Submillimetre Observations of the Subaru Deep Field by Kr is ten E r i n Ka th ryn Coppin B.Sc. (Physics and Astronomy) University of V ic tor ia , 2000 A T H E S I S S U B M I T T E D IN P A R T I A L F U L F I L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F S C I E N C E in T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Department of Physics and Astronomy) We accept this thesis as conforming to the require^ standard T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A October 10, 2003 © Kr is ten E r i n Ka th ryn Coppin , 2003 In presenting this thesis in part ial fulfilment of the requirements for an advanced degree at the University of Br i t ish Columbia, I agree that the L ibrary shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publ icat ion of this thesis for f inancial gain shall not be allowed without my writ ten permission. Department of Physics and Astronomy The University Of Br i t ish Columbia Vancouver, Canada 11 A B S T R A C T We have measured the submill imetre wavelength continuum emission from the Subaru Deep F ie ld (SDF) at 450 and 850 (j,m wi th the Submil l imetre Common-User Bolometer Ar ray ( S C U B A ) detector on the James Clerk Maxwel l Telescope ( J C M T ) . The S D F is as deep a near-IR image as is available and contains four 'hyper extremely red objects' ( H E R O s ) . These data have allowed us to test the connection between 'extremely red objects' (EROs) found in IR surveys and the populat ion of bright submil l imetre sources found wi th S C U B A . Th is is an important bui lding block in our understanding of the star formation history of the universe. We examine the entire submil l imetre map of the S D F region and perform correlation analyses of a /^'-selected catalogue of galaxies in the S D F wi th our submil l imetre S D F map. We find that there is no clear correlation between the near-Infrared positions and submill imetre flux. We also present upper l imits to the fluxes of the four H E R O s , or a weak measurement of the average flux for the four of them. Our data are consistent wi th the H E R O s being Arp220-like objects out to z ~ 1.6. However, they would need to be one and a half times as luminous as Arp220 and extincted in the J -band by about 1 magnitude; a plausible scenario since only modest adjustments are made to Arp220's luminosity and dust content. i i i CONTENTS Abstract i i Contents i i i List of Tables v List of Figures . v i 1 Introduction 1 1.1 Galaxy Formation.and Evolut ion 1 1.2 Submil l imetre Galaxies 2 1.2.1 The Orig in of Submil l imetre Emission 2 1.2.2 Inhabitants of the High-Redshift Universe 3 1.2.3 Submil l imetre Emission as a High-Redshift Probe 3 1.2.4 E R O S 7 1.2.5 Hyper-Extremely Red Objects 8 1.3 The Subaru Deep F ie ld 10 1.4 Guide to This Thesis 15 2 The Instrument and Data 17 2.1 The J C M T and S C U B A 17 2.1.1 Observing Modes 18 2.2 Observations 21 2.2.1 Added Noise in the 2003 Data 24 2.3 Da ta Reduct ion 30 2.3.1 Prel iminary Reduction wi th S U R F . 30 iv 2.3.2 Mak ing the Maps 31 2.3.3 Cal ibrat ion 34 2.3.4 F l ux Measurements 35 3 H E R O s 40 3.1 Cirrus Contr ibut ion 40 3.2 H E R O I R Colours and 850 /mi Fluxes 41 3.3 Using Arp220 as a Template 41 3.4 Using a NGC3938 as a Template 51 3.5 Results 51 3.6 Are the Models Feasible? 51 3.6.1 Number Counts of Submil l imetre Sources 54 3.6.2 Cou ld the H E R O s be Starburst Galaxies? 54 3.6.3 Cou ld the H E R O s be Normal Spiral Galaxies? 61 4 S D F S o u r c e s 68 4.1 Correlat ing the IR Galaxies wi th the Submil l imetre Da ta 68 4.1.1 Detections 91 5 C o n c l u s i o n s 96 5.1 Future Work 98 B i b l i o g r a p h y 99 V LIST OF T A B L E S 1.1 Summary of IR information on the H E R O s 13 2.1 Summary of S C U B A observations 23 2.2 Summary of F lux Conversion Factors 36 2.3 Summary of S C U B A data for the H E R O s wi th upper l imi t flux estimates and photometric redshifts 39 3.1 Submil l imetre sources wi th red IR colours and known redshifts 49 3.2 Summary of S C U B A data for the H E R O s wi th upper l imi t f lux estimates and photometric redshifts 53 4.1 Detected sources 95 v i LIST O F FIGURES 1.1 The extragalactic background 4 1.2 Negative K-correction 6 1.3 I — KAB versus KAB 9 1.4 The Subaru Deep F ie ld i f ' -band image 11 1.5 J and i f ' - band images of the H E R O s 12 1.6 The 850 pm cumulative source counts 16 2.1 S C U B A wideband filter profiles 19 2.2 The S C U B A bolometer arrays 20 2.3 F F T s of jiggle map data . 25 2.4 F F T of the average of 2003 jiggle map data 27 2.5 Standard deviation of 10 integrations for 64 positions of the central bolometer 28 2.6 Contour plots of the standard deviation of 10 integrations for 64 positions of the central bolometer 29 2.7 Chopping strategy for photometry observations of the S D F 33 2.8 850 am S D F S N R map 37 2.9 450 pm S D F S N R map . . 38 3.1 Is there a colour-flux correlation for the H E R O s ? 42 3.2 S E D of Arp220 44 3.3 Comparison of 3 different luminosity S E D s 45 3.4 Effective wavelengths and relative extinction values from Schlegel, Finkbeiner & Davis [53] 46 3.5 Creat ing an extinction law 47 3.6 Testing our model against sources wi th known redshifts 50 v i i 3.7 S E D o f N G C 3 9 3 8 52 3.8 Expected 850 / im versus redshift using Arp220 55 3.9 /(' '-band brightness versus redshift using Arp220 56 3.10 Required J a n d I f ' -band dust extinction using Arp220 58 3.11 The Totani et al . [63] model of a typical el l ipt ical galaxy 60 3.12 Model of HR10 based on an Arp220 S E D 62 3.13 Expected 850 /tin versus redshift using NGC3938 63 3.14 iT -band brightness versus redshift using NGC3938 64 3.15 Required J and i f ' -band dust extinction using NGC3938 . . 65 4.1 Histogram of number of pixels at each flux level for the 850/ im map . . . 69 4.2 The difference of the flux bins and the flux bins reflected about 0 m J y for the 850 (ira. map 70 4.3 Histogram of number of pixels at each flux level for the 450 /tm map . . . 71 4.4 The difference of the flux bins and the flux bins reflected about 0 m J y for the 450 / im map 72 4.5 Histogram of the mean of the 20 brightest i f ' -band objects per redshift b in 74 4.6 Histogram of observed average 850 /tin flux per redshift bin 75 4.7 Histogram of observed average 450/tm flux per redshift bin 76 4.8 Distr ibut ion of correlation coefficients obtained from a set of Monte Car lo simulations of stacked 850 /zm flux in redshift bins 78 4.9 Frequency of obtaining a certain correlation coefficient from a set of Monte Car lo simulations of stacked 850 /tm flux in redshift bins 79 4.10 Distr ibut ion of correlation coefficients obtained from a set of Monte Car lo simulations of stacked 450 /tm flux in redshift bins 80 4.11 Frequency of obtaining a certain correlation coefficient from a set of Monte Car lo simulations of stacked 450 /tm flux in redshift bins 81 4.12 Average 850 /tm flux per redshift bin wi th an equal number of objects per b in 83 v i i i 4.13 A plot of observed average 450 am flux per redshift b in wi th an equal number of objects per bin 84 4.14 Distr ibut ion of correlation coefficients obtained from a set of Monte Car lo simulations of stacked 850 um flux wi th an equal number of objects per redshift bin 85 4.15 Frequency of obtaining a certain correlation coefficient f rom a set of Monte Car lo simulations of stacked 850 /um flux with an equal number of objects per redshift bin 86 4.16 Distr ibut ion of correlation coefficients obtained from a set of Monte Car lo simulations of stacked 450 /xm flux wi th an equal number of objects per redshift b in 87 4.17 Frequency of obtaining a certain correlation coefficient from a set of Monte Car lo simulations of stacked 450 am flux wi th an equal number of objects per redshift bin 88 4.18 Stacked 850 jum flux measured in K magnitude bins 89 4.19 Stacked 450 pm flux measured in K magnitude bins 90 4.20 Scatter-plots of 850 am flux for F-band detected objects 92 4.21 Stacked 850 pm flux measured in M v bins 93 4.22 Stacked 850 / /m flux measured in M v bins for objects wi th z > 2 and z > 4 94 ix Acknowledgements I would like to thank my research supervisor, Dr . Mark Halpern, for his expertise, guid-ance, patience, enthusiasm and financial support through N S E R C . Add i t iona l funding was received from the Nat ional Research Counci l of Canada and was used for tr ips to the J C M T to collect data for this project. Much thanks also goes out to the staff of the J C M T for their assistance wi th the S C U B A observations. The James Clerk Maxwel l Telescope is operated on behalf of the Part ic le Physics and Astronomy Research Counc i l of the Uni ted K ingdom, the Netherlands Organisation for Scientific Research, and the Nat ional Re-search Counci l of Canada. This research has made use of the N A S A / I P A C Extragalact ic Database ( N E D ) which is operated by the Jet Propulsion Laboratory, Cal i forn ia Institute of Technology, under contract with the Nat ional Aeronautics and Space Admin is t ra t ion. We acknowledge the use of N A S A ' s Sky View facil ity (http://skyview.gsfc.nasa.gov) lo-cated at N A S A Goddard Space Fl ight Center. In addit ion, Gui l la ine Lagache supplied her group's starburst and normal galaxy templates and an unpublished version of a pa-per describing these galaxy templates which was much appreciated. I would also like to thank A n n a Saj ina for help using the templates and for getting me started wi th IDL . M y gratitude goes out to Co l in Borys who helped me wi th using his map-making codes and for generally useful advice. Douglas Scott and James Dunlop also provided many useful suggestions and comments about the work contained in this thesis. 1 C H A P T E R 1 INTRODUCTION 1.1 Galaxy Formation and Evolution Two fundamental unsolved problems in modern cosmology are: how and when did the first galaxies form; and how long did their initial starbursts last? There are mainly two theories depicting how galaxies may have formed: the monoli thic collapse and hierarchical merging scenarios. The current consensus is that massive galaxies formed v ia hierarchical structure formation, but there is possible contrary evidence suggesting that a monoli thic collapse scenario may sti l l play a role in some instances. Based on the motions of old stars in our Galaxy, Eggen, Lynden-Bel l &: Sandage [27] conceived the monoli thic col-lapse scenario as a theory of galaxy formation, involving a single rapid collapse of stars at high redshift, producing a violent and short-duration burst of star-formation followed by a quiescent evolution of the stellar population. Th is theory was questioned by Searle & Z inn [54] as new and better data became available and they suggested that star formation in our Galaxy was a more prolonged chaotic process. New theories were subsequently developed (see Davis et al . [16]) describing galaxy formation as a "bottom-up" process, i.e. hierarchical merging, or the gradual formation of a galaxy by the merging of smaller collections of stars at moderate redshifts of z < 1.5 producing moderate and continuous star-formation (e.g. Whi te & Frenk [66], Kauffmann [44]). We see evidence of hierarchi-cal merging as our neighbouring satellite, the Sagittarius dwarf galaxy, is being t idal ly disrupted by our Galaxy 's gravitational field (see Ibata, Gi lmore &; Irwin [39]). Bu t older stellar populations of galactic spheroids seem to be better explained by a mono-l i thic collapse scenario. So we see evidence of both processes throughout the history of the Universe but we do not know how much each process plays into the entire galaxy formation picture. CHAPTER 1. INTRODUCTION 1.2 Submillimetre Galaxies 2 Very luminous high-redshift galaxies have recently been discovered at submil l imetre wave-lengths and may be significant players in deducing the history of galaxy formation and evolution. Far-Infrared and submill imetre galaxies are responsible for approximately 2 /3 of the total power output of al l known galaxies and they are excellent tracers of massive regions of star-formation at high redshifts (Gispert, Lagache & Puget [33]). Most of the star formation history of the Universe is "hidden" from optical sight, but remains trans-parent in the Infrared (IR) and submill imetre regimes. Careful submil l imetre surveys of distant galaxies should be able to shed some light on the relative importance of each for-mat ion process since submill imetre emission can indicate galaxies undergoing hightened rates of star-formation. (Bla in et al . [3]). 1.2.1 The Origin of Submillimetre Emission There are two different mechanisms responsible for submill imetre emission from galaxies: continuum thermal emission and line emission. Supernovae are thought to be the most l ikely dominant mechanism of dust production at high redshifts (Dunne et a l . [25]). These micron-sized interstellar dust grains, pr imari ly composed of silicates and polycycl ic aro-matic hydrocarbons (PAHs) , absorb ha rd -UV photons from regions of intense high mass star-formation or from Act ive Galact ic Nuclei ( A G N ) accretion disks and are heated to temperatures between 20-200 K. The obscuring dust grains re-radiate this absorbed ra-diat ion as thermal continuum emission, peaking in the Far-Infrared (F IR) . Th is emission becomes visible at submill imetre wavelengths when the cosmological expansion of the Universe redshifts the F I R peak into this regime. The other mechanism involves atomic and molecular transitions in the interstellar gas resulting in line emission. Approximately 99 per cent of the power output of galaxies at submill imetre and far-IR wavelengths is produced by continuum thermal emission and the rest comes from line emission. (Bla in et al . [3]). CHAPTER 1. INTRODUCTION 3 1.2.2 Inhabitants of the High-Redshift Universe In the 1980's, a ~ U y sensitive all-sky survey at 12, 25, 60 and 100 / im by the Infrared Astronomical Satellite (IRAS), revealed a populat ion of previously unseen optically-faint galaxies out to a redshift of ~ 0.3 and that the amount dust reprocessing increases wi th star production and cannot be measured solely wi th optical or U V data (see Sanders & Mi rabe l [52]). After the dominant C M B emission is accounted for, about half of the remaining energy budget of the extragalactic background is contained in the Far-Infrared Backgound ( F I R B ) (see F ig . 1.1). The Infrared Space Observatory (ISO), sensitive only to the low redshift region, resolved about 10 per cent of the F I R B into discrete galaxies ly ing at redshifts less than about 1 (Dole et al . [20]). The discovery of a submill imetre extragalactic background by COBE (COsmic Back-ground Explorer) was an amazing find and requires the existence of far-IR emitt ing galaxies in the early throes of their evolution or el l ipt ical galaxies burst ing wi th star formation. Th is submil l imetre background implies a strong cosmological evolution, or that early type galaxies in the distant Universe must be very different from galaxies in the local Universe which we know to be predominantly composed of relatively old stars. This progression greatly affects the star formation history, as a large part of the stars being formed are most l ikely hidden in dust-enshrouded galaxies, eluding opt ical surveys. What are the objects responsible for producing this background emission and what is their role in galaxy formation? The S C U B A camera on the J C M T is able to detect rest-frame far-IR radiat ion from high-redshift galaxies and has revealed a galaxy populat ion w i th star-formation rates of l O O M Q y r " 1 , similar to that expected from dusty protogalaxies (Steidel [60]). There are relatively few secure identifications and redshifts unfortunately, but most S C U B A galaxies are thought to be at redshifts greater than 1. 1.2.3 Submillimetre Emission as a High-Redshift Probe Submil l imetre emission may be used as a probe of the high-redshift Universe since i t can penetrate dust, and because of another powerful effect called the negative K-correction CHAPTER 1. INTRODUCTION 4 energy (keV) 10 3 1 lO"3 lO"6 10"9 10- 1 2 wavelength (mm) Figure 1 . 1 : Th is figure depicts the energetics of the extragalactic background which is comprised of six main components: the Cosmic Microwave Background ( C M B ) , the Cosmic Infrared Background (CIB) or Far-Infrared Background (F IB) , the Near-Infrared (NIR) , the Opt i ca l -UV, the X - R a y Background ( X R B ) and the Gamma-Ray Background ( G R B ) . Use of this figure courtesy of Scott et al . [55]. CHAPTER 1. INTRODUCTION 5 which facilitates submill imetre observations of galaxies out to the distant reaches of the Universe. The negative K-correction is strongest for wavelengths longer than about 250/xm, being the most pronounced at mil l imetre wavelengths as seen in F i g . 1.2, and causes a galaxy's flux density to remain about constant as it is moved to higher redshifts (between 1 < z < 10). Resolved submillimetre-wave observations exist only for the lowest redshift galaxies and it is usual practice to use the results of these observations as templates for galaxies of the more distant Universe (which may or may not be a reasonable assumption given the lack of understanding of high-redshift galaxies and of the feedback processes involved in star formation). IRAS uncovered a distinct class of extremely luminous objects of 1 0 1 1 L o , dubbed Luminous Infrared Galaxies or L I R G s . Not-so-distant cousins to L I R G s , U l t ra Luminous Infrared Galaxies (UL IRGs) are probably more similar to galaxies in the high-redshift world wi th respect to their luminosities and physical properties. U L I R G s are a small class of special objects, believed to be interacting or merging pairs of spiral galaxies. They make up less than 0.1 per cent of galaxies in the local Universe but were more important in the past (Sanders [51]). They represent the most luminous of galaxies in the local Universe wi th bolometric luminosities in excess of 10 1 2 L© and the dust emission dominated peak of their spectral energy distr ibut ion (SED) lies in the F I R region of the spectrum. This peak gets redshifted into the submil l imetre regime, making submil l imetre observations an ideal way to see the highest redshift dusty U L I R G s . U L I R G s have irregular morphologies and are bursting wi th star-formation, most l ikely indicat ing recent galaxy mergers. U L I R G s predominantly display el l ipt ical galaxy surface brightness profiles (e.g. Zheng et al . [69]), hint ing that U L I R G formation could be just the way to form ell ipt ical galaxies at high redshifts. Submill imetre-selected galaxies are an important piece of the galaxy formation and evolution puzzle but are difficult to study, as they are usually very faint in the com-plementary wavelength regimes. L imi ted progress has been made in determining the relationship of the submill imetre population to other high-redshift galaxy populations. Of al l the classes of galaxies, Extremely Red Objects (EROs) have emerged to be one CHAPTER 1. INTRODUCTION 6 1 0 0 0 . 0 0 1 0 0 . 0 0 b £ c Q > 0) OT o 1 0 . 0 0 b 1.00 0 . 10 0.01 1 1 1 1 1 1 1 1 1 1 1 | 1 1 1 1 1 1 1 1--\ \ r \ — : \ \ -\^ \ -" N. \ - \ --: \ \ \ \ ^ \ - \ ' . ^ S s s . \ \ \ ~ -\ \ >v : \ \ \ : \ \ \ \ \ \ \ \ \ — \ \ \ : \ \ \ : \ \ \ ; \ \ \ \ \ \ \ \ \ \ \ l \ \ i i i i i i i l i . i \ i i i i i M i l \ 0.1 1.0 10.0 R e d s h i f t 1 0 0 . 0 Figure 1.2: Th is figure depicts the effect of negative K-correction on the predicted flux density of Arp220, a dusty U L I R G galaxy, as a function of redshift. Note the powerful K-correction at submill imetre wavelengths of 850 / im (solid line) and 450/ im (dashed line), where the flux density is almost independent of redshift past z « 1. The sudden drop in flux at high redshift occurs because we begin to sample the S E D on the other side of the thermal F I R peak. The i f -band does not benefit from this effect and is shown for comparison (dot-dashed line). CHAPTER 1. INTRODUCTION 7 of the most promising and important tools for studying the history of galaxy formation and star formation in the universe (Dickinson et a l . [18], Yaha ta et a l . [68]), since dusty objects at high redshifts are expected to be strongly reddened. 1.2.4 EROS The recent development of large near-IR imaging cameras led to the discovery of a faint class of extremely red objects, or E R O s . The E R O class contains objects w i th very red opt ical to near-IR colours, typical ly R — K > 5 to 6 or J — i f > 4. Suggested explanations for the very red spectra include the possibil i ty that they might be dusty galaxies at z < 1, extraordinari ly high-redshift 'J-dropout ' galaxies, or that they are old el l ipt ical galaxies containing low-mass cool stars seen at z « 1 or 2, passively evolving after their in i t ia l burst of star formation, and that line emission might be involved. The E R O class may very well contain several types of object. Dickinson et a l . [18]. Th is selection criterion was originally designed to pick out old (> 1 Gyr ) passively evolving el l ipt ical galaxies at redshifts between 1 and 2. Bu t a smal l fraction of al l E R O s selected this way seem to be very dusty ell ipticals undergoing their in i t ia l burst of star formation. It is important to find the dominant populat ion amongst the E R O s since the high-redshift old ell ipticals would indicate a "monoli thic collapse" scenario of galaxy formation whereas the "hierarchical" scenario would be better supported by evidence of many dusty starbursts l iv ing at intermediate redshifts. Many of the most luminous submill imetre sources discovered have been identified as convincing counterparts to E R O s . For example, C U D S S 1 4 A (Gear et a l . [32]), S M M J00266+1708 (Frayer et al . [31]), W - M M D 1 1 (Chapman et al . [12]), S M M J09429+4658, and S M M J04431+0210 (Smail et al . [58]) have al l been identified wi th E R O s . These submil l imetre sources have fluxes in the range of 7 to 20 mJy. Because of the high frequency wi th which very bright submill imetre sources are identified wi th E R O s , Smai l et a l . [58] suggest that submil l imetre sources comprise the major i ty of the reddest E R O s . Totani et a l . [63] discovered four objects in the Subaru Deep F ie ld that were even more CHAPTER 1. INTRODUCTION 8 red than the reddest known E R O s . So the U B C submill imetre group collaborated wi th the Subaru team and applied for t ime to search for these rare objects at submil l imetre wavelengths, to test the conjecture that very red objects are submil l imetre-bright. The search for dusty starbursts at high redshifts is a key step to unraveling the mystery of galaxy formation at early epochs. 1.2.5 Hyper-Extremely Red Objects Among the E R O s which are faintest in A' -band, K > 22, an increasing fraction are extraordinari ly red, even compared to other E R O s , with J — K > 3 or 4. A ' -band sources which are very faint or undetected in optical data hold a strong possibi l i ty of being S C U B A bright. F i g . 1.3 is the near-IR colour-magnitude diagram for the C F R S (Canada-France Redshift Survey) and demonstrates that U L I R G s make up a large fraction of the reddest objects in this field. Bu t does a redder colour necessarily guarantee that an object wi l l be bright in the submill imetre? Totani et al . [63] cleverly dubbed this very distinct populat ion of galaxies Hyper-Extremely Red Objects, or H E R O s , and in detailed modeling find that they are much too red to be passively evolving el l ipt ical galaxies at z < 2. These objects are most l ikely to be either very dusty el l ipt ical galaxies which formed at redshift 2 « 4 - 7 and are st i l l undergoing rapid star formation when seen at redshift z > 2, or they are clean, Lyman break galaxies seen at extraordinary redshifts of z > 10. Red optical colours sometimes suggest that dust extinction is present, and this is precisely what submil l imetre instruments can see. Knowing the submill imetre flux of these H E R O s would distinguish between the galaxy formation models and could provide very valuable information for understanding when galaxies first formed and how long their in i t ia l starbursts lasted. CHAPTER 1. INTRODUCTION 9 Figure 1.3: The near-IR colour-magnitude diagram for the identifications the C F R S 14h and 3h fields (see Clements et al . , in preparation). The solid circles corre-spond to radio-detected objects, and the solid diamonds denote radio non-detections. The open circles show the best identif ication for the remaining objects. For comparison, the magnitudes and colours of al l the galaxies in the C F H T I R C U D S S + 1 4 image are also shown (plus signs). Not included in this plot are the possible L B G identifications (whose colors are in the grz filter system). The sol id, diagonal black line denotes the I detection l imi t of the C F R S . The dashed lines are tracks of I — KAB, KAB wi th redshift for sources 14.1, 14.3, 14.18, 3.10, and 3.15. The three solid lines show the predicted col-ors for the three U L I R G s studied by Trentham, Kormendy & Sanders [64]. These have been scaled to M K a b = 24.4 (approximately M* — 2). The tracks begin at z = 0.5 and are marked (small triangles) every 0.5 step in redshift. Th is plot il lustrates that U L I R G s comprise most of the reddest populations. Use of this plot and caption courtesy of Webb et a l . [65]. CHAPTER 1. INTRODUCTION 10 1.3 The Subaru Deep Field The Subaru Telescope, atop Mauna Kea , has been used to produce deep images in J-band A = 1.16 — 1.32 pm and in i f ' - b a n d 1 A = 1.96 — 2.30 / im of a 'blank' 2 arcmin x 2 arcmin field near the north Galact ic pole (Maihara et a l . [46]) (see F ig . 1.4). The Subaru Deep F ie ld (SDF) is a blank survey region and was chosen according to the following criteria: 1. to be in a different location than the Hubble Deep F ie ld ( H D F ) since the Universe may have different characteristics depending on the direction you look (the H D F also has a higher airmass than the S D F from Mauna Kea) , 2. to have a reference star nearby for future adaptive optics (AO) observations, 3. to have low Galact ic HI column density, no close-by stars (except for the A O guide star), galaxies or clusters of galaxies. Subaru's large aperture and 0.3 arcsec seeing have allowed extremely deep images which r ival any other images at these wavelengths (5 SH cu co o o cu •'—: X ! O CO d O XS co CP CP s-i faO CP o CP co o SH CT3 I—1 I—1 i — i OS 00 o eo o CN o o CM O o lO OS 00 o 00 © 00 o CO CO o o " o OS o O i o OS o OS o os o os o O i o OS o os o os o OS CP -a o o ©" co o o o o m I C CM CM > s-> S-l l - i (-I CS cc3 cc3 a N N N < < < CM o I CM co o o > S-l + 3 CP a o + 3 o X i PM X fo CM O o CN co" >> S-l CP a o + 3 O XS cu 7, 2002 , 2003 2003 , 2003 2003 o T—1 os" o in" X SH cc3 fo fo fo SH SH SH CP "cp CP a a a o o X5 OH O o -d OH o o X ! fo fo P CO fo Q CO CM fo P GO CO fo Q GO fo P GO SH CP XS O CO SH O Xi 00 cd a 13 CP a CP o SH o co i d o "(3 o + 3 CP X -u O CO SH M H CP SH CP a PH .2 a SH faO CP .a o a o CO cc3 CP X3 H SH CP CO X O SH CP X3 O cc3 CO Cl T3 O Cl CP - c | bO a • 1-4 T3 d o .a CP ^3 d o •fH += ce Cl CP • - H SH O CP X ! o CP OH co CP SH X ! CP -u cc3 H J O SH O H-3 O "73 CO CP XI faO Cl "8, OH O XI o 'rt X3 d a • -H SI ct3 O +s CO SH ^ H CP SH SI d cc3 d o • i—i £ SH 0 ) CO X O d CP CO SH d o o CP XI -u fcuO d • - H SH d -o cc? SH SH Cfi CHAPTER 2. THE INSTRUMENT AND DATA 24 2.2.1 Added Noise in the 2003 Data In June 2003, Co l in Borys, a former U B C P h D . student, discovered a large power spike at « 1/16 Hz coincident wi th the frequency of the secondary chopper in de-nodded, flatfielded, and extinction corrected 64-point jiggle map data. The telescope perfoms 1.024 second integrations on each of the 16-point jiggle pattern positions and nods to one side before repeating this procedure and nodding to the opposite side. Th is pattern is repeated two more times to complete a 64-point jiggle pat tern 3 . Thus, the telescope is moving by a large amount every 16 seconds (1 nod every 16 points x 1.024s/point w 16nods/s) . We hypothesize that there may be some microphonic noise pickup when the telescope moves during the nod, corresponding to the noise appearing stronger in the bolometers at a frequency of « 1/16 Hz. 300 jiggle map observations between 1998 and 2003 were analyzed by Co l in Borys and Remo Ti lanus of the Joint Astronomy Centre ( J A C ) , and only observations taken between December 2002 and March 2003 seem to be affected by this anomaly (see F i g . 2.3). The S C U B A fridge cycle at the end of March is l ikely responsible for f ixing the problem (Borys 2003, private communication). Upon analysis of the fourier spectrum of our data, we discovered that this power spike occurs on the long-wave array in about two-thirds of the bolometers, while the remaining th i rd had nominal signals, similar to what Borys found. The effect is also present in the short-wave array, but to a lesser extent. On ly about 10 per cent of the bolometers show this noise spike. Usual ly this effect would be automatically removed after subtracting off the sky signal if al l of the bolometers had this same spurious signal present in their frequency spectra. Bu t since only a fraction of the bolometers have this signal, the effect of removing the mean sky background actually imposes this signal on the non-affected bolometers, cor-rupt ing them as well. The concern is that the presence of a periodic noise fluctuation on a similar timescale to the jiggle pattern may be affecting the data, thereby imposing 3When SURF de-nods the data, each data point it reports is actually the sum of two 1.024 second measurements taken 16 seconds apart. CHAPTER 2. THE INSTRUMENT AND DATA 25 0.10 0.15 Frequency (Hz) 0.20 0.25 > 1.2x1 0 " 4 ' 1 . 0 x 1 0 - 4 -8 . 0 X 1 0 - 5 o 6 . 0 x 1 0 - 5 4 . 0 x 1 0 ~ 5 2 . 0 X 1 0 - 5 0.10 0.15 Frequency (Hz) 0.20 0.25 Figure 2.3: The top plot is the F F T of the timestream for bolometer G15 (or number 12) of the long-wave array, from 64-point jiggle map data taken of the S D F in May 2002. The bottom plot shows the F F T of the t imestream for the same bolometer from 64-point jiggle map data taken of the same field in March 2003. Note the power contained in the large noise spike at about 1/16 Hz in the lowermost plot. CHAPTER 2. THE INSTRUMENT AND DATA 26 unwanted structure on the map data to some degree. (Borys 2003, private communica-tion). The data were examined in a variety of ways. F i rs t , we tr ied three different methods of separating the data: averaging al l of the 2003 data together, separating the data according to chop throw and posit ion angle and taking the average, and concatenating al l of the t imestream data for one night. Upon examination of the fourier spectrum of each of these new averaged data-sets we observed that the signal persisted. In the 2003 averaged data-set, the signal decreased by about an order of magnitude, but so d id the overall noise level, resulting in a constant signal-to-noise ratio for the spike. See F ig . 2.4 for an example. We first investigate if and how this addit ional noise affects the data in the t ime domain (i.e. the bolometer timestream). We trace how the standard deviation (a) of the signal varies wi th time. One jiggle observation consists of 10 integrations of a 64-point jiggle pattern, adding up to 640 timestream elements for each bolometer. For the central bolometer H7 (number 19), we calculate a noise of o = 0.410 V over the course of one ten-integration observation. For a single bolometer, we calculate the mean and error of each 64-point jiggle map integration and see no trend for the error (noise) either increasing or decreasing over the set of ten integrations. We therefore conclude that the noise is randomly fluctuating, as it does not vary significantly in the bolometer t imestream over the course of an observation. Next, we investigate if and how the noise spike affects the data in the spatial domain (i.e. the map structure). Obviously, one might worry that unwanted structure is being imposed on the map. We trace how the signal varies across the 64-point jiggle pattern array. A bolometer wi l l hit 64 different positions in the sky dur ing the course of one 64-point jiggle integration. Since each observation is made up of ten integrations, each posit ion wi l l be hit 10 times during the course of an observation. We add up the contr ibu-t ion of each integration at each jiggle posit ion, calculate the mean, a and the error in cr, and look for a trend in the noise. We do not see any trends of the noise either increasing or decreasing. For an error of a — 0.410 V (found above), we determine the typical error CHAPTER 2. THE INSTRUMENT AND DATA 27 > 1 . O X 1 0 - 4 5 . O X 1 0 - 5 0 . 00 0 . 05 0 .10 0 . 15 Frequency ( H z ) 0 . 20 0 . 2 5 4 X 1 0 " 5 3 x l 0 " 5 E -> o 2 x 1 0 ' X 1 0 - 5 0 . 0 0 0 .05 0 . 10 0..15 Frequency ( H z ) 0 . 20 0 . 2 5 Figure 2.4: The top plot is the F F T of the timestream for bolometer G15 (or number 12) of the long-wave array, from 64-point jiggle map data taken of the S D F in March 2003. The bottom plot shows the F F T of the t imestream for the same bolometer from the average of al l the 2003 64-point jiggle map data taken of the same field. Note the factor of « 10 decrease in the power of the noise spike at about 1/16 Hz and in the overall noise level in the lowermost plot from the one above. CHAPTER 2. THE INSTRUMENT AND DATA 28 Jiggle map position Figure 2.5: Th is plot shows the standard deviation of ten points (ten integrations) for each of the 64 different jiggle positions for the central bolometer, H7. The solid line represents the mean standard deviation value. The dashed lines represent the la error in the standard deviation (~ ±l/y/2N). CHAPTER 2. THE INSTRUMENT AND DATA 29 B o l o m e t e r 19 B o l o m e t e r 1 3 0 2 4 6 8 10 12 X grid B o l o m e t e r 2 0 0 2 4 6 8 10 12 X grid 0 2 4 6 8 10 12 X grid B o l o m e t e r 2 6 0 12 Figure 2.6: The 64-pt jiggle pattern is observed in 4 sections of 16 points (solid lines). For 4 different bolometers on the array, we plot the standard deviation of ten measurements (ten integrations) at each posit ion in the 64-point jiggle pattern using contours to specify the noise level. Contours begin at a = 0.0 and increase in steps of 0.05 up to a maximum of a = 1.0. It is evident that the noise varies randomly across the array for any given bolometer. CHAPTER 2. THE INSTRUMENT AND DATA 30 in a to be about 0.0920 V . Figs. 2.5 and 2.6 il lustrate that the noise appears to vary randomly over the course of 64-point jiggle pattern for a given bolometer. We conclude that the noise is randomly fluctuating, does not depend on posit ion, and therefore does not impose any structure on the map. Co l in Borys postulates that there was some k ind of vibration-induced microphonic pickup when the telescope suddenly moves at the end of the 16 point jiggle pattern to nod, making the noise on the array stonger at a frequency of 1/16 Hz. It may also be due to a noisy atmosphere or to a temperature instabil i ty which makes the bolometer gains fluctuate in a random way (which would have been fixed by the S C U B A fridge cycle). In any case, we conclude that the large power spike at 1/16 Hz is just introducing more random noise l / \ / 2 ) into the data and not imposing any structure on our maps. 2.3 Data Reduction 2.3.1 Preliminary Reduction with SURF The S U R F ( S C U B A User Reduction Faci l i ty; Jenness & Lightfoot [40]) reduction pack-ages were used for nod compensation ( r e d u c e _ s w i t c h ) , flat-fielding ( f l a t f i e l d ) and extinction correction ( e x t i n c t i o n ) of the data. The r e d u c e . s w i t c h command subtracts the off-position from the on-position and splits the raw data array into separate com-ponents. Next, the f l a t f i e l d command "flatfields" or corrects the gain of the array by mult ip ly ing each bolometer by the volume flatfield value relative to a reference pixel, usually the central pixel of the array. The e x t i n c t i o n command corrects the data for source airmass (elevation) and sky opacity, r. A great deal of care was taken to eliminate any spikes in the C S O r readings using the polynomial fits provided by the J A C , and a constant r value was used over the duration of each observation. The e x t i n c t i o n command has the capabil i ty to interpolate between two specified r values (before and after the observation) but the r did not vary significantly during the course of any one observation so this feature was not used. Th is last command also splits up the data into CHAPTER 2. THE INSTRUMENT AND DATA 31 the two different wavelength components, namely 850 and 450 pm. The work produced in this thesis made use of locally developed code wri t ten in C + + by Co l in Borys (Borys [6]) for his U B C doctoral dissertation. We converted the observa-t ion files into the more versatile " F I T S " format using the program SCUBA2FITS, making it readable for the next series of locally developed programs. The files al l now contain a header, and an array of three vectors containing: data timesteps, bolometer number, and the data. The data vector itself contains the signal in Volts and three R A / D e c offset positions corresponding to the three positions on the sky that the bolometer looked at while taking the difference measurement. 2.3.2 Making the Maps A program called B S _ S C A N _ R E D U C E was then used to 4 a despike and rebin a l l data types (photometry and jiggle maps) into a single map, producing three useful data products (similar to those that can be produced by S U R F ) : a signal map, a noise map (the stan-dard deviation or a), and a signal-to-noise ratio (SNR) map. B a d pixels were essentially el iminated by weighting each pixel by its t imestream inverse variance relative to the cen-tral pixel. The data were despiked by means of calculating the root-mean-square (rms) for each bolometer and rejecting any data 4 times greater than this value. Sky subtrac-t ion was performed by calculating the mean sky at each timestep using the median of al l of the bolometers save the noisiest bolometers and subtracting this result from the data. A bolometer was deemed too noisy to include in the sky subtraction by calculat ing its variance and cutt ing out the bolometers that were above a certain threshold (usually a > 0.005). The flux from the negative off-beams were then "folded i n " to increase the overall sensitivity of the 850 pm map. For each sample in the bolometer t imestream, we add -0.5 times the measured flux to the position of the off-beam. Since a bolometer spends only half as much time in each of the off-positions (reference positions that do not lie on the target) as in the on-position during the course of a single observation, we assign the CHAPTER 2. THE INSTRUMENT AND DATA 32 on-position a weight of it; = 1 and the off-positions weights of w — —0.5 each 4 . Fol lowing the treatment by Saj ina [50], the folding can be represented by: = and (2-4) where /x is the mean, a is the standard deviation, and x is the voltage measured. It should be noted that the beam is assumed to be Gaussian in shape which is usually the case at 850 fim but not at 450 fim, where the beam is often more severely distorted and non-Gaussian in shape. Therefore we do not fold in the off-beams to produce the 450 /zm map. In this way, folding in the off-beams wi l l improve the sensitivity of the map by about a factor of: 2 (2-5) V 0 . 5 2 + 1.0 + 0.5 2 V 3 The "-f" option of the B S _ S C A N _ R E B I N program was used to produce this result. Th is step is essentially equivalent to a single iteration in the iterative deconvolution technique described in Borys [6]. Th is method is only val id if sources are far enough apart and are not being chopped onto one another. We have taken great care to use a chopping strategy that avoids the H E R O s being chopped onto each other by selecting appropriate chop throws in azimuthal coordinates, allowing the chop posit ion to rotate on the sky during the observation (see F ig . 2.7). This procedure of folding in the off-beams was used successfully by Borys [6] to produce the Hubble Deep F ie ld Supermap and by Eales et al. [26] to produce the 14 Hour F ie ld , for example. The 850 /tm data were then regridded onto a rectangular grid of 3 arcsec pixels and smoothed using the program B S J 3 M O O T H F I T to a final F W H M of 14.7 arcsec (the size 4This can also be seen in that for an observation, both values vary as N"1 since /t = (WotKES4 )^-CHAPTER 2. THE INSTRUMENT AND DATA 33 SDF- 1 SDF-2 • • SDF-4 SDF-3 ^^ ^^ ^^ ^ JlP^ i^lllk • 9 £r^^k> H 0 0 • Ik ^^mw M Figure 2.7: We plot the chopping strategies for photometry observations of the four H E R O s . For each plot, the shaded source is the H E R O that the central bolometer is pointed at during photometry, and the other smal l circles (scaled to the represent the beam-size at 850 fxm) represent the other H E R O s in the field. We also plot the path that the bolometer wi l l chop on the sky as the sky rotates during the night. We label the chop throws in arcseconds on each path. CHAPTER 2. THE INSTRUMENT AND DATA 34 of the J C M T beam at 850 /mi) and the 450 /zm data were smoothed to a F W H M of 7.5 (the beamsize at 450/ im). B y rebinning the 850 / im map using 3 arcsec pixels, the map wi l l have pixels which are correlated since the beam is 4.93 of these new pixels wide. Another reason for smoothing is that the 3 arcsecond pixel grid gives an art i f icial ly smal l resolution which is not the true resolution of the image (given by the beam size) and wi l l have the effect of making the map appear more noisy than it t ruly is. Neglecting the undersampled and hence noisier edges of the map, we get a mean signal of —0.047 (±0.03) m J y from the central 3425 pixels of the 850 am map. Th is is in agreement wi th the expected map average of 0 m J y for differential measurements. A s a check, we examine the rms of the map without the off-beams folded in and compare it to the rms of the map wi th the off-beams folded in . The rms before off-beam folding is 2.26 m J y and the rms post off-beam folding is 1.80 mJy , which is entirely consistent wi th an estimated resulting rms of 2.26 x = 1.85. Similarly, we measure the mean signal of the central 1662 pixels of the 450 /zm map to be 0.32 (±0.24) mJy . We measure an rms for the 450 / im map of 9.7mJy. 2.3.3 Calibration Because the map is output in Volts and we wish to measure flux in an astronomically-meaningful units (i.e. mJy) we must calibrate the data-set by observing objects wi th known submil l imetre fluxes. For consistency, the calibrators were reduced in the same way as the S D F map and the mean of the peak pixel values of the calibration sources were used to convert the measurements into units of flux. See table 2.2 for the list of average F C F s used. We adopt the average value of al l of the flux conversion factors (FCFs) obtained over one night of observing and use this value as the F C F for every observation on the same night. The average values and standard deviations of al l of the F C F s we obtained are 224(±27) m J y at 850 / im and 339(±120) mJy at 450 / im. These flux conversion factors are consistent wi th the standard gains (within the uncertainties) at 450 and 850 / im of CHAPTER 2. THE INSTRUMENT AND DATA 35 308(±109) and 219(±21) mJy , respectively, from the J C M T website. For photometry mode, the standard gains are slightly different: 384(±82) and 197(±13) m J y / V o l t at 450 and 850 fj.m. Rms uncertainties in the calibrations are approximately 10 per cent for the 850 fim j iggle map and photometry data and 30 per cent for the 450 /zm jiggle map data-set. These uncertainties are mainly caused by the variabi l i ty of the atmospheric transmission and changes in the dish surface accuracy induced by temperature fluctuations (Jenness et al . [41]). 2.3.4 Flux Measurements S C U B A operates at wavelengths appropriate,to detect thermal emission from dust wi th temperatures of 3-30K. A t higher temperatures (excluding high-redshift objects), dust radiates mostly in the far-IR, so high-redshift far-IR emitters w i l l have this emission redshifted into the submill imetre (Holland et al . [36]). Since the angular extent of any of these sources is l ikely to be much smaller than the 3 arcsec pixel size (for example a l k p c source at a distance of z = 3 would appear to be 0.14 arcsec in the sky), we simply measure the flux of each source in the pixel corresponding to the near-IR source coordinates on the beam-convolved map. The best estimate of the noise associated wi th each pixel is given in the noise map. We ignore the cal ibrat ion uncertainty, which is unimportant wi th these low signal-to-noise ratio data. The flux measurements include contributions from al l jiggle map and photometry data which have al l been inverse-variance weighted in the co-addition process. The final maps are displayed in Figs. 2.8 and 2.9. The results for the H E R O s are summarized in Table 2.3. It is interesting to note that the 3 reddest objects (J — K' > 3) a l l have positive submil l imetre flux greater than the least red H E R O , S D F 3 . CHAPTER 2. THE INSTRUMENT AND DATA 36 Date Observing Number of FCF450 Number of FCF850 Mode Calibrators (mJy /Vo l t ) Cal ibrators ( m J y / V o May 27, 2001 Jiggle M a p 5 325 8 235 May 16, 2002 Photometry - - 0 *197 May 17, 2002 Photometry - - 0 *197 Feb 6, 2003 Photometry 2 324 2 201 Feb 9, 2003 Photometry 0 *384 0 *197 Feb 10, 2003 Photometry 2 404 2 233 M a r 2, 2003 Jiggle M a p 2 292 2 200 M a r 3, 2003 Jiggle M a p 3 402 3 232 Mar 4, 2003 Jiggle M a p 1 318 1 219 M a r 5, 2003 Photometry 3 417 3 230 M a r 8, 2003 Jiggle M a p 4 260 5 215 Table 2.2: For each night of data, a different F l ux Conversion Factor ( F C F ) , or gain, is derived from the average value of calibration observations taken near to the observations. In cases where the cal ibration source was too extended to give a useful measurement or where too few cal ibrat ion observations were made (denoted by an asterisk), we just use the "standard" gain value from the J A C website. The 450 fxm data from May 2002 were not used because the weather was very poor (see Table 2.1). The standard gains are 308(±109) and 219(±21) m J y / V o l t at 450 and 850 pm respectively in mapping-mode and 384(±82) and 197(±13) m J y / V o l t in photometry mode (cf. S C U B A cal ibra-t ion webpage). CHAPTER 2. THE INSTRUMENT AND DATA 37 Figure 2.8: The Subaru Deep F ie ld 850 pm signal-to-noise ratio (SNR) map. The bright-est points have a S N R of w 3 — 4 and the lowest points have a S N R of w —3. The circles are roughly 14.7 arcsec in diameter, the size of the S C U B A 850 / im F W H M beam size. The circles with crosses inside mark the location of the four H E R O s , and the other circle marks a > 3 a detection. CHAPTER 2. THE INSTRUMENT AND DATA 38 Figure 2.9: The Subaru Deep F ie ld 450 /zm signal-to-noise ratio (SNR) map. The bright-est points have a S N R of w 3 — 4 and the lowest points have a S N R of « —3. The circles are roughly 7.5 arcsec in diameter, the size of the S C U B A 450 /zm F W H M beam size. The circles wi th crosses inside mark the location of the four H E R O s , and the other two circles mark > 3 a detections. CHAPTER 2. THE INSTRUMENT AND DATA 39 ID S450 5 g 5 o Upper L imi ts (mJy) (mJy) g 8 5 0 S D F 1 i:47 ( ± 5.94) 2.15 ( ± 0.92) < 3.46 S D F 2 -7 .84 ( ± 6.36) 1.01 ( ± 0.94) < 2.64 S D F 3 6.10 ( ± 6.48) - 0 . 22 ( ± 0.88) < 1.63 S D F 4 1.73 ( ± 5.76) 1.85 ( ± 0.98) < 3.45 Mean 0.42 (±3.06) 1.15 (±0.46) < 1.76 Table 2.3: Measured flux densities wi th errors and predicted redshift upper l imits for the four S D F H E R O sources. The fourth column lists the 95 per cent Bayesian upper l imits (in mJy) at 850/ im. The bottom row lists the error-weighted mean of the 850 / im fluxes, and the 95 per cent Bayesian upper confidence l imi t to this mean flux. CHAPTER 3. HEROS 40 C H A P T E R 3 HEROS In the following chapter, we estimate the cirrus contribution to our 850 / /m map. We also briefly discuss the relation between H E R O colour and submil l imetre flux. We then create two template S E D s , a starburst galaxy and a normal galaxy, and use them to facil itate estimates of the redshifts of the H E R O s . We compare and evaluate the models to determine how realistic these templates are at describing the H E R O s . 3.1 Cirrus Contribution We need to estimate the amount of cirrus contribution to our map in case it is providing a large fraction of the flux, which would mean we are just measuring galactic emission and not extragalactic sources at al l . We use a port ion of the Schlegel, Finkbeiner & Davis [53] ful l sky 100 am map, a reprocessed composite image of the COBE/DIRBE (Diffuse InfraRed Background Experiment) and IRAS/ISSA (IRAS Sky Survey At las) maps. These authors carefully removed zodiacal foreground emission, artefacts from the IRAS scan pattern, and confirmed point sources, resulting in a map wi th DIRBE-qu&lity cal ibrat ion and IRAS resolution. From this map we obtain 0 .9MJy /s te rad ian in the direction of the S D F point ing centre coordinates. Franceschini [30] presents a plot of the S E D of the diffuse dust due to cirrus for a nearby prototype starburst galaxy, M82. From this plot, we take the ratio of the cirrus contr ibution at 100 /mi to the contribution at 850 /mi and scale the 100 /mi cirrus measurement by this ratio in order to estimate the 850 /mi cirrus contr ibution. We obtain a cirrus flux contribution of 5 x 10~ 4 MJy / s te rad at 850 /mi . Given that there are about 60,000 14.7 arcsecond S C U B A beams in 1 square degree of sky, the flux contribution becomes 2 x 10~ 3 m J y per S C U B A beam at 850 /mi . Th is is much less than CHAPTER 3. HEROS 41 the confusion l imit of our map l m J y ) and is extremely negligible. A t 450 / im we obtain a cirrus flux contribution of 3 x 1 0 - 3 MJy / s te rad . The 450 /zm cirrus contr ibution is 4 x 1 0 - 3 m J y due to the smaller beam size of 7.5 arcseconds F W H M . We can therefore safely ignore the neglible contribution of cirrus in our measurements. 3.2 HERO IR Colours and 850 jam Fluxes S C U B A galaxies are often found to be associated wi th E R O s but we have shown that hyper -EROs are not always necessarily associated wi th S C U B A - b r i g h t galaxies. We do not find that the H E R O s are SCUBA-b r i gh t , and we also do not find any evidence for their 850 / im submil l imetre flux increasing with J — K' colour. The latter finding is i l lustrated in F ig . 3.1, where we plot 850 /tm against J — K' colour and get a slope of 0.81 ± 0.89 and a x2 value of 3.3 for the best-fit line through the data. Given that extremely red objects associated wi th S C U B A sources tend to be dusty starburst galaxies, we might expect an overlap between the H E R O s and the U L I R G populat ion at high redshifts. We use the S E D of Arp220, constrained to fit the near-IR and submil l imetre observations, to investigate if this is a reasonable assumption. We compare the results against a normal galaxy template as well. 3.3 Using Arp220 as a Template We created a template S E D using public photometric data available for A r p 220 from the N A S A / I P A C Extragalact ic Database ( N E D ) . A r p 220 is the most luminous object in the nearby universe (at z = 0.018) and is the most well-studied example of a U L I R G . It is necessary to use this low-redshift analogue of a S C U B A source since higher redshift sources have more sparsely sampled SEDs , due to the difficulty of observing them in mult iple wavelength regions. We use the Lagache, Dole & Puget [45] U V to radio S E D models of a typical starburst galaxy to fill in the gaps in the data for Arp220. The Lagache, Dole & Puget [45] CHAPTER 3. HEROS 42 E =4. o m oo .j i i i_ J - K Figure 3.1: Is there a colour-flux correlation for the H E R O s ? We plot 850 / /m flux versus J — K' colour. The best-fit line through these data has a slope of 0.81 ±0 .89 and a goodness-of-fit x2 value of 3.3, suggesting that there is no correlation between the measured 850 fim submill imetre flux and IR colour in the case of the S D F H E R O s . CHAPTER 3. HEROS 43 templates evolve (become hotter) with luminosity over a range of luminosity levels from about L— 10 9 — 1 O 1 3 L 0 . These templates cover the range of bolometric luminosities of galaxies that comprise the Cosmic Infrared Background up to z < 2. For example, F I R B A C K (Far InfraRed B A C K g r o u n d ) sources comprise a cold nearby populat ion wi th L= 10 9 — l O n L 0 and a more distant (z < 1.2) cold (or warm and highly luminous) populat ion wi th L= 1 0 1 2 L Q . We employ their model number 30 (albeit heavily modified in terms of its dust content in the IR portion of the spectrum) in order to draw a smooth line through Arp220 (see F ig . 3.2). F ig . 3.3 shows that for the i f ' - band and at 850 /un, the flux does not vary by more than about 10 per cent from the Lagache, Dole & Puget [45] template number 41, the most luminous template available. We have constructed a three parameter model based on the S E D of Arp220, con-strained only by the K' magnitude and the J — K' colour. We effectively redshift, dust extinct, and brighten the S E D by known amounts and then read off the flux at 850 /on. After choosing a redshift, we find the J — K' colour and add dust to fix the colour to the observed value for each H E R O . The model we have employed fits known sources even though it is very simple (one-parameter), and this gives it an advantage over more complicated multi-parameter models. We add dust by mult ip ly ing our Arp220 template by an extinction function which varies wi th wavelength. We determine such an extinction law by constructing a power-law fit to the relative extinction values in the Landolt V , R, I and U K I R T J, H, K and V bandpasses versus effective wavelength from Schlegel, Finkbeiner & Davis [53] (see Figs. 3.4 and 3.5). They employ a 'diffuse I S M ' mean value of Rv = 3.1 for the extinction laws of Cardel l i , C layton & Math is [10] and O'Donnel l [47]. We note that the difference between the extinction curves of the M i l k y Way, the Magel lanic Clouds and starburst galaxies is almost negligible at wavelengths longer than w 0.26 /tm (see Calzet t i , K inney & Storchi-Bergmann [9], Cardel l i , C layton & Math is [10]). Therefore, using a different extinct ion law should not affect our results since only galaxies at redshifts of z < 3 are considered, corresponding to a rest wavelength of 0.31 u-m being redshifted into the J-band. CHAPTER 3. HEROS 44 Figure 3.2: The model spectral energy distr ibution of a nearby U L I R G : the Lagache, Dole & Puget [45] starburst template (dashed line represents template number 30) fit at z — 0.018 to Arp220 photometric data (solid l ine), which are marked wi th the plus signs. The energy output is dominated by a modified blackbody and the mid- IR emission features seen here are attr ibuted to P A H molecules. The slope of the drop at « 1 /mi in the original S E D has been extrapolated to shorter wavelengths and fits an Arp220 ROSAT X- ray data point (not shown in the plot) wi th in its error bar. CHAPTER 3. HEROS 45 1 0 2 7 f m22r , , , , 1 0.1 1.0 1 0 . 0 1 0 0 . 0 1 0 0 0 . 0 1 0 0 0 0 . 0 W a v e l e n g t h [ ^ m ] Figure 3.3: The upper panel shows 3 different luminosity templates from Lagache, Dole & Puget [45]. The lowest line is the template used for Arp220 (number 30), the middle line is template number 35, and the highest line is the brightest template available (number 41). The lower panel gives plots of the ratio of the Arp220 template wi th the other two higher luminosity templates. Note how the template SEDs actually change intr insically in shape wi th luminosity. Vert ical lines indicate the port ion of the spectrum that we wi l l be sampling over a redshift range 0 < z < 3 at wavelengths of 850 fim and K'. For an observed band, the ratio of the AT'-band to 850 fj,m flux does not vary by more than 10 per cent across the models chosen. CHAPTER 3. HEROS 46 TABLE 6 RELATIVE EXTINCTION FOR SHLECTED BANDPASSBS Filter (A) A/A(V) A/E(B-V) Filter A*t A A/A(V) A/E(B-V) Landolt V 3372 1.664 5.434 Stromgren u 3502 1.602 5.231 Landolt B 4404 1.321 4315 4676 1.240 4.049 Landolt V , , 5428 1.015 3.315 4127 1.394 4.552 Landolt R 6509 0.819 2.673 Stromgren ft 4861 1.182 3.858 Landolt I 8090 0.594 1540 Stromgren y 5479 1.004 3.277 CTIO U 3683 1.521 4568 3546 1.579 5.155 CTIO B 4393 1.324 4.325 4925 1.161 3.793 CTIO V 5519 0.992 3.240 Sloan f 6335 0.843 2.751 CTIO R 6602 0.807 2.634 7799 0.639 2.086 CTIO/ 8046 0.601 1.962 9294 0.453 1.479 UKIRT J 12660 0.276 0.902 WFPC2 F300W 3047 1.791 5.849 UKIRT H 16732 0.176 0.576 WFPC2 F450W 4711 1.229 4.015 UKIRT K 22152 0.112 0.367 WFPC2 F555W 5498 0596 3.252 UKTRTE 38079 0.047 0.153 WFPC2 F606W 6042 0.885 2.889 5244 1.065 3.476 WFPC2 F702W 7068 0.746 1435 6707 0.793 Z590 WFPC2 F814W 8066 0.597 1.948 Gunn i 7985 0.610 1.991 DSS-II g 4814 1.197 3.907 Gunn z 9055 0.472 1.540 DSS-II r 6571 0.811 2.649 6993 0.755 2.467 DSS-II/ 8183 0.580 1.893 APMb, 4690 1.236 4.035 NOTB.—Magnitudes of extinction evaluated in different passbands using the Rr = 3.1 extinction laws of Cardelli et al. 1989 and ODonneB 1994. The final column normalizes the extinction to photoelectric measurements of E(B—V). Figure 3.4: We use the effective wavelengths (Aeff column) and relative extinction values (A/A(V) column) for the Landolt V, R, I and U K I R T J , H, K and 11 bandpasses in order to construct an extinction law. Th is table has been extracted from Schlegel, Finkbeiner & Davis [53], and they employ a 'diffuse I S M ' mean value of R v = 3.1 for the extinction laws of Cardel l i , C layton & Math is [10] and O'Donnel l [47]. CHAPTER 3. HEROS 47 Figure 3.5: We plot the magnitudes of extinction relative to the F-band (A/A (V)) for the Landolt (triangles) and U K I R T (squares) bandpasses (see F i g . 3.4 for the complete table). We fit 2 .51og( l / ( l + (1^3)2.6)0.2x) w i t h x = 1 0 ^ e 2 .51og of Eqn . 3.1) to the data points (solid line). The values in the function were chosen arbitrar i ly to give a reasonable fit to the data points: 1.23 places the line at the correct vertical posit ion relative to the points, 2.6 gives the correct line steepness to fit the data, and 0.2 places the graph in the correct horizontal location relative to the data points and also adjusts the line steepness to a lesser degree. CHAPTER 3. HEROS 48 We redden the original template by a factor of: 1 (3.1) (l + ( L|3)2.6)0.2a: where x is our free parameter which adjusts the reddening (i.e. J — K' colour) and A is wavelength in microns. The function was fabricated to fit the relative extinction values of the Landol t and U K I R T bandpasses (which are similar enough to the Subaru bandpasses), and the values in the function were chosen arbitrar i ly to give the best fit to the data points (see caption of F ig . 3.5 for details). Choosing template number 30, redshifted to z = 0.018 and using a value of x = 2.5 in the equation above gives a good fit to the Arp220 photometric data (see F ig . 3.2 solid line). We alter the luminosity of the template to fit the observed K' flux, as determined from the observed K' magnitude following Skinner [57]: where m is the magnitude and F0 is the zero-point flux for the photometric band. The zero-point for the i f ' -band is F 0 = 718.903 Jy and was extracted from the interpolation of the broad-band flux of Vega (Motohara 2002, private communication). where dL is the luminosity distance. Following Carro l l , Press & Turner [11], the lumi-nosity distance, 1.76 m J y (the 95 per cent upper l imit to the average measured flux of the H E R O s ) . So the flux level we measure for the H E R O s is consistent wi th the number of H E R O s in the S D F . 3.6.2 Could the H E R O s be Starburst Galaxies? We use the modified starburst template to create F i g . 3.8, a plot of expected 850 /tm flux as a function of redshift for a H E R O which is representative of the whole populat ion wi th the average colour and magnitude of the four-source sample. Since we cannot claim detection above 3 a of any of our faint sources, Bayesian 95 per cent upper confidence l imit flux estimates can be calculated for each source by integrating over the non-negative flux regions of a normalized Gaussian probabi l i ty function. These upper flux l imits were provided earlier in Table 3.2. Combin ing the results from the four sources allows us to obtain an average object flux for our sample of H E R O s . The error-weighted average flux density for the whole sample at 850 am is 1.15 ± 0.46 mJy , and has a 95 per cent upper l imi t of 1.76 mJy . The average flux density at 450 m J y is 0.42 ± 3.06 mJy . When we force our template S E D to match this same 850 fj,m flux, constrained by a faint K' magnitude and very red J — K' colour, we estimate a corresponding redshift of 1.6. Th is result is consistent with the median redshift estimates for S C U B A sources (see e.g. Dunlop [23]). Based on the assumption that hyper-extremely red objects are well-represented by an S E D that fits Arp220, then they are at a redshift less than ~ 1.6. F ig . 3.9 investigates how much we need to brighten the S E D to get the correct K' magnitude after adding in an amount of dust to the Arp220 S E D to get the correct CHAPTER 3. HEROS 55 10.00 F 1 1 1 1 I 1 1 1 1 — r — i — i — i — i — | — ' — i — i — ' — q 0.01 I i i , i | , i i , | i , i , i l i! . > , I 0.0 0.5 1.0 1.5 2.0 R e d s h i f t z Figure 3.8: A plot of expected 850 pm versus redshift for our average H E R O . Calculated data points are marked wi th plus signs and the asterisks represent where we read off a redshift, based on the mean and upper l imit fluxes. We expect this semilog plot to have a linear shape since we add dust to the SED by applying a power-law function to the template. We expect a range of models to occupy a "band" around this line, predicting different flux levels for a given redshift. CHAPTER 3. HEROS 5 6 1 0 0 . 0 0 F o 1 0 . 0 0 b-1.00 0 . 10 0.01 0.0 R e d s h i f t z Figure 3.9: A plot of the brightness in the A" -band (giving the correct K' magnitude) compared to the brightness of Arp220 (after adding in the right amount of extra dust extinction included to obtain the observed J — K' colour) versus redshift for our average H E R O . Based on the assumption that this model describes the H E R O s , then they are approximately as luminous as Arp220. CHAPTER 3. HEROS 57 J — K' colour. We notice that we need the K' flux to be dimmer than that of this 'dust-enriched' Arp220 unt i l z ~ 1.5, where the H E R O starts becoming intr insical ly brighter and more luminous than the dusty template. For z ~ 1.6, we need to mul t ip ly the S E D by a factor of ~ 1.5 to get the correct K' magnitude for the H E R O sample, which is reasonable (i.e. not brighter than a factor of a few) since we believe that more distant galaxies w i l l be more intr insically luminous (B la in et al . [3]). Even the brightest H E R O in this sample, S D F 1 , only needs to be brightened by a factor of ~ 3 at its upper redshift l imit of z ~ 1.65 F ig . 3.10 shows that the amount of dust reddening applied to the Arp220 S E D in order to match J — K' decreases wi th redshift. (Recall that our Arp220 template S E D is comprised of Lagache et al.'s (2002) starburst template number 30 wi th dust added v ia equation (3.1) wi th m = 0.5). After z ~ 1, the amount of required dust decreases very slowly but never reaches zero. The curves show an asymptotic behaviour, i l lustrat ing that some amount of dust extinction must always be applied to the S E D . A t z ~ 1.6, we add in only 0.4 magnitudes in K' and « 1.2 magnitudes in J. For the reddest H E R O , S D F 4 , we need to add « 1.8 magnitudes of dust in J a t the upper redshift l imit of z = 1.66. The H E R O s could be about the same luminosity as Arp220 and only slightly more dusty. Aga in , we stress that the model possesses an advantage over more complicated mult i -parameter models in that it fits known sources well, despite its single-parameter s im-plicity. In F i g . 3.11, we compare our S E D wi th that of Totani et a l . [63] to get an idea of the amount of uncertainties in our model and which way a more sophisticated model would affect the results. We note that we must add about 1.80 magnitudes in J and 0.57 magnitudes in K' i n order to obtain the correct colour of 4.12 at a redshift of z = 2.3. Th is results in a predicted 850 /zm flux of ~ 0.5 mJy , close to predicted value of 1 m J y that Totani et al . [63] determined. In comparison, when we use the Arp220 S E D and place it at this same redshift of z = 2.3 we predict a 850 /zm flux of « 20 mJy . Th is demonstrates how different redshifts wi l l be inferred, given the large difference in the predicted 850 / im flux, depending entirely on the type of S E D that is chosen. The result CHAPTER 3. HEROS 58 Figure 3.10: The amount of dust extinction that required to redden our we our template S E D to match the data versus increasing inferred redshift at the J (solid line) and K' (dotted line) rest wavelengths. The curves display an asymp-totic behaviour to values above zero, i l lustrating that some amount of dust extinction must always be applied to the S E D . CHAPTER 3. HEROS 59 of 0.5 m J y from the typical el l ipt ical S E D is plausible given our measured average H E R O flux. The H E R O s could in fact be typical dusty starburst galaxies at z ~ 2.3 which formed at z ~ 3 and are the progenitors of present-day giant ell ipticals, as Totani et a l . [63] suggest. We make note here of some caveats wi th this model and address some issues. We have assumed that the far-IR luminosity does not increase or decrease as we alter the amount of dust or extinction in the opt ical /near-IR port ion of the S E D v ia equation 3.1. Performing numerical integration over the S E D of F i g . 3.2 reveals that the total power absorbed in reddening the starburst galaxy template is only about 3 — 5 per cent of the total galactic emission. The smal l increase in the far-IR luminosity can be neglected as ) the reddening is altered. We also scale the galaxy brightness uniformly across the spectrum, which inherently assumes that intrinsic properties such as temperature and density remain constant, which essentially has the effect of simply changing the physical size of the galaxy. The simple one-to-one relation between dust temperature and I R luminosity which is widely used empir ical ly is also assumed here and is l ikely to break down for high-redshift galaxies from a theoretical standpoint (see Totani & Takeuchi [62]). Dust grains in a real galaxy always span a range of temperatures (20-200K), but in reality a large amount of it is cold (i.e. at the lower end of the 20-200K range). Therefore, the underlying assumption of a constant temperature seems reasonable here. The redshift upper l imi t would be weakened if the dust temperature was allowed to increase over a modest range. As alluded to earlier, high-redshift source SEDs may be systematically different, since pr imordial galaxies w i l l contain smaller-sized and more chemically simpler dust which wi l l heat up to higher temperatures due to the higher reflectivity and grain physics involved (see for e.g. Takeuchi et al . [61]). Since the H E R O s are extreme examples of E R O s , it is useful to consider how the results would change by fitt ing an ERO-speci f ic S E D to the data aswell. H R 1 0 was detected by H u & Ridgway [37], has subsequently been well-studied at many wavelengths, and was the first E R O discovered to be associated wi th a U L I R G . Not suprisingly, the CHAPTER 3. HEROS 60 1 1 i i i 111111 i n—i i 111111 1—i—i 111111 1 —i i 111 i|i | 1—i i 1 1 1 1 1 O b s e r v e d W a v e l e n g t h [fim] Figure 3.11: We scale and plot the Totani et al . [63] model of a typical el l ipt ical galaxy wi th present-day absolute magnitude of M B = —20, J — K' colour of 4, and formation redshift z? = 3 at a redshift of z = 2.3. The model contains emission from the direct stellar light that survived absorption by dust, as well as emission from heated dust. We scale the brightness to match the A'-band brightness of S D F 4 and adjust the J — K' colour slightly to that of S D F 4 (4.12) using our extinction law (cf. Eqn . 3.1) (solid l ine). For compar-ison, we plot Arp220 (dashed line) at z = 2.3 and adjust the brightness and . colour to match that of S D F 4 . The three vertical lines represent (from left . to right), the J-band, the A^'-band, and 850 /mi . Th is plot demonstrates how different redshifts wi l l be inferred, given the large difference in the predicted 850 /an flux, depending entirely on the type of S E D that is chosen. CHAPTER 3. HEROS 61 S E D for HR10 matches the S E D shape of Arp220, a nearby U L I R G , but is brighter (i.e. more intr insically luminous) by a factor of 3.8 (see F ig . 3.12 and Elbaz et al . [28]). HR10 lies at a redshift of z — 1.44 and is thought to be no more than a distant clone of Arp220. If we examine this template specific to E R O s , and fit it to the H E R O s , we find that the brightness factor is reduced to the H E R O s being 0.34 times at bright as HR10 as opposed to being 1.3 times as bright as Arp220 at a redshift of 1.6. Ma ihara et a l . [46] state that the H E R O s are likely to be remote galaxies, based on their "stellarity" indices and their apparent spatial extent. F rom the S D F A' ' -band image, we estimate the spatial extent of the H E R O s to be between 1.2 to 2.0 arcsec in diameter. Based on this estimate from the image, if the H E R O s are at z ~ 1.6, then this would make them ~ 4 — 6.5 kpc in size. This is consistent wi th the known range of galaxy sizes of 1 kpc up to 50 kpc. 3.6.3 Could the HEROs be Normal Spiral Galaxies? We use the modified normal spiral galaxy template to create F i g . 3.8, a plot of expected 850 / /m flux as a function of redshift for a H E R O which is representative of the whole populat ion wi th the average colour and magnitude of the four-source sample. A s before, we use the Bayesian 95 per cent upper confidence l imit flux estimates to derive the upper redshift l imits l isted in the last column of Table 3.2. When we force our template S E D to match the 95 per cent upper l imi t to the 850 fim flux of 1.76 mJy , constrained by a faint K' magnitude and very red J — K' colour, we estimate a corresponding redshift of 1.74. Based on the assumption that hyper-extremely red objects are well-represented by an S E D that fits NGC3938 , then they are at a redshift less than ~ 1.74. F i g . 3.14 investigates how much we need to brighten the S E D to get the correct K' magnitude after adding in an amount of dust to the NGC3938 S E D to get the correct J — K' colour. We notice that we need the K' flux to be dimmer than that of this 'dust-enriched' NGC3938 unt i l z ~ 0.6, where the H E R O starts becoming intr insical ly brighter CHAPTER 3. HEROS 62 1 0 - 1 1 0 ° 1 0 1 1 0 2 1 0 3 1 0 4 1 0 5 Observed Wavelength [ / i . m ] Figure 3.12: Th is is a model of HR10 based on an Arp220 S E D brightened by a factor of 3.8 to fit the S E D of HR10 at a redshift of z = 1.44. No dust has been added to this model. The crosses mark photometric data points and the arrows denote upper l imit flux estimates from N E D . CHAPTER 3. HEROS 63 10.00 F 1 1 1 1 I 1 1 1 1 I 1 1 I— 1— 1— 1— 1—q 0.01 I i i i i I , , i i I , i i , L_1J U_J i _ l 0.0 0.5 1.0 1 .5 2.0 R e d s h i f t z Figure 3.13: A plot of expected 850/ im versus redshift for our average H E R O , based on the normal spiral galaxy template. To reach a flux level of 1.76 m J y at 850/tm, the template must be moved out to z ~ 1.74. Using the Arp220 S E D , the template only needed to be moved out to z ~ 1.61 in order to reach the same 850 am flux. Calculated data points are marked wi th plus signs and the asterisks represent where we read off a redshift, based on the mean and upper l imit fluxes. CHAPTER 3. HEROS 64 Figure 3.14: A plot of the brightness in the i f ' -band (giving the correct K' magnitude) compared to the brightness of spiral galaxy NGC3938 (after adding in the right amount of extra dust extinction included to obtain the observed J — K' colour) versus redshift for our average H E R O . CHAPTER 3. HEROS 65 Figure 3.15: The amount of dust extinction that we must add to our normal galaxy S E D template for increasing redshift at the J (solid line) and K' (dotted line) rest wavelengths. and more luminous than the dusty template. For z ~ 1.7, we need to mul t ip ly the S E D by a factor of ~ 30 to get the correct K' magnitude for the H E R O sample. The brightest H E R O , S D F 1 , requires the S E D to be brightened by a factor of 60 at its upper redshift l imit of z = 1.79. F ig . 3.7 reveals the energetics of the normal cold spiral galaxy, and it is clear that there is just as much power in the near-IR peak as in the far-IR peak. Th is was not the case in the starburst galaxy template, where the far-IR peak dominated the energy output of the galaxy. W i t h this in mind, it is clear that for such galaxies we cannot CHAPTER 3. HEROS 66 ignore the effect of dust extinction on the total power of the galaxy. The power that would be "lost" due to reddening the spiral galaxy must be added back into the far-IR region of the spectrum in order to conserve energy. We calculate the total power of the galaxy before adding dust, and calculate the power in both the far-IR and near-IR peaks after adding some dust to the galaxy. We subtract the power in the near-IR peak from the total power "before dust" to find the amount of energy that must be elsewhere in the spectrum in order to conserve energy. We achieve this by first spl i t t ing the spectrum into two different components (the near-IR bump and the far-IR bump) and mul t ip ly ing each bump by a smooth function of the form: A(s) = e<£>" (3.7) for a decreasing exponential, or f2(x) = 1 - e<£> n (3.8) for an increasing exponential. A* is the cut-off wavelength where the function drops off steeply and n controls the steepness of the function's drop-off (we choose a large enough value, n = 20, to get sufficient steepness). Subsequently, we add the two components back together to create one unified template. We take the ratio of the "missing power" (i.e. the power not in the near-IR peak) to the amount of energy in the far-IR to find the far-IR "boost ing" factor. Mu l t ip ly ing the far-IR bump by this factor effectively moves the entire far-IR vFv spectrum up by the same amount. B y conserving energy in this way, the total power absorbed in reddening the galaxy is extremely negligible. Even though this scaling method is rather crude, it avoids changing the temperature and other intrinsic physics of the galaxy. F i g . 3.15 shows that the amount of dust reddening applied to the NGC3938 S E D in order to match J — K' decreases wi th redshift. (Recall that our NGC3938 template S E D is comprised of Lagache et al.'s (2002) normal cold spiral galaxy template wi th dust added v ia E q n (3.1) wi th x = 0.1). After z ~ 1, the amount of required dust decreases very slowly. A t z = 1.74, we add in only « 0.7 magnitudes in K' and « 1.2 magnitudes CHAPTER 3. HEROS 67 in J. For the reddest H E R O , S D F 4 , we need to add « 1.8 magnitudes of dust in J at the upper redshift l imit of z = 1.76 NGC3938 is not a reasonable analogue for the H E R O s , since we need to brighten the galaxy by an extraordinary amount to fit the entire S E D (i.e. more than a factor of a few). A n 850/ im flux of 1.76mJy indicates that the galaxy lies at z = 1.74, the galaxy would need to be intrinsically brighter by a factor of about 30 compared wi th a normal spiral galaxy, not a very realistic scenario for this type of galaxy. The possibi l i ty of the H E R O s being normal spiral galaxies is therefore eliminated. Even if NGC3938 was at a lower redshift and would therefore not need to be brightened by such a large factor, we would need to add in more dust than would be reasonable (i.e. more than one or two magnitudes of extinction in J, c f . F ig . 3.15). If the flux is just above the confusion l imi t at ~ 0.6 m J y then the redshift is about z ~ 1.3 and the galaxy would be st i l l need to be brighter than NGC3938 by a factor of 8 and contain about 1.3 magnitudes of extinction in J-band. CHAPTER 4. SDF SOURCES 68 C H A P T E R 4 SDF SOURCES A n image is "confused" when multiple unresolved faint objects cluster in one beam-size. A s we look fainter and fainter, d im objects become more numerous, superimposing their signals on each other, unt i l the confusion l imit is reached. We investigate the number of pixels wi th in a certain flux level in F ig . 4.1, a histogram of the number of pixels at each flux level in the 850 / im S D F map. The positively skewed non-Gaussian single peak distr ibut ion of flux demonstrates that a population of submil l imetre sources lies in the positive ta i l of the distr ibut ion, below the confusion l imit of these data (Condon [15]). If we subtract the flux bins reflected about 0 mJy from the original flux bins, this effect becomes apparent (see F ig . 4.2). We can therefore infer that a populat ion of sources lies just below the confusion l imit of the 850 / im S D F map. Figs. 4.3 and 4.4 demonstrate this also for the 450 / im map. This result hints that we are detecting submil l imetre emission from the S D F and that it could be correlated wi th S D F data in another wavelength region. 4.1 Correlating the IR Galaxies with the Submillimetre Data Kashikawa et a l . [43] have constructed a deep i f ' -band selected B,V,R,I,z',J,K' mul t i -colour sample of 439 galaxies (K'< 24.0) in the Subaru Deep F ie ld and estimated a photometric redshift for each galaxy using the public domain HYPERZ code wri t ten by Bolzonel la, Miral les & Pello [5]. HYPERZ finds the redshift of a galaxy using a standard S E D f i t t ing procedure, i.e. comparing the observed magnitudes wi th those computed from template S E D s . Throughout this thesis we use the more complete, unpublished 526 i f ' -selected galaxy catalogue based on the criterion of K{sophot< 24.6 (Kashikawa 2002, private communication). In F i g . 4.5, we plot the mean of the 20 brightest i f ' - band CHAPTER 4. SDF SOURCES 69 250 T 1 1 1 1 r Figure 4.1: We plot the number of pixels at each brightness level in the 850 pm map as a histogram wi th 100 bins of width 0.2 mJy (solid jagged line). The overlayed dashed histogram represents the flux bins reflected about 0 m J y , demonstrat-ing that there is clearly more positive flux in the map than negative flux. The flux distr ibut ion is clearly non-Gaussian and has enhanced high and low flux tails when compared to the overlaid Gaussian. The Gaussian fit to the data (smooth solid line) has a height of 186 ± 0.27, center of - 0 . 1 2 ± 2.3, and a half-width half-maximum of 2.26 ± 0.004. CHAPTER 4. SDF SOURCES 70 Figure 4.2: We plot the difference of the flux bins and the flux bins reflected about 0 mJy . If we take the difference between a positively skewed single peaked distr ibut ion and its reflection about OmJy, we reproduce this result. We can therefore infer a population of sources lying just below the confusion l imi t of the 850 / im map. CHAPTER 4. SDF SOURCES 71 1 2 0 I 1 1 1 I 1 1 1 I 1 1 1 1 1 1 1 1 1 1 ' 1 1 ' r Flux B i n s ( m J y ) Figure 4.3: We plot the number of pixels at each brightness level in the 450 pm map as a histogram wi th 120 bins of width 1.0 m J y (solid jagged line). The overlayed dashed histogram represents the flux bins reflected about 0 mJy , demonstrat-ing that there is clearly more positive flux in the map than negative flux. The flux distr ibut ion is clearly non-Gaussian and has enhanced high and low flux tails when compared to the overlaid Gaussian. The Gaussian fit to the data (smooth solid line) has a height of 93 ± 0.27, center of - 0 . 7 3 ± 0.04, and a half-width half-maximum of 11.4 ± 0.04. CHAPTER 4. SDF SOURCES 72 c 'CD > o I " D (D x> o o Q X I c o Q c in CQ O _ -60 -40 -20 0 20 Flux Bins (mJy) 40 60 Figure 4.4: We plot the difference of the flux bins and the flux bins reflected about 0 mJy . If we take the difference between a positively skewed single peaked distr ibut ion and its reflection about 0 mJy , we reproduce this result. We can therefore infer a population of sources lying just below the confusion l imit of the 450 fim map. CHAPTER 4. SDF SOURCES 73 objects in redshift bins of 0.5 w id th 1 , and observe that the faintest i f ' - band objects are also at the highest redshifts. It is interesting to investigate the measured submil l imetre emission of the A"'-selected galaxies as a function of redshift, since i t can potential ly reveal if amounts of hidden star formation evolve wi th redshift. Peacock et a l . [48] have performed a similar analysis wi th ultraviolet posit ional information for the Hubble Deep Fie ld and found l itt le correlation at z < 1, but a definite positive signal spread out over higher redshift bins (z > 2.5). Th is demonstrates that their submil l imetre map of the H D F receives emission from galaxies over a wide range of redshifts. A A' -band source which is very faint or undetected in optical data (i.e. J or / -band) holds a strong possibil i ty of being S C U B A bright - or at least we know the converse to be true. We currently do not know how to predict which E R O s w i l l be S C U B A - b r i g h t . We would like to test if faint near-IR emission is correlated wi th submil l imetre emission. We might expect higher redshifted near-IR objects to be stronger submil l imetre sources, since the far-IR peak of the S E D wi l l be significantly shifted into the submil l imetre regime. The best way to see if we are detecting IR-galaxy emission statistically in the 850 /mi map is to measure the flux density at the location of each galaxy and look for a correlation. We take the value from the map at the posit ion of each IR source as the best estimate of its 850 /im flux density. These flux densities are then averaged or "stacked" into redshift bins (see F i g . 4.6). We also perform the same analysis on the 450/mi map (see F i g 4.7). B y div iding the maps up into redshift slices, we can now see if one redshift band tends to dominate more in the submill imetre than another. In order to quantify the degree of correlation we use the linear Pearson correlation coefficient, a number between —1 and +1 which measures the degree to which two variables are l inearly related using a least-squares fit. A value of 0 indicates no correlation and values of —1 and +1 indicate a strong correlation (inverse proportionality and direct proportionality, respectively). We weight the mean fluxes by a factor pf 1/cr2 (cr is the error bar on each bin) in order to take the error bars into account before computing the correlation coefficient. We obtain 1We choose the number of bins and the bin occupancy based on the best compromise between noise reduction and a sensible number of points. CHAPTER 4. SDF SOURCES 74 2 4 2 2 h & 2 0 o E 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1111111111 1 1 1 1 1 1 1 1 1 1 i » J 1 6 I n 111 1111 I i i i 111 I I 0 1 2 3 4 5 6 R e d s h i f t z Figure 4.5: A plot of the 20 brightest i f ' -band objects per redshift b in. The vertical error bars show the 1 a standard deviation of the K' magnitudes in each redshift b in. The asterisks mark the mean redshift of each bin and the horizontal error bars represent the coverage of each redshift b in. The 7th bin only contains 14 objects due to the low number of objects in this redshift b in. We remark that the the faintest i f ' -band objects also have the highest photometric redshifts. CHAPTER 4. SDF SOURCES 75 E o LD oo R e d s h i f t z Figure 4.6: A plot of observed average 850 //m flux per redshift b in (Az = 1). Start ing wi th the lowest redshift bin, the number of galaxies contained in each bin is 183, 150, 100, 36, 46, 6, 4, 1. It is evident that the highest redshift bins contain the smallest number of objects and so wi l l have the largest error bars. The lowest redshift bins contain a large number of objects and so the average flux wi l l tend to be closer to 0 mJy , since the measurement is beginning to be constrained by the map average. CHAPTER 4. SDF SOURCES 76 E a. o in X - 1 0 h R e d s h i f t z Figure 4.7: A plot of observed average 450fim flux per redshift b in (A.2 = 1). Start ing wi th the lowest redshift b in, the number of galaxies contained in each bin is 183, 150, 100, 36, 46, 6, 4, 1. It is evident that the highest redshift bins contain the smallest number of objects and so wi l l have the largest error bars. The lowest redshift bins contain a large number of objects and so the average flux wi l l tend to be closer to 0 mJy , since the measurement is beginning to be constrained by the average of the map. CHAPTER 4. SDF SOURCES 77 correlation coefficients of 0.49 and 0.38 at 850 and 450 jura respectively. In order to see how likely it is to achieve these outcomes and correlation coefficients simply by chance, we perform 1000 statistical Monte Car lo simulations of this procedure (see Figs. 4.8, 4.9, 4.10, 4.11). The Monte Car lo simulation selects positions (the same number of times as there are objects in the bin) from the S D F map at random and mea-sures the flux there. For each bin, we calculate the mean and error, again weighting the measurements by the corresponding value on the noise map. We calculate a correlation coefficient between the error-weighted (1/tr2) mean flux and redshift b in for a complete set of redshift bins and repeat the whole procedure 1000 times in order to get a well-defined distr ibut ion of correlation coefficients. It is now possible to estimate how likely it is to obtain a certain correlation coefficient at random. Looking at these plots, getting a correlation coefficient of 0.49 or higher by chance occurs about 80 per cent of the time for the 850 pm map and getting a coefficient of 0.38 or higher by chance occurs about 40 per cent of the t ime for. the 450 fxm map. From these results, there is no strong case for a correlation between the stacked submil l imetre fluxes and redshift. A l though there does seem to be a hint of positive flux spread over the redshift range 3 < z < 6 in the 450 pm map (see F ig . 4.7), these results suggest that the flux we detect in the submill imetre is not well-described by the AT'-band galaxy photometric redshifts. Because the first few bins each contain a large number of objects, we are inherently just taking the average of the map, and we therefore expect a stacked flux close to 0 mJy . The bins wi th fewer objects wi l l inherently posess very large error bars, as they suffer from small number statistics. It is therefore more useful to rebin the objects into bins wi th an equal number of objects. We redo the analysis for Figs. 4.6 and 4.7 but instead divide the objects into bins wi th equal numbers of objects to see if this makes a trend visible (see Figs. 4.12 and 4.13). We get correlation coefficients of 0.40 and 0.43 for the 850 and 450 fj,m data-sets, respectively. We perform 1000 Monte Car lo simulations again (see Figs. 4.14 and 4.16), but this t ime wi th an equal number of objects per b i n 2 , 2 A l l bins contain 40 objects except for the last bin which contains 46 objects, mimicking how the real CHAPTER 4. SDF SOURCES 78 2 5 0 2 0 0 1 5 0 CL I 1 0 0 5 0 •1.0 n 1 1 1 1 1 1 r _i I i _ -0 .5 0.0 0.5 C o r r e l a t i o n C o e f f i c i e n t B in 1.0 Figure 4.8: Th is plot shows the distr ibution of correlation coefficients obtained from 850 / im stacked fluxes versus redshift for a set of Monte Car lo simulations of stacked 850/xm flux in redshift bins of Az = 1. CHAPTER 4. SDF SOURCES 79 Figure 4.9: Th is plot shows how many times a correlation coefficient is obtained for the 850 / im stacked fluxes versus redshift from a set of Monte Car lo simulations of stacked 850 / im flux in redshift bins of Az = 1. CHAPTER 4. SDF SOURCES 80 100 80 c CD 0 (all objects), z > 2, and z > 4. We notice that the mean 850 /xm flux does not vary wi th My, but this can be more clearly seen if we stack the submil l imetre flux. The leftmost plots in Figs. 4.21 and 4.22 represent stacked 850/an flux for equal-size bins of My. These plots generally give nul l results, except for a hint of a correlation for objects wi th z > 4 of 850/xm flux increasing wi th the faintness in My. Th is would be plausible considering that strong submil l imetre detections are associated w i th a great deal of dust extincting the optical light. The rightmost plots in Figs. 4.21 and 4.22 represent stacked 850 /xm flux wi th an equal number of objects in each bin and al l show no trends. We conclude that the lack of a stong trend of submil l imetre flux wi th My indicates that the submill imetre emission in the S D F map is not well-described by the absolute V magnitudes of A ' - b a n d selected objects in the S D F . 4.1.1 Detections We use Co l in Borys ' B S _ F I N D S O U R . C E program to pick out sources by fitting the point spread function (PSF) of the beam to each pixel on the map. Br ight peaks in the convolved (i.e. smoothed) map are selected as "detected" sources when the P S F matches a source in the map (see Borys [6] for further details). We note detections of three S N R > 3.0 a S C U B A sources at 450 and 850 /xm combined, but unfortunately they al l lie outside of the region covered by this deep A^'-band image (see Figs. 2.8 and 2.9 for the maps). Table 4.1 lists the detections in the 450 and 850 /xm maps. Using the source counts of 850/xm submil l imetre sources (cf. F i g . 1.6), we expect to see « 1 source detected at the 3.5 o~ level, given a map rms of 2 mJy . Th is estimate is consistent wi th the single 3.5 a detection we make. The number counts at 450 /xm are not as well known, however, CHAPTER 4. SDF SOURCES 92 -20 -15 Absolute V magnitude E a. -20 -15 Absolute V mognitude -20 -15 Absolute V magnitude Figure 4.20: From left to right: these scatter-plots represent 850 am flux for every object detected in the V-band (488 objects total), for objects detected in the V-band wi th a photometric redshift higher than 2 (158 objects total), and for objects detected in the V-band wi th a photometric redshift higher than 4 (31 objects total). CHAPTER 4. SDF SOURCES 93 - 2 5 - 2 0 - 1 5 - 1 0 Absolute V magnitude E X 0.6 0.4 0.2 £ 0.0 r LO CO -0.2 r -0.4 r - 0 . 6 * * - 2 5 - 2 0 - 1 5 - 1 0 Absolute V magnitude Figure 4.21: The leftmost plot represents stacked 850 fxm flux measured in M y bins, two magnitudes in width, for al l objects detected in the F-band (488 objects total). The rightmost plot represents 850/zm flux measured in M y bins, wi th an equal number of objects per bin (49), for al l objects detected in the V-band. The horizontal bars represent the actual range of magnitudes in a bin and the vertical error bars show the standard deviation of the mean stacked flux in each bin. CHAPTER 4. SDF SOURCES 94 E 3. o in oo X 3 - 2 5 - 2 0 - 1 5 - 1 0 Absolute V magni tude - 2 5 - 2 0 - 1 5 - 1 0 Absolute V magn i tude E =1 o t n co X 1 3 - 2 0 - 1 5 - 1 0 Absolute V magni tude - 2 5 - 2 0 - 1 5 - 1 0 Absolute V m a g n i t u d e Figure 4.22: The leftmost plots represent stacked 850 / im flux measured in M v bins, two magnitudes in width, for al l objects detected in the 7-band wi th z > 2 (158 objects total) (top plot) and for al l objects detected in the F-band wi th z > 4 (31 objects total) (bottom plot). The rightmost plots represent 850 /xm flux measured in M y bins, with an equal number of objects per b in, for al l objects detected in the F-band wi th z > 2 (40 objects per bin) (top plot) and for al l objects detected in the V-band wi th z > 4 (10 objects per bin) (bottom plot). The horizontal bars represent the actual range of magnitudes in a bin and the vertical error bars show the standard deviation of the mean stacked flux in each bin. CHAPTER 4. SDF SOURCES 95 and we make no attempt to estimate how many sources we expect to see at 450 / im, given the depth of the map and the area surveyed. ID Posit ion (2000.0) F l ux (mJy) S N R S D F 850.1 13 h24m16?91 +27°29'46'.'01 S 8 5 0 = 4.40 ( ± 1.25) 3.5 cr S D F 450.1 13 h24m25?25 +27°30'3r.'01 S 4 5 0 = 75.45 ( ± 18.52) 4.1 cr S D F 450.2 13 h24m19?39 +27°30'3r.'01 S 4 5 0 = 50.29 ( ± 16.50) 3.0 cr Table 4.1: The positions, fluxes and signal-to-noise ratios (SNR) of significant (> 3cr) S C U B A detections in the S D F . CHAPTER 5. CONCLUSIONS 96 C H A P T E R 5 CONCLUSIONS We have mapped the 850 and 450 /zm continuum emission of the Subaru Deep F ie ld using S C U B A on the J C M T . We found that the observed high J - K' (K 3 - 4) sources produce an amount of S C U B A flux which is near to or below the ~ 1 m J y confusion level, making their true flux difficult to estimate. The four H E R O sources have an average 850 /zm flux of 1.15 ± 0.46 m J y or a combined 95 per cent upper confidence l imi t of 1.76 mJy , a hint that they are dusty galaxies emitt ing a small amount of submill imetre flux. Naively, the H E R O s are most l ikely to be starburst galaxies based on their apparent irregular morphologies rather than compact regularly-shaped el l ipt ical galaxies. Based on the assumption that H E R O s are well-represented by an S E D that fits Arp220, then they are at z < 1.6, consistent wi th known redshifts for S C U B A sources, but at the low redshift end 1 . If we assume instead that the H E R O s lie at average S C U B A redshifts (2 « 2 — 5), then the typical el l ipt ical galaxy model from Totani et a l . [63] would also fit the data for at a redshift of z = 2.3. Using an Arp220-l ike S E D , we need to increase the brightness of the average H E R O by a reasonable amount (about a factor of 1 — 1.5) and add a modest amount of dust (about 1 magnitude in J ) in order to obtain an adequate fit. Thus, there is no reason to suspect that these objects are J-dropout galaxies at extraordinary redshifts. The H E R O s are not well-represented by an NGC3938- l ike normal face-on spiral galaxy, given the flux constraints, since they would need to be brightened by about a factor of about 20 — 30, JAt optical and near-IR wavelengths, deep images are crowded with galaxies with no more than a few percent of them lying at z > 3 (see Guhathakurta, Tyson & Majewski [34]), as the galaxies do not benefit from the negative K-correction at near-IR wavelengths as they do in the submillimetre. We note that we may be only be picking up the low end of the redshift distribution by selecting objects in the iY'-band. CHAPTER 5. CONCLUSIONS 97 a larger than acceptable amount for a galaxy of this nature. Because the H E R O s are faint in optical-IR-submil l imetre wavebands wi th current instrument sensitivities, these objects may st i l l play an important role in tracing the dust-enshrouded part of the early star formation picture. If the H E R O s are dusty starbursts, as this work suggests, we see that the optically derived star formation rate (SFR) would be underestimated due to the presence of dust. Our strongest detection (S 85o = 2.15 ± 0.92 mJy , is the most convincing case for the H E R O s being dusty starbursts. However, they could well be a mixed class of objects. Even for this object, its A ' - b a n d morphology looks superficially to be a merging pair of galaxies with' z < 1.6, as determined from the Arp220 starburst template, consistent with a hierarchical formation scenario, where galaxies are made over a long period of t ime through successive mergers or through accretion of clumps of matter. Barger et a l . [2] find that most of the submil l imetre extragalactic background is emitted by 1 < z < 3 sources, so the peak of the starburst act ivi ty lies at moderate redshifts. If there are many H E R O s of this type at these redshifts then this has serious implicat ions for the S F R as a function of redshift, since we may be missing more than was previously thought. Our measurements of the H E R O s do not support the prediction that H E R O s should be S C U B A - b r i g h t . S C U B A galaxies are very often identified wi th E R O s and we have tested if the inverse statement also holds true. Our results clearly demonstrate that measuring the submil l imetre flux of H E R O s wi l l not necessarily select very bright S C U B A sources. S C U B A galaxies may just comprise one sub-set of a more diverse populat ion of E R O s . From Figs. 4.1 and 4.3, we infer that we are detecting a populat ion of sources just below the confusion l imit of the maps. We correlate the IR A ' - b a n d selected galaxy positions wi th 450 and 850 / im flux and find that there is no correlation at the sensitivity l imit of our measurements of the S D F . We conclude that the submil l imetre flux we detect in our data is not characterized by the A ' - b a n d selected IR galaxies or by their photometric redshifts, and is presenting a view of the Universe different from that of the IR S D F . Al though there is no apparent correlation wi th the IR S D F galaxies, the S C U B A map could st i l l potential ly be correlated wi th U V , optical, X - ray or radio data. CHAPTER 5. CONCLUSIONS 5.1 Future Work 98 Future S C U B A observations could provide even deeper l imits (down to an rms of « l m J y ) , but wi l l be constrained by the confusion l imi t (~ 0.5 mJy) . Deep V L A ob-servations would allow us to obtain further redshift constraints through the well known radio/ far- IR correlation 2 (see Chapman et al . [13]). A L M A (submm), S I R T F and J W S T (IR) wi l l have the required sensitivities to detect these faint objects and possibly reveal their nature and redshifts wi th wide-band spectroscopy instruments. If the H E R O s are at z < 1.6 then they should be detectable by B L A S T (Balloon-borne Large Aperture Submil l imetre Telescope) or with S C U B A at 450 / im in very good dry weather. Those data, in addit ion to C S O / S H A R C - I I 350/tm observations, could help to better constrain the assumed shape of the S E D , thereby providing tighter constraints on the inferred red-shift. A more complete study might include the use of a wider range of S E D s , rather than just the two templates that were employed here. In addit ion, without spectroscopic redshifts, substantial progress may also require model SEDs wi th a proper treatment of the dust at the same time as the stellar population evolution. 2 The underlying power causing the emission originates from the same short-lived massive stars 99 BIBLIOGRAPHY I] Arch iba ld E . N . , Wagg J .W. , Jenness T., 2000, Calculat ing Sky Opacit ies: a re-analysis for S C U B A data, S C D system Note 2.2 2] Barger A . J . , Cowie L .L . , Smai l I., Ivison R . J . , B la in A . W . , Kne ib J . - P . , 1999, A J , 117, 2656 3] B la in A . W . , Smai l I., Ivison R . J . , Kne ib J . - P , Frayer D.T. , 2002, preprint (astro-ph/0202228) 4] B la in A . W . , Barnard V . E . , Chapman S.C. , 2003, M N R A S , 338, 733 5] Bolzonel la, M . J . , Miral les, M . , Pel lo, R., 2000, A & A , 363, 476 6] Borys C , 2002, P h D . thesis, Univ. of Br i t ish Columbia 7] Borys C , Chapman S .C. , Halpern M . , Scott D., 2000, preprint (astro-ph/0009143) 8] Borys C , Chapman S.C. , Halpern M . , Scott D., 2002, M N R A S , 330, L63 9] Calzet t i D., K inney A . L . , Storchi-Bergmann T. , 1994, A p J , 429, 582 10] Cardel l i J . A . , Clayton G . C . , Math is J .S . , 1989, A p J , 345, 245 II] Carro l l S . M . , Press W . H . , Turner E .L . , 1992, A R A & A , 30, 499 12] Chapman S .C . et a l . , 2000, M N R A S , 319, 318 13] Chapman S . C , Lewis G . E . , Scott D., Borys C , Richards E. , 2002, A p J , 570, 557 14] Chapman S . C , Scott D., Borys C , Fahlman G . G . , 2002, M N R A S , 330, 92 15] Condon J . J . , 1974, A p J , 188, 279 100 [16] Davis M . , Efstathiou G . , Frenk C.S. , Wh i te S . D . M . , 1985, A p J , 292, 371 [17] Dey A . , Graham J .R . , Ivison R . J . , 1999, A p J , 519, 610 [18] Dickinson M . E . et al . , 2000, A p J , 531, 624 [19] Dole H. , 2000, P h D . thesis, Univ. of Par is-Sud X I , Orsay [20] Dole H. et a l . , 2001, A & A , 372, 364 [21] Downes D. et a l , 1999, A & A , 347, 809 [22] Downes A . J . B . , Peacock J .A . , Savage A . , Carrie D.R. , 1986, M N R A S , 218, 31 [23] Dunlop J .S . , 2001, N e w A R , 45, 609 [24] Dunlop J .S. et a l , 2002, preprint (astro-ph/0205480) [25] Dunne L., Eales S., Ivison R., Morgan H. , Edmunds M . , 2003, Nat , 424, 285 [26] Eales S., L i l l y S., Webb T. , Dunne L., Gear W . , Clements D., Y u n M . , 2000, A J , 120, 2244 [27] Eggen O . J . , Lynden-Bel l D., Sandage A . R . , 1962, A p J , 136, 748 [28] E lbaz D., Flores H. , Chania l P., Mi rabe l I.F., Sanders D., Due P. -A. , Cesarsky C . J . , Aussel FL, 2002, A & A , 381, L I [29] Ferguson H.C. , Dickinson M . , Wi l l iams R., 2000, A R A & A , 38, 667 [30] Franceschini A . , 2000, preprint (astro-ph/0009121) [31] Prayer D.T., Smai l I., Ivison R . J . , Scoville N.Z., 2000, A J , 120, 166 [32] Gear W . K . , L i l l y S . J . , Stevens J .A . , Clements D.L. , Webb T . M . , Eales S.A. , Dunne L., 2000, M N R A S , 316, L51 [33] Gispert R., Lagache G . , Puget J . -L . , 2000, A & A , 360, 1 [34 [35; [36 [37; [38 [39 [40 [41 Jenness T. , Stevens J .A . , Arch iba ld E . N . , Economou F., Jessop N .E . , Robson E.I., 2002, M N R A S , 336, 14 [42] Johnstone D., Wi lson C D . , Moriarty-Schieven G . , Giannakopoulou-Creighton J . , Gregersen E. , 2000, A p J S , 131, 505 [43; [44; [45; [46 [47 [48 [49 [50 [51 [52; 101 Guhathakur ta P., Tyson J .A . , Majewski S.R., 1990, A p J , 357, L9 Hi ldebrand R . H . , 1983, Q J R A S , 24, 267 Hol land W . S . et a l , 1999, M N R A S , 303, 659 H u E . M . , Ridgway S .E . , 1994, A J , 107, 1303 Hughes D .H . et a l . , 1998, Nat , 394, 241 Ibata R .A . , Gi lmore G . , Irwin M . J . , 1994, Nat , 370, 194 Jenness T. , Lightfoot J . F . , 1998, Starl ink User Note 216, Starl ink Project, C L R C Kashikawa N. , et al . , 2003, A J , 125, 53 Kauf fmann G . , 1996, M N R A S , 281, 487 Lagache G . , Dole H , Puget J . -L . , 2002, M N R A S , in press Ma ihara T . et a l . , 2001, P A S J , 53, 25 O'Donnel l J . E . , 1994, A p J , 422, 1580 Peacock J . A . et al . , 2000, M N R A S , 318, 535 Rowan-Robinson M . , 2001, A p J , 549, 745 Saj ina A . , 2002, MSc . thesis, Univ . of Br i t ish Columbia Sanders D.B. , A p & S S , 266, 331 Sanders D.B. , Mi rabe l F., 1996, A R A & A , 34, 749 102 [53 [54 [55; Scott D. et a l . , The Promise of the Herschel Space Observatory. Eds. Phi lbrat t G . L . , Cernicharo J . , Heras A . M . , Prus t i T. , Harris R., 2001, E S A - S P , 460, 305 [56 [57; [ss; Smai l I., Ivison R . J . , Kne ib J . -P . , Cowie L .L . , B la in A . W . , Barger A . J . , Owen F . N . , Morr ison G . , 1999, M N R A S , 308, 1061 [59; [60; [61 [62 [63 [64 [65 [66 [67; Schlegel D . J . , Finkbeiner D.P. , Davis M . , 1998, A p J , 500, 525 Searle L., Z inn R., 1978, A p J , 225, 357 Scott S .E . et a l , 2002, M N R A S , 331, 817 Skinner C . J . , 1996, Instrument Science Report N ICMOS-014 Smai l I., Ivison R . J . , B la in A . W . , Kneib J . -P . , 2002, M N R A S , 331, 495 Steidel C . C . , 1999, Proc. Nat l . Acad . Sci . U S A , 96, 4232 Takeuchi T .T . , Hirashi ta H. , Ishii T .T . , Hunt L .K . , Ferrara A . , 2003, M N R A S , 343, 839 Totani T. , Takeuchi T .T . , 2002, A p J , 570, 470 Totani T. , Yoshi i Y . , Iwamuro F., Ma ihara T. , Motohara K., 2001, A p J , 558, L87 Trentham N. , Kormendy J . , Sanders D.B. , 1999, A J , 117, 2152 Webb T . M . A , L i l l y S . J . , Clements D.L. , Eales S., Y u n M . , Brodwin M . , Dunne L., Gear W . K . , 2003, preprint (astro-ph/0308439) Whi te S . D . M . , Frenk C.S . , 1991, A p J , 379, 52 Wright E .L . , Hinshaw G . , Bennett C .L . , 1996, A p J , 458, L53 [68] Yahata N. , Lanzetta K . M . , Chen H.-W., Fernndez-Soto A . , Pascarelle S . M . , Yah i l A . , Puetter R . C . , 2000, A p J , 538, 493 [69] Zheng Z., W u H , Mao S., X i a X . - Y . , Deng Z . - G , Zou Z. -L . , 1999, A & A , 349, 737