|Annu. Rev. Astron. Astrophys. 2014. 52:
Copyright © 2014 by Annual Reviews. All rights reserved
4.1. Gravitational lenses as detailed probes of the interstellar medium at high redshift
Gravitational lensing magnification by individual foreground galaxies or by clusters of galaxies permits detailed studies that are impossible for equivalent sources in blank fields, due to sensitivity limitations. Starting with the detection of IRAS F10214+4724 in a redshift survey of the IRAS catalog (Rowan-Robinson et al. 1991), a handful of lensed z > 2 sources have become accessible for in-depth infrared to millimeter studies. The first few such objects were detected through a variety of techniques that often selected systems hosting powerful AGN. Early on, a number of studies realized that large area surveys at submillimeter and millimeter wavelengths will provide an efficient and less biased method for detecting lensed dusty galaxies at high redshift (e.g. Blain 1996, Negrello et al. 2007). At these wavelengths, the combination of the steep decay of unlensed source counts towards bright fluxes and the approximate constancy of source flux with redshift due to the negative K-correction on the Rayleigh-Jeans side of the far-infrared SED ensures that a significant fraction of bright sources in a large area survey will be lensed high redshift galaxies, rather than local objects. This differs from optical and X-ray wavelengths, where lensed objects are relatively faint compared to the multitude of foreground sources, due to positive K-corrections.
Herschel-SPIRE surveys and ground based South Pole Telescope (SPT) mm surveys have now covered about 1000 and 2500 square degrees respectively, sufficient for application of these methods. Flux cuts at S500 μm = 100 mJy or S1.4 mm = 20 mJy produce samples that contain candidate lensed dusty galaxies at the tens of percent level. Separating these from radio galaxies, low redshift galaxies, and galactic dust clouds at the same flux level is relatively straightforward, based on (sub)mm colors as well as readily available radio and optical data (Vieira et al. 2010, Negrello et al. 2010), see also Marsden et al. (2013) for a search from Atacama Cosmology Telescope data. Hundreds of lensed galaxies are expected to be present at these levels in the existing surveys, given surface densities of candidate lensed galaxies of 0.14 to 0.26 ± 0.04 deg-2 at S500μ m > 100 mJy (Wardlow et al. 2013a, Bussmann et al. 2013) and ~ 0.1 deg-2 at S1.4 mm > 20 mJy (Vieira et al. 2010). Refined selection methods including color information may push the number of candidates to more than a thousand (González-Nuevo et al. 2012). Related searches for somewhat fainter sources with extremely red SPIRE colors (Dowell et al. 2014) have the potential of uncovering highest redshift dusty star forming galaxies (e.g. Riechers et al. 2013), likely with a smaller fraction of lenses. Firming up the census of z > 4 sources from these searches may tighten the contraints that very high star formation rate high-z sources place on models of galaxy evolution, beyond those known from the z < 4 SMG and Herschel populations.
Most redshifts of these lensed dusty star forming galaxies have been obtained from direct CO detections. CO lines (or alternatively rest frame mid-infrared features observable with space telecopes) are quite closely linked to the rest frame far-infrared emission that defines these targets. Their direct detection largely avoids the effort, mis-identification risks and biases that are inherent to the traditional way of obtaining first a radio- or mm-interferometric accurate position, then identification in the optical or near-infrared, and finally an optical spectroscopic redshift of the counterpart. The high flux of these lensed objects and the recent availability of wideband spectrometers at both single dish telescopes and interferometers have contributed to a rather rapid buildup of redshift information (e.g. Negrello et al. 2010, Frayer et al. 2011, Cox et al. 2011, Scott et al. 2011, Combes et al. 2012, Harris et al. 2012, Lupu et al. 2012, Weiß et al. 2013, Bussmann et al. 2013). Spectroscopic redshift completeness well exceeds 50% for the best studied lensed samples.
Herschel- and SPT-detected systems include some of the highest redshift z ~ 6 dusty star forming objects that are currently known and do not host a QSO (Riechers et al. 2013, Weiß et al. 2013), and have already been followed up in a variety of other tracers of the warm and dense interstellar medium (e.g., Omont et al. 2011, Cox et al. 2011, Omont et al. 2013, Rawle et al. 2013, Bothwell et al. 2013). Corrected (where known) for the lensing magnification they typically show very high star formation rates ranging from several hundred to few thousand solar masses per year (e.g. Negrello et al. 2013), including extreme cases that reach ≳ 2000 M⊙ yr-1 and are single galaxies or physically associated pairs according to interferometric followup (e.g. Fu et al. 2012, Riechers et al. 2013, Fu et al. 2013, Ivison et al. 2013). These last cases have passed the detection thresholds simply by their huge luminosities, with only weak amplification suggested by the detailed followup (e.g. Fig. 10). In their intrinsically large luminosities (LIR ~ 1.5 × 1013 L⊙), compact sizes (Rhalf,880 μm ~ 1.5 kpc) and warm dust temperatures (Tdust ~ 39 K), the lensed sources studied by Bussmann et al. (2013) resemble high luminosity blank field SMGs, with the effect of lensing biases still under study.
Figure 10. Followup of bright Herschel 500 μm or SPT millimeter sources uncovers a large fraction of lensed high redshift galaxies. Properties range from only weakly magnified, intrinsically extremely luminous targets as in the example of the weakly amplified galaxy pair HXMM01 (Figure reproduced from Fu et al. 2013), to strongly magnified targets. The left panel shows a SPIRE three color image of HXMM01, the right panel high resolution follow up in near-infrared stellar continuum, submm dust continuum and JVLA CO emission.
Placing these sources in the context of other galaxy evolution studies is slowed down by the need to construct reliable lensing models on the basis of < 1" resolution data either from (sub)mm interferometry, or near-infrared imaging from HST or adaptive optics, even if methods for crude magnification estimates on the basis of line widths have also been proposed (Harris et al. 2012). Of the ~ 30 published lensing magnifications, many are based on ~ 0.5" resolution data, coarser than desirable given the likely presence of differential lensing effects especially for objects that are strongly amplified by galaxy lenses or near the caustics of cluster lenses. In a statistical sense, lenses from flux limited wide area submm surveys prefer intrinsically compact sources (Hezaveh, Marrone & Holder 2012) with comparison of the expected magnification distributions to observations going on (Bussmann et al. 2013). This size bias will also affect the ISM conditions. The difficulty of separating lens and lensed object at optical/near-infrared wavelengths severely limits the number of lensed dusty star forming galaxies from wide area searches for which key host parameters such as the stellar mass are currently known. Cases with published stellar mass (Fu et al. 2012, 2013, Negrello et al. 2013, see also Swinbank et al. 2010) mostly lie above the main sequence, resembling bright blank field SMGs also in that respect.
4.2. UV selected galaxies
Over the last two decades, galaxies selected in the rest frame UV via either broad band photometry and the Lyman break dropout technique (Lyman break galaxies, LBGs) or by detection as Lyman α emitters (LAEs) in narrowband searches have played a pivotal role for our understanding of z ≳ 3 galaxy evolution. An important uncertainty concerns the obscured fraction of their star formation rate. One typical way of deducing obscuration (often expressed as the infrared excess IRX = LIR / LUV) is from the slope β of the rest frame UV continuum (fλ ∝ λβ), applying locally calibrated relations (e.g. Meurer, Heckman & Calzetti 1999). Alternatively, SED fitting techniques are applied to the same plus longer wavelength photometric information. Direct detection of rest frame far-infrared or submillimeter emission from these systems turned out to be difficult (Chapman et al. 2000). Concerning typical members of the population, such detections have been restricted mostly to a few lensed systems (e.g. Baker et al. 2001, Siana et al. 2009).
4.2.1. LYMAN BREAK GALAXIES The direct detection of typical blank field LBGs remains out of reach of Herschel data. In large samples, a rather small fraction is individually detected but biased towards LBGs that are more massive and have redder rest frame UV colors, for both GALEX-selected 0.7 < z < 2.0 LBGs (Burgarella et al. 2011, Oteo et al. 2013c and for z ~ 3 LBGs (Oteo et al. 2013b). These Herschel-detected LBGs have large IR luminosities LIR > 1011 … 1012 L⊙. For larger z ≳ 3 LBG samples (sometimes pre-selected for larger expected SFR), even stacking analyses are just able to sometimes achieve detections in Herschel or groundbased (sub)millimeter data (Magdis et al. 2010, Rigopoulou et al. 2010, Lee et al. 2012, Davies et al. 2013). The SED information that can be extracted is limited given the modest S/N of the stacks, but in some cases (unless affected by sample contamination) suggests moderately cold dust temperature (Tdust ~ 30-40 K, Lee et al. 2012, Davies et al. 2013). Perhaps the best SED constraints for z ≲ 3 LBG-like galaxies come from small numbers of gravitationally lensed systems, where Saintonge et al. (2013) find warm dust temperature reaching Tdust ~ 50 K (see also Sklias et al. 2013). There remains uncertainty concerning the rest frame far-infrared SEDs of z > 3 LBGs. This implies limitations for interpreting detections that ALMA will obtain on the long wavelength tail of the SED, in terms of the balance of obscured and unobscured star formation.
4.2.2. THE IRX-β RELATION AT HIGH REDSHIFT In the absence of infrared data, obscured star formation has to be quantified from the rest frame UV emission. Discussing and testing at high redshift the derivation via the IRX-β relation mentioned above is of relevance also for more complex SED fitting technigues that are making use of the same photometric information and similar underlying concepts. An empirical IRX-β relation such as the classical one of Meurer, Heckman & Calzetti (1999) for local UV-selected starbursts (and many revisions since) encodes the properties of the unobscured stellar population such as star formation history and metallicity, and an attenuation law. The attenuation law is sensitive not only to the properties of the obscuring dust that determine the classical extinction curve, but also to the geometric configuration of stars and obscuring dust. Inevitably, stellar emission and dust obscuration in a galaxy will be spatially mixed to some extent. In this approach, however, attenuation is formally treated as a foregound screen.
Several papers have used Herschel data to test the IRX-β relation outside the local universe. A main finding is the wide spread in the IRX-β plane, also depending on the specific sample studied, which makes it impossible to distill a unique and reliable recipe to estimate the obscured star formation. A sample of z < 0.3 250 μm selected galaxies (Buat et al. 2010) is found to spread from roughly the IRX-β relation for local starbursts to that of local normal star forming galaxies, with IRX varying by at least an order of magnitude at given β. A similarly large spread is observed for 1 < z < 2.5 PACS detections (Buat et al. 2012, Nordon et al. 2013), and in a number of other studies (Wijesinghe et al. 2011, Oteo et al. 2012b, 2013b, 2013a).
Analysing Herschel stacks of individually undetected z ~ 2 and z ~ 1.5 UV selected galaxies, Reddy et al. (2012) and Heinis et al. (2013) find that the IRX for such stacks increases with β, similar to the trends for local galaxies. There is, however, a noticeable factor > 2 difference between these two studies in the absolute IRX level at given β, reflected in a better match to local starbursts (Reddy et al. 2012) vs. a better match to local normal star forming galaxies (Heinis et al. 2013). A similar spread may be indicated in the 3.3≲ z ≲ 4.3 stacking study of Lee et al. (2012).
Buat et al. (2010) report a decrease in IRX for z ~ 1 galaxies compared to local galaxies of the same infrared luminosity. Nordon et al. (2013) observe that the offset of 1 < z < 2.5 PACS galaxies from a mean IRX-β relation correlates strongly with the offset from the star forming main sequence, with larger IRX above the main sequence. This finding likely relates to the offset from a standard IRX-β that is found for local dusty ULIRGs (Goldader et al. 2002) as well to the connection that has been found locally between the IRX-β relation and the stellar birthrate parameter (Kong et al. 2004). UV obscuration properties may be closely connected to the position of a galaxy with respect to the evolving main sequence, in analogy to the infrared SEDs (Section 3).
In their detailed study combining short wavelength and Herschel data, Buat et al. (2012) report detection of a 2175&$197; UV bump in the attenuation curves for 20% of their sample, as well as a diversity of attenuation law slopes. Both findings might relate to variations between the galaxies in dust geometry as well as dust properties.
Obscuration of star forming galaxies (IRX) increases with stellar mass (e.g. Wuyts et al. 2011b, Buat et al. 2012, Heinis et al. 2014). Grouping galaxies detected in both IR and UV by UV luminosity yields a trend in the sense that the most FUV luminous galaxies avoid large IRX (Burgarella et al. 2011, Buat et al. 2012, Oteo et al. 2013c). Selection as well as evolution may play a role here, and no obscuration trend is obvious in stacks of z ≲ 2 UV-selected galaxies that are grouped by LUV (Reddy et al. 2012, Heinis et al. 2013). Some decrease of IRX towards high LUV may be present for stacks of z ~ 3-4 UV selected galaxies (Heinis et al. 2014).
In summary, there is a large spread in the IRX-β properties of Herschel detected galaxies. Combined with the tendency for underpredicting the SFRs of the most star forming galaxies when using rest UV-optical SED fitting (Wuyts et al. 2011a) or a local IRX-β relation (e.g. Oteo et al. 2013b, 2013a), this finding supports to use a hierarchy of SFR indicators, where SFR is based on IR (plus unobscured UV where available) for the most heavily star forming systems detected with Herschel or in the mid-infrared, and SFR is based on rest frame UV to NIR SED fitting for lower star formation rate systems (Section 2.2).
4.2.3. LYMAN α EMITTERS Detection of dust continuum emission from z ≳ 5 Lyα emitters is firmly in the regime requiring ALMA, although an initial upper limit indicates a complex picture (Ouchi et al. 2013). At lower redshifts, insights can be gained from single dish submillimeter telescopes or Herschel. A small subset of z ~ 2-3.5 LAEs is detected by Herschel (e.g. Bongiovanni et al. 2010, Oteo et al. 2012a), in another z ~ 2.8-4 study there are no individual detections (Wardlow et al. 2013b). Finding a few LAEs to be IR-bright is in line with the earlier realisation of significant Lyα emission from many SMGs (Chapman et al. 2005). These individual identifications with dusty star forming galaxies form a biased subset of LAEs, though, calling for stacking analyses or study of local analogs. At z ~ 0.3, Herschel detection rates of GALEX-selected Lyman α emitters are more favorable (Oteo et al. 2011), and LAEs are found to be biased to lower obscuration compared to similarly UV-luminous systems that are lacking Lyman α emission (Oteo et al. 2012b). In all properties, classical LAEs selected from narrow band Lyα searches should be contrasted with the extremely IR-luminous dust-obscured population that was recently retrieved from a WISE selection by Bridge et al. (2013), which also exhibits prominent and extended Lyα emission.
4.3. Submillimeter galaxies from ground-based surveys
Groundbased 850 μm and millimeter observations provided the first detections of luminous z ~ 2 infrared galaxies. These `submillimeter galaxies (SMGs)' continue to be of interest, with ever larger samples being obtained with improving detector technology at ground-based single dish telescopes. Reproducing such a significant high redshift population with star formation rates reaching the regime of 1000 M⊙ yr-1 has long been known to be a challenge to theoretical models of galaxy evolution (e.g. Baugh et al. 2005, Davé et al. 2010). Since SMG luminosities traditionally had to be extrapolated from the submillimeter and radio or from the mid-infrared using locally calibrated relations, confirmation of these large luminosities is important.
For the difficult task of identifying short wavelength counterparts to SMGs, Herschel photometric surveys can provide better spectral continuity between the detection wavelength and mid- and near-infrared counterparts (Dannerbauer et al. 2010). But given Herschel beamsizes, they cannot replace radio and in particular (sub)mm interferometry for reliable identification and for characterizing the fraction of single dish detections that breaks up into several sources, either physically associated and possibly interacting galaxies, or chance projections (e.g. Hayward et al. 2011, Barger et al. 2012, Smolcic et al. 2012, Karim et al. 2013, Hayward et al. 2013).
PACS and SPIRE studies sample the SED peaks of bright z ≲ 4 SMGs with good S/N (Magnelli et al. 2010, Chapman et al. 2010, Magnelli et al. 2012a, Swinbank et al. 2013). These results strongly confirm the dust temperature selection effects that ground-based SMG samples suffer at given IR luminosity and redshift. In particular for z ~ 1, LIR ≳ 1012L ⊙ galaxies, an object has to be colder than representative for bolometrically selected samples, in order to be detectable in a typical groundbased submm survey. These selection effects are less severe for the most luminous z ~ 2 SMGs. Herschel data also confirmed that a large fraction of previously known `optically faint radio radio galaxies' (OFRGs) with radio continuum indicative of strong star formation but nondetection in submm surveys are indeed strongly star forming objects but with warmer dust temperatures (Magnelli et al. 2010, Chapman et al. 2010). A further confirmation of the dust temperature biases in the SMG population stems from the comparison to galaxies selected solely via the stellar rest frame 1.6 μm bump. These extend to warmer dust temperatures than SMGs, at same redshift and IR luminosity (Magdis et al. 2010).
The huge luminosities of SMGs previously extrapolated from other wavelengths are broadly confirmed by Herschel. As for other high redshift galaxies (Section 2.2) IR luminosities estimated from 24 μm emission via locally calibrated luminosity-dependent templates are too high and show large scatter (Magnelli et al. 2012a). Luminosity estimates based on radio and submm flux, redshift, and adoption of the local radio-far-infrared correlation do better, but also require downward revision for the SMG sample with spectroscopic redshifts that was studied with Herschel by Magnelli et al. (2012a). It would however be premature to interpret this as evidence for evolution of the radio-far-infrared correlation, since these particular SMGs will be somewhat biased to high radio fluxes due to the role radio interferometry plays in the traditional SMG identification and redshift determination process. In line with this, the small SMG sample of Barger et al. (2012) with full submm and radio identification is consistent with the local radio-FIR correlation.
A picture emerges in which traditional SMG samples that are selected over just a small S850 ~ 3-10 mJy range are in fact rather heterogeneous (e.g., Magnelli et al. 2012a). Such samples cover two orders of magnitude in total IR luminosity, range from objects near the star forming main sequence to more than an order of magnitude above it (but note discussion on SMG stellar masses, e.g. Michaowski et al. 2012), and cover almost a factor of three in dust temperature. In studying the detailed nature of SMGs and their interstellar medium, it is hence key to differentiate these groups rather than globally refer to `SMGs'. A z ≳1 LIR ~ 1012 L⊙ sample (as, e.g., studied in [OI] 63 μm by Coppin et al. 2012) will almost certainly differ in its properties from ten times more luminous galaxies at z ~ 2.5, despite sharing detection in a ground based SMG survey. This caveat will become ever more important with improving single dish submillimeter surveys and with ALMA detections of significantly fainter objects at similar wavelengths. The heterogeneity of the SMG population is further emphasized by the analysis of Hayward et al. (2013). They use a semi-empirical model based on evolving stellar mass functions and scalings for star formation rate and gas fractions, combined with hydrodynamical and radiative transfer simulations for isolated disks and mergers, to model the SMG population. Number counts are fit without a need to invoke a top-heavy IMF. At faint fluxes S850 ~ 1 mJy, isolated normal star forming disks and pairs that are blended by the large single dish beam dominate somewhat expectedly. But isolated disks and blended pairs still do so at S850 ~ 3 mJy, and even at the S850 ~ 10 mJy bright end, only 2/3 of the SMGs are suggested to be luminous mergers. In going from single dish to interferometric studies, number counts should then be reduced by as much as a factor 2 as blended pairs break up, as also suggested by recent observational studies (Barger et al. 2012, Smolcic et al. 2012, Karim et al. 2013).
4.4. Dust-obscured galaxies
A considerable portion of Spitzer observations and follow up has been devoted to a population of z ~ 2 dust-obscured galaxies (DOGs), defined by a high ratio of mid-infrared to rest UV (observed R band) flux density S24 μm / SR ≳ 1000 (e.g. Dey et al. 2008). At the S24 μm ≳ 1 mJy bright end, these samples include a large fraction of obscured AGN, while at fainter levels often obscured star formation is suggested, from the presence of a rest near-IR stellar bump rather than an AGN power law continuum.
Several studies address the far-infrared properties of this population and are in broad agreement with this general picture (see also the related discussion of a bright 24 μm selected sample by Sajina et al. 2012). Herschel detects half of the S24 μm > 100 μJy z ~ 2 dust obscured galaxies, at average infrared luminosity 2.8 × 1012 L⊙ (Calanog et al. 2013). Sources with bump rest frame near-IR SED require a star formation dominated overall SED, while sources with power-law near-IR SEDs are better fit with SEDs such as that of Mrk 231, including a strong AGN component (Melbourne et al. 2012). For a sample of dust obscured galaxies that emphasises the faint mostly star forming end of the population (median S24μm = 161 μJy), Penner et al. (2012) analyse the far-infrared to mid-infrared flux ratio. This ratio is found to be similar as in other dusty star forming galaxies at these redshifts, suggesting that the large mid-IR to rest UV ratio is not due to an extra mid-infrared component but due to atypically large obscuration of the star forming regions in the rest frame UV.
The WISE color-selected population with strong Lyα emission studied by Bridge et al. (2013) appears to form a yet more extreme end of the dust obscured population, with extreme LIR > 1013 - 1014 L⊙ luminosities and warm dust temperatures, and likely a dominant AGN.