From Nuclear to Circumgalactic: Zooming in on AGN-driven Outflows at z ∼ 2.2 with SINFONI

K20-ID5 (also known as 3D-HST GS4-30274, GS3-19791, and GMASS 0953) is a well studied star-forming galaxy at z ∼ 2.224. It was identified as a high-redshift candidate (z_phot > 1.7) in the K20 survey of infrared bright galaxies (Cimatti et al. 2002), spectroscopically confirmed by Daddi et al. (2004), and followed up with deeper long slit spectroscopy as part of the GMASS survey (Kurk et al. 2013) and integral field spectroscopy as part of the SINS/zC-SINF (Förster Schreiber et al. 2009), KMOS^3D (Wisnioski et al. 2015), and KASHz (Harrison et al. 2016) surveys.

The left-hand panel of Figure 2 shows an IJH color HST composite image of the region around K20-ID5. The galaxy has a bright and compact nucleus surrounded by fainter emission, which is stronger and more extended on the western side of the galaxy than the eastern side. The blue galaxy to the north of K20-ID5 is a lower redshift foreground object.

Zoom In Zoom Out Reset image size

K20-ID5 has a stellar mass of log(M_*/M_⊙) = 11.2 and a UV + IR SFR of 335 M_⊙ yr⁻¹ (Wuyts et al. 2011a, 2011b), placing it on the upper envelope of the main sequence (purple star in Figure 1).¹¹ The galaxy is classified as an AGN host based on the X-ray luminosity, radio luminosity, mid-infrared colors, and optical line ratios (e.g., van Dokkum et al. 2005; Genzel et al. 2014; Newman et al. 2014). The black hole accretion is driving a galactic wind, which has been detected as a broad blueshifted component in the rest-frame optical emission lines (Förster Schreiber et al. 2014; Genzel et al. 2014; Loiacono et al. 2019; Scholtz et al. 2020), the rest-frame UV absorption lines (Cimatti et al. 2013), and tentatively in the CO(6-5) emission line (Talia et al. 2018).

K20-ID5 has an effective radius of r_e = 2.5 kpc (van der Wel et al. 2014) and is classified as a compact star-forming galaxy (van Dokkum et al. 2015; Wisnioski et al. 2018). Molecular gas observations suggest that K20-ID5 has a starburst-like molecular gas depletion time and may therefore quench on a few-hundred-megayear timescale (Popping et al. 2017; Talia et al. 2018; Loiacono et al. 2019).

Our SINFONI-AO observations of K20-ID5 cover the white dashed region in the left-hand panel of Figure 2. Over the same region, we also have deep seeing-limited KMOS K-band observations and seeing-limited SINFONI H-band observations. The right-hand panel of Figure 2 shows the distribution of the SINFONI-AO Hα flux (mapped in colors), the KMOS Hα and SINFONI [O iii] flux (measured from single Gaussian fits; blue and yellow contours, respectively), and the dust distribution traced by the ALMA band 4 continuum emission (program 2015.1.00228.S, PI: G. Popping) (brown contours). The SINFONI-AO Hα flux map has a similar asymmetric distribution to the broadband flux, but the KMOS^3D data indicate that Hα emission is indeed present on the fainter eastern side of the galaxy, and further analysis reveals that the ionized gas follows a regular rotation curve on both sides of the nucleus (Wisnioski et al. 2018). The ALMA data suggest that there is 2.5 × 10⁸ M_⊙ of dust centered just to the east of the nucleus (Popping et al. 2017; Talia et al. 2018), and the Balmer decrement indicates significant attenuation of nebular emission in the nuclear region of the galaxy (A_V = 2.7; see Section 4.3). The large amount of dust could explain the asymmetry in the UV/optical emission, as well as the significant offset between the Hα and [O iii] flux peaks (also seen in Loiacono et al. 2019 and Scholtz et al. 2020).

The obscuration also complicates the measurement of the AGN luminosity. Using the X-ray hardness ratio as a tracer of the obscuration yields a column density of log(N_H/cm⁻²) = 23.2, corresponding to an intrinsic hard (2–10 keV) X-ray luminosity of log(L_X/erg s⁻¹) = 43.0 and a bolometric luminosity of log(L_AGN/erg s⁻¹) = 44.2 (using the bolometric correction adopted by Rosario et al. 2012; their Equation (1)). However, ongoing X-ray spectral modeling efforts suggest that the AGN may be Compton thick (M. E. Dalla Mura et al. 2020, in preparation), indicating that the hard X-ray luminosity is likely to be underestimated. Therefore, we also estimate the AGN luminosity from the [N ii] line (as described in Förster Schreiber et al. 2019), yielding log(L_AGN/erg s⁻¹) = 45.9. This value is much closer to what one would derive from the X-ray flux if the AGN was assumed to be Compton thick. Finally, we use the Spitzer/MIPS 24 μm (rest-frame 7.4 μm) flux density to derive an upper limit on the AGN luminosity. Lutz et al. (2004) found a linear correlation between $\nu {F}_{\nu }$(6 μm) and F(2–10 keV), with a typical slope of 0.23 for Seyfert 2 AGN. The observed mid-infrared flux includes contributions from both SF and AGN activity, and therefore, the total $\nu {F}_{\nu }$ provides an upper limit on the AGN luminosity for K20-ID5. We measure log($\nu {L}_{\nu }$[7.4 μm, AGN]/erg s⁻¹) < 45.2, corresponding to a hard X-ray luminosity of log(L_X/erg s⁻¹) < 44.6 and a bolometric luminosity of log(L_AGN/erg s⁻¹) < 46.4. The upper limit is 0.5 dex higher than the luminosity calculated from [N ii]. Therefore, we adopt the average of the X-ray and [N ii] estimates as our fiducial AGN luminosity; i.e., log(L_AGN/erg s⁻¹) = 45.6.

COS4-11337 and COS4-11363 are two massive (log(M_*/M_⊙) ∼ 11), compact (R_e ≤ 2.5 kpc; van der Wel et al. 2014) galaxies that lie very close in redshift (Δv = 150 km s⁻¹) and have a projected separation of only 065 (5.4 kpc), and they may therefore be in the early stages of a major merger. The dwarf starburst galaxy COS4-11530 lies 33 kpc to the northwest of the pair at a very similar spectroscopic redshift (z = 2.097). The left-hand panel of Figure 3 shows an IJH color HST composite image of the triplet.

Zoom In Zoom Out Reset image size

The stellar masses and A_V values for COS4-11337 and COS4-11363 were initially derived using standard SED modeling (following Wuyts et al. 2011a), assuming an exponentially declining star formation history and considering e-folding times in the range log(τ/yr) = 8.5–10.0. The best-fit galaxy parameters for COS4-11337 are log(M_*/M_⊙) = 11.3 and A_V = 0.8. We found that the fit for COS4-11363 could be significantly improved by adopting either a shorter e-folding time (100 Myr) or a truncated star formation history. Both models give best-fit galaxy parameters of log(M_*/M_⊙) = 11.1 and A_V = 0.9.

We calculate the IR SFR of the system using the Herschel PEP 160 μm flux, which we convert to ${L}_{\mathrm{IR}}$ using the Wuyts et al. (2008) SED template. The galaxies are strongly blended in the far-infrared imaging, so it is only possible to calculate the total (combined) SFR of the system, which is 424 M_⊙ yr⁻¹. The SED fitting suggests that COS4-11337 has a ∼ 10× higher SFR than COS4-11363, and the Hα flux ratio between the two nuclei is 47, or 14 when removing the contribution of the outflow to the Hα flux of COS4-11337 (see Section 5.1). Therefore, we divide the IR SFR between the galaxies in a 10:1 ratio; i.e., SFR_{IR, 11337} = 385 M_⊙ yr⁻¹ and SFR_IR,11363 = 39 M_⊙ yr⁻¹. We add the UV SFRs measured from the SED fitting for the individual systems and obtain UV + IR SFRs of 395 M_⊙ yr⁻¹ and 50 M_⊙ yr⁻¹ for COS4-11337 and COS4-11363, respectively.

Based on our derived parameters, COS4-11337 lies on the upper envelope of the star formation main sequence (blue star in Figure 1), and COS4-11363 lies about a factor of three below the main sequence (green star in Figure 1).

Grism spectroscopy from the 3D-HST survey (Brammer et al. 2012; Momcheva et al. 2016) indicates that COS4-11337 has strong [O iii] emission, while COS4-11363 has very weak line emission. COS4-11337 is classified as an AGN based on both the optical line ratios (Genzel et al. 2014) and the hard X-ray luminosity (log(L_X/erg s⁻¹) = 44.5; Luo et al. 2017), which corresponds to an AGN bolometric luminosity of log(${L}_{\mathrm{bol},\mathrm{AGN}}$/erg s⁻¹) = 46.2 (Rosario et al. 2012). We note that a very similar bolometric luminosity is obtained from the extinction-corrected [O iii] luminosity, assuming a bolometric correction factor of 600 (Netzer 2009).

Seeing-limited H- and K-band observations of the COS4-11337/COS4-11363 system were obtained as part of the KMOS^3D survey (Wisnioski et al. 2015, 2019). Broad forbidden lines reveal a strong outflow originating from COS4-11337. The beam convolved line emission from COS4-11337 is dominant everywhere line emission is detected, but the [N ii]/Hα ratio increases toward the nucleus of COS4-11363, indicating that COS4-11363 does have some (weak) line emission. We targeted the pair with SINFONI + AO to allow us to robustly separate the line emission from the two galaxies. The right-hand panel of Figure 3 shows a map of the Hα flux from our SINFONI-AO data, with contours of the HST F160W emission overlaid. The Hα flux primarily originates from COS4-11337 but is also detected at the location of COS4-11363, consistent with the results from KMOS^3D and the 3D-HST grism spectroscopy.

SDSS J090122.37 + 181432.3 (abbreviated to J0901) is a strongly lensed, triply imaged galaxy at z = 2.259, first reported by Diehl et al. (2009). The left-hand panel of Figure 4 shows an IJH color HST composite image of the region surrounding the lensing cluster, which is at a redshift of z = 0.346. The green circles indicate the locations of the northeastern (NE) and western (W) images of J0901. The southeastern (SE) image is at the center of the white and red rectangles, which indicate the coverage of our SINFONI seeing-limited and adaptive optics data sets, respectively. We targeted the SE arc because the similarly bright NE fold arc is strongly distorted and images only part of the source.

Zoom In Zoom Out Reset image size

Rest-frame UV and optical spectra of J0901 reveal high [N ii]/Hα and [O iii]/Hβ ratios (Hainline et al. 2009) and clear [N V] emission (Diehl et al. 2009), indicative of AGN activity. The [N ii]+Hα emission line complex shows a clear broad component that likely traces an AGN-driven outflow (Genzel et al. 2014).

We obtained K-band SINFONI observations of J0901, using the seeing-limited mode to probe the extended flux of the galaxy and the adaptive optics mode to obtain a high-resolution view of the center of the galaxy. The right-hand panel of Figure 4 shows an image plane Hα flux map constructed from the combination of the seeing-limited and AO data sets. The fluxes for the pixels inside the red rectangle were measured from the AO data, and the fluxes for the pixels in the outer region were measured from the seeing-limited data. The white contours show the spatial distribution of the [N ii] emission.

We also observed J0901 in the H band using the LUCI long slit spectrograph on the Large Binocular Telescope (LBT), aided by the Advanced Rayleigh guided Ground-layer adaptive Optics System (ARGOS; Rabien et al. 2019). The observations consist of one 45 minute block from LUCI-1 on 2018 March 2 and one 45 minute block from each of LUCI-1 and LUCI-2 on 2018 March 3. The three blocks were reduced independently using the Flame pipeline (Belli et al. 2018) and then combined, weighted by the uncertainty on each pixel. The spatial resolution of the combined data set is 037, and the spectral resolution calculated from a stack of skylines close to the wavelength of [O iii]λ5007 is 55 km s⁻¹. The final 2D spectrum was flux calibrated using the slit alignment star.

The strong gravitational lensing causes magnification and distortion of the light, which must be accounted for in order to derive the intrinsic properties of J0901. We use Lenstool and archival HST imaging to build a model for the mass distribution of the lensing cluster, as described in Appendix A.1. This model allows us to convert from observed (image plane) flux distributions to intrinsic (source plane) flux distributions and derive the intrinsic properties of the galaxy.

We measure an intrinsic [O iii] luminosity of log(${L}_{[{\rm{O}}\mathrm{III}]}$/erg s⁻¹) = 43.5, corresponding to an AGN bolometric luminosity of log(${L}_{\mathrm{bol},\mathrm{AGN}}$/erg s⁻¹) = 46.3 (see Appendix A.2). We note that the AGN luminosity is relatively uncertain due to both the lack of hard X-ray observations and the uncertainties associated with the source plane reconstruction. We fit SED models to the magnification corrected F160W, F814W, and F435W fluxes, yielding a stellar mass of log(M_*/M_⊙) = 11.2 and an A_V of 1.2 (Appendix A.3). Using the 160 μm flux measurement presented in Saintonge et al. (2013), we derive an SFR of 200 M_⊙ yr⁻¹ (Appendix A.4), which places J0901 on the SFR main sequence (red star in Figure 1).

Ionized gas outflows are generally observed as broadened (FWHM ≳ a few hundred km s⁻¹) emission line components underneath the narrower emission produced by gas in the disk of the galaxy. The outflow and disk components can be robustly separated for individual spaxels in integral field observations of local galaxies, but even with our deep observations, the emission line signal-to-noise ratio (S/N) is only high enough to permit a robust disk-outflow decomposition in the brightest spaxels. Therefore, we calculate the properties of the AGN-driven outflows in K20-ID5, COS4-11337, and J0901 using spectra integrated across the nuclear regions of the galaxies.

The method used to extract the nuclear spectra is described in detail in Section 2.5.1 of Förster Schreiber et al. (2019). In short, each datacube was median subtracted to remove continuum emission, σ-clipped to remove skyline residuals, and smoothed spatially using a Gaussian filter with a width of 3–4 pixels (FWHM). For each spaxel in each datacube, we simultaneously fit the [N ii]λ6548, Hα, and [N ii]λ6583 lines as single Gaussians with a common velocity offset and velocity dispersion, and then we shifted the spectrum so that the narrow line cores were centered at zero velocity. This velocity shifting removes (and therefore prevents artificial line profile broadening associated with) large-scale gravitationally driven velocity gradients, but it has a minimal impact on the shapes of the outflow line profiles because their line widths are ∼ 5–10× larger than the maximum velocity shifts. From the velocity-shifted cubes, we extracted spectra integrated over the region where broad outflow emission was detected (∼1–3 kpc in radius).

We fit the line profiles in each nuclear spectrum as a superposition of two kinematic components—one for the galaxy and one for the outflow. An example fit (to the nuclear spectrum of COS4-11337) can be seen in the left-hand panel of Figure 8. We again assumed a common velocity offset and velocity dispersion for all lines in each kinematic component, and we fixed the [N ii]λ6583/[N ii]λ6548 and [O iii]λ5007/[O iii]λ4959 ratios to 3.0 (the theoretical value set by quantum mechanics).

We use the two-component fits to the nuclear spectra to determine which spectral channels are dominated by the outflow component, and we integrate the flux over the outflow channels in each spaxel to create maps of the outflow emission. We compare the curves of growth of the outflow emission and the point-spread function (PSF) to confirm that the outflows are resolved. We calculate the intrinsic (PSF-corrected) size of the outflows by modeling the observed outflow emission as a 2D Gaussian (representing the intrinsic outflow emission) convolved with the PSF, and adopt the HWHM of the Gaussian as the radius of the outflow. In this modeling, we use the empirically derived average AO PSF from the SINS/zC-SINF AO Survey (Förster Schreiber et al. 2018), which has sufficiently high S/N that both the core (AO corrected) and wing (uncorrected) components of the PSF can be robustly characterized. The average PSF is constructed from data sets obtained under similar conditions to our targets, and the FWHM measured from the curve of growth of the average PSF is 018, which is similar to the FWHM values measured for our data sets (see Table 2).

The outflow velocity is calculated by taking the full width at zero power (FWZP) of the entire [N ii]+Hα complex, subtracting the velocity separation of the [N ii] doublet lines (1600 km s⁻¹), and dividing by two (following Förster Schreiber et al. 2019). This measurement yields the maximum line-of-sight velocity of the outflowing material, which should be close to the true outflow velocity for a constant velocity wide angle outflow.

The mass outflow rate ${\dot{M}}_{\mathrm{out}}$ of a constant velocity spherical or (multi-)conical outflow can be calculated from the Hα luminosity of the outflow (${L}_{{\rm{H}}\alpha ,\mathrm{out}}$) as follows:

$$\begin{eqnarray}\begin{array}{rcl}{\dot{M}}_{\mathrm{out}}\,({M}_{\odot }\,{\mathrm{yr}}^{-1}) & = & 33\left(\displaystyle \frac{1000\,{\mathrm{cm}}^{-3}}{{n}_{e}}\right)\left(\displaystyle \frac{{v}_{\mathrm{out}}}{1000\,\mathrm{km}\,{{\rm{s}}}^{-1}}\right)\\ & & \times \,\left(\displaystyle \frac{1\,\mathrm{kpc}}{{R}_{\mathrm{out}}}\right)\left(\displaystyle \frac{{L}_{{\rm{H}}\alpha ,\mathrm{out}}}{{10}^{43}\,\mathrm{erg}\,{{\rm{s}}}^{-1}}\right)\end{array}\end{eqnarray} \tag{ 1 }$$

${R}_{\mathrm{out}}$ is the radial extent of the outflow, ${v}_{\mathrm{out}}$ is the outflow velocity, and n_e is the local electron density of the ionized gas in the outflow (Genzel et al. 2011; Newman et al. 2012b).¹² The mass loading factor η is defined as η = ${\dot{M}}_{\mathrm{out}}$/SFR_best.

The biggest uncertainty in the calculation of ${\dot{M}}_{\mathrm{out}}$ and η is the electron density, which is extremely challenging to constrain for AGN-driven outflows at z ∼ 2. In principle, the electron density of the ionized gas in the outflow can be measured from the [S ii]λ6716/[S ii]λ6731 ratio in the outflow component (Osterbrock & Ferland 2006). However, the [S ii] lines are weak compared to Hα and [N ii], and they become strongly blended in the presence of a broad outflow component, making it very difficult to constrain a two-component decomposition of the [S ii] line profile. Förster Schreiber et al. (2019) fit two kinematic components to a high S/N stacked spectrum of 30 AGN-driven outflows at $0.6\lt z\lt 2.6$ and found that the average electron density of the outflowing material is ∼ 1000 cm⁻³. Their stack includes two of the three galaxies in our sample. Several recent studies of AGN-driven outflows at low and intermediate redshift have found similarly high densities in the outflowing gas (e.g., Perna et al. 2017; Kakkad et al. 2018; Husemann et al. 2019; Shimizu et al. 2019). We therefore adopt n_e = 1000 cm⁻³.

The Hα emission from the outflowing gas is attenuated by dust along the line of sight to the nuclear regions of the galaxies. To calculate the intrinsic mass outflow rates, we need to correct the observed Hα fluxes for extinction.

Our observations cover both the Hα and Hβ emission lines. When Hβ is detected, the best method for correcting emission line luminosities for extinction is to use the Balmer decrement, which directly probes the attenuation along the line of sight to the nebular line-emitting regions. The Hβ line is too weak to robustly separate into disk and outflow components, and therefore, we adopt the integrated Balmer decrement. The theoretical unattenuated Balmer decrement for Case B recombination at T = 10⁴ K is 2.86 (Osterbrock & Ferland 2006). However, this value can increase to 3.1 in the presence of an AGN (Gaskell & Ferland 1984). Therefore, we adopt an intrinsic Balmer decrement of 3.1. We assume the nebular extinction follows the Cardelli et al. (1989) curve.

In cases where the Balmer decrement cannot be measured, we correct for extinction using the global continuum A_V of the galaxy and account for extra attenuation of the nebular emission using the empirical formula presented by Wuyts et al. (2013; A_Hα = 1.9 × ${A}_{\mathrm{stars}}$–0.15n × ${A}_{\mathrm{stars}}^{2}$). The empirical correction assumes that the extinction follows the Calzetti et al. (2000) curve, but the functional form was chosen to produce the best agreement between the Hα and UV + IR SFRs for SFGs at 0.7 < z < 1.5 (i.e., A_Hα ∼ −2.5 log₁₀ (SFR_Hα/SFR_UV+IR)), and therefore, the derived ${A}_{{\rm{H}}\alpha }$ should be largely independent of the chosen attenuation curve.

We note that in both cases, we assume that the outflow emission and galaxy nebular emission experience similar attenuation, which may not be the case if the two components have different spatial distributions. Higher S/N and/or space-based observations of the Hβ line are required to calculate the Balmer decrements of the galaxy and outflow components separately.

Our SINFONI-AO data reveal previously unresolved kinematic structures in K20-ID5, providing key insights into the nature of the line emission in the nuclear region of the galaxy. The kinematic properties of K20-ID5 are summarized in Figure 5.

Zoom In Zoom Out Reset image size

The first row of panels shows the velocity fields measured from the KMOS and SINFONI-AO data. At the lower resolution and coarser spatial sampling of the KMOS data, the velocity field looks quite regular, but the higher-resolution SINFONI-AO data reveal a twist and a steepening of the velocity gradient in the central 04. The bottom panel of Figure 5 shows the 1D velocity profile, extracted in 5 × 5 spaxel apertures along the kinematic major axis of the galaxy (PA = −845). The colored dots trace the velocity of the ionized gas as a function of galactocentric distance in the SINFONI-AO data, and the purple squares show the velocity profile extracted from the KMOS data. The velocities are derived from single-component Gaussian fits to the Hα and [N ii] lines, but we verified that very similar profiles are recovered when performing two-component fitting to account for the presence of the broad emission component in the nuclear region. Therefore, the SINFONI-AO data reveal irregular narrow component kinematics in the nuclear region of K20-ID5.

We exploit the regularity of the KMOS velocity field to model the kinematics of the extended disk. We model the velocity field using Dysmalpy, a Python implementation of the dynamical fitting code Dysmal (Cresci et al. 2009; Davies et al. 2011; Wuyts et al. 2016; Übler et al. 2018). Dysmal is a forward modeling code that builds a model for the mass distribution (including one or more of a disk, bulge, and dark matter halo), produces a mock datacube with a given spatial and spectral sampling, convolves the datacube with the given line spread function and spatial PSF, produces model velocity and velocity dispersion fields, and compares the model to the data. We fit the KMOS velocity field using an exponential disk model with the mass, effective radius, inclination, and position angle of the disk as free parameters. We do not fit the velocity dispersion field because it is heavily influenced by the outflow in the central regions.

The best-fit model for the extended velocity field of K20-ID5 is shown in the second row of Figure 5, at the resolution/sampling of the KMOS data in the left column and at the resolution/sampling of the SINFONI-AO data in the right column. The third row of panels shows the residuals after subtracting the best-fit model from the measured velocity field for each of the data sets. The best-fit model reproduces the KMOS velocity field very well, leaving only small amplitude residuals (less than 55 km s⁻¹ in 90% of spaxels) with no clear spatial structure. On the other hand, the discontinuity in the SINFONI-AO velocity field is visible as a strong velocity gradient in the residuals, going from −120 km s⁻¹ to +240 km s⁻¹ at an angle of ∼ 25° to the major axis of the disk.

The twist in the narrow component kinematics could trace either a misaligned core or an outflow. In order to determine which of these is more likely, we examine how the line profiles and the single-component [N ii]/Hα ratio and velocity dispersion differ between the nuclear region and the extended disk of the galaxy. The left-hand panels of Figure 6 show maps of the single-component [N ii]/Hα ratio (top panel) and velocity dispersion (bottom panel) from the SINFONI-AO data. The [N ii]/Hα ratio is elevated in the nucleus compared to the extended disk region, and more interestingly, it shows a sharp boundary between the nucleus and the extended disk on the eastern (receding) side of the galaxy. This sharp boundary coincides with an abrupt change in the magnitude of the velocity residuals. The velocity dispersion is also elevated in the nuclear region but decreases gradually with increasing galactocentric distance.

Zoom In Zoom Out Reset image size

The right-hand panels of Figure 6 show a position–velocity (p–v) diagram extracted along the E–W direction (i.e., the p–v diagram and the maps on the left side of the figure share the same x-axis) and spectra extracted in eight slices that are numbered on both the p–v diagram and the corresponding panels. Slices 1, 6, 7, and 8 trace the regularly rotating extended disk of the galaxy and show narrow line emission and low [N ii]/Hα ratios. Slices 2–4 trace the kinematically anomalous nuclear region and show strong broad emission and high [N ii]/Hα ratios. Slice 5 traces the western edge of the nuclear region where the broad outflow emission is still visible but is overpowered by narrow emission from the galaxy disk. The [N ii]/Hα ratio is intermediate between the nuclear and extended disk regions, but the narrow lines follow the velocity field of the extended disk.

The kinematically anomalous nuclear region of K20-ID5 is characterized by broad line profiles and high [N ii]/Hα ratios, whereas the regularly rotating regions at larger radii are dominated by narrow disk emission with much lower [N ii]/Hα ratios. This provides strong circumstantial evidence that the residual nuclear velocity gradient is tracing the outflow.

The nuclear velocity gradient reflects the kinematics of the narrow line peaks, leading to the conclusion that both the narrow and broad line emission in the nuclear region of K20-ID5 must be primarily associated with outflowing material. In other words, the outflow has a non-Gaussian line-of-sight velocity distribution that can be approximated by the superposition of a narrow and a broad Gaussian component. Three-dimensional biconical outflow models suggest that outflow emission line profiles can have non-Gaussian shapes and may resemble the superposition of a narrow and a broad component depending on the outflow velocity and geometry (e.g., Bae & Woo 2016). AGN-driven outflows with non-Gaussian line profiles and/or requiring multiple Gaussian components have been observed both in the local universe (e.g., Fischer et al. 2018; Shimizu et al. 2018; Davies et al. 2020) and at high redshift (e.g., Liu et al. 2013; Vayner et al. 2017). There are also many examples of AGN host galaxies with outflow-dominated narrow line region kinematics (e.g., Fischer et al. 2013; Liu et al. 2013), similar to what we observe in K20-ID5.

We note that the clear distinction between the kinematics and line ratios of the nuclear outflow and the extended gas in K20-ID5 confirm that the extended gas is tracing the rotating disk of the galaxy rather than a galaxy-scale outflow (see discussion in Loiacono et al. 2019).

We measure the properties of the nuclear outflow from a deep K-band spectrum extracted over the kinematically anomalous nuclear region of K20-ID5, where the outflow dominates the line emission (see Section 4.1). We combine the SINFONI-AO and KMOS data sets for a total on-source integration time of 31 hr. The spectrum is shown in the left-hand panel of Figure 7.

Zoom In Zoom Out Reset image size

Based on the FWZP of the [N ii]+Hα complex, we measure an outflow velocity of 1410 ± 56 km s⁻¹.¹³ Our best-fit exponential disk model yields a circular velocity of ∼ 240 km s⁻¹, corresponding to a halo escape velocity of ∼720 km s⁻¹ (assuming ${v}_{\mathrm{escape}}\sim 3\,{v}_{\mathrm{circ}}$; Weiner et al. 2009). The outflow velocity is a factor of two larger than the escape velocity, suggesting that a significant fraction of the outflowing material could be ejected from the halo.

The mass outflow rate is derived from the Hα luminosity of the outflow, as described in Section 3.4. Our analysis in Section 4.1 revealed that the vast majority of the (narrow and broad) line emission in the nuclear region of K20-ID5 is associated with the outflow. Therefore, we calculate the outflow properties assuming that 100% of the nuclear Hα flux originates from the outflow ("outflow" model).

Even though the outflow dominates the line emission, it is necessary to fit two Gaussian components to recover the shape of the line profiles, as shown in the left-hand panel of Figure 7. We sum the Hα fluxes of the two components to obtain the total Hα flux. We do not use the seeing-limited SINFONI H-band data to constrain the emission line fitting, because the nuclear [O iii] emission is strongly attenuated (see Figure 2), and we do not find any evidence for a broad or blueshifted component in the [O iii] line profile.

There is a significant amount of dust attenuating the nebular line emission from the nuclear region of K20-ID5 (see Section 2.1). We use the SINFONI H-band data to measure the Hβ flux and find Hα/Hβ = 8.0 ± 0.7, corresponding to an A_V of 2.7. This is significantly larger than the global continuum A_V of 1.3 but is consistent with the results of Loiacono et al. (2019), who measured a global Balmer decrement of 8.3 ± 1.8, and with those of Scholtz et al. (2020), who measured a nuclear Balmer decrement of ${8.7}_{-1.8}^{+2.3}$. We use the measured Balmer decrement to correct the Hα luminosity of the outflow for extinction.

The outflow is well resolved (see right-hand panel of Figure 7) and has a PSF-deconvolved HWHM of 1.0 ± 0.2 kpc. Combining all of these quantities, we measure a mass outflow rate of 262 ± 76 M_⊙ yr⁻¹, corresponding to a mass loading factor of η = 0.78 ± 0.23.

For comparison, we also calculate the outflowing mass using just the Hα luminosity of the broad component ("galaxy + outflow" model). In this case, we measure ${\dot{M}}_{\mathrm{out}}$ = 103 ± 30 M_⊙ yr⁻¹ and η = 0.31 ± 0.09. The outflow parameters for the outflow only (fiducial) and galaxy + outflow models are listed in Columns 2 and 3 of Table 3, respectively.

Table 3.  Derived Outflow Parameters

Galaxy	K20-ID5		COS4-11337	J0901
Model Type	Outflow	Galaxy + Outflow	Galaxy + Outflow	Galaxy + Outflow
(1)	(2)	(3)	(4)	(5)
(a) ${R}_{\mathrm{out}}$ (kpc)	1.0 ± 0.2		0.9 ± 0.2	0.47 ± 0.07
(b) ${v}_{\mathrm{out}}$ (km s−1)	1410 ± 56		1459 ± 66	650 ± 46
(c) ${\dot{M}}_{\mathrm{out}}$ (M⊙ yr−1)	262 ± 76	103 ± 30	61 ± 6	25 ± 8
(d) η (=${\dot{M}}_{\mathrm{out}}$/SFRbest)	0.78 ± 0.23	0.31 ± 0.09	0.15 ± 0.03	0.12 ± 0.04

Note. For COS4-11337 and J0901, we consider only a Galaxy + Outflow model, while for K20-ID5, we also consider a model where all of the nuclear line emission is associated with the outflow (Outflow). The boldface font indicates the fiducial model for each galaxy. The rows are as follows: (a) half-light radial extent of the outflow emission; (b) outflow velocity, defined as (FWZP_{[N ii]+Hα}–1600)/2; (c) mass outflow rate, calculated using Equation (1); and (d) mass loading factor.

Download table as:  ASCIITypeset image

The detection of a velocity gradient across the nucleus of K20-ID5 makes it possible to place some constraints on the geometry of the outflowing material. The velocity difference between the approaching and receding sides of the outflow is Δv ∼ 360 km s⁻¹ (see Figure 5), which is significantly smaller than the outflow velocity (1410 km s⁻¹). This suggests that the outflow is quasi-spherical, because a large opening angle (≳60°) is required to produce a large range of projected outflow velocities at every radius, while maintaining a similar average velocity across the entire outflow (see, e.g., models in Liu et al. 2013). The small Δv could also be produced by a collimated outflow almost perpendicular to the line of sight, but in a collimated outflow, there would only be a small range of velocities at each radius, and therefore, this scenario cannot account for the large observed line width.

It is also possible to place some constraints on the velocity profile of the outflowing material. There is no evidence for any radial variation in the FWZP of the [N ii]+Hα complex, suggesting that the outflow velocity is approximately constant out to the maximum radius at which it is detected (at least ∼ 5 kpc; see Figure 6).

We measure the properties of the outflow from COS4-11337 by fitting the KMOS H- and K-band nuclear spectra as a superposition of two Gaussian components. We use the KMOS K-band data in favor of the SINFONI-AO K-band data because the KMOS K-band observations have 3.3 times the integration time (corresponding to a factor of ∼1.8 higher S/N) and allow us to perform a robust galaxy + outflow decomposition of the nuclear line profile, which is not possible using the SINFONI data alone. Although COS4-11363 and COS4-11337 are partially blended in the KMOS data, the SINFONI-AO data indicate that the line emission from COS4-11363 is weak and is confined to the nucleus of the galaxy (see Figure 3). Therefore, the contribution of COS4-11363 to the nuclear spectrum of COS4-11337 should be negligible.

The two-component fit to the spectrum of COS4-11337 is shown in Figure 8. The K-band spectrum shows very broad wings, revealing the presence of a fast outflow. Based on the FWZP of the [N ii]+Hα complex, we measure an outflow velocity of 1459 ± 66 km s⁻¹. COS4-11337 has a circular velocity of 150 ± 60 km s⁻¹ (Wisnioski et al. 2018), corresponding to an escape velocity of ∼ 450 km s⁻¹. The outflow velocity is more than a factor of three larger than the escape velocity, indicating again that a significant fraction of the outflowing material could potentially be expelled from the host halo.

Zoom In Zoom Out Reset image size

The redshift of this system places the Hβ line at a wavelength with bad skyline residuals (see Figure 8), and as a result, we cannot derive a reliable Balmer decrement for COS4-11337. Therefore, we correct the Hα luminosity of the outflow for extinction using the global A_V, as described in Section 3.5. The outflow is resolved (see Figure 9) and has a PSF-deconvolved HWHM of 0.9 ± 0.2 kpc. We find a mass outflow rate of 61 ± 6 M_⊙ yr⁻¹ and a mass loading factor of η = 0.15 ± 0.03. The derived outflow parameters are listed in Column 4 of Table 3.

Zoom In Zoom Out Reset image size

COS4-11337 and COS4-11363 are resolved and clearly separated in the SINFONI-AO K-band data, allowing us to extract and analyze the spectrum of COS4-11363, which is shown in the left-hand panel of Figure 10. Only one emission line is detected. It is relatively narrow and lies close to the observed wavelength of [N ii]λ6583 in COS4-11337, which is indicated by the blue dashed line. The single emission line in the spectrum of COS4-11363 could trace either [N ii]λ6583 at z = 2.097, in which case the dv between the galaxies would be ∼ 140 km s⁻¹, or Hα at z = 2.107, in which case the dv would be ∼ 1100 km s⁻¹. The 3D-HST grism redshift is z_grism = 2.103, in between the two possible spectroscopic redshifts.

Zoom In Zoom Out Reset image size

We break the redshift degeneracy by utilizing archival ALMA observations. COS4-11337/11363 was observed for 90 minutes in Band 4 as part of program 2016.1.00726.S (PI: A. Man). The observations cover the CO(4-3) line and have a spatial resolution of 039, which is sufficient to resolve the two galaxies. The left-hand panel of Figure 11 shows the F160W image of the system, with contours of the CO(4-3) emission (at levels of 2σ and 3σ) overlaid. Despite the relatively short integration time, CO(4-3) emission is detected near the nucleus of COS4-11363.¹⁴ The right-hand panel shows the CO(4-3) spectrum associated with the peak of the emission. The spectrum is plotted as a function of velocity offset from COS4-11337. The purple and green dashed lines indicate where CO(4-3) would fall if the redshift of COS4-11363 were z = 2.097 or z = 2.107, respectively. The ALMA data clearly favor the z = 2.097 scenario, indicating that the line detected in the SINFONI-AO K-band spectrum is [N ii]λ6583 and that the Δv between COS4-11363 and COS4-11337 is $\leqslant$ 150 km s⁻¹.

Zoom In Zoom Out Reset image size

The detection of [N ii]λ6583 emission without strong Hα emission indicates that the [N ii]/Hα ratio in this galaxy must be significantly higher than observed in normal star-forming galaxies. The average [N ii]/Hα ratio for a pure star-forming, log(M_*/M_⊙) = 11.1 galaxy on the mass–metallicity relation at z = 2.1 is [N ii]/Hα ∼ 0.4 (based on the mass-redshift-metallicity parameterization in Tacconi et al. (2018) and the [N ii]/Hα-metallicity calibration from Pettini & Pagel 2004). However, to determine the intrinsic [N ii]/Hα ratio in the nucleus of COS4-11363, we must account for the fact that the Hα emission line coincides with a deep photospheric absorption feature in the spectrum of A stars. The best-fit SED models for COS4-11363 support the presence of strong Balmer absorption features, regardless of whether we adopt a truncated or exponentially declining star formation history. We scale the best-fit SED to match the continuum level of COS4-11363 and subtract this scaled best-fit SED from the observed spectrum (shown in the left-hand panel of Figure 10) to obtain a pure emission line spectrum (right panel of Figure 10). We fit the [N ii] and Hα lines in this emission line spectrum as single Gaussians, using the same fitting algorithm with the same parameter restrictions as described for our multicomponent fitting process (see Section 3.1). We measure an [N ii]/Hα ratio of 2.6 ± 0.4—a factor of 6.5 higher than expected for a pure star-forming galaxy.

The high [N ii]/Hα ratio indicates that the line-emitting gas must be collisionally excited and/or ionized by sources other than young stars. We measure an Hα equivalent width of 5.8 ± 2.5 Å, which exceeds the maximum value of 3.0 to be consistent with ionization by evolved stellar populations (e.g., Cid Fernandes et al. 2011; Belfiore et al. 2016). We therefore suggest that the line emission is most likely to be powered by either shock excitation or AGN activity.

The strongest evidence for the source of the line emission comes from the velocity dispersion map, shown in Figure 12. The velocity dispersion peaks at ∼ 800 km s⁻¹ at the nucleus of COS4-11337, where the outflow is launched. However, it remains elevated above 500 km s⁻¹ along the entire region connecting COS4-11363 and COS4-11337, before dropping to ∼ 250 km s⁻¹ at the nucleus of COS4-11363. This suggests that the outflow from COS4-11337 may be propagating toward its companion. Based on the outflow velocity and the projected separation between the galaxies, the travel time between the nuclei is ∼4 Myr. If the outflow collided with the ISM of COS4-11363, it is likely to have driven large scale shocks, producing the high observed [N ii]/Hα ratios. This is a potential example of AGN feedback acting on both galactic scales (by transporting mass and energy in the outflow) and circumgalactic scales (by driving shocks through the ISM of the companion galaxy).

Zoom In Zoom Out Reset image size

Tidal torques are likely to be an additional source of shock excitation in both COS4-11363 and COS4-11337. In the local universe, interacting/merging systems show prominent line emission from shock excited gas with a typical FWHM of 250–500 km s⁻¹ (e.g., Monreal-Ibero et al. 2006; Farage et al. 2010; Rich et al. 2011, 2015). This FWHM is similar to the width of the line emission from COS4-11363 but is a factor of ∼4 narrower than the broad line emission from COS4-11337, suggesting that an outflow is most likely required to explain the kinematics of the gas in COS4-11337. However, we cannot rule out a scenario where the shock excitation in COS4-11363 is purely induced by the interaction.

It is interesting to speculate on the possible impact that the outflow from COS4-11337 may have had on the star formation activity of COS4-11363. The 3D-HST grism spectrum and the best-fit SED for COS4-11363 both support the presence of a prominent Balmer break, and the SED fitting favors a rapid decrease in the SFR over the last hundred megayears. From the Hα flux, we measure an instantaneous SFR of <2 M_⊙ yr⁻¹, which places the galaxy two orders of magnitude below the main-sequence SFR. The detection of CO associated with this galaxy indicates that the recent truncation of the SF activity was not simply the consequence of an exhausted gas reservoir. The quenching could plausibly have been triggered by the outflow from COS4-11337 plowing into the ISM of COS4-11363, driving large scale shocks and preventing the gas from collapsing to form stars. However, the galaxy–galaxy interaction is also likely to have had a significant impact on the SF activity in this system. Deep spectroscopy covering the region around 4000 Å will assist in more accurately constraining the star formation history and evolution of this galaxy.

The nuclear spectrum of J0901 was extracted from the source plane datacube (which was created by applying the source plane reconstruction to each spectral channel individually) and is shown in Figure 13. The [N ii] line is significantly broader than the Hα line, and our two-component fitting reveals that this is because the Hα emission peak is dominated by the galaxy component whereas the [N ii] emission peak is dominated by the broader outflow component. From the FWZP of the [N ii]+Hα complex, we measure an outflow velocity of 650 ± 46 km s⁻¹. This is a factor of ∼2 smaller than the outflow velocities measured for K20-ID5 and COS4-11337. Kinematic modeling indicates that J0901 has an inclination-corrected circular velocity of ∼ 260 km s⁻¹ (Rhoads et al. 2014; Sharon et al. 2019), corresponding to an escape velocity of 780 km s⁻¹. The outflow velocity is approximately 85% of the escape velocity, and therefore, the majority of the outflowing material detected in our data is unlikely to be able to escape from the galaxy halo.

Zoom In Zoom Out Reset image size

Figure 14 shows source plane maps of the narrow galaxy emission (left panel) and the broad outflow emission (middle panel) over the region covered by the SINFONI-AO data. The approximate shape of the source plane PSF at the location of the nucleus is indicated by the hatched ellipse at the bottom right of the middle panel. Both the galaxy and outflow emission show a single peak in the nucleus, unlike the image plane [N ii] map (white contours in the right-hand panel of Figure 4) in which two clear peaks are visible. Our lens modeling reveals that the secondary peak in the image plane flux distribution is a re-image of the northern side of the nucleus. This result is confirmed by Sharon et al. (2019), who independently developed a lens model for the J0901 lensing cluster in parallel with our team.

Zoom In Zoom Out Reset image size

The narrow emission is extended and traces the disk of the galaxy. We note that this figure only shows the central region of the galaxy covered by our AO data, and the true extent of the disk is significantly larger. In contrast, the broad outflow emission is very centrally concentrated. The right-hand panel of Figure 14 compares the curves of growth for the outflow emission and the PSF. The source plane spatial resolution of the SINFONI-AO data at the nucleus is 122 × 56 mas (FWHM), corresponding to a circularized FWHM of 83 mas and a physical resolution of 680 pc. Despite the factor of two improvement in spatial resolution due to the lensing, the outflow emission is only slightly more extended than the PSF, indicating that it is marginally resolved. The deconvolved Gaussian HWHM of the outflow emission is 470 ± 70 pc—a factor of ∼ 2 smaller than the outflows from K20-ID5 and COS4-11337.

The outflowing mass is calculated from the source plane Hα luminosity of the outflow component. We robustly detect the Hβ line in our LUCI long slit spectrum of J0901, and therefore we correct the Hα luminosity for extinction using the Balmer decrement. We measure a Balmer decrement of Hα/Hβ = 6.8 ± 0.6, corresponding to A_V = 2.2. We find a mass outflow rate of 25 ± 8 M_⊙ yr⁻¹, and a mass loading factor of η = 0.12 ± 0.04 (listed in Column 5 of Table 3).

Although all three of our galaxies host AGNs, the observed ionized gas outflows could plausibly be driven by SF, AGN activity, or both. Outflows can be energy conserving or momentum conserving depending on how fast the hot wind material cools as it interacts with the ISM (e.g., King 2010; King & Pounds 2015).

The AGN ionizing radiation field injects momentum into the surrounding material, driving a hot wind with a momentum rate of ${\dot{p}}_{\mathrm{wind}}$ ∼ L_AGN/c. If the kinetic energy of the wind couples efficiently to the ISM, it can drive a galaxy-scale energy conserving outflow with kinetic power ${\dot{E}}_{\mathrm{out}}$ ≲ 0.05 L_AGN and momentum rate ${\dot{p}}_{\mathrm{out}}$ ≲ (5–20) L_AGN/c (Faucher-Giguère & Quataert 2012; Zubovas & King 2012). On the other hand, if the wind kinetic energy is efficiently radiated away, the result is a smaller-scale momentum conserving outflow with ${\dot{E}}_{\mathrm{out}}$ ≲ 10⁻³ L_AGN and ${\dot{p}}_{\mathrm{out}}$ ≲ L_AGN/c (King 2010; King & Pounds 2015).

Supernovae and stellar winds deposit energy at a rate of ${\dot{E}}_{\mathrm{out}}$ ≲ 10⁻³ L_SF (Murray et al. 2005). Momentum is deposited through both radiation pressure from massive stars (${\dot{p}}_{\mathrm{rad}}$ ∼ L_SF/c) and supernova explosions. The initial momentum of the ejecta from a single supernova is ${p}_{{\rm{i}},\mathrm{SN}}$ ∼ 3000 M_⊙ km s⁻¹, and assuming one supernova per 100 yr per solar mass of SF, this corresponds to a total momentum injection rate by supernovae of ${\dot{p}}_{{\rm{i}},\mathrm{SN}}$ ∼ 0.7 L_SF/c (Murray et al. 2005). However, if the cooling time of the supernova ejecta is sufficiently long, the hot wind can sweep up a significant amount of ISM, and the final momentum rate of the outflow can be a factor of ∼ 10 larger than the initial wind momentum rate (${\dot{p}}_{{\rm{f}},\mathrm{SN}}$ ∼ 6 L_SF/c; Kim & Ostriker 2015). SF-driven outflows have been observed to propagate to distances of a few kiloparsecs at z ∼ 2 (e.g., Newman et al. 2012a, 2012b; Davies et al. 2019).

We calculate the kinetic powers and momentum rates of the outflows in our three galaxies and compare them to L_SF and L_AGN to determine if the outflows are driven by SF or AGN activity and if they are momentum or energy conserving (see Table 4).

Table 4.  Outflow Kinetic Powers and Momentum Rates, and Comparison to the Bolometric Luminosities of the AGN and the Young Stars

Galaxy		K20-ID5	COS4-11337	J0901
Model Type	Outflow	Galaxy + Outflow	Galaxy + Outflow	Galaxy + Outflow
(1)	(2)	(3)	(4)	(5)
log(${\dot{E}}_{\mathrm{out},{n}_{{\rm{e}}}=1000}$/erg s−1)	44.2 ± 0.1	43.8 ± 0.1	43.6 ± 0.1	42.5 ± 0.2
log(${\dot{p}}_{\mathrm{out},{n}_{{\rm{e}}}=1000}$/dyn)	36.4 ± 0.1	36.0 ± 0.1	35.7 ± 0.1	35.0 ± 0.1
${\dot{E}}_{\mathrm{out},{n}_{{\rm{e}}}=1000}$/LAGN	(4.0 ± 1.2) × 10−2	(1.6 ± 0.5) × 10−2	(2.8 ± 0.3) × 10−3	(1.5 ± 0.9) × 10−4
${\dot{p}}_{\mathrm{out},{n}_{{\rm{e}}}=1000}$/(LAGN/c)	17 ± 5	6.8 ± 2.0	(2.8±0.3) × 10−3	0.14 ± 0.08
${\dot{E}}_{\mathrm{out},{n}_{{\rm{e}}}=380}$/LSF,best	(3.4 ± 1.0) × 10−2	(1.3 ± 0.4) × 10−2	(7.1 ± 1.2) × 10−3	(1.1 ± 0.5) × 10−3
${\dot{p}}_{\mathrm{out},{n}_{{\rm{e}}}=380}$/(LSF,best/c)	14 ± 4	5.6 ± 1.7	2.9 ± 0.5	1.0 ± 0.4

Note. The columns are the same as in Table 3. When comparing to L_AGN, we adopt n_e = 1000 cm⁻³, and when comparing to L_SF, we adopt n_e = 380 cm⁻³, reflecting the different ionized gas densities of AGN-driven and SF-driven outflows at z ∼ 2 (see discussion in Section 7.1).

Download table as:  ASCIITypeset image

In this comparison, it is important to account for the 1/n_e dependence of the mass outflow rate. In our calculations, we have adopted n_e = 1000 cm⁻³, under the assumption that the outflows are AGN-driven. However, the typical electron density of the ionized gas in SF-driven outflows at z ∼ 2 is 380 cm⁻³ (Förster Schreiber et al. 2019), a factor of 2.6 lower. Therefore, when calculating ${\dot{E}}_{\mathrm{out}}$/L_SF and ${\dot{p}}_{\mathrm{out}}$/(L_SF/c), we multiply the mass outflow rates by a factor of 2.6. It is also important to consider that L_SF represents the global bolometric output of the young stars, but the AGN-driven outflows are launched from the nuclear regions of the galaxies, and therefore, only some fraction of L_SF will be available to drive the outflows.

The kinetic powers of the outflows from K20-ID5 and COS4-11337 are too large for the outflows to be SF-driven. The kinetic power of the K20-ID5 outflow is 0.04(±0.01) × L_AGN and the momentum rate is 17(±5) × L_AGN/c, suggesting that the outflow is energy conserving. However, we note that the energy and momentum ratios vary inversely with the AGN luminosity, which is poorly constrained due to the probable high column densities toward the nucleus (see Section 2.1). The coupling between the AGN radiation field and the outflow is less efficient in COS4-11337, which has an outflow kinetic power of 2.8(±0.3) × 10⁻³ L_AGN and a momentum rate of 1.1(±0.1) × L_AGN/c. The inefficient coupling suggests that the outflow could plausibly be momentum driven. However, the outflow extends to a distance of ∼5 kpc (see Section 5.4 and Figure 12), which is more in favor of the energy driving scenario.

The J0901 outflow has a smaller kinetic power than the outflows from K20-ID5 and COS4-11337, primarily due to the lower outflow velocity. The outflow has a kinetic power of 1.1(±0.5) × 10⁻³ L_SF and a momentum rate of 1.0(±0.4) × L_SF/c, and could therefore potentially be SF-driven. However, the spatial distribution of the Hα emission suggests that only ≳1/3 of the SF is occurring within the outflow region, in which case, the bolometric luminosity of the young stars would likely be insufficient to power the observed outflow. In addition, the outflow emission has a very high [N ii]/Hα ratio (3.3; see Figure 13), which is typical of AGN-driven outflows but is not observed in SF-driven outflows at z ∼ 2 (e.g., Newman et al. 2012b; Genzel et al. 2014; Davies et al. 2019; Förster Schreiber et al. 2019). Therefore, we suggest that the outflow is most likely to be AGN-driven; although, we cannot rule out a significant contribution from the SF. Only a very small fraction of the energy released by the AGN needs to couple to the ISM in order to drive the observed outflow. Given the very low coupling efficiency and fact that the outflow is confined to the circumnuclear regions, it seems likely that the outflow in J0901 is a momentum conserving AGN-driven outflow.

An important ingredient in our understanding of how AGN-driven outflows interact with their host galaxies is accurate measurements of the radial extents of AGN-driven outflows. In all three of our systems, we find that the majority of the outflow emission is concentrated within approximately the central kiloparsec, consistent with the findings of Förster Schreiber et al. (2014). However, with the exception of J0901, the outflows are not confined to the nuclear regions but extend well beyond the effective radii of the galaxies. In K20-ID5, the outflow propagates at an approximately constant velocity to a radius of at least 5 kpc (see Section 4.4). In COS4-11337, we see compelling evidence that the outflow has traveled beyond the galaxy and is shock heating the ISM in its companion galaxy COS4-11363 (see Section 5.4). Other studies of luminous AGNs at similar redshifts have found evidence that ionized outflows can propagate 5–10 kpc from the galaxy nuclei (e.g., Nesvadba et al. 2006, 2008; Harrison et al. 2012; Cresci et al. 2015; Brusa et al. 2018; Herrera-Camus et al. 2020). Put together, these results suggest that AGN-driven outflows have steep luminosity profiles, with a luminous core component in the central kiloparsec and a faint tail extending to several kiloparsecs, which may reflect a decrease in the surface brightness and/or density of the outflowing material as it expands out from the galaxy nuclei (e.g., Kakkad et al. 2018).

However, the detection of a confined (r_e = 470 ± 70 pc) outflow in J0901 emphasizes that not all AGN-driven outflows extend beyond the nuclear region. Fischer et al. (2019) reported another hundred-parsec-scale, ∼ 500 km s⁻¹ AGN-driven outflow in the lensed galaxy SGAS J003341.5 + 024217 (SGAS 0033+02 hereafter) at z = 2.39. The outflows in both J0901 and SGAS 0033 + 02 are so compact that they would not be resolved without gravitational lensing. Fischer et al. (2019) state that the outflow in SGAS 0033 + 02 would probably not be detectable if the system was not lensed, because the outflow emission would be overpowered by emission from the galaxy. The J0901 outflow is clearly visible even in an 06 (source plane) diameter aperture, primarily because the large [N ii]/Hα ratio in the broad component increases the contrast between the broad component and the surrounding continuum. However, it is possible that confined nuclear outflows are present but undetected in a nonnegligible fraction of massive galaxies at z ∼ 2.

The similarity in the AGN luminosities measured for COS4-11337 and J0901 suggests that the outflow extent is not determined by the current AGN luminosity; although, we reiterate that the uncertainty on the AGN luminosity of J0901 is relatively large. The AGNs in J0901 and SGAS 0033 + 02 could potentially have "switched on" relatively recently, so that the outflows have not yet had time to propagate beyond the circumnuclear regions. Alternatively, the AGN accretion energy may not couple efficiently to the gas in the nuclear regions (as appears to be the case for J0901), giving the outflows insufficient energy to propagate to larger radii. In their current states, these outflows are unable to directly impact gas on kiloparsec scales. However, the outflows are depositing a significant amount of kinetic energy within a few hundred parsecs of the galaxy nuclei. This deposition of energy will increase the amount of turbulence in the circumnuclear regions, and if the turbulent pressure becomes large enough, the star-forming gas could become stabilized against collapse (e.g., Guillard et al. 2012; Alatalo et al. 2015). Therefore, these small, lower-velocity outflows could potentially lead to a reduction of the star formation efficiency in the nuclear regions of their host galaxies. Further studies of confined outflows will assist in better characterizing this unique class of objects.

The near-infrared spectroscopic data analyzed in this paper probe only the ionized gas phase of the AGN-driven outflows in K20-ID5, COS4-11337, and J0901. However, galaxy-scale outflows are intrinsically multiphase and contain not only warm ionized gas but also cooler molecular and atomic gas as well as hotter X-ray emitting gas. Multiphase observations of outflows in local AGN host galaxies suggest that on galaxy scales, the molecular and neutral phases dominate the outflow mass and the mass outflow rate, but the ionized gas has a higher outflow velocity (e.g., Rupke & Veilleux 2013; Veilleux et al. 2013; Fiore et al. 2017; Fluetsch et al. 2019; Husemann et al. 2019; Shimizu et al. 2019; Herrera-Camus et al. 2020). In two quasar-driven outflows at z ∼ 1.5, the molecular phase has a factor of ∼2–5 higher mass outflow rate than the ionized phase but a factor of ∼2–4 lower outflow velocity (e.g., Vayner et al. 2017; Brusa et al. 2018). For a typical star-forming galaxy at z ∼ 2, Herrera-Camus et al. (2019) found that the molecular outflow rate is a factor of ∼5 higher than the ionized outflow rate.

Simulations predict that the hot (∼10⁷ K) phase should carry at least as much mass as the cooler gas phases (e.g., Nelson et al. 2019), but so far, the majority of the observational constraints come from studies of X-ray Broad Absorption Line (BAL) winds and Ultra-Fast Outflows (UFOs) on very small spatial scales, and these appear to have mass outflow rates similar to or lower than those of ionized gas outflows (e.g., Feruglio et al. 2015; Fiore et al. 2017; Tombesi et al. 2017).

Even when outflows can be observed in multiple phases, constructing an accurate budget of the mass and energy in the different phases is very challenging. Ionized gas outflow masses scale with the inverse of the electron density, which is a relatively poorly constrained quantity. Recent studies suggest that the luminosity-weighted density of ionized gas in AGN-driven outflows is ∼ 1000 cm⁻³ (e.g., Perna et al. 2017; Kakkad et al. 2018; Förster Schreiber et al. 2019; Husemann et al. 2019; Shimizu et al. 2019), but many studies in the literature adopt n_e ≲ 100 cm⁻³, indicating potential discrepancies on the order-of-magnitude level. CO-based molecular gas outflow rates scale with the CO-to-H₂ conversion factor (α_CO), for which the typically adopted values vary between 0.8 (the "ULIRG" value; e.g., Cicone et al. 2014) and 4.3 (the Milky Way value; Bolatto et al. 2013), and optically thin outflows with even lower conversion factors have been reported in two objects (Dasyra et al. 2016; Lutz et al. 2020).

Although the exact distribution of mass and energy between different outflow phases is poorly constrained, it is clear that the mass outflow rates and mass loading factors listed in Table 3 and the outflow coupling efficiencies listed in Table 4 are only lower limits. This must be taken into consideration when evaluating the potential impact of outflows on the evolution of their host galaxies.

The M_BH–σ (Ferrarese et al. 2001) and M_BH–M_bulge (Magorrian et al. 1998) relations provide indirect evidence to suggest that black holes coevolve with their host galaxies. The gravitational energy released by accretion onto supermassive black holes greatly exceeds the binding energy of the bulge, and therefore, AGN feedback is widely considered an important mechanism for shaping this relationship. Analytical theories predict that AGN-driven winds should have ${\dot{E}}_{\mathrm{out}}\,\sim$ 0.05 L_edd and that this relationship should naturally give rise to the M_BH–σ relation as a locus of balance between the momentum injection rate from the AGN and the gravitational potential of the bulge. Black holes above the M_BH–σ relation are predicted to drive galaxy-scale energy conserving outflows that eject gas from the bulge and prevent further black hole growth (e.g., King 2003; Zubovas & King 2012; Lapi et al. 2014). Various studies based on numerical simulations have reported that a 5% coupling efficiency is sufficient to drive strong outflows that halt star formation and black hole growth and leave galaxies on the M_BH–σ relation (e.g., di Matteo et al. 2005).

Figure 15 shows the relationship between the outflow kinetic power and the AGN bolometric luminosity for K20-ID5, COS4-11337, and J0901. For comparison, we also plot data for a literature compilation of AGN-driven ionized outflows at z ∼ 1–3 (Genzel et al. 2014; Harrison et al. 2016; Fiore et al. 2017; Fischer et al. 2019; Förster Schreiber et al. 2019; Herrera-Camus et al. 2019; Leung et al. 2019). All of the literature measurements have been scaled to n_e = 1000 cm⁻³ for consistency.

Zoom In Zoom Out Reset image size

There is a clear correlation between the AGN bolometric luminosity and the outflow kinetic power. The average ratio of the outflow kinetic power to the AGN bolometric luminosity is 0.02% (black dotted line). There is a large scatter around the average (primarily driven by variations in outflow velocity at a fixed AGN luminosity), but in the vast majority of cases, the coupling efficiency is well below the 5% level suggested by the models (black dashed line). K20-ID5 is one of the extreme cases falling close to the 5% line, but the coupling factor for COS4-11337 is a factor of 10 lower at 0.3%. J0901 is another factor of 10 lower at 0.02%, but this is not surprising given the likely momentum conserving nature of the outflow. Even if we were to assume an electron density of 100 cm⁻³, the average coupling factor for the full sample would be 0.2%—still a factor of 25 too low. The coupling between the AGN ionizing radiation field and the ionized gas outflows does not appear to be efficient enough for the M_BH–σ relation to be the consequence of self-regulating black hole feedback. Anglés-Alcázar et al. (2013) showed that a torque-limited accretion model (in which the inflow rate onto the black hole accretion disk is driven by gravitational instabilities in the galaxy disk) naturally reproduces the M–σ relation without the need for any coupling between the AGN accretion energy and gas on galaxy scales.

Accounting for the mass and energy in other phases of the outflows would result in higher ${\dot{E}}_{\mathrm{out}}$ values and may partially alleviate the discrepancy with first-order expectations for self-regulated black hole growth. However, for zC400528, a normal AGN host galaxy at z ∼ 2, the molecular and ionized phases have similar kinetic powers (Herrera-Camus et al. 2019), and therefore, the overall coupling efficiency does not change significantly depending on whether only the ionized phase (filled yellow pentagon in Figure 15) or both the ionized and molecular phases (open yellow pentagon) are considered. Further multiphase studies of outflows in individual galaxies as well as better constraints on uncertain parameters such as n_e and α_CO will be crucial for gaining further insights into the primary mode of black hole growth and the degree of coupling between the AGN accretion energy and gas in the host galaxy.

K20-ID5 and COS4-11337 are particularly interesting systems because they provide insights into the role of AGN feedback in driving the evolution of compact star-forming galaxies (cSFGs). cSFGs lie on or above the SFR main sequence but have sizes more similar to those of compact quiescent galaxies. There is growing evidence that the build-up of large stellar mass surface densities is closely linked to the quenching of star formation (e.g., Martig et al. 2009; Bluck et al. 2014; Lang et al. 2014), and cSFGs may represent an intermediate population of galaxies that have already undergone morphological transformation but have not yet ceased forming stars (e.g., Barro et al. 2013; Nelson et al. 2014; Williams et al. 2014). Most cSFGs appear to have lower gas fractions and shorter depletion times than normal star-forming galaxies, suggesting that they will indeed quench on relatively short timescales (e.g., Barro et al. 2016; Spilker et al. 2016; Popping et al. 2017; Tadaki et al. 2017; Talia et al. 2018).

Compaction occurs when a large amount of gas is funneled toward the center of a galaxy—for example, as a result of disk instabilities (e.g., Bournaud et al. 2007; Dekel & Burkert 2014; Brennan et al. 2015) or galaxy–galaxy interactions (e.g., Hopkins & Elvis 2010). The presence of a large nuclear gas reservoir can trigger star formation and/or AGN activity, which can subsequently drive outflows. cSFGs exhibit a higher incidence of AGN activity than normal star-forming galaxies at fixed stellar mass (Kocevski et al. 2017) and are also expected to have a high incidence of AGN-driven outflows based on their large stellar masses and central stellar mass densities (e.g., Förster Schreiber et al. 2019). Therefore, it is important to consider the potential role of AGN feedback in quenching star formation in cSFGs.

K20-ID5 and COS4-11337 are both classified as "Strong Outflows" by Förster Schreiber et al. (2019) because an unusually large fraction of their [N ii]+Hα emission is associated with their ∼1500 km s⁻¹ AGN-driven outflows. This is exemplified in K20-ID5, for which our analysis suggests that almost all of the nuclear line emission is associated with the outflow. Strong Outflows are rare, occurring in ∼5% of massive (log(M_*/M_⊙) ≳ 10.7) galaxies and accounting for ∼10% of AGN-driven outflows. They have similar outflow velocities and global mass loading factors to normal AGN-driven outflows but have ∼ 2.5× higher mass outflow rates and are found in galaxies that are smaller and have higher SFRs and specific AGN luminosities (sometimes used as a proxy for Eddington ratio) compared to typical AGN host galaxies at the same redshift. Strong Outflows may therefore trace a "blowout" phase, which is also associated with strong SF and black hole accretion activity.

The impact of these extreme outflow phases on SF in the host galaxy is unclear. The subdominant contribution of SF to the Hα emission in the nuclear regions could indicate either that there is very little nuclear SF or that the nuclear SF is heavily obscured, which would be expected if the gas mass surface densities in the central regions are high. The strong outflows on average have similar global mass loading factors to normal AGN-driven outflows, suggesting that the SF activity in the host galaxy is relatively unaffected by the extreme nuclear blowout, at least in the early stages. However, the outflows carry significant amounts of kinetic energy into the circumgalactic medium, which may help to maintain the presence of a hot halo and therefore impede replenishment of the molecular gas reservoir (e.g., Bower et al. 2006; Croton et al. 2006; Bower et al. 2017; Pillepich et al. 2018).

The outflows from K20-ID5 and COS4-11337 decrease the already short molecular gas depletion times in these systems. K20-ID5 has a molecular gas mass of log(${M}_{{{\rm{H}}}_{2}}$/M_⊙) = 11.0 (calculated by re-scaling the CO-based gas mass from Popping et al. 2017 to the metallicity-dependent α_CO from Tacconi et al. 2018), corresponding to an SF depletion time of 280 Myr. If we assume that the molecular gas outflow rate is at least as large as the ionized gas outflow rate (see discussion in Section 7.3), the overall depletion time (including the contribution of the outflow) is $\leqslant$ 160 Myr, compared to an average depletion time of 520 Myr for galaxies at the same stellar mass, SFR, and redshift (Tacconi et al. 2018). There are no existing gas mass measurements for COS4-11337, but using the upper limit on the CO(4-3) flux (Figure 11), we find a 3σ upper limit on the gas mass of log(${M}_{{{\rm{H}}}_{2}}$/M_⊙) < 9.9 (assuming CO(1-0)/CO(4-3) = 2.4, and the metallicity-dependent α_CO from Tacconi et al. 2018). The gas mass upper limit corresponds to an SF depletion time of <18 Myr and an overall molecular gas depletion time of <16 Myr, compared to an average of 540 Myr for galaxies at the same stellar mass, SFR, and redshift. The molecular gas depletion time for COS4-11337 is very short and may indicate that the AGN radiation field is heating some of the molecular gas and causing it to emit primarily in higher excitation CO transitions (e.g., Gallerani et al. 2014; Mingozzi et al. 2018; Rosario et al. 2019).

Both K20-ID5 and COS4-11337 could deplete their entire molecular gas reservoir within a couple of hundred megayears, and the kinetic energy injected into the halos by the AGN-driven outflows could suppress the accretion of fresh molecular gas, supporting the notion that these galaxies may be the direct progenitors of compact quiescent systems. However, both galaxies are currently located on the upper envelope of the star-forming main sequence, suggesting that the outflows have not yet had any significant impact on the SF activity in their host galaxies.

J0901 at z = 2.2586 (Hainline et al. 2009) is lensed by a poor cluster at z ∼ 0.3459 (Diehl et al. 2009). Three images are apparent in Figure A1: the southeast (SE) arc that is the target of our high-resolution SINFONI observations, a relatively undistorted western (W) arc, and the northeastern (NE) arc, which is a fold arc that images part of the source twice but misses the H-band nucleus (see also Tagore 2014; Sharon et al. 2019). To construct a lens model for interpretation of the SINFONI adaptive optics and HST data, we rely on archival HST F814W (rest-frame UV) and F160W (rest-frame optical) images, and we use version 7.0 of the parametric gravitational lens modeling code lenstool (Kneib et al. 1996; Jullo et al. 2007; Jullo & Kneib 2009). All HST data have been astrometrically registered for consistency with each other and with five stars from GAIA DR1 (Gaia Collaboration et al. 2016).

Zoom In Zoom Out Reset image size

The model is constrained by the locations of 15 I-band (rest UV) emission knots (Figure A2), with positions that have been manually measured in the W arc, in the SE arc (one of them imaged three times within the SE arc), and for five of them in the NE fold arc (all of these imaged twice). In addition, we use the four HST H-band images of a reddish second lensed object with unknown redshift, nicknamed "Sith" by Tagore (2014). The redshift of this object is left free for the lens modeling and is estimated at $z\approx 3.1$ by the adopted lenstool model.

Zoom In Zoom Out Reset image size

The lens model includes a general cluster potential for which we adopt an NFW profile (Navarro et al. 1997). We fix the NFW concentration parameter at c = 6, reasonable for an ${M}_{200}\approx {10}^{14}$ M_⊙ cluster reported below (Dutton & Macciò 2014; Merten et al. 2015; Umetsu et al. 2016) and considering some bias toward more concentrated halos for strong lensing clusters. The position, axial ratio, position angle, and radius of the cluster potential are left free for the fit. We note that our data do not strongly test or constrain this adopted cluster radial profile shape, since all J0901 and Sith constraints are at similar cluster-centric radii.

Galfit (Peng et al. 2002, 2010) was applied to the HST I-band and H-band images in order to derive parameters of the foreground cluster members. For the two galaxies near the cluster center, we adopt dPIE (Elíasdóttir et al. 2007) profiles with fixed position and core radius but fit the axial ratio, position angle, velocity dispersion, and cut radius. Close proximity to the cluster center makes their individual parameters less constrained. For the perturber galaxy near the J0901 SE arc, we fix position and position angle to the I-band observed values, but we fit axial ratio, velocity dispersion, and cut radius. For 13 additional foreground cluster galaxies, we follow the common procedure (e.g., Limousin et al. 2007) of fixing position, position angle, and axial ratio individually to the I-band observed values but fitting $\propto {L}^{1/4}$ relations for velocity dispersion and cut radius, with L based on the H-band magnitude. Finally, we use the crossing of the NE fold arc by the critical line and the arc orientation at that point as constraints. Table A1 lists the resulting parameters.

The model fits the overall lens morphology. The rms difference in the image plane between observed position of a knot and position predicted from the knot's location in another arc plus lens model is 045. This reduces to 016 when comparing only SE and W arc, i.e., excluding the strongly distorted NE arc and Sith. The overall morphology and in particular the need to consider the perturber near the SE arc match the findings that Sharon et al. (2019) obtained from a lens model that is based on 11 resolution IRAM CO data. Since our focus is on interpretation of higher spatial resolution SINFONI adaptive optics data, we adopt the model built from high-resolution HST data. Figure A3 provides a sanity check of the lens model, by comparing the SE and W I-band arcs, projected back to the source plane and applying a small scaling/rotation correction to the W reconstruction. Before applying the lens model to SINFONI data, we shifted the astrometry of the SINFONI cubes to obtain a K-band continuum position consistent with the HST F160W image.

Zoom In Zoom Out Reset image size

We calculate the luminosity of the AGN in J0901 from the [O iii] luminosity. It is not possible to directly compute the source plane [O iii] luminosity because we only have long slit data. Instead, we assume that the [N ii] and [O iii] emission have similar spatial distributions, and we multiply the image plane [O iii] luminosity by the ratio of the [N ii] flux in the source plane nuclear spectrum to the [N ii] flux in the image plane long slit spectrum. We correct the [O iii] luminosity for extinction using the Balmer decrement (described in Section 6.3) and convert the [O iii] luminosity to the AGN bolometric luminosity using a bolometric correction factor of 600 (Netzer 2009), yielding an AGN luminosity of log(L_AGN) = 46.3.

Saintonge et al. (2013) estimated the stellar mass of J0901 by making use of empirical correlations between the 3.6 μm and 4.5 μm luminosities and the stellar mass. The observed luminosities were corrected for lensing magnification using a single wavelength-independent correction factor. We re-calculate the stellar mass and A_V for J0901 by utilizing archival HST imaging in the F475W (λ_rest = 1481 Å), F814W (λ_rest = 2500 Å), and F160W (λ_rest = 4959 Å) bands (Program ID 11602, PI: S. Allam). The imaging has a spatial resolution of 01–015, which is sufficient to apply the source plane reconstruction and calculate the magnification factor for each band individually. We calculate the source plane luminosities in these three bands and fit the SED using the FAST code (Kriek et al. 2009). We consider solar metallicity models from the Bruzual & Charlot (2003) library, with dust extinction following the Calzetti et al. (2000) curve and A_V = 0–3. We assume an exponentially declining star formation history (τ model), allowing for ages between 50 Myr and 2.86 Gyr (the age of the universe at z = 2.259), and log(τ/yr) = 8.5–10.0. The best-fit model gives log(M_*/M_⊙) = 11.2 and A_V = 1.2. Our stellar mass estimate is close to the value that is obtained by scaling the Saintonge et al. (2013) value to our Hα magnification factor (log(M_*/M_⊙) = 11.03 ± 0.14).

We calculate the SFR of J0901 from the 160 μm flux presented in Saintonge et al. (2013). We scale the total 160 μm flux by a factor of 0.4 to isolate the flux originating from the SE arc (Saintonge et al. 2013). We cannot measure the magnification factor in the FIR, because the spatial resolution of the imaging is insufficient to perform the source plane reconstruction. Instead, we adopt the Hα magnification factor (μ = 9.9), which provides a good first-order approximation of the lensing correction. We convert the 160 μm luminosity to ${L}_{\mathrm{IR}}$ using the SED template from Wuyts et al. (2008) and convert the ${L}_{\mathrm{IR}}$ to an SFR using SFR = 1.09 × 10⁻¹⁰ L_IR/L_⊙ (Wuyts et al. 2011b). The resulting SFR is 200 M_⊙ yr⁻¹.