A CENSUS OF BROAD-LINE ACTIVE GALACTIC NUCLEI IN NEARBY GALAXIES: COEVAL STAR FORMATION AND RAPID BLACK HOLE GROWTH

The SDSS provides magnitudes in five ugriz filters for all broad-line AGNs and inactive galaxies. We use both the total magnitude m integrated across the entire galaxy,⁴ and the inner aperture magnitude m_in measured within a 3'' diameter of the galaxy center.⁵ An outer aperture magnitude m_out is calculated from the light outside the 3'' diameter, given by

$$\begin{equation} m_{{\rm out}} = -2.5\log ( 10^{-0.4m}-10^{-0.4m_{{\rm in}}}). \end{equation} \tag{ 1 }$$

Note that m_in is calculated after convolving the image to 2'', which ensures uniform seeing for all objects. While convolving to lower resolution slightly worsens AGN contamination, it actually helps our AGN/galaxy decomposition by ensuring that all active galaxies have the same 2'' resolution as the point-source stars and inactive galaxies used to calibrate the method.

We K-correct the observed photometry in both samples to the z = 0.05 frame using the public kcorrect IDL software (Blanton & Roweis 2007). The prime (') notation is used to denote colors and magnitudes K-corrected to z = 0.05. Figure 1 shows the K-corrected outer (u − z)' color with the outer M'_z absolute magnitude for inactive galaxies and broad-line AGNs in the faint sample, and Figure 2 similarly shows the luminous sample. Naively one might assume that the outer magnitudes do not contain any light from the AGN point source. However, both figures show that many broad-line AGNs have brighter and bluer outer magnitudes than the inactive galaxy population. Light from the SDSS point-spread function (PSF) extends beyond the inner 3'' aperture, and a more sophisticated process is necessary to recover the uncontaminated galaxy properties. This technique is derived and applied in Section 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Galaxy stellar masses come from the MPA-JHU value-added catalog, derived according to Kauffmann et al. (2003b). Derived masses for the inactive galaxies have 1σ errors of ∼0.05 dex, with no error dependence on mass or magnitude. However, most of the broad-line AGN hosts are absent from this catalog, and those included in the catalog probably have inaccurate stellar masses due to the AGN contamination. To estimate masses for AGN hosts, we first calculate the mass-to-light ratios of inactive galaxies as a function of color and luminosity.

Figure 3 shows the median z'-band mass-to-light ratio in bins across the color–magnitude diagram. We use this figure to estimate masses for broad-line AGN hosts, applying the mass-to-light ratio from the bin corresponding to their corrected (AGN-subtracted) host galaxy color and absolute magnitude.

Zoom In Zoom Out Reset image size

We begin by comparing the inner (<3'') and outer (>3'') aperture magnitudes of inactive galaxies and broad-line AGNs. Figure 4 shows inner and outer magnitudes in three different filters (u, r, z) and redshift ranges (0.02 < z < 0.03, 0.04 < z < 0.05, 0.08 < z < 0.09). Light from broad-line AGNs has the strongest contaminating effect in blue light and inner magnitudes, but also affects outer magnitudes. For each of the five filters and in 10 bins of redshift (0.01 < z < 0.11 in Δz = 0.01 intervals), we find the best-fit line describing outer (m_{out, GAL}) and inner magnitude (m_{in, GAL}) for inactive galaxies (shown by the dashed lines for the filters and redshift ranges in Figure 4). The offset above this line is defined by

$$\begin{equation} {\Delta }m = A + B m_{{\rm out,GAL}} - m_{{\rm in,GAL}} . \end{equation} \tag{ 2 }$$

The best-fit line is given by Δm = 0, and broad-line AGNs typically have Δm > 0. However, even inactive galaxies have a large scatter in Δm, presumably because there is significant structural variation in galaxies of a given outer apparent magnitude. Adding a structural measurement could better describe the typical relationship for inner and outer magnitudes in inactive galaxies and explain their scatter about Δm = 0. In particular we use z-band concentration, defined as the ratio between the radius containing 90% of the z-band light and the radius containing 50% of the light, C_z = R_{90, z}/R_{50, z}.

Zoom In Zoom Out Reset image size

Figure 5 shows Δm from Equation (2) versus z-band concentration C_z. Galaxies with Δm < 0 (from fainter inner magnitudes than the average given their outer magnitude) have low concentration, while galaxies with Δm > 0 (and brighter inner magnitudes) are more concentrated. The inactive population shows a tighter relation after using C_z to describe the structural variation, and broad-line AGNs scatter to brighter inner magnitudes and higher Δm.

Zoom In Zoom Out Reset image size

A cubic line is fit to the inactive galaxies in Figure 5 (shown by the dashed line), with the offset from this line given by

$$\begin{equation} {\delta }m = {\Delta }m - C - D C_z - E C_z^2 - F C_z^3. \end{equation} \tag{ 3 }$$

Figure 6 shows δm with C_z for inactive galaxies and AGNs. Broad-line AGN hosts tend to have δm > 0, as the AGN light causes the inner aperture magnitude to be brighter than expected given the galaxy's z concentration. Inactive galaxies have δm ∼ 0 with small scatter: the standard deviation is typically only σ_δm = 0.3 mag, with slightly larger σ_δm = 0.5 mag scatter for low-concentration (C_z < 2.5) galaxies in the u band. We use δm as an estimate of the brightness excess in the inner aperture due to AGN light.

Zoom In Zoom Out Reset image size

Using Equation (3) for AGN/host decomposition has two important assumptions. First, we assume that C_z is not contaminated by the broad-line AGN. We originally chose C_z with this in mind: z-band light is the least affected by the (typically blue) AGN, and R₅₀ and R₉₀ are large radii well beyond the point-source AGN. Figure 7 directly tests if the AGN affects C_z by plotting AGN luminosity against z-band concentration and comparing the distributions of C_z among broad-line AGNs and inactive galaxies. In general, both the faint and luminous AGN samples span a wide range of concentrations. However, the most luminous (M'_{u, AGN} < −20) AGNs are typically more concentrated: this may be evidence that very luminous AGNs contaminate the concentration measurement, or it may be that more luminous AGNs prefer more bulge-like hosts (Bahcall et al. 1997; Cisternas et al. 2011; Kocevski et al. 2012). In case it is the result of a bias, we flag these high-luminosity AGNs when discussing any connections between AGN strength and host galaxy properties. Meanwhile we conclude that our assumption that C_z is unaffected by the AGN remains valid for the bulk of broad-line AGNs with M'_{u, AGN} > −20.

Zoom In Zoom Out Reset image size

The second assumption is that all inactive galaxies have δm ∼ 0 with some random scatter σ_δm, independent of the presence of an AGN. There is evidence that this scatter is not due to measurement error: as Figure 8 shows, inactive galaxies have values of δm that are correlated across the ugriz filters. In other words, an inactive galaxy with δm ∼ 1 in the u band will also tend to have δm ∼ 1 in g, r, i, z. (This effectively means that the δm = 0 assumption causes a much smaller scatter in color than in luminosity or stellar mass: see Section 3.4) We tested if, in addition to C_z, δm was connected to galaxy properties like color or Sérsic (1968) index. However, we found no additional correlation and were unable to find the physical basis for the small scatter of inactive galaxies about δm = 0. Defining δm using color and Sérsic index instead of C_z also proved ineffective at reducing the scatter. Instead σ_δm is treated as a random error, and we investigate its effects in Section 3.4.

Zoom In Zoom Out Reset image size

Inactive galaxies with δz > 1 (in the upper left of the right panels in Figure 6) are potentially interesting because they have similar relationships between inner and outer magnitudes to bright AGNs. However, after visually inspecting their images and spectra we determined that these galaxies probably have nuclear starbursts and are not some class of misclassified broad-line AGNs.

Most of the light from a point source is detected in the inner aperture, but the PSF of the SDSS causes light from point sources to "leak" into the outer aperture as well. Stars can be used to model the relationship between inner and outer aperture magnitudes for AGN-only light (m_{in, AGN} and m_{out, AGN}), since both are point sources. We select a sample of 362 stars with r < 16 and −0.2 < r − z < −0.1. The brightness ensures their photometry is well measured, and the color cut makes them similar to the colors of bright AGNs.

Figure 9 shows the inner and outer magnitudes for the sample of 362 stars. These point sources have a well-defined relationship between inner and outer aperture magnitude, and the best-fit line in each filter describes the relationship between inner and outer magnitude for the AGN-only light:

$$\begin{equation} m_{{\rm in,AGN}} = G + H m_{{\rm out,AGN}} . \end{equation} \tag{ 4 }$$

Figure 9 shows that the brightest AGNs behave like point sources in blue light and lie on this line, while in red light AGNs behave more like inactive galaxies. Because all the images are convolved to 2'' before measuring the aperture photometry, m_{in, AGN} ≈ 1 + m_{out, AGN} in all five filters.

Zoom In Zoom Out Reset image size

Given the derived relationships between inner and outer magnitudes for galaxy-only (Equations (2) and (3)) and AGN-only (Equation (4)) light, we can solve for the unknown quantities m_{in, AGN}, m_{out, AGN}, m_{in, GAL}, and m_{out, GAL}. First, the total outer and inner magnitudes (both measured quantities) are simply the sum of the flux contributions from the AGN and the galaxy:

$$\begin{equation} f_{{\rm in}} = f_{{\rm in,AGN}} + f_{{\rm in,GAL}} \end{equation} \tag{ 5 }$$

$$\begin{equation} f_{{\rm out}} = f_{{\rm out,AGN}} + f_{{\rm out,GAL}}. \end{equation} \tag{ 6 }$$

The relationship between flux and AB magnitude is defined by m ≡ −2.5log (f_ν) − 48.6. Combining Equations (2) and (3) and converting to flux results in

$$\begin{equation} f_{{\rm in,GAL}} = X \cdot f_{{\rm out,GAL}}^B, \end{equation} \tag{ 7 }$$

with $X = 10^{-0.4(A + 48.6 - 48.6B - C - D C_z - E C_z^2 - F C_z^3)}$.

Equation (4) is similarly converted from magnitude to flux:

$$\begin{equation} f_{{\rm in,AGN}} = Q \cdot f_{{\rm out,AGN}}^H, \end{equation} \tag{ 8 }$$

with Q = 10^{−0.4(G + 48.6 − 48.6H)}.

There are now four equations (Equations (5)–(8)) that describe the relationships between inner and outer magnitudes for each of the broad-line AGN and galaxy contributions. The quantities f_in, f_out, and C_z are measured, and the coefficients A, B, C, D, E, F, G, and H come from our line fits in each filter and redshift bin (with three example filters and redshift bins shown in Figures 4–6). Solving these four equations for f_{in, AGN} results in

$$\begin{equation} f_{{\rm in,AGN}} = f_{{\rm in}}-X [f_{{\rm out}}-(f_{{\rm in,AGN}}/Q)^{1/H}]^B \end{equation} \tag{ 9 }$$

(with the constants X and Q given in Equations (7) and (8)). We numerically solve this equation using a bisector root-finding method. After solving for f_{in, AGN}, the other three unknowns can be computed from Equations (5), (6), and (8). This process is repeated to derive AGN-subtracted host galaxy magnitudes in all five bands in each of the 10 redshift bins. Table 1 shows the corrected magnitudes and derived stellar masses for the 561 broad-line AGNs in the luminous and faint samples.

It is necessary to ensure that our AGN/host decomposition method does not introduce errors which bias the resultant host galaxy colors and stellar masses. One assumption of our method is that galaxies have δm = 0: Figure 6 demonstrates that this is true on average, but the δm values of individual galaxies scatter about this value. Similarly the true inner and outer magnitudes of AGN-subtracted host galaxies may have nonzero values of δm, and our correction to δm = 0 may introduce systematic errors in their resultant colors and masses.

We investigate the errors of the δm = 0 assumption by applying the decomposition method to inactive galaxies. Figure 10 shows the changes in (u − z)' color and M_* introduced by "correcting" inactive galaxies to δm = 0. The dispersions in these shifts represent the errors in color and mass from correcting AGN host galaxies with nonzero δm. Errors in color are small because δu and δz are correlated (see Figure 8). The dispersion in stellar mass is slightly larger, as indicated by the error bar in the bottom right of each panel, but mass shifts and dispersions are small among blue cloud and green valley galaxies (the typical hosts of broad-line AGNs). From Figure 10, we do not expect the δm = 0 assumption to bias the observed colors and masses of broad-line AGNs.

Zoom In Zoom Out Reset image size

If our other assumption is incorrect and AGNs contaminate z-band concentration, this could also introduce significant errors in the AGN/host decomposition method. Figure 7 shows that the most luminous AGNs (M'_{u, AGN} < −20) may bias C_z, and such systems may have larger systematic or random errors than those estimated in Figure 10. For this reason we eliminate the most luminous AGNs from the discussion and conclusions below. However, the low and moderate luminosity broad-line AGNs that make up the bulk of our sample do not exhibit biased C_z. For these objects the assumption that AGNs do not affect z-band concentration is unlikely to cause errors in the AGN/host decomposition.

While broad-line AGNs are the brightest and most rapidly accreting AGNs (e.g., Kollmeier et al. 2006; Trump et al. 2009), many of the weakest broad-line AGNs have significant galaxy contribution to their continua. Dim AGNs could be more difficult to detect in bright galaxies, with the broad lines diluted by host galaxy starlight. More distant galaxies also contain more starlight within the 3'' spectroscopic fiber aperture and could similarly dilute the broad emission lines of AGNs.

We test the potential selection bias against dim AGNs in luminous host galaxies by computing the AGN fraction across the color–mass diagram for dim, moderate, and luminous AGNs, as shown in Figure 13. AGN strength is quantified using the AGN-only u-band absolute magnitude, with M_{u, AGN} divisions chosen to put the same number of AGNs in each set. AGNs were drawn from the faint sample because it includes a large range in stellar mass. From Figure 13 it is clear that dim AGNs are not found less frequently in massive, luminous galaxies, and so this potential selection effect does not occur in our sample.

Zoom In Zoom Out Reset image size

Figure 14 similarly tests the potential selection effect of higher distance biasing against AGN identification. The luminous sample is most appropriate for this test because it includes luminosity-limited AGNs over the full redshift range 0.01 < z < 0.11. In addition to measuring the broad-line AGN fraction over all galaxies in each redshift bin, a set of "massive" galaxies is also defined with log (M_*/M_☉) > 9.5 + 0.5(u − z)'. This mass limit corresponds to the most massive 10% of galaxies along the blue cloud and green valley.

Zoom In Zoom Out Reset image size

Over the limited 0.01 < z < 0.11 redshift range of our sample, there is no evidence for a decline in the fraction of galaxies hosting a broad-line AGN at higher distance. For this reason we do not expect distant galaxies to bias against AGN identification through starlight dilution in the spectroscopic aperture.

A major finding of this study is that broad-line AGNs tend to lie in hosts with young stellar populations, and avoid red and quiescent host galaxies. The simplest explanation for this is if the same gas that fuels rapidly accreting, unobscured nuclear activity also rapidly formed stars in the recent past. Figure 13 suggests that the more powerful the broad-line AGN, the bluer the host galaxy: more rapidly accreting AGNs may have more powerful recent star formation in their host galaxy. Once again, this suggests that star formation and nuclear activity are fueled by the same material.

Figure 15 directly compares the broad-line AGN luminosity to the color and stellar mass of the host galaxy. AGN luminosity is quantified by AGN-only M'_u. In addition to corrected color and mass for broad-line AGN hosts, Figure 15 also shows the offsets in each of these quantities from inactive galaxies of similar (within 0.1 dex) mass or color. These offsets are useful to remove the degeneracy between a galaxy's mass and its color. For example, the apparent anti-correlation between stellar mass and AGN luminosity in panel (c) is a result of the degeneracy between mass and color. Comparing AGNs and inactive galaxies of similar color, panel (d) shows no significant trend between stellar mass offset and AGN luminosity.

Zoom In Zoom Out Reset image size

Panel (b) in Figure 15 shows that more luminous AGNs have bluer host galaxies in both the luminous and faint samples. This may represent a physical connection between AGN luminosity and the host galaxy's recent star formation history. AGN luminosity probably translates to accretion rate, and so the relationship between AGN M'_u and host color indicates that AGN accretion rate is correlated with the number of recently formed stars. These observations favor scenarios with similar fuel sources for both star formation and nuclear activity (e.g., Salim et al. 2007; Silverman et al. 2009), with little or no time delay between the processes fueling each.

This interpretation is robust even given the limitations of our AGN/galaxy decomposition method. Recall the assumption that AGN host galaxies have the same light distributions as inactive galaxies (quantified by δm = 0). Given the M − σ relation (e.g. Park et al. 2012) and the requirement of a bulge to host an AGN, it is unlikely that broad-line AGNs have hosts with intrinsic δm < 0. On the other hand, the unusual blue colors of broad-line AGN hosts might indicate nuclear starbursts and intrinsic δu > 0. Applying our decomposition method to an AGN host with a nuclear starburst would oversubtract the blue light, resulting in a redder corrected galaxy (u − z)' color. With the presence of a nuclear starburst it is possible that the galaxy colors of AGN hosts may be even bluer than we observe. Similarly the connection between AGN luminosity and blue host color in Figure 15 would steepen if these blue hosts contain nuclear starbursts and unusually concentrated blue light.

Another potential concern is that the broad-line AGN host galaxies appear blue due to extended scattered light from the AGN. Zakamska et al. (2006) noted that in extreme cases Type 2 quasars with M_{B, AGN} < −24 have scattered light on >1 kpc scales that dominates over the host galaxy starlight. Measuring extended scattered light for our sample is beyond the scope of this work, given the requirement for long-slit spectropolarimetry on a large number of broad-line AGNs with varying luminosities. However, we note that our broad-line AGNs are several (4–7) magnitudes fainter than the Zakamska et al. (2006) sample, and so we assume the amount of extended scattered light in this work to be negligible.

Our large sample of broad-line AGNs from the SDSS shows a strong preference for blue host galaxies. This matches earlier studies with small broad-line AGN samples, which found a similar preference for hosts with young stellar populations (Bahcall et al. 1997; Jahnke et al. 2004a, 2004b). The most [Oiii]-luminous narrow-line AGNs also prefer star-forming host galaxies (Kauffmann et al. 2003a; Silverman et al. 2009). These classes of luminous AGNs all represent "feast mode" rapid accretion with high Eddington ratios (λ_Edd): Kauffmann & Heckman (2009) estimate that the most [Oiii]-luminous narrow-line AGNs have λ_Edd ∼ 0.1, and broad-line AGNs also have Eddington ratios of 1%–100% (Kollmeier et al. 2006; Trump et al. 2009). In future work we will directly measure and compare λ_Edd in our broad-line AGN sample, but for now we simply assume that they are rapidly accreting AGNs with λ_Edd > 0.01.

The blue hosts of our broad-line AGNs are in contrast to the hosts of fainter narrow-line AGNs, which instead prefer green valley galaxies (Nandra et al. 2007; Salim et al. 2007; Georgakakis et al. 2008; Silverman et al. 2008; Gabor et al. 2009; Schawinski et al. 2009; Hickox et al. 2009; Kocevski et al. 2009) or all galaxy types equally (Xue et al. 2010). The host-dominated AGNs of these previous studies are, by construction, fainter than broad-line AGNs, due to either obscuration or lower accretion rates (λ_Edd < 0.01). Different host galaxies for broad-line AGNs and weaker host-dominated AGNs are incompatible with the historical AGN unified model (Antonucci et al. 1993). In its simplest version, the unified model uses only geometrical orientation to explain the different observed properties of broad-line and host-dominated AGNs. Both broad-line and host-dominated AGNs would then be observed in the same host galaxy types, as has been observed among intermediate-luminosity AGNs at z ∼ 1 (Ammons et al. 2011). Our observations instead favor AGN unified models where broad-line and faint narrow-line AGNs have physically different accretion properties (Ho 2008; Trump et al. 2011; Antonucci et al. 2011). Broad-line AGNs might be in star-forming galaxies because such galaxies have the most abundant or efficient fuel supply for the SMBH, while other galaxy types can fuel only "famine-mode" host-dominated AGNs with low accretion rates (Kauffmann & Heckman 2009; Schawinski et al. 2010).

Kauffmann et al. (2003a) used a large sample of host-dominated AGNs to show that there may be a minimum stellar mass of log (M_*/M_☉) > 10.5 required for a galaxy to host an AGN. Recently, however, Aird et al. (2012) argued that the apparent preference for massive AGN hosts is caused by selection effects, and there is instead a universal Eddington-ratio distribution of AGNs in galaxies of any stellar mass. Figure 16 shows the percentage of broad-line AGNs with host stellar mass for both the luminous and faint samples. Given the narrow distribution of high Eddington ratios for broad-line AGNs (0.01 < L/L_Edd < 1, e.g., Kollmeier et al. 2006; Trump et al. 2009), our AGN sample is complete to log (M_*/M_☉) ∼ 10 (Kelly & Shen 2012). Thus the luminous broad-line AGN sample (also limited by log (M_*/M_☉) > 10) provides a direct test of the AGN dependence on host stellar mass. After controlling for galaxy color (right panel of Figure 16), we find a flat AGN fraction with stellar mass. This suggests that there is no stellar mass threshold for broad-line AGNs, in agreement with Aird et al. (2012).

Zoom In Zoom Out Reset image size

Theoretical simulations invoke feedback from luminous AGNs to rapidly quench star formation and transform their host galaxy colors from blue to red (Silk & Rees 1998; Fabian 2002; Di Matteo et al. 2005). Several authors have claimed observational evidence for this scenario, with host-dominated AGNs apparently preferring recently quenched green valley galaxies (Nandra et al. 2007; Georgakakis et al. 2008; Schawinski et al. 2009; Hickox et al. 2009; Kocevski et al. 2009). Since radiative-mode feedback scales with AGN strength, the broad-line AGNs in our sample would be expected to have an even greater effect in quenching star formation. However, we find precisely the opposite effect: more powerful AGNs seem to lie in bluer galaxies. If quasar winds cause significant feedback, then their galaxy-wide effects are not visible until after the broad-line AGN disappears.

Figure 17 outlines three scenarios connecting AGNs and galaxy quenching: scenario A has star formation quenched before the broad-line AGN phase, scenario B quenches during the broad-line AGN peak, and scenario C has star formation quenched well after the broad-line AGN phase. The increase in star formation during the AGN peak in all scenarios is motivated by our observed connection between AGN luminosity and blue host galaxy color. The broad-line AGN phase occurs only during the most luminous AGN activity, which means that our study probes only the region limited by the dashed "BL AGN Visible" lines in each panel.

Zoom In Zoom Out Reset image size

Scenario A is immediately ruled out by the observed absence of broad-line AGNs among quenched red galaxies. However, our observations cannot distinguish between quenching during or after the broad-line AGN phase (scenarios B and C). The broad-line AGN lifetime is less than ∼100 Myr (Martini 2004), which is roughly the same amount of time it takes for galaxy colors to change from blue to green or red after star formation quenches (e.g., Bruzual & Charlot 2003). Even if quenching occurs during the AGN peak, a change in galaxy colors would not be observed until after the broad-line AGN fades. This is evident in the scenario B curve in the right panel of Figure 17. So while our observations exclude AGN quenching scenarios that occur before the broad-line AGN phase, only observations of fading AGNs can distinguish between simultaneous or delayed AGN quenching.