A RUNAWAY BLACK HOLE IN COSMOS: GRAVITATIONAL WAVE OR SLINGSHOT RECOIL?

Figure 3 (left) shows a 9''× 9''zoom around source CID-42 in the ACS/HST image (003 pixel⁻¹; filter F814W). The distance measured between the two optical centers is 0495 ±  0005 (16.5 pixels) which corresponds to a projected separation of 2.46 ± 0.02 kpc at the source's redshift.

Zoom In Zoom Out Reset image size

We modeled the optical surface brightness of CID-42 with GALFIT (Peng et al. 2002; Version 3.0²³). We use an empirical point-spread function (PSF) created by averaging stars in the ACS frames (Jahnke et al. 2004, 2009b) to represent a point source in the field and to use for the convolution of the parametric galaxy models.

We tested several combinations and numbers of point sources and Sersic profiles (Sersic 1968) with a free Sersic-parameter n. We also made use of two new features in GALFIT V3.0, a Fourier mode modification of isophotes, restricted in this case to F1 (lopsidedness); and a mode to model the NE tidal arm. The goals of this exercise were to (1) test whether the two nuclei in the system are point sources, (2) extract reliable point source fluxes, and (3) put constraints on the relative fluxes of the galaxy components around the two sources.

The best description of the system consists of one point source and three Sersic components. The model and the residuals are shown in Figure 3 (middle and right). We find that, above the general background, only the southeastern (SE) nucleus is well modeled with a point source, while the northwestern (NW) nucleus is extended though compact. This finding is independent of the exact number and geometry of large-scale galaxy components added to the model. By this model, we estimate the point-like SE nucleus to have F814W = 20.51 mag (square in Figure 2, left) with an uncertainty of ∼0.1 mag. The first Sersic component corresponds to the NW nucleus and is very compact, with a scale length of ∼01 (∼0.5 kpc), smaller than the typical dimension of host galaxies of X-ray selected AGNs (e.g., Schade et al. 2000), and F814W = 19.67 mag (triangle in Figure 2, left). The second Sersic component is a large spheroid (r_eff = 3'', ∼15 kpc) and the third Sersic component describes part of the overall light distribution plus the spiral arm. The latter two components combined have F814W = 18.6 mag and together describe the host galaxy light distribution, likely extending the profile of the NW source. The overall system brightness obtained with the fit is F814W = 18.11 mag (pentagon in Figure 2, left), in agreement with the photometry used for the SED.

In order to test the goodness of the fit, we ran the GALFIT analysis with a simple model (one point-like source and one Sersic component), as was reported by Gabor et al. (2009). This fitting runs into calculation boundaries, indicating that a two-component model is not adequate to represent this complex system. A minimum of three components (one PSF and two Sersic) are required; the best fit is obtained using four components, as explained above.

Given the point-like unresolved profile of the SE source, it could be considered as an unobscured BH (i.e., Type 1 AGN) and it is possible to compute its bolometric luminosity by normalizing to its F814W magnitude a typical Type 1 AGN SED (Elvis et al. 1994; red line in Figure 2, left). This gives a bolometric luminosity for the SE source of L_{24 μm−UV} = 2.8 × 10⁴⁴ erg s⁻¹, consistent with being a moderately luminous AGN.

The merging system CID-42 is not an isolated source, as is evident in Figure 1. Using combined XMM-Newton and Chandra data (with prior point source removal), extended emission has been detected around CID-42. The X-ray detected group (Finoguenov et al. 2007; A. Finoguenov et al. 2010, in preparation) is spectroscopically confirmed with 7 galaxy members in the zCOSMOS bright catalog (Lilly et al. 2007, 2009) and 30 members in the photometric redshift catalog (Ilbert et al. 2009), of which 10 are visible in Figure 1. The extended X-ray luminosity in the 0.1–2.4 keV band is 1.4 × 10⁴² erg s⁻¹. The X-ray centroid, though uncertain (30''positional error), is consistent with the position of CID-42. The detection of extended X-ray emission from the hot gas of the group allows us to estimate the mass of the halo in which the merging galaxy resides. The total gravitational mass of the group is M₂₀₀ = (2.5 ± 0.7) × 10¹³ M_☉,²⁴ based on the average L_x–M₂₀₀ calibration of Leauthaud et al. (2010). The corresponding R₂₀₀ is 106''(∼0.5 Mpc), so CID-42 is located in the core of the group. The source CID-42 has also been identified as the second most massive group galaxy (MMGG; details in Leauthaud et al. 2010). Furthermore, the galaxy group hosting CID-42 is embedded in a large over-dense region at z ∼ 0.37 covering the whole COSMOS field (Scoville et al. 2007b, Gilli et al. 2009).

CID-42 was imaged with Chandra in six ACIS-I pointings with different exposure times (14–46 ks) and off-axis angles (3'–75) spread over four months. Given the Chandra ACIS-I resolution (05 pixel⁻¹), it is not possible to perform an accurate decomposition analysis as performed with HST/ACS data, but it is possible to study the emission in different energy bands.

We analyzed the X-ray emission in both soft (0.5–2 keV) and hard (2–7 keV) band images, using the observation in which the source has been observed at the smallest off-axis angle (∼3') and with the longest exposure time (46 ks). In this observation, the source has 187 net counts in the full band in a region corresponding to 95% of the encircled energy fraction for a point source at that position.

Most of the counts (123 out of 187) are detected in the 0.5–2 keV band. Comparing to the PSF that applies to this energy band, off-axis angle, and azimuth, there is some evidence for extension along the axis joining the two optical nuclei. To characterize this better, we did a detailed ray-trace simulation using ChaRT²⁵ and MARX²⁶ with the assumption of equal flux from the two optical nuclei. The results are illustrated in Figure 4 where we show the Chandra soft band X-ray image (top right) and the simulation (lower right). The observed and simulated images are qualitatively consistent, indicating that there is no evidence for galaxy-scale X-ray diffuse emission in this source. One would like to constrain usefully the flux ratio of the two point sources, but this is not possible here given the PSF extent and astrometric uncertainty at 3'off-axis. For completeness, we also simulated the 2–7 keV hard band data as coming from the two nuclear point sources and also in this case we found that the observed image (Figure 4, lower left) is consistent with the simulated image.

Zoom In Zoom Out Reset image size

Optical spectra are available from SDSS (Adelman-McCarthy et al. 2008), IMACS/Magellan (Trump et al. 2007, 2009a), and VIMOS/VLT (zCOSMOS; Lilly et al. 2007, 2009). The source has been observed on 2005 February with IMACS and then twice with VIMOS on 2005 May and June. Given the 1''width of the slits (oriented N–S) both in the VIMOS and IMACS observations, and the small E–W separation of the sources (∼03), the flux from both falls within the slit. The two-dimensional (2D) spectra have been analyzed, but none of them spatially resolve the two nuclei, while in Comerford et al. (2009) the two nuclei seem to be resolved in a 2D DEIMOS spectrum; the spatial resolution of the DEIMOS spectrum is similar to the IMACS one (IMACS: 1 pixel = 0.11'', DEIMOS: 1 pixel = 0.11'', VIMOS: 1 pixel = 0205) but its spectral resolution (R ∼ 3000) is higher. The two zCOSMOS spectra (5500–9500 Å) have an average S/N of 6 (11 around the Hβ line; Figure 5) and a spectral resolution of R ∼ 600. The IMACS spectrum (5600–9200 Å, R ∼ 700) has a detector chip gap at 6450–6560 Å, cutting off part of the Hβ line. The Hα emission line is in a region where fringing becomes strong in VIMOS. Although Hα is visible in the IMACS spectrum, which uses the nod and shuffle techniques to avoid fringing, the peak of the line is saturated. The SDSS spectrum (4000–9000 Å) has an S/N of ∼6 around the interesting emission lines, lower than the Very Large Telescope (VLT) and Magellan spectra. The SDSS spectrum uses a 3'' fiber instead of a 1'' wide slit, and so the spectrum contains more extended emission diluting the AGN continuum.

Zoom In Zoom Out Reset image size

For these reasons, we started the spectral analysis by fitting the blue part of the zCOSMOS spectrum around the Hβ and the [O iii] doublet. The fitting was performed using the IRAF routine SPECFIT (Kriss 1994). We modeled the spectrum with a linear component for the continuum plus four Gaussian lines, one broad and three narrow: the [O iii] doublets, and the broad and narrow Hβ. The central wavelength, width, and intensity of the Gaussians were left free to vary. The redshift measured from the peak of the narrow emission lines ([O iii] and Hβ) is z = 0.359, coincident with the redshift listed in Lilly et al. (2007, 2009) and Comerford et al. (2009). This redshift is also consistent with the one measured from absorption lines (e.g., Hγ). The FWHM of the narrow emission lines, corrected for instrumental broadening, is 560 ± 120 km s⁻¹.

The three optical spectra show a typical Type 1.5 (Osterbrock & Ferland 2006) AGN feature: a broad Hβ component plus a narrow core that has a line width similar to the forbidden [O iii] lines. Unusually though, the broad and narrow Hβ components of CID-42 have widely offset peaks, potentially implying that the emission comes from different regions, consistent with the presence of two unresolved sources. The same offset is measured in both the zCOSMOS spectra and so, in order to have a better measurement of this offset, we summed them. We measure an offset Δλ∼ 30 ± 7 Å (1360 ± 320 km s⁻¹) between the centroids of the broad and narrow Hβ. Small residuals at the peak of the broad line possibly suggest the presence of another narrow component at the same wavelength that, when fitted, unfortunately has a flux below the S/N of the spectrum.

If the offset between the two components is due to the narrow lines being produced by one of the two optical sources and the broad line by the other source, then an instrumental shift due to the spatial separation and the orientation of the sources with respect to the E–W VIMOS dispersion axis has to be considered. This effect shifts the broad component toward the red by 4 Å, which should be subtracted from the measured offset. The net offset (26 ± 7 Å) implies a difference in velocity of ∼1180 ± 320 km s⁻¹ between the Hβ narrow and broad components under this assumption.

The 2D IMACS spectrum (see Figure 6) shows a clear shift of the peak of the broad line with respect to the narrow line, but does not show the spatial separation seen in Comerford et al. (2009), despite the same pixel scale of the two instruments (IMACS and DEIMOS); this is most likely due to different seeing conditions.

Zoom In Zoom Out Reset image size

The presence of a detector chip gap close to the Hβ region in the IMACS spectrum prevents a good fitting of the line profile. However, a narrow line is present with a broad peak redshifted from the narrow core. Using the shift measured in the zCOSMOS spectrum, we predicted the wavelength where the broad line in IMACS should peak. The instrumental shift due to position and slit orientation in the IMACS spectrum would move the broad component toward the blue (i.e., in the direction opposite to the zCOSMOS spectrum shift) because of the opposite dispersion direction of the two instruments, so that the two peaks have a different observed offset with respect to zCOSMOS. The predicted offset in IMACS is ∼20 Å, which matches with the observed line peak (red dashed line in Figure 5, top). If the lines are not coming from two different sources, the predicted offset in IMACS would not correspond with the observed line peak, although it could be consistent within the errors.

We analyzed the SDSS spectrum to check for the presence of the offsetted Hβ line, but, given the lower S/N and the higher flux of the continuum, the broad component of the Hβ line is barely above the noise in the spectrum and so we do not use it for our analysis.

We performed further checks to test the possible presence of a second system of lines with the same offset of the broad Hβ. The first would be the detection of a second [O iii] doublet with the same shift measured in the Hβ but, at the expected wavelength, a Telluric feature (∼6825 Å) prevents the detection of the corresponding redshifted [O iii] lines. The second test would be the detection of a redshifted broad Hα line. As explained above, Hα lies in a region dominated by fringing in the zCOSMOS spectrum, while in IMACS the line peak is saturated. The fit of the Hα and [N ii] doublet in the IMACS spectrum returns three narrow lines at the same redshift of the narrow [O iii] doublet (Figure 5, bottom). The broad Hα component peak is slightly offset (∼5 Å) to the red from the narrow one but this shift is not enough to be consistent with the shift predicted from Hβ, even when the instrumental shift is taken into account. Although proper fitting of the continuum has been performed, a 1:2 ratio for the [N ii] doublet flux is observed instead of the 1:3 ratio required by atomic physics. On the contrary, the expected 1:3 ratio has been measured in the [O iii] doublet flux in all spectra. This suggests that, around Hα, the IMACS spectrum is not well behaved and so the measurement of the broad-line peak is still uncertain.

We also performed a simple two-component fit to the IMACS spectrum to determine the relative nuclear and host contributions in the 4000–9000 Å range. Following Trump et al. (2009b), we find the combination of quasar (Vanden Berk et al. 2001) and galaxy (Eisenstein et al. 2001) that maximizes the Bayesian probability function and best describes the spectrum. The estimated nuclear-to-host ratio is ∼3:2. Systematic errors, arising from neglecting the contribution from other components, could affect this estimated ratio. However, this simple fit shows that the spectrum is not consistent with a simple blue quasar but instead includes a significant host contribution (consistent with the GALFIT decomposition of Section 3.1). The significant host contribution may also be part of the reason for which the fitting around the Hα region returns different results from those obtained for Hβ and [O iii].

The X-ray data available (∼800 counts in Chandra ACIS and ∼2200 counts in the XMM-Newton pn) allow us to perform good quality X-ray spectral analysis when XMM-Newton pn and Chandra ACIS spectra are fitted together. The results are reported in Table 1. Due to the modest number of counts, we used the Cash statistic (Cash 1979). A fit with an absorbed power law returns a spectral index Γ = 2.11 ± 0.06, with no cold intrinsic absorption, in good agreement with the results of Mainieri et al. (2007; Γ = 2.18 ± 0.07). Residuals in the very soft energy band (<0.5 keV, mostly in the XMM-Newton spectrum) suggest the presence of a thermal component (kT < 0.13 keV) that accounts for 8%–9% of the 0.3–2 keV luminosity, which could be due to either a nuclear soft-excess or hot gas in the galaxy. The addition of the soft component to the fit produces a slightly flatter spectral slope with Γ = 1.95^+0.07_−0.06.

The most striking features of the X-ray spectrum are around 4.5 keV (observed frame; Figure 7, left). The residuals suggest the presence of both emission and absorption features in both spectra, forming a kind of inverted P-Cygni profile, i.e., the absorption component is redshifted with respect to the emission component (Figure 7, right).

Zoom In Zoom Out Reset image size

The emission feature, which is consistent with being a neutral iron line, has a constant flux and is more prominent in the Chandra spectrum (EW = 570 ± 260 eV, in the rest frame), where the continuum is fainter, than in the XMM-Newton one (EW = 142⁺¹⁴³₋₈₆ eV, in the rest frame), where the continuum is 2 times higher.

The absorption feature instead shows strong variability between XMM-Newton and Chandra data, both in flux and width: the line is well fitted with a broad (σ = 0.22^+0.33_−0.11 keV, 1σ error) Gaussian line in the XMM-Newton spectrum, where it is more prominent, while it is unresolved in the Chandra data (σ < 0.5 keV). Assuming z = 0.359, the central rest-frame energy of the absorption line is around 6 keV, and therefore it seems to be a redshifted absorption iron line.

The shape and intensity of the background spectrum have been studied in order to discard the possibility for the absorption line to be a background artifact. The fact that the absorption line has been detected in spectra from two different instruments (EPIC and ACIS), already rejects this hypothesis.

In order to check if the inverted P-Cygni feature can be mimicked by other spectral models, we also performed the spectral fitting using more complex models—a broken power law and a partial covering model. The two models reproduce the continuum better than a single power law, given the larger number of parameters, but none of them reproduces the absorption and emission features. In particular, the edge around 5–6 keV in the partial covering model does not reproduce the absorption feature. The ΔCash statistic obtained by adding the two Gaussian lines to the models is the same as that obtained with a single power law, meaning that the absorption feature is real and independent of the model adopted to reproduce the continuum. We note that the partial covering model returns N_H∼10²³ cm⁻² and a covering fraction of 40%, which can be translated as the sum of two power laws of almost the same intensity, one obscured and the other not.

A spectral fitting with a photoionized gas model to the total XMM-Newton spectrum has also been performed. Unfortunately, given the low CCD resolution, it is not possible to distinguish single absorption lines in the soft band. The equivalent width of the absorption feature with a photoionized-gas model implies a high column density of the absorber (N_H∼ 5 × 10²³ cm⁻²); and the modest absorption measured in the soft band (N_H< 0.9 × 10²⁰ cm⁻² in the rest frame) requires the absorber to be highly ionized (log ξ∼3). From the centroid of the line, we measure the velocity of the outflow of 30,000 ± 4000 km s⁻¹.

In order to check the nature of the broadening of the absorption line, we analyzed the spectra individually. Given that the XMM-Newton data come from eight observations taken over three years, variability may be responsible for the broadening. Nine (five XMM-Newton and four Chandra) out of the fourteen available observations have enough counts (>125) to perform an individual spectral fit. The absorption feature is present at 2σ–3σ in all the spectra, apart from one Chandra observation where, although the number of counts is enough for fitting (187 net counts), the continuum flux is faint and thus the absorption line is not detected. The intensity (with 1σ error), width, and rest-frame energy centroid of the line measured in the eight spectra are plotted in Figure 8 (red = XMM-Newton, black = Chandra). In contrast to the summed spectrum, the absorption features show a narrow profile, with only upper limits on the width in all the spectra (σ< 250 eV) and the EW is consistent with being constant, implying a column density of gas of ∼5 × 10²³ cm⁻² (Bianchi et al. 2005).

Zoom In Zoom Out Reset image size

The line energy centroid in the XMM-Newton spectra changes in time over the range 5.9–6.3 keV, while in the Chandra spectra it is almost the same, within the errors. This variation explains the broadening measured in the total XMM-Newton spectrum as due to the superposition of narrow lines with different energy centroids. The variability of the energy centroid is of ΔE = +500 eV in the first two years and ΔE = −400 eV in the last year. These shifts are comparable to the 400 eV separation of the Fe i and the Fe xxvi Kα absorption lines, so a change in ionization could, in principle, be responsible for the observed shifts. We analyzed the hardness ratio in all the spectra to check for column density variability to be related with variability of the ionization state, but, given the large error bars, we do not detect statistically relevant variability.

Instead, if this variation is converted in a pure velocity variation, it implies a range in velocity of 0.02c–0.07c for neutral iron (Fe i), with a deceleration of 10,000 km s⁻¹ yr⁻¹ in the first two years and a faster acceleration in the last year of observations. If the lines are due to completely ionized iron (Fe xxvi) absorption, the velocity is higher (0.09–0.14c) and the implied acceleration is slightly lower.

Because of the degeneracy between velocity and ionization state, the same conclusions on the absorber parameters (density, ionization state, and velocity) are obtained doing the single-line analysis and the fitting with a photoionized model.

The spatial offset of the SE point-like source with respect to the NW source and the kinematic offset of the broad Hβ line with respect to the narrow line system can be explained by an SMBH ejected by gravitational radiation produced in the BH merger.

According to theoretical models, during the final coalescence of two BHs in a merger event, the GWs produced can impart a large kick to the resultant BH, which recoils in the opposite direction with respect to the center of the galaxy (Peres 1962). Although some approximate methods indicated that the kick velocities are most likely small (<few × 100 km s⁻¹; Blanchet et al. 2005; Damour & Gopakumar 2006), the results from full numerical relativity simulations show that the recoil speed can be up to 4000 km s⁻¹ (Campanelli et al. 2007a, 2007b).

It is difficult to estimate how often significant recoil kicks occur because the recoil kick distribution depends on the spin and mass ratio distributions of the progenitor BH binaries, which are not well known. Several authors (Schnittman & Buonanno 2007; Campanelli et al. 2007a, 2007b; and Baker et al. 2008) have estimated, using results from full numerical relativity simulations, that 12%–36% of recoiling BHs will have velocities >500 km s⁻¹ and 3%–13% will have velocities >1000 km s⁻¹, assuming a spin of 0.9 with random orientations and using different assumptions on how the kick velocity scales with BH progenitor mass ratio (0.1 < q< 1). These fractions might be smaller if BH spins are on average lower or if gas torques are efficient at aligning the spin axes in a way that is unfavorable for large kicks (Bogdanović et al. 2007; Dotti et al. 2010). To constrain the actual recoil kick distribution, better observational constraints on merging BHs or observations of individual recoiling BHs are required. A candidate recoiling SMBH with a large kick velocity of 2650 km s⁻¹ has been spectroscopically discovered by Komossa et al. (2008), although alternative models have been proposed to explain the two systems of lines in the optical spectrum. A second possible candidate has also been proposed by Shields et al. (2009).

The high-velocity offset (v > 1000 km s⁻¹) measured in the optical spectra of CID-42 is consistent with the numbers obtained in simulations, although the probability of finding such high-velocity recoiling BH is low (<10%).

The surface brightness decomposition of the CID-42 ACS image is consistent with the ejected BH hypothesis, too: the SE point-like source, the ejected BH, is spatially separated from the compact, but not point-like, NW bright source, which can easily represent the galaxy core from which the BH has been ejected.

The fact that the broad lines are redshifted with respect to the stellar absorption lines is consistent with the recoiling BH picture because they are following the BH, as explained at the beginning of Section 5, while the NLRs are left at the center of the NW source. The narrow lines have the same redshift of the stellar absorption lines but, given that the BH is more than 2.5 kpc away, the NLRs are no longer powered by it. Another way to produce the narrow lines is needed. The line flux ratios indicate that the emission is produced by nuclear activity and not by star formation (Comerford et al. 2009). The narrow lines could be produced by the ejected BH, ionizing the local galaxy ISM on its way out from the center. In this case, the narrow lines should have the same redshift of the galaxy, as observed, implying that the BH is still inside a region of the galaxy disk with high ISM density or is passing by molecular clouds. In this case, the velocity offset measured in the optical spectrum, not corrected for instrumental shift (see Section 4.1), is 1360 ± 320 km s⁻¹, which represents the radial component of the total velocity.

Lousto et al. (2010) showed, in a statistical study of BH binaries during the inspiral and merger, that when the kick velocity is ≳1000 km s⁻¹ the direction of the kick is most likely perpendicular to the BH orbital plane. They found a probability of 6.3% for large recoil velocities (>1000 km s⁻¹) along the line of sight. If the accretion disk is aligned with the nearly face-on galaxy disk as a whole, then a large kick closely aligned with the line of sight is expected, and it is consistent with that observed in CID-42, if the source is a recoiling BH.

The time since the BH has been kicked (∼4 × 10⁶ yr), as computed at the beginning of Section 5, is in agreement with the prediction of the effective time for the persistence of off-center quasar activity (10^6.2–10⁷ yr; Blecha & Loeb 2008). This number is a lower limit but it should be a good estimate, assuming that the total kick velocity is not much higher than the radial velocity, given that this is already in the extreme tail of the kick velocity distribution predicted by models and simulations. According to the Blecha & Loeb (2008) model, for a kick velocity in the range 1100–1500 km s⁻¹ and a BH mass of 10⁷–10⁸ M_☉, the mass of the BH disc, in order to be still emitting, is ∼2%–3% of the BH mass.

Models predict that the effect on galaxy morphology of BH recoil is to produce significant asymmetry (Kornreich & Lovelace 2008). This is seen in the morphology of CID-42 where the galaxy core (the NW source) is slightly misplaced with respect to the geometric center of the galaxy disc.

The variable Fe–K absorption line can be alternatively explained, as already explained at the beginning of this section, by material inflowing into the BH, as it has been proposed in a few other sources (Cappi 2006; see Section 5). In order to have such a high velocity, the inflow of material would have to be very close to the nucleus (few tens of Schwarzschild radii). One way of achieving high inflow velocity is with a photon bubble instability (Arons 1992; Gammie 1998; Begelman 2006) which produces a propagating pattern of low- and high-density regions through which the radiation has to go. The photon bubble instability is currently used to explain the inverted P-Cygni profiles detected in the spectra of luminous blue variable (LBV) stars (e.g., van Marle et al. 2008). Although this would be a physical explanation, it requires super-Eddington accretion, which is not observed in CID-42.

Although the high-velocity inflow possibility is very interesting, without detailed modeling it remains a mere hypothesis.

In the following, we list the evidences in favor of a picture in which CID-42 contains two AGNs, an unobscured Type 1 AGN in the SE source, and an obscured Type 2 AGN in the NW one.

• 1.  
  The surface brightness decomposition of the ACS image finds a compact, but not point-like, bright source in the NW nucleus.
• 2.  
  The shift (Δv∼ 1180 ± 360 km s⁻¹) between the broad and narrow components of the Hβ line in the optical spectrum also suggests the presence of two active nuclei, one responsible for the broad-line emission (i.e., a Type 1 AGN), the other responsible for the narrow emission lines only (i.e., a Type 2 AGN). A possible explanation for the presence of only one narrow emission-line system could be that the NLRs of the two sources are mixed up and, for this reason, are at the same redshift measured for the galaxy from the absorption lines. However, we discard this hypothesis as NLRs, at the luminosity of CID-42, are expected to extend 0.1–1 kpc from the nucleus (Schmitt et al. 2003), much less than the projected separation measured between CID-42 nuclei.
• 3.  
  The X-ray flux variability and the presence of a strong iron emission line (EW ∼ 500 eV) when the continuum is weak also suggest a combination of a variable Type 1 AGN and a constant obscured Type 2 AGN. Usually, strong iron emission lines (EW ∼ 1 keV) are associated with heavily obscured (Compton thick) sources and are less prominent in the unobscured ones (EW ∼ 150 eV; e.g., see Figure 9 of Guainazzi et al. 2005a). As already noted by Guainazzi et al. (2005b), all the AGN pairs observed so far are found to be heavily obscured or Compton thick in the X-rays, in both components. The presence of at least one obscured AGN in our system is consistent with this picture but the possibility that we have a Type 2 plus a Type 1 AGN embedded in the same galaxy is new, although we believe CID-42 is not a binary.

Assuming that the [O iii] emission lines are produced by the Type 2 AGN only, it is possible to roughly estimate the contribution of the two sources to the total X-ray luminosity. The Type 2 luminosity is estimated from the relation between the L_2–10 keV–L$_{[\rm O\,\mathsc {iii}]}$ measured in the COSMOS field (Silverman et al. 2009) and the Type 1 AGN luminosity is then found by subtraction. Given the [O iii] luminosity (∼3 × 10⁴¹ erg s⁻¹) and the uncertainties on this relation, the luminosity of the NW Type 2 nucleus would be ∼9 × 10⁴² erg s⁻¹ in the 2–10 keV band (not absorption corrected). The derived Type 1 AGN luminosity (2–10 keV) is in the range (0.3–1.8) × 10⁴³ erg s⁻¹ due to the strong measured variability (Figure 2, right). The last number is consistent with the 2–10 keV luminosity obtained in the previous section with the Eddington ratio.

As explained at the beginning of Section 5, the large velocity shift between the two nuclei implied by the optical spectra would require a large mass of the system, assuming it is virialized (i.e., the two nuclei are gravitationally bound in a circular orbit). Comerford et al. (2009) detected two sets of [O iii] and Hα lines in a Keck/DEIMOS spectrum,²⁷ from which a Δv = 150 km s⁻¹ is measured by detecting two sets of emission lines. This lower velocity shift allows a more "reasonable" mass for a virialized system; however, virialization, which is a fair assumption for close pairs, is probably not appropriate for the kiloparsec-scale separation of the two CID-42 sources. Most likely, the velocity measured by Comerford et al. (2009) is due to the rotation of the gas in the galaxy as a whole, as the system is in a violent state of gravitational interaction.

The large velocity offset between the two sources is more likely due to a nearly radial orbit, with the Type 1 SE AGN recoiling with respect to the Type 2 AGN. A velocity difference of this size may result from a slingshot ejection in a triple BH encounter (Saslaw et al. 1974; Hoffman & Loeb 2006, 2007). If a BH binary stalls for a ∼few Gyr before it coalesces, due to the depletion of stars in the loss cone in a gas-poor merger (Merritt & Milosavljević 2005), its host galaxy could merge with another galaxy and a third BH may spiral in and undergo a strong three-body interaction with the binary. The system that just arrived will merge with the original galaxy. The SFR (∼100 M_☉ yr⁻¹) measured from the SED fitting supports the scenario of a merger in an early phase. The merger will drive mass and gas to the binary, refilling its loss cone. The configuration will become gravitationally unstable and this event will end with the further reduction of the separation or the coalescence of the binary and the ejection of the lightest body at a speed comparable to the binary's orbital speed (Heggie 1975).

The large projected distance (2.4 kpc) between the two sources and the high velocity measured in CID-42 are possible for this kind of ejection (Hoffman & Loeb 2006, 2007), although the probability to have a certain velocity at a certain separation depends from the initial condition of the system. Given that all massive galaxies have possibly experienced a merger event in the last ten billion years, assuming uncorrelated events, and a typical binary lifetime of one billion years, then 10% of SMBH binaries may form a triplet. However, even though the triple interaction would possibly lead to an ejection of one or even all SMBHs (Valtonen et al. 1994), most of the triple systems live long (∼10⁹ yr; Hoffman & Loeb 2007), and final coalescence is more common (85% versus 15%) than ejection.

5.2.1. Backlit Wind Model

On the basis of the above analysis, the SE nucleus is the unobscured Type 1 recoiling AGN, while the NW is an obscured Type 2 AGN resulting from the coalescence of a binary. The two nuclei lie close to the plane of the sky possibly moving in a radial orbit in a direction close to the line of sight, though a small angle has to be considered given the 2.4 kpc separation. Relative to the mean velocity of the system, the SE nucleus is moving away from the observer and the NW one toward the observer. The suggested geometry of the system is shown in Figure 9.

Zoom In Zoom Out Reset image size

The SE Type 1 nucleus then has (1) a strong unobscured optical continuum, leading to the point-like HST/ACS image; (2) the broad Hβ emission line; and (3) a strong, variable, X-ray continuum emission. The obscured NW Type 2 nucleus has (1) a weak optical continuum, leading to a compact but extended HST/ACS source dominated by the galaxy light; (2) only narrow emission lines of Hβ and [O iii]; and (3) a weak X-ray continuum, with a strong fluorescent iron emission line (EW ⩾ 570 eV).

However, the high-velocity infall scenario could still be a possible interpretation of the Fe–K absorption line (see Section 5.1); the proximity of the two sources allows special interpretation as a "Backlit Wind." In this scenario, a fast, highly ionized BAL-like wind is emitted by the NW Type 2 nucleus (located in front). In almost all circumstances, this wind would be undetectable. However, in the case of CID-42, the receding wind crosses our line of sight to the SE Type 1 nucleus, leading to absorption of the Type 1 continuum in redshifted iron. The blueshifted emission of the approaching wind remains invisible as there is no background source to illuminate it. The column density of the flow derived from the EW of the absorption line is consistent with what is seen in BAL wind (>10²³ cm⁻²; Green et al. 2001; Gallagher et al. 2002; Chartas et al. 2009). The fact that we measure no X-ray photoelectric absorption at low energies implies that the wind is highly ionized, something also observed in other BAL (e.g., Telfer et al. 1998; Hamann 1998; Chartas et al. 2009) and X-ray absorbers (e.g., Vignali et al. 2000; Reeves et al. 2009).

The variations of the iron absorption line energy centroid on a timescale of years can be explained either by velocity variations (∼10,000 km s⁻¹ per year) in the flow, or by a change in the ionization state, and so in the density (Proga et al. 2000), or, possibly, to a precession of the flow, so that a change in inclination can produce a difference in velocity, or by a combination of all these effects. BALs, when we look directly along the flow, are known to vary on year timescales (e.g., Lundgren et al. 2007; Gibson et al. 2008), so variability should be more likely in a section viewed across the flow. The variability implies that the BAL-like flow must be highly structured, even 2.5 kpc away from the nucleus. This is not unusual as it has been shown by Chandra detection of radio jets still relativistic at megaparsec distance from their origin (e.g., PKS 0637-752; Schwartz et al. 2000).

As we have only one example of a redshifted iron absorber and only one recoiling BH, this scenario is only plausible if the BAL-like wind has a large opening angle, making an interception of the rear recoiling Type 1 nucleus likely.

Winds at a few 1000 km s⁻¹ are normal in Type 1 AGNs (Reynolds 1997; Piconcelli et al. 2005) and much faster (0.1c–0.2c) BAL winds, confined to modest solid angles, are not unusual (Weymann et al. 1991; Hamann et al. 1993; Ogle et al. 2000). Fast winds at kiloparsec-scale distances from Type 1 AGNs have been found as well (Arav et al. 2001) and have been proposed to be universal in AGNs (Elvis 2000). However, there has been no possibility to test this idea in Type 2 AGNs. CID-42 provides this opportunity and gives us the first detection of a wind in a Type 2 AGN.