dominated sources. In Section 4.6,1 derived the total IR luminosities by fitting templates to the IRS spectrum, together with the submm and radio photometry. I used a scaled Mrk231 template to determine the amount of IR luminosity coming from the AGN. This leaves a value of Lf^, the total IR luminosity of a galaxy coming from star formation activity. These estimates of ZfR are listed in Table 4.6. Now that I have an estimate of the IR luminosity free of any AGN contamina-tion, I can explore the connection between the individual PAH luminosities and the infrared luminosity. Both processes, dust and PAH emission, are expected to trace the star formation in galaxies, and so I anticipate that they will be correlated. In Fig. 4.16 Iplot the PAH line luminosities as a function of L I R for SMGs, local SBs and local ULIRGs. For the SMGs, I plot the Lf^ coming from star formation. In all panels, the straight line is the best-fit to the local SBs and it is consistent with the fit to both the SBs and the SMGs. The best-fit relation to both the SMGs and SBs is given by log(£e.2) = ( 1.8 ± 0.6) + (0.93 ± 0.05) x log(Lm) log(L7.7) = ( 0.3 ± 0.6) + (0.83 ± 0.05) x log(LIR) 4.7. PAH LUMINOSITIES 125 SFRdVIeyr1) 1 10 100 1000 Figure 4.16: Correlations between L I R and PAH luminosities. For the SMGs, larger dots indi-cate the L i R once the AGN contribution has been removed. The lines show the best fit to the SBs. SMGs clearly follow the same relation established for local SBs and the ULIRGs are systematically below the lines. 4.7. PAH LUMINOSITIES 126 log(In.3) = (-1.4 ± 0.8) + (0.89 ± 0.07) x log(LIR). These parameters also show that the relation between the PAH line luminosity and Lm is similar, within the errors, regardless of which PAH line is used. Again, the slopes are close to one indicating a direct proportionality between LiR and LPAH- If I assume a slope of one, the relation becomes L1K ~ 550 x L 6 . 2 220 x L7.7 ~ 680 x L 1 L 3 . (4.4) The first thing to note here is that the SMGs lie on the relation established for local SB galaxies extrapolated to very high luminosities. This is what I expect if both are dominated by the same emission mechanism, namely star formation. B06 also showed that the PAH line luminosities are correlated with the IR luminosities within their SB sample and Schweitzer et al. (2006) showed that L7,7 is loosely correlated with the 60 pm luminosity for a sample of Quasi-Stellar Objects (QSOs). Peeters et al. (2004) showed a correlation between L F i R and L 6 .2 for a range of scales of star formation, going from local massive star-forming regions to normal and SB galaxies, but they also note that PAHs may be better suited as tracers of B stars than massive star formation. The fact that the SMGs lie on this relation demonstrates the reliability of the full SED fitting to derive the total infrared luminosity and confirms that SMGs do have incredibly high infrared luminosities powered primarily by star formation. Note that CI does not stick out as an outlier in these plots, since I have removed the AGN component of its L I R . Again I find that most of the local ULIRGs deviate from this relation; this is most striking for the 6.2 and 11.3 pm panels. The ULIRGs which are known to be SB-dominated from Armus et al. (2007) are among the closest to the line. No attempt has been made to remove the contribution of AGN from the L I R estimates for the ULIRGs, and therefore this plot is probably showing that ULIRGs contain a higher AGN contribution to the IR luminosities or more extinction in the mid-IR. A high AGN contribution will causes the PAH lines to be suppressed by the strong continuum and larger amounts of silicate absorption and the L i R will also include a larger contribution from the AGN. Correcting for both of these effects will 4.7. PAH LUMINOSITIES 127 move the ULIRG points much closer to this correlation. The continuum component diluting the RAHs in ULIRGs could also arise from hot dust in energetic HII regions, in which case it would also be powered by star formation but under different environments than those observed in the SMGs. It is worth noting that in the centre panel of Fig. 4.16, the ULIRGs are not as far off the SB and SMG relation. This is likely due to the fact that the 7.7 pm PAH lines are much harder to measure, because of blending with other nearby emission lines and Si absorption at 9.7 pm. I found that the latter affects the spectra of ULIRGs more than that of local SB or high redshift SMGs. The second thing which stands out in Fig. 4.16 is that there is significant scatter in this relation for both the low redshift SBs and the high redshift SMGs. This scatter is considerably higher than would be implied by the measurement uncertainty. Other properties of individual objects, such as geometry of the SB region, spatial distribution of the dust, viewing angle relative to a merging event, dumpiness of the star formation, etc. are likely to play a role in this scatter. I investigated whether the scatter might arise from different dust temperatures (e.g. warm vs cool as defined by IRAS colours, Sanders et al. 1988b) but I did not find any significant trend. In addition, I found no correlations between the IR luminosity and equivalent widths of the PAH lines in SMGs. This is consistent with what was found in B06 for the local SB galaxies, and shows that both the mid-IR continuum and the PAH strength scale together with increasing L I R . I also find no correlation between X-ray luminosity and L I R (or the PAH luminosities), a correlation which would have been expected had the IR emission been dominated by an AGN. Assuming that the PAH and dust emission are tracing the star formation, I can use the Kennicutt (1998) relation between L I R and SFR and the above PAH line luminosity to L I R cor-relations to determine the relation between SFR and the PAH line luminosity. I chose to focus on the 6.2 pm PAH line here, since it is the least affected by line blending and Si absorption. I find that given the PAH luminosity, one can calculate the SFR using: S F R [ M 0 y r - 1 ] ~ l O - 7 x L 6 . 2 [L0]. 4.8. DISCUSSION 128 Similar (although perhaps slightly less accurate) relations can be found for L 7 . 7 and L 1 L 3 using Eq. 4.4. Recall that because of the large amount of scatter seen in Fig. 4.16 the resulting SFRs will only be good to roughly half an order of magnitude. That being said, in the absence of any far-IR/submm data, it is useful to be able to put constraints on the SFRs using the only the PAH line luminosities. 4.8 Discussion In the previous sections, I have shown from the mid-IR spectra of high redshift SMGs that these systems are star-burstdominated and contain a negligible contribution from the AGN. Only a small fraction of SMGs (2/13) contain an AGN which contributes significantly to the mid-IR luminosity, and when you extrapolate to the total IR luminosity, all SMGs are dominated (> 50% of L I R ) by star formation and not AGN activity. X-ray observations suggest that most SMGs may harbour an AGN. However, I have found that that they are not important for the bolometric luminosity. I also showed that, although ULIRGs are-often thought to be the local analogues to SMGs, their mid-IR spectra can be quite different. In this section I explore why these two populations may be different and what this might indicate about SMGs in the context of galaxy evolution. Using the usual Kennicutt (1998) relation between LjR and SFR, I can calculate the SFR for each SMG. I use the L f R value, since it does not include any emission from the AGN. Fig. 4.16 shows the corresponding SFR on the top axis. The median Lf^ for the sample is 6 x 10 1 2 L o , which corresponds to an SFR of around 1000 M 0 y r - 1 . This is an extremely high rate of star formation, capable of creating a stellar mass of 1 O U M 0 in only 108 years. A galaxy is not able to sustain this level of SF forever and together with the fact that the number density of SMGs is quite low, this leads to the suspicion that SMGs are a phase in the evolution of massive galaxies. SMGs are only one of several populations of high redshift galaxies which are ultraluminous in the infrared. Many galaxies selected in deep 24 yum surveys also qualify as ULIRGs (Daddi 4.8. DISCUSSION 129 et al, 2005; Yan et al. 2005, 2007). These other ULIRGs are generally not detected at submm wavelengths. What then are the fundamental differences between various populations of high redshift ULIRGs? One of the main factors which could cause a bifurcation in the population of ULIRGs at high redshift is the presence or absence of a significant AGN. The evolutionary scenario proposed by Sanders et al. (1988a) shows massive galaxies going through several main stages en route to becoming a massive elliptical. The process starts with an IR luminous phase, most likely triggered by a massive merger. During this stage of intense star formation, the AGN is also growing. As the AGN becomes larger it begins to feed back on the galaxy, eventually quenching the star formation completely by blowing off all the remaining dust and gas. Thus begins the QSO phase, where the AGN is free to dominate the emission. After exhausting it fuel, eventually the QSO settles down and we end up with a quiescent massive elliptical galaxy. In this scenario, the submm emission will be at a maximum during the initial IR luminous phase (e.g. Springel, Di Matteo & Hernquist 2005). However, one uncertainty in this scenario is the timescales. In particular, how long is the IR luminous phase, how long does it take the AGN to develop and can there be multiple episodes of ULIRG and QSO activity? Since local ULIRGs show a mix of AGN and SB activity they fit into this evolutionary scenario at the stage where the AGN has developed but before it has had a chance to blow off all the gas and dust. Local ULIRGs also show messy morphologies indicative of recent mergers (e.g. Sanders & Mirabel 1996; Rigopoulou et al. 1999). Since SMGs show less of an AGN contribution, it is likely that they represent a slightly earlier phase of this scenario, before the AGN has had time to develop fully and is signifiant to the IR luminosity. This is also consistent with the findings that SMGs have cooler dust temperatures than local ULIRGs, as the AGN is not strong enough to begin to heat the dust and destroy the PAH molecules. Other high redshift ULIRGs, like those detected at 24 //m (e.g. S07), must have warmer dust temperatures, otherwise they would be detected in the submm. Therefore they are more like local ULIRGs and in this evolutionary scenario represent the stage after the SMG phase. For local ULIRGs, there is evidence which shows that the luminosity of CO emission decreases as the merger progresses (Rigopoulou et al. 1999), which is consistent with decreasing submm emission in 4.8. DISCUSSION 130 the later stages of the merger. This idea is also consistent with morphological studies which show that SMGs are often very extended and messy looking (Chapman et al. 2003b; Conselice et al. 2003; Pope et al. 2005). Alternatively, SMGs and other high redshift ULIRGs might appear different because of different merger progenitors. Since the number density of SMGs is much less than that of 24 /mi-selected IR-luminous galaxies, then this phase in the evolutionary sequence must be shorter. I can make a rough estimate for the duration of the submm luminous phase. In the 2Gyrs between 1.5 < z < 3 there are roughly 20 SMGs and 500 24 /mi-selected galaxies (with Lj.R > 10 n L o ) within a portion of the GOODS-N field. Assuming that all IR-luminous 24 /im-selected galaxies will undergo a submm luminous stage during this period of time, I calculate the duration of the submm luminous phase to be on the order of 108 yrs12. As mentioned above, this timescale is what is needed to create a galaxy with a stellar mass of ~ 1 O U M 0 (Borys et al. 2005) at a rate of lOOOM 0yr _ 1. This timescale is consistent with those derived from gas masses of SMGs (Greve et al. 2005). Rigopoulou et al. (1999) looked at whether the stage of the merger in local ULIRGs was correlated with the dominant energy source (SB or AGN), but they only looked at separations of up to 10 kpc and were also limited by low number statistics. The apparently high occurrence of double counterparts to SMGs (see Chapter 3; Ivison et al. 2007) indicates that many could be merging systems with separations between the merging components of several 10s of kpc. Such separations seem consistent with the galaxy evolution models of Springel, Di Matteo & Hernquist (2005), for example. Of course, in reality I expect this evolutionary picture to be only a basic description of the average SMG properties. Individual SMGs will have a variety of detailed properties because of the specific merging histories as well as the geometry of the star-forming regions or merg-ing clumps. This is seen through the scatter among the properties of the SMGs, even in this modest-sized sample. One could imagine learning more from this scatter in future studies with 12It is not clear what fraction of 24 /on-selected galaxies will undergo a submm-luminous phase since SMGs are known to be massive and could represent only the most extreme galaxies. Nevertheless, this calculation provides a useful reality check on the expected duration of the SMG phase. 4.9. CONCLUSIONS 131 significantly larger samples. There are several complications with this emerging picture of SMGs. The disturbed mor-phology and cold dust observed in SMGs (Chapman et al. 2003b; Conselice et al. 2003; Pope et al. 2005; Chapter 3) seems to indicate that they are spatially extended, which would make them optically thin in the optical and submm. However, the dust corrected UV luminosities of SMGs underestimate the SFRs (C05) which suggests that these systems are optically thick. In a compact, optically thick SB, one would expect to see substantial silicate absorption. However, the IRS spectra of SMGs show only modest Si absorption. If SMGs really are a snapshot of a massive merging event, then one would expect them to have a very complicated geometry, in which case there would be patchy regions of star formation, some which are optically thin and some which are optically thick. Our impression of SMGs will be based entirely on the viewing angle of the disturbed system and therefore, for individual objects, we may not be seeing the full picture. As we collect much larger samples of these extreme objects in the future, with surveys like SCUBA-2 and high resolution imaging with the Atacama Large Millimeter Array (ALMA), we can make progress on understanding the detailed physical conditions and spread of properties within the SMG population. 4.9 Conclusions Spitzer IRS spectroscopy has been obtained for a sample of 13 SMGs brighter than 200 piy at 24 pm. This has effectively confirmed the identification of these SMG counterparts, and in some cases has provided a new redshift estimate. The SMGs show strong PAH emission and, on average, very little in the way of a rising continuum. I have explored several diagnostics from the mid-IR spectra of SMGs to determine the level of AGN contributing to the luminosity at these wavelengths. All of them seem to converge on a picture in which typical SMGs are SB dominated systems with at most a 30% contribution from an AGN at mid-IR-wavelengths. Their mid-IR spectra are similar to a scaled spectra of local SB galaxies like M82. " The classification of an SMG as AGN or starburst-dominated from the mid-IR often dis-4.9. CONCLUSIONS 132 agrees with the classification from the X-rays. While X-ray observations are better suited for determining the presence of an AGN in an SMG, the mid-IR spectra can determine how im-portant the AGN is to the total infrared luminosity since it is directly detecting the hot dust emission and not subject to details about the geometry Full IR SED fits to the IRS spectra and mid-IR through to radio photometry show that SMGs are best fit by scaled up versions of local IR-luminous galaxy templates with additional extinction from the Draine (2003) extinction curves. These models have cool dust temperatures (T ~ 32 K) and high L I R s which imply SFRs of - 1000 Moyr" 1 . SMGs lie on thelocal relation between L m and L 7 . 7 (or L 6 2 or L n , 3 ) , which means that the PAH line flux can be used to estimate the SFRs in these systems, albeit with large uncertainties. Equivalent widths, which have been the focus of some other studies, are much less useful in this regard. . SMGs are consistent with being a stage in the evolution of massive galaxies. This is an ear-lier phase than other high redshift ULIRGs, which show warmer dust and less PAH emission, presumably because the AGN has not yet become strong enough to start heating the dust and destroying PAHs. I have been able to put constraints on the contribution from AGN activity to the restframe 5-12 Aim mid-IR emission, and have extrapolated this to far-IR wavelengths. Without more data points from 30-100 fim in the rest frame it is difficult to directly probe the emission at far-IR wavelengths. We can begin to achieve this with deep 70 ^ m surveys with Spitzer (e.g. the Far-Infrared Deep Extragalactic Legacy, FIDEL, Survey, PI M. Dickinson), and we will be able to model this part of the spectrum in more detail once data from the Herschel Space Observatory13 and SCUBA-2 (Holland et al. 2006) become available. Details of the SMG phase will await the spatially resolved spectroscopy obtainable when A L M A 1 4 comes online. 1 3 h t t p : / / h e r s c h e l . e s a c . e s a . i n t / 1 4http://www.eso.org/proj ects/alma/ C H A P T E R 5 133 T H E STORY OF GN20 Enormous effort has gone into determining the redshift distribution of SMGs through opti-cal spectroscopy and photometric redshift estimates (Chapman et al. 2005; Pope et al. 2005; Aretxaga et al. 2007). All of the different methods seem to converge on a redshift distribution which peaks around 2-2.5, with very few sources at redshifts less than 1.0 and greater than 3.5. In fact, currently the highest spectroscopically confirmed blank-field SMG is at 3.623 (Chapman et al. 2005). Due to the negative K-correction at submm wavelengths (see Blain et al. 2002), SMGs should be visible out to z ~ 10 if they existed at this time. However, at very high redshifts optical/near-IR spectroscopy of SMGs becomes increasingly difficult, due to both distance dimming and the high dust obscuration in.these systems. The redshift distribu-tion of the essentially complete sample of SMGs in the GOODS-N SCUBA 'super-map' seems to imply that the fraction of 850 pm selected galaxies at z > 4 is below 14% (see Chapter 3). However there is evidence suggesting that galaxies selected at longer wavelengths (~ 1 mm) may have a higher fraction of sources at z > 4 (e.g. Eales et al. 2003 and preliminary studies with the Astronomical Thermal Emission Camera - AzTEC). The hunt for SMGs at the highest redshifts remains one of the outstanding issues in submm astronomy. In this chapter, I present the identification and redshift for the most distant SMG and the first direct probe of a high redshift tail of SMGs. I present extensive follow-up observations of GN20 at other wavelengths, including high resolution Institut de Radio Astronomie Mil-limetrique (IRAM) Plateau de Bure (PdB) and the SubMillimeter Array (SMA) interferometry and optical spectroscopy. This bright SMG has led to the discovery of a high redshift galaxy cluster in the foreground. I present spectroscopic redshifts for the galaxies which are thought to form a rich cluster lensing the submm system. These results have important implications for future wide areas submm surveys and demonstrate the potential for these surveys to identify high redshift clusters. 5.1. FOLLOW- UP OBSERVATIONS OF GN20 134 Note that not all of data presented in the chapter was reduced and analyzed as part of this PhD thesis, however I summarize all of the existing data here for completeness. 5.1 Follow-up observations of GN20 GN20 was discovered in SCUBA imaging taken in June 2003 as part of the continuing JCMT programme to fill in the data deficient regions in the GOODS-N field. GN20 was first published in Pope et al. (2005) and its secure counterpart was first discussed in Pope et al. (2006) and Iono et al. (2006). Because of its unusually high submm flux, GN20 quickly became the target of several multi-wavelength follow-up campaigns. In particular, it was targeted with the IRAM PdB and SMA interferometers to localize the submm emission. 5.1.1 IRAM continuum imaging The six antennas of the IRAM PdBI were used to image the 1.3 and 3 mm continuum of GN20. The primary beam of the antenna is about 55 arcsecs and 20 arcsecs at 86.2 GHz and 240 GHz, respectively. The observations covered approximately 13.6 hrs on-source at 3mm and 13.4hrs at 1.3mm (after flagging low-quality visibilities) for the central pointing at ct=12:37:11.17, 5=62:22:10.0. All these observations were carried out in the D configuration between 17 and 20 May 2004. 1044+719,1458+718 and 1150+497 were used as phase and amplitude calibrators, and 0716+714, 1749+096, 3C273 and MWC349 as bandpass calibrators and to define the flux density scale. The flux densities at 3 mm are accurate to better than 10%, and to better than 20% at 1.3 mm. Data were reduced by J.-P. Kneib. Natural-weighted dirty maps were made of the contin-uum emission at 3 mm and 1.3 mm by uw-averaging over all the channels. These were cleaned using the Clark algorithm, restored with a Gaussian beam of 3.9" x 2.0 arcsec with a posi-tion angle of 35° at 1.3-mm, and 9.2" x 5.4 arcsec with a position angle of 38° at 3-mm, and corrected for primary beam attenuation. See Kneib et al. (2005) for a description of the data reduction procedure for a similar data-set. 5.1. FOLLOW-UP OBSERVATIONS OF GN20 135 J — i — i — i — I — i — i i i I i i i i i • • • • Figure 5.1: IRAM PdB 1.3 mm image of GN20. Contours are from the 1.4 GHz VLA radio image and are shown at 3, 4 and 5a. The inset in the bottom right corner shows the beam at 1.3 mm used to restore the IRAM image. The counterpart to GN20 is clearly detected at 5a in the same position at both wavelengths. 5.1. FOLLOW-UP OBSERVATIONS OF GN20 136 The absolute precision of the astrometry of the maps is estimated to be 0.3", sufficient to accurately locate the continuum emission relative to sources in the HST imaging which has an absolute astrometry at the 0.2" level. GN20 is detected at 1.3 mm localizing the position to a =12:37:11.17, 6 =62:22:11.9 (Fig. 5.1). The estimated flux density is 3.5 ± 0.7 mJy. GN20.2 is outside the primary beam at 1.3mm, so no detection is possible with the IRAM PdB data to confirm the position of its counterpart. 5.1.2 Submillimeter Array observations GN20 was observed with the SMA at 890 /j,m for 4 tracks in the period February to April 2005. These observations and data are presented in Iono, Peck, Pope et al. (2006). GN20 was de-' tected at high significance, confirming the position obtained with the IRAM PdB. The position uncertainty from the SMA observations is ~ 0.1 arcseconds in RA and DEC. Again, GN20.2 was beyond the half power point of the SMA primary beam and therefore is undetected in the SMA image. GN20 is unresolved in the SMG image which implies a size of < 1.2 arcseconds. The accuracy of the position from the IRAM and SMA observations allow me to unam-bigously identify the optical and infrared galaxy counterpart to GN20 (see Section 5.2). 5.1.3 SHARC-II imaging GN20 was observed with the Submillimeter High Angular Resolution Camera (SHARC-II) on the Caltech Submillimeter Observatory (CSO) on 22 April 2005. The weather was very good: T~225GHZ ~ 0.04. The integration time was 1.5 hours at 350 //m and 1.0 hours at 450 ^m. I used a small scan pattern centred on the counterpart of GN20. Data were reduced by D. Dowell using his own reduction scripts. Despite the fantastic observing conditions, GN20 and GN20.2 were undetected in the deep SHARC-II maps. I extract the flux at the IRAM/SMA counterpart position as well as 3a upper limits. 5.1. FOLLOW-UP OBSERVATIONS OF GN20 137 3 0 - 5 Figure 5.2: SMA image of GN20. From left to right the images in the three panels are; SMA 890 /im, IRAC 3.6/im and HST V-band. Note that each image is on a different scale. Contours in all panels are from the SMG image: in (a) they start from 2a for the lowest contour and increase by la, while for (b) and (c) the 4, 6 and 8<r contours from (a) are plotted. The SMA position is coincident with a radio and 24 /xm counterpart which I identified in Chapter 3. This figure is published in Iono, Peck, Pope et al. (2006) and reprinted by permission of the American Astronomical Society. 5.1. FOLLOW- UP OBSERVATIONS OF GN20 138 5.1.4 Millimetre imaging The whole GOODS-N field was imaged at 1.2 mm with MAMBO on the IRAM 30m telescope in Winter 2006. Full details of the observations and data analysis are presented in Greve et al. (in preparation). GN20 is the brightest source in the map and is detected at > 10cr. GOODS-N was also imaged at 1.1 mm with AzTEC on the JCMT in Winter 2006. Full de-tails of the observations and data analysis will be presented in Perera et al. (in preparation) and Chapin et al. (in preparation). Not surprisingly, GN20 is again detected at high significance. The millimetre fluxes of GN20 are listed in Table 5.1. 5.1.5 GMOS optical spectroscopy Armed with an accurate position for the submm emission from the IRAM and SMA obser-vations, I targeted the counterpart of GN20 with GMOS spectroscopy on the Gemini North telescope. I used the B600 grating centred on a wavelength of 5500A which provided spectral coverage from 4100-6900A. This is an optimal range for detection of Lyman a emission for galaxies in the range of z=2.5^ 1.5. The slits were 1 arcseconds wide. I binned the data in both the spectral and spatial directions for final resolutions of 1.8A and 0.145 arcseconds, respec-tively. I took 13 exposures of 30 minutes each, between May and June 2005, for a total on source integration time of 6.5 hours. Weather conditions were poor, with seeing ranging from 0.8-1.7 arcseconds. Initial data reduction, including bias subtraction, flatfielding and wavelength calibration, is performed using scripts in the Gemini Image Reduction and Analysis Facility (IRAF) pack-age1. The sky subtraction, spectral extraction and co-adding of the ID spectra are performed using my own IDL scripts. For the sky subtraction, I characterize the sky emission as a function of wavelength using the off source positions in the slit and subtract this from the 2D science files. For optimal extraction of this faint target, I collapse (sum) the spectrum along the spectral direction to increase the SNR. I then fit a Gaussian to the total signal across the slit to define the extraction window. The width of the window varies depending on the weather conditions 'http://www.gemini.edu/sciops/instruments/gmos/ 5.2. COUNTERPART OF GN20 139 of each observations and is on average ~ 14—17 pixels (out of 36 along the length of the slit), which corresponds to ~ 2 arcseconds. I use a boxcar extraction window, since this is better for faint sources, to extract the ID spectrum for each of the 13 exposures. 1 then use a 3<7-clipped median to combine these into one final spectrum for GN20. In addition to GN20, the GMOS mask contains 24 other galaxies with colours consistent with z ~3-5. I also targeted several sources with known spectroscopic redshifts in the range of z — 3-4 to check the consistency of the data. I confirm the redshifts of several galaxies from the Team Keck Redshift Survey (TKRS2) sample (Wirth et al. 2004). 5.1.6 Keck optical spectroscopy Additional optical spectroscopic observations of GN20 were made with the DEep Imaging Multi-Object Spectrograph (DEIMOS) on the Keck II telescope in February 2007. The expo-sure time was 9000 seconds in good weather. Data were taken and reduced by S.C. Chapman. 5.2 Counterpart of GN20 GN20 provides the rare situation of having a high SNR detection at submm wavelengths. While most SCUBA galaxies hover around the detection threshold of 3.5-4.0a, GN20 is detected at 10cr. Given that the positional uncertainty is supposed to scale as FWHM/SNR (see Chapter 3 and Ivison et al. 2007), this implies that the counterpart to GN20 should not be much further than 1-2 arcseconds from the peak of the SCUBA emission. Fig. 5.1 shows the 1.3 mm IRAM image of GN20 with dark colours indicating a positive detection. The Gaussian beam used to restore the image is shown in the bottom left corner. The contours on Fig. 5.1 are from the re-reduced VLA 1.4 GHz radio image (Morrison et al. in preparation). Although there was no detection of GN20 in the original Richards (2000) radio catalogue, it is clearly detected here with a flux of 70 /Jy. The offset between the VLA and IRAM positions is 0.2 arcseconds, which is well within the positional uncertainty of each. The 2http://tksefver.keck.hawaii.edu/tksurvey/ 5.2. COUNTERPART OF GN20 140 Figure 5.3: i-band and IRAC 3.6/xm images of GN20. Each image is 5 arcseconds on a side. Contours are the same radio contours shown in Fig. 5.1. While the optical emission is offset by 0.5 arcseconds from the radio emission (which is beyond the offset expected from astrometry uncertainties), the IRAC emission is only offset by 0.2 arcseconds. Both the optical and radio emission are extended in the N-S direction. 5.2. COUNTERPART OF GN20 141 Table 5.1: Positions and fluxes of counterparts to GN20. All GOODS positions have been corrected for the known offset of -0.38 arcseconds in declination. CSO SHARC-II values are not detections but simply measurements of the flux at the position of the radio counterpart. ACS fluxes are measure through isophotal apertures. Wavelength Instrument/Telescope RA DEC Flux 1.4 GHz VLA 12:37:11.86 62:22:11.92 70.0 ± 16.3 pJy 1300 Aim IRAM PdB 12:37:11.83 62:22:11.9 3 .5±0.7mJy 1200/mi MAMBO/IRAM 30m 12:37:11.7 62:22:11 10.0 ±0 .95 mJy 1100 ^ m AzTEC/JCMT 12:37:1.2.1. 62:22:12 10.0 ± 0 . 9 3 mJy 890 pm SMA 12:37:11.92 62:22:12.10 22.9 ± 2.8 mJy 850 ^m SCUBA/JCMT 12:37:11.7 • 62:22:12 20.3 ± 2.1 mJy 450 SHARC-II/CSO 1 0 ± 7 m J y 350 pm SHARC-II/CSO 18 ± 8 mJy 70 pm MlPS/Spitzer < 3.3 mJy 24 pm MIFS/Spitzer 12:37:12.0 62:22:11.9 68.9 ± 4.8 pJy 8.0 yum IRAC/Spitzer 12:37:11.88 62:22:12.1 25.28 ± 1.48/Jy 5.8 pm IRAC/Spitzer 15.93 ± 1.28/uJy 4.5 pm IRAC/Spitzer 9.23 ±0 .79 yuJy 3.6 pm IRAC/Spitzer 6.79 ± 0.55 pJy 2.2 pm Flamingos/KPNO 12:37:11.90 62:22:12.2 3.13 ± 0.73 pJy 0.850 yum ACS/HST 12:37:11.812 62:22:12.22 0.492 ± 0.018 pJy 0.775 pm ACS/HST 0.409 ± 0.015 pJy 0.606 pm ACS/HST • 0.194 ± 0.009 pJy 0.435 pm ACS/HST 0.011 ± 0.008 AiJy 5.2. COUNTERPART OF GN20 142 SMA position of GN20 is also consistent with these positions. The separation between the radio/IRAM counterpart and the submm position is 1.4 arcseconds, which is approximately equal to the FWHM/SNR. The accuracy of the position from the IRAM, SMA and radio observations allow me to unambigously identify the optical and infrared galaxy counterpart to GN20. Fig. 5.2 shows the high resolution SMA imaging overlaid on the IRAC 3.6 pm and F-band images. The SMA contours are clearly offset from the disturbed optical system beyond what you expect from the uncertainties in the astrometry. Fig. 5.3 shows the radio contours overlaid on the HST ACS i-band image (left panel) and the Spitzer IRAC 3.6 pm image (right panel). The centre of the radio emission is offset from the peak optical emission by 0.5 arcseconds. However, the IRAC and MIPS emission is closer, with an offset of only 0.2 arcseconds relative to the radio position. Another thing to note from this figure is that both the optical and radio emission look extended in the N-S direction. I discuss this further in Section 5.6, when I investigate a lensing scenario for this SMG. GN20 is also detected in the AzTEC and MAMBO millimetre imaging of GOODS-N. These independent surveys both converge on a ~1 mm flux of about 10 mJy for GN20. This is substantially higher than the 3.5 mJy detected by IRAM PdB at 1.3 mm. However, the latter are interferometry observations and since GN20 looks like an extended system, some fraction of the flux could be resolved out by the IRAM PdB. Furthermore, the calibration of these observations are known to be problematic. Regardless of the reason, I go with the majority and proceed with a millimetre flux for GN20 of ~10 mJy. There is no X-ray source within the search radius of GN20, however there is a very bright X-ray source located only 20 arcseconds to the west. GN20 is not detected at 70 pm, with an upper limit on the flux of 3.3 mJy (M. Huynh, private communication). The positions and fluxes from all multi-wavelength data available for GN20 are summarized in Table 5.1. 5.3. REDSHIFT OF GN20 143 5.3 Redshift of GN20 5.3.1 Photometric redshift GN20 was discovered in the on-going SCUBA imaging campaign of the GOODS-N field (see Chapter 2). While this SMG is 20mJy (10a) at 850 ^m, it remains undetected at 450 ^m, which provides the first clue to it lying at very high redshift, certainly z > 2 for any reasonably SED. The optical photometry can be used to predict the photometric redshift of GN20. The high B — V colour of the galaxy counterpart places it in the z = 3.3 — 4.3 73-dropout range for the Lyman-break technique (Steidel et al. 1998). Using the full range of UV/optical/near-IR photometry available in GOODS-N, the photometric redshift of GN20 has been estimated us-ing both the x2 minimization technique and the Bayesian method (Mobasher et al. 2004, see Pope et al. 2005 for more details on how the photometric redshifts for SMGs in GOODS-N were estimated). The photometric redshift for GN20 is 3.9 (note that in Pope et al. 2006a I reported a lower value of 2.95, however this estimate did not include the Spitzer photom-etry). The optical/near-IR photometric redshift for GN20 has been calculated independently by another group using different methods, also converging on the same solution (J. Dunlop, priv. communication). The optical photometric redshift is consistent with the redshift implied from the low radio to submm ratio (e.g. Carilli & Yun 1999) and from the lack of detection at 350 and 450 /xm with SHARC-II. While far-IR/radio photometric redshifts can be uncertain to ±0.5 (Hughes et al. 2002), the simple lack of detection of this bright source at 350 and 450 pm indicates that it must be at z > 3. Interestingly, I derive the same photometric redshift of 3.9 for GN20.2 (Pope et al. 2005). I will discuss the companion source in more detail in Section 5.5. 5.3.2 Spectroscopic redshift Fig. 5.4 shows the GMOS spectrum for GN20. This spectrum is too noisy (due to poor weather conditions) to show any emission or absorption lines. However, the continuum is detected and 5.3. REDSHIFT OF GN20 144 4500 5000 5500 6000 6500 Observed wavelength (angstroms) Figure 5.4: GMOS spectrum of GN20 smoothed to a resolution of 10A for clarity. The red curve at the bottom is the sky signal and this has been offset in the y-direction for clarity. The green dashed line shows a best-fit step model, and the main spectral lines redshifted to z=3.955 (see next section) are indicated by the blue vertical lines. there is evidence for a break in the spectrum around 6000A. I fit the spectrum to a model which has two flat continuum levels with a break. I solve for 3 parameters: the height of each line and the position of the step. The best-fit model is shown as the green dashed line in Fig. 5.4 and has the step at 6140A which is consistent with z ~ 4 if it is the break just bluewards of the Lyman a emission line. A decrease in the flux blueward of Lyman a emission is expected due to absorption of the Lyman a photons by intervening neutral hydrogen between us and the galaxy. This signature in the spectra of faint high redshift galaxies has been used many times before to assign redshifts (e.g. Spinrad et al. 1998). The positions of the main expected spectral features redshifted to 3.955 are indicated by the vertical blue lines in Fig. 5.4 -1 do not claim a redshift based on detection of any of these lines. Fig. 5.5 shows the final Keck DEIMOS spectrum from 5900-7900 A. Based on absorp-tion lines from Lya, NV, SilV and Sill/CIV, a redshift of 3.955 is calculated for GN20. The redshift of GN20 based on the Keck spectrum is among the bottom 30% of absorption line 5.4. THESEDOFGN20 145 identifications made in Chapman et al. (2005) for SMGs. The agreement between the spectro-scopic redshift and the photometric redshift estimates increases our confidence in the faint Keck spectrum. Spectroscopic confirmation of other lines such as CO emission lines are needed to guarantee the redshift of GN20. This spectroscopic redshift estimate is in very good agreement with the photometric red-shift estimates and therefore 1 am confident that the redshift of GN20 has been successful de-termined. At this redshift, the size of the SMG from the SMA observations (< 1.2 arcseconds) implies a physical size of < 11 kpc. GN20 is currently the highest redshift SMG detected with SCUBA which has spectroscopic confirmation. To put GN20 into context I compare it to HDF850.1, the brightest SCUBA source detected in the central HDF (S^so = 6-7 mJy, Hughes et al. 1998) and, prior to GN20, the most studied blank-field SMG. After extensive follow-up with IRAM PdB, MERLIN and near-IR imaging, the counterpart to HDF850.1 is thought to lie behind a foreground elliptical galaxy very near the submm position, making further study of its multi-wavelength properties very difficult (see Dunlop et al. 2004, and references therein). HDF850.1 is also thought to lie near z ~ 4, although this redshift estimate is much less certain than that of GN20, since it relies solely on fitting the far-IR through radio photometry, and moreover, additional prospects for constraining the redshift are very limited. Therefore GN20 presents the best current opportunity to probe the high redshift, high luminosity tail of SMGs. 5.4 The SED of GN20 Armed with a plethora of multi-wavelength detections and redshift information, I am well po-sitioned to plot a full spectral energy distribution for this galaxy and use its shape to determine key physical parameters. 5.4. THESEDOFGN20 146 7000 7200 7400 7600 7800 Observed wavelength (Angstroms) Figure 5.5: Keck DEIMOS spectrum of GN20. The spectrum has been smoothed to a resolu-tion of 10A for clarity. The red curve at the bottom is the sky signal and has been slightly offset in the /^-direction. The shaded regions represent areas where there are bright sky lines which may contaminate the spectrum of GN20. The main spec-tral lines are shown in blue. The identification of the redshift for GN20 is based on absorption lines for Lya, NV, SilV and Sill/CIV. 5.4. THESEDOFGN20 147 10 100 1000 10000 Rest Wavelength (jam) Figure 5.6: IR-radio SED of GN20. The blue curve is the best-fit modified blackbody model, while the red curve is the best-fit modified CE01 template. Both models agree well at A > 100 pm, while at shorter wavelengths the CE01 template undoubtedly provides a better estimate of the shape in the mid-IR, although I currently do not have data to constrain this portion of the SED. The total infrared luminosity of GN20 is ~ 1 0 1 3 L o implying an SFR of around 2000 Moyr"1. These values have not been corrected for possible lensing magnification. 5.4. THE SED OF GN20 148 5.4.1 Dust properties I start with the far-IR dust peak. This part of the SED is often characterized by a modified blackbody (or greybody) spectrum by including a v& term, where /3 is the dust emissivity. (1 has been found to range from 1-2 (Hildebrand 1983) and, due to a lack of data points, it is often set to be 1.5. I fit the 350-1200 /tm photometry of GN20 to a modified blackbody spectrum and solve for j3, temperature and normalization. The cold dust characterized by the modified blackbody template dominates the luminosity of the galaxy from ~ 50-1000 /xm. Since I am interested in obtaining an estimate of the total IR luminosity (often defined as the integrated luminosity from 8-1000 /tm), I also fit the data to the CE01 models. As I did in Chapter 3,1 have modified the CE01 templates by allowing for additional extinction from the Draine models and allowing them to scale independently in luminosity and temperature. I fit to all the data from.24/xm to radio to these templates. • " -Fig. 5.6 shows the IR through radio SED of GN20 and the best-fit models. The blue curve is the modified blackbody model, while the red curve is the modified CE01 template. The vertical lines indicate the region within which I measure the infrared luminosity. From the greybody fit I get L8_iooo = 8.2 x 1 0 1 2 L o and from the modified CE01 fit I get L8_i0oo = 1-3 x 10 1 3 L o . These values have not been corrected for possible lensing magnification (see Section 5.6). In the absence of any lensing magnification, GN20 would be considered a hyper-luminous infrared galaxy. The estimate of the infrared luminosity from the modified CE01 templates is 1.6 times higher than from the simple greybody fit. This difference is due to the fact that there is negligible emission at wavelengths shorter than 30 /xm in the greybody fit, whereas the modified CE01 template has significant emission from 8-30/xm. Using the formula in Kennicutt (1998) this amount of infrared luminosity corresponds to an SFR of 1400-2300 M 0 y r - 1 . Again, these values are for the case when GN20 is not lensed (discussed further in Section 5.6). Both fits converge on the same dust temperature, 30 K with f3 = 1.4. This is consistent with values found for SMGs and also consistent with the finding that SMGs do not follow the local luminosity temperature relation (see Chapter 3). Galaxies of this luminosity in the 5.4. THE SED OF GN20 149 local Universe have very warm dust emission and hence there are different physical conditions in high redshift SMGs than in local ULIRGs. At z. — 3.955, the temperature of the Comic Microwave Background (CMB) was about 13.5 K and the contrast of dust temperature relative to the C M B is lower than in local ULIRGs. Using the formula for dust mass from Dunne et al. (2000), I calculate 2 x 1 0 7 M Q . Both the dust mass and emissivity of GN20 are consistent with values found for IRAS bright galaxies (Dunne et al. 2000). The dust emissivity in IR luminous galaxies does not seem to have evolved significantly since z — 4. According to the correlation found in Dunne et al. (2000), this dust mass implies a total gas mass of ~ 1 0 1 0 M G . 5.4.2 Stellar properties Even with significant amounts of dust within it, GN20 is still detected at rest-frame UV/optical wavelengths in the deep GOODS HST images. Furthermore, at z ~ 4, the longest IRAC wavelength, 8 pm, samples the peak of the stellar bump at 1.6 pm. Therefore, I can use the HST and Spitzer photometry to investigate what stellar populations look like in a z = 4 star-forming galaxy. I use the Bruzual & Chariot (2003) stellar population models which provide stellar SED templates over a wide range of metallicities and ages. I specifically adopt the suite of Simple Stellar Population (SSP) instantaneous burst templates which range from 5 Myr to 11 Gyr. However, I restrict to the range from 5 Myr to 1.4 Gyr, since the Universe itself was only 1.6 Gyr old at z = 3.955 (for the adopted cosmology). These templates use the Padova 1994 evolutionary tracks and assume a Chabrier (2003) IMF. I assume solar metallicity, which is consistent with optical spectra of SMGs (Swinbank et al. 2004). I also have the option of using models with constant star formation, but this is not likely to be a good representation of what is going on in this extremely luminous galaxy. I fit the photometry of GN20 to a model consisting of two stellar populations - a young and an old population, and apply a different level of extinction to each. The model, S, is given by 5.4. THESEDOFGN20 150 Observed wavelength (Angstroms) Figure 5.7: Stellar population fits to the optical plus IRAC photometry. The red curve is the best-fit model from a combination of BC03 templates. The green and blue curves are the young and old stellar populations which make up the total model (red curve). 5.4. THESEDOFGN20 151 S = d (Iy e" C 3 T + c-210 e - C 4 T ) (5.1) where Iy and I0 are the intensities of the young and old stellar populations and r is the extinc-tion curve as a function of wavelength. c\ is the overall normalization and c 2 is the fraction of the total which comes from the old stellar population. c 3 and c 4 quantify the amount of extinction applied to the young and old stellar populations, respectively. I fix the age of the old population and solve for 5 parameters: age of the young population; fraction of the total population from old stars; opacity for the young stars; opacity for the old stars; and the nor-malization. I repeat this fit for different ages of the old stellar population and find the fit that gives the lowest x2- I originally tried a simpler model consisting of only one stellar population and/or one extinction parameter, but found that the x2 values were much higher. Fig. 5.7 shows the stellar SED for GN20. A l l of the ACS and IRAC points are shown in addition to the marginal Ks band detection with Flamingos. The red curve is the best-fit model template and the green and blue curves show the contribution to the total from the young and old stellar populations, respectively. The ages of the young and old stars are 25 Myr and 640 Myr, and they have extinction parameters of c 3 = 3.4 and c 4 = 10.5 (r is normalized at 550 nm). As I varied the age of the older population I found that the properties of the younger stellar population (age and opacity) were very stable. Regardless of the ages found for the stellar populations, the flux at 8 pm (1.6 pm rest frame) implies a very high stellar mass. I can calculate the total stellar mass of GN20 from the rest-frame absolute Xf-band magnitude, M#. From the best-fit SED, I calculate an absolute if-band magnitude of MK = -27.4. Using the average value of LK/M = 3.2 derived for SMGs in Borys et al. (2005), I derive a total stellar mass of 6.0 x 1 0 n M G . At this wavelength, 75% of the stellar mass comes from the old stellar population. The remaining 1.5 x 1 0 n M Q of stars are young, consistent with the idea that SMGs are massive galaxies currently undergoing intense star formation because of their high LIRs. These values are lower limits since they represent the stellar mass inferred from the optical/near-IR light which is observed, but much of the optical/near-IR light from SMGs is obscured. So instead I can calculate M ^ from the best-fit model without any extinction to give an estimate of the obscured and unobscured stellar 5.4. THESEDOFGN20 152 mass in GN20 which gives a value ~ 2 times larger. None of the parameters here have been corrected for any effects of lensing which will amplify the fluxes and derived masses. What is surprising from this analysis is the existence of a massive population of old stars at z = 4. I find that the older population dominates the rest-frame iY-band luminosity in GN20. Given the best fit parameters, 3/4 of the stars were formed at z ~ 6 (i.e. about 700 Myr before the time I observe this galaxy). At these early times, GN20 might have been like the i-dropout galaxies (e.g. Dickinson et al. 2004). Instead of invoking an old population of stars to fit the higher fluxes seen in the IRAC channels, I can also fit a power-law component. Power-law emission might be expected if there is an AGN in GN20 that is emitting strongly in the near-IR. In Chapter 4, I found that there was little evidence from the mid-IR spectra of SMGs that there is a strong AGN component contributing to the luminosity at these wavelengths. Furthermore, there is no indication from the X-rays that GN-20 contains a bolometrically important AGN. Therefore I do not expect much emission in GN20 at near-IR wavelength from an AGN. Nevertheless, I redid the stellar population fits by adding a power-law component normalized to the 24 pm flux (4.8 rest frame). The result was that I still require a significant (about 3/4 L K ) old population of stars to fit the data. The reduced x 2 values when I included the power-law component were not any better than the previous model, and so I stick with the parameters derived from the latter. I cannot rule out the possibility that a power-law component is contributing to the near-IR luminosity of GN20, however I found that if it does exist, it contributes only a small fraction and does not change the derived parameters by much. It might seem surprising that I found the old stellar population to have much more extinction than the young stars. In spiral galaxies, the old stars are in the bulge and the young star-forming regions are in the arms. However, the morphologies of SMGs are very disturbed and are possibly massive mergers in progress, so it is not surprising that there is a variation in the extinction. This might indicate that the older stars are near the centre of the SMG while the younger ones are spread over more extended regions, but a complex dust geometry would also create this effect. The fit shown in Fig. 5.7 is still not perfect but recall that this is an average over the whole galaxy. It is realistic to assume that the star-formation is clumpy in these 5.5. COMPANION SOURCE: GN20.2 153 Table 5.2: Summary of physical parameters for GN20. This values do not account for the effects of lensing magnification. Parameter Value Redshift 3.955 Temperature 30 K Stellar mass 6 x 1 O U M 0 Dust mass 2 x 1 O 7 M 0 Stellar age: young 25 Myr Stellar age: old 640 Myr L8-1000 \im 1.3 x 1 O 1 3 L 0 SFR 2000 M o y r - 1 Size (submm) < 11 kpc disturbed systems and that there is some combination, of constant star formation and bursts with a continuum of stellar ages. I also caution the reader that the use of stellar population models for studying very high redshift galaxies has been quite controversial in the literature (see for example Mobasher et al. 2005 and Chary et al. 2007) and one should not necessarily take the individual parameters at face value. With the current 9 data points I am not able to explore more complicated models. In Table 5.21 summarize all the physical parameters derived for GN20. 5.5 Companion source: GN20.2 As previously mentioned, GN20 has a companion submm source nearby. GN20.2 has a secure radio, MIPS and IRAC counterpart (see Chapter 3). While I do not have a spectroscopic redshift for GN20.2, the photometric redshift is 3.9 (Pope et al. 2005), placing it at the same distance as GN20. Fig. 5.8 shows the i-band image of GN20.2. The green square indicates the position of the 5.5. COMPANION SOURCE: GN20.2 154 Figure 5.8: ?'-band image of the counterpart to GN20.2. The small square indicates the position of the radio, IRAC and MIPS counterpart. The closest optical emission is the two knots 0.8 arcseconds to the north. The large circle has a diameter of 3 arcseconds which is the aperture used for the photometric redshift calculation. 5.6. DISCOVERY OF A HIGH REDSHIFT CE USTER 155 radio, IRAC and MIPS counterpart, which is 0.8 arcseconds from the two knots to the north in the optical image. This is similar to the 0.5 arcsecond separation seen between the optical and radio emission from GN20's counterpart. This suggests that the GN20.2 system is disturbed with the optical and radio/IR emission coming from different parts of the galaxy. The yellow circle shows the aperture used to calculate the photometric redshift. Given that both of these SMGs are brighter than the majority of SCUBA galaxies, there is a high probability that they are related in some way. At this redshift, the 24 arcsecond separation between their counterparts corresponds to a physical separation of 170 kpc. This separation is quite large for these to be part of the same system; even in a major merger the largest separations between bright components is ~ 50 kpc (Springel, Di Matteo & Hernquist 2005). They could be two SMGs in the same cluster or group or they could be two images of the same lensed galaxy (see Section 5.6). If GN20 and GN20.2 are two images of the same galaxy then I expect them to have similar fluxes at all wavelengths, since the lensing amplification is not a function of wavelength. However, in practice lensing of extended systems is more complicated and the geometry of the galaxy being lensed could lead to differential lensing amplification (e.g. Franx et al. 1997; Blain et al. 1999). I have seen that the counterparts of GN20 and GN20.2 show offsets between the radio/IR and optical emission, and so they might experience different lensing effects. The submm flux of GN20.2 is half of the submm flux of GN20. While on average the ratio between the IRAC and MIPS 24 /im fluxes of GN20 and GN20.2 is around 2.5, the radio flux of GN20.2 is 2.5 times larger than the radio flux of GN20. Therefore if these are two lensed images of the same object, differential lensing would have to be occuring, presumably due to the complex geometry of the system lying across the critical lensing line. 5.6 Discovery of a high redshift cluster Most rich galaxy clusters which have been targeted show strong lensing of background SMGs. This effect is more important than for other galaxy populations, because of the steep number counts of bright SMGs, and was used to great effect with the SCUBA instrument (e.g. Cowie et al. 2002; Smail etal. 2002; Borys et al. 2004a). But how important is lensing for the SMGs 5.6. DISCOVERY OF A HIGH REDSHIFT CLUSTER 156 Figure 5.9: Biz HST image of central field surrounding GN20. Ovals indicates the counter-parts of GN20 and GN20.2 (the image is centred on GN20). The squares indicate the confirmed cluster galaxies within this area. The triangle indicates the massive X-ray source which 1 think is the centre of the cluster. This image is 50 by 50 arcseconds. 5.6. DISCOVERY OF A HIGH REDSHIFT CLUSTER 157 found in untargeted wide area surveys? Because of the unusually bright submm flux- of GN20 and the steep submm number counts (Coppin et al. 2006), lensing is likely to play a role in this system. GN20 looks extended in the optical and radio images and also has a companion submm source 24 arcseconds away (170 kpc if they are at the same redshift). This prompted me to search for a possible lens for the SMG(s). I caution that much of the following analysis is speculative, since I have no way to prove that the SMGs are lensed with current data. That being said, all observations that are currently available are consistent with the lensing scenario. I searched the deep HST images around GN20 and GN20.2 using all available redshift information and found 2 galaxies between the two SMGs with exactly the same redshift of 1.3644 from the TKRS survey (Wirth et al. 2004). I also found many more galaxies in the field with photometric redshifts consistent with this redshift. One of these galaxies is a very bright X-ray source just 18 arcseconds from both GN20 and GN20.2. This galaxy is very red and consistent with being near the top of the red sequence (e.g. Gladders & Yee 2000). This type of galaxy is often found near the centre of galaxy clusters. During the Keck run in February 2007, several of these candidate cluster members were targeted and 7 of them were confirmed to be at the same redshift. Redshifts for the new cluster members were identified by the Oil doublet at a rest wavelength of 3727 A. Combined with other galaxies in the TKRS survey, I identified 8 galaxies within a radius of 1 Mpc of the original two TKRS galaxies consistent with a spectroscopic redshift of 1.364 ± 0.003. There is also a larger structure which extends out to a diameter of 6 Mpc containing 18 galaxies with consistent spectroscopic redshifts. Table 5.3 lists the positions, redshifts and fluxes of the cluster members. Unfortunately, I was not able to obtain a redshift for the bright X-ray source, but a full fit to its photometry confirms a photometric redshift of ~ 1.4. This overdensity of galaxies at z = 1.364 represents one of the highest redshift galaxy clus-ters discovered to date. Currently there have only been a handful of galaxy clusters discovered above a redshift of 1 (e.g. Stanford et al. 2005; McCarthy et al. 2007 and references therein). The velocity dispersion of the central 8 galaxies is ~ 350 km s_ 1. Due to a bright X-ray point source within the field (which I believe may be at the centre of the cluster), I am unable to 5.6. DISCOVERY OF A HIGH REDSHIFT CLUSTER 158 constrain the diffuse X-ray emission that may be coming from this cluster. There is also a faint detection of weak lensing around this cluster (L. Van Waerbeke, private communication). Since this is the first galaxy cluster discovered because of a bright submm source, I refer to this cluster as SMM1. Making a few assumptions about the system, I can put constraints on the central mass of the cluster. The Einstein radius is a characteristic angle for gravitational lensing. Following Peacock (1999), the Einstein radius assuming a point mass, 6E is defined as where D$ and L \ are the angular diameter distances (see Appendix A.l) to the source and lens, respectively, and DLS is the angular diameter distance between the source and the lens. M is the total mass within the Einstein radius that is lensing the background galaxy. Assuming that GN20 and GN20.2 are both images of the same lensed galaxy and that the bright X-ray source is the centre of the cluster at z = 1.364, then I can estimate #E ~ 18 arc-seconds (150 kpc at z = 1.364). Substituting into Equation 5.3,1 calculate M ~ 1O 1 4 M 0 . This is the total mass within this radius including both visible and dark matter, and is comparable to other known clusters; for example Abell 2218 has a total mass within the Einstein radius of 6 x 10 1 3 M o (Makino & Asano 1999). The above has assumed a point mass, although in reality the geometry and mass distribution in the cluster is likely to be much more complicated, which will require a larger mass and introducing additional errors into this estimate of the mass. The mass within the Einstein radius will be split between several galaxies, but a large fraction of it should come from the massive galaxy at the center. For cluster SMM1,1 suspect that this is the bright X-ray galaxy. Fitting the optical/IRAC photometry of the bright X-ray source I calculate a total stellar mass of 3 x 1 0 n M o . The fit shows an old population of stars dominating this galaxy, consistent with it being a massive elliptical galaxy with little on-going star formation. Once I consider the dark matter from this galaxy (e.g. M t o t /M* ~ 100, Balogh (5.2) (5.3) 5.6. DISCOVERY OF A HIGH REDSHIFT CLUSTER 159 Table 5.3: Positions and fluxes of galaxies in cluster SMM1. The error on the spectroscopic redshifts are typically dz~ 0.0001. The first 8 galaxies are within 1 Mpc while the last 10 lie between 3-6 Mpc from the centre of the cluster. Within each section, the sources are ordered according to RA. At z = 1.364, the rest-frame i\~-band shifts to 5 . 2 psm therefore these IRAC fluxes should be roughly proportional to the stellar masses of these galaxies but of course theconversion will vary with galaxy type. ID RA DEC Redshift *775 •S'4.5 S5.8 Ours TKRS AB mag (uJy) (uJy) 16676 12:36:56.65 62:20:49.2 1.36488 23.7 8.38 6.62 19659 12:37:06.24 62:21:11.7 1.36316 23.0 18.72 13.99 19762 12:37:06.62 62:21:08.3 1.36404 22.9 15.55 12.28 10817 12:37:01.31 62:20:51.0 1.359 23.5 19.18 12.42 20730 12:37:10.13 62:22:06.6 1.36442 23.3 30.98 22.32 10297 12:37:10.83 62:22:11.1 1.365 1.36444 24.6 2.90 1.41 10386 12:37:15.45 62:22:02.8 1.363 1.36261 23.8 4.26 2.99 10687 12:37:24.42 62:21:32.2 1.360 27.3 1.56 0.88 3660 12:35:53.11 62:10:37.3 1.37091 21.5 88.39 89.57 1543 12:36:13.34 62:17:15.4 1.361 -99 34.30 29.88 1608 12:36:19.80 62:16:01.3 1.35672 23.3 14.18 10.67 14103 12:36:24.46 62:17:45.2 1.365 25.4 31.35 26.81 15822 12:36:42.59 62:16:29.7 1.363 24.5 4.57 3.64 6681 12:36:52.74 62:13:54.7 1.35907 22.8 15.00 11.13 11312 12:37:32.23 62:13:04.8 1.35846 23.9 10.27 7.15 11594 12:37:39.82 62:13:52.5 1.35934 23.6 8.41 4.56 8133 12:37:45.84 62:18:16.5 1.36435 24.4 5.80 2.65 9696 12:37:53.19 62:17:16.6 1.36516 23.9 6.04 6.08 5.6. DISCOVERY OF A HIGH REDSHIFT CLUSTER 160 et al. 2007), I can account for a significant fraction of the total mass within the Einstein radius. Without more constraints on the mass distribution of the cluster, I cannot calculate the amount of amplification that GN20 is undergoing due to the gravitational lensing. I can, how-ever, roughly estimate the lensing magnification for GN20 by comparing to results for other strong-lensing clusters. The field surrounding the cluster Abell 2218 (which has a similar mass within the Einstein radius as SMM1) has a triply-imaged SMG system with an amplification factor of 45 (Kneib et al. 2004). Assuming GN20 is lensed by the same amount, then all mass and luminosity values in Table 5.2 need to be scaled down by this factor. With this magnifi-cation, GN20 would still be classified as a massive luminous infrared galaxy, the presence of which at z = 4 has implications for galaxy formation models. 5.6.1 Future potential Our analysis of the complex system is on-going and there are still many unknowns concerning the lensing model. That being said, one thing is clear, the presence of this bright SMG enabled the discovery of a high redshift galaxy cluster. This cluster would not have been found in X-ray surveys or shallow Red Sequence surveys, since its members are too faint at these wavelengths. In future SCUBA-2 Legacy surveys, 20 square degrees will be surveyed deep enough to reveal tens to hundreds of systems like GN20. This study has shown that these bright SMGs may act like pathfinders for identifying high redshift galaxy clusters. Identification and characterization of clusters at z > 1 has important implications for the formation of massive early type galaxies and for probing the underlying density of dark matter and dark energy. • 161 C H A P T E R 6 RESOLVING THE COSMIC INFRARED BACKGROUND In Fig. 1.1, I showed the background radiation in the Universe as a function of wavelength. This plot shows that the background at IR wavelengths, the cosmic infrared background (CIB), is roughly equal to that at optical wavelengths. A key step in understanding the formation and evolution of galaxies is determining the nature and redshift distribution of the galaxies which make up the CIB. While deep imaging with ISO and Spitzer have resolved the majority of the CIB at 15, 24, 70 and 160 ^m (Elbaz et al. 2002; Papovich et al. 2004; Dole et al. 2006), the CIB at longer wavelengths remains largely unresolved. Furthermore, the slope of the long-wavelength part of the CIB in Fig. 1.1, is much less steep than the SEDs of galaxies, indicating that the millimetre background is' not due to millimetre emission from the same galaxies that account for the peak at the CIB. This implies that the far-IR production rate must increase dramatically between z = 0-1 and then remain fairly constant out tb z ~ 4 (Gispert, Lagache & Puget 2000). At 850 pm the observed value of the CIB is between 31 Jy deg - 2 (Puget et al. 1996) and 44 Jy deg - 2 (Fixsen et al. 1998). The difference between these estimates comes from the two different methods for removing the foreground emission. The background at 850 ^m is 10 ± 4 Jy deg - 2 resolved with SCUBA galaxies detected down to 2 mJy (e.g. Coppin et al. 2006). In order to probe fainter flux densities, Smail et al. (2002) have taken advantage Of the strong gravitational lensing observed in cluster fields to probe SCUBA galaxies down to ~ 1 mJy which, together with detections in blank fields, accounts for 26 ± 9 Jy deg - 2 of the CIB at 850 ^m (see also Cowie et al. 2002 and Knudsen et al. 2006). The upper limit at 850 ^m (350 GHz) in Fig. 1.1 shows the contribution to the CIB from galaxies detected with SCUBA down to 1 mJy. What is the nature and redshift distribution of the galaxies which account for the remaining (5-18) ± 9 Jy deg - 2 of the CIB at submm wavelengths? Observations with SCUBA are limited by confusion noise due to the large beam size at 6.1. STACKING METHOD 162 submm wavelengths. There is a flux density level below which all sources become blended together and this is referred to as the confusion limit. Confusion noise becomes important when the number density of sources is above 0.03 beam -1 (e.g. Scheuer 1957, 1974; Condon 1974). For 850 /im SCUBA observations with a beam size of 15 arcseconds, the confusion limit corresponds to an RMS contribution around ~ 0.5 mJy RMS (Blain et al. 1998; Barger et al. 1999; Eales et al. 2000). While it is difficult to extract sources with flux densities lower than ~ 2 mJy (assuming a Aa detection) directly from SCUBA maps, it is possible to probe fainter flux densities using a statistical stacking analysis. With this technique, the average flux density of a population of galaxies is measured in the SCUBA maps using their known positions. By performing a variance-weighted average, it is possible to statistically measure flux densities below the confusion limit. In this chapter I present the overlap between high redshift Spitzer and SCUBA populations using a statistical stacking analysis to measure the contribution of Spitzer galaxy populations to the 850 nm submm background. 6.1 Stacking method As mentioned in Chapter 2, the 'super-map' gives the best estimate for the flux density of a point source centred on each pixel. Therefore in stacking the flux density of a list of objects I simply have to pull off the value of the pixel at each location. The average submm flux density, (5g5o), for a population of galaxies, and its error, o"(s850), are defined as (585o) = 2 ^ 0 - , (6.1) -2 ' JI ~ i (^Ss.so) = , (6.2) -2 i " i where Sj and at are the flux density and noise taken at the position of the ith Spitzer source in the beam-convolved 'super-map'. Recall that because the observations were obtained by chopping (see Chapter 2), the average in the super-map is zero and stacking on random posi-6.1. STACKING METHOD 163 tions produces a stacked submm flux density of zero. This has been verified through Monte Carlo simulations, and therefore I can be confident that any signal that I find from stacking is genuine. Of course the super-map also contains ~ 40 resolved sources detected at > 3.5<r. These sources have already been identified with unique Spitzer galaxies (see Chapter 3), and so I can separately calculate the contribution of these galaxies to the CIB. Since galaxies detected with Spitzer are known to be clustered (e.g. Fang et al. 2004; Oliver et al. 2004), my results will be biased high if I do not remove the submm sources which are > 3.5a from the super-map before stacking the flux density of the Spitzer galaxies. In Pope et al. (2005), I describe a method for 'cleaning' the submm detections from the SCUBA map. This analysis removes both the positive signal from the detections and the negative sidelobes from the chopping pattern from the SCUBA map. In the stacking analysis in this chapter, I use the cleaned 850 pm SCUBA map and calculate the contribution of the resolved SCUBA sources to the background separately. Once I have determined the average submm flux density from a list of objects, I can deter-mine their contribution to the diffuse submm background. The total intensity at 850 pm, isso. from the stacked galaxies is defined as = ^ , (o.3) where (585o) is the average submm flux density, N is the number of sources stacked on the submm map and A is the area of the submm map. For the GOODS-N super-map, if I restrict to regions of the map with noise < 4 mJy, the area which contains IRAC and MIPS coverage is 0.025 deg 2 . One needs to be careful here though, since ls 5 0 is really an integral over the source counts (= J Sj^dS). This will not necessarily be equal to (Sgso) x A/TOT, depending on the slope and curvature of the source counts. Simply stacking the whole population and using a single A/TOT can result in a biased background estimator. To avoid this I always split the target population into several flux density bins, effectively approximating the integral as a sum. To choose how many bins are required, I can double the number and check that this has negligible effect on the final results; this is the procedure I followed for all of my background estimates described 6.2. STACKING IRAC GALAXIES 164 0.35 F (nJy) (nJy) Figure 6.1: Average submm flux density of Spitzer 3.6 pm galaxies. In the left panel I show the average submm flux density in each 3.6 ^ m flux density bin, while the right panel shows the cumulative average submm flux density. The horizontal line indicates the total average submm flux density for all these 3.6 pm sources. The number of sources per bin is roughly 300. below. 6.2 Stacking IRAC galaxies The analysis performed in the previous section can be performed with the whole catalogue or with various sub-populations of galaxies. Since I have very deep Spitzer observations in GOODS-N (around 20,000 galaxies detected in the IRAC 3.6 ^ m image), I can stack the submm flux density of these galaxies as a function of their mid-IR flux density. For each IRAC flux density bin, I can calculate J 8 5 0 , where (5850) and M are determined for each flux density bin sample. To obtain the total contribution to the background, I just take the sum of all the / 8 5 0 values for each bin. This allows me to determine if the average submm flux density is 6.2. STACKING IRAC GALAXIES 165 dominated by galaxies which are brighter or fainter in the mid-IR. This method of stacking as a function of flux density was used by Dole et al. (2006) in determining the contribution from mid-IR detected galaxies to the far-IR background. For the IRAC wavelengths, I used the deep GOODS catalogue (> 5a) and imposed a flux density cut equal to the 50% completeness limit (see Table 6.1). I also excluded any galaxies which were flagged in the source extraction as having problems such as being close to the edge or having blending issues. The total numbers of sources which I stacked at each wavelength are shown in Table 6.1. The difference between each sample is simply the wavelength that they are selected at, and therefore they are not independent. Fig. 6.1 shows average 850 /xm flux density from galaxies detected at 3.6 /xm as a function of their 3.6 /xm flux density. In the left panel, I show the average flux density in each bin and the right panel shows the cumulative average flux density. Each bin contains an equal number of galaxies (~ 300). The horizontal curve shows the average submm flux density for all 3.6 /xm galaxies that were stacked, (S850) = 0.10 ± 0.02 mJy. The entire 3.6 /xm galaxy population is thus detected at 5a. At fainter 3.6 /xm flux density levels, I see that the stacked submm flux density is sometimes negative, with individual measurements consistent with zero flux density. I cut the 3.6/xm catalogue off at the 50% completeness limit of 0.4/xJy, and therefore some fraction of the galaxies which I am stacking are not genuine 3.6 /xm detections, while I am also missing some sources which genuinely have £3.6 > 0.4/Jy. This incompleteness in the catalogue I stack will act to dilute the stacked signal, since I am averaging in random galaxies with statistically zero flux density. Therefore, the contribution of these galaxies to the CIB that I derive will be a lower limit. Since IRAC is such a sensitive instrument, it detects a wide variety of galaxies. Of course, not all of the 5,000 3.6 /xm galaxies will emit with this average submm flux density. In order to weed out the galaxies which are dominating the stacked signal, the sample would need to be sub-divided further using other properties such as redshift or radio flux density. The cumulative contribution to the 850 /xm CIB of IRAC galaxies as a function of IRAC flux density is shown in Fig. 6.2. The value of the total CIB in the submm lies somewhere between the two horizontal lines. Again in this plot, I see the effect of incompleteness of the 6.2. STACKING IRAC GALAXIES 166 Figure 6.2: Contribution of IRAC-detected galaxies to the CIB as a function of IRAC flux density. The two horizontal dashed lines indicate the two estimates of the total CIB at 850 /xm: 31 Jy deg - 2 (Puget et al. 1996); and 44 Jy deg - 2 (Fixsen et al. 1998). 6.3. STACKING MIPS 24 y.M GALAXIES 167 IRAC catalogues, as the cumulative power begins to drop at the faintest IRAC flux density levels. The false positive IRAC sources included at these flux density levels will only act to decrease the total power, and so I can take the peak power seen on these plots as a lower limit on the contribution of these galaxies to the CIB. These values are listed in Table 6.1 for each wavelength. The total value of I85Q for galaxies selected at IRAC wavelengths ranges from 21-26 Jy deg - 2. The two longer wavelength IRAC channels are somewhat less sensitive, as seen by their higher 50% completeness limit. Recall that this analysis excludes the contribution from galaxies which are individually detected at > 3.5<J in the submm image. If I sum the submm flux density for all the SCUBA sources and divide by the area I get a value of 10.9 ± 0.4 Jydeg - 2. This is consistent with what is found in Coppin et al. (2006) for the SHADES survey. Combined with the intensity from stacking the 4.5 /mi-selected galaxies, I get a total intensity of 37 ± 3 Jy deg - 2. This lies right between the two estimate for the total background at 850 //m. This is the first time that the 850 jim CIB has been resolved into individual sources. It implies that the counterparts to all galaxies detected in future submm surveys which go much deeper than current limits will be detected in IRAC imaging to the GOODS depth. I note that this estimate may still be biased high due to the fact that the IRAC-selected galaxies are clustered. I have tried to account for this by removing the submm detections (> 3.5a) from the map before stacking but there will still be additional contributions from sources which fall within the PSF of other fainter 0-3.5a sources. This bias can be quantified by performing simulations of the clustered galaxy population which I intend to perform in the future. 6.3 Stacking MIPS 24 galaxies In an attempt to refine the sample of galaxies which are responsible for the CIB, I have repeated this stacking analysis on MIPS 24 selected galaxies. This selection is less sensitive to 'normal' galaxies and preferentially, picks out star forming or active galaxies. The sensitivity of the instrument limits its galaxy samples to z < 4 (e.g. Chary et al. 2004). 6.3. STACKING MIPS 24 pM GALAXIES 168 As mentioned in Chapter 2, the 24 pm catalogue was created using the position of the IRAC galaxies as priors. This allows me to push to lower SNR at 24 pm, since the positions of the galaxies are known. For the stacking analysis, I use the full 3a 24 pm catalogue which goes down to a la depth of ~ 5 /Jy. At this detection limit ( > 15 /Jy) the catalogue is roughly 50% complete according to simulation of the 24 pm images (Chary et al. in preparation). There are just over 1,000 24 pm selected galaxies which overlap with the SCUBA coverage. In Fig. 6.3,1 plot the distribution of submm SNR for all 24 pm galaxies as the thick solid curve. The lighter histogram shows the distribution of all pixels in the cleaned super-map. As expected it is consistent with a Gaussian centred on zero. The dashed red curve shows the best-fit Gaussian to the distribution for the 24 pm galaxies, and it is clearly offset from the total distribution of pixels in the map - this effect is also seen in the top panel which shows the difference between the two distributions. This overall shift of the distribution, as opposed to a tail at positive SNR, shows that the population of 24 pm galaxies as a whole is detected in the map. , The contribution of the 24 pm galaxies to the submm CIB as a function of 24 pm flux is shown in Fig. 6.4. The solid circles are the measured values, whereas the open circles have been corrected for completeness (Chary et al. in preparatipn). Even though I am stacking a catalogue which pushes to low SNR, there does not seem to be much of a decline in the contribution to the CIB from the faintest 24 pm flux density bins for the uncorrected points. This indicates that by using the IRAC positions as priors ensures that a large fraction of the sources are genuine 24 pm galaxies. However, at this SNR there will be a substantial number of galaxies which are actually brighter than the flux density limit but that miss the detection threshold because of noise excursions. To account for the contribution to the background from these sources I have applied a completeness correction to each point based on simulations of the GOODS 24 pm images (Chary et al. in preparation). These are plotted as the open circles and show a total contribution of 34 ± 4 Jy deg - 2 when I include the faintest 24 pm galaxies. If I do not correct for completeness then the contribution from 24 pm selected galaxies is 1 7 ± 2 Jy deg"2, which is only about 1/2 of the total CIB. When I add in the SCUBA detected sources which have 24 pm detections, I resolve 45 ± 4 Jy deg - 2, which is essentially all of the CIB at 850 pm. The value 6.3. STACKING MIPS 24 pM GALAXIES 169 40 r "i 1 1 1 1 1 1 1 1 1 1 1 1 r-Figure 6.3: Distribution of submm SNR for the positions of 24 /zm-selected galaxies. The thick curve is the distribution of SNR from the submm super-map at the positions of the 24 pm galaxies, while the thin histogram is the distribution of SNR for all pixels in the cleaned super-map. The panel at the top shows the difference between the two distributions. The whole map is consistent with a mean of zero, as expected, while the distribution for the 24 pm galaxies is offset towards positive SNR, indicating that the population as a whole is detected in the super-map. 6.3. STACKING MIPS 24 fjM GALAXIES 170 CN 00 CN A O 100 ,24um (MJy) Figure 6.4: Contribution of 24 /xm galaxies to the CIB as a function of 24um flux density. The filled circles are the measured values whereas the open circles have had a completeness correction applied to them. The two horizontal dashed lines indicate the two estimates of the total CIB at 850 /xm: 31 Jy deg - 2 (Puget et al. 1996); and 44 Jy deg - 2 (Fixsen et al. 1998). The number of sources per bin is roughly 100. 6.3. STACKING MIPS 24 p,M GALAXIES 111 Table 6.1: Contributions from Spitzer-selected galaxiesto the CIB at 850/mi. C 5 0% is the 50% completeness limit and A/TOT is the total number of galaxies used in the stacking analysis. (5850) is the average submm flux density from all A/TOT galaxies and J 8 5 0 is their total intensity at 850 /mi. The value of / 8 5 0 listed for 24 /mi has not been corrected for completeness. The difference between these samples is simply the wavelength that they are selected at, and therefore they are not independent. Wavelength C50% A/TOT 8^50 (/tm) 0"Jy) (mJy) (Jydeg"2) 3.6 0.4 5271 0.10 ± 0 . 0 2 23 ± 3 4.5 0.4 5124 0.11 ± 0 . 0 2 26 ± 3 5.8 0.9 2696 0.19 ± 0 . 0 2 21 ± 3 8.0 0.9 2488 0.21 ± 0.03 23 ± 2 24 16 1152 0.37 ± 0.04 17 ± 2 is slightly high, given the two estimates for the background, however it is consistent within the errors quoted. Furthermore, there are errors associated with the completeness correction which I have not applied here. Given the sensitivity of MIPS 24 /mi observations (see Chary 2006), this suggests that all of the 850 /mi background comes from star-forming galaxies with z < 4. Note that I am concerned that the trend for the open circles in Fig. 6.4 does not appear to be flattening out at the lowest flux levels. This suggests that extrapolating to even lower 24 /im flux densities will resolve more than the total amount of the submm CIB. Several sources of uncertainty exist in this analysis which need to be explored further including simulations to predict the bias from the clustering of the Spitze r-selected population and estimating the errors in the completeness simulations. I have demonstrated that estimating the contribution of galaxy populations to the diffuse CIB is difficult and requires proper treatment of all sources of error and bias. 6.4. COMPARISON TO OTHER STUDIES 6.4 Comparison to other studies 172 Several other studies have attempted to resolve the CIB with samples of Spitzer-selected galax-ies, albeit with much shallower Spitzer observations. Serjeant et al. (2004) stacked galaxies detected in IRAC and MIPS 24 pm from the Spitzer Early Release Observations of the Lockman Hole field. The depths of their Spitzer catalogues for stacking are roughly an order of magnitude shallower than the GOODS Spitzer catalogues. While they are successful in statistically detecting the Spitzer-selected galaxies in their SCUBA maps, this accounts for only a small contribution to the background at 850 nm. Serjeant et al. (2004) do not weight by the variance in calculating the stacked average. Their SCUBA maps show variable noise levels and so this may bias their stacked results. However, they have removed the submm detections prior to stacking. Using observations in the GOODS-N field, Wang, Cowie & Barger (2006) studied the galaxies which dominate the CIB. Their submm map contains less data than the super-map and they followed different analysis procudures in their data reduction (see Borys et al. 2004b for a more detailed comparison of the maps). Their main results come from stacking on the submm map which still contains all the submm detections, and therefore, as noted previously in this Chapter, their results will be biased high, since Spitzer-selected galaxies are strongly clustered. In fact, they also list the values for their clean map in the paper and indeed those values are a factor of about 2 lower. They conclude that roughly 2/3 (really 1/3 using the cleaned map) of the submm CIB comes from near-IR selected galaxies. Furthermore, they stack as a function of redshift and find that most of this background comes from galaxies at z — 0-1.5. While these results are certainly useful, since they have stacked on their uncleaned map, the background contributions are all overestimated by a factor of about 2, and so they are talking about only a small fraction of the background in this paper. Another SCUBA survey was the Canada-UK Deep Submillimetre Survey (CUDSS, Eales et al. 1999) which surveyed two fields down to ~ lmJy RMS. Dye et al. (2006) performed stacking of 8 and 24 /mi Spitzer-selected galaxies on these submm maps. Again their Spitzer observations are substantially shallower (by a factor of about 5) than those in GOODS-N. They 6.4. COMPARISON TO OTHER STUDIES 173 take care to remove the submm detections from their map and use variance-weighting in their stacking analysis. They positively detect both samples of Spitzer galaxies in the submm map, but this resolves only about 5-16% of the submm CIB, or about 30% when they include the contribution from submm detections. Contrary to Wang et al. (2006), they find that the 850 pm flux density is emitted almost exclusively from Spitzer galaxies at z > 1.3. My study is the first one which may have resolved all of the submm background. This is mainly because the GOODS-N Spitzer data are substantially deeper than those used in other studies. The differences in the data or stacking methods I list above can account for all of the differences seen between my results and those from other studies. Future work will involve further sub-dividing the IRAC and MIPS catalogues to hone in on the properties of the sample of galaxies which dominate this background and understanding all sources of bias and error in this difficult analysis. 174 CHAPTER 7 HOW BRIGHT ARE SUBMILLIMETRE GALAXIES? The high redshift submm-bright galaxy population is known to be faint at most other wave-lengths - but how faint exactly? It has been impossible to answer this question until now, because data at other wavelengths have not been deep enough to identify counterparts to all SMGs in a sample. In this chapter, I present the cumulative flux distribution at X-ray, optical, infrared and radio wavelengths for an essentially complete sample of 850 /im SCUBA galax-ies in the GOODS-N field. This sample contains sources with submm fluxes from 2-20 mJy, which covers the full range for all galaxies detected in SCUBA blank-field surveys. Using deep VLA and Spitzer (IRAC and MIPS) imaging I have counterparts for 94% (33 out of 35) of the SMGs; I also present results for a subset of 21 (60%) which were identified using very conservative criteria. I demonstrate how new data and improved reduction of the VLA imaging in the field has increased the number of radio counterparts to submm sources. With this nearly completely identified sample of 850 /tm sources, I can determine the depths at which one can expect to detect the majority of SMGs in future mm/submm surveys, such as with SCUBA-2. I discuss which multiwavelength data are most useful for learning about the nature of SMGs. 7.1 Multiwavelength flux distributions In order to clearly display the fluxes of submm counterparts, I plot cumulative flux distributions and normalize them to the total number of sources in the sample. These distributions can be used to easily read off the fraction of submm counterparts expected above each flux density level. Throughout this chapter, I plot the 21 secure counterparts and all 33 (secure or tentative) counterparts as separate curves (solid and dashed lines, respectively). Both the Poisson statis-tics and Monte Carlo simulations tell me that on average one of the secure counterparts will be a random association. While individually the tentative counterparts have a higher chance of 7.1. MULTIWAVELENGTH EE UX DISTRLB UTIONS 175 1.0 20 10 1 S850um (mJy) Figure 7.1: Cumulative distribution of deboosted 850 /jm flux density for submm sources in GOODS-N. The solid curve is for the submm sources with secure (P < 0.05) counterparts while the dashed curve contains all 35 submm sources. All cumula-tive distribution plots in this chapter have been normalized to the total number of sources in the sample so as to clearly display the identified fraction at each wave-length. 7.1. MULTIWAVELENGTHFLUX DISTRIBUTIONS 176 being random associations, statistically I expect that most of them are genuine. If I consider all 33 counterparts then this sample is nearly complete. Other possible counterparts for the submm sources have higher probabilities of random association and are typically fainter, therefore the multiwavelength fluxes for the tentative counterparts are, at worst, upper limits. I am therefore confident that my catalogue provides an accurate (or at least conservative) representation of the submm population with which to explore the multiwavelength behavior of 850 pm SCUBA galaxies. I now describe each waveband in turn. 7.1.1 Submillimetre In Fig. 7.1,1 plot the cumulative distribution of deboosted 850 jum flux for the submm sources with secure (P < 0.05) counterparts and for all 35 submm sources in the sample. My sample contains a significant fraction of faint (Ssso^m < 5.0 mJy) sources, whereas other SCUBA samples are mostly limited to sources at Sgso^m > 5.0 mJy (Scott et al. 2002; Chapman et al. 2005). However, since the dynamic range in flux of blank sky SCUBA sources is small I do not expect very different multiwavelength fluxes for the brighter submm sources. I have checked this for a sub-sample of S85oMm > 5.0 mJy sources and found qualitatively the same distributions of multiwavelength fluxes for the counterparts, with the exception of i 7 7 5 and X-ray. The brighter submm sources appear to have fewer optically bright counterparts and have a lower X-ray identification rate. These results are qualitatively consistent with those reported by Pope et al. (2005, see Figure 3 in that paper), who found that fainter submm sources tend to have brighter optical counterparts and lower redshifts. 7.1.2 Radio Fig. 7.2 shows the cumulative distribution of 1.4 GHz radio flux for the submm counterparts in GOODS-N. I compare the identification rate using the Richards (2000) 5<7 catalogue (thin lines) and the new Morrison et al. (in preparation) 3a catalogue (thick lines). Note that in both curves I am plotting the fluxes as measured from the new Morrison et al. (in preparation) radio map. 10% of the improvement is due to the lower signal-to-noise ratio (SNR) threshold now 7.1. MULTIWAVELENGTH FLUX DISTRIBUTIONS 177 Figure 7.2: Cumulative distribution of 1.4 GHz radio flux density for submm counterparts in GOODS-N. The solid curves are for the secure ( P < 0 .05) counterparts while the dashed curves contain all counterparts. The thick lines show the distributions using the new Morrison et al. (in preparation) reduction of the VLA data, while the thin lines contain identifications made using only the initial Richards ( 2 0 0 0 ) catalogue of the same data. 7.1. MULTIWAVELENGTHFLUX DISTRIBUTIONS 178 adopted. My confidence in these new low SNR radio detections is considerably bolstered by the fact that they are all coincident with IRAC and MIPS detections (see Chapter 3 for more details on the new radio counterparts). Going deeper in the radio observations has revealed a significant number of new counterparts to the SMGs. This is expected, because while some submm sources have bright radio counterparts, a significant number (~ 30% in Chapman et al. 2005) are very near the detection limit in radio observations. The new radio reduction tech-niques have increased the radio-detected fraction of all submm sources in GOODS-N to 74%, which is consistent with the radio identification rate in the 8 mJy survey, where the radio RMS varies between ~ 5-10 /Jy (Ivison et al. 2002). Considering only the secure radio counterparts, the new radio identification fraction is close to 60%. This emphasizes that, when quoting radio identification rates for submm sources, it is important to consider the depth, area, configura-tion, etc. of the radio observations, as well as considering the probability that the associations are random. Fig. 7.2 shows that one needs to be able to detect radio sources at the < 20 ply level in order to have a hope of identifying most of the SMGs. These depths will be extremely challenging for ~ 10 deg2 SCUBA-2 surveys. 7.1.3 Mid-infrared Both SCUBA and Spitzer are able to select distant star-forming galaxies which contribute sig-nificantly to the cosmic infrared background. In blank field surveys, SCUBA is primarily sen-sitive to the thermal dust emission from ultra-luminous galaxies at high redshift. While Spitzer also detects these galaxies, its sensitivity is such that it detects galaxies over a wider range of luminosity and galaxy type. Fig. 7.3 shows the MIPS 24 /tm flux density distribution for the 850 /im selected sample. The 3a sensitivity limits for other Spitzer surveys are indicated by the vertical dashed lines. The Guaranteed Time Observation (GTO) shallow survey of the Bootes field (E. Le Floc'h priv. communication) is deep enough to detect 25% of the galaxies detected with SCUBA at 850 /mi, whereas the deep GTO surveys (Papovich et al. 2004, 2006) should be able to detect 50-75%. With the MIPS depth of the First Look Survey (FLS1), or the Spitzer ' h t t p : / / s s c . s p i t z e r . c a l t e c h . e d u / f l s / e x t r a g a l / s p i t z e r . h t m l 7.1. MULTIWAVELENGTH FLUX DISTRIB UTIONS 179 £ 0.6 r (N A 0.4 100 '24nm (nJy) 10 Figure 7.3: Cumulative distribution of 24 pm flux density for submm counterparts in GOODS-N. The solid curve is for the secure (P < 0.05) counterparts while the dashed curve contains all counterparts. The vertical dash-dot lines indicate the 3<r limits of several other Spitzer surveys: GTO shallow survey of the Bootes field (E. Le Floc'h priv. communication), FLS, SWIRE (D. Farrah priv. communication), the Early Release Observation (ERO) survey of the Lockman Hole East region (Egami et al. 2004), and the GTO deep survey of the Chandra Deep Field South (CDF-S, Papovich et al. 2004, 2006). 7.1. MULTIWAVELENGTHFLUX DISTRIBUTIONS 180 Wide-area InfraRed Extragalactic survey (SWIRE, Lonsdale et al. 2004), one would find coun-terparts for 30^ 10% of the SMGs. Still, a future SCUBA-2 survey of a 5 square degree SWIRE field will detect ~ 15,000 SMGs brighter than 2 mJy at 850 /mi. Given my analysis, this equates to ~ 5,000 850 /im SCUBA galaxies with MIPS 24 /mi counterparts at the SWIRE depths, but a further ~ 10,000 fainter MIPS counterparts which will be undetected at this depth. Going to shorter wavelengths, Fig 7.4 shows the distribution of IRAC flux densities for the submm sample. The top two panels show the 3.6 and 4.5 /mi IRAC channels. Submm surveys in the SWIRE fields will be able to find the majority of the submm counterparts in the SWIRE 3.6 and 4.5 /im images, although identifying the correct counterpart without deeper data at longer wavelengths will be challenging (in GOODS-N, the typical ratio of the number density of IRAC sources to MIPS sources is roughly 7). The flattening out of the dashed curves at the faintest IRAC fluxes seems to indicate that observations fainter than this will not gain you many more counterparts although this could also be complicated by the lower completeness level in the IRAC catalogues at the faintest fluxes. The bottom two panels of Fig. 7.4 show the flux distributions for the IRAC 5.8 and 8.0 /im channels. A striking feature in Fig. 7.4 is the different fraction of submm counterparts found to the depth of the SWIRE survey between the shorter and longer IRAC channels. The longer IRAC channels provide invaluable photometry for spectral energy distribution fitting to de-termine physical parameters for the submm systems, such as stellar mass, opacity, and active galactic nuclei/starburst contribution to the infrared luminosity, and also provide an impor-tant link between 24 /im and the shorter IRAC channels for counterpart identification. IRAC colours using all 4 channels also provide strong redshift constraints if the spectrum shows a significant stellar component (e.g. Sajina, Lacy, & Scott 2005, Chapter 3). However, as seen in Fig. 7.4, the majority of submm counterparts would not be detected at 5.8 and 8.0 /im to the SWIRE survey depths. Recently, Younger et al. (2007) have localized the submm emission from seven2 1.1 mm-selected galaxies with the SMA. Using the Spitzer images in the field, two sources are detected 2Note that one of the sources is confused by a bright foreground source in the Spitzer images and it is not possible to assess whether or not it is detected. 7.1. MULTIWAVELENGTH FLUX DISTRIBUTIONS 181 0.8 h 50 10 2 1 50 10 2 1 S (pJy) S (uJy) Figure 7.4: Cumulative distributions of IRAC flux for submm counterparts. The panels show the 4 IRAC wavelengths: 3.6, 4.5, 5.8 and 8.0 /xm. In each panel, the solid curve is for the secure (P < 0.05) counterparts while the dashed curve contains all counter-parts. For each wavelength, the 5cr sensitivity limit of the SWIRE survey is shown by the vertical dashed line. 7.1. MULTIWAVELENGTHFLUXDISTRIBUTIONS 182 at 24 /mi while six are detected with IRAC at 3.6 pm. This sample has quite a different submm flux limit compared to my submm sample; the average 890 jum flux of the sources in Younger et al. (2007) is ~ 12mJy whereas the average 850/im flux of my sample is twice as faint, ~ 6 mJy. Nevertheless, given the shallower depths of their Spitzer observations, the Younger et al. (2007) results are completely consistent with the predictions based on my sample of SMGs. 7.1.4 Near-infrared Compared with the imaging at other wavelengths, the near-infrared data in GOODS-N are currently not as deep as originally planned, due largely to bad weather. Fig. 7.5 shows the distribution of Ks magnitude for the sample and it is clear that one needs to get to Ks ~ 23.5-24 before one can expect to see the majority of the submm counterparts. Obtaining this depth over wide areas is now possible with the wide-field infrared camera (WIRCAM) on the Canada-France-Hawaii Telescope (CFHT) and the wide-field Camera (WFCAM) on the UK Infrared Telescope (UKIRT), for example. 7.1.5 Optical Fig. 7.6 shows what the submm counterparts look like in two optical wavebands. The solid and dashed curves show the distribution of z 7 7 5 and .B435 magnitudes, respectively. The dash-dot line indicates the 10<7 limit for point sources in the 2 square degree i&u Cosmic Evolution Survey (COSMOS3, Koekemoer & Scoville 2005). SMGs are at high redshift (< z > ~ 2 for my sample), and are also heavily dust-obscured. At z — 2, the ACS z 7 7 5 and 5435-bands correspond to rest-frame emission at 258 and 145 nm, respectively. Because submm systems are very dusty, the optical emission becomes increasingly extinguished at shorter optical and UV wavelengths. Combined with the effects of redshift, this can be seen in Fig. 7.6. To detect 50% of all submm counterparts at these optical wavelength, a depth of i775 ~ 25 AB mag would be sufficient, although the depth required at £435 is close to 28 AB mag. It is worth noting that even in the exquisite HST images of GOODS-N, 25% of the submm sources are 3http://www.astro.caltech.edu/cosmos/ 7.1. MULTIWAVELENGTH EL UX DISTRTB UTIONS 183 1.0 n I i I i I i I i i r "1 1 1 1 1 1 1 1 1 1 T 0.8 s £ 0.6 03 v, 0.4 0.2 0.0 _J I I I I I I u I , , _l I I I I I I I I 1_ 20.0 20.5 21.0 21.5 22.0 Ks magnitude (AB) 22.5 Figure 7.5: Cumulative distribution of iv~s magnitude for submm counterparts in GOODS-N. The solid curve is the secure (P < 0.05) counterparts while the dashed curve contains all counterparts, offset slightly in the y-direction for clarity. 7.1. MULTIWAVELENGTH FLUX DISTRIBUTIONS 184 i.o 0.8 0.6 oo g V 0.4 0.2 0.0 T 1 I 1 1 1 ! 1 COSMOS j r 2 2 2 3 2 4 2 5 2 6 i115 A B magnitude 2 7 22 23 24 25 26 6 7 7 5 A B magnitude 27 Figure 7.6: Cumulative distribution of i775 (left panel) and P 4 3 5 (right panel) AB magnitudes for submm counterparts in GOODS-N. In each panel, the solid curve is for the secure (P < 0.05) counterparts while the dashed curve contains all counterparts, offset slightly in the y-direction for clarity. The dash-dot line shows the 10cr limit for point sources in the i 8 1 4 COSMOS survey (Koekemoer & Scoville 2005). 7.2. DISCUSSION 185 still undetected. These sources are detected in radio and/or Spitzer, but have ^775 > 28 AB mag. Interestingly, this sub-sample of sources does not show systematically different fluxes at 850 or 24 /jm, which might be expected if they are significantly more dusty. The lack of optical flux for these sources means that I cannot obtain an optical photometric redshift, although their IRAC colours are consistent with them being at higher redshift (z > 1.5). 7.1.6 X-ray The distribution of full band (0.5-8 keV) Chandra X-ray fluxes for my submm sample is shown in Fig. 7.7. Note that only 13/35 of the 850 /im SCUBA super-map sources in GOODS-N are detected in the full band 2 Msec exposure with Chandra, and less than half of these are detected in both the hard and soft X-ray bands. All but 2 of these X-ray counterparts are also detected in the radio. The stacked X-ray spectrum for a sub-sample of the Chapman et al. (2005) sources is presented in Alexander et al. (2005b). This gives a good characterization of the X-ray emission from SMGs and demonstrates what could be achieved for individual objects with deeper Chandra .observations. 7.2 Discussion One of the main goals of the SCUBA-2 extragalactic surveys (see Holland et al. 2006) is to place SMGs in the context of galaxy evolution using clustering analysis, redshift distributions, luminosity functions and knowledge of the role of active galactic nuclei (AGN) and starbusts (SB) in these systems. For the first time, SCUBA-2 surveys will produce large samples of high signal-to-noise ratio sources. However, the success of achieving these goals relies on having deep multiwavelength follow-up data. I have presented radio, infrared, optical and X-ray flux distributions for a complete sam-ple of 850/im selected sources in GOODS-N. Table 7.1 summarizes the results by giving the depths needed at several wavelengths to detect a given fraction of 850 /im SCUBA galaxies. Note that simply having the counterpart in an image is not the end of the story, since there are 7.2. DISCUSSION 186 1.0 0.1 S0.5-8kev(10-18Wm-2) Figure 7.7: Cumulative distribution of X-ray flux (0.5-8keV) for submm counterparts in GOODS-N. The solid curve is for the secure (P < 0.05) counterparts while the dashed curve contains all counterparts. 7.2. DISCUSSION 187 Table 7.1: Depths needed to detect 850 //m SCUBA galaxies at other wavelengths based on all 33 counterparts in'the GOODS-N sample/The first column lists the percentage of SCUBA galaxies that will be detected at the given multiwavelength depths. Entries with a dash indicate that 1 still do not have deep enough data at these wavelengths to identify 90% of the sample. Radio and MIPS sources are rare enough on the sky that I can use a 3cr catalogue when searching for submm counterparts, but I need to go to a high SNR catalogue at optical wavelengths. ID Depth needed rate 1.4 GHz (3a) 24//m(3a) i775 (5a) 50% 30//Jy 100 /Jy 25 AB mag 70% 18/Jy 70/Jy 27.8 AB mag 90% — 20 /Jy — often still several possible counterparts to choose from (see e.g. Chapter 3). Given that I have a template for determining how deep one needs to go at other wavelengths to detect submm counterparts, one can ask which wavelengths are most useful for learning more about the nature of SMGs. Throughout much of the submm literature, it is clear that high angular resolution radio observations are invaluable for localizing the submm emission (e.g. Ivison et al. 2002). With SCUBA-2, there will be large samples of submm sources detected at high signal-to-noise ratio, therefore the submm positional accuracy should be greatly improved, making statistically se-cure identifications a bit easier. Apart from helping with the identifications, radio observations are also key to establishing the radio-IR correlation at high redshift in different systems and investigating the presence of AGN using radio spectra and morphology (Chapman et al. 2004a). Obtaining accurate redshifts and stellar masses for SMGs is a key step to determining their role in galaxy evolution. While spectroscopic observations are costly for large samples of 7.2. DISCUSSION 188 optically-faint (or nearly invisible) SMGs, progress can be made using optical/IR photometric redshifts (Pope et al. 2005; Chapter 3) and/or submm/radio photometric redshifts (Hughes et al. 2002, Aretxaga et al. 2003). Combining the stellar mass estimates with dust mass estimates (Dunne et al. 2000) from well-sampled IR SEDs provides a more complete picture of the ob-served and obscured star formation in SMGs and how it is evolving. Borys et al. (2005) looked at the stellar masses of 13 SMGs which have spectroscopic redshifts and found significant stel-lar populations already in place at high redshift. For SMGs at z ~ 2, the peak of the emission from evolved stars lies in the middle of the 4 IRAC channels and therefore deep IRAC data in all channels are needed to break degeneracies between stellar age, opacity and redshift. In ad-dition, a long baseline in the mid-IR (at least 3 points observed between 3.6-24 pm, including a detection at 5.8 and/or 8.0 imi) is needed to weed out a power-law component indicative of an AGN in these high redshift systems (e.g. Sajina et al. 2005). X-ray observations are also important here, as they provide an independent probe of AGN activity. As I have shown, a large fraction of SMGs are still undetected in the deepest optical data, although deep J and if-band observations provide useful constraints on the optical SEDs (Borys et al. 2005). Deep near-IR data will also allow for a more detailed comparison between galaxies selected at near-IR wavelengths, such as BzK galaxies (Daddi et al. 2004) and distant red galaxies (DRGs, Franx et al. 2003) and the SMGs; currently there is little overlap between these populations and the small samples of SMGs (Daddi et al. 2005, Knudsen et al. 2005). The unfortunate conclusion from all of this is that comprehensive studies of the SMG population will require deep and wide imaging over a broad range of wavelengths. Hence, the detailed study of these obscured star-formers seems destined not to be easy and progress will likely be made by focusing on sub-samples which are brightest at one or more other wavebands. In addition to these more traditional wavebands, follow-up studies of SMGs suffer from a lack of high resolution observations surrounding the peak of the IR SED (30-200 pm rest frame). This gap will be filled by future missions such as the Herschel Space Observatory and currently with SHARC-II on Caltech Submillimeter Observatory (CSO, e.g. Kovacs et al. 2006). Combined with redshift information, this will allow for an accurate measure of the submm luminosity function. 7.2. DISCUSSION 189 While progress can be made in understanding SMGs using only portions of follow-up data (e.g. only near-IR data), a fully sampled SED from from X-ray to radio wavelengths is required to provide a complete picture. For future ambitious mm/submm surveys, such as with SCUBA-2, the challenge will be to obtain deep multiwavelength coverage over wide areas. 190 C H A P T E R 8 S U M M A R Y AND FUTURE WORK 8.1 Summary In the thesis, I have studied a sample of high redshift SMGs in order to put constraints on their role in galaxy evolution. Starting with the largely uncertain submm positions, I used multi-wavelength observations and statistical criteria to assign counterparts. I found statistically likely counterparts for the majority of my sample, which means that I can, for the first time, put constraints on essentially the whole submm population. I estimate redshifts and find that the bulk of SMGs live near z ~ 2 with very few galaxies at z < 1 and z > 3. This is a busy time in the history of the Universe when a large fraction of the stars and galaxies seen locally were assembled. I studied the spectral energy distributions of SMGs from near-IR to radio wavelengths and found that they have cooler dust temperatures than local ULIRGs. This implies that the luminosity-temperature relation at high redshift differs from that seen locally. The SEDs of SMGs also differ from those of their high redshift neighbors, the BzK near-IR selected galax-ies. The SED fits show that SMGs are consistent with the correlation between radio and IR luminosity observed in local galaxies. I explore the role of AGN, or black hole accretion, in these galaxies through mid-IR spec-troscopy. I find the mid-IR regime to be dominated by strong emission features due to dust, which rules out a major contribution from AGN emission at these wavelengths. Extrapolating to the far-IR I find that SMGs are truly starburst dominated systems, forming stars at an intense rate of ~ 1000M o yr _ 1 . This cannot be said for local ULIRGs, which often show a mix of AGN and SB activity. Assuming an evolutionary link, this places SMGs before ULIRGs in the sequence from a major merger to a quasar to a massive elliptical galaxy. I also presented the highest redshift SMG discovered to date at z = 3.955 and showed 8.1. SUMMARY 191 how it is likely complicated by lensing from a foreground galaxy cluster. This SMG led to the discovery of this high redshift galaxy cluster and therefore this is a potentially new method for discovering galaxy clusters. I resolved essentially the entire 850 pm background by stacking the submm flux from galax-ies detected with Spitzer. While it is still not understood which galaxies are dominating the sig-nal, it is clear that there is little room for a significant contribution to the submm background from galaxies beyond z = 4. And finally, I have presented the multi-wavelength flux distribu-tions for this SMG sample in order to help plan follow-up efforts of future submm surveys. In the next section, I demonstrate how I can use this complete sample of SMGs to constrain the SFRD. 8.1.1 Star formation rate density At the beginning of this thesis I showed the plot of the SFRD as a function of redshift for a sub-sample of SMGs. Now that I have a complete sample of SMGs, which I know are powered by star formation and not AGN activity, I can return to this plot. Fig. 8.1 shows the SFRD as a function of redshift from Wall, Pope & Scott (2007). The light blue points are the sample of identified SMGs from this thesis. Using a maximum likelihood analysis, the luminosity function for the complete sample of SMGs with Sg50 > 2 mJy was calculated. The other points in Fig. 8.1 are for galaxies detected in UV, optical and mid-IR surveys and the curves are for quasars. The dust correction for the UV and optical points is fraught with uncertainties, especially beyond z ~ 2. The distribution of the SMGs in this plot follows very closely that of the quasars, with a peak near z ~ 2 and a redshift cut-off around z ~ 4. This is again consistent with the idea that the two populations are linked in an evolutionary sequence. Furthermore, the SMG points are among the highest in this plot, particularly for z < 2. This means that the SMGs, while much less numerous than optically selected galaxies, are more important for the build-up of stars and massive galaxies in the Universe. The error bars for the SMG points in Fig. 8.1 are still fairly large. This is understandable given that my sample only contains ~ 40 objects. With future wide-field submm surveys, this 8.2. FUTURE TELESCOPES AND INSTRUMENTS 192 sort of detailed luminosity function analysis will be able to put much tighter constraints on the contribution to the SFRD from SMGs and hence on their evolution. 8.2 Future telescopes and instruments The sample of observed SMGs will soon increase dramatically with the commissioning of SCUBA-2 (Holland et al. 2006) on the JCMT (expected to start in early 2008). One of the main goals of the SCUBA-2 extragalactic surveys (see Holland et al. 2006) is to place SMGs in the context of galaxy evolution using clustering analysis, redshift distributions, luminosity functions and knowledge of the role of AGN and starbursts in these systems. The success of achieving these goals relies on having deep multiwavelength follow-up data. In addition to mapping large areas up to a thousand times faster than SCUBA (Holland et al. 2006), SCUBA-2 will provide access to a new population of galaxies. Because of the slow mapping and limited number of nights with really good weather, no more than a dozen robust 450 pm galaxies were discovered in blank-field extragalactic SCUBA maps. With the increased mapping and sampling speed of SCUBA-2, we will take advantage of the limited nights with really good weather to obtain the first detailed picture of the 450 pm sky. The sources which make up the 450 pm background are likely to be at lower redshifts and hotter than the 850 pm population, and could contain a new population of sources missing in other multi-wavelength surveys. The 450 pm galaxy population is likely to provide a link between the 850 pm galaxies and the sources detected in deep MIPS surveys. Multi-wavelength observations of the SCUBA-2 survey fields are already underway. In particular, the new Spitzer Legacy Survey: Far-Infrared Deep Extragalactic Legacy (FIDEL) survey (PI: Mark Dickinson) will image the deep SCUBA-2 fields at 24 and 70 pm. The depths of the 70 pm data in this program are very well matched to the deep 450 pm observations in the SCUBA-2 Cosmology Survey and therefore together they provide a valuable probe of the star formation activity in star forming galaxies over a wide redshift range. Using these two datasets (24, 70, 450 and 850 pm) together, we will be able to characterize the shape of the IR SED to obtain accurate estimates of the star formation rates in these IR-luminous galaxies. An 8.2. FUTURE TELESCOPES AND INSTRUMENTS 193 1 1 1 1 1 1 T I _ I i i i i i i i u 0 2 4 6 8 r e d s h i f t Figure 8.1: Star formation rate density (SFRD) as a function of redshift. This figure is taken from Wall, Pope & Scott (2007). Light blue points are the GOODS-N SMGs, while the black and grey curves are the current estimates for quasars. Data from the optical and near-IR are denoted by squares and the solid green circles indicate the extinction corrected values for some of these points - the light blue shaded region is a model prediction. Orange solid circles are from mid-IR observations and the dark red and blue circles are the Chapman et al. (2005) points (see Fig. 1.2). See Wall et al. (2005) for a full list of references for these data. 8.2. FUTURE TELESCOPES AND INSTRUMENTS 194 uncertainty of 15% in the dust temperature leads to an uncertainty of 50% in the IR luminosity, as derived from a single long wavelength observation (Blain et al. 2002). Therefore, having detection at both IR and submm wavelengths is the only way to accurately determine the IR luminosity and calibrate the star formation rate in different types of galaxies. For populations of galaxies which remain undetected in one of these IR or submm wavelengths, we can use stacking techniques, where global properties of the population can be determined (e.g. Daddi et al. 2005, Huynh et al. 2007). Additionally we will be able to compare the dust temperatures and masses for galaxies selected from the 24, 70, 450 and 850 pm surveys to investigate the evolution of these properties with luminosity and redshift. Specifically, we can investigate how the luminosity-temperature (L-T) relation changes with redshift and galaxy type. This should provide valuable insights into the selection effects at different wavelengths and provide, for the first time, a more complete picture of the star formation at these redshifts, including both warm and cool LIRGs and ULIRGs. Fig. 8.2 shows the S 4 5 0 / 5 7 0 flux ratios as a function of redshift for models with different average dust temperatures. I also plot the current estimate of this ratio for 850 /urn-selected galaxies (in two redshift bins) and BzK galaxies based on their best-fit SEDs (Chapter 3; Daddi et al. 2005; Huynh et al. 2007). This hints at a variation in the dust temperatures of these systems. Whether this variation is due to selection effects or true evolution requires populating this plot with larger samples of galaxies detected at both IR and submm wavelengths. A key step in this analysis is establishing the contribution of an AGN to the IR luminosity in these systems.. The. success'Of my current IRS program at revealing the nature of SMGs (see Chapter 4) has demonstrated the potential for using mid-IR spectroscopy to separate the AGN and SB components. IRS observations of high redshift (z > 1) sources selected at 70 pm will provide a complementary dataset for determining the differences between high redshift ULIRGs selected in different ways. Combined with results from the literature and the archive for other high redshift IR-luminous systems (e.g. Yan et al. 2007; Valiante et al. 2007), we will be able to characterize the mid-IR SED as a function of redshift, luminosity and sample selection. Combined with X-ray, optical and IRAC data, we will be able to separate the AGN and starburst components in these systems. This will put constraints on the nature of the AGN 8.2. FUTURE TELESCOPES AND INSTRUMENTS 195 in these systems and help determine how the growth of the AGN is related to the build-up of dust and stars. In addition to SCUBA-2 there are several other up-coming projects which will allow us to take the analysis of SMGs to the next level. BLAST, the Balloon-borne Large-Aperture Sub-millimeter Telescope1 (Devlin et al. 2004), flew from Antarctica in December 2006 and observed several deep extragalactic fields. With three bolometer arrays at 250, 350 and 500 /im, BLAST surveys will be able to put constraints on the dust properties of dusty galaxies. The AzTEC camera, which had a successful run on the JCMT in 2005/2006, and will be mounted on the 50 m diameter Large Millimeter Telescope2 (LMT) in 2007 for continuum observing at 1.1 /im. In the south, the Large Bolometer Camera for APEX (LABOCA) is expected to start observations on the 12 m Atacama Pathfinder Experiment (APEX3) at 870 pm any day now. The Herschel Space Observatory* and the Planck Surveyor5 are both scheduled for launch to-gether in late 2008. Herschel, a 3.5 m telescope, has two imaging instruments, the Spectral and Photometric Imaging REceiver (SPIRE) and the Photodetector Array Camera and Spectrom-eter (PACS), which together cover the wavelength range of 57-670 pm enabling us to place constraints on both sides of the far-IR dust peak. Planck will produce all-sky maps at millime-ter wavelengths to provide a complete census of the brightest SMGs in the Universe, but with very crude angular resolution. A key step in localizing the submm emission is observations at radio wavelengths. The Expanded VLA project (EVLA 6) has begun operations and will eventually achieve sensitivities of better than 1 /dy (at 2-40 GHz) at high resolution (down to several milliarcseconds). And finally within the next 5-10 years we will see the commissioning of the Atacama Large Millimeter Array (ALMA 7) and the James Webb Space Telescope (7WST8) to provide exquisite ' h t t p : / / c h i l e l . p h y s i c s . u p e n n . e d u / b l a s t p u b l i c / 2http://www.lmtgtm.org/ 3http://www.apex-telescope.org/ 4http://herschel.esac.esa.int/home.shtml 5http://www.rssd.esa.int/index.php?project=PLANCK 6http://www.aoc.nrao.edu/evla/ 7http://www.eso.org/projects/alma/ 8http://www.jwst.nasa.gov/ 8.2. FUTURE TELESCOPES AND INSTRUMENTS 196 100.00 F 0.011 • • • • i • • • • i • • • • i • • • • i • . . . i . . . . I 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Redshift Figure 8.2: Evolution of the S450/1S70 ratio with redshift for different SED models. The curves show the Chary & Elbaz (2001) templates with the effective dust temperature in-creasing from 20-55 K from the top to bottom. The symbols show the current estimate of this colour for 850 pm selected (split into low and high redshift bins) and BzK galaxies assuming the best-fit SEDs from Huynh et al. (2007), Pope et al. (2006), and Daddi et al. (2005), respectively. A 4a detection at both 70 and 450 pm (either for individual galaxies or for stacked averages) in future surveys will be able to distinguish between these different templates. 8.2. FUTURE TELESCOPES AND INSTRUMENTS 197 sensitivity and resolution at the longest submm and shortest IR wavelengths, respectively, to trace the the dust in normal to extreme star-forming galaxies out to the highest redshifts. Given that the submm selected galaxies are among the most massive galaxies at z ~ 2, an understanding of their role in galaxy evolution is crucial for constraining structure formation at these redshifts and understanding the downsizing phenomenon. These future projects will pro-vide a key contribution to our understanding of these outstanding issues in galaxy evolution. 198 BIBL IOGRAPHY Adelberger K.L., Steidel C.G., Pettini M , Shapley A.E., Reddy N.A., Erb D.K., 2005, ApJ, 619, 697 Allamandola L.J., Hudgins D.M., Sandford S.A.,1999, ApJ, 511, LI 15 Alexander D.M., et al., 2003a, AJ, 125, 383 Alexander D.M., et al., 2003b, AJ, 126, 539 Alexander D.M., Smail I., Bauer F.E., Chapman S.C., Blain A.W., Brandt W.N., Ivison R.J., 2005a, Nature, 434, 738 Alexander D.M., Bauer EE., Chapman S.C., Smail I., Blain A.W., Brandt W.N., Ivison R.J., 2005b, ApJ, 632, 736 Appleton P.N., et al., 2004, ApJS, 154, 147 Archibald E.N., et al., 2002, MNRAS, 336, 1 Aretxaga I., Hughes D.H., Chapin E.L., Gaztanaga E., Dunlop J.S., Ivison R.J., 2003, MN-RAS, 342, 759 Aretxaga I., et al. 2007, MNRAS, 379, 1571 Armus L., et al., 2004, ApJS, 154, 178 Armus L., et al., 2007, ApJ, 656, 148 Balogh M.L., et al., 2007, MNRAS, 374, 1169 Barger A.J., Cowie L.L., Sanders D.B., Fulton E., Taniguchi Y., Sato Y, Kawara K., Okuda H., 1998, Nat, 394, 248 Bibliography 199 Barger A.J., Cowie L.L., Sanders D.B., 1999, ApJ, 518, L5 Barger A.J., Cowie L.L., Richards E.A., 2000, AJ, 119, 2092 Bauer EE., Alexander D.M., Brandt W.N., Hornschemeier A.E., Vignali C , Garmire G.R, Schneider D.P., 2002, AJ, 124, 2351 Bertin E., Arnouts S., 1996, A&AS, 117, 393 V ' " " ' Blain A.W., Ivison R.J., Smail I., 1998, MNRAS, 296, L29 Blain A.W., 1999, Ap&SS, 266, 261 Blain A.W., Smail I., Ivison R.J., Kneib J.-P., Frayer D.T., 2002, PhR, 369, 111 Blain A.W., Chapman S.C., Smail I., Ivison R., 2004, ApJ, 611, 52 Borys C.J.K., 2002, PhD thesis, University of British Columbia Borys C , Chapman S., Halpern M., Scott D., 2003,' MNRAS, 344, 385 (Paper I) Borys C , et al., 2004a, MNRAS, 352, 759 (Paper II) Borys C , Scott D., Chapman S., Halpern M., Nandra K., Pope A., 2004b, MNRAS, 355, 485 Borys C , Smail I., Chapman S.C., Blain A.W., Alexander D.M., Ivison R.J., 2005, ApJ, 635, 853 Brandl B.R., et al., 2006, ApJ, 653, 1129 (B06) Brandt W.N., Hasinger G., 2005, ARA&A, 43, 827 Bruzual G., Chariot S., 2003, MNRAS, 344, 1000 Capak P., et al., 2004, AJ, 127, 180 Carilli C.L., Yun M.S., 1999, ApJ, 513, L13 Chabrier G., 2003, PASP, 115, 763 Bibliography 200 Chapman S.G., etal., 2000, MNRAS, 319, 318 >* Chapman S.C., Richards E.A., Lewis G.F., Wilson G., Barger A.J., 2001, ApJ, 548, L147 Chapman, S.C., Blain, A.W., Ivison, R.J., Smail, I.R., 2003a, Nature, 422, 695 Chapman S.C., Windhorst R., Odewahn S., Yan, H., Conselice C , 2003b, ApJ, 599, 92 Chapman S.C., Smail I., Windhorst R., Muxlow T., Ivison R.J., 2004a, ApJ, 611, 732 Chapman S.C., Smail I., Blain A.W., Ivison R.J., 2004b, ApJ, 614, 671 Chapman S.C., Blain A.W., Smail I., Ivison R.J., 2005, ApJ, 622, 772 (C05) Charmandaris V., Stacey G.J., Gull G., 2002, ApJ, 571, 282 Carilli C.L., Yun M.S., 1999, ApJ, 513, L13 Charmandaris V., et al., 1999, Ap&SS, 266, 99 Chary R., Elbaz D., 2001, ApJ, 556, 562 Chary R., et al., 2004, ApJS, 154, 80 Chary R., Dickinson M.E., Teplitz H.I., Pope A., Ravindranath S., 2005, ApJ, 635, 1022 Chary R.-R., 2006, ASPCS Conf. Proc. At the edge of the Universe, eds. J.Afonso, H. Fergu-son and R. Norris, in press (astro-ph/0612736) Chary R.-R., Teplitz H.I., Dickinson M.E., Koo D.C, Le Floc'h E., Marcillac D., Papovich C , Stern D., 2007, ApJ, 665, 257 Clavel J., et al., 2000, A&A, 357, 839 Cohen J.G., Hogg D.W., Blandford R., Cowie L.L., Hu E., Songaila A., Shopbell P., Richberg K., 2000, ApJ, 538, 29 Condon J.J., 1974, ApJ, 188, 279 Bibliography 201 Condon J.J., Anderson M.L., Helou G., 1991a, ApJ, 376, 95 Condon J.J., 1992, ARA&A, 30, 575 Condon J.J., 1997, PASP, 109, 166 Connolly A.J., Csabai I., Szalay A.S., Koo D.C, Kron R.G., Munn J.A., 1995, AJ, 110, 2655 Coppin K., Halpern M., Scott D., Borys C , Chapman S., 2005, MNRAS, 357, 1022 Coppin K., et al., 2006, MNRAS, 372, 1621 Coppin K., et al., 2007, MNRAS submitted Conselice C.J., Chapman S.C., Windhorst R.A., 2003, ApJ, 596, L5 Cowie L.L., Songaila A., Hu E.M., Cohen J.G., 1996, AJ, 112, 839 Cowie L.L., Barger A.J., Kneib J.-P, 2002, AJ, 123, 2197 Cowie L.L., Barger A.J., Hu E.M., Capak P., Songaila A., 2004, AJ, 127, 3137 Daddi E., Cimatti A., Renzini A., Fontana A., Mignoli M., Pozzetti L., Tozzi P., Zamorani G., 2004, ApJ, 617, 746 Daddi E., et al., 2005, ApJ, 631, L13 ' Daddi E., et al., 2007, ApJ, in press (astro-ph/0705.283) Dale D.A., Helou G., 2002, ApJ, 576, 159 Dale D.A., et al., 2005, ApJ, 633, 857 Dannerbauer H., et al., 2006, ApJ, 637, L5 Desai V., et al., 2007, ApJ in press (astro-ph/0707.4190) Bibliography 202 Devlin M.J., et al., 2004, Millimeter and Submillimeter Detectors for Astronomy II. Edited by Jonas Zmuidzinas, Wayne S. Holland and Stafford Withington Proceedings of the SPIE, Volume 5498, pp. 42-54 Dickinson M., et al., 2004, ApJ, 600, L99 Dole H., et al., 2006, A&A, 451, 417 Downes D., et al., 1999, A&A, 347, 809 Downes A.J.B., Peacock J.A., Savage A., Carrie D.R., 1986, MNRAS, 218, 31 Draine B.T., 2003, ARA&A, 41, 241 Draine B.T., Li A., 2007, ApJ, 657, 810 Dudley C C , Wynn-Williams C.G., 1997, ApJ, 488, 720 Dunlop J.S., Peacock J.A., Savage A., Lilly S.J., Heasley J.N., Simon A.J.B., 1989, MNRAS, 238,1171 Dunlop J.S., et al., 2004, MNRAS, 350, 769 Dunne L., Eales S., Edmunds M., Ivison R., Alexander P., Clements D.L., 2000, MNRAS, 315,115 Dunne L., Eales S.A., 2001, MNRAS, 327, 697 Dye S., etal., .2006; ApJ, 644, 769, ' . Eales S., Lilly S., Gear W., Dunne L., Bond J.R., Hammer E , Le Fevre O., Crampton D., 1999, ApJ, 515, 518 Eales S., Lilly S., Webb T., Dunne L., Gear W., Clements D., Yun M., 2000, AJ, 120, 2244 Eales S., Bertoldi E , Ivison R., Carilli C , Dunne L., Owen E , 2003, MNRAS, 344, 169 Egami E., et al., 2004, ApJS, 154, 130 Bibliography 203 Elbaz D., Cesarsky C.J., Chanial P., Aussel H., Franceschini A., Fadda D., Chary R.R., 2002, A&A, 384, 848 Elston R., Raines S.N., Hanna K.T., Hon D.B., Julian J., Horrobin M., Harmer C.F.W., Epps H.W., 2003, SPIE, 4841, 1611 Fang E , et al., 2004, ApJS, 154, 35 Farrah D., Afonsb J., Efstathiou A., Rowan-Robinson M., Fox M., Clements D., 2003, MN-RAS, 343, 585 Farrah D., Fox M., Rowan-Robinson M., Clements D., Afonso J., 2004, ApJ, 603,489 Fazio G.G., et al., 2004, ApJS, 154, 10 Fischer J., et al., 1999, Ap&SS, 266, 91 Fixsen D.J., Dwek E., Mather J.C., Bennett C.L., Shafer R.A., 1998, ApJ, 508, 123 Forster Schreiber, N.M., et al., 2003, A&A, 399, 833 Fox M.J., et al., 2002, MNRAS, 331, 839 Franx M., Illingworth G.D., Kelson D.D., van Dokkum P.G., Tran K.-V., 1997, ApJ, 486, L75 Franx M., et al., 2003, ApJ, 587, L79 Frayer D.T., et al., 2004, ApJS, 154, 137 Garrett M.A., de Bruyn A.G., Giroletti M., Baan W.A., Schilizzi R.T., 2000, A&A, 361, L41 Genzel R., et al., 1998, ApJ, 498, 579 Giavalisco M., et al., 2004, ApJ, 600, L93 Gispert R., Lagache G., Puget J.L., 2000, A&A, 360, 1 Gladders M.D., Yee H.K.C., 2000, AJ, 120, 2148 Bibliography 204 Gonzalez-Alfonso E., Smith H.A., Fischer J., Cemicharo J., 2004, ApJ, 613, 247 Greve T.R., et al.', 2005, MNRAS, 359, 1165 Halpern M., Benford R., Meyer S., Muehlner D., Weiss R., 1988, ApJ, 332, 596 Hauser M.G., Dwek E., 2001, ARA&A, 39, 249 Helou G., Soifer B.T., Rowan-Robinson M., 1985, ApJ, 298, L7 Higdon S.J.U., et al., 2004, PASP, 116, 975 Hildebrand R.H., 1983, QJRAS, 24, 267 Holland W.S., et al., 1999, MNRAS, 303, 659 Holland W.S., et al., 2006, SPIE conference, 6275, 45 Houck J.R., et al., 2004, ApJS, 154, 18 Hudgins D.M., Allamandola L.J., 2004, ASPC, 309, 665 Hughes D.H.et al, 1998, Nature, 394, 241 Hughes D.H., et al., 2002, MNRAS, 335, 871 Huynh M.T., Pope A., Frayer D.T., Scott D., 2007, ApJ, 659, 305 Iono D., et al., 2006, ApJ, 640, L l Ivison R.J., Smail I., Le Borgne J.-F., Blain A.W., Kneib J.-R, Bezecourt J., Kerr T.H., Davies J.K., 1998, MNRAS, 298, 583 Ivison R.J., Smail I., Barger A.J., Kneib J.-R, Blain A.W., Owen F.N., Kerr T.H., Cowie L.L., 2000, MNRAS, 315, 209 Ivison R.J., et al., 2002, MNRAS, 337, 1 Ivison R.J., et al., 2004, ApJS, 154, 124 Bibliography 205 Ivison R.J., et al, 2005, MNRAS, 364, 1025 Ivison R.J., et al, 2007, MNRAS, 380,199 Jaffe W., et al., 2004, Natur, 429,47 Jenness T., Lightfoot J.R, 2000, Starlink User Note 216, Starlink Project, CLRC Kasliwal M.M., Charmandaris V., Weedman D., Houck J.R., Le Floc'h E., Higdon S.J.U., Armus L., Teplitz H.I., 2005, ApJ, 634, LI Kaviani A., Haehnelt M.G., Kauffmann G., 2003, MNRAS, 340, 739 Kennicutt R.C., 1998, ARA&A, 36, 189 Kennicutt R.C., et al., 2003, PASP, 115, 928 Kneib J.-P, van der Werf P.P., Kraiberg Knudsen K., Smail I., Blain A., Frayer D., Barnard V., Ivison R., 2004, MNRAS, 349, 1211 Kneib J.-P, Neri R., Smail I., Blain A., Sheth K., van der Werf P., Knudsen K.K., 2005, A&A, 434, 819 Knudsen K.K., et al., 2005, ApJ, 632, L9 Knudsen K.K:, et al., 2006, MNRAS, 368, 487 _ ' ' Koekemoer A.M., Scoville N.Z., 2005, NewAR, 49, 461 Kovacs A., Chapman S.C., Dowell CD. , Blain A.W, Ivison R.J., Smail I., Phillips T.G., 2006, ApJ, 650, 592 Lacey C , Cole S., 1993, MNRAS, 262, 627 Lagache G., Dole H., Puget J.-L., 2003, MNRAS, 338, 555 Lagache G., Puget J.-L., Dole FL, 2005, ARA&A, 43, 727 Bibliography 206 Le Floc'h, E., et al., 2005, ApJ, 632, 169 Lilly S.J., Eales S.A., Gear W.K.P., Hammer F., Le Fevre O., Crampton D., Bond J.R., Dunne L., 1999, ApJ, 518, 641 Lonsdale C., et al., 2004, ApJS, 154, 54 Lutz D., Spoon H.W.W., Rigopoulou D., Moorwood A.F.M., Genzel R., 1998, ApJ, 505, L103 Lutz D., Genzel R., Kunze D., Rigopoulou D., Spoon H.W.W., Sturm E., Tran D., Moorwood A.F.M., 1999, Ap&SS, 266, 85 Lutz D., Sturm E., Genzel R., Spoon H.W.W.,'Moorwood A.F.M., Netzer H., Sternberg A., 2003, A&A, 409, 867 Lutz D., Valiante E., Sturm E., Genzel R., Tacconi L.J., Lehnert M.D., Sternberg A., Baker A.J., 2005a, ApJ, 625, L83 Lutz D., Yan L., Armus L., Helou G., Tacconi L.J., Genzel R., Baker A.J., 2005b, ApJ, 632, L13 Makino N., Asano K., 1999, ApJ, 512, 9 McCarthy P.J., et al., 2007, ApJL, 664, 17 Menendez-Delmestre K., et al., 2007, ApJ, 655, L65 Mobasher B., etal.. 2004, ApJ, 600, L167 *. Mobasher B., et al., 2005, ApJ, 635, 832 Murphy T.W., Jr., Armus L., Matthews K., Soifer B.T., Mazzarella J.M., Shupe D.L., Strauss M.A., Neugebauer G., 1996, AJ, 111, 1025 Muxlow T.W.B., et al., 2005, MNRAS, 358, 1159 Oke J.B., & Sandage A., 1968, ApJ, 154, 21 Bibliography 207 Oliver S., et al., 2004, ApJS, 154, 30 Owen F.N., Keel W.C., Ledlow M.J., Morrison G.E., Windhorst R.A., 2005, AJ, 129, 26 Papovich C , et al., 2004, ApJS, 154, 70 Papovich C , et al., 2006, ApJ, 640, 92 Peacock J.A.-, 1999, Cosmological Physics. 1st ed. Cambridge University Press. 702 p. Phillips A.C., Guzman R., Gallego J., Koo D.C, Lowenthal J.D., Vogt N.P, Faber S.M., Illingworth, G.D., 1997, ApJ, 489, 543 Pilbratt G.L., 2001, phso.conf, 13 Pope A., 2004, MSc thesis, University of British Columbia Pope A., Borys C , Scott D., Conselice C , Dickinson M., Mobasher B., 2005, MNRAS, 358, 149 (Paper ill) Pope A., et al., 2006, MNRAS, 370, 1185 (Paper IV) Puget J.L., Leger A., 1989, ARA&A, 27, 161 Puget J.-L., Abergel A., Bernard J.-P, Boulanger F., Burton W.B., Desert F.-X., Hartmann D., 1996, A&A, 308, L5 Reddy N.A., Steidel C.C., Fadda D., Yan L., Pettini M., Shapley A.E., Erb D.K., Adelberger K.L., 2006, ApJ, 644, 792 Richards E.A., Kellermann K.I., Fomalont E.B., Windhorst R.A., Partridge R.B., 1998, AJ, 116,1039 Richards E.A., 2000, ApJ, 533, 611 Rieke G.H., et al., 2004, ApJS, 154, 25 Bibliography 208 Rigopoulou D., Spoon H.W.W., Genzel R., Lutz D., Moorwood A.F.M., Tran Q.D., 1999, AJ, 118,2625 " " • • Sajina A., Lacy M., Scott D., 2005, ApJ, 621, 256 Sajina A., 2006, PhD thesis, University of British Columbia Sajina A., Scott D., Dennefeld M., Dole FL, Lacy M., Lagache G., 2006, MNRAS, 369, 939 Sajina A., Yan L., Armus L., Choi P., Fadda D., Helou G., Spoon FL, 2007, ApJ, 664, 713 (S07) Salpeter E.E., 1955, ApJ, 121, 161 Sanders D.B., Soifer B.T., Elias J.H., Neugebauer G., Matthews K., 1988b, ApJ, 328, L35 Sanders D.B., Mirabel LE, 1996, ARA&A, 34, 749 Scheuer P.A.G., 1957, Proceedings of the Cambridge Philisophical Society, 53, 764 Scheuer P.A.G., 1974, MNRAS, 166, 329 Schweitzer M., et al., 2006, ApJ, 649, 79 Scott S.E., et al, 2002, MNRAS, 331, 817 Serjeant S., et al., 2004, ApJS, 154, 118 Shapley A.E., Steidel C C , Erb D.K., Reddy N.A., Adelberger K.L., Pettini M., Barmby P., Huang J., 2005, ApJ, 626, 698 Smail I., Ivison R.J., Blain A.W., 1997, ApJ, 490, L5 Smail I., Ivison R.J., Owen F.N., Blain A.W., Kneib J.-R, 2000, ApJ, 528, 612 Smail I., Ivison R.J., Blain A.W., Kneib J.-P, 2002, MNRAS, 331,495 Smith J.D.T., et al., 2007, ApJ, 656, 770 Bibliography 209 Soifer B.T., Sanders D.B., Madore B.F., Neugebauer G., Danielson G.E., Elias J.H., Lonsdale C.J., Rice W.L., 1987, ApJ, 320, 238 Spergel D.N., et al., 2007, ApJS, 170, 377 Spinrad H., Stern D., Bunker A., Dey A., Lanzetta K., Yahil A., Pascarelle S., Fernandez-Soto A., 1998, AJ, 116,2617 Spoon H.W.W., Keane J.V., Tielens A.G.G.M., Lutz D., Moorwood A.F.M., Laurent O., 2002, A&A, 385, 1022 Spoon H.W.W., et al., 2004a, ApJS, 154, 184 Spoon H.W.W., Moorwood A.F.M., Lutz D., Tielens A.G.G.M., Siebenmorgen R., Keane J.V., 2004b, A&A, 414, 873 Spoon H.W.W., Marshall J.A., Houck J.R., Elitzur M., Hao L., Armus L., Brandl B.R., Char-mandaris V., 2007, ApJ, 654, L49 Springel V., Di Matteo T., Hernquist L., 2005, MNRAS, 361, 776 Stanford S.A., et al., 2005, ApJ, 634, L129 Steidel C.C., Adelberger K.L., Dickinson M., Giavalisco M., Pettini M., Kellogg M., 1998, ApJ, 492, 428 "" . Sturm E., LutzD., Tran D., Feuchtgruber H., Genzel R., KunzeD., Moorwood A.F.M., Thorn-ley M.D., 2000, A&A, 358; 481 , Swinbank A.M., Smail I., Chapman S.C., Blain A.W., Ivison R.J., Keel W.C., 2004, ApJ, 617, 64 Tacconi L.J., et al., 2006, ApJ, 640, 228 Teplitz H.I., Charmandaris V., Chary R., Colbert J.W., Armus L., Weedman D., 2005, ApJ, 634,128 Bibliography 210 Teplitz H.I., et al., 2007, ApJ, 659, 941 Tilanus R.P.J, 2004, JCMT Technical Report TR/001/106, http://www.jach.hawaii.edu/-JACdocs/JCMT/tr/001/106 Tran Q.D., et al., 2001, ApJ, 552, 527 Valiante E., Lutz D., Sturm E., Genzel R., Tacconi L.J. , Lehnert M.D. , Baker A.J. , 2007, ApJ, 660, 1060 Vega O., Silva L. , Panuzzo P., Bressan A. , Granato G.L., Chavez M . , 2005, M N R A S , 364, 1286 Vignati P., et al., 1999, A & A , 349, L57 Wall J.V., Jackson C.A., Shaver P.A., Hook I.M., Kellermann K.I., 2005, A & A , 434, 133 Wall J.V., Pope A. , Scott D., 2007, M N R A S , accepted (astro-ph/0702682) Wang W.-H., Cowie L .L . , Barger A.J. , 2006, ApJ, 647, 74 Webb T.M., et al., 2003, ApJ, 582, 6 Wirth G.D., et al., 2004, AJ, 127, 3121 Yan L. , et al., 2005, ApJ, 628, 604 Yan L. , et al., 2007, ApJ, 658, 778 Younger J.D., et al., 2007, ApJ, in press (astro-ph/07081020) ' ' " Yun M.S., Reddy N.A., Condon J.J., 2001, ApJ, 554, 803 211 APPENDIX A R E L E V A N T F O R M U L A E A . 1 Observational cosmology Here I provide some of the key cosmological formulas used in this thesis. I follow the same notation as in Peacock (1999). At cosmological distances, the flux that is measured at the telescope for these galaxies, S, is related to its luminosity, L by the luminosity distance, L \ . L = AnDlS, (A.l) where L \ is the luminosity distance defined as DL = (l + z)R0Sk(r). (A.2) The luminosity distance depends on the comoving radial distance which is defined as R0Sk(r) = 4r / = , (A.3) #o h Vi1 - + z)2.+ ^ + On(l + zf + a.(i + zy where Q = Qv + Q m + Q r and Qv, Qm and Q r are the fractions of the critical density in vacuum, matter and radiation, respectively. Alternatively, the luminosity density at the emitted frequency, Lvt, is calculated using L " ~ TiT^ ' ( A- 4 ) where Su is the flux density at the at the observed frequency, v = z/( l + z). An alternate distance measure is the angular diameter distance which is defined as DA = (l + z)-1R0Sk(r). (A.5) A.l. POISSON PROBABILITIES 212 The comoving volume element is given by ; * dV = 4ir[P^Sk{r)fRodr. (A.6) In this thesis I assume a standard cosmology with with H0 = 7 3 k m s - 1 M p c - 1 , Qm = 0.3 and Q A = 0.7 (Spergel et al. 2007). A.2 Poisson probabilities As I explained in Chapter 3, the counterpart identification procedure relies strongly on the use of probabilities to asses the reality of the association. I followed the prescription for calculating these probabilities described in Downes et al. (1986). If the expected number of objects is the area multiplied by the surface density, n, then the Poisson probability of finding at least one object brighter than S within a radius 8 is given by P* = 1 - e x p ( - 7 m ( > S ) 0 2 ) . (A.7) I want to calculate the probability of finding an object with P < P* which depends on both the search radius, 8S, and the number density of all objects in the image, n T . The range within which P < P* is a function of the surface density. The expected number of events with P < P* is given by />71T E= ndinfdn. (A.8) Jo Since I am dealing with small probabilities, P* -C 1, then nd(n)2 = P*/n and the integral above is a logarithmic function. E must be finite, since I have a defined search radius, and so there is a critical density, nc when 9 — 8S. In that case P* = 7mc#2 and I define P c = im-rd2. The expected number of events becomes E = Pc, P* > Pc P* < Pc-A3. MAGNITUDES 213 Given the expect number of events, I can calculate the a posteriori probability of finding an object brighter than S within a radius 9 by For shallow catalogues the correction term of Pc becomes negligible. However, for deep catalogues, like the deep radio and IR catalogues used in this thesis, the correction term be-comes important and can make the probabilities several times greater. A. 3 Magnitudes Due to the multi-wavelength nature of this thesis, I often have to convert between magnitudes and fluxes. In order to keep things simple, I work with the AB magnitude system, where P — 1 — exp -E (A.9) m A B = -2.5 log S + 8.9. (A. 10) Here ?TZAB is the magnitude in the AB system and S is the flux in Jy. APPENDIX B 214 N O T E S O N I N D I V I D U A L S O U R C E S Here I summarize the identifications for the full sample of submm sources in GOODS-North, which are found at > 3.5a in the SCUBA super-map. There are 40 submm sources from the original super-map presented in Pope et al. (2005). In Section 2.1, I discussed the flux-deboosting of this sample. The following Paper III sources are not included in the analysis of this paper due to the fact that they are very noisy and have a non-negligible probability (> 5%) of having essentially zero flux density: GN27, GN29, GN33, GN36 and GN38. However, for completeness (and since it is likely that many of them are real sources), I also discuss the possible counterparts for these 5 less secure sources here. In 2006,1 re-reduced the super-map including some new photometry observations. This resulted in 3 new sources GN39, GN40 and GN41. I also discuss the counterparts to these 3 sources in this section. My procedure for counterpart identification is outlined in Section 3.1. Using the Spitzer data and the new radio reduction, I am now able to find unique counter-parts for 35/38 submm sources, 23 of these are secure counterparts (P < 0.05) and 12 are ten-tative counterparts (0.05 < P <0.20). The remaining three sources, GN08, GN35 and GN41, have multiple IRAC counterparts within the search region which appear equally likely and therefore I am unable to assign a unique identification. All of the submm sources with previ-ous VLA 1.4 GHz detections (Richards 2000) were confirmed in the new reduction (Morrison et al.,in preparation), and, in addition, there are four'sources from the Paper III radio-detected list which now have a second radio source within the search radius, with three of these being close enough to be regarded as interacting systems. Of the 12 optical counterparts suggested in Paper III for radio-undetected sources using the optical data, 7 are confirmed as being correct using the Spitzer data and the new radio images. The new data favour different counterparts for four of these sources, although the Paper III counterparts are still feasible in some cases and may make a sub-dominant contribution to the submm emission in others. The last source B.l. SECURE COUNTERPARTS 215 has several possible counterparts, including the Paper III counterpart. Six of these previously radio-undetected submm sources are now detected above 3a in the 1.4 GHz radio image. B . l Secure counterparts For each of the 23 sources described in this section, the probability that the counterpart is associated with the submm source at random, based on having a > 3a radio detection, MIPS 24pm detection, and/or red IRAC colours is < 5%. I now,discuss each of these sources in turn and refer to Fig. 3.1, which shows postage stamp images for each source at optical, IRAC, MIPS and radio wavelengths. GN03: There is no MIPS/IRAC detection of the Paper III identification. However, there is a relatively bright MIPS source, with a 4.3a radio detection, within the search radius. There is an IRAC detection of this source, although there is no ACS detection. GN04: There are two radio sources within the search radius, separated by 2.5 arcsec, with one of them twice as bright as the other. Both have an IRAC and a hard X-ray counterpart, but the fainter radio source lacks an optical counterpart. There is one 24 pm source located between the two, but it is difficult to separate this into two sources, given the 24 pm resolution. The original V L A source has a spectroscopic redshift of 2.6 (Chapman et al. 2005). At this redshift, the separation between the two radio sources is only ~ 20 kpc. The IRAC flux densities of the two sources are comparable. While the ACS image does not show any interaction between them, the morphology of the brighter radio source is disturbed and has multiple components (Paper III). I conclude that the submm emission is coming from two interacting galaxies at the same redshift. This submm source also has a relatively close companion (25 arcsec away), GN04.2 discussed in Section A3. However, the counterparts of each seem to imply that they are at different redshifts and not part of the same system. GN05: The only MIPS source in the search radius is coincident with the Paper III identifi-cation and has a new 3.9a radio detection. GN06: There is one likely counterpart which is detected in radio, IRAC, MIPS, and ACS, making this an unambiguous identification. B.l. SECURE COUNTERPARTS 216 GN07: There are two radio sources within the search radius, separated by 2.5 arcsec. One of these radio sources is three times brighter than the other. Both have an ACS, IRAC and MIPS counterpart. Interestingly the fainter radio source is brighter in the optical and mid-IR. The IRAC fluxes for the two sources are almost identical at all four wavelengths, and the shape of the IR SED is consistent with both sources being at the same redshift. Chapman et al. (2004a) report a spectroscopic redshift of 1.99 for the brigher optical, fainter radio source. At this redshift 2.5 arcseconds corresponds to about 20 kpc, so it is likely that this is an interacting or multi-component system. GN10: The Paper III identification has an IRAC detection, however there is a new 4.5cr radio detection 2.5 arcseconds from this system. The new radio source has a faint IRAC de-tection but no optical counterpart. There is very faint MIPS emission which extends between these two sources. The Paper III counterpart is at z = 1.344 and therefore it should be fairly bright at 24 ^ m if it is a submm source. Since it is not, I conclude that the new radio source is the most likely submm counterpart, but note that the Paper III counterpart and this new ra-dio source could be associated. The optical counterpart from Paper III is classified as being asymmetric, which is an indication of a merging, or disturbed, system. GNU: In Papers II and III this submm source was identified with a Westerbork Synthesis Radio Telescope (WSRT) radio source (Garrett et al. 2000). However, the positional uncer-tainly of the WSRT source is ~ 3 arcseconds. There is no MIPS/IRAC detection of the WSRT source, but on the other hand there is a new radio source with a MIPS/IRAC counterpart al-most 5 arcseconds from the WSRT position which is undetected in the optical. By fitting the 8 5 0 / 2 4 pm flux to a suite of templates, I find that neither of the two other MIPS sources in the search radius are likely to be the source of the submm emission, and because the radio source has a low probability of being there at random, I assign the new radio source as the counterpart. This is the only radio-detected Paper III counterpart which has changed. This submm detection is also in the Chapman et al. (2005) catalogue, although the suggested identification there is different, and there is no detection of their counterpart in the new radio image. At a redshift of 0.5, the source in the Chapman et al. (2005) catalogue is not likely to correspond to the submm source, given the lack of a bright VLA radio detection and low 24 pm flux. B.l. SECURE COUNTERPARTS 217 GN12: There is one likely counterpart, which is detected in radio, IRAC, MIPS and ACS, making this an unambiguous identification. The MIPS image in this region is crowded and blending is an issue, therefore there is a large systematic uncertainty in the 24 pm flux. I found that the addition of the IRAC photometry improved the photometric redshift estimate significantly, and therefore I use the new redshift in this thesis. In the ongoing campaign to obtain more redshifts in GOODS-N, several of the submm counterparts have been targeted with the Low Resolution Imaging Spectrometer (LRIS) on Keck. However, due to poor weather and the faintness of the sources in the optical, many of the spectra have come up blank. The LRIS slit for GN12 happened to pick up another 24 pm galaxy ~ 8 arcsec away from the SCUBA position. Strong Lyman alpha emission was detected at z = 2.006 from this serendipitous source (12h36m46.73s, +62°14'46.3"). While this is near the peak of the redshift range for submm counterparts, it is elimimated as a possible counterpart because: 1) it is too far from the SCUBA position; and, 2) it is bright in the ultraviolet, which is not a typical characteristic of submm counterparts. In fact, the submm observations of ultraviolet bright galaxies such as LBGs have shown that while both populations are star forming galaxies at similar redshifts, the bright submm and LBG populations overlap very little (Chapman et al. 2000; Webb et al. 2003). However, there is evidence for strong clustering between bright submm and LBG galaxies which might indicate that they lie at the same redshifts (Webb et al. 2003; Borys et al. 2004a). Therefore, while this new z = 2.006 24 /mi galaxy is not likely to be producing the submm flux, it may be part of the same high redshift system. GN13: The counterpart to this source is detected in radio, IRAC, MIPS, and ACS. While there are two additional MIPS sources within the search radius capable of producing the 850 pm emission, neither of them have a radio detection, which might be an indicator of the correct identification. This submm counterpart has the lowest redshift in my sample, with £ s p e c = 0.475 and is also detected in 70 pm MIPS imaging (Huynh et al. 2007). The coun-terpart to this submm source is a supernova host (Chary et al. 2005), which is the highest IR luminosity supernova host in GOODS-N at z < 0.5, and therefore it is not surprising that it is the only one which is detected with SCUBA. At this low 850 pm flux level, it is reasonable for the submm emission to be associated with the supernova host galaxy (see Farrah et al. 2004). B.l. SECURE COUNTERPARTS 218 GN14: This source is also known as HDF850.1, and is discussed in detail in several papers (see Dunlop et al. 2004, and references therein). There is a low redshift (zspec = 0.300), radio detected galaxy 6 arcseconds to the south-west of the submm position with a P value of 0.04. While this would qualify as a secure counterpert under my criteria, there is additional evidence which suggests that the counterpart for GN14 is much more complicated. Due to a detection with IRAM and MERLIN, the counterpart to GN14 is thought to lie behind a foreground elliptical galaxy very near the submm position. In the new reduction of the VLA data, a 2.7a peak is found at the position of the Dunlop et al. (2004) counterpart. The position of the elliptical and the submm counterpart are less than an arcsecond apart. There is an IRAC and MIPS detection of this system, but given the resolution of the Spitzer data, it is difficult to determine the contributing flux from the submm counterpart and elliptical separately. There is also an additional complication from the foreground elliptical galaxy possibly lensing the submm system. Nevertheless, the probability of randomly finding a red IRAC source which is also detected at 24 pm this close to the submm position is < 0.05. The faintness of the MIPS flux is consistent with the photometric redshift for the submm source being ~ 4 (Dunlop et al. 2004). I list this source in Table 2.1 and Table 3.1 and consider it a secure identification. However, since I can only put upper limits on the IRAC and MIPS fluxes, I leave it out of the figures and analysis in this thesis. GN16: There is one likely counterpart which is detected in radio, IRAC, MIPS and ACS, making this an unambiguous identification. GN17: There are two radio sources within the search radius, separated by 5.5 arcsec. One of them is three times brighter then the other. Both sources have an IRAC/MIPS counterpart, although the fluxes of the IRAC and MIPS counterparts are different by factors of 5 and 10, respectively. The brighter radio source is much fainter optically and has a photometric redshift of 1.72 (Paper III) which would indicate a separation of 45 kpc if both sources are at the same redshift. This is perhaps a little too large to be considered one system and, in fact, the fainter radio source has a photometric redshift of 1.2. By fitting the MIPS flux to a range of templates I find that the brighter radio source is more likely to be the dominant submm emitter, and therefore assign this as the counterpart. Cohen et al. (2000) list a spectroscopic redshift for this B.l. SECURE COUNTERPARTS 219 galaxy of 0.884, but given its optical and IR colours, such a redshift seems very unlikely. This counterpart satisfies the 'BzK' criteria for classification as a massive galaxy at z > 1.4 (see Daddi et al. 2005 for more details), and therefore I use the photometric redshift rather than the apparently discrepant spectroscopic redshift. GN18: There is only one MIPS source in the search radius and it has an IRAC counterpart but no optical detection. There is also a new 3.7a radio detection of this source. GN19: There are two equally bright radio sources within the search radius, separated by 3.0 arcsec. Both have an IRAC/MIPS counterpart, but the brighter MIPS source lacks an asso-ciated optical source. Nevertheless, both of these objects are confirmed to have a spectroscopic redshift of 2.484 (Chapman et al. 2005). At this redshift, the separation is 24 kpc and I conclude that the submm emission is coming from two interacting galaxies at the same redshift. GN20: The counterpart to this submm source is now unambiguously known, due to detec-tions with both IRAM and the SMA (Pope et al. in preparation; Iono et al. 2006). The new radio reduction has also revealed a source at 1.4 GHz which looks extended. This counterpart is detected in MIPS, IRAC and ACS. GN20 is discussed in detail in Chapter 5. GN20.2: The counterpart to this source is detected in radio, IRAC, MIPS and ACS. It is the companion to the bright submm source GN20 (18 arcsec away) and is also a B-dropout galaxy. Note that GN20.2 was first published in Chapman et al. (2001) when they targeted optically faint radio sources. However, so far it has eluded a spectroscopic identification (S. Chapman, private communication). GN22: There is one likely counterpart, which is detected in radio, IRAC, MIPS and ACS, making this an unambiguous identification. The other MIPS source within the search radius which is coincident with an edge-on galaxy (see Fig. 3.1) must contribute less than half of the submm emission since it is at lower redshift (zspec = 0.6393). GN25: There is one likely counterpart, which is detected in radio, IRAC, MIPS and ACS, making this an unambiguous identification. Note that the radio emission is extended (see Fig. 3.1). GN26: There is a bright radio, MIPS, IRAC and optical counterpart with a spectroscopic redshift of 1.219. Given the high flux density at radio and mid-IR wavelengths, this is consid-B.l. SECURE COUNTERPARTS 220 ered a secure counterpart even though it has a relatively large offset from the submm position. Because of its relatively low redshift, this source is also detected in the deep 70 pm imaging (Huynh etal. 2007)./- v -GN30: This source was identified with a faint 8.5 GHz radio source in Paper II, although there was no detection at 1.4 GHz in the Richards (2000) catalogue. In Paper III I found an optical counterpart at the 8.5 GHz position. There is an IRAC/MIPS detection of this galaxy, and in addition there is a 3a peak in the new 1.4 GHz radio image. This is a very blue, high surface brightness 'chain galaxy', more similar to typical LBGs than most submm counterparts. However, given the low 850 pm flux, it is perfectly reasonable that this is the counterpart. Given the radio and mid-IR detection, it qualifies as a secure counterpart. There is another MIPS source in the search radius, although it is at lower redshift (zspec = 0.557). GN31: There is no MIPS or IRAC detection of the Paper III counterpart, but there is one relatively bright MIPS source in the search radius, which has a 3a peak in the new radio image. This MIPS source is detected with both IRAC and ACS, and it is one of the lowest redshift counterparts, zspec = 0.935. The following two SMGs were added to my submm sample in 2006 after including some additional photometry data to the super-map. Since they joined the sample later, they are not included in all of the analysis in this thesis (e.g. Chapter 3), however I discuss them here for completeness. GN39: This is another double counterpart system. Two radio/MIPS/IRAC/X-ray counter-parts are separated by 7 arcseconds and both have spectroscopic redshifts of around 2 (Swin-bank et al. 2004; Chapman et al. 2005). The northern counterpart, GN39a, shows several knots in the optical images while the other counterpart, GN39b, is very faint and diffuse in the opti-cal images. I obtained mid-IR spectra of both counterparts to GN39 (see Chapter 4 for more details on this SMG system). GN40: The counterpart to this source is detected in the radio, MIPS, IRAC and X-ray images but it is not detected in the optical images. I use the IRAC photometry to estimate a photometric redshift of 2.6. The radio flux is relatively strong and the X-ray detection is very hard, both of these indicate that this source may be more like an AGN/QSO type source. B.2. TENTATIVE COUNTERPARTS B. 2 Tentative counterparts 221 The following 12 sources have likely radio, 24 yum, and/or IRAC counterparts but, because the probability of random association is above 0.05, the identification is listed as tentative. GN01: There is only one MIPS source in the search radius. It is coincident with the Paper III optical identification and has a 3.5a detection in the new radio map. GN02: There is a new 5.6a radio source just beyond the search radius. The new radio source has a bright MIPS, IRAC and ACS counterpart. There are no other MIPS sources within the search radius, and I therefore assign this radio source as the counterpart. This is the only proposed counterpart which is beyond the search radius, but since the submm noise is high, it is not unreasonable for it to have a greater positional offset. GN04.2: There is only one MIPS source in the search radius and it has an IRAC and ACS detection and a new 3a radio detection. This submm source is the companion to GN04. However, the redshift estimates of the counterparts seem to imply that they are independent systems. GN09: There are two MIPS/IRAC sources within the search radius. By fitting the 850/24 yum flux to a suite of templates, I find that the second source is fainter at 24 yum and is not capable of contributing more than 10 % of the submm flux. In addition the brighter 24 pm source has red IRAC colours, which means that it has a low probability of being within the search radius at random. I therefore assign the brighter MIPS source as the counterpart. This counterpart is not detected in the ACS or radio images, and is therefore likely to be at high redshift. GN15: There is a MIPS/IRAC detection of the Paper III counterpart. This source is bright at 24 yum and has red IRAC colours. The VLA source to the North, discussed in Paper III, is detected in the new radio image and has a P value of 0.05. Although there appears to be extended MIPS emission close to the radio source, it is very faint, and moreover, there is no IRAC or ACS galaxy associated with it, although there are several nearby. All of the other SMGs which are radio detected are bright at 24 yum and therefore the mid-IR faint, radio source is not typical of SMG counterparts. Since the 24 yum flux of the Paper III counterpart is so high there is a lower probability that it is a random association. This is a rare case where a radio B.2. TENTATIVE COUNTERPARTS 222 detection is not considered the most likely counterpart. GN21: There is only one MIPS source in this region and it has a 4.7a radio detection and a red IRAC counterpart. This source is undetected in the optical and the IR photometric redshift puts it at 3.3. There are no other likely counterparts in the search radius. GN23: This source has a radio, IRAC, and MIPS detection. However, it is only a faint smudge in the ACS image, and therefore I am not able to obtain an accurate optical photometric redshift, but I do have an estimate of the redshift from the Spitzer photometry (see Section 3.3.3 and Table 3.2). GN24: There is only one MIPS source in the search radius and it is coincident with the Paper III counterpart. There is no radio detection but it is coincident with a red IRAC source. GN28: This source is unique in that it is within 8 arcseconds of a massive radio jet (Borys et al. 2002). Within the search radius, there is faint MIPS emission, elongated in the NE-SW direction, which connects several IRAC sources. One of these IRAC sources is 0.6 arcseconds from the end of a radio jet and so the usual assumption that radio flux is indicative of the submm counterpart is more complicated here. The ~ 15 arcsec long radio feature is the most prominent in the entire GOODS-N field (Richards et al. 1998). There is a disturbed optical detection coincident with the end of the radio jet, which is confirmed to be at a similar redshift (zspec =. 1.020, Phillips et al. 1997) to that of the central AGN (z s p e c = 1.013, Richards et al. 1998). The fact that the submm emission seems coincident with the end of the jet is suggestive of jet-induced star formation. This is not the only 850 pm selected source which lies near a radio jet. LE850.07 (also known as LH1200.096, Ivison et al. 2002; 2005) is a submm source associated with a radio lobe, however spectroscopy of the lobe and the core fail to put them at the same redshift and there are other candidate counterparts in the field. In both of these cases, it will, be necessary to. better localize the submm emissionbefore drawing firm conclusions. There is also another IRAC source with faint MIPS emission which is an optically-invisible X-ray source in the region: While there are several possible counterparts, the fact that other submm sources have been seen close to radio jets, which has a low probability of happening at random, convinces me to assign the MIPS/IRAC source at the end of the radio jet as a tentative counterpart. B.3. MULTIPLE COUNTERPARTS 223 GN32: The Paper III optical counterpart has a MIPS and IRAC detection, but it is not detected in new radio image. The 24 pm flux is fairly high and it has red IRAC colours. The only other MIPS source in the search radius is radio-detected and known to be at lower redshift ( A p e c — 0.761). This radio/MIPS source has a P value of 0.06, therefore appears statistically likely to be the submm counterpart. However, based on reasonable SEDs which are consistent with the rest of the sample, this source is not likely to dominate the submm flux, although of course it may contribute. Because of this I maintain that the Paper III source is the major submm emitter. • GN34: The Paper IIP counterpart has a MIPS and IRAC detection, however there are also two other MIPS sources within the search radius, both of which are capable of contributing to the submm flux. Based on Poisson statistics, the red IRAC/MIPS source nearest the submm centre has, by far, the lowest probability of random association and therefore I assign it as the tentative counterpart. GN37: The Paper III counterpart has a MIPS and IRAC detection, but it not detected in the new radio image. The MIPS flux is fairly low but it has red IRAC colours. B. 3 Multiple counterparts The following three submm sources have multiple possible counterpart within the search radius which are equally likely. Since there is not one single counterpart which sticks out as having a significantly low probability of random association, I am unable to assign a unique counterpart for these sources. The submm emission from these sources could be coming from one of the multiple possible counterparts or from several of them. GN08: There is an IRAC detection of the Paper III counterpart, but no MIPS or radio detection. However, the lack of 24 pm detection may be because it is in the wings of much brighter MIPS sources, which are outside the search radius and therefore I can only obtain an upper limit on the 24 pm flux density. There are two MIPS detected sources within the search radius, but neither of these can contribute more than 10% of the submm flux, because they are at low redshift and faint at 24 pm. Since the probability of random association is B.4. POSSIBLE SPURIOUS SOURCES 224 not significantly low for either of these sources, I am unable to assign a unique identification. It will be possible to make further progress on this by localizing the submm emission using IRAM or SMA interferometry. • -,< ,, -v -GN35: There are no MIPS sources in the search radius, although there are three faint IRAC/ACS sources. Two of these can be eliminated as being likely counterparts, since their optical photometric redshifts put them at z < 0.1. The third IRAC source is detected in the ACS image and has a photometric redshift of 2.20. Given that all other submm sources in GOODS-N are detected in IRAC, there is no reason to believe that this one is not. However, since the probility of random association is not significantly low, I am unable to assign a unique identification. GN41: This is the 3rd of the 3 new SMGs which were discovered in the 2006 super-map. It does not have any radio or MIPS sources within the search radius, but there are several IRAC sources to choose from and therefore I am unable to assign a unique counterpart for this SMG. B.4 Possible spurious sources The following sources are no longer in my sample, due to the fact that each of them individually has a non-negligible probability (> 5%) of being at essentially zero flux density once I apply flux-deboosting (see Coppin et al. 2005). However, I discuss their possible counterparts here for completeness and because the chances are high of several of them being real. GN27: There is only one MIPS source in the search radius, and it has an IRAC and ACS counterpart. This is the most likely identification. GN29: There is only one MIPS source in the search radius and it has an IRAC and ACS counterpart. This is the most likely identification. GN33: There are three possible IRAC sources and no MIPS sources in the search radius, therefore I cannot assign a unique identification. GN36: There are no MIPS sources in the search radius. There are three IRAC/ACS sources, but nothing that stands out and therefore I cannot assign a unique counterpart. GN38: This source is near the edge of the original GOODS-N region, therefore there is B.4. POSSIBLE SPURIOUS SOURCES 225 no deep ACS coverage here, although there are radio and Spitzer data. There are no Richards (2000) VLA sources within the search radius, but there are three sources in the new radio reduction. The furthest radio source is coincident with a zspec — 0.1136 galaxy and therefore is not likely to be the submm counterpart. The two closer radio sources have similar radio flux, are separated by 4.5 arcseconds, and each of them is associated with an IRAC and MIPS source of comparable flux. Using just the infrared, submm and radio information, these two sources have very similar SEDs, which may indicate that they are at the same redshift. If this is true then these sources are both contributing to the submm flux (approximately equally in this case), and are possibly an interacting system at high redshift. Given the three other submm sources in this sample with two radio counterparts, I consider this to be the most likely scenario and consider both radio sources as the correct identification. Since they are radio-bright and close to the submm position, this system would classify as a secure counterpart. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0085807/manifest

Comment

Related Items