5. ULIRGs AT HIGH REDSHIFT

5.1. The cosmic infrared background

The existence of a significant population of IR-luminous galaxies at high redshift, though hinted at in followup of IRAS, was thrown to the fore of astronomical debate by results from the Cosmic Background Explorer (COBE) satellite. Launched on November 18, 1989, COBE was responsible for two of the most significant astronomical results in the 20th century. The first was measuring the spectral shape and level of anisotropy in the Cosmic Microwave Background (CMB, peaking at ~ 2000 µm), providing overwhelming evidence for the Hot Big Bang model and giving a glimpse at the early structures that would eventually evolve into todays galaxies and clusters. The second result, more relevant for this review, was the discovery of a Cosmic Infrared Background (CIB) with FIRAS at 240 µm [Puget et al. 1996], and at 140 µm and 240 µm with DIRBE [Schlegel, Finkbeiner & Davis 1998, Hauser et al. 1998, Fixsen et al. 1998] (later detections with DIRBE of the CIB at 2.4 µm and 3.5 µm have been published, see [Hauser & Dwek, 2001; and Kashlinsky, 2005] for reviews). Such an infrared background had been predicted many years previously [Partridge & Peebles 1967], but had proven fiendishly difficult to detect, remaining invisible both to rocket-borne IR observatories [Kawada et al. 1994], and to IRAS [Rowan-Robinson et al. 1990, Oliver, Rowan-Robinson & Saunders 1992]. COBE was the first observatory with both the right instrumentation and sufficient sensitivity to detect the CIB.

The cosmological implications from the discovery of the CIB were profound [Dwek et al. 1998]. The total background detected by COBE between 140 and 5000 µm is ~ 16 nW m^-2 sr^-1, or 20%-50% of the total background light expected from energy release by nucleosynthesis over the entire history of the Universe, implying that 5%-15% of all baryons are or have been parts of stars. And while the CIB itself amounts to less than about 2% of the CMB, the intensity of the CIB is still surprisingly high, comparable to, or exceeding, the integrated optical light from the galaxies in the Hubble Deep Field [Hauser et al. 1998].

When compared to the cosmic history of star formation derived from optical and UV surveys [Madau et al. 1996] a serious discrepancy became apparent; the CIB detected by COBE requires at least a factor of two more star formation than was apparent in optical and UV surveys, meaning that the integrated star formation rate at z ~ 1.5 must be higher than that implied from UV/optical observations by a comparable factor, and that this star formation must be largely surrounded by dust. COBE however could provide little further constraint on the form of the star formation history implied by the CIB; in principle there could be large numbers of galaxies at high redshift that are faint in the IR, or a few extremely IR-luminous sources ⁷. Further progress required that the CIB be resolved into its constituent sources.

5.2. Resolving the CIB: LIRGs and ULIRGs at 0 < z < 1.5

The first major steps in resolving the CIB came from extragalactic surveys carried out by ISO at 7 µm and 15 µm with ISOCAM, and at 90 µm and 170 µm with ISOPHOT, most notably observations of the HDF [Oliver et al. 1997, Rowan-Robinson et al. 1997], the European Large Area ISO Survey (ELAIS; [Oliver et al., 2000; Rowan-Robinson et al., 2004]), and the FIRBACK survey [Puget et al. 1999, Dole et al. 2001] (see also reviews in [Elbaz, 2005; Verma et al., 2005; Oliver & Pozzi, 2005]). The 15 µm surveys [Elbaz et al. 2002] were particularly successful; the sources seen in these surveys could, with considerable extrapolation of the SED shape at longer wavelengths, account for around 80% of the CIB. Followup observations showed these sources have <z> ~ 0.8, and that the comoving density of infrared light due to these 15 µm sources is at least 40 times greater at z ~ 1 than in the local Universe (compare this to the B band luminosity density, which is only about three times the local value at z ~ 1). Not to be outdone, the FIRBACK survey found that the 170 µm source counts show very strong evolution with redshift (reaching z ~ 1), directly resolve 5% of the CIB [Dole et al. 2001], and are responsible for a dramatic rise in the integrated star formation rate, with a value at least ten times that seen locally at z ~ 1 [Rowan-Robinson et al. 1997, Flores et al. 1999, Pozzi et al. 2004]. Overall, therefore, the ISO deep surveys put the IRAS discovery of strong IR galaxy evolution onto a very firm footing. Out to z ~ 1 the ISO sources are chiefly LIRGs, not ULIRGs, and they are generally similar in many ways to lower redshift LIRGs, although possibly with lower average metallicity [Franceschini et al. 2003b, Elbaz 2005, Liang et al. 2004].

The 170 µm surveys also resulted in another interesting discovery; a large number of the 170 µm-selected objects have cooler dust temperatures and larger dust masses than those seen in starburst galaxies selected at mid-IR wavelengths. This is not unexpected for a long wavelength-selected survey, since a large luminosity from a cool dust component requires a larger dust mass than a warmer source. ISO however was the first observatory to show conclusively that large masses of 'cool' dust existed in many galaxies, from local spirals to distant ULIRGs. In local spirals this cold "cirrus" component is expected to be diffuse dust heated by the interstellar radiation field from later type stars. In ULIRGs however this cold component could alternatively be a compact dusty starburst with colder than average dust. This discovery has interesting implications for studies of ULIRGs at higher redshifts, as we will discuss in Section 5.3.

Focusing on ULIRGs; the largest ISO survey, and therefore the one most likely to discover ULIRGs in any number, was ELAIS (see Fig. 1). The most sensitive ELAIS band for LIRGs and ULIRGs was 15 µm, resulting in the detection of just under 100 ULIRGs, comprising > 10% of the 15 µm sample. Around 10 of these are HLIRGs; the first of these to be identified being a QSO at z = 1.01 [Morel et al. 2001]. These ULIRGs range in redshift up to z 3 and many are fit well in colour by an Arp 220-like SED. The longer wavelength ELAIS surveys result in many fewer sources due to decreased sensitivities; [Taylor et al. 2005] have identified 4 likely ULIRGs in the FIRBACK-ELAIS N2 (FN2) 170 µm population, while [Dennefeld et al. 2005] identify 1 or 2 in the FIRBACK-ELAIS N1 (FN1) field. [Sajina et al. 2003] identified a population of z ~ 0.5-1 ULIRGs, representing ~ 1/6 of the total 170 µm FN1 sample, and [Chapman et al. 2002] identified two of these at z = 0.5 and 0.9 to be unusually cold systems with merger morphologies. Several sources in each field remained optically unidentified and are likely to be ULIRGs at moderate redshifts (0.5-1). Several ULIRGs and HLIRGs have been discovered in other ISO surveys, including a z = 1.7 FeLoBAL QSO behind the z = 0.56 cluster J1888.16CL [Duc et al. 2002], and a radio detected ERO at z = 1.5 [Pierre et al. 2001], probably a QSO with very hot dust.

Follow-up of z 0.5 ULIRGs revealed that many are involved in ongoing interactions, but that interestingly a significant number are QSOs with very massive (> 2L^*) elliptical host galaxies [Farrah et al. 2002a]. Detailed SED modeling [Rowan-Robinson & Crawford 1989, Rowan-Robinson 2000, Verma et al. 2002, Farrah et al. 2002b] showed evidence for both starburst and AGN activity in most sources, with colossal implied star formation rates of up to ~ 1000 M yr^-1. X-ray observations however found unexpectedly weak X-ray emission, implying either no obscured AGN or extremely high obscuration levels [Wilman et al. 1998]; the latter interpretation is favored by deeper observations which found evidence for Compton thick AGN in some (but not all) of these systems [Wilman et al. 2003]. Very recently, [Iwasawa et al. 2005] have used XMM to detect faint X-ray emission with a hint of the 6.4 keV Fe K line from F15307+3252, which they attribute to a Compton thick AGN with L(2 - 10kev) > 10⁴⁵ erg/s, which can account for a significant fraction of the infrared luminosity.

The picture that emerged from ISO, therefore, was one where very IR-luminous galaxies become substantially more numerous with increasing redshift, making z ~ 1 LIRGs and ULIRGs a cosmologically significant population. The 0.5 < z < 1.0 LIRG population seems quite similar to LIRGs at lower redshift, but also revealed a population of moderate redshift LIRGs with relatively cool dust emission. The higher redshift ULIRG population exhibits a more prevalent level of QSO activity than found in their local cousins, a result that could be either luminosity or redshift related since the most luminous ULIRGs are found at the higher redshifts.

5.3. Resolving the CIB: ULIRGs at z > 1.5

The CIB had another startling surprise in store. Though ISO had proven itself remarkably adept at resolving much of the CIB, there remained a significant shortfall between the CIB as measured by COBE, and the sources detected by ISO. The background levels at > 200 µm determined by FIRAS onboard COBE [Fixsen et al. 1998] implied a population of "colder" sources, probably z 1 systems with IR emission redshifted to longer wavelengths than the sources detectable in shallower, < 200 µm, ISO surveys.

Fortunately, the perfect instrument to search for just this kind of source had just been commissioned at the James Clerk Maxwell Telescope, namely the Sub-millimetre Common User Bolometer Array (SCUBA, [Holland et al. 1999]). SCUBA heralded the advent of bolometer arrays for sub-mm observations, and was one of the first sub-mm instruments that could map large areas of sky relatively quickly in the sub-mm; operating at substantially longer wavelengths than ISO, namely 850 µm and 450 µm, SCUBA had the sensitivity to detect these distant `cold' ULIRGs. Furthermore, the strongly negative k-correction at sub-mm wavelengths [Blain & Longair 1993] meant that a ULIRG at z=5 is (depending on the choice of cosmology) almost as easy to detect as a ULIRG of comparable luminosity at z=1. Other important sub-mm bolometer arrays include the 350 µm optimized SHARC-II camera and the 1.1mm Bolocam instrument, both at the Caltech Sub-mm Observatory, and the MAx-Planck Millimetre Bolometer (MAMBO) at the Institut de Radioastronomie Millimetrique (IRAM) 30m telescope, which operates at 1.2mm.

The advent of sub-mm array instruments therefore prompted a plethora of surveys try to find these distant "cold" sources, ranging from ultra-deep surveys of lensing cluster fields [Smail, Ivison & Blain 1997, Smail et al. 2002, Ivison et al. 2000] through deep blank-field surveys of small areas [Hughes et al. 1998, Barger et al. 1998], to wider area blank-field surveys to shallower depths [Eales et al. 2000, Scott et al. 2002, Borys et al. 2003, Greve et al. 2004, Laurent et al. 2005]. These surveys found a huge population of sub-mm bright, optically faint sources, with source counts of 320_-100⁺⁸⁰ and 180 ± 60 per square degree at S₈₅₀ > 10 mJy [Scott et al. 2002]. Taken together, the sources found in the various sub-mm & mm surveys can directly account for around 50% of the CIB detected by FIRAS, and with reasonable extrapolation, can account for all of it [Barger et al. 1999b, Blain et al. 1999b].

The first step was to determine whether these sub-mm sources really are distant ULIRGs, or some other class of object. If, as is very plausible, the sub-mm emission from distant galaxies can be modeled with a simple modified blackbody, where the dust temperature distribution is characterised by T, then the flux at frequency , F(), is given by F() B(, T), where is the spectral index (also referred to as the emissivity), and B(,T) is the Planck function. As a virtue of the shape of this function, sources with a redshift of z 1, a dust temperature of 25K, an 850 µm flux of S₈₅₀ > 1 mJy, and a reasonable choice of emissivity will, under most circumstances, be ULIRGs. Simple photometric redshift estimates [Bertoldi et al. 2000, Fox et al. 2002, Borys et al. 2004, Webb et al. 2003] place virtually all the sources from sub-mm surveys at z > 1, and most at z > 2, meaning that the sub-mm sources are likely to be extremely luminous ULIRGs, with star formation rates substantially exceeding 1000 M yr^-1. This implies that there are of order a few hundred ULIRGs per square degree at z > 1, compared to about one ULIRG every four square degrees locally; though contributing less than 1% of the total extragalactic background light locally, ULIRGs were therefore shown to be a major, perhaps even dominant contributor at z > 1.

Before proceeding further, there is an important cautionary note. A single sub-mm flux cannot be used to infer dust mass or source luminosity without assuming a dust grain size and temperature distribution [Hildebrand 1983], thus, for example, large cold disks can be confused with compact warmer starbursts of very different bolometric luminosity [Kaviani, Haehnelt & Kauffmann 2003, Efstathiou & Rowan-Robinson 2003, Farrah et al. 2004a, Almaini et al. 2005]. If the redshift is also unknown, a much lower luminosity foreground cold disk or even a Galactic dust cloud could easily mimic a high-redshift ULIRG [Lawrence 2001]. As we describe below, however, there is now good evidence that at least ~ 70% of sub-mm sources are indeed z > 1 ULIRGs, but within the remaining 30% there remains the possibility of significant sample contamination. Moreover, a strong temperature selection bias does exist in these very long-wavelength surveys; for example, a ULIRG at 2 < z < 3 with L_ir ~ 10¹³ L and a dust temperature of 60K would not be detected in the wide, shallow surveys with SCUBA [Blain et al. 2004b]. Such "hot" sources do appear to exist [Chapman et al. 2004b, Lutz et al. 2005], and could be responsible for up to 1/3 of the DIRBE detected CIB. Therefore, sub-mm surveys alone do not necessarily provide a hyaline view of dust-shrouded star formation, or even of ULIRG activity, at high redshift. Spitzer is proving to be extremely important for discovering the warmest high-redshift ULIRGs (see Figure 1), as we shall describe further below.

Figure 4. An example of a recent submm survey; in this case the HDF-N SCUBA "Supermap" [Borys et al. 2003]. circles = > 4, squares = 3.5-4D. The white line encompasses the [Hughes et al. 1998] SCUBA-HDF observations.

Due to the coarse angular resolution of SCUBA (and indeed most sub-mm & mm observatories up to now) and the optical faintness of most of these sources, determining reliable ID's for follow-up proved to be extremely difficult (see [Dunlop et al. 2004], for a perfect example). Nevertheless, since 1997, progress has slowly but surely been made in following up Sub-mm Galaxies (SMGs), and a clear picture of their nature has now started to emerge. A prime question is: how well do these distant systems resemble local ULIRGs? We would not necessarily expect them to be closely similar in nature because conditions are very different at high redshift, notably an earlier stage in the clustering evolution of the underlying dark matter, and larger gas fractions.

Optical and near-IR imaging [Smail et al. 1999, Webb et al. 2003, Smail et al. 2004, Pope et al. 2005] has shown that SMGs have a diverse range in optical properties, ranging from optically bright sources, to sources undetected in even the deepest optical imaging. Radio observations, particularly with the VLA, provided a major step forward in followup of SMGs [Ivison et al. 2002], detecting around 70% of the SMGs found in wide-field surveys with sufficiently high spatial resolution to allow reliable identification of optical/near-IR counterparts. Later, very high resolution radio studies have shown, intriguingly, that the radio emission in a surprisingly high fraction of SMGs, around 70%, is spatially extended on scales of ~ 10 kpc [Chapman et al. 2004a], significantly larger than the dust emission seen in most local ULIRGs, which are generally less than a kpc across [Downes & Solomon 1998]. The beam shape, especially when beam smearing is accounted for, is probably not well enough known to provide much evidence on the intrinsic shape of these radio sources, so possible explanations include large scale star formation over large disks, multiple compact starburst sites in a large disk or a coalescing group, or to AGN jet-induced star formation as seen in some distant radio galaxies [van Breugel et al. 2004]. Another class of z > 2 large scale systems inferred to have ULIRG-level luminosities are the so-called Ly blobs [Geach et al., 2005].

Early efforts to obtain spectroscopic redshifts for SMGs had been mostly stymied (but not entirely, see [Barger et al., 1999a]), not only by uncertain IDs and extreme optical faintness, but also because many SMGs lie within the so-called redshift `desert', 1 < z < 2, within which the bandpasses of most optical spectrographs could only sample weak absorption features, from which redshifts are hard to determine. But with high resolution radio ID's in hand, and significant tenacity, spectroscopic redshifts for SMGs were soon forthcoming [Chapman et al. 2005, Simpson et al. 2004]. These surveys showed that SMGs have a median redshift of 2.4, removing any last lingering doubts that the (radio detected) SMGs were indeed high-redshift ULIRGs. The optical spectra themselves show evidence for both starbursts and AGN in most sources, high unobscured star formation rates, and that SMGs are generally metal rich [Swinbank et al. 2004]. Very recent optical spectroscopic studies have focused on integral field spectroscopy, though instrumentation is only now becoming capable of integral field studies of distant SMGs. Early results however look very promising; the host galaxies of SMGs are metal rich, and appear to contain surprisingly large numbers of evolved stars and show evidence for ongoing interactions [Tecza et al. 2004, Swinbank et al. 2005].

With accurate redshifts it was possible to use heterodyne instruments (which generally have a very narrow bandpass, making spectroscopic redshifts essential for meaningful observations) to look at far-IR fine structure lines, particularly those of CO. CO surveys of SMGs [Genzel et al. 2003, Neri et al. 2003, Greve et al. 2005] revealed that they have enormous masses (often exceeding ~ 1 × 10¹¹ M) of molecular gas, usually unresolved but in some cases extended over 3-5 kpc [Genzel et al. 2003], much greater than those masses seen in local ULIRGs and comparable to those seen in high-redshift radio galaxies. HST imaging shows that SMGs often possess disturbed morphologies, consistent with (but not solely supportive of) ongoing major mergers, similar to the local ULIRG population [Smail et al. 1998, Conselice et al. 2003]. Very recent results based on deep X-ray observations have provided convincing evidence that the bulk of SMGs harbour a Compton thick AGN in addition to a starburst, implying that a sub-mm galaxy signposts a major period of growth of the central supermassive black hole in these systems, as well as an intense starburst [Alexander et al. 2005]. There is some evidence that the periods of black hole growth and stellar mass buildup in these systems may not be coeval [Borys et al. 2005], although there are uncertainties in the derivation of the X-ray obscuring columns, and therefore of the black hole mass [Polletta et al. 2006].

Further results have come from the first observations of SMGs with Spitzer. Results from IRAC and 24 µm imaging [Egami et al. 2004, Frayer et al. 2004] have shown that SMGs have a wide range of observed mid-IR colours, and that their rest-frame mid-IR SEDs can in most cases be fitted with a starburst SED template, with the remainder being well fitted by a power-law AGN-dominated template. Photometric redshifts estimated from the near-infrared hump peaking at rest-frame 1.6 µm and detected in the Spitzer IRAC bands (3.6-8.0 µm) span 1 < z < 3.5. [Ivison et al. 2004] observed 9 MAMBO 1200 µm sources in the same region, finding 75% of them to be starburst-dominated using Spitzer mid-infrared spectral shapes (assuming redshifts ~ 2.5), and concluding that the more AGN-like SEDs are consistent with AGN indications in these objects from UV/optical spectroscopy and X-ray imaging.

Spitzer will be a very powerful tool for finding large populations of ULIRGs at z > 1.5, with many hundreds of candidates appearing in the Spitzer surveys such as SWIRE [Lonsdale et al. 2006], the Bootes Shallow field, the First Look Survey, the GTO deep surveys and GOODS (see Figure 1). Moreover Spitzer complements the cold dust selection function of the sub-mm and mm surveys with its mid-IR selection function and enhanced sensitivity to warm systems.

The first source counts and luminosity functions at 24 µm, 70 µm and 160 µm from Spitzer extend the results seen by ISO. Spitzer is revealing large numbers of systems with significantly more extreme infrared/optical luminosity ratio than seen in the local Universe [Rowan-Robinson et al. 2005], a result also known from moderate redshift ISO surveys ([Rowan-Robinson et al. 2004; Oliver & Pozzi 2005]). [Rowan-Robinson et al. 2005] also find a significant population of very cool luminous Spitzer systems, which could be very large disks with quiescent star formation rates rather than starbursts. The 24 µm counts in the GOODS fields [Papovich et al. 2004] show evidence for strong evolution, exceeding the `no evolution' counts predictions by a factor of at least 10, and implying evolution in the comoving IR energy density of the form (1 + z)^{3.9 ± 0.4} up to z ~ 1 [Le Flóch et al. 2005]. The Spitzer counts at 70 µm and 160 µm [Dole et al. 2004] directly resolve 20% and 7% of the CIB at these wavelengths, respectively, and also show strong evolution, by a factor of ~ 3 over no evolution models, implying that the galaxies responsible for this background mostly lie in the redshift range 0.7 < z < 0.9. [Pérez-González et al. 2005] used find that LIRGs and ULIRGs are increasingly important contributors to the infrared energy density as redshift increases to z ~ 3, being responsible for half of all star formation by z ~ 1.5. The characteristic luminosity of the luminosity function, L^*, increases steadily with z, consistent with the cosmic star formation rate density going as (1 + z)⁴ to z = 0.8, then flattening somewhat, and with ULIRGs playing a rapidly increasing role above z = 1.3. [Le Flóch et al. 2005] came to similar conclusions out to z ~ 1, and also compared the UV and IR star formation history since z = 1 using deep observations from Spitzer and GALEX, finding that the SFH has evolved much more strongly in the IR than in the UV, thus confirming the ISO-based results of [Rowan-Robinson et al. 1997, Flores et al. 1999, Pozzi et al. 2004].

[Yan et al. 2005] and [Houck et al. 2005] have reported the first Spitzer low resolution IRS spectroscopy for extreme IR/optical z ~ 2 ULIRG candidates, demonstrating that Spitzer is capable of determining mid-infrared redshifts for the brightest mid-infrared galaxies (f₂₄ > 0.75 mJy). The space density of L_ir > 10^12.3 L z ~ 2 Spitzer mid-infrared-selected ULIRGs may be similar to that of SMGs (Yan et al.) at this redshift, though, the sample sizes are as yet small and incomplete. Yan et al. infer a higher fraction of starburst-dominated ULIRGs in their sample than Houck et al., which is probably due to the additional selection criterion on high 24/8 µm colour (ie a rejection of mid-IR warm sources). [Lutz et al. 2006] report a low 1.2mm detection rate these systems using MAMBO, and suggest that their mid-infrared-selected sample may have significantly warmer dust than submm-selected samples. [Franceschini et al. 2004, Martinez-Sansigre et al. 2005, Donley et al. 2005 and Polletta et al. 2006] have used Spitzer surveys to discover extensive populations of heavily obscured AGN populations that can have very high X-ray obscuring columns, including two z > 2, Compton-thick SWIRE HLIRGs with AGN torus-dominated mid-IR SEDs [Polletta et al. 2006]. These new Spitzer results indicate that in many ULIRGs highly obscured QSOs are extremely difficult to detect at any wavelength, including hard X-ray and mid-infrared, due to extreme optical depths.

[Daddi et al. 2005] have reported the first Spitzer results from the Great Observatories Origins Deep Survey (GOODS) for z ~ 2 ULIRGs, finding that typical massive (M_stellar ~ 10¹¹ M) star forming galaxies at this redshift are ULIRGs, based on 24 µm and radio detections, with a co-moving space density of ULIRGs at z = 2 at least 3 orders of magnitude greater than the local one.

In summary it would be fair to say that surveys with SCUBA and other sub-mm/mm observatories, and their followup, have revolutionized our perceptions of ULIRGs, while Spitzer stands poised to contribute its own revolution of knowledge about them. From being little more than an interesting oddity in the local Universe, ULIRGs have become a crucially important population at z > 1. Furthermore, whilst z > 1 ULIRGs are similar in many ways to their local counterparts - both populations apparently being heavily dust-obscured starbursts and/or AGN triggered by interactions between large, evolved systems - there are important differences. High-redshift ULIRGs are substantially more gas rich than their local counterparts, and some may have extended rather than compact starbursts. Moreover the star formation rates implied (if AGN do not strongly dominate the energetics) are enormous, pushing the limits on theoretical ideas for a "maximal" starburst in a massive galaxy: M_gas / t_dyn = 10¹¹ M / 10⁸ yrs ~ 1000 M/yr [Eggen, Lynden-Bell & Sandage 1962]. To understand these differences, and place these high-redshift ULIRGs into a cosmological context, we must turn to theories of galaxy and large-scale structure formation.

⁷ Though even at this stage source count models that could explain the CIB all invoked a population of high-redshift IR-luminous starbursts [(Guiderdoni et al. 1998, (Blain et al. 1999a, (Rowan-Robinson 2001, (Xu et al. 2001] Back.