THE SINS SURVEY: BROAD EMISSION LINES IN HIGH-REDSHIFT STAR-FORMING GALAXIES^*

Observations of galaxies in the early universe are a unique probe of matter assembly during its most active epoch; at z ∼ 2, both the cosmic star formation rate and the luminous quasar space density are at their peaks (e.g., Fan et al. 2001; Chapman et al. 2005). Galaxies themselves undergo corresponding growth during this time, with the total stellar mass density in galaxies increasing from ∼ 15% to 50%–75% its current value between z∼ 3 and z∼ 1 (e.g., Dickinson et al. 2003; Rudnick et al. 2003, 2006). Constraining the dynamical and baryonic processes driving this rapid evolution is, therefore, central to our understanding of galaxy formation and to informing cosmological simulations.

At the relevant epochs, key spectral diagnostic features are redshifted into the near-infrared. In recent years, a number of surveys have therefore begun to systematically probe high-redshift populations with near-infrared spectroscopy (e.g., Erb et al. 2006a, 2006b, 2006c; Swinbank et al. 2004; Takata et al. 2006; Kriek et al. 2008). Our recently completed SINS (Spectroscopic Imaging in the Near-infrared with SINFONI) survey has combined the resolving power of 8–10 m class telescopes with high-resolution integral field spectrographs to study the detailed internal processes at work within massive, star-forming galaxies at z ∼ 2 (Förster Schreiber et al. 2006, 2009; Genzel et al. 2006, 2008; Bouché et al. 2007; Shapiro et al. 2008; Cresci et al. 2009; see also related work by Puech et al. 2006; Swinbank et al. 2006; Wright et al. 2007; Law et al. 2007, 2009).

In this paper, we combine the spectra of 47 galaxies detected in Hα emission to study the average spectral properties of star-forming galaxies at z ∼ 2 (Section 2). In particular, we report the discovery of broad emission lines in these galaxies (Section 3) and interpret this high-velocity warm gas as arising either in large-scale galactic winds driven by the high star formation rates (SFRs) in these galaxies or in the broad-line regions (BLRs) surrounding active galactic nuclei (AGNs). We explore the implications of both scenarios in Section 4 and conclude in Section 5.

Throughout this paper, we assume a Λ-dominated cosmology with H₀= 70 km s⁻¹ Mpc⁻¹, Ω_m= 0.3, and Ω_Λ= 0.7. For this cosmology, 1'' corresponds to ≈ 8.2 kpc at z= 2.2.

In the context of the SINS program, 80 z = 1–3 systems were observed in emission lines in the infrared (rest-frame optical) with VLT/SINFONI (Eisenhauer et al. 2003; Bonnet et al. 2004) for an average of 3.5 hr per band and pixel scale on each target (Förster Schreiber et al. 2009). These galaxies were largely (62/80) taken from the (rest-frame) UV-selected samples of Erb et al. (2006b, 2006c) and the (rest-frame) optically-selected samples of Abraham et al. (2004), Daddi et al. (2004b), Kong et al. (2006), Lilly et al. (2007), and J. Kurk et al. (2010, in preparation). From these magnitude- and color-defined samples, suitable SINS targets were culled, with the main selection criteria being a combination of target visibility during the observing runs, night sky line avoidance in the emission lines of interest, and an estimated integrated emission-line flux ≳ 5 × 10⁻¹⁷ erg s⁻¹ cm⁻², such that high-quality data could be obtained in reasonable integration times. Of the 62 rest-frame UV/optically-selected galaxies chosen in this manner, 52 were well detected in Hα in our SINFONI observations.

Förster Schreiber et al. (2009) discuss the selection of this sample in detail and show that the SINS galaxies are representative of the z ∼ 2 star-forming galaxy population, with some bias toward the more rapidly star-forming (and therefore more luminous in Hα emission) systems. They also note that the SINS sample (and some of the parent samples) selects against known AGNs and quasars, in the interest of studying the dynamic and evolutionary state of z ∼ 2 star-forming galaxies. However, a small number of previously known AGNs were in fact observed in the SINS program; in the UV/optically-selected part of the sample, there are five such systems, as originally identified with the UV/optical spectroscopy of the parent surveys. For further details about the SINS sample, observations, and data reduction, we refer interested readers to Förster Schreiber et al. (2009).

Here, we analyze our SINS observations of the galaxies that were UV/optically identified and that are well detected in Hα in individual spatial elements; this population (totalling 47 of 52 sources detected in Hα, including 4/5 of the previously known AGNs) comprises the majority of the SINS sample. To study the average properties of these z ∼ 2 star-forming galaxies, we generated stacked, high-signal-to-noise (S/N) spectra representative of the population as a whole.

From our integral field data, we first created a spatially integrated one-dimensional spectrum for each galaxy by shifting each spectrum within the galaxy datacube by its measured (Hα) velocity and then collapsing the datacube into a single spatially integrated spectrum. In this manner, the spatially integrated spectrum contains no systematic velocity broadening (e.g., by large-scale rotation) on scales larger than the point-spread function (PSF; FWHM ∼ 4 kpc). Testing confirmed that this approach did not affect the properties of the broad Hα component (see below) but did improve the S/N of the detection. This technique has the additional benefit of randomizing OH atmospheric emission lines in the Hα rest frame and therefore effectively eliminating residuals from the OH line removal. The remaining residuals from this process were inspected and masked out by hand.

The 47 spatially integrated spectra were then combined into a single spectrum (with equivalent integration time of 195 hr) by interpolating all spectra onto a common wavelength axis, converting their measured fluxes to luminosities using their luminosity distances, weighting each spectrum by the S/N of the Hα emission line, and averaging the resulting spectra. During this process, we do not correct for extinction (but see Section 4). Typical extinctions of our sample galaxies have been measured to be A_V∼ 1 (Förster Schreiber et al. 2009; see also, e.g., Daddi et al. 2004a; Erb et al. 2006c), which translates to an underestimation of our Hα luminosities by at most a factor of ∼ 2.

The average spectrum for the SINS z ∼ 2 star-forming galaxies is presented in the top panel of Figure 1. We also created average spectra of subsets of the SINS galaxy sample, in order to test the dependence of spectral properties on other known galaxy properties.

Zoom In Zoom Out Reset image size

Our average spectra reveal a broad emission component underneath the bright narrow lines. We quantify this feature in each average spectrum by simultaneously fitting a combination of a constant continuum offset, narrow lines (Hα, [N ii], [S ii]) of identical kinematics (velocity and velocity dispersion), and a single broad component, whose kinematics are allowed to vary. All lines are assumed to be well described by a single Gaussian; the validity of this assumption is confirmed by a reduced χ² close to unity (χ²_dof ∼ 0.9–1.9) for all fits (see Table 1). Our data can also be well fitted by a combination of a constant continuum offset, narrow lines with shared kinematics, and broad forbidden and permitted lines with shared kinematics, with the [N ii]/Hα ratio identical in the narrow and broad components. All lines are assumed to be well described by this combination of two Gaussians (dotted green line in Figure 1; χ²_dof= 1.4), which has the same number of free parameters as the previous fit. In the limit of the S/N of our data, we cannot add additional free parameters to the fits, nor can we identify a preferred model. For clarity, we refer throughout to the former (single broad Gaussian under the Hα+[N ii] complex) as "broad lines" and the latter (double Gaussians for both Hα and [N ii]) as "broad wings." For simplicity and for comparison with the literature, we primarily quantify the observed high-velocity feature with a single broad Hα line in Section 3. However, we also discuss the implications of the broad wings scenario on the derived properties of the broad emission (Section 3) and on the interpretation of this emission (Section 4).

From these fits to the average spectra, we measure the fractional contribution of broad emission to the overall emission-line flux, the kinematics of the broad component, and the line ratios of the narrow-line components. The significance of these measurements is quantified in two ways. First, for each average spectrum, we recreated 100 spectra by randomly sampling (with replacement) and combining the individual contributing galaxy spectra. The properties of the broad component were measured in each of the resulting 100 average spectra, yielding confidence intervals for all derived quantities. Second, the probability of false positive detections P_false was tested by creating 1000 simulated spectra with the narrow line and observational properties (noise and spectral resolution) characteristic of each average spectrum. Comparing the derived broad components in these spectra to those in the real SINS spectra, we estimate the rate of spurious detections of broad components equal to or more prominent (in luminosity and FWHM) than those in the actual data. The results of this analysis for the different average spectra are presented in the following section.

We find that the average SINS galaxy spectrum includes a significant amount of broad emission (top panel of Figure 1), with χ²_dof = 4.9 for a fit with only narrow lines (dashed magenta line in the top right panel of Figure 1) and χ²_dof = 1.8 and 1.4 for fits including a broad Hα line and broad Hα and [N ii] wings, respectively (red and dotted green lines in the same panel; P_false= 2% and 1%). To ensure that this signature is not the result of the four known AGNs included in this sample, we also create stacked spectra of the AGNs and the rest of the sample. While the broad emission from the AGN host systems alone is quite substantial, a comparison of the average SINS spectrum and the average non-AGN spectrum illustrates that the broad emission in the average SINS spectrum is not dominated by that coming from the four AGNs. These results are summarized in Table 1.

We test the dependence of the presence of broad emission on galaxy properties by dividing the sample into three stellar mass bins, using the results from the spectral energy distribution (SED) fitting of Förster Schreiber et al. (2009) for the 45 of our 47 targets for which sufficient broadband data exist. The average spectra for these three bins show an increasing presence of a broad component with stellar mass (Figure 2; P_false= 11%, 5%, and 3%, respectively). However, the spectrum of the highest mass bin is significantly affected by the contribution of the four previously known AGNs in our sample, which all fall into this bin. The nuclear emission in these systems can bias the results of SED fitting toward larger masses, so we confirm their high masses with the dynamical mass measurements made by Cresci et al. (2009); in all cases, the dynamical masses of these galaxies are consistent with the stellar masses used here and remain among the highest in the SINS sample. Nevertheless, we confirm that a (weaker) broad component is also present in the nonactive galaxies in this bin (green line in Figure 2), with P_false= 9%.

Zoom In Zoom Out Reset image size

Several other key properties of galaxies, including SFR, size, stellar age, and metallicity, have well established correlations with stellar mass in high-z galaxies (e.g., Noeske et al. 2007; Trujillo et al. 2006; Erb et al. 2006b). Both the SFR–M_* relation and the mass–metallicity relation are apparent in Figure 2, via the increasing narrow Hα luminosity and increasing [N ii]/Hα ratio with stellar mass, respectively. Since the [N ii]/Hα ratio remains well below levels expected of shock heating or AGN activity, this latter is most likely tracing variations in metallicity (see below, as well as P. Buschkamp et al. 2010, in preparation). To these established mass-dependent properties in high-z galaxies, we now add the presence and strength of a broad component.

With the spatially resolved data, we can also compare the integrated spectra from the central (R < 3 kpc) regions of high-z galaxies to those from extended (R = 3–15 kpc) regions, in order to determine which regions in these systems are generating broad emission. For this analysis, we use galaxies from the intermediate and high mass bins of Figure 2 in which the intensity distribution of the stellar continuum defines a clear center of the system (totalling six systems). The average spectra of the central and extended regions of these systems are shown in Figure 3, normalized to the spatial area over which the spectra were extracted.

Zoom In Zoom Out Reset image size

In these spectra, a broad component is preferred by the best-fitting models; however, the significance of this result is low. The detection of broad emission in galaxy centers at z ∼ 2 (with P_false = 8%) is somewhat more robust than that in the extended regions (P_false = 14%), in which the best-fit broad component is very shallow. If real, the broad feature in the extended regions accounts for a comparable fraction of the total Hα luminosity to that in the central regions (Table 1). Tests of simulated galaxies indicate that such a broad line in the extended regions cannot be reproduced by a nuclear point source of broad emission (i.e., AGN) broadened by the PSF.

In the spectra shown in Figure 3, we note that the average [N ii]/Hα ratio is comparable in the central and extended regions, despite the difference in broad emission component, in support of the above interpretation of the [N ii]/Hα feature as primarily reflecting the metallicity of these systems. However, this should not be interpreted as a lack of metallicity gradient in these galaxies, since the spectra shown here are averaged over a number of galaxies, in each of which the extended region is spatially integrated over a large range in radii. Detailed studies of metallicity gradients within and between individual SINS systems will be presented in P. Buschkamp et al. (2010, in preparation).

The low-luminosity broad emission seen in our SINS galaxies can be interpreted in one of two ways: either as evidence of large-scale galactic outflows, presumably driven by the galaxies' very high SFRs, or as a tracer of the BLRs surrounding AGNs in these early galaxies. The current data do not allow us to robustly distinguish between these two scenarios empirically; in the following, we therefore examine both in detail.

4.1. Broad Emission from Starburst-driven Winds

One possible explanation of the broad emission is a starburst-driven wind. This scenario is in keeping with the positive correlation between broad emission and stellar mass (and therefore SFR) seen in Figure 2 and with the possible broad emission from non-nuclear regions seen in Figure 3. Moreover, starburst-driven winds are expected to be ubiquitous in the rapidly star-forming populations common at high redshift (e.g., Pettini et al. 2001; Shapley et al. 2003; Smail et al. 2003; Weiner et al. 2009) and are probably expelling mass from their host galaxies at rates comparable to the SFR (below; see also Martin 1999; Pettini et al. 2000; Martin 2003; Erb 2008; Weiner et al. 2009).

At low redshift, star-forming systems are known to drive galactic winds with observable signatures in the wings of the permitted and forbidden emission lines. In dwarf starburst galaxies, Westmoquette et al. (2007a) find that the broad wings of the emission lines can be modeled as a second Gaussian component with FWHM ⩽ 300 km s⁻¹. In contrast, observations of the more massive and more rapidly star-forming IR-luminous galaxy population reveal higher FWHM in the broad wings (300–800 km s⁻¹; Arribas et al. 2001), whose line widths and f_broad are comparable to those observed in the SINS galaxies (550 km s⁻¹; see Table 1; Armus et al. 1989, 1990; Lehnert & Heckman 1996).

Indeed, the relationship between wind speed and SFR (and the equivalent properties, galaxy mass and B-band magnitude) has been demonstrated by Rupke et al. (2005), Martin (2005), and Tremonti et al. (2007), respectively. In particular, Lehnert & Heckman (1996) and Rupke et al. (2005) showed that systems with higher SFR have faster winds, whose velocities exceed ∼1000 km s⁻¹ in ionized gas tracers. We recover similar trends and velocities with the three mass bins of the SINS data, as shown in Figure 4. Note, however, that this figure should not be directly compared with those in the works listed above, in which the wind speeds are probed with interstellar absorption features, whose velocities are typically lower than those of ionized emission-line gas (e.g., Veilleux et al. 2005; Rupke et al. 2005). Nevertheless, the similarity between the properties of local winds and those of the broad wings seen in the SINS galaxies makes it plausible that this emission is due to the galactic winds that certainly exist in these systems.

Zoom In Zoom Out Reset image size

These broad wings can be produced via several mechanisms, including turbulent mixing of the hot wind fluid and cool gas clumps at the base of the wind (Westmoquette et al. 2007a, 2007b) or within the large-scale bubble along the galaxy minor axis (e.g., Armus et al. 1990; Arribas et al. 2001). In the idealized model of a wind plowing into a spherically symmetric ambient medium, Heckman et al. (1993) provide analytic expressions for the velocity and Hα luminosity of the shocked gas as functions of the bolometric luminosity of the star-formation event, the duration of the event, and the ISM density. Converting the SFR (∼ 100 M_☉ yr⁻¹) of the SINS galaxies to a bolometric luminosity and accounting for the correspondingly denser ISM than in local spiral galaxies (by a factor of 10–30; Bouché et al. 2007), these relations predict a Hα luminosity of (0.7–10) × 10⁴² erg s⁻¹ generated from gas moving at ∼ 70–300 km s⁻¹. These properties are broadly consistent with the high-velocity wings observed in the SINS galaxies.

However, our spectra can also be fitted by a broad Hα line, and we therefore explore whether such a feature could likewise be generated in galactic winds. For simplicity, we assume that these winds are powered by supernova remnants (SNRs), in local examples of which broad Hα (FWHM = 500–8000 km s⁻¹) is observed throughout the Sedov–Taylor expansion phase, the result of charge exchange of the electrons from slow neutral atoms to the fast postshock protons (e.g., Chevalier & Raymond 1978; Smith et al. 1991; Heng & Sunyaev 2008). Although such broad Hα emission is not observed on the galactic scale in local starbursting systems, we note that none of these systems are appropriate analogs to the star formation mode that dominates at z ∼ 2. Only local ULIRGs have SFR comparable to that in the SINS galaxies, and the ionized gas emission in ULIRGs suffers significantly more extinction (A_V = 5–1000; Genzel et al. 1998) than in SINS galaxies (A_V∼ 1). We therefore briefly examine the possibility that galactic winds at z ∼ 2 emit broad Hα lines via a superposition of SNRs.

A simple test of this scenario is whether there are a sufficient number of SNRs in the SINS galaxies to drive the observed broad Hα luminosity. Locally, SNRs are observed to have broad Hα luminosities of 10³⁰−10³⁴ erg s⁻¹, generated by change exchange as the shock encounters the ISM and therefore proportional to the square of the gas density. In the SINS galaxies, the high SFR (∼ 100 M_☉ yr⁻¹) and the dense ISM (10–30 times denser than in local spiral galaxies) imply an increase in broad Hα luminosity per SNR by a factor of 100–1000, to 10³²–10³⁷ erg s⁻¹. The SFR in the SINS galaxies yields ∼ 1 supernova explosion per year, each of which will have expected Sedov–Taylor lifetimes of ∼ 10⁴ yr in the dense ISM, resulting in roughly 10⁴ SNRs radiating broad Hα at any given time in the average SINS galaxy, for a total broad Hα luminosity of 10³⁶–10⁴¹ erg s⁻¹. This number is lower than the observed broad Hα luminosity in the SINS galaxies by a factor of a few (see Table 1); however, real galactic winds penetrate much further through their host galaxies' ISM than do the sum of individual SNRs, implying that additional broad Hα emission is expected from the interaction of the large-scale winds with the ambient medium. This effect brings the predictions into even closer agreement with the observations and thus makes this mechanism an energetically feasible explanation for the luminosity and FWHM of the broad emission in the SINS galaxies.

Regardless of whether the broad emission is a broad Hα line or broad wings on all emission lines, if the emission is in fact due to starburst-driven winds, it is worth examining what the fate of this high-velocity gas may be. The average SINS galaxy (top panel of Figure 1) has either broad wings in Hα and [N ii] with velocity dispersions of ∼ 250 km s⁻¹ or a broad Hα line with velocity dispersion ∼ 700 km s⁻¹. We can compare these values directly with the escape velocity v_esc≈ 450 km s⁻¹ for a typical SINS galaxy (〈M_dyn〉 = 8 × 10¹⁰ M_☉, 〈R_1/2〉= 3.4 kpc; Förster Schreiber et al. 2009). A significant fraction (7% or 55%, respectively) of the high-velocity gas has velocity exceeding the host galaxy's escape velocity; assuming that this gas is distributed throughout the star-forming disk, a non-negligible amount of it should be expected to escape the galaxy. Furthermore, if the surrounding dark matter halo has a flat rotation curve, this gas would also be expected to escape the halo into the intergalactic medium.

We can then estimate the mass outflow rate that would correspond to such superwinds. The escaping Hα-emitting gas is ∼ 3%–17% of the total Hα-emitting gas in the SINS galaxies (i.e., 30% of the emission is broad and 55% of the broad emission escapes). Assuming that the fraction of gas in the ionized phase is roughly the same in the galaxies' star-forming disks and in the outflows (compare ∼ 1% in the Milky Way to ∼ 0.1%–1% in winds; Veilleux et al. 2005), this implies that ∼ 3%–17% of the galaxies' gas reservoirs are being expelled by the observed star-forming event. The dynamical times associated with these outflows can be approximated as the radius of the star formation event divided by the velocity of the flow; for the SINS galaxies, this yields dynamical times of ∼ 10 Myr. The average SINS galaxy has a dynamical mass of 8 × 10¹⁰ M_☉ and a gas fraction of 0.2–0.4 (e.g., Erb et al. 2006c; Bouché et al. 2007), yielding an expected outflow rate of 50–500 M_☉ yr⁻¹. This is consistent with the results of Erb (2008), who argue that the observed z ∼ 2 mass–metallicity relationship requires outflow rates slightly larger than the SFR (SFR = 1–800 M_☉ yr⁻¹ with median 72 M_☉ yr⁻¹ in the SINS sample; Förster Schreiber et al. 2009). If the broad emission we observe is due to galactic winds at z ∼ 2, we thus find mass outflow rates consistent with the observed metallicity evolution of these galaxies.

4.2. Broad Emission from Active SMBHs

Another possible, and common, interpretation of broad emission lines is as signatures of nuclear activity. For our galaxies, this interpretation is supported by the strength of the broad Hα emission increasing with stellar mass (and therefore bulge mass and possibly black hole mass) seen in Figure 2, by the less-luminous broad emission in the nonactive systems (the driving mechanism being "turned off" or obscured; Figures 1 and 2), and by the more significant detection of broad emission in the centers of galaxies than in their non-nuclear regions. Moreover, studies of z ∼ 2 star-forming galaxies in infrared and X-ray emission suggest that active nuclei may be a common feature of this population (Daddi et al. 2007).

Early work in the local universe has shown that broad Hα is by definition omnipresent in Type 1 AGN, in which the BLRs around the nuclei are unobscured. Moreover, the kinematics and sizes of these regions can be used to infer "virial" supermassive black hole (SMBH) masses, via calibrated relations between the observed AGN continuum luminosities at 5100 Å (L₅₁₀₀) and the sizes of the BLR (Kaspi et al. 2000), which can then be used in combination with the width of the broad emission lines to estimate SMBH masses (Vestergaard 2002). Greene & Ho (2005) have additionally related L₅₁₀₀ to the luminosity of the broad emission lines, making it possible to measure virial black hole masses using only a single broad line,

$${\fontsize{7.5}{10}\selectfont{\begin{equation} {M_{\rm{BH}}} = \big(2.0^{+0.4}_{-0.3}\big)\ \times \ 10^6 \left(\frac{L_{\rm H\alpha }}{10^{42}\ {\rm erg\ s^{-1}}} \right)^{0.55 \pm 0.02} \left(\frac{\rm FWHM_{ H\alpha }}{10^3\ {\rm km\ s^{-1}}} \right)^{2.06 \pm 0.06}\;{M_\odot }. \end{equation}}} \tag{ 1 }$$

Such virial black hole mass estimates have been verified against stellar and gas kinematic determinations of M_BH and found to be accurate to within a factor of ∼3 (Onken et al. 2007; Hicks & Malkan 2008; Netzer 2009, but see also Marconi et al. 2008). As a result, they have been utilized in the high-z universe to probe the masses of SMBHs in quasars (e.g., Willott et al. 2003; McLure & Dunlop 2004; Vestergaard 2004) and in the submillimeter-bright galaxy population (Alexander et al. 2008).

Additionally, the broad-line Hα luminosity can also be used to probe the accretion rate of SMBHs. Assuming the AGN continuum luminosity L₅₁₀₀ is roughly 1/10 the AGN bolometric luminosity L_bol, the accretion rate can be estimated by $\dot{ M}_ {\rm BH} = L_{\rm bol} / (\eta c^2)$ (La Mura et al. 2007). Using a typical value of η ∼ 0.1 and the calibrated relationship between L_Hα and L₅₁₀₀ (Greene & Ho 2005), we can then estimate $\dot{M}_{\rm BH}$. Although approximate, this estimate of $\dot{M}_{\rm BH}$ and the estimate of M_BH derived using Equation (1) nevertheless allow us to study both the putative black hole masses in our SINS galaxies and their accretion rates using our measurements of the broad Hα features.

With this procedure, we derive for the average SINS spectrum (top row in Figure 1) a black hole mass of M_BH= 4⁺³₋₂ × 10⁶ M_☉ with an Eddington ratio of ∼0.2. Furthermore, we infer black hole masses of M_BH< 9 ×  10⁵ M_☉, M_BH= 2⁺²₋₁ ×  10⁶ M_☉, and M_BH= 1⁺¹_−0.5 ×  10⁷ M_☉ in each of the three stellar mass bins, respectively. In the latter two bins, the Eddington ratios are estimated to be 0.3 and 0.4, respectively.

We note that the Hα luminosities used in these black hole mass estimates have not been corrected for extinction, and any such correction would increase the derived masses. However, broad emission lines associated with Type 1–1.5 AGN in local galaxies are consistent with no additional extinction along the line of sight to the BLRs (e.g., Rhee & Larkin 2000; Alonso-Herrero et al. 2003; Greene & Ho 2005). Likewise, at z ∼ 2, Alexander et al. (2008) have used the Hα/Hβ broad-line Balmer decrement to measure only small amounts of nuclear extinction (A_V ≈ 1.2) in their dust-rich submillimeter population. These authors postulate that the plentiful dust in the submillimeter galaxies obscures regions of star formation and not the BLRs. In the SINS galaxies, the average extinction in the star-forming regions is A_Hα∼ 0.8; if the BLR is similarly obscured, correcting for this effect would yield an increase of at most a factor of 2 in broad Hα luminosity and, from Equation (1), a factor ∼ 1.5 in derived black hole mass. This suggests that it is unlikely that the black hole masses measured here suffer significantly from extinction.

It is then naturally interesting to compare the estimated black hole masses with large-scale galaxy properties. In the left panel of Figure 5, we find the expected trend that black holes of increasing mass are found in galaxies of increasing stellar mass, with black-hole-to-stellar mass ratios of ∼ 7 × 10⁻⁵. For such black holes to be consistent with local scaling relations between black holes and bulges (M_BH/M_bulge ∼ 10⁻³), the bulge-to-total ratio B/T of these systems would need to be quite small (≲ 0.07). This is in marked contrast to results from detailed dynamical modeling of the five SINS galaxies with the highest resolution observations (Genzel et al. 2008), which yield B/T= 0.15–0.4 (center panel of Figure 5). Parameterizing these results with a simple polynomial, we estimate "bulge" masses (∼ dynamical mass within 3 kpc) for the three mass bins, albeit with large uncertainty. The resulting B/T for these mass bins are < 0.1, < 0.4, and 0.4, respectively. These bulge masses are then plotted against our measured black hole masses in the right panel of Figure 5.

Zoom In Zoom Out Reset image size

Comparing to the local M_BH–M_bulge relation measured by Häring & Rix (2004), we find that our z ∼ 2 galaxies lie significantly below the local relation, implying that black holes in the star-forming galaxies in the early universe may have lagged significantly behind their host bulges in assembly. Genzel et al. (2008) have shown that z ∼ 2 is, for many of these systems, an era of bulge formation via smooth but rapid secular processes, during which massive bulges are assembled from large star-forming clumps (M ∼ 10⁸–10⁹ M_☉) on timescales of ≲ 1 Gyr. Elmegreen et al. (2008) have demonstrated in simulations that, assuming each such clump contains a black hole of 10⁻³ of its total mass, these black holes would migrate to the galaxy center with their host clumps and form central SMBHs that are somewhat undermassive for the resulting bulges. The location of our z ∼ 2 systems significantly below the local relation likewise suggests that the bulges in these galaxies form first, through rapid secular processes, with the assembly of the central SMBHs following later.

The timescale for this final SMBH growth can be estimated for the high mass bin, in which the galaxy bulges are probably largely in place at z ∼ 2 (Genzel et al. 2008). Accretion onto the black hole at the current rate ($0.4\ \dot{M}_{\rm Edd} \sim 5 \times 10^{-2}$ M_☉ yr⁻¹) will bring these galaxies onto the local M_BH–M_bulge relation (M_BH∼ 10⁸ M_☉) in ∼ 2 Gyr. Similarly, galaxies in the intermediate mass bin will acquire SMBHs with final masses ∼ 2 × 10⁷ M_☉ in ≲ 1 Gyr. The evolutionary link between these final SMBHs at z= 0 (M_BH ≳2 × 10⁷ M_☉) and their putative z ∼ 2 host galaxies (M_* >10¹⁰ M_☉) is supported by the similar space densities of these two populations (∼3 × 10⁻³ h³₇₀ Mpc⁻³ and 2 × 10⁻³ h³₇₀ Mpc⁻³, respectively; McLure & Dunlop 2004; Daddi et al. 2004b, 2005; Reddy et al. 2005; Grazian et al. 2007). Moreover, the final black holes (M_BH ≳2 × 10⁷ M_☉) are found at z= 0 in elliptical and bulge-dominated spiral galaxies (e.g., Tremaine et al. 2002; Marconi et al. 2004), consistent with the probable descendants of the z ∼ 2 star-forming galaxy population (Genel et al. 2008; Conroy et al. 2008). This evidence thus confirm the plausibility of the rapid bulge formation seen at z ∼ 2 in the SINS galaxies (Genzel et al. 2008) being followed by a few Gyr of rapid SMBH assembly, ultimately resulting in spheroids and bulges that obey the local M_BH–M_bulge relation.

Alexander et al. (2008) found similar delayed SMBH formation in the submillimeter galaxy population at z ∼ 2 (see Figure 5), suggesting that the time lag between black holes and bulges may be a common phenomenon in rapidly forming galaxies at high redshift. However, this trend is not universal; quasars and radio galaxies at similar redshifts are suspected to lie above local black hole scaling relations, with black holes that are overmassive for their host bulges by up to and exceeding an order of magnitude (e.g., Walter et al. 2004; Shields et al. 2006; McLure et al. 2006; Peng et al. 2006; Maiolino et al. 2007, but see also Shields et al. 2003). These systems populate the highest mass end of the black hole mass function, with black hole masses of >10⁸–10⁹ M_☉ already in place at z ∼ 2. It may therefore be that black holes in these different mass/activity regimes grow in very different circumstances and consequently relate to their bulges very differently. If this is the case, the challenge is then to locate the mechanism(s) that bring these varied high-redshift formation processes together into the black hole scaling relations observed at z = 0.

In stacked, average spectra of SINS z ∼ 2 star-forming galaxies, we have detected broad emission underneath the much brighter narrow Hα and [N ii] emission lines. This broad emission accounts for ∼ 30% of the total Hα luminosity of these galaxies and can be parameterized equally well with a single broad Hα line ("broad line" of FWHM ∼ 1500 km s⁻¹) and with a two Gaussian fitted to both the permitted and forbidden lines ("broad wings" of FWHM ∼ 550 km s⁻¹). The luminosity and FWHM of the broad component increases with increasing galaxy mass and therefore with SFR. This broad component is found both in known AGNs and in stacked spectra of systems that have not been previously identified as AGNs. There is some evidence that the broad emission is more luminous in galaxy centers, as opposed to in the outer regions, but the significance of these detections are low.

We cannot empirically determine whether this broad emission is due to high-velocity galactic winds and the associated shocks or to the BLR emission of AGN. In the former case, we find that simple scaling arguments show that the luminosity and FWHM of the broad emission can plausibly be accounted for via shocking of the ambient interstellar media from supernovae-driven galactic winds. These winds would then be ejecting matter from the host galaxy at rates slightly exceeding the SFR, in keeping with expectations from the metallicity evolution of these galaxies and with ultraviolet interstellar absorption-line studies at similar redshifts.

On the other hand, the broad emission may be generated in a BLR; in this case, we can estimate the black hole masses and luminosities necessary to fuel the observed emission for each of the three galaxy mass bins. We find that the measured SMBH masses correlate with the host galaxy masses, as expected from local scaling relations, but that the SMBHs are significantly undermassive for their bulges when compared with local relations. While this result has large uncertainties, it is consistent with the emerging picture of galaxy assembly at z ∼ 2, in which a gas-rich disk fragments into large (⩾1 kpc) super-star-forming clumps that then migrate into the galaxy center on Gyr-timescales to form a nascent bulge. The bulge would then form first through this process and only later completely assemble its black hole.

The obvious direction for future research is to determine the source of the broad Hα emission in high-redshift star-forming galaxies. This will most likely require detailed examination of individual galaxies. Among the diagnostics that will be useful for this task are comparisons with X-ray data and deep integrations in rest-frame UV/optical wavebands to spatially resolve, e.g., UV interstellar absorption lines and broad Balmer emission.

We thank the ESO staff, especially those at Paranal Observatory, for their helpful and enthusiastic support during the many observing runs and several years over which the SINS project was carried out. We also acknowledge the SINFONI and PARSEC teams, whose hard work on the instrument and laser paved the way for the success of the SINS observations. This paper has additionally benefited significantly from many enlightening conversations with colleagues, including Frédéric Bournaud, Mohan Ganeshalingam, Kevin Heng, Phil Hopkins, Chris McKee, Jeffrey Silverman, and Thea Steele. Finally, we thank the referee, whose detailed and insightful comments greatly improved the quality of this paper.