The complexity of the ULIRG phenomenon requires a multi-wavelength approach and we have learned an enormous amount from detailed studies at many wavelengths. For this review we focus on those observational areas which we believe are key to understanding the ULIRG phenomenon, in part because they represent the spectral regions of lowest optical depth, and in which key advances have been made in the years since the Sanders & Mirabel review. We stress, however, that even the relatively low-optical depth radio, mid-infrared, and X-ray regimes may still show substantial obscuration due to starburst-related free-free absorption [Condon et al. 1991], compact nuclear molecular clouds or molecular tori, which may be optically thick at wavelengths as long as 30 µm, or Compton-thick absorbing columns with NH 1024 cm-2.
The compactness of the IR-emitting regions in luminous IR galaxies (c.f. [Condon et al. 1991]) suggests two possible origins for the high ULIRG luminosities: compact nuclear starbursts and/or highly obscured AGN activity. Much early effort was given to determining which phenomenon powers ULIRGs; with the understanding that starburst and AGN emission are frequently found together in luminous galaxies, the emphasis has shifted to determining which mechanism is dominant and to understanding the relationship between co-existing AGN & starburst emission.
3.1. Optical to mid-IR Imaging
Early imaging surveys of IR-luminous sources spanned a wide range in luminosity (from < 109 L up to ~ 1012 L, and revealed an interesting picture [Soifer et al. 1984, Rieke & Lebofsky 1986]. Sources with Lir < 109 L are almost exclusively confined to undisturbed E and S0 systems, with few spirals. The fraction of spirals increases sharply with increasing IR luminosity however, with most systems in the range 1010 L < Lir < 1011 L being Sb or Sc type systems. At luminosities above about 1011 L, the majority of systems are still spirals, but an increasing fraction (up to ~ 25%) appear to be involved in interactions, or to show signs of morphological disturbance.
This apparent increase in the number of interacting IRAS systems was thrown into sharp relief by early ULIRG imaging studies, showing that interactions and mergers are much more common amongst ULIRGs than in lower luminosity systems, though the exact fraction in ongoing interactions remained contentious for some years. The first imaging surveys [Armus, Heckman & Miley 1987] showed that at least 70% of systems with ULIRG or near-ULIRG luminosities are interacting, with morphologies expected from the collision of two disk galaxies. Later optical and near-IR studies of ULIRGs [Melnick & Mirabel 1990, Hutchings & Neff 1991, Clements et al. 1996] found a higher fraction of ULIRGs involved in interactions, at least 90%, and that there are a wide range of merger stages present in the ULIRG population, from widely separated systems to advanced mergers [Murphy et al. 1996]. Other studies, however, found a much lower fraction of ULIRGs involved in interactions, fewer than 70% [Lawrence et al. 1989, Zou et al. 1991, Leech et al. 1994].
These results demonstrate that interactions and mergers play an important role in the ULIRG population. This provided a plausible trigger for the immense IR luminosities seen in ULIRGs. In order to both fuel and enshroud the power sources in ULIRGs (irrespective of whether they are starbursts, AGN, or both), a large quantity of gas and dust needs to be channeled into a small volume, most plausibly sited in the nucleus of the host galaxy. Results from N-body modeling of galaxy collisions (see [Barnes & Hernquist 1992] for a review) suggest that this is readily achieved during the course of a merger [Barnes 1989, Barnes & Hernquist 1996, Mihos & Hernquist 1996, Dubinski, Mihos & Hernquist 1999], though details depend on many variables (e.g. angle of approach, relative velocity, disc inclinations, bulge size and gas and dark matter masses). The majority of mergers involve an initial close approach, followed by a maximum separation of up to ~ 50kpc, reached after ~ 2.5 × 108 years, then a second close approach a few tens of millions of years later, which is rapidly followed by coalescence, and relaxation towards an elliptical profile. Total times scales to coalescence range from ~ 7 × 108 years to ~ 2 × 109 years, depending on the parameters of the encounter. Gas and dust can be channeled into the nuclear regions of the progenitors in one of two ways; either before coalescence, when tidal forces during the first close approach form bars in one or both progenitors (particularly if the progenitors have small bulge components) which are very efficient at channeling material into the central regions of a galaxy, or during coalescence, when shocks drive very large quantities of gas and dust into the nuclear regions. More recent N-body simulations have in general supported these conclusions, but added an intriguing possibility; under certain conditions (particularly if the progenitors are gas dominated), a merger between two disk galaxies can result in a disklike remnant, rather than an elliptical [Naab & Burkert 2003, Springel & Hernquist 2005, Robertson et al. 2005]. It seemed therefore that galaxy mergers could provide all the necessary physical conditions for triggering a ULIRG, and that this implicated ULIRGs in the formation of elliptical galaxies. Furthermore, these simulations could explain a result that had been puzzling; that ULIRG activity could apparently be triggered in mergers when the progenitors are still physically separate.
Optical imaging from the Hubble Space Telescope (HST) offered enhanced resolution and sensitivity over ground based facilities. An early HST study of ULIRGs using WFPC2 [Surace et al. 1998] focused on a small sample with `warm' infrared colours 5, which biases towards systems likely to contain an obscured AGN. All of the sample were found to be interacting, with complex structures in their nuclear regions. Several systems showed a large number of compact bright `knots' a few hundred pc in diameter, whose ages suggest they result from the merger. A complementary ground-based survey of ULIRGs with `cool' IR colours [Surace et al. 2000] found a similar picture; all the galaxies show signs of interactions, from early to late stage, and many systems harbour `knots' similar to those seen in the warm sample. The optical magnitudes are in most cases relatively modest (at least compared to the enormous IR luminosities); most systems have ~ L* luminosities, with very few being substantially brighter. A later HST survey of a larger, unbiased ULIRG sample [Farrah et al. 2001] found similar results; nearly all of the sample are interacting, with a wide range of merger stages. A small number host optical QSOs whose host galaxies are either interacting, or elliptical-like. Other authors have, on the basis of HST data, suggested that some ULIRGs show evidence for mergers between more than two galaxies, suggesting that ULIRGs may be the remnants of compact galaxy groups [Borne et al. 2000]. HST imaging of ULIRGs in the near-infrared with NICMOS [Colina et al. 2001, Bushouse et al. 2002] has produced very similar results to those from optical imaging, with at least 90% being interactions between two or (possibly) more progenitors over a wide range of merger stages, and that very few are much brighter than L*.
To date however, the largest optical/NIR imaging survey of ULIRGs has been done from the ground; [Veilleux, Kim, & Sanders 2002] present R and K images for 118 ULIRGs from the IRAS 1Jy spectroscopic survey, and find that virtually all show signs of interactions, though very few showed definite signs of interactions between more than two progenitors. Most of the systems appear to be late-stage mergers, especially for the more luminous systems and/or those with spectroscopic signatures of an AGN, and typically have luminosities (in the K band) of L* or greater. The most advanced mergers show evidence for emerging elliptical profiles.
Near diffraction-limited Keck mid-infrared imaging of ULIRGs has been obtained by Soifer et al. ([1999; 2000]) who find extremely compact structures, with spatial scales smaller than 0".3 in six of the seven ULIRGs observed. These compact sources emit between 30% - 100% of the mid-infrared energy from these galaxies. In Mrk 231, IRAS 05189-2524 and IRAS 08572+3915 there is strong evidence that the source size increases with increasing wavelength, suggesting heating by a central rather than extended source, consistent with the optical classification as an AGN. Spitzer mid-IR imaging programs of over 200 nearby ULIRGs (ongoing programs of J. Surace (Fig. 2) and J. Mazzarella) will provide great sensitivity in this important wavelength region, if limited spatial resolution.
Figure 2. IRAS 08572+3915, as seen at many wavelengths. The colour composite at the center encodes wavelengths from the far-UV to the near-IR as blue-to-red, and illustrates the complex composite nature of ULIRG systems. The ultraviolet emission is dominated by one of the two merger nuclei and by young super-star clusters along the leading edge of one of the tidal tails. Almost no ultraviolet or blue emission is seen from the NW nucleus. At optical wavelengths, increasingly dominated by old stars, we see the merger galaxy body, which appears to be two spirals, accompanied by tidal tails. The NW nucleus has a complex structure of dust lanes. As we proceed into the thermal infrared, all emission sources other than a single compact emission source in the NW nucleus fade away. The images from 0.16-2.2 µm are from STIS, WFPC2, and NICMOS [Goldader et al. 2002, Surace et al. 1998, Surace et al. 2000, Scoville et al. 2000]. The 8 µm image is from Spitzer (Surace et al, in prep), and has a beam size ~ 5 times greater than the HST data, but is sensitive enough to exclude much emission from the SE nucleus. Keck observations [Soifer et al. 2000] at similar wavelengths constrain the size of the nucleus to < 0.3 arcseconds, or 300 pc. (J. Surace, priv. comm.).
3.2. Optical & near-IR Spectroscopy
It might be suspected that spectroscopy at wavelengths shortward of a few microns would be useful mainly for redshift surveys, given the dust-enshrouded nature of ULIRGs. This, however, is a misconception; optical spectroscopic signatures of starburst and AGN activity are still apparent even when such activity is moderately obscured (up to and surpassing AV ~ 10 depending on the observations), and starbursts and AGN can be detected in polarized light even when the direct line of sight is completely hidden. Furthermore, spectroscopic diagnostics from the UV through to the near-IR are, on the whole, more mature than those at longer wavelengths [Baldwin, Phillips & Terlevich 1981, Veilleux & Osterbrock 1987, Osterbrock, Tran & Veilleux 1992, Dopita et al. 2000], and line strengths, ratios and profile shapes can provide powerful constraints on the nature of the source of excitation. Recent advanced spectral synthesis codes (e.g. [Leitherer et al., 1999; Kewley et al., 2001]) allow for insightful diagnostics of starburst events, particularly in the UV where direct emission from hot young star photospheres, rather than reprocessed light from dust, is being sampled.
Early optical spectroscopic surveys of ULIRGs generally showed that they were mostly starburst-like in the optical [Elston, Cornell & Lebofsky 1985], while samples with `warmer' infrared colours appeared more biased towards Seyferts or LINERS [de Grijp et al. 1985, Osterbrock & de Robertis 1985], or even optical QSOs [Beichman et al. 1986]. Conversely, studies of ULIRGs with `cool' IR colours generally indicated the presence of starbursts [Heckman, Armus & Miley 1987, Armus, Heckman & Miley 1988], sometimes accompanied by spectral signatures of Wolf-Rayet stars, indicating that the starburst was likely only a few Myr old. Also noted at this time was a tendency for more IR-luminous objects to exhibit Seyfert spectra, with narrow lines in direct light [Cutri et al., 1994] and broad lines in polarized light (e.g. [Hines et al., 1995].
To gain a complete picture of the spectroscopic properties of ULIRGs, however, requires large-scale surveys, which were soon forthcoming as part of IRAS followup. One of the largest to date is the IRAS 1Jy survey [Kim, Veilleux & Sanders 1998, Veilleux, Kim & Sanders 1999a], which showed that the majority of ULIRGs have optical spectra reminiscent of starbursts, but with a systematic increase in the fraction of ULIRGs with Seyfert (1 or 2) spectra with increasing IR luminosity. Most of the ULIRGs with Seyfert spectra however also show evidence for ongoing or recent star formation. Approximately 30% of the Seyferts are Sy1's, with a systematic increase compared to Sy2's with IR luminosity. Other notable findings included that optical reddening generally decreases with increasing distance from the nuclear regions and that the optically derived star formation rates are in most cases many times lower than those derived from mid-/far-IR data. Followup spectroscopy in the near-IR [Veilleux, Sanders & Kim 1997, Veilleux, Sanders & Kim 1999b] refined and extended this picture, showing that, overall, around 25% of ULIRGs show evidence for an AGN and that this fraction increases with increasing IR luminosity (reaching ~ 50% at Lir > 1012.3 L). Those ULIRGs with `warm' IR colours are more likely to show broad lines in the near-IR than `cool' ULIRGs, and there is no observed correlation between extinction in the NLR and the presence of broad lines in the near-IR, suggesting that the NLR and BLR do not lie along the same line of sight. An intriguing further result is that, of all the objects that show broad lines in the near-IR, all are Sy2's in the optical, with no LINERs or HII's. Furthermore, most ( ~ 70%) of the ULIRGs that show a Sy2 spectrum in the optical show broad lines in the near-IR.
Very recent spectroscopy of ULIRGs from the UV through to the near-IR has revealed some further important details. High spatial resolution UV and optical spectroscopy using the Space Telescope Imaging Spectrograph onboard HST of four `warm' ULIRGs [Farrah et al. 2005] has shown that the `knots' seen in optical imaging in many cases harbour very luminous starbursts and AGN, implying that these optically bright knots may also be the sites of the heavily obscured power sources behind the IR emission. The spectral properties of some of these knots also suggest further links between ULIRGs and (low ionization) broad absorption line (BAL) QSOs, and the forming cores of elliptical galaxies. The starbursts in these knots were all observed to be young, with ages of 4Myr-20Myr, supersolar metallicities, and an IMF (Salpeter) slope of less than about 3.3. Near-IR spectroscopy with the Very Large Telescope (VLT) in Chile, and with Keck [Genzel et al. 2001, Tacconi et al. 2002] has shown that the host galaxy kinematics of ULIRGs resemble those of ellipticals with luminosities of ~ L* (but with a large scatter), and that the host galaxy properties of those ULIRGs that contain an AGN and those without are very similar.
Finally, there is strengthening evidence that nuclear and galactic scale outflows may be common in ULIRGs [Heckman, Armus & Miley 1987, Wilman, Crawford & Abraham 1999], and high resolution spectroscopy has discovered galactic-scale outflows with high ejection efficiencies in many ULIRGs that are dominated by star formation [Rupke, Veilleux & Sanders 2002]. [Lipari et al. 2003] noted that relatively low velocity outflows are present in starburst dominated ULIRGs, but that higher velocity outflows are present in systems that show evidence for both a starburst and an AGN, though even in composite systems, the starburst is still probably the dominant mechanism behind galactic-scale outflows [Rupke, Veilleux & Sanders 2005].
3.3. Mid-infrared spectroscopy
An area where the Infrared Space Observatory, ISO [Kessler et al. 1996], excelled in the study of ULIRGs was in spectroscopy of nearby systems using SWS [de Graauw et al. 1996], LWS [Clegg et al. 1996], ISOCAM [Cesarsky et al. 1996] and ISOPHOT-S [Lemke et al. 1996], because for the first time high-sensitivity and high-resolution spectroscopy could be obtained in a wavelength region with minimal effects of dust extinction, at least compared to the optical. Moreover the mid-infrared region holds several diagnostic lines which are very useful for characterization of the major power source for ULIRGs. The mid-infrared continuum shape can be a powerful diagnostic itself, betraying the existence of warm dust in the close vicinity of an AGN. We refer the reader to several previous reviews on ISO's legacy on star forming galaxies, AGN and ULIRGs [Genzel & Cesarsky 2000, Elbaz 2005, Verma et al. 2005, Oliver & Pozzi 2005] for more details.
[Genzel et al. 1998] demonstrated the power of SWS spectroscopy to separate AGN from starbursts by comparing the strength of the 7.7 µm PAH equivalent width to high/low excitation line ratios, such as [OIV]25.9 µm / [NeII]12.8 µm; PAH molecules will be destroyed by high-intensity AGN radiation fields so low PAH/continuum ratios should correlate with strong [OIV]/[NeII] ratios. They concluded that at leat half of their sample of 15 local ULIRGs have simultaneous starburst and AGN activity, but 70-80% of the sample is predominantly powered by star formation (a result mirrored by mid/far-IR SED fitting, [Farrah et al. 2003]). The method requires high sensitivity to determine the fine structure line ratios, and the PAH emission can also be contaminated by strong silicate absorption at 9.7 µm. Moreover dense nuclear environments can hide AGN activity even at mid-IR wavelengths, as discovered by [Clavel et al. 2000] in Seyfert 2 nuclei in which the the mid-infrared continuum from the AGN is sufficiently absorbed to allow extranuclear PAH emission to dominate the spectrum. [Laurent et al. 2000] developed a diagnostic diagram based on the 6.2 µm PAH feature relative to continuum strength versus mid-IR continuum colour, which could be used for systems lacking fine-structure line spectroscopy, and [Peeters et al. 2004] extended this approach to include far-infrared continuum colours. [Lutz et al. 1998, Rigopoulou et al. 1999] and [Tran et al. 2001] applied these techniques to large ULIRG samples studied with ISOPHOT-S, concluding, in agreement with [Genzel et al. 1998], that starbursts predominantly power these systems, though the presence of heavily obscured AGN cannot be ruled out.
[Soifer et al. 2002] obtained low spectral resolution, but high angular resolution, Keck mid-IR spectra of five LIRGs-ULIRGs demonstrating that PAH emission, when present, generally is circumnuclear in origin, extended over scales of 100-500pc. The silicate optical depths in these sources can be as high as 15, suggesting that even mid-IR spectroscopy may not be probing the true nuclei in the most compact sources.
Mid-IR spectral classifications, however, generally agree with optical line classifications of AGN vs star formation power, although some ULIRGs with LINER-like optical spectra were interpreted as starburst dominated in the mid-IR, attributed to starburst wind-driven ionising shocks [Lutz, Veilleux & Genzel 1999, Sugai & Malkan 2000] instead of to low-level AGN. However, the LINER situation is complex as demonstrated by subsequent comparison of ISO-SWS fine structure line spectroscopy with Chandra X-ray imaging of a sample of LINERs. [Satyapal et al. 2004] has shown that LINERs are intermediate between starbursts and AGN in mid-infrared line excitation, and that most LINERs contain a compact hard X-ray source characteristic of an AGN. They also found some anti-correlations between mid-IR fine structure line diagnostics and hard X-ray AGN diagnostics: their highest excitation mid-IR spectrum source NGC 404 shows only weak soft X-ray emission while the low mid-IR excitation LINER NGC 6240 shows an extremely luminous binary X-ray AGN [Komossa et al., 2005]. The most likely explanation for objects of this kind are extremely high optical depths even at mid-infrared and X-ray wavelengths, and potentially different lines-of-sight to the AGN core at different wavelengths [Risaliti et al. 2000], and/or unusual ratios of gas-to-dust optical depths.
[Sturm et al. 2002] have developed diagnostic diagrams using [OIV]25.9 µm, [SiII]34 µm, [NeVI]7.65 µm and [NeII]12.8 µm and other fine structure lines, coupled with Brackett at 2.63 µm, to separate high from low excitation systems and estimate the fractional contribution of star formation and AGN to the integrated light using mixture lines, in analogy to the well-known optical line diagnostic diagrams [Veilleux & Osterbrock 1987]. Most ULIRGs were not detectable by ISO in these lines, though Arp 220 was shown to exhibit low excitation in these diagrams.
[Spinoglio, Andreani & Malkan 2002] proposed an analogous far-infrared approach using [CII]158 µm, [OI]63 µm and [OII]88 µm, though again very few ULIRGs were detectable in these lines by ISO. Arp 220 has been observed extensively by ISO, as the nearest and brightest ULIRG. Arp 220's FIR spectrum is dominated by molecular absorption from species common to Galactic photo-dissociation regions (PDR): OH, H2O, CH, NH, NH3 [Fischer et al. 1999] and is very weak in the fine structure emission lines, with [OI] 63 µm in absorption. The [CII]158 µm fine structure transition decreases with increasing IR luminosity (and more strongly with FIR colour - S60μ m/S100μ m) in infrared galaxies [Malhotra et al. 1997], and in ULIRGs is only 10% of that measured in less luminous starbursts [Fischer et al. 1999, Luhman et al. 2003]. Since [CII] is the primary coolant in the ISM of normal galaxies it was expected to be strong in high-star formation systems, and its apparent lack is interpreted to indicate a stronger radiation field in luminous starbursts producing charged grains which result in a lower heating rate in PDRs [Malhotra et al. 2001]. [González et al. 2004] have modeled the FIR spectrum of Arp 220 over the spectral range 40-200 µm. Their model requires three components: the compact nuclei, modeled as a unit, with T = 106 K which are optically-thick throughout the infrared, an extended region which is dominated by PDR emission at T ~ 40-90 K, and a halo which produces absorption from the low-lying levels of OH and H2O plus the CH. Given that the two nuclei show distinct mid-IR spectra [Soifer et al. 1999] it is unclear what effect modeling the nuclei as a single source may have. In this model the [CII] emission is produced in the extended (PDR) region with little [CII] emission from the nucleus, which has distinctly non-PDR conditions.
For the majority of ULIRGs for which spectroscopy is not available broadband colours have been used to characterize the broad spectral type and major infrared energy source, following the early IRAS schemes [Klaas et al. 2001, Spinoglio, Andreani & Malkan 2002, Farrah et al. 2003]. ULIRGs are found to fall into two general classes by these methods: "cool" systems dominated by star formation, which show a relatively flat mid-infrared spectral shape and a steep rise towards longer wavelengths, and "warm" systems with a red power-law-like SED through the mid-infrared, usually assumed to be AGN-dominated, although compact HII regions can also show a warm mid-IR SED [Dopita et al., 2005]. Warm ULIRGs tend to have Seyfert 1 like optical spectral features while cool ones have starburst or Seyfert 2 optical spectra. The two clases were found to be indistinguishable in the far-infrared [Klaas et al. 2001], suggesting that at these wavelengths the emission may not be dominated by an AGN in either class, however [Peeters et al. 2004] find ULIRGs to have more prominent far-infrared emission than AGN or lower luminosity starbursts such as M82.
Several lines of evidence indicate that AGN become more bolometrically significant with increasing IR luminosity in ULIRGs. The incidence of both AGN optical line indicators and mid-infrared continuum and line AGN diagnostics were found to increase with increasing infrared luminosity [Shier, Rieke & Rieke 1996, Genzel et al. 1998, Lutz et al. 1998, Rigopoulou et al. 1999, Tran et al. 2001]. However, these ISO HLIRG samples are biased towards previously known AGN so Spitzer confirmation of these result is important, using unbiased samples of infrared-selected HLIRGs.
Observations with the ISO satellite greatly expanded our understanding of the mid-infrared spectra of ULIRGs, however, many ULIRGs were beyond the reach of many of the diagnostic methods until the advent of Spitzer. Several extensive Spitzer programs are underway to adequately sample the local ULIRG population, the most comprehensive being a study of the mid-infrared spectra of a large number (> 100) of ULIRGs having 0.02 < z < 0.93 with the Infrared Spectrograph (IRS), as part of the IRS guaranteed time program. These sources are chosen primarily from the IRAS 1-Jy [Kim & Sanders 1998], 2-Jy [Strauss et al., 1990], and the FIRST/IRAS radio-far-IR sample of [Stanford et al. 2000]. In Figure 3 we show IRS spectra for three nearby ULIRGs (Mrk 1014, UGC 5101, and NGC 6240) whose spectra serve to highlight the range in properties seen throughout much of the sample, and the power of the IRS in wavelength coverage and sensitivity for studies of the nuclei and interstellar media in dusty galaxies [Armus et al., 2004; 2005].
Figure 3. IRS low-resolution spectra of Mrk 1014, UGC 5101 and NGC 6240. The positions of prominent emission features and absorption bands (the latter indicated by horizontal bars) are marked on the UGC 5101 spectrum. Figure courtesy of L. Armus and J. Houck.
Mrk 1014 (z = 0.1631) is a radio-quiet, infrared luminous QSO with broad optical emission lines and twin tidal tails indicative of a recent interaction [MacKenty & Stockton 1984]. UGC 5101 (z = 0.039) has a single, very red nucleus within a disturbed morphology suggestive of a recent interaction. Optically, UGC 5101 is classified as a LINER [Veilleux et al., 1995]. It has a high brightness temperature (T > 107K) radio nucleus at 1.6 GHz which is resolved with the VLBA [Lonsdale et al., 1995]. ISO SWS and PHT-S spectroscopy [Genzel el al., 1998] indicate a powerful, circumnuclear starburst. Based upon its IRAS colours, UGC 5101 is classified as a cold, starburst-dominated, far-infrared source. However, XMM data indicate an obscured, but luminous, hard X-ray source with Lx(2-10 keV) ~ 5 × 1042 erg s-1 and Lx(2-10 keV) / Lir ~ 0.002 suggestive of a buried AGN [Imanishi et al., 2003]. NGC 6240 (z = 0.0245), is a double-nucleus, merging galaxy [Fosbury & Wall 1979], with an 8-1000 µm luminosity of ~ 7 × 1011 L. The optical nuclear spectrum of NGC 6240 is classified as a LINER [Armus, Heckman & Miley 1987], and the extended optical nebula reveals the presence of a starburst-driven superwind [Heckman, Armus & Miley 1987]. X-ray observations with ASCA [Turner et al., 1998], Beppo-Sax [Vignati et al., 1999], Chandra [Komossa et al., 2003; Ptak et al., 2003], XMM-Newton [Netzer et al., 2005] provide clear evidence for the presence of one (or two) AGN behind significant columns of absorbing material (NH = 1-2 × 1024 cm-2). Relatively strong [OIV] 25.89 µm line emission in the ISO SWS spectrum of NGC 6240 led [Lutz et al. 2003] to suggest that up to 50% of the infrared energy emitted by NGC 6240 could be powered by a buried AGN.
The spectra of Mrk 1014, UGC 5101, and NGC 6240 are strikingly different. Mrk 1014 has a steeply rising mid-infrared spectrum with weak emission features and little or no silicate absorption. The spectra of UGC 5101 and NGC 6240, on the other hand, are dominated by strong silicate absorption at 9.7 µm and 18 µm, and PAH emission at 6.2, 7.7, 11.3, and 12.7 µm [Armus et al. 2004, 2005]. The extinctions toward the nuclei in UGC 5101 and NGC 6240, as estimated from the depths of the silicate absorption, are at least AV = 15-35 and AV = 60 mag, respectively. UGC 5101 also shows strong absorption between 5-7.5 µm from water ice and hydrocarbons. As suggested by [Spoon et al. 2002], the water ice features may indicate the presence of shielded molecular clouds along the line of sight to the nucleus. NGC 6240 has little or no water ice absorption, but very strong emission lines from warm (T ~ 300-400 K) H2. The mass of this warm gas is estimated to be approximately 1.4 × 108 M - about 1% of the cold molecular gas mass derived from single-dish millimeter CO line measurements [Solomon et al. 1997], but up to 3-7% of the cold molecular gas within the central 1 kpc measured by [Tacconi et al. 1999].
The IRS high-resolution (R = 650) spectra (not shown) provide important diagnostic measures of the dominant ionizing sources in the ULIRGs because it is possible to accurately measure unresolved atomic, fine-structure lines of Ne, O, Si, and S, covering a large range in ionization potential. The [NeV] 14.3 / [NeII] 12.8 and [OIV] 25.9 / [NeII] 12.8 line flux ratios in Mrk 1014 (0.9 and 1.7, respectively) suggests that nearly all the ionizing flux comes from the central AGN, although the obvious presence of PAH emission suggests some extranuclear star formation. In UGC 5101 and NGC 6240 weak [NeV] emission has been detected, with 14.3 µm line fluxes of 5-6 × 10-21 W cm-2, indicating buried AGN. While the [NeV] / [NeII] and the [OIV] / [NeII] line flux ratios imply an AGN contribution of < 10% to the total luminosity in both sources [Armus et al. 2004, 2005], the large optical depth to the nuclei, as evidenced by the deep silicate absorption and X-ray columns, leaves open the possibility that the true contribution of the AGN to the bolometric power output in UGC 5101 and NGC 6240 may be much larger than revealed by the mid-IR emission lines. In NGC 6240, the extinction-corrected hard X-ray data are consistent with the buried AGN producing 50-100% of the luminosity. An inclined, dusty torus, patchy extinction, and/or a low covering factor for the [NeV]-emitting clouds, could reconcile these apparently discrepant estimates - an explanation often evoked to explain observations of other type-2 AGN.
[Spoon et al. 2004] observed the more distant ULIRG IRAS F00183-7111 at z = 0.327, detecting strong absorption from CO2 & CO gas, water ice, hydrocarbons and silicates, indicating high obscuration, a complex line of sight, and the presence of very warm, dense gas (720K). Direct signs of an obscured AGN are not found but Spoon et al. conclude that an obscured AGN probably accounts for most of the luminosity based on similarities to other systems with highly obscured AGN.
3.4. Radio Continuum Studies: AGN vs Luminous Radio Supernovae
The existence of a tight correlation between integrated far-infrared flux-density and radio continuum emission - the "Radio-FIR Relation" [Helou et al. 1985, Yun, Reddy, & Condon 2001] - over four orders of magnitude in IR luminosity, allows the use of high-resolution radio interferometric techniques to study the compact nuclei of luminous IR galaxies in one of the few spectral regions with relatively low optical depth. The Radio-FIR relation is not well understood, but is believed to be produced by nonthermal radiation from relativistic electrons in, or leaking out of, starburst-related supernova remnants. The seminal study of a complete sample of luminous IR galaxies by [Condon et al. 1991] with the VLA at 1.49 and 8.44GHz demonstrated the extreme compactness of luminous IR nuclei and concluded that most of their 40 galaxies - 6 of which are ULIRGs - are consistent with a starburst-related origin for the radio emission.
[Lonsdale et al. 1993] used a global VLBI array to detect 18cm high brightness temperature, Tb 106 K, emission cores from 17 of 31 luminous (log LFIR 11.25 (L)) infrared galaxies, consistent with such cores existing in all such LIRGs at ~ 10% of the total 18cm flux density. This result indicates either obscured radio-quiet AGN in most LIRGs and ULIRGs, possibly energetically dominant [Lonsdale et al. 1995], or clumps of starburst-related luminous radio supernovae (LRSN) and remnants [Smith et al. 1998]. Similar studies in the South [Norris et al. 1990, Kewley et al. 2000] detected compact nuclear radio cores with lower frequency, but their results are consistent, allowing for differences in sample selection and sensitivity.
Subsequent VLBI imaging dramatically revealed a cluster of luminous RSN in the nearest ULIRG, Arp 220 [Smith et al. 1998] - clear evidence for a high star formation rate, estimated at that time as ~ 100 M, yr-1 by [Smith, Lonsdale & Lonsdale 1998], which would be enough to power the infrared luminosity by star formation without resorting to an obscured AGN. [Momjian et al. 2003, Neff, Ulvestad & Teng 2004, Bondi et al. 2005] have reported possible "supernova factories" in IRAS 17208-0014, Mrk 299 and Mrk 273. respectively. The original dozen LRSN in Arp 220 have been monitored with VLBI [Rovilos et al. 2005] and more sensitive VLBI imaging studies [Lonsdale et al. 2006a] have revealed a number of new and fainter LRSN. These Luminous RSN were originally modeled after RSN 1986J in NGC 891 [Weiler et al. 1990], which exhibited a maximum radio power, log P1.5 GHz1986J = 21.15. The LRSN in Arp 220, however, exhibit much slower decay in their radio light curves [Rovilos et al. 2005] than RSN 1986J; longer RSN lifetimes reduce the inferred star formation rate, therefore it is not clear from the VLBI imaging whether a starburst is the chief power source for Arp 220.
On the other hand [Lonsdale et al. 2003a] demonstrated that the radio emission in UGC 5101, Mrk 231 and NGC 7469 is AGN dominated. Mrk 231, in some senses the classical "infrared quasar", has been studied with the VLBI by [Ulvestad et al. 1999]. Their images (see also [Lonsdale et al. 2003a] show a triple structure, with a core and two lobes which classify it as a Compact Symmetric Object (CSO). It has been suggested that CSOs are young, << 106 yr, with the hot spots representing the working surface of a relativistic jet upon the ambient medium [Readhead et al. 1996]. If the southern (primary) lobe/hot-spot in Mrk 231 is confined by ram pressure, [Lonsdale et al. 2003a] estimate a lobe advance speed, va ~ 10-4 c and an age for the jet/compact source, < 106 yr. Despite the clear evidence for AGN domination of the radio structure in these systems, the radio power is small compared to the infrared/bolometric emission, which may still therefore be starburst dominated. In Mrk 231, for example, several studies [Carilli et al. 1998, Lonsdale et al. 2003a, Farrah et al. 2003] suggest that more than half of the total luminosity comes from a circumnuclear starburst in a molecular ring rather than from the AGN.
3.5. Molecular gas: CO & HCN Observations
The interpretation of Luminous Infrared Galaxies as starburst systems was strengthened soon after their identification with IRAS by studies of neutral hydrogen [Mirabel & Sanders 1988] molecular gas, principally CO [Sanders et al. 1986, Sanders et al. 1991] but more recently HCN [Gao & Solomon 2004] and OH maser emission [Baan et al. 1982, Baan 1989]. These early studies demonstrated that ULIRGs as a class exhibit compact nuclear reservoirs of high-density gas, and, with mass estimates of order 109 - 1010 M in HI and in H2, are consistent with the interpretation that star formation accounts for a substantial fraction of the FIR luminosity in Luminous IR Galaxies.
Millimeter-wave interferometer measurements of CO emission (e.g. [Scoville et al. 1991; Solomon et al. 1997; Downes & Solomon 1998; Bryant & Scoville 1999] have demonstrated that nearly half of the CO mass in Luminous IR Galaxies is contained within the central regions, r < 0.5-1 kpc, with as much as 1010 M of molecular gas distributed in nuclear disks of radius a few hundred pc, but thicknesses of one-tenth the radius and densities over 104 cm-3 [Bryant & Scoville 1999]. The mean molecular surface densities in these structures may exceed 104 M pc-2 with the molecular gas providing a large fraction of the dynamical mass. The picture of molecular gas in ULIRGs is very different from that in the Galaxy, with much higher surface densities and inferred optical depths (Av 102 - 103). Early CO studies assumed a Galactic conversion ratio between CO Luminosity and H2 mass, MH2 / L'CO = 4.6 M (K km s-1 pc-2), however, recent analyses suggest that this may overestimate the molecular mass, either owing to high-brightness temperatures in the CO emission (e.g. Mrk 231; [Bryant & Scoville 1999]) or because the line width/velocity dispersion reflects the total dynamical mass in the central nuclear region, rather than that within virialized molecular clouds [Solomon et al. 1997].
Again, Arp 220 provides a convenient laboratory to study molecular gas, if not always in a typical ULIRG environment. [Sakamoto et al. 1999] have studied the Arp 220 nuclei in the CO(2-1) transition and accompanying 1mm continuum. Steep velocity gradients are found in the CO associated with each of the two nuclei, which are misaligned with each other and with the outer CO disk. Sakamoto et al. interpret these as molecular disks associated with the merging nuclei, counter-rotating with respect to each other and with respect to the outer disk. The dynamical masses inferred from CO kinematics are of the order of 2 × 109 M. The central molecular gas is inferred to have high filling factor, more like a uniform disk than individual clouds, which will have an inward accretion rate of approximately 100 M yr-1 [Scoville, Yun & Bryant 1997] which is similar to the star formation rate of Arp 220 inferred from the FIR Luminosity [Smith, Lonsdale & Lonsdale 1998, Farrah et al. 2003].
Early studies placed considerable emphasis on the ratio of FIR Luminosity to CO Luminosity - often called the Star Formation Efficiency - demonstrating that the ratio of FIR/CO increases at higher LFIR. This is interpreted as either an increase in Star-Formation Rate per unit molecular gas mass, i.e. more efficient star formation, or as possible evidence for an alternative, AGN, contribution to the FIR luminosity [Sanders & Mirabel 1996]. More recently, studies of HCN, which, owing to the higher dipole-moment of the HCN molecule is a tracer of warmer (Tkin ~ 60-90K), higher-density (nH2 ~ 105 - 107 cm-3) gas, shows a much higher ratio of HCN/CO luminosity in Luminous IR Galaxies than in quiescent spirals like the Milky Way [Solomon, Downes & Radford 1992]. Furthermore the FIR/HCN luminosity (HCN Star Formation Efficiency, as above) shows an approximately linear relationship in Luminous IR Galaxies with Lfir from 1010 - 1013 L indicating the presence of an abundant warm, high-density molecular environment within ULIRGs [Gao & Solomon 2004]. ULIRGs thus have a much greater dense, warm medium like that of the star-forming cores of Galactic molecular clouds - perhaps the central regions of the molecular disks, although little spatial information is available about the HCN distribution in ULIRGs. Since there is little doubt that the lower luminosity systems amongst this sample are dominated by star formation, the continuity of this relation up to luminosities characteristic of HLIRGs is one of the most compelling pieces of evidence, albeit statistical, that most ULIRGs are dominated by starburst power rather than AGN power.
3.6. Maser Emission
Luminous IR Galaxies have been known to be strong emitters of OH maser emission since the discovery of OH 1667MHz emission in Arp 220 by [Baan et al. 1982], who dubbed it a "megamaser", having an OH luminosity roughly 106 times that of masers in the Galaxy. H2O maser emission, which has been used to map torus/disklike structures around compact supermassive AGN nuclei such as NGC 4258 [Miyoshi et al. 1996] is not generally detected in classical Luminous IR Galaxies (see [Lo 2005] for a review).
OH megamasers are found preferentially in the most luminous IR galaxies [Baan 1989], and a rough dependence of the OH luminosity on the square of the FIR luminosity was originally found, suggesting that the FIR emission provided the pumping mechanism to provide the necessary population inversion (e.g. [Henkel & Wilson 1990]) within a foreground molecular screen amplifying the diffuse radio continuum, which is in turn correlated with the FIR emission. The OH gas may trace very high density regions (nH2 = 105-7 cm-3), although an apparently separate component of the OH megamasers is associated with a much lower-density high-velocity outflow [Baan, Haschick & Henkel 1989].
The initial interpretation of the maser phenomenon of a foreground, low-gain molecular screen amplifying the diffuse radio continuum in Luminous IR Galaxies, however, has been questioned by the VLBI observations of Arp 220 [Diamond et al. 1989, Lonsdale et al. 1994, Lonsdale et al. 1998, Rovilos et al. 2003] demonstrating that over two-thirds of the masing gas in Arp 220 is compact, produced in structures a few pc3 in volume with amplification ratios of order 103 or higher. These compact masers have complex spatial and velocity structure arising in clouds generally within the Arp 220 nuclei, but not coincident with nuclear radio continuum (or any other detectable radio continuum). The southern component within the western nucleus (the one with the majority of LRSN) shows a velocity gradient in excess of 18,000 km s-1 [Rovilos et al. 2003], which, interpreted as rotation, implies a mass of order 2 × 107 M.
Similar compact maser emission has been detected with Global VLBI in IIIZw35 and IRAS17208-0014 [Diamond et al. 1999], and in 12032+1707 [Pihlstrom et al. 2005]. On the other hand Mrk 231 [Lonsdale et al. 2003a, Klöckner et al. 2003], Mrk 273 [Klöckner & Baan 2004] and 14070+0525 [Pihlstrom et al. 2005] show more extended emission in a circumnuclear disk with conditions more like the classical maser model. IIIZw35 also shows diffuse ring of maser emission with radius ~ 20pc and central mass ~ 7 × 106 M [Pihlstrom et al. 2001]. In this case the compact emission occurs at the tangent points of the ring, and the maser structure has been modeled by multiple high density clouds within the diffuse ring. Such a structure does not, however, appear to be able to explain the complex, compact maser emission in Arp 220.
In the case of Mrk 231, the OH emission appears to be the central portion of a circumnuclear HI disk detected in absorption by [Carilli et al. 1998], though it is misaligned with the central CO distribution [Klöckner & Baan 2004]. [Lonsdale et al. 2003a] speculate that the ignition of Mrk 231's Seyfert nucleus may have disrupted compact maser emission in the nucleus itself.
An extensive survey for OH maser emission in over 300 IRAS galaxies from the PSCz Survey [Saunders et al. 1990] with z > 0.1 has been carried out with the upgraded Arecibo Telescope by Darling & Giovanelli ( and references therein). This survey has expanded the sample of known megamasers in luminous infrared galaxies to over 100 with detections in nearly 20% of the galaxies surveyed. Their analysis suggests few correlations between OH and FIR properties. A re-analysis of the LOH - LFIR relationship in their data suggests a much flatter relation LOH LFIR1.2 consistent with other recent analyses [Kandalian 1996] which finds a slope of approximately 1.4, suggesting a mixture of extended, unsaturated emission plus compact, saturated clouds. One source in the [Darling & Giovanelli 2002] sample, IRAS 21272+2514, shows apparent variability which the authors interpret as due to interstellar scintillation. This interpretation requires ~ 30-60% of the OH maser emission to originate in saturated clouds of dimension smaller than ~ 2pc.
A satisfactory model for the complexities of OH masers in ULIRGs has yet to be proposed. Almost certainly the diffuse emission follows the classical model with largely unsaturated OH maser clouds, pumped by the strong FIR radiation field, amplifying the nuclear 1.6GHz radio continuum. The compact, pc-scale structures in Arp 220 and other galaxies, coupled with the lack of detectable associated 1.6GHz continuum emission, requires a substantial saturated maser component. These structures subtend solid-angles too small to intercept sufficient FIR photons to excite the OH molecules, suggesting that collisional excitation must be important [Lonsdale et al. 1998]. A tentative correlation between OH line-width and X-ray luminosity [Kandalyan 2003] may indicate that X-ray heating of the molecular gas plays a role in collisional excitation.The possible roles of shocks and/or AGN activity remain to be explored.
3.7. X-Ray Emission
The importance of X-rays has been recognized not only because of the diagnostic ability of the X-ray to discriminate between AGN and Starburst emission [Rieke 1988], but also because models of the X-ray background (XRB) require substantial populations of highly-obscured AGN at redshifts, z ~ 0.5-1.5, to reproduce the observed XRB spectrum [Worsley et al. 2005]. Rieke's early analysis from HEAO A-1 demonstrated that ULIRGs are underluminous in the 2-10keV band compared to classical Seyfert galaxies or QSOs. It has only been recently with the Chandra and XMM-Newton X-ray observatories that the implications of this early discovery could be investigated in significant ULIRG samples with sufficient resolution and signal-to-noise to provide reliable diagnostics of the nature of the nuclear X-ray sources.
Further evidence of the weakness of X-ray emission in ULIRGs comes from a ROSAT survey [Boller et al. 1998] of 323 ULIRGs which detected fewer than 10%. Probably the most significant starburst-related discovery of the ROSAT soft X-ray satellite is the detection of extended thermal outflows, first in nearer, lower-luminosity starbursts, and then in Arp 220, dubbed "superwinds" [Heckman et al. 1996]. Similar thermal components appear to be common in ULIRGs and they are described as "ubiquitous" in starbursts with Star Formation Rates > 10-1 M yr-1 kpc-2 in [Heckman's 2001] review; see also the recent review by [Veilleux, Cecil, & Bland-Hawthorn 2005]. The superwinds are believed to be driven by supernova supplied kinetic energy with outflow rates comparable to the star-formation rates in these galaxies. Heckman speculates that these superwinds may be the principal "polluters" of metals and dust into the IGM. Further Chandra studies [McDowell et al. 2003] of the Arp 220 superwind reveal extended, faint, edge-brightened, soft X-ray lobes outside the optical galaxy out to a distance of 10-15 kpc. Bright plumes inside the optical isophotes coincide with the optical line emission [Colina et al. 2004] and extend 11 kpc from end to end across the nucleus. The data for the plumes cannot be fitted by a single-temperature plasma and display a range of temperatures from 0.2 to 1 keV. There is a close morphological correspondence between the H and soft X-ray emission on all spatial scales.
There have been three recent ULIRG X-Ray surveys with XMM-Newton [Franceschini et al. 2003a] which observed 10 ULIRGs, and with Chandra [Ptak et al. 2003, Teng et al. 2005] which observed 8 and 14 respectively. All of the ULIRGs surveyed were detected, with X-ray luminosities typically L2-10 keV < 1042 - 1043 erg s-1. These luminosities represent < 1% of the infrared luminosities in these systems, confirming that ULIRGs are much less luminous in the X-ray than classical AGN. Furthermore, the soft X-ray emission from all systems is dominated by extended, thermal emission with kT ~ 0.7 keV, and is uncorrelated with IR luminosity. In at least two XMM-Newton systems the emission is extended on scales of 10s of kpc, suggesting a superwind origin. In at least 5 galaxies a hard X-ray (2-10keV) component and/or the presence of 6.4 keV Fe K-line emission suggests the presence of an AGN which is not energetically dominant [Franceschini et al. 2003a]. Similarly, in the nearer Chandra sample of [Ptak et al. 2003] all galaxies exhibit hard components which are interpreted as AGN sources. 5 galaxies in the Chandra samples which are classed as AGN (including IRAS 05189-2524, Mrk 231, Mrk 273) show order of magnitude greater X-ray luminosities than the Starburst ULIRGs (UGC 5101, IRAS 17208-0014, IRAS 20551-4250, IRAS 23128-5919) with NGC 6240 being an intermediate case 6. As already noted UGC 5101 and Mrk 231 are believed to harbor obscured AGN based on their Spitzer spectra. Mrk 273, NGC 6240 and UGC 5101 exhibit Fe K-line emission, but sufficiently weak that the authors argue against X-rays reflected from a Compton-thick (NH > 1024 cm-2) absorber as the origin of the observed X-ray emission. [Teng et al. 2005] use hardness ratios to estimate the X-Ray spectral properties of their fainter galaxies. The photon indices for the combined Chandra samples peak in the range 1 < < 1.5, with a tendency for Seyfert ULIRGs to have steeper spectra, > 2. Although the X-ray properties of this fainter sample are consistent with a Starburst origin, the presence of Compton-thick AGN cannot be ruled out and may be expected in many cases given the results discussed above.
Although the situation is complicated, certainly some ULIRGs must harbor luminous X-ray AGN, though perhaps behind large absorbing columns and thus only visible in hard X-rays. In Mrk 231 BeppoSAX revealed a highly absorbed (NH ~ 2 × 1024 cm-2) power-law component [Braito et al. 2004] and analysis of XMM-Newton data indicates that below 10 keV only scattered or reflected X-rays escape. In a 40ks observation of of Mrk 231, [Gallagher et al. 2002] find the majority of the X-ray luminosity is emitted from an unresolved nuclear point source with a very hard spectrum, the majority of the flux emitted above 2 keV. The source is also variable on a timescale of a few hours. They argue against a Compton-thick reflection model of Mrk 231 [Maloney & Reynolds 2000], proposing a Compton-thick absorber which allows scattered light from multiple lines of sight to be detected. Highly absorbed AGN power-law hard X-ray sources are also reported in Mrk 273 [Xia et al. 2002, Balestra et al. 2005], NGC 6240 [Iwasawa & Comastri 1998, Vignati et al. 1999, Netzer et al. 2005], and Mrk 1014 [Boller et al. 2002] although the starbursts in these galaxies dominate in softer X-rays, and possibly also bolometrically.
Again the nearby system, Arp 220, provides more details, as well as unanswered questions. BeppoSAX observations [Iwasawa et al. 2001] placed severe constraints on an energetically significant AGN in Arp 220, requiring an absorbing column, NH > 1025 cm-2 to hide an X-ray AGN. They suggested X-ray binaries as the source of Arp 220's hard X-rays. [Clements et al. 2002] detected several sources near Arp 220's nucleus, including a mildly absorbed point source with a hard spectrum coincident with the western radio nucleus, plus a fainter source which may coincide with the eastern nucleus. A classical X-ray AGN cannot be ruled out, but again, columns greater that 5 × 1024 cm-2 would be required to hide it. [Iwasawa et al. 2005] have reported the detection of Fe K emission in the XMM-Newton spectrum of Arp 220. A supernova shocked bubble as suggested by the VLBI observations of radio supernovae could provide an explanation. However, the apparent lack of emission from X-ray binaries is incompatible with the high supernova rate ( ~ 2 SNe yr-1) required.
5 i.e systems with f25 / f60 > 0.2, where f25 and f60 are the 25 µm and 60 µm IRAS fluxes respectively Back.
6 Mrk 231, IRAS 17208-0014 and IRAS 23128-5919 are included in both the [(Ptak et al. 2003] and [(Franceschini et al. 2003a] samples Back.