An incredible amount of energy has been spent trying to understand ULIRGs in the decade since the [Sanders & Mirabel 1996] review, and this review has not been able to do adequate justice to all this effort. To summarize the current state of knowledge we address the a few of the most pressing questions from our own perspective.
In the relatively local Universe, ULIRGs are almost all associated with major mergers of gas rich systems. Evolutionary schemes whereby ULIRGs transition into optical QSOs at the end stages of a merger, with an elliptical as the end result, are highly attractive and well supported by simulations. The reality is obviously complex, however, depending strongly on the details of the merger encounter, in particular bulge mass. Observationally it is not necessarily the case that all ULIRGs represent late stages in the merger sequence, however the most luminous systems do seem to represent the late stages, implying that it requires the extraordinary high density conditions of the final merger stages to produce these tremendous levels of power. Evidence that some optical QSOs may have more massive hosts than ULIRGs, or inhabit richer environments, would indicate that the true situation is more complex, but matching AGN/ULIRG samples for such studies is very challenging.
Merger sequences connect starbursts and AGN tightly together via the availability of fuel to build the stellar and SMBH masses, and to fuel the AGN. AGN orientation schemes also tie ULIRGs and QSOs tightly together since off-axis QSOs will be obscured and may appear as ULIRGs. Therefore the empirical determination of the relative power of AGN and starbursts for both infrared-selected ULIRGs and optically-selected ULIRG QSOs is a paramount question. Most ULIRGs are thought to be powered mostly by starbursts however indicators of AGN activity increase with luminosity, recognised from near- and mid-infrared spectroscopy, X-ray detections or VLBI radio imaging. All of these indicators are tricky however and, because nuclear (Starburst) and gravitational accretion (AGN) power often compete for dominance we need more robust estimators of the SFR and AGN activity to compare with merger stage, morphology, mass, etc. to disentangle an evolutionary sequence from other effects. For example, low/high-resolution mid-IR spectroscopic indicators can be obscured; Spitzer will do an excellent job of measuring diagnostic features from 5-40 µm, but interpretation will require an understanding of extinction as a function of wavelength throughout the spectrum as well as modeling with full radiative transfer treatment. X-ray columns can be high enough to severely limit AGN detection; separating soft (starburst or reflected AGN) vs hard (AGN or X-ray binary) emission will require sensitivity at higher energies with Con-X and XEUS. High-resolution radio imaging has to be capable of mapping at brightness temperatures above 106K to distinguish jet structures from RSN complexes, because starburst and radio-quiet AGN emission cannot be distinguished on the basis of power alone. But we do not yet even understand the complex radio supernovae in the one, closest, well studied system, Arp 220, where questions remain about the nature of the LRSN, the IMF in the luminous starburst, etc. Such imaging observations are time consuming, but need to be expanded to much larger samples. One possible line of inquiry may be to obtain BH mass estimates from the molecular gas kinematics with ALMA (r ~ 20 pc) as has been done for NGC 4258 in H2O. Other questions remain, eg.: What are the compact masers telling us? Are they pointers to AGN/mass concentrations?
ULIRGs increase dramatically in importance with redshift, from being a very minor component at z ~ 0 to possibly dominating the energy density at z > 2, and ULIRG populations at z>2 are much more extreme beasts than at low redshift, with larger IR luminosities, gas masses and, in some cases, larger ratios of infrared-to-optical luminosity than local ULIRGs, by factors of 10 to 100. They also show an increase in space density of a factor of 100-1000 [Chapman et al. 2005, Le Flóch et al. 2005, Pérez-González et al. 2005], though this number is very uncertain because the low and high redshift LFs barely overlap in luminosity range; at low redshift the number density of HLIRGs is intrinsically extremely small, while at high redshifts, submm surveys are only sensitive only to systems above a few × 1012 L. Wide area Spitzer luminosity functions can be expected to bridge this gap ([Lonsdale et al. 2006b, Babbedge et al. 2006]). There is clear evidence that at high redshift many ULIRGs are triggered by mergers of gas rich systems, but there are also some striking differences from local systems: extended radio emission and some evidence for very large cool disk systems, which may not even qualify as starbursts, but rather star formation ocurring in a more-or-less quiescent fashion in these very large disks.
As noted above, the incidence of AGN seems to increase with luminosity, and this is also a strong redshift effect since the most luminous systems are only known at higher redshifts. As also noted above, it remains remarkably difficult to determine the relative power contribution from star formation vs. AGN activity in many ULIRGs. Many lines of argument are used to conclude that star formation powers most ULIRGs at all redshifts. These include the energy reservoir available from nucleosynthesis relative to SMBH accretion; low AGN power deduced from absorption-corrected hard X-ray detections; the number of obscured AGN required to explain the XRB and the correlation between bulge and SMBH masses which suggest bulge growth events must accompany SMBH growth. Arguments for a stronger importance for AGN power include that luminosities above 1013 L are very extreme for a starburst event while many high-redshift QSOs are known with such luminosities; a large number of QSO-powered ULIRGs is expected in AGN unification-by-orientation schemes; hard X-ray absorption correction is difficult, and Compton thick AGN such as Mrk 231 are very difficult to detect at high redshift.
Perhaps one of the strongest lines of evidence one way or the other comes from the impressive correlation between warm dense molecular gas traced by the HCN molecule with infrared luminosity [Gao & Solomon 2004], which extends across 4 orders of magnitude in luminosity from 109 to 1013. There's little doubt that low luminosity systems are powered by disk star formation, so this relation provides support for the picture that even HLIRGs can be powered by star formation. HCN ULIRG detections are limited at high luminosity/redshift, however, and they tend to be displaced above the lower luminosity redshift relation towards excess Lir / MHCN [Carilli et al. 2005], which is consistent with the presence of an an additional, AGN, power source but is also within the scatter of the lower redshift starburst relation. Spitzer will be able to provide deep insights into this debate because the presence of a significant AGN power source will be evident the mid-infrared spectral shape, as demonstrated by the recent flurry of papers reporting the discovery of large populations of highly obscured QSO/ULIRGs [Martinez-Sansigre et al. 2005, Polletta et al. 2006, Yan et al. 2005, Houck et al. 2005, Donley et al. 2005].
The causes for the dramatic differences in ULIRG number density and nature since z ~ 3 must lie with the evolving matter density field and galaxy formation processes. Since interactions and mergers are clearly implicated in most ULIRG activity, the most natural assumption to make is that these objects follow 'the action'. They are, after all, the most dramatic active events that occur in the Universe, the prima donnas of the show. At a given epoch they are likely to occur where the most dramatic gas rich mergers are happening, as illustrated in the simulation of K. Nagamine (private communication) in Figure 5. At the earliest epochs of galaxy formation the action is at the sites of future rich clusters, while at current times loose groups are ideal future ULIRG sites since there are a few close galaxies with moderate relative velocity, and gas has not been exhausted yet. At intermediate redshifts, the best ULIRG sites may be the interaction zones in cluster-cluster mergers. These ideas are consistent with the sparse currently available observations of ULIRG environments at various redshifts.
Figure 5. An example of the cosmological hydrodynamic CDM simulation (SPH G6 run) described in Nagamine et al. [2005a, b]. Each panel has a comoving size of 143 Mpc on a side, and the starforming galaxies with instantaneous SFR > 100 M / yr at each epoch are indicated by the circles (K. Nagamine, priv. comm.)
So what are the key unanswered questions and where do we go from here to answer them? So far ULIRG samples are quite sparse - numbering in the hundreds over a huge redshift range. We clearly need much larger, volume-limited samples. Moreover there are serious selection biases since high-redshift systems have been selected at sub-mm wavelengths by their cool dust, and substantially warmer systems will have been missed. Therefore we need to understand the full range of ULIRG SEDs and obtain complete samples with well measured SEDs via selection from the mid-IR to the sub-mm. Spitzer will clearly revolutionize this field, bringing in the warm-samples to complement the sub-mm ones, but also selecting PAH-dominated objects as the mid-IR PAH features redshift into the 24 µm band. Larger, deeper sub-mm & mm surveys are also required to reach down the luminosity function at z>1 and combat the effects of cosmic variance.
To complement the ULIRG samples it's obviously important to study complete QSO samples in the same volumes of space and to determine the lifetimes of both types of object, to directly investigate their evolutionary relationships. A tricky aspect to this work will be the fact that ULIRG and QSO luminosities are predicted by the most recent simulations to vary very dramatically on short timescales (as are SMBH and stellar masses), especially as the SMBH building and AGN accretion phase accelerates, so matching samples by luminosity or mass is probably going to match apples and oranges.
Key to understanding the role of ULIRGs in galaxy formation - their connections to other kinds of systems, their progenitors and their descendents, the connections between bulge and SMBH building and the relative importance of starburst vs quiescent star formation modes - is to determine the host galaxy properties of ULIRGs and QSOs, especially stellar masses and morphologies, and to determine the richness of their environments and their clustering properties as a function of redshift.
The future holds great promise for answering these questions. Measuring the clustering of ULIRGs requires wide area far-IR/sub-mm surveys, which are or will be forthcoming from Spitzer, SCUBA-2, APEX, ASTRO-F, Herschel and WISE. These surveys will also provide the large samples of ULIRGs needed to measure the range in dust temperatures in the distant ULIRG populations, with detailed followup provided by other groundbased facilities. Measuring sizes, morphologies, and stellar masses, assessing the impact of gravitational lensing and determining environmental richness, on the other hand, require deep pointed observations in the near-IR to radio. Here the key facilities are Spitzer, JWST, ALMA, the EVLA and SKA. Imaging in molecular and other submm/mm lines will also be possible with ALMA.