DIRECT EVIDENCE FOR TERMINATION OF OBSCURED STAR FORMATION BY RADIATIVELY DRIVEN OUTFLOWS IN REDDENED QSOs

The last decade has seen substantial progress in understanding how galaxies assemble their stars and central supermassive black holes (SMBHs). Recent results almost all point to the same conclusion that the bulk of galaxy assembly occurred at z ≳ 0.7, and that a significant fraction of it took place in obscured "bursts" of intense star formation and SMBH accretion. Indirect evidence for this comes from, for example, the stellar ages of low-redshift galaxies (Heavens et al. 2004), the existence of evolved elliptical galaxies at high redshifts (Dunlop et al. 1996; Ellis et al. 1997; Rakos et al. 2007), and studies of stellar mass assembly over wide redshift ranges (Glazebrook et al. 2004; Fontana et al. 2006; Pérez-González et al. 2008; Marchesini et al. 2009; Ilbert et al. 2010). Direct evidence mostly comes from extragalactic infrared and millimeter imaging surveys, which find that the number density of IR-luminous galaxies increases dramatically with increasing redshift (e.g., Le Floc'h et al. 2005; Chapman et al. 2005; Babbedge et al. 2006; Coppin et al. 2006; Shupe et al. 2008; Austermann et al. 2010; Eales et al. 2010), and that much of the growth period of SMBHs was shrouded in dust (Martínez-Sansigre et al. 2005; Alexander et al. 2008).

The importance of obscured starburst and active galactic nucleus (AGN) activity in assembling galaxies suggests that the two phenomena may affect each other. A link between them is implied by, for example, the tight correlation between the mass of the SMBH and stellar bulge mass (Magorrian et al. 1998; Ferrarese & Merritt 2000; Tremaine et al. 2002; Marconi & Hunt 2003; Shields et al. 2003; Häring & Rix 2004), though the correlation probably does not hold for dark matter halo mass (Kormendy & Bender 2011, but see also Ferrarese 2002). A link is also implied by the coeval presence of both starburst and AGN activity in many IR-luminous systems at all redshifts (see, e.g., Blain et al. 2002; Lagache et al. 2005; Lonsdale et al. 2006; Hernán-Caballero et al. 2009).

Recently, interest into the relationship between starburst and AGN activity has been stimulated by what first appeared to be irreconcilable differences between observational and theoretical results on the assembly history of galaxies. Foremost among these were the difficulties that models faced in explaining the observed galaxy luminosity function (LF) at low and high redshifts simultaneously; if the models were tuned to match the local galaxy LF then they underpredicted the number of massive galaxies observed at high redshifts, whereas the models that matched the number of high-redshift galaxies gave poor fits to the local galaxy LF (Kauffmann et al. 1999; Cole et al. 2000; Somerville et al. 2001; Benson et al. 2003). Other problems included (1) the prediction that a higher fraction of rich galaxy clusters should harbor cooling flows than is observed (Peterson et al. 2003; Xu et al. 2002; Peterson & Fabian 2006), (2) the difficulties that early semi-analytic models faced in reproducing the large number of IR-luminous galaxies observed at high redshift (e.g., Baugh et al. 2005), and (3) the expectation that the mass return rate from stars would cause central SMBHs to be larger than is observed (Ciotti et al. 1991; Ciotti & Ostriker 2007).

Several solutions have been proposed to resolve these issues. Among the most promising is an idea termed "AGN feedback." AGN feedback is the exertion of influence of an SMBH on kpc to Mpc scales to curtail star formation in the host galaxy, and/or further accretion onto the SMBH itself.¹² This can occur in four ways. Radiation from regions immediately local to the SMBH can (1) heat gas in the interstellar medium (ISM) so it cannot collapse to form stars, and/or (2) radiatively drive gas out of the galaxy (or, equivalently, stop it falling in from the intergalactic medium), thus emptying the galaxy of fuel for further star formation. Additionally, matter expelled from regions immediately local to the SMBH can (3) heat gas in the ISM, and/or (4) drive gas out of the galaxy via kinetic pressure.

Most discussions of AGN feedback condense the four feedback mechanisms described above into two simplified paradigms; "quasar" mode feedback and "radio" mode feedback. Quasar mode feedback assumes that radiation from an accretion disk terminates star formation in the host galaxy, usually by coupling some fraction of the QSO luminosity to kinetic energy injected into the ISM. Quasar mode feedback is thought to be short term, acting for only ∼10⁸ years over which the quasar is accreting rapidly. Radio mode feedback on the other hand is assumed to occur via a relativistic jet that transfers momentum to the ISM. Radio mode feedback can occur over longer timescales (∼10⁹ years) than quasar mode as it requires only a low accretion rate to produce jets sufficiently powerful to affect the ISM. Nevertheless, radio mode and quasar mode feedback are not mutually exclusive.

For individual galaxies, AGN feedback is predicted to have a significant impact, both from analytic (Fabian 1999; Wyithe & Loeb 2003; Sazonov et al. 2005; Fabian et al. 2006; Pope 2009; Power et al. 2011; Kaviraj et al. 2011) and numerical studies (Omma et al. 2004; Ciotti & Ostriker 2007; Antonuccio-Delogu & Silk 2008; Tortora et al. 2009; Hopkins & Elvis 2010; Ostriker et al. 2010; Wagner & Bicknell 2011; Hambrick et al. 2011). For galaxy mergers, quasar mode feedback is predicted to have a profound influence (Di Matteo et al. 2005; Springel et al. 2005; Debuhr et al. 2011; Snyder et al. 2011). Models for the cosmological evolution of galaxies and clusters have incorporated one or both of these feedback paradigms, often with noticeable improvements in reproducing observations. These include semi-analytic models using quasar mode (Granato et al. 2004; Menci et al. 2008), radio mode (Bower et al. 2006, 2008; Gonzalez-Perez et al. 2009), or both (Somerville et al. 2008), and numerical models with quasar mode (Booth & Schaye 2009; Sales et al. 2010; McCarthy et al. 2011; Teyssier et al. 2011; Chatterjee et al. 2011), or both forms of feedback (Croton et al. 2006; De Lucia et al. 2006; Sijacki et al. 2007; Kitzbichler & White 2007; Monaco et al. 2007; Puchwein et al. 2008; Short & Thomas 2009; Gaspari et al. 2011).

Observationally, evidence for AGN feedback is mounting. Studies of early-type galaxies suggest that some form of feedback may have terminated their star formation (Schawinski et al. 2007; Roseboom et al. 2009; Kormendy et al. 2009, but see also Shin et al. 2011). Far-IR observations have also found powerful, plausibly AGN-driven outflows in OH⁻ and CO 1–0 (Feruglio et al. 2010; Chung et al. 2011; Sturm et al. 2011) in local ultraluminous infrared galaxies (ULIRGs) and QSOs that could exhaust the fuel supply for star formation within ∼10⁷ years, while X-ray observations have found mildly relativistic outflows in radio-quiet and radio-loud AGNs, with an origin close to the SMBH (Chartas et al. 2003; Braito et al. 2007; Tombesi et al. 2010). Further observations have provided indirect evidence for "quasar mode" feedback in some galaxies (Dunn et al. 2010; Alexander et al. 2010; Müller-Sánchez et al. 2011; Rupke & Veilleux 2011), and for radio mode feedback in galaxies and clusters (Best et al. 2006; McNamara & Nulsen 2007; Mittal et al. 2009; Nesvadba et al. 2010; Werner et al. 2010, 2011; Ehlert et al. 2011; Shabala et al. 2011; Ma et al. 2011). Controversies do, however, remain. For example, Jahnke & Macciò (2011) suggest that there is no need for AGN feedback to explain the bulge–SMBH mass relation, Ammons et al. (2011) suggest that only a small fraction of intermediate-luminosity AGNs at z ∼ 1 can be undergoing galaxy-wide AGN feedback, and Lutz et al. (2010) find that there is no simple inverse relation between star formation rate and AGN luminosity in X-ray selected AGNs. It is also worth mentioning that the effectiveness of AGN feedback may depend on the mode of star formation, since spatially compact, "bulge" star formation is likely easier to turn off than spatially extended, "disk" star formation.

Our group has been examining the role of AGN feedback by looking at systems in which such feedback may be ongoing. To do so, we have been studying the "FeLoBAL" class of QSOs (Hazard et al. 1987), which comprise part of the broad absorption line (BAL) QSO population.¹³ We selected this population for three reasons. First, the UV absorption troughs are unambiguous signatures of radiatively driven outflows powered by an AGN. They have velocities of up to 0.2c, widths of at least a few thousand km s⁻¹, and are usually very deep or black, implying high column densities moving at high velocities, which plausibly implies high mass-loss rates (Arav et al. 1994; Proga et al. 2000; Chartas et al. 2003; Crenshaw et al. 2003). Therefore, they cannot be driven by even the most extreme starbursts. Furthermore, in at least some FeLoBAL QSOs the outflows may extend up to several kpc into the host galaxy (Arav et al. 2008; Moe et al. 2009; Bautista et al. 2010; Dunn et al. 2010). Second, they are invariably reddened objects with high IR luminosities (Farrah et al. 2007), and sometimes harbor intense starbursts (Farrah et al. 2010), suggesting they may be a transition phase in the lifetime of an AGN (FeLoBAL features are also occasionally seen in ULIRGs, see, e.g., Farrah et al. 2005). Third, recent results have shown that FeLoBAL QSOs are much more common at z ≳ 0.5 than was originally thought, by up to a factor of 10 (Dai et al. 2008; Urrutia et al. 2009; Allen et al. 2011).

In this paper, we compare the strengths of the AGN-driven outflows to the luminosities of obscured star formation in a sample of 31 FeLoBAL QSOs, selected purely on the basis of their rest-frame UV spectral properties. We combine data from the Spitzer Space Telescope (Werner et al. 2004) with data from the Sloan Digital Sky Survey (SDSS; York et al. 2000), the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006), the UKIRT Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007), and from the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010; Jarrett et al. 2011) in order to measure the luminosities of both obscured AGN activity and star formation. We measure the strengths of the AGN-driven outflows from the UV absorption troughs in the SDSS spectra. We assume a spatially flat cosmology, with H₀ = 70 km s⁻¹ Mpc⁻¹, Ω = 1, and Ω_Λ = 0.7. We use the term "IR luminosity" to refer to the luminosity integrated over 1–1000 μm in the rest frame. We quote luminosities in units of bolometric solar luminosities, where L_☉ = 3.826 × 10²⁶ watts.

2.1. Sample Selection

2.2. Observations

The MIPS photometry is presented in Table 2. We combine the MIPS data with archival photometry from SDSS, 2MASS (or UKIDSS if available), and WISE. These archival data are presented in Table 1. Additional submillimeter photometry for two objects is presented in Lewis et al. (2003).

As WISE returned data only recently, we provide more detail on the WISE data. WISE completed its first full coverage of the sky in 2010 July. A Preliminary Release Catalog¹⁴ covering 57% of the sky, was made available to the community starting 2011 April 14. Point sources were extracted with 5σ sensitivities greater than 0.08, 0.11, 0.8, and 4 mJy, respectively, for the four bands. Photometric calibration, source counts, colors, and population statistics at the ecliptic poles are presented by Jarrett et al. (2011). Of our 31 sources, 10 come from the Preliminary Release Catalog, and the rest come from the proprietary WISE First Pass Internal Source Database, which performs internal verification, quality, and photometric analysis. In the cases where there were multiple measurements of the same source (due to the overlap between WISE orbit-to-orbit scans), we used quality indicators to choose the best measurement. Overall, 25 of our sources have >2σ detections in at least one of the four WISE bands.

We measure total IR luminosities, and the contributions to the total IR luminosities from star formation and AGN activity, by fitting the optical through MIPS photometry for each object with radiative transfer models for AGNs and starbursts, following the methods in Farrah et al. (2003) and F07. We describe the models and the fitting methods in Sections 3.1 and 3.2.

3.1. The Models

3.2. Fitting and Results

We fit all possible combinations of starburst plus AGN models to the SDSS through MIPS photometry of each object in order to extract starburst and AGN luminosities.

The models do not include a prescription for the broad absorption features in the rest-frame UV that arise from radiatively driven outflows, so we do not include the SDSS spectra in the fits. We do not include photometry shortward of ∼0.35 μm in the rest frame for the same reason; this means that we do not use the Ugri photometry for any object, and in some cases do not use the z-band data either. We do however use photometry at rest-frame wavelengths of 0.35 < λ(μm) <1; while we are interested in starburst and AGN luminosities at rest-frame 1–1000 μm, including these data limits which models are acceptable fits in the optical, and thus limits which models can be used to fit the IR data. For the MIPS data we fit to the measurements in all cases even if they are of low significance, or negative. We allowed the contribution from each component to vary from 0% to 100% to see if the IR emission was consistent with arising from a single origin. The model libraries span several free parameters (Section 3.1), but our data are not comprehensive enough to constrain all of them. Therefore, we use the complete model libraries to determine only the range in total, starburst, and AGN luminosities that are consistent with the data.

The best-fit SEDs are presented in Figures 1–6. In most cases the models provide an excellent fit to the data. We obtain χ²_red < 1 in 12/31 objects, and 1 < χ²_red < 2 in 15/31 objects. The remaining four objects have 2 < χ²_red < 3. The most difficult points to fit are usually the near-IR and/or the two shorter wavelength WISE bands, which are somewhat underpredicted in many cases. This could be due to the still preliminary WISE photometric calibration, and/or the lack of a host galaxy contribution in our models, so we defer exploration of this until more comprehensive near-IR data are available, and the WISE calibration is refined. In nearly all cases the ⩾24 μm photometry is well fitted by the models, though in a few objects the fit is relatively poor at these wavelengths. This effect is most noticeable in objects 3, 11, and 24, in which the 70 μm flux is significantly underpredicted. This could be due to an additional "hot" dust component or an extremely strong spectral feature that is not reproduced in the models. Since the models still reproduce the slope and approximate normalization of the far-IR SEDs even for these objects, we consider the results to still be usable.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To extract the most probable total, AGN, and starburst luminosities for each object, and their confidence intervals, we first construct a discrete probability distribution function (PDF) for the total, AGN, and starburst luminosities of each source. We weight each fits contribution to the PDF by its χ²_red value. From these PDFs, we then construct cumulative distribution functions, from which the most probable luminosities, and their confidence intervals, can be derived straightforwardly. Ideally, we would show the PDFs for all objects. This would however be unwieldy, so, as an example of the method, we show the distribution of χ²_red values and the total, starburst, and AGN PDFs for one object in Figure 7. In most cases the PDFs have a single peak and an approximately Gaussian shape. In a few cases though there are minor secondary maxima, and/or the shape is non-Gaussian. We therefore quote as the positive and negative errors the 90% confidence intervals. The luminosities, their confidence ranges, and the χ²_red values of the best fits are presented in Table 2.

Zoom In Zoom Out Reset image size

4.1. Infrared Luminosities

We find that FeLoBAL QSOs are luminous in the IR. All of our sample have (best-fit) total IR luminosities (L_Tot) in excess of 10¹² L_☉, with nine objects, or 29%, exceeding 10¹³ L_☉. There is no other sample to which we can directly compare ours, since no sample matched to ours in optical continuum luminosity and redshift has been observed in the mid/far-IR. We do however find that FeLoBAL QSOs are more IR-luminous than the Palomar–Green (PG) QSOs in Haas et al. (2003). Instead, they appear to have comparable IR luminosities to the wider population of BAL QSOs (Gallagher et al. 2007), reddened QSOs (Georgakakis et al. 2009), the most IR-luminous members of the general QSO population (Lutz et al. 2008; Orellana et al. 2011), and to ULIRGs (Genzel et al. 1998; Farrah et al. 2002, 2003; Desai et al. 2007).

The dominant power source behind the IR emission is, in most cases, AGN activity. A pure AGN is either the most likely power source, or consistent within the 90% confidence interval, for 35% of the sample (11/31 objects). A starburst component is required (at ⩾90% confidence) for the remaining objects, but in only twelve of these objects is the starburst more luminous than 10¹² L_☉, and in only three objects is the starburst more luminous than the AGN. The mean AGN contribution to the total IR luminosity is ∼76%, comparable to that seen in local ULIRGs with "warm" IR colors,¹⁵ but lower than that seen in PG QSOs (Veilleux et al. 2009). The spread in AGN contribution to L_Tot is however wide, spanning 0.2–1.0. We find a strong correlation between L_Tot and the AGN IR luminosity (L_AGN), with a Spearman rank correlation coefficient¹⁶ of ρ = 0.92 and a significance of deviation from zero of P < 0.001. We also find a correlation between L_Tot and starburst IR luminosity (L_SB), with ρ = 0.66 and P < 0.001. Conversely, we find no trend between L_Tot and the starburst contribution to L_Tot (f_SB), with ρ = −0.23, P = 0.21 (Figure 8, left). Finally, we find a weak anticorrelation between L_AGN and f_SB (ρ = −0.47, P = 0.007; Figure 8, right). This is most apparent when the AGN IR luminosity exceeds ∼10¹³ L_☉, at which point the starburst contribution decreases noticeably.

Zoom In Zoom Out Reset image size

The star formation rates for the objects with starbursts detected at ⩾90% confidence, computed by extracting the monochromatic 60 μm luminosities from the SED fits and then using Equation (7) of Rowan-Robinson et al. (1997), lie between several hundred to a few thousand solar masses per year. For the six objects in common with Farrah et al. (2010), these star formation rate estimates agree with those from the Infrared Spectrograph, though the large systematics on both measurements renders this comparison of little value.

We now examine the IR emission from AGNs and starbursts in FeLoBAL QSOs using a different approach. We use all the solutions from all the individual fits to construct a single PDF for the whole sample¹⁷ (Figure 9). From this, we reach similar conclusions to those described above. An AGN supplies the bulk of the IR emission in the majority of cases. Starburst activity on the other hand is significantly less luminous; we find that the starburst is fainter than 10¹² L_☉ in just over half (56.7^+5.2_−5.3%) of cases, and almost never exceeds 10¹³ L_☉ (Table 4, first row).

Zoom In Zoom Out Reset image size

Overall, we find compelling evidence from the IR properties of FeLoBAL QSOs that they are, as a class, distinct from the general QSO population. FeLoBAL QSOs have higher IR luminosities and a smaller average AGN fraction. Instead, the IR properties of FeLoBAL QSOs are consistent with those of the reddened and LoBAL QSO populations.

The relationship between FeLoBAL QSOs and ULIRGs is harder to discern. It has been suggested (F07) that FeLoBAL QSOs are, as a class, a ULIRG-to-QSO transition phase. There is one result from our study that supports this, namely, that the mean AGN fractional IR luminosity of FeLoBAL QSOs is similar to that of "warm" ULIRGs. Conversely, if FeLoBAL QSOs were such a transition phase then we may have expected a higher fraction of them to host luminous starbursts, up to ∼50% if they represented the entirety of the starburst to AGN transition (see, e.g., Farrah et al. 2009). Now, this fraction is an upper limit to what we might see, and such a high fraction with luminous starbursts is (just) consistent with the PDF in Figure 9. Other evidence for this comes from the unusual mid-IR spectral shapes of FeLoBAL QSOs in comparison to other IR-luminous QSOs (Farrah et al. 2010). From the results in this paper, however, the scenarios that FeLoBAL QSOs are (1) randomly drawn from the reddened QSO population and (2) a ULIRG to QSO transition phase are both plausible. Since the former scenario is simpler, we conclude that, while some fraction of reddened/IR-luminous QSOs are almost certainly the endpoint of a ULIRG–QSO transition, we see no evidence in this paper that demands that FeLoBAL QSOs are more likely to be such a transition stage than the (presumably) parent reddened QSO population.

4.2. Starbursts and Outflows

We now examine whether or not there is a relationship between the AGN-driven outflows and the obscured star formation in our sample. In Section 4.2.1 we discuss how to measure the strength of the BAL features. In Section 4.2.2 we examine the relationship between AGN-driven outflows and obscured star formation using the best-fit luminosities and their confidence ranges to start with, and then examine it using all the information in the PDFs. We interpret our results in Section 4.2.3.

There is one point we note first. We cannot here measure how much kinetic energy the outflow is injecting into the ISM of its host galaxy. Even with well-resolved BALs in multiple species, such a measurement involves in-depth radiative transfer calculations (e.g., Casebeer et al. 2008). The SDSS optical spectra used here are however of low resolution, and usually contain BALs only for one or two species. So, even attempting such calculations for our sample is futile. It is however plausible that the depth, width, and velocity offset of BALs in a single species, as quantified in measures such as balnicity index and absorption index (Hazard et al. 1987; Hall et al. 2002; Trump et al. 2006) do scale with increasing outflow strength. So, we here use the properties of the Mg ii BAL features solely as an estimate of relative outflow strength within the sample.

4.2.1. Measuring Outflow Strengths

The strength of absorption of the BALs in a given species is traditionally defined as the total velocity width over which the absorption exceeds a minimum value. This however gives rise to four problems with interpretation. First, different authors use different parameterizations for this strength, which in some cases can determine whether an object is identified as a BAL QSO (of any variety) or not. Second, the derived depths and widths of the BAL troughs are sensitive to the choice of continuum level, which can vary significantly between automated and manual measurements even for the same input data (e.g., compare the results in Trump et al. 2006; Gibson et al. 2009; Allen et al. 2011). Third, since our sample is at z < 1.8 we see absorption only in a limited number of species. Fourth, the BALs in some QSOs show temporal variation on approximately decadal timescales (Arav et al. 2001).

The latter two issues cannot be addressed with the data we have. To mitigate the first two as far as is possible we proceeded as follows. First, we used the following parameterization for the absorption strength (AS) of BALs.

$$\begin{equation} {\rm AS} = \int _{v_{0}}^{v_{1}}\left[1 - \frac{f(v)}{a}\right]Bdv, \end{equation} \tag{ 1 }$$

where v is velocity, f(v) is the (normalized) flux at that velocity, and a is a scaling factor. The quantity B is set to unity if the absorption is below 10% of the continuum and if the width of the trough is greater than some minimum value, otherwise it is set to zero. We define the systemic redshift to be at zero velocity, and positive velocity to be blueward of this redshift. Equation (1) includes all previous definitions of BAL strength as special cases (Table 3). We chose the AS₂ parameterization as our primary measure since it is a reasonable compromise between strictness and inclusiveness. Second, we measure absorption strengths for the same, single species across the whole sample. We chose Mg ii λ2799 as it is the only species present in all the SDSS spectra of our sample, though we note that it can be contaminated by Fe ii absorption (see Table 1 of Hall et al. 2002). We remeasured all the Mg ii absorption strengths by hand, using the methods described in Urrutia et al. (2009). Our measurements, together with one set of comparison measurements from Gibson et al. (2009), are given in Table 2. We check our absorption measures against those in the literature, and explore the effects on our results of using different measurements of absorption strength, in the Appendix. The formal error on the absorption strengths from the fits is usually of order 100 km s⁻¹, but there are significantly larger systematic uncertainties (see the Appendix) that are hard to quantify. These systematic uncertainties should however not dramatically change the absorption strengths of the sample relative to each other, as long as the measurements are done in an internally consistent way.

Table 3. Examples of Different Absorption Strength Measures from Equation (1), Ordered by Decreasing Strictness

Name	v0	v1	a	Min. Velocity Width
AS0 (BI)a	3000	25000	0.9	2000
AS1	3000	29000	0.9	2000
AS2 (BI0)b	0	25000	0.9	2000
AS3	0	29000	0.9	1000
AS4 (AI)c	0	29000	1.0	1000

Notes. The names in parentheses are those of measures used in the literature. We do not use AS₁ or AS₃ here, but present them to illustrate other possible parameterizations. ^aBalnicity index (Weymann et al. 1991; Gibson et al. 2009; Allen et al. 2011). ^bModified balnicity index (Gibson et al. 2009). ^cAbsorption index (Hall et al. 2002; Trump et al. 2006).

Download table as:  ASCIITypeset image

4.2.2. Outflows versus Infrared Properties

We first use the luminosities in Table 2 to investigate if absorption strength depends on total, AGN, and starburst IR luminosities. We find no correlation between absorption strength and L_Tot (ρ = 0.31, P = 0.09; Figure 10, top left), or between absorption strength and L_SB (ρ = −0.10, P = 0.58; Figure 10, bottom left), though we do find a hint of a correlation between absorption strength and L_AGN (ρ = 0.39, P = 0.04; Figure 10, top right). It is also interesting that (1) the dispersion in absorption strength is greater at higher total and AGN luminosities, and (2) all but two of the luminous (>10¹² L_☉) starbursts lie at AS₂ < 5000 km s⁻¹. We conclude that the mechanism that determines the strength of the outflows is not directly responsible for heating the dust near the AGN, and does not have a strong effect on the absolute luminosity of the starburst.

Zoom In Zoom Out Reset image size

If however we plot absorption strength against f_SB then we see a weak but clear anticorrelation (ρ = −0.49, P = 0.005; Figure 10, bottom right). Moreover, there seems to be a change in the distribution of starburst contributions with absorption strength at AS₂ ≃ 3500 km s⁻¹; all the systems with AS₂ > 3500 km s⁻¹ have a starburst contribution of less than 25%, while the systems with AS₂ < 3500 km s⁻¹ have a wide dispersion in starburst contributions, from 0% to ∼80%.

To estimate the probability that the distribution of starburst contributions changes at AS₂ = 3500 km s⁻¹, we employ the two-sided Kolmogorov–Smirnov test. We find the distributions of the objects above and below AS₂ = 3500 km s⁻¹ in the bottom right panel of Figure 10 are different at 99.84% confidence.¹⁸ The number of objects in the two subsamples is however low enough that the Kolmogorov–Smirnov test can be unreliable. So, we employ a cruder test. We take the null hypothesis to be that there is no correlation between absorption strength and f_SB, and that the systems with f_SB < 25% represent the "true" underlying distribution. The probability of finding all eight systems with f_SB > 25% at AS₂ < 3500 km s⁻¹ is then (8/19)⁸ ≃ 1%, i.e., a ≃99.0% probability of a difference. We regard this latter figure as a more reliable measure of the significance of a difference. Hence, our finding all of the f_SB > 25% systems at AS₂ < 3500 km s⁻¹ is only weak evidence that an anticorrelation exists.

Nevertheless, this result is consistent with the idea that the radiatively driven outflows negatively affect star formation. The systems with f_SB < 25% would then be those in which an outflow has curtailed star formation, and those in which such an outflow has subsequently waned, making the observed dispersion in absorption strengths wide. The systems with f_SB > 25% would be those in which an outflow has started to develop, but has not yet affected the starburst.

To explore the relationship between absorption strength and infrared properties further, we use all of the information in the PDFs, in a manner similar to that used in Section 4.1. We adopt a boundary value of absorption strength motivated by Figure 10 of AS₂ = 3500 km s⁻¹, and divide our sample into two subsamples at this boundary.

We first construct PDFs of L_AGN and L_SB for these two subsamples, and extract from them the probabilities of obtaining luminosities in excess of certain values (rows 2 and 3 of Table 4). We see similar results to those seen in the first three panels of Figure 10. We see no convincing differences in the starburst luminosities between the low and high absorption strength subsamples, compared to either each other or to the sample as a whole. We also see that we are only marginally more likely to see AGNs with L_IR < 10^12.5 L_☉ in the low absorption strength subsample.

Table 4. Probabilities of Obtaining Starburst and AGN Luminosities Below Certain Boundaries

Selection	Probability of Obtaining
	LSb(L☉)			LAGN(L☉)
	<1011	<1012	<1012.5	<1012	<1012.5
All objects	28.3+2.8−4.8%	56.7+5.2−5.3%	83.7+4.1−3.2%	21.4+3.8−10.5%	57.0+7.3−7.9%
AS2 < 3500 km s−1	27.7+3.2−7.0%	55.2+5.6−6.9%	80.0+5.7−5.2%	32.1+6.2−14.1%	68.7+8.8−8.3%
AS2 > 3500 km s−1	29.6+6.2−5.5%	59.7+11.0−9.0%	91.0+4.2−2.8%	0%	33.8+10.8−13.6%
LAGN < 1012.5 L☉	38.8+3.8−7.1%	69.4+5.2−4.6%	93.7+2.5−2.0%	37.3+7.4−15.7%	90.9+5.6−3.0%
LAGN > 1012.5 L☉	16.2+2.5−4.9%	43.2+6.9−10.4%	69.8+8.2−7.2%	0%	10.4+2.0−6.2%

Notes. We give probabilities for the complete sample, the sample divided by absorption strength, and the sample divided by AGN luminosity (see also Table 6). Errors were derived using jack-knife resampling, removing one source at a time and computing the ∼1σ confidence interval from all the resulting realizations. The subsamples divided by L_AGN were divided on their peak luminosities, hence the non-zero probabilities of obtaining luminosities outside the boundaries.

Download table as:  ASCIITypeset image

If however we consider the PDFs for the contribution of the starburst to the total IR luminosity for the two subsamples (Figure 11), then we see a clear difference. The low absorption strength subsample shows a higher chance of a higher f_SB than the sample as a whole, and a much higher chance than the high absorption strength subsample. We quantify this by extracting the probabilities of obtaining starburst contributions to the total IR luminosity in excess of 25% and 50% from the whole sample and the two absorption strength subsamples (Table 5, Rows 1–3). We find, at >5σ significance, a higher chance of obtaining f_SB > 25% and f_SB > 50% in the low absorption strength sample compared to the high absorption strength sample. These results do not change if we exclude the objects noted as potential contaminants in Section 2.1, or if we exclude the objects with χ²_Red > 2 in Table 2.

Zoom In Zoom Out Reset image size

Table 5. Probabilities of Obtaining Starburst Contributions to the Total IR Luminosity, Above Two Boundaries

Selection	P(fSB)
	>25%	>50%
All objects	50.3+5.3−5.4%	17.9+2.5−4.2%
AS2 < 3500 km s−1	67.3+4.5−4.1%	26.6+3.2−4.8%
AS2 > 3500 km s−1	17.8+3.7−6.5%	<1.5%
LAGN < 1012.5 L☉	57.8+6.1−6.4%	18.6+3.1−5.0%
LAGN > 1012.5 L☉	38.7+8.5−10.0%	15.1+4.0−7.0%

Notes. As with Table 4, we give probabilities for the complete sample, the sample divided by absorption strength, and our sample divided by AGN luminosity (see also Table 6). Errors were derived using jack-knife resampling.

Download table as:  ASCIITypeset image

4.2.3. Interpretation

The anticorrelation that we observe between absorption strength and contribution from star formation to the total IR emission is straightforwardly interpreted as the outflow from the AGN curtailing star formation in the host galaxy. There are however four other ways that we could see this anticorrelation.

The first is that stronger outflows reflect an increase in the IR emission from the AGN, but have no effect on the starburst; if this is the case we would see a decline in f_SB, but without there being a direct relationship behind the two phenomena. This possibility is apposite if there is a conspiracy of fortuitous timescales, where the peak starburst luminosity precedes the peak AGN luminosity by a few Myr. The second is that starburst activity suppresses AGN outflows, so when the starburst wanes (via a cause unrelated to the AGN) an AGN driven outflow can appear. The third is an observation bias; e.g., if a high f_SB meant that the Mg ii troughs were observed to be weaker than they really are. The fourth is a selection bias, e.g., if QSOs with strong BALs and strong starbursts drop out of the initial SDSS QSO selection and were thus not included in Trump et al. (2006).

We first consider the last three of these alternatives. The second alternative has traditionally been considered unlikely, the argument being that for it to happen then the starburst would have to act near the origin of the outflow, but AGN broad-line regions are around five orders of magnitude smaller than starburst regions. Nevertheless, it is not completely implausible. One possible scenario is as follows. If the ISM was initially dense and the SMBH was initially small, then the SMBH may not be at first capable of powering outflows that extend ∼kpc into the host galaxy, but as the density of the ISM was reduced by the ongoing starburst (which thus waned due to the reduction in fuel supply) and the SMBH grows, then large-scale outflows would subsequently appear. This is not a scenario we can test, but it would likely require a serendipitous conjunction of ISM and SMBH parameters, so we do not consider it further.

The third possibility is one that we again cannot test, so we cannot formally discount the possibility of a very large population of OB stars acting to suppress the observed depth of the UV absorption troughs. Conversely, the rest-frame UV spectra of starbursts in ULIRGs reveal continua that are at least an order of magnitude too weak to provide this effect, and sometimes show absorption in the same species (Farrah et al. 2005). The fourth possibility is also not testable, but the SDSS is now turning up FeLoBAL QSOs in large numbers, and the initial QSO follow-up color selections are fairly relaxed, so we do not consider this possibility likely either.

The first alternative is however one that we can test, from which we propose that it is unlikely as well. The test is as follows. If it is the case that outflow strength is a proxy for AGN luminosity, then we should see a bigger difference between the starburst contribution PDFs for subsamples divided by AGN luminosity than between subsamples divided by absorption strength.¹⁹ In Figure 12 we show the starburst contribution PDFs for two subsamples divided by AGN luminosity, one for objects with L_AGN > 10^12.5 L_☉ and one for objects with L_AGN < 10^12.5 L_☉. Qualitatively, the difference between the PDFs in Figure 12 is weaker than the difference between the PDFs divided by absorption strength in Figure 11. If we extract probabilities of obtaining the same starburst and AGN luminosities, and starburst contributions as we did for the absorption strength subsets (fourth and fifth rows of Tables 4 and 5), then we see three interesting results. First, lower luminosity star formation is now more likely (at ≳ 3σ) to be seen in the lower luminosity AGN subsample (e.g., for P(L_Sb < 10¹² L_☉), the L_AGN < 10^12.5 L_☉ subsample is 69.4^+5.2_−4.6% while the L_AGN > 10^12.5 L_☉ subsample is 43.2^+6.9_−10.4%). Second, we are more likely (albeit only at just over 2σ) to obtain a smaller starburst contribution by selecting high absorption strength systems than we are by selecting high AGN luminosity systems (e.g., for P(f_SB > 25%), the AS₂ > 3500 km s⁻¹ subsample is 17.8^+3.7_−6.5% while the L_AGN > 10^12.5 L_☉ subsample is 38.7^+8.5_−10.0%). In other words, we are more successful in finding systems with a large starburst contribution to the total IR emission by selecting on weak outflows than we are by selecting on low AGN luminosity. Third, the probabilities of a starburst contribution in excess of 25% (or 50%) are statistically indistinguishable between the high and low AGN luminosity subsamples, but are different at >5σ between the high and low absorption strength subsamples (Table 5).

Zoom In Zoom Out Reset image size

Overall, therefore, we find that radiatively driven outflows from an AGN with absorption strengths ≳ 3500 km s⁻¹ act to curtail star formation in their host galaxies. We also find that this effect is (at least largely) relative; such outflows reduce the contribution from star formation to the total IR luminosity to less than ∼25%. We also propose that the infrared luminosity of the AGN is not a good proxy for the degree of AGN feedback that is taking place. Finally, since the IR properties of our sample are consistent with being drawn randomly from the reddened QSO population, we conclude that this is true generally for reddened QSOs.

These results are consistent with the idea that starburst and AGN activity crudely correlate with each other in active galaxies—a more luminous starburst means we are more likely to see a luminous AGN²⁰—but where we add that a radiatively driven outflow can curtail the relative luminosity of the starburst on much shorter timescales than the typical lifetimes of a starburst or AGN.

We conclude on a cautionary note. Our results appear solid, but they are based on a small sample, which makes the errors difficult to estimate. We have used what we believe to be a robust error estimation method (jack-knife resampling, Tables 4 and 5), but with only 31 objects we cannot reliably measure the error function, since comprehensive resampling methods are not possible. Moreover, we cannot test the robustness of the adopted boundary for dividing the PDFs of AS₂ = 3500 km s⁻¹. This boundary was motivated by the analysis in Section 4.2.2, but we would ideally like to explore the consequences of varying this boundary by up to a few thousand km s⁻¹ in either direction. We did perform a basic test of this in the Appendix, by adopting instead a boundary of AS₂ = 5000 km s⁻¹, but we do not consider this to be a robust assessment of how sensitive our results are to the choice of boundary. Given the distribution of AS₂ values of our sample, however, we cannot perform more comprehensive tests. A similar argument applies to the choice of luminosity cut for dividing the AGN PDFs in Figure 12, and to the choice of template library used to model the luminosities (since several alternatives are available, see Section 3.1). To explore these issues properly would require a sample at least a factor of two larger than that used here. Aside from sample size, we would also like photometry longward of 200 μm so as to measure the emission from cold dust heated mainly by star formation, and higher quality optical spectra to resolve Mg ii kinematics and look at the absorption strengths of at least one other species.

We have presented a sample of 31 FeLoBAL QSOs from the SDSS at 0.8 < z < 1.8. These QSOs have broad, deep absorption troughs in their rest-frame UV spectra that are unambiguous signatures of radiatively driven outflows powered by an AGN. Previous work has suggested that FeLoBAL QSOs are IR-luminous, and sometimes harbor star formation rates of up to a few thousand solar masses per year. Furthermore, there is evidence that the AGN-driven outflows can extend up to several kiloparsecs into the host galaxies. FeLoBAL QSOs are thus ideal laboratories for studying the effects of radiatively driven outflows from an AGN on obscured star formation.

We selected our sample purely on the basis of their rest-frame UV spectral properties and assembled for them optical through far-IR photometry from the SDSS, 2MASS, UKIDSS, WISE, and Spitzer. We then fit these data with radiative transfer models for the IR emission from AGNs and starbursts to extract best-fit AGN and starburst IR luminosities. We then compared these luminosities to the strength of their outflows as inferred from the absorption properties of the Mg ii λ2799 line in their SDSS spectra. Our conclusions are as follows.

• 1.  
  FeLoBAL QSOs are luminous in the IR. All of our sample have total IR luminosities in excess of 10¹² L_☉. Nearly one-third of the sample exceed 10¹³ L_☉. The dominant power source behind the IR emission is, in most cases, AGN activity. A pure AGN is either the most likely power source, or consistent within the 90% confidence interval, for eleven of the sample. A starburst component is required for the remaining objects, but in only twelve of these objects is the starburst more luminous than 10¹² L_☉, and in only three objects is the starburst more luminous than the AGN. The mean AGN contribution to the total IR luminosity is ∼76%. The spread in AGN contribution to the total IR luminosity is however wide, spanning 0.2–1.0. Overall, the IR properties of FeLoBAL QSOs appear consistent with those of the general red/dusty QSO population. We do not however find convincing evidence that FeLoBAL QSOs are more likely to be a ULIRG to QSO transition phase than the general red QSO population.
• 2.  
  We find no convincing relationship between the strength of the outflows and the total IR luminosity, or between the strength of the outflows and either the starburst or AGN luminosities. Conversely, we find a clear relationship between the strength of the outflows and the contribution from star formation to the total IR luminosity. If we divide our sample in two at an outflow strength boundary of AS₂ = 3500 km s⁻¹ and construct PDFs for the starburst contribution for the two subsamples, then the low absorption strength subsample shows a higher chance of a higher starburst contribution than the sample as a whole, and a much higher chance of a higher starburst contribution than the high absorption strength subsample (Figure 11). We quantify this by extracting the probabilities of obtaining starburst contributions to the total IR luminosity in excess of 25% (Table 5, Rows 1–3). We find, at >5σ significance, a higher chance of obtaining a starburst contribution in excess of 25% in the low absorption strength sample compared to the high absorption strength sample.
• 3.  
  This anticorrelation between outflow strength and the contribution from star formation to the total IR luminosity is straightforwardly interpreted as the outflow from the AGN curtailing star formation in the host galaxy. There are however several other ways that it could arise. The most obvious alternative is that stronger outflows reflect an increase in the IR emission from the AGN, but have no effect on the starburst; if this is the case then the starburst contribution would decline with stronger outflows, but without there being a direct relationship behind the decline. To test this possibility we divided our sample into two at an AGN IR luminosity boundary of L_AGN < 10^12.5 L_☉ and constructed PDFs for the starburst contribution for these two subsamples (Figure 12). If it is the case that outflow strength is a proxy for AGN luminosity, then we should see a bigger difference between these two PDFs than that seen between the PDFs in Figure 11. Instead, the PDFs show a smaller difference. Furthermore, we find that (a) we are (marginally) more successful in finding systems with a large starburst contribution to the total IR emission by selecting on weak outflows than we are by selecting on low AGN luminosity, and (b) the probabilities of a starburst contribution in excess of 25% are statistically indistinguishable between the high and low AGN luminosity subsamples, but are different at >5σ between the high and low absorption strength subsamples (Table 5). We considered several further alternative possibilities (Section 4.2.3) but did not find any of them convincing.
• 4.  
  We therefore conclude that strong, radiatively driven outflows in FeLoBAL QSOs can have a dramatic, negative effect on obscured star formation in their host galaxies. This is the most direct evidence yet obtained for "quasar mode" feedback in the QSO population at high redshifts. We find that outflows with an absorption strength in Mg ii λ2799 of greater than AS₂ = 3500 km s⁻¹ (Table 3) act to curtail luminous star formation in their host galaxies. We also find that this effect is at least largely relative; the starburst luminosity is reduced to less than about 25% of the total luminosity. Finally, we propose that the magnitude of this effect is not deducible from the IR luminosity of the AGN.

We thank the referee for a very helpful report, Sandra Blevins for help with reducing the Spitzer data, and Karen Leighly, Amanda Truitt, and Adrian Lucy for helpful comments. This work is based on observations made with the Spitzer Space Telescope, the Wide-field Infrared Survey Explorer (WISE), the Two Micron All Sky Survey (2MASS), and the Sloan Digital Sky Survey (SDSS). Spitzer is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. WISE is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. 2MASS is a joint collaboration between the University of Massachusetts and the Infrared Processing and Analysis Center (JPL/Caltech), with funding provided primarily by NASA and the NSF. The SDSS and SDSS-II are funded by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web site is http://www.sdss.org/. The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. This research has made extensive use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA, and of NASA's Astrophysics Data System. This research has also made use of Ned Wright's online cosmology calculator (Wright 2006). D.F. acknowledges support from the Science & Technology Facilities Council via an Advanced Fellowship (PP/E005306/1). J.A. acknowledges support from the Science and Technology Foundation (FCT, Portugal) through the research grant PTDC/CTE-AST/105287/2008.

Facilities: Spitzer - Spitzer Space Telescope satellite, WISE - Wide-field Infrared Survey Explorer, 2MASS - , Sloan - Sloan Digital Sky Survey Telescope, UKIRT - United Kingdom Infrared Telescope

As described in Section 4.2.1, measuring the properties of BALs is not straightforward. So, we here summarize the checks we performed to see if our results are robust against the choice and method of BAL measurement.

First, we check our measures by comparing them to the three independent measurements already in the literature (Trump et al. 2006; Gibson et al. 2009; Allen et al. 2011, though we note that the different absorption strength measures used by these authors together with the evolving reduction of the SDSS spectra means the samples in them are not identical). We obtain generally reasonable agreement (Figure 13). We conclude that our measurements of absorption strength are acceptable. We note though that the scatter in this figure is larger than the formal errors on the fits, by a factor of two or more in some cases. This likely reflects systematic uncertainties arising from continuum placement. We do not attempt to account for these systematics, or quote them, as they are difficult to estimate robustly. Instead, we use our absorption strength measurements only to make comparisons to each other, where they should be reliable. As a consequence, we do not quote individual errors on the fits (which are usually of order 100–200 km s⁻¹) as we feel these numbers are misleadingly small.

Zoom In Zoom Out Reset image size

Next, we check whether the results in Figure 10 depend on the adopted parameterization of absorption strength. In Figures 14 and 15, we replot these figures with both a stricter and more relaxed definition of absorption strength (AS₀ and AS₄, see Table 3). In all cases we recover comparable results. It is interesting that the relation is slightly offset from zero in the bottom right panel of Figure 15, perhaps suggesting that the anticorrelation is not driven by troughs much narrower than 2000 km s⁻¹. We conclude that our results are reasonably robust to the choice of parameterization of absorption strength.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Finally, we check whether the results in Section 4.2.2 depend on the method used to measure the absorption strengths. As our sample is small, we cannot robustly test the effects of different techniques to measure absorption strength. We can however do two basic checks. First, we keep our AS₂ measurements, but change the boundary for the low versus high absorption strength samples from 3500 km s⁻¹ to 5000 km s⁻¹. Second, we substitute our AS₂ measurements with the AS₀ measurements for the same objects in Gibson et al. (2009), adjusting the boundary to AS₀ = 4000 km s⁻¹. Since Gibson et al. (2009) are more stringent about what constitutes a BAL QSO in comparison to Trump et al. (2006) (from where we chose our sample), this is also a test on a smaller, "golden" sample. We plot the resulting PDFs for AGN fractions in Figure 16 and present the confidences on luminosity ranges and AGN fractions in Table 6 for both tests. We obtain the same results—an anticorrelation between absorption strength and fractional AGN luminosity—at a similar degree of significance. We conclude that our result is robust against the method used to measure absorption strengths.

Zoom In Zoom Out Reset image size