Annu. Rev. Astron. Astrophys. 2014. 52: Copyright © 2014 by Annual Reviews. All rights reserved

5. STAR FORMATION IN HOSTS OF ACTIVE GALACTIC NUCLEI

Ideally, co-evolution of supermassive black holes and their host galaxies should be studied in samples with measurements for all of the four key quantities that capture the status and growth of the supermassive black hole (BH mass and accretion rate) as well as the host galaxy (assembled stellar mass and star formation rate). Concerning host properties, many of the popular star formation indicators (rest frame UV-optical SED fitting, optical emission lines, mid-infrared flux) are significantly affected by AGN driven emission, in particular for type 1 AGN. Since in the rest frame far-infrared the contrast between the typical SED of a star forming galaxy and an AGN SED is the largest, many workers have exploited Herschel surveys to study star formation in the hosts of high redshift AGN. One goal of these surveys is to search for correlations between phases of intense host and black hole growth that may occur in the `merger scenario' (e.g. Sanders et al. 1988, Di Matteo, Springel & Hernquist 2005) where major galaxy mergers play a strong role in shaping the local scaling relations. Another one is to test to what extent (on the other hand) AGN are hosted by galaxies with already quenched star formation.

Even in the far-infrared part of the SED, the host dominance is not granted. However, to explore the conditions where the host is dominant, theoretical models of AGN heated dust are of somewhat limited value. The freedom to distribute dust outside the sublimation radius in different ways leads to a large variety of warm to cold SEDs, not all of which are actually present around real AGN. Two different empirical approaches have been pursued instead. Netzer et al. (2007), Mullaney et al. (2011) and Mor & Netzer (2012) have decomposed the total AGN+host SEDs of local systems using Spitzer-IRS spectra, and spectral templates of star forming galaxies. The resulting `intrinsic AGN SEDs' drop steeply from the rest mid- to far-infrared. Hatziminaoglou et al. (2010) and Rosario et al. (2012) have compared infrared colors of high redshift AGN hosts with inactive reference samples, finding the mid-infrared flux often boosted by AGN emission but rest frame far-infrared colors similar in AGN hosts and reference samples. Both approaches suggest that for z ≲ 3 AGN with luminosity similar to typical L_{2-10 keV} ≲ 10⁴⁵ X-ray deep field sources, the rest frame far-infrared emission is on average dominated by the host. This may not be the case for some of the most luminous AGN, or for AGN in hosts with particularly low star formation rates.

In practice, the mid- to far-infrared intrinsic AGN SEDs mentioned above, scaled according to the AGN luminosity, are often used to assess whether a given study is still in a regime where host dominance in the rest frame far-infrared can be confidently assumed or not. This is illustrated in Figure 11, which shows a summary of average far-infrared luminosities (tracing host star formation rates) for AGN of different luminosities and redshifts. Pre-Herschel studies of samples of powerful AGN indicated correlations between SFR and AGN luminosity that are suggestive of merger-induced correlations (e.g. Lutz et al. 2008, Netzer 2009). The much better statistics and the breaking of redshift-luminosity degeneracies by the Herschel samples provides a more differentiated picture. Host SFRs at a given AGN luminosity rise steeply with redshift. At z < 0.8, dependence of SFR on AGN luminosity is not significant for moderate luminosity AGN but appears to be present for the most luminous ones (bolometric L_AGN ~ 10⁴⁵erg s^-1). At 0.8 < z < 2.5, no significant SFR variations are seen over several orders of magnitude in AGN luminosity. Typical star formation rates of optically selected QSOs at similar redshifts, as derived from stacking or maximum-likelihood analysis, also find a clear increase of host SFRs with redshift, and typically weak dependence on AGN luminosity (Serjeant et al. 2010, Bonfield et al. 2011). Powerful z > 2 optically selected QSOs are found to be hosted on average in highly star forming systems (Serjeant et al. 2010, Mor et al. 2012, Leipski et al. 2013, Netzer et al. 2013). This reinforces previous ground based (sub)mm studies of powerful high-z QSOs, by new measurements which are probing closer to the peak of the star forming SED component. The AGN luminosity range covered by these high-z QSOs is too small to probe for trends with luminosity in the same way as for X-ray AGN at z ≲ 2.5, but some correlation of host star formation and AGN luminosity among very luminous AGN is suggested (e.g. Netzer et al. 2013).

Figure 11. Growth of galaxies and their black holes. Average rest frame far infrared emission of AGN hosts expressing the star formation rate is plotted as a function of bolometric AGN luminosity, in different redshift bins. Data for local BAT AGN and for z < 2.5 X-ray selected AGN are reproduced from Rosario et al. (2012). Data for optically selected high redshift QSOs are from Serjeant et al. (2010, and priv. comm.) and the z ~ 4.8 sample of Mor et al. (2012, square). Far-infrared luminosities are mean values that include direct detections as well a stacked nondetections. They include all AGN in a bin, and are plotted at the median AGN luminosity with horizontal error bars showing the range including 80% of the bin's sources. The dotted line indicates the proportionality for a continuous host and black hole growth that would produce the local universe relation between black hole mass and bulge mass (from Häring & Rix 2004 assuming black hole accretion efficiency of 0.1). The diagonal dashed line is the correlation for local AGN-dominated sources as proposed by Netzer 2009, and the diagonal grey band the approximate 1σ range exhibited by empirical pure AGN `intrinsic' SEDs (see Rosario et al. 2012 for details).

This absence or weakness of trends in average SFR with AGN luminosity among moderate luminosity (L_AGN < 10⁴⁵erg s^-1) X-ray selected AGN is reported by several studies with partly overlapping deep X-ray and Herschel datasets (Shao et al. 2010, Mullaney et al. 2012b, Rosario et al. 2012, Rovilos et al. 2012). The increase of AGN host SFRs with redshift (Figure 11) is similar to the increase with redshift in SFR of main sequence star forming galaxies, a link that is strongly reinforced by studies that compare active and inactive galaxies of the same stellar mass and redshift. AGN host SFRs and SSFRs are enhanced with respect to mass matched galaxies if both star-forming and passive galaxies are included in the matched reference sample (Santini et al. 2012), but are close to those of mass-matched star forming main-sequence galaxies only (Mullaney et al. 2012b, Santini et al. 2012, Rosario et al. 2013a). In other words, the hosts of moderate luminosity z ≲ 2 AGN resemble typical massive `main sequence' star forming galaxies at these redshifts, and are less likely to be quenched (Mullaney et al. 2012b, Rosario et al. 2013a). The similarity between AGN hosts and star forming galaxies of the same stellar mass extends to the distribution of rest frame U-V colors (Rosario et al. 2013a). Because of dust, those colors are poor tracers of star formation, however. AGN populate the `green valley' of a color-mass diagram in a similar way as massive star forming galaxies, and far-infrared based SSFR locates many green valley AGN on the main sequence. Location on the green valley then does not correspond to ongoing quenching of star formation, and the green optical colors are indicative of dust reddening.

The link between the hosts of moderate luminosity AGN and normal star forming galaxies provided by these Herschel studies is fully consistent with the lack of evidence for enhanced merger fractions, in morphological studies of AGN hosts at these redshifts (e.g. Cisternas et al. 2011, Kocevski et al. 2012 and references therein). It is also consistent with the absence of some correlations that are expected in at least some versions of the merger scenario: Obscured AGN as defined by either high X-ray obscuring column or by optical type 2 are not more star forming than unobscured AGN at the same redshift (Rosario et al. 2012, Rovilos et al. 2012, Merloni et al. 2014), there is no evidence for enhanced star formation in the hosts of those AGN accreting with the highest Eddington ratios (Rosario et al. 2013b), and HiBal and non-BAL QSOs do not differ significantly in their far-infrared properties (Cao Orjales et al. 2012).

Focussing on more luminous L_{2-10 keV} > 10⁴⁴ erg s^-1 1 < z < 3 AGN, (Page et al. 2012) reported reduced Herschel detection rates and a drop in average SFRs in comparison to moderate luminosity AGN. They interpret this in terms of star formation suppression by the AGN. This finding is in tension with the better statistics study of Rosario et al. (2012, Figure 11) which finds no such drop, and the significant Herschel detection rates of luminous AGN at similar redshift in other studies (e.g. Hatziminaoglou et al. 2010, Dai et al. 2012). It is also not reproduced in a larger statistics follow-up (Harrison et al. 2012) with methodology closely matching that of Page et al. (2012). The differences between these studies are probably caused by the combination of small number statistics and a considerable dispersion in individual SFRs that enter the average SFRs for a certain L_{2-10 keV}, z bin. There likely is a fraction of hosts of luminous z ~ 2 AGN that are quenched (i.e. have very low SFR), even if the average host SFR of such luminous AGN is not lower than the average SFR for lower luminosity ones.

Evidence for suppressed star formation has been reported in massive radio galaxies at z < 0.8 and L_K > 1.5 L_*, in comparison to magnitude and color-matched inactive galaxies (Virdee et al. 2013). Similarly, radio excess AGN at 0.2≲ z ≲ 3 show lower specific star formation rates than X-ray AGN at similar redshifts (Del Moro et al. 2013). These two results may currently present the strongest Herschel based evidence for quenching in a subset of the high redshift AGN population. Of course, radio galaxies are not all passive. Strong star formation in some powerful high-z radio galaxies has been implied by previous ground-based submillimeter observations and is being studied with detailed Herschel SEDs (e.g. Seymour et al. 2011, Barthel et al. 2012, Wylezalek et al. 2013).

Quenching by an AGN may also be responsible for the fast (up to ~ 1000 km/s) massive (outflow rates up to ~ 1200 M_⊙ yr^-1) molecular outflows found in local ULIRGs via both CO emission and Herschel OH absorptions (e.g. Feruglio et al. 2010, Fischer et al. 2010, Sturm et al. 2011, Veilleux et al. 2013, Cicone et al. 2013). These could indicate a phase where the cold gas reservoir of these galaxies is expelled on short timescales within ~ 10⁶ - 10⁸ yr.

The studies discussed above address the star formation rate of various AGN populations. Taking a reverse view, several recent works indicate a roughly proportional increase of mean black hole accretion rate with host star formation rate at z < 1 (Rafferty et al. 2011, Chen et al. 2013). Mullaney et al. (2012a) group z ~ 1 and z ~ 2 star forming galaxies by stellar mass. Averaging over the samples in these stellar mass bins, they derive mean accretion and mean star formation rates which increase with host stellar mass at both redshifts. Absolute growth rate of the black hole and star formation is larger at z ~ 2, but ratios of black hole growth and star formation are similar at both redshifts and all stellar masses (Figure 12). In all three studies, the mean ratios of black hole to host growth are close to, but somewhat lower than the local ratio of black hole and bulge mass. A variety of factors may contribute to such mismatch: For example, obscured AGN may be missing from the samples studied. Also, it would be necessary to compare black hole accretion to only the fraction of the SFR that corresponds to bulge growth rather than disk growth.

Figure 12. (a) Average SFRs (right-hand axis) vs. stellar mass for z ~ 1 (open circles) and z ~ 2 (filled squares) samples of star forming galaxies (left-hand axis gives equivalent infrared luminosity for illustrative purposes only). Lines are least squares fits to the data. (b) Average X-ray luminosities of these star forming galaxies (same symbols as top panel) after accounting for any host galaxy contribution. Lines have the same gradients as in the top panel, only normalized to fit the inferred BH which is indicated in the right-hand axis. (c) Average SMBH accretion rate to SFR ratio for the two redshift samples. The uncertainties on these points are consistent with a flat BH / SFR ratio with respect to M∗ for both samples, indicated by the dotted and dashed lines, respectively. Reproduced from Figure 1 of Mullaney et al. 2012a.

Returning to Figure 11, we conclude that the lack of correlation between SFR and AGN luminosity for z<2.5 moderate luminosity AGN, and the similarity of their hosts to main sequence star forming galaxies implies that they largely are part of the same population. The classification as AGN or as inactive as well as AGN luminosity are varying rapidly through processes that do not strongly relate to the global star formation rate.

Merger induced effects may still play a role in shaping the increase in SFR with AGN luminosity for low-z and high luminosity AGN. This is in qualitative agreement with a visualisation of recent merger hydrodynamical models into this plane (Rosario et al. 2012), showing that these models spend considerable time in the region near the correlation line proposed by Netzer (2009). Fast AGN variability very likely plays a role in shaping this diagram, but different mechanisms deserve further scrutiny. Scenarios qualitatively reproducing Figure 11 range from one where accretion tracks constant star formation in the long term average but with a time-variable Schechter-like accretion rate distribution (Hickox et al. 2013), to one where accretion events and short star bursts mediated by (minor) mergers are placed on top of a slowly varying background star formation (Neistein & Netzer 2014).

In summary, Herschel studies suggest that z ≲ 2.5 AGN are hosted mostly in massive main sequence star forming galaxies with properties that are typical for their redshift. Special events such as major mergers likely play a less important role. Black hole feeding and star formation seem connected by the common gas reservoir and supply on a long galaxy evolution time scale. Given observational limitations such as the widespread use of averaging over subsamples, and the theoretical uncertainties, the detailed mechanisms of feeding the AGN are not strongly constrained.