GALAXY ZOO: THE FUNDAMENTALLY DIFFERENT CO-EVOLUTION OF SUPERMASSIVE BLACK HOLES AND THEIR EARLY- AND LATE-TYPE HOST GALAXIES^*

The photometric and spectroscopic data used here are taken from the Sloan Digital Sky Survey (SDSS) DR7 (York et al. 2000; Strauss et al. 2002; Abazajian et al. 2009). Throughout the paper, we use modelMags for galaxy colors, as they are derived from the best-fitting exponential or de Vaucouleurs galaxy profile and are thus best suited for measuring galaxy colors.

Our sample selection criteria ensure a large and complete sample of galaxies to low stellar masses whose SDSS images are suitable for detailed morphological classification. We select all galaxies with SDSS spectra classified as GALAXY (Strauss et al. 2002) in the redshift interval 0.02 < z < 0.05 and limit our selection to galaxies with r < 17 AB mag. In order to create a magnitude-limited sample that is close to a stellar mass-limited sample, we further limit this sample to M_z,Petro < −19.5 AB mag. We choose the z band rather than the r band since the redder z band is closer to stellar mass. This yields a total of 47,675 galaxies.

The SDSS fibers are circular with a diameter of 3'' (York et al. 2000; Strauss et al. 2002), which corresponds to a physical footprint of 1.2–2.9 kpc in diameter at the lower and upper redshift limits of our sample. The spectroscopic information used in this work therefore samples only the central regions of galaxies, rather than the bulk light which is sampled in more distant (z ∼ 0.1) galaxies. We describe the selection of AGNs from this parent sample in Section 3.

In the past, obtaining accurate morphologies for large samples of galaxies has been difficult. Only with the involvement of over 200,000 citizen scientists in the Galaxy Zoo project (Lintott et al. 2008) we have been able to gather multiple independent visual classifications for the entire SDSS main galaxy sample that are comparable to those of professional astronomers. Galaxy Zoo provides a web interface¹⁹ where volunteers from the general public are shown randomly selected images from the SDSS main galaxy sample. Since every object supplied for inspection by Galaxy Zoo has been viewed multiple times by independent classifiers, we are able to use the distribution of votes to choose the purity of the classifications of individual galaxies by specifying the required level of agreement among classifiers.

We use the definition for the clean sample of Land et al. (2008), which requires an 80% majority agreement on the morphology of any object. We combine all spirals regardless of orientation into the late-type class and all the spheroids into the early-type class, while those objects without a consensus at the 80% level are placed in the indeterminate class. We use the label "early-type" rather than "elliptical," as we specifically include lenticular galaxies in this class, while the late-type class contains anything with a discernible disk, from an Sa to an Sd galaxy.

We include merging galaxies with the indeterminate-type class; in any case, major mergers in the local universe are rare (1%–2%; Darg et al. 2010a, and references therein), and the incidence of emission-line AGNs in them is not significantly enhanced over the general population (Darg et al. 2010b). In Table 1, we present the total numbers and population fractions of the various morphology classes defined by the Galaxy Zoo classifications.

In order to give a sense what these three morphological classes mean, we show in Figure 1 a randomly selected sample of AGN host galaxies in each morphology class. We stress that these are AGN host galaxies and not objects taken from the entire population. For examples of images of normal galaxies, we refer the reader to Lintott et al. (2008).

Zoom In Zoom Out Reset image size

We measure stellar masses for all objects in this sample by fitting the five SDSS photometric bands to a library of 6.8 × 10⁶ model star formation histories generated from Maraston (1998, 2005) stellar models. The star formation histories are parameterized as two bursts (e.g., Schawinski et al. 2007; Kaviraj et al. 2007) of varying ages and mass fractions, fixed solar metallicity, varying dust extinction (0< E(B − V) <0.5; Calzetti et al. 2000) and exponential e-folding time (τ = 10, 100, 1000 Myr); all models assume a Salpeter (1955) initial mass function. Stellar masses are measured by finding the minimum of the χ² statistic in the parameter space probed, while the errors are determined from the range of stellar masses within 1 − σ of the minimum. We do not vary the initial mass function.

We test the reliability of our stellar mass measurements for AGN host galaxies with the methodology described in Appendix A. We find that our stellar mass measurements are robust for the AGN present in our sample.

We measure the emission-line fluxes using an analysis tool called Gas AND Absorption Line Fitting algorithm (GANDALF; Sarzi et al. 2006), which uses the penalized pixel-fitting(PPXF) method of Cappellari & Emsellem (2004) to simultaneously fit the stellar continuum and nebular emission lines.²⁰ The advantage of this tool is that it is designed to minimize template mismatch, where emission and absorption lines are difficult to separate (see Falcón-Barroso et al. 2006 and Sarzi et al. 2006 for example applications). We use a set of stellar templates from Tremonti et al. (2004) plus Gaussian emission-line templates. We measure [O iii] λ5007 fluxes for each processed SDSS spectrum and compute the total luminosity $L[\mbox{O\,{\sc iii}}]$ by correcting for extinction based on the measured Balmer decrement. GANDALF also provides measurements of the stellar velocity dispersion. Since the fixed SDSS spectroscopic fiber samples the light out to a different fraction of the effective radius, we use the empirical correction derived by Cappellari et al. (2006) to correct the velocity dispersion to the effective radius of the bulge surface-brightness profile.

Our AGN selection technique is the same as used by Schawinski et al. (2007): we first separate galaxies whose narrow emission lines are dominated partially or wholly by star formation on the [N ii]/Hα diagram (Figure 2, left) by using the extreme starburst line of Kewley et al. (2001) to retain only those beyond this line (i.e., the top right of the diagram). Between this extreme starburst line and the empirical pure starburst line of Kauffmann et al. (2003b), there is a substantial population of composite objects where both AGNs and star formation are comparable in ionizing luminosity. We exclude this class for this study and note that by removing this class of potential AGNs, we may be removing an important phase in the AGN–galaxy co-evolution at low redshift. The main problem with the composite class is the lack of easily accessible spectral indicators that would allow a clean separation of AGNs and star formation.

Zoom In Zoom Out Reset image size

We are thus left with galaxies whose emission lines are dominated by sources of ionization other than young stars. These sources may be AGNs, but can also include slow-moving shocks and gas excited by post-AGB stars and horizontal branch stars (see Ho 2008; Stasińska et al. 2008; Sarzi et al. 2010). These non-stellar sources are empirically divided into two branches, a division that is most obvious in the [O i]/Hα diagram (Figure 2, right). The lower branch of this population was first designated as "low-ionization narrow emission-line regions," or LINERs, by Heckman (1980). The upper branch is identified with galaxies hosting Type 2 (Seyfert) AGN.

Originally thought to be low-luminosity AGN, the nature of the LINER ionization mechanism has since been debated, and it is becoming increasingly clear that the majority of galaxies with LINER spectra are unlikely to be low-luminosity AGNs. The physical footprint of the line emitting area in nearby LINERs has been determined to be extended on kpc scales (e.g., Keel 1983; Phillips et al. 1986; Sarzi et al. 2006) and thus is unlikely be photoionized by a nuclear source. A detailed analysis of the distribution of this extended LINER emission in nearby SAURON galaxies by Sarzi et al. (2010) demonstrates that the majority of LINERs are incompatible with an ionization source in the nucleus, even in the case of galaxies known to host (X-ray or radio-detected) AGN, but instead are powered by some component of the underlying old stellar population. Sarzi et al. (2010) further show that this extended LINER emission is sufficiently strong to be detected in SDSS spectra of more distant z ∼ 0.05 galaxies. Cid Fernandes et al. (2009) have recently similarly argued that many of the SDSS objects in the LINER box are "retired" galaxies powered by old stellar populations.

While it is certainly the case that some X-ray detected AGNs exhibit nuclear LINER spectra (e.g., González-Martín et al. 2009), the presence of LINER-like line ratios over the large physical footprint of the SDSS spectroscopic fiber does not imply the presence of a low-luminosity AGN. LINERs are thus potentially highly diverse and the majority of them are not low-luminosity AGN. Even if a low-luminosity AGN were present, the $L[\mbox{O\,{\sc iii}}]$ emission sampled by the spectroscopic fiber cannot disentangle the relative contributions. We therefore decided to eliminate them from our sample and to retain only the definite AGN for our analysis.

Since galaxies exhibiting LINER spectra are different from those hosting Seyfert AGNs, we verify that the inclusion of the small number of LINERs that are above the luminosity limit of $L[\mbox{O\,{\sc iii}}] \sim 10^{40}$ erg s⁻¹ established in the next section does not significantly change our results, as most LINERs have substantially lower [O iii] λ5007 luminosities and Eddington ratios compared to Seyfert AGNs and are thus not sites of substantial black hole growth (see Appendix B for details).

The [O i]/Hα diagram allows for the best separation of AGNs and LINERs, so we use it for those spectra where [O i] λ6300 is detected with signal-to-noise ratio, S/N > 3. In cases where [O i] λ6300 is not detected, but [S ii] λλ6717, 6731 is, we use the [S ii]/Hα diagram (Figure 2, center). In both cases, we use the AGN–LINER division of Kewley et al. (2006). If neither [O i] λ6300 nor [S ii] λλ6717, 6731 is detected, but [N ii] λ6584 is, we make use of the original [N ii]/Hα diagram, where we use the empirical division line defined by Schawinski et al. (2007).

Out of a parent sample of 47,675 galaxies, we thereby select a sample of 942 narrow line AGNs (Table 2). This corresponds to a lower limit on the AGN fraction of 2%, as we do not include broad-line AGNs, highly obscured AGNs or LINERs. As we show later in this paper, the AGN fraction is highly dependent on position in the color–mass diagram and ranges from effectively zero to ∼10%.

Table 2. Typical [O iii] λ5007 luminosities, Eddington Ratios, and Black Hole Masses of AGNs by Host Galaxy Morphology

Host Galaxy	Number	Percentage of	Mean	Median	Mean	Median	Mean	Median
Morphology		Host Galaxies	log $L[\mbox{O\,{\sc iii}}]$	log $L[\mbox{O\,{\sc iii}}]$	log $L[\mbox{O\,{\sc iii}}]$/σ4	log $L[\mbox{O\,{\sc iii}}]$/σ4	log MBH	log MBH
			( erg s−1)	( erg s−1)			( M☉)	( M☉)
Late types	402	42.68%	40.40	40.39	0.295	0.306	6.52	6.63
Indeterminate types	431	45.75%	40.40	40.39	0.456	0.442	6.52	6.63
Early types	109	11.57%	40.27	40.30	0.326	0.397	6.36	6.45
All	942	100%	40.38	40.38	0.372	0.370	6.43	6.53

Download table as:  ASCIITypeset image

Since we wish to study the properties of AGN host galaxies as a class, we need to understand how effective our AGN selection technique is and to what degree it is biased. Emission-line selection and characterization of AGNs has been successfully employed in many works (e.g., Baldwin et al. 1981; Veilleux & Osterbrock 1987; Ho et al. 1995, 1997a, 1997b, 1997c; Kauffmann et al. 2003a; Schawinski et al. 2007; Constantin et al. 2008; Kewley et al. 2006). Deep Chandra studies of such emission-line selected AGNs by Ho et al. (2001) show that the majority of such AGNs also exhibit a nuclear X-ray source. For emission-line selection of AGNs to be reliable and reasonably complete, two assumptions must be true.

• 1.  
  The AGN must always excite a narrow-line region.
• 2.  
  The emission lines from the narrow-line region must never be overwhelmed by lines excited by other processes, in particular star formation.

Many recent studies of AGN host galaxies have noted that AGNs tend to concentrate in the green valley on the color–mass diagram, i.e., at intermediate optical colors, and that there are very few AGNs in the blue cloud and the red sequence. This raises the possibility that AGNs in star-forming blue cloud galaxies could be overwhelmed by emission from star formation. The lack of relatively luminous AGNs (L_X ∼ 10⁴³–10⁴⁵ erg s⁻¹) in the blue cloud has been recently been established by Schawinski et al. (2009a) using hard X-ray-selected AGNs, demonstrating that, at these luminosities, the absence of the AGN is not a selection effect.

But could there still be a population of low-luminosity emission-line AGN in the blue cloud whose emission lines are overwhelmed by star formation? For substantially lower luminosity AGNs than those studied by Schawinski et al. (2009a), any X-ray detection in blue cloud star-forming galaxies is likely ambiguous in the 0.2–10 keV window accessed by XMM-Newton and Chandra  as this could be due to star formation and X-ray binaries (e.g., Persic et al. 2004). Can we do any better with emission-line diagnostics?

In order to test whether emission-line AGNs similar to those we select on BPT diagrams might be hidden in blue cloud star-forming galaxies, we perform a simple empirical experiment. We pair a randomly selected star-forming galaxy from our galaxy sample with another randomly selected AGN. We rescale the AGN emission-line fluxes to a specific $L[\mbox{O\,{\sc iii}}]$ to simulate the effect of adding an AGN of a certain luminosity. For example, we compute the composite [N ii] λ6584 line from the line of a random star-forming galaxy and the line from a random AGN as follows:

$$\begin{equation} [\mbox{N\,{\sc ii}}]_{\rm model} = [\mbox{N\,{\sc ii}}]_{\rm SF} + \left(\frac{L_{\rm s}}{L_{\rm AGN}} \right) [\mbox{N\,{\sc ii}}]_{\rm AGN,} \end{equation} \tag{ 1 }$$

where L_s is the specific $L[\mbox{O\,{\sc iii}}]$ to which the AGN contribution to [N ii] λ6584 is scaled.

We perform this pairing of star-forming and AGN host galaxies 1000 times scaling the AGN contribution to L_s = 10³⁹, 10⁴⁰, and 10⁴¹ erg s⁻¹ in $L[\mbox{O\,{\sc iii}}]$, to cover the range of star formation rates²¹ and line ratios of blue cloud star-forming galaxies and of AGNs.²² We show the results of this exercise in Figure 3, where each panel shows a [N ii]/Hα diagram with the entire galaxy sample as shades. On top of each, we plot the 1000 composite objects scaled to the three specific $L[\mbox{O\,{\sc iii}}]$ values. This addition of an AGN component moves the position of a star-forming object off the star-forming locus, into the composite region and finally onto the AGN region. For added Sefyerts with $L[\mbox{O\,{\sc iii}}]$ =10⁴¹ erg s⁻¹, we find that only 1.4% remain below the Kauffmann et al. (2003a) empirical starburst line. However, for $L[\mbox{O\,{\sc iii}}]$ =10⁴⁰ erg s⁻¹ and =10³⁹ erg s⁻¹, this fraction increases to 12.8% and 56.6%, respectively. Similarly, the fractions of AGNs that move below the Kewley et al. (2001) extreme starburst line into the composite region are 85.5%, 54.4%, and 12.9%, respectively.

Zoom In Zoom Out Reset image size

From this experiment, we conclude that toward the higher luminosities in our sample (∼10⁴¹ erg s⁻¹), our sample is complete even in the blue cloud, and that there is no substantial population of such blue-cloud AGN. For lower luminosities, $L[\mbox{O\,{\sc iii}}]$ ≲10⁴⁰ erg s⁻¹, this is no longer true, and there may potentially be AGNs in the blue cloud whose narrow-line region is overwhelmed by current star formation. If there is such a population of AGNs, they cannot be detected via emission-line diagnostics, just as selection in the X-rays would be challenging. However, a recent search for broad-line AGNs hidden in star-forming galaxies by Mao et al. (2009) found only three such objects in a parent sample of over 3200 objects, establishing a lower limit on the very low-luminosity AGN fraction in the blue cloud of ∼0.1%, which suggests that the true abundance of such objects is very low, though broad lines may be similarly diluted by host galaxy light.

We diagnose the accretion state of the black hole using the [O iii] λ5007 line. Observed correlations between $L[\mbox{O\,{\sc iii}}]$ and other indicators of AGN luminosity imply that $L[\mbox{O\,{\sc iii}}]$ is a reasonably good indicator of the current accretion rate, though with considerable scatter (Heckman et al. 2004). More recent work by Diamond-Stanic et al. (2009) comparing a variety of indicators of the intrinsic bolometric output of local AGN do show that $L[\mbox{O\,{\sc iii}}]$ may not be an isotropic indicator and that Type 2 AGN—such as those in our sample—are intrinsically more obscured, and therefore more luminous, than previously thought, even based on hard X-ray observations (Rigby et al. 2009; cf., Lamastra et al. 2009). Given these caveats, $L[\mbox{O\,{\sc iii}}]$ will be used cautiously as a probe of black hole accretion in this work, and we restrict its use to statements about populations and trends rather than about individual objects. $L[\mbox{O\,{\sc iii}}]$ is the only indicator of the current accretion rate in low-luminosity AGNs available for large SDSS samples.

We infer supermassive black hole masses indirectly from the stellar velocity dispersion at the effective radius via the M_BH–σ relation (Gebhardt et al. 2000; Ferrarese & Merritt 2000; Tremaine et al. 2002; Häring & Rix 2004) using the slope and normalization of Tremaine et al. (2002). While the resolution of the SDSS spectra and the templates used is approximately 60 km s⁻¹, we are able to use GANDALF to probe slightly below this resolution limit. Whenever we quote black hole masses, we include all objects with velocity dispersion measurements down to 40 km s⁻¹, but caution that in the range from 40 to 50 km s⁻¹(log  M_BH =5.3–5.7), the quoted black hole masses are increasingly uncertain. The median error in this lowest velocity dispersion range is ∼10 km s⁻¹, corresponding to an error of ∼0.4 dex in M_BH.

To estimate the Eddington ratio of the AGN, we adopt the measure $L[\mbox{O\,{\sc iii}}]$/M_BH as a rough proxy (Kewley et al. 2006). Kauffmann & Heckman (2009) normalize the Eddington parameter to argue that the Eddington limit corresponds to log ($L[\mbox{O\,{\sc iii}}]$/M_BH) = 1.7 dex. Our AGN sample contains 33 (3.5%) objects with log $L[\mbox{O\,{\sc iii}}]$/M_BH > 1.7 dex, though the large scatter in $L[\mbox{O\,{\sc iii}}]$/M_BH makes it likely that none of our AGNs are super-Eddington. Kauffmann & Heckman (2009) argue that the bolometric correction for the extinction-corrected [O iii] λ5007 luminosity is a factor of 300–600, so the bolometric luminosities of the AGN in our sample are between 2.5 and 2.8 dex higher than the [O iii] λ5007 luminosities quoted throughout.

We summarize some of the key AGN properties, including the AGN fraction in each morphology class, in Table 2 and plot the distribution of the stellar velocity dispersion, $L[\mbox{O\,{\sc iii}}]$ and $L[\mbox{O\,{\sc iii}}]$/M_BH for both the entire AGN sample, and for each morphology class, in Figure 4. We note that the majority of the AGNs in our sample (∼90%) reside in indeterminate- and late-type hosts.

Zoom In Zoom Out Reset image size

Figures 5 and 6 suggest that the early-type and late-type host galaxies of AGN are fundamentally different populations, and that analyzing the entire AGN host galaxy population as a whole obscures these differences. There is no "typical" AGN host galaxy.

We note at this point that, out of the entire sample of 942 AGN host galaxies, only 12 (1.3%) were classified as ongoing major mergers with serious disturbances by Darg et al. (2010a); the vast majority of emission-line selected AGNs in the local universe are not hosted by galaxies undergoing a major merger.

From their distribution in Figure 5, we see that the AGN host galaxies span nearly the entire observed range of the general galaxy population (in this sample) in stellar mass. This is not the case for color; many works have already noted that AGN host galaxies preferentially reside in the green valley at intermediate host galaxy colors and that they are less common in the blue cloud and the red sequence (e.g., Schawinski et al. 2007, 2009a; Nandra et al. 2007; Wild et al. 2007; Salim et al. 2007; Constantin et al. 2008; Martin et al. 2007; Georgakakis et al. 2008; Silverman et al. 2008; Treister et al. 2009b).

The concentration of AGN host galaxies in the green valley is strongly dependent on whether we are considering the relative or absolute AGN distribution. Figure 5 shows that a substantial number of AGNs lie on the red sequence. However, when we consider the AGN fraction in Figure 6, this population is put into context—compared to the total number of galaxies on the red sequence, the AGN population is lower by at least a factor of 10–20 than in the peak regions, and therefore so is the duty cycle of these AGNs. Hence, the importance of AGNs to galaxy evolution on the red sequence is likely very small: we note that the population of AGN host galaxies at the massive end of the red sequence is almost entirely of indeterminate- and late-type morphology; massive early-type hosts are rare.

The AGN fraction is strongly concentrated below the red sequence and at relatively high host stellar masses of M_* ∼ 10¹¹ M_☉. Furthermore, the AGN fraction rapidly drops toward lower stellar masses and is very low toward the lower mass range.

The absence of emission-line AGN host galaxies on the red sequence is significant, as the emission-line signature of even a very low-luminosity AGN superimposed on a passive stellar population would be detected. Interpreting the absence of AGN host galaxies from the blue cloud is more complex: as discussed in Section 3.3, their absence is not significant for AGNs at the lower end of the $L[\mbox{O\,{\sc iii}}]$ of AGNs in our sample (see Figure 4), because the signature of the AGN narrow-line region could be overwhelmed by star formation. At higher luminosities of ≳10⁴¹ erg s⁻¹, the absence of blue cloud AGN is highly significant, as we would have been able to detect them. At even higher luminosities of L_X ≳ 10⁴³ erg s⁻¹, Schawinski et al. (2009a) have established the complete absence of AGNs in the blue cloud using hard X-ray observations by the Swift satellite.

What happens when we split the AGN population by morphology? In the remaining panels of Figures 5 and 6, a dramatically different picture emerges wherein the early- and late-type AGN host galaxies are shown to be significantly different populations. The absolute distribution of AGN host galaxies in Figure 5 already shows that the typical host stellar mass early-type AGN hosts are lower than that of late-type AGN hosts.

Turning to the AGN fraction and therefore to where the AGN duty cycle is highest, this picture becomes even more extreme. The AGN fraction peaks at a stellar mass of M_* ∼ 10¹¹ M_☉ in the general population, and ranges from 10^9.5to10^11.5 M_☉. Once we split by morphology in Figure 6, the difference between early- and late-type hosts becomes even more striking than for the absolute distribution: while for the late types the AGN fraction still peaks strongly at 10¹¹ M_☉, the AGN fraction in the early-type population peaks at M_* ∼ 10¹⁰ M_☉, an order of magnitude lower than in both the late types and the general population. This division has not been apparent in previous works due to a lack reliable morphological classification for very large numbers of galaxies. Because the early-type AGNs make up only ∼11% of all AGN (see Table 2), their properties are overwhelmed by the late types in a general AGN host galaxies sample.

The most striking aspect of the AGN fraction in Figure 6 is that in the early-type population, the AGN duty cycle is highest in an extreme population of early types that do not reside on the red sequence, but rather below it at the low-mass end and above the bulk of the blue cloud, i.e., in the green valley. Similarly, the AGN duty cycle for late-type galaxies is highest in objects that are more massive and redder than the typical late-type population. These red late types will include a substantial fraction of inclined spirals with intrinsically blue colors, but dimmed and reddened by dust. If the detected AGN fraction is independent of host galaxy inclination, which is expected if the bulk of the nuclear obscuration comes from the central regions orientated randomly with respect to the galaxy, then these contaminants to the red part of the late-type color–mass diagram from the part with much lower AGN fractions will make this conclusion even stronger.

The indeterminate-type galaxies exhibit an AGN fraction distribution similar to the late types, though their AGN fraction does not extend into the higher mass red sequence. This may be explained by the fact that the incidence of indeterminate-type galaxies in the red sequence is much higher than that of late types. We therefore tentatively conclude that the indeterminate-type AGNs, which are 45% of the AGN population after all, appear to behave more similar to the late types than the early types.

When we make a cut in Eddington parameter of $L[\mbox{O\,{\sc iii}}]$/ M_BH >0.7, corresponding to an Eddington ratio of L/L_Edd ∼0.1, an even more extreme picture occurs. By making a cut in Eddington ratio, we avoid any bias toward more massive black holes which radiate at higher luminosities at fixed L/L_Edd than their low-mass counterparts. Figure 7 is identical to Figure 6 except for the cut in L/L_Edd. This reveals that the duty cycle of high-Eddington AGNs is significant in only one population of low-redshift galaxies: low-mass early-type galaxies in the green valley.

Zoom In Zoom Out Reset image size

We conclude that the early- and late-type host galaxies of AGNs are fundamentally different populations: for late-type hosts, AGNs preferentially reside in massive (10¹¹ M_☉) galaxies either with green colors or on the red sequence. In contrast, early-type AGNs tend to have moderate mass (10¹⁰ M_☉) and strongly cluster in the green valley in between the blue cloud and the red sequence. We further showed that the duty cycle for high-Eddington AGNs is high only in green valley early-type galaxies.

The AGN fraction on the color–mass diagram is a new and powerful tool to probe the duty cycle of AGNs. Under the assumption that a high duty cycle AGN phase is more likely to affect the host galaxy evolution, the duty cycle is in turn a measure of the importance of AGNs for galaxy formation for specific galaxy populations.

We have shown that the properties of early- and late-type AGN host galaxies are different. This in turn strongly suggests that the triggering and fuelling of the AGN, as well as their role in the host evolution, are different. We therefore continue our discussion with the black hole masses and accretion efficiencies of black holes in early- and late-type AGN host galaxies.

Note that the characteristic black hole mass of the AGN is M_BH ∼10^6.5 M_☉ and that this value is approximately independent of the host galaxy morphology (Table 2). The active black holes in late-type hosts are typically more massive than those in early types by only ∼0.2 dex, or less than a factor of 2. Combined with the fact that the typical accretion rate estimated by $L[\mbox{O\,{\sc iii}}]$ is the same in early and late types, we find that the Eddington ratios as inferred via the Eddington parameter $L[\mbox{O\,{\sc iii}}]$/M_BH are similar, with the AGN in early types having Eddington ratios that are only about a factor of 1.5 lower than those in late types. Thus, the typical black hole mass growing today in early- and late-type galaxies is the same, with the same range of Eddington ratios. In addition, the host galaxy colors of the AGN host galaxies place them preferentially in the green valley, adding to an apparent picture where the conditions for black hole growth are approximately independent of host morphology.

However, as we have shown in the previous section, the stellar masses of early- and late-type host galaxies are very different, so taking only population averages into consideration can obscure crucial variations between populations. Similarly, it is not just relevant which black holes are currently accreting, but also to what degree these growing black holes are different from those in their quiescent parent population, and crucially, how this difference depends on host morphology.

In Figure 8, we plot the distribution of inferred supermassive black hole masses for both the AGN (striped) and the normal galaxy population (solid), both for the entire sample, and split by host morphology. These histograms highlight that taking the mean, and ignoring the distribution of black hole masses in the parent population masks intrinsically very different distributions of black hole growth. Even though the median black hole masses of active early- and late-type galaxies are comparable (Table 2), a Kolmogorov–Smirnov test shows that they are drawn from different parent distributions at the 99.7% significance level.

Zoom In Zoom Out Reset image size

The distribution of active black holes in late-type hosts skewed toward higher black hole mass, peaks at M_BH ∼10⁷ M_☉ and cuts off above M_BH ∼10⁸ M_☉. This contrasts sharply with the distribution of black hole masses in all late-types galaxies, which decreases with increasing black hole mass. Plotting the fraction of active supermassive black holes in late-type galaxies in Figure 9, we find a strongly increasing fraction from ∼1% at 10^5.5 M_☉ to ∼8% at 10^7.25 M_☉. At greater black hole masses, the fraction may begin to decline again, until we run out of number statistics.

Zoom In Zoom Out Reset image size

The black holes of early-type galaxies show a radically different picture (Figure 8). The mass distribution of active black holes in early-type galaxies is skewed toward the lowest inferred masses in our sample. Plotting the active fraction as for the late types in Figure 9, we find a very different behavior. The fraction of active black holes in early-type galaxies is decreasing with increasing black hole mass, but we run out of galaxies to provide adequate number statistics.

These two very different trends in early- and late-type hosts are unlikely to be selection effects. Since we are able to detect AGNs in low-mass early-types down to host galaxies with stellar masses of 10^9.5 M_☉; we do not see AGNs in late-type hosts where the lowest-mass hosts are ∼10^9.5 M_☉. In order to substantially change the trend of increasing AGN fraction with increasing black hole mass, we would have to have missed a very large number of AGNs in low mass blue cloud late-type galaxies, more than 10 times the number in our sample. To change the distribution to one skewed to low masses similar to the distribution of the early types, an increase by a factor of ∼1000 would be required (Figure 6). At luminosities above the ∼10⁴⁰ erg s⁻¹ completeness limit in Section 3.3, we are not missing any AGN in star-forming galaxies that might change this trend. Since the majority of the AGNs contributing to the trends in Figure 9 are above this limit (see Figure 4), only lower-luminosity AGN might change these trends (see also Appendix B). Currently, there are no indications that such a large population of low-luminosity AGN exists (factor of ∼1000 greater in number than observed), but our results are an additional motivation to search for them.

We restrict our AGN sample by Eddington ratio as we did in Section 4.1 by limiting the AGN sample to those AGN with $L[\mbox{O\,{\sc iii}}]$/M_BH >0.7 (which corresponds to L/L_Edd ∼0.1) in order to test whether the trend we see in Figure 9 is due to a bias against low-mass black holes at a fixed Eddington ratio. Such a bias is unlikely to be the cause of the different trends we see for early- and late-type hosts as it would imply vastly different distributions of Eddington ratios. Imposing this cut in Eddington ratio to Figure 9 yields qualitatively the same trend, though the increase in the AGN fraction for the late-type hosts is less steep.

In summary, dividing the local AGN host galaxy population by morphology reveals that that there are two very different modes of black hole growth: in early types, it is preferentially the least massive black holes that are actively accreting material in an AGN phase. In late-type galaxies, it is preferentially the most massive black holes that are active, with a potential decline at the very highest masses. Since the early-type AGNs are only 11% of the population, this effect can easily be masked by studies of the AGN population as a whole.

As discussed in Section 4, while there are a number of AGNs in early-type galaxies on the red sequence, the fraction of early-type galaxies on the red sequence that host AGNs is very small, i.e., the duty cycle of AGNs in this population is small as well when compared to the peak duty cycle in the green valley.

We see from Figure 6 that the duty cycle of AGNs in early-type host galaxies is peaked in a well-defined band at intermediate green colors precisely in the green valley above the blue cloud and below the low mass end of the red sequence. If the early-type AGN hosts in the green valley are post-starburst objects, then there is clearly an ample supply of star-forming progenitor galaxies at the same mass in the blue cloud. These progenitors would evolve to the green valley if star formation is suppressed, and will continue to fade onto the low-mass end of the red sequence. Such progenitors may already have early-type morphology (e.g., Schawinski et al. 2009b), or they may first transform their morphology in a major merger, as simulations seem to show (e.g., Toomre & Toomre 1972; Barnes 1992; Barnes & Hernquist 1996). Such a major merger may also fuel the black hole, giving rise to the AGN phase observed at a later stage (Sanders et al. 1988; Hopkins et al. 2006). Simulations by Kaviraj et al. (2009) show that minor mergers with a gas-rich dwarf are not sufficient to move a red early-type progenitor to the optical blue cloud (though they may reach UV-blue colors) and so are not a good candidate for the trigger mechanism.

Of course, it does not follow from the intermediate optical colors of green valley galaxies that they must be in transition from blue to red. There are multiple, very different star formation histories that can result in green colors. A slightly enhanced dust screen covering an actively star-forming galaxy would suffice. More complex scenarios are also imaginable. That early-type AGN host galaxies genuinely are a transition population shown by Schawinski et al. (2007); however, this does not imply that the AGN phase in the green valley is necessarily the cause for the shutdown of star formation and the transition from blue to red. Rather, the absence of active black hole accretion in blue cloud objects, established by Schawinski et al. (2009a), implies that there must be a significant time delay on the order of at least 100 Myr, more likely 500 Myr, between whichever process suppresses star formation and the onset of substantial AGN radiation. Furthermore, Schawinski et al. (2009c) showed that the extremely rapid destruction of the molecular gas reservoir that fuels star formation coincides with a very low level of accretion preceding the AGN phase in so-called composite objects²³ and that this very low-luminosity phase—perhaps accompanied by a radiatively inefficient kinetic outflow (e.g., Narayan & Yi 1994)—may be responsible for the shutdown of star formation by removing the molecular gas reservoir.

Regardless of whether any AGN phase is responsible for the shutdown of star formation, the duty cycle of AGN in the present day early-type galaxy population is highest in low-mass green valley objects in which star formation has already been shut down and which are in the process of moving to the red sequence. The duty cycle of AGN in massive, red sequence early types on the other hand is very low. The majority of black hole growth in early-type galaxies at low redshift thus appears to be a post-starburst phenomenon, rather than part of a coeval starburst. Figure 7 furthermore illustrates that it is only this population of low redshift galaxies that has a substantial duty cycle of high Eddington accretion.

The late-type AGN host galaxies are genuine late-type spirals, not S0/Sa galaxies that consist of a large bulge and a small disk, even at the highest masses, where the duty cycle is highest. We show example images of massive late-type AGN hosts in Figure 10. While the host galaxies are quite massive, their bulges are not—this is apparent from their bulge velocity dispersions, and therefore their modest black hole masses, which are comparable to those of active early-type galaxies; they therefore must have substantial disks which make up the remaining stellar mass (Figure 8; Table 2). To put it another way, the typical black hole masses, bulge masses, and Eddington ratios of early- and late-type AGN hosts are similar, but the late-type hosts are substantially more massive due to the presence of a massive disk.

Zoom In Zoom Out Reset image size

At the same time, the distribution of Eddington parameters (Figure 4; Table 2) does not provide a comprehensive picture: comparing Figure 6 shows that the duty cycle of AGN—as we are able to detect them via emission-line diagnostics—is high in high-mass, green late types. However, Figure 7 shows that the duty cycle of high-Eddington-rate accretion is low in those late types, especially when compared to the green valley early types.

Given the distinct location of late-type AGN host galaxies on the color–mass diagram, what role do the AGNs play in the evolution of late-type galaxies? We argued previously that the early-type AGNs represent the true green valley transition population. The majority of AGNs hosted by late-type galaxies do not reside between the blue cloud and the low-mass end of the green valley and therefore cannot be part of the same blue–green–red transition. A minority of the late-type AGN hosts do however have sufficiently low masses to plausibly originate from the blue cloud. The absence of a substantial duty cycle for high-Eddington accretion in these objects further supports this, as the AGN of the early-type hosts do have such a high duty cycle. There is also an absence of AGNs at the very highest masses ≳10^11.5 M_☉ which may indicate a feedback process that has shut down accretion in those objects, or this may be due to insufficient number statistics.

We consider a number of scenarios that may account for the presence of AGNs in late types; any such scenario must account for three observational characteristics: (1) characteristic host mass of ∼10¹¹ M_☉, (2) late-type host galaxy morphology, and (3) intermediate/green host galaxy color.

5.2.1. Major Mergers of Two Disk Galaxies

5.2.2. Mixed Mergers of a Disk and a Spheroid Galaxy

5.2.3. Concluding Remarks concerning Late-type AGN Host Galaxies

The only part of the early-type galaxy population that is experiencing a high duty cycle of black hole growth at high efficiencies are low-mass post-starburst early-type galaxies transitioning from the blue cloud to the low-mass end of the red sequence. This population is continuing the build-up of the red sequence, so it is natural to speculate that the process is a downsized version of the process that led to the formation and quenching of more massive early-type galaxies at high redshift which are known to be older and to have formed on shorter timescales.

The late-type AGNs are not associated with the same transition from blue cloud to the red sequence and their role in the evolution of their hosts is less clear. The majority of the black hole growth in the local universe is associated with late-type hosts (at least ∼40%, and almost 90%, if we include all indeterminate types), and only ∼11% is associated with early types. Thus, the mode of black hole growth at work in late-type galaxies is significantly more common in the local universe than the one at work in early types, and so most black hole growth is not associated with the blue–red transition seen in early types.

Recent observations of X-ray-selected AGN host galaxies in deep fields by Georgakakis et al. (2009) suggest that the fraction of AGN host galaxies with late-type morphology is only ∼30% at z ∼ 1, indicating a decrease of the late-type fraction with increasing redshift, though the AGN population studied by Georgakakis et al. (2009) is somewhat more luminous that in our sample. This decrease may suggest that the appearance of the mode of black hole growth at work in late-type galaxies is relatively recent, perhaps becoming apparent only around z ∼ 1 when the assembly of the Hubble sequence begins to finalize. Another possibility is that this mode has always been active at a moderate level, but that the AGN host galaxy population at high redshift is dominated by the downsizing mode at work in the early-type population. The two modes at work in early- and late-type galaxies, respectively, change in importance with cosmic time, the late-type mode may account for the the observed peak at z ∼ 0.7 in the number density of moderate-luminosity X-ray selected AGNs at z ∼ 0.7, while the mode at work in late types is responsible for the high redshift peak of luminous AGNs (Giacconi et al. 2001; Alexander et al. 2001; Ueda et al. 2003; Gilli et al. 2005). As star formation and black hole growth move to less massive early-type galaxies (e.g., Thomas et al. 2005, 2010), the black hole growth mode in early-type galaxies also becomes less luminous and thus by ≲1 becomes overwhelmed by the late-type mode.

Perhaps the local universe does not tell us much about the AGN-host galaxy connection because black hole growth rates and star formation rates have declined so substantially since z ∼  1 except in a few rare systems more reminiscent of activity at high redshift (e.g., NGC 6240)? Further work on the morphology of AGN host galaxies at high redshift, in particular detailed restframe optical morphological classification, will be required to investigate these suggestion. Near-IR imaging data from the new Wide Field Camera 3 on board Hubble could be used to reproduce this work at z ∼ 2.

Throughout our entire discussion, we have mostly considered the early- and late-type samples as the two morphological extremes. However, both in terms of the entire galaxy sample and the AGN sample, the indeterminate types are the most numerous class (45.3% of galaxies and 44.5% of AGNs; see Tables 1 and 2). Future imaging data, such as from PanSTARRS,²⁴ VST,²⁵ or LSST,²⁶ classified via a future Galaxy Zoo project would enable us to better separate them into early and late types.

Figure 6 shows that the duty cycle for AGNs in late-type galaxies is very strongly peaked at a particular stellar mass and color. The Milky Way most likely occupies this sweet spot for late-type AGNs. Estimates of the stellar mass of the Milky Way vary, but are around ∼10¹¹ M_☉, while the current star formation rate is approximately 3 M_☉ yr⁻¹ (e.g., Scalo 1986; Diehl et al. 2006; Calzetti et al. 2009). These parameters place the Milky Way right where the AGN fraction for late-type galaxies is highest, between 5% and 10% and therefore the duty cycle is the highest. However, this also implies that the black holes in galaxies like the Milky Way are not active ∼92%–95% of the time at the luminosity of the typical Seyfert in the SDSS universe (∼10⁴⁰ erg s⁻¹; see Sections 3.3 and 3.4).

The black hole at the center of the Milky Way at Sagittarius A* (Sgr A*) is, apart from occasional weak flares, remarkably quiet with an extremely sub-Eddington accretion rate (e.g., Marrone et al. 2006, 2008). However, there is evidence that Sgr A* was substantially more active in the very recent past. Hard X-ray observations with ASCA showed Fe Kα lines around Galactic Center emitted by the molecular cloud complex surrounding the black hole. Follow-up observations with more recent hard X-ray facilities indicate that Sgr A* was orders of magnitude, perhaps 10⁶ times more luminous in the last few hundred years than it is currently. Suzaku observations by Nobukawa et al. (2008) detect a Kα luminosity consistent with Sgr A* having been at 10³⁸–10³⁹ erg s⁻¹ a mere 300 yr ago, while Revnivtsev et al. (2004) used INTEGRAL data to argue that Sgr A* was at a 2–200 keV luminosity of L ∼ 1.5 × 10³⁹ erg s⁻¹ just 300–400 years ago. This is not quite the luminosity necessary to excite a narrow-line region sufficiently luminous to be detected by an SDSS fiber at z ∼ 0.02, but these observations clearly imply that Sgr A* is active at slightly lower luminosity on short timescales. Together with the duty cycle for Milky Way-like galaxies derived here, these observations may give a first hint of the lifetime of AGNs in such galaxies. A caveat to this statement is the fact that the luminosity of this very recent flare is still below the AGN luminosity limit of our sample.

Under what conditions would the black hole at the center of the Milky Way reach a sufficient luminosity so that it would be included in our AGN sample? To reach a typical luminosity of a few percent Eddington $({\sim} \rm {few} \times 10^{42}$ erg s⁻¹ bolometric luminosity) requires a mass-accretion rate of only 10⁻³ M_☉ yr⁻¹. This only a factor of 100 higher than Bondi accretion rate inferred from simulations, and there are 10⁴ M_☉ of molecular gas within 2 pc of the central black hole (Herrnstein & Ho 2002), with large clouds at further distances. It would not necessarily be surprising if dynamical friction or a minor merger, e.g., the swallowing of the Sagittarius dwarf galaxy drove some of these clouds in, temporarily re-igniting the black hole at the center of the Milky Way. All the ingredients thus appear to be on hand for turning our galaxy into a Seyfert AGN. Indeed, the mystery may rather be: why is our central black hole so dark and why is gas accretion so infrequent?

The implication of this work is that the Milky Way is in a location on the color–mass diagram where late-type galaxies have the highest duty cycle for AGN phases. The mass of the Milky Way black hole has been inferred to be 4.1 × 10⁶ M_☉ (Ghez et al. 2008) which places it in the typical range of black hole masses in late-type galaxies that are currently growing in the local universe (Figures 8 and 9). Thus, 5%–10% duty cycle for galaxies like the Milky Way places into context the current quiescence of Sgr A*, together with the evidence for accretion in the recent past and apparently favorable conditions for a resumption of accretion.