EVIDENCE FOR THREE ACCRETING BLACK HOLES IN A GALAXY AT z ∼ 1.35: A SNAPSHOT OF RECENTLY FORMED BLACK HOLE SEEDS?^*

We discovered the galaxy in the Chandra Deep Field-North (CDFN) while examining the WFC3/IR grism data of Chandra X-ray sources. Barger et al. (2008) report a spectroscopic redshift of z = 1.3550 (ID 1163), with ground-based coordinates of (J2000) 12:36:52.77 + 62:13:54.8. It is listed as an individually detected X-ray source (ID 272) in the 2 Ms catalogue of Alexander et al. (2003), who report an X-ray flux of $F_{0.5-8\, \rm {keV}} = 1.3\times 10^{-16}\ \rm erg\ s^{-1}\ cm^{-2}$ and a hardness ratio of HR = (H − S)/(H + S) ∼ −0.03. At the redshift of the source, this flux corresponds to a modest observed luminosity of $L_{0.5-8\, \rm keV} = 1.4\times 10^{42} \, \rm erg\ s^{-1}$ and an obscuring column density of N_H ∼ 5 × 10²² cm⁻² (Treister et al. 2009), which corresponds to a obscuration conversion of 1.5 for the observed 0.5–8 keV band (rest-frame 1.2–19 keV) for a typical AGN spectrum. The (optical) galaxy is larger than the Chandra point-spread function but the low count number does not allow us to associate counts to individual components. The HST optical image is shown in Figure 1 and the infrared image and grism data are shown in Figure 2.

Zoom In Zoom Out Reset image size

We retrieved the WFC3 imaging and grism data of the CDFN field⁷ from the archive and processed them with standard software. The detection F140W image was reduced using the STSDAS pyraf task multidrizle (Koekemoer et al. 2002). The dispersed (G141 grism) data were reduced using the aXe slitless spectroscopy package (Kümmel et al. 2009).

In Figure 2, we show both the undispersed F140W (H-band) image and the dispersed G141 grism image. The source shows two complexes of two components each, which we label A, B, C, and D. The orientation angle of the dispersed observation means that while two of these components, A and D, are cleanly dispersed, two others, B and C, contaminate each other. Fortunately, source B is significantly brighter than C, so the dispersed data are dominated by component B. We extract three different spectra from the dispersed image containing source A, B(+C) and D, respectively, and analyze them separately (Figure 3).

Zoom In Zoom Out Reset image size

For each of the three extracted spectra, we measure the [O iii] λ5007 and Hβ line strengths, as the [O iii]/Hβ ratio is a diagnostic of the dominant source of ionization (AGN versus star formation; Baldwin et al. 1981; Kewley et al. 2001). At the resolution of the WFC3 G141 grism, the Hα line is completely blended with the [N ii] λ6548 and λ6583 doublet surrounding Hα. For AGN photoionization, the Hα and [N ii] doublet become comparable in strength, so we cannot use the blended line as a diagnostic as the [N ii]/Hα ratio. We account for the blending of [O iii] λ4959 and λ5007.

We therefore measure only the [O iii]/Hβ ratio for each of the three components by fitting Gaussians using the mpfit algorithm (Markwardt 2009) from the extracted grism spectra in Figure 3. All three spectra show a strong [O iii] λ5007 line and a weak Hβ line yielding line ratios of log₁₀([O iii]/Hβ)∼0.4–0.5 which are commonly associated with photoionization by AGNs. In two of the three components (A and B), the [S ii] λλ6717, 6731 line is clearly visible, supporting the identification of these components as AGNs and ruling out the low-metallicity starburst scenario. The spectrum of component D is too noisy to reliably detect the [S ii] line but there appears to be a marginal detection. We report these results in Table 1.

Theoretical models could produce such a high [O iii]/Hβ ratio by an extreme combination of significantly sub-solar metallicity and very high ionization parameters (e.g., Kewley et al. 2001) but this alternative is not consistent with the X-ray data. Specifically, if we attribute all the flux in the blended Hα+[N ii] λ6584 region to a single Hα line to star formation, the implied maximum star formation rates are 11.0, 6.9 and 2.1 M_☉ yr⁻¹ for components A, B, and D, respectively (Kennicutt 1998), while the star formation rate implied by the X-ray luminosity is 420  M_☉ yr⁻¹ (Ranalli et al. 2003). Furthermore, the star formation rates and stellar masses of each clump imply gas-phase metallicities that are too high to explain the observed [O iii]/Hβ ratios were they due to star formation. Following the "fundamental plane" of Mannucci et al. (2010) and assuming that all the blended Hα+[N ii] λ6584 is due to star formation, we compute 12+O/H for the three components of 8.49, 8.77 and 8.82, respectively. According to Maiolino et al. (2008), these high gas phase metallicities would imply log₁₀[O iii]/Hβ ratios of 0.41, 0.03 and −0.06, i.e., high metallicity. The first of these is marginally below the measured log₁₀[O iii]/Hβ =0.53 ± 0.1 for component A, but the other two are completely inconsistent. As with the other arguments here, a low metallicity starburst cannot be completely ruled out but the preponderance of the evidence points to three AGNs. Future observations, e.g., with the James Webb Space Telescope, could settle the debate.

We retrieved the Advanced Camera for Surveys (ACS) imaging data in the B, V, I, and z bands taken as part of the GOODS survey from the archive and determined the morphologies and luminosities of each component source using the two-dimensional image decomposition program GALFIT (Peng et al. 2002). We find that each of the components in all ACS bands has morphologies consistent with disks. Based on analysis of more than 50, 000 simulated AGN+host galaxy morphologies at z ≲ 1 (Simmons & Urry 2008), this indicates that no more than 20% of each object's rest-frame UV light comes from a bulge.

In order to investigate whether any of the components have detected central point sources in the ACS images, we fit each component with both a single Sérsic component and a Sérsic + point source. We assess the robustness of the point-source detection via analysis of the χ² goodness-of-fit parameter and examination of the fit residuals. Only for component D is the improvement in χ² marginally significant, and the fit is unable to converge to physically realistic parameters without a central point source. We therefore determine that component D has a detected point source in the rest-frame UV bands, supporting the claim that D is an AGN. From our extensive simulations (Simmons & Urry 2008), the likelihood of a falsely detected point source is very small (<1%), while there is a 5% chance of missing a central point source. While the addition of a central point source does not improve the fit of component A, the addition of a second Sérsic parameter does. Component A is best fit in all bands by two disks, one compact and one more extended.

Locally, black hole mass is correlated with galaxy bulge velocity dispersion (Ferrarese & Merritt 2000; Gebhardt et al. 2000) and stellar mass (Häring & Rix 2004). We obtain a photometric spectral energy distribution (SED) of all four clumps using the ACS BVIz and WFC3 F140W images and fit a two-component star formation history to each clump to estimate its stellar mass (Schawinski et al. 2007). From these stellar masses, reported in Table 1, we infer black hole masses using the local relation of Häring & Rix (2004), yielding masses in the range of 3 × 10⁶ to 2 × 10⁷ M_☉. We caution that there are two major systematics that could affect these mass estimates: (1) the mass in each clump is assumed to be all bulge, and (2) the local black-hole–bulge relation may evolve to z = 1.3 (e.g., Bennert et al. 2011). The first consideration means that these masses are upper limits to the true black hole mass; since the morphological analysis indicates that any bulge is less than 20% of the observed light, the black hole masses are plausibly at least five times smaller.

The [O iii] λ5007 luminosity has been used as a proxy for black hole accretion rate (e.g., Heckman et al. 2004). We correct the observed [O iii] luminosity measured in the grism spectra for the contribution of the blended [O iii] λ4959 contribution to use it as a basis for calculating the accretion rate and therefore Eddington ratio of each black hole. We scale L[O iii] to an X-ray luminosity using the empirical correlation between the two for moderate-redshift AGNs (a factor of ∼10; Figure 5 of Trouille & Barger 2010). We then scale this X-ray luminosity to a bolometric luminosity. The local relation by Marconi et al. (2004) suggests a factor of ∼20 which is supported by recent measurements of z ∼ 1 AGNs by Simmons et al. (2011). Using bolometric luminosities obtained this way results in Eddington ratios of L/L_Edd = 1.4, 0.3 and 0.09 for components A, B, and D, respectively. We note that these Eddington ratios are likely underestimates due to the assumption that the whole stellar mass in each component represents pure bulge light and the 20% limit on the bulge light would lower the black hole masses, and raise the Eddington rations, by a factor of five respectively.

The observed Chandra X-ray properties are derived from 11 photons that cannot be localized to individual clumps and so represent the summed emission from the three AGNs. It is possible that the unobscured X-ray light is dominated by the AGN in component D which features a point source, with the other two AGNs being more obscured. This yields an Eddington ratio of 0.03 for component D, assuming a bolometric correction factor of 20. The AGNs in components A and B are likely more obscured but still visible in [O iii] and therefore intrinsically more luminous.

The in situ formation scenario is plausible due to the fact that these black holes have low mass and are accreting at a high Eddington rate. As we estimate below, even a low mass black hole seed can reach the observed mass range in a short time without requiring any extreme super-Eddington accretion episodes. To illustrate, we estimate below the growth as a function of time for a black hole accreting at the Eddington limit:

$$\begin{equation} M(t)=M(0) \exp \left(\frac{1-\epsilon }{\epsilon } \frac{t}{t_{\rm Edd}} \right), \end{equation} \tag{ 1 }$$

where the radiative efficiency ≈ 0.1 and the characteristic time scale for Eddington-limited growth is t_Edd = 0.45 Gyr. The growth curves for black hole seeds ranging from 1 to 10⁶ M_☉ in Figure 4 show that even very low mass seeds of ∼10² M_☉ can comfortably reach the observed black hole mass range of 10⁶–10⁷ M_☉ in ∼500 Myr, while a more massive ∼10⁶ M_☉ initial seed will reach the observed range in under 100 Myr. Their location at the centers of clumps embedded in a larger galaxy means that there likely is gas to sustain accretion to the observed masses. Similarly, the growth times for massive seed models approach the dynamical timescale of the host galaxy.

Zoom In Zoom Out Reset image size

Various mechanisms for producing black hole seeds, especially massive black hole seeds at extremely high redshift to account for z ∼ 6 quasars, have been proposed (Lodato & Natarajan 2006, 2007; see also Natarajan 2011). Two general channels have been theoretically proposed and extensively studied: (1) the remnant of a Population III star, (2) massive primordial gas disks that can either collapse directly into a black hole or fragment into a dense star cluster that collapses into a black hole via runaway collisions and stellar-dynamical processes (Lodato & Natarajan 2006, 2007; Devecchi & Volonteri 2009; Volonteri 2010). The location of the three AGNs at the centers of clumps embedded within a galaxy does not allow us to distinguish between these scenarios.

If indeed these observed black holes formed in situ as recently as z = 1.35, our theoretical models should be re-examined, as all the channels considered so far have been worked out in the context of primordial gas disks. The presence of metals at late times will rapidly enhance cooling and the outcomes for this case are yet to be explored. Even with the one proposal to form black holes as late as z = 3–4, Jimenez & Haiman (2006) require pockets of unenriched gas to exist, but it is unlikely that such conditions would survive to z = 1.35. A channel allowing seed formation at non-primordial metallicities may be required.

We note that the possibility that these black holes formed significantly earlier, perhaps even via the Pop III mechanism, cannot be ruled out. In this case, the present accretion event represents a late-stage re-activation.

With the current data, we cannot fully determine whether the host galaxy of the Triple AGN is a major merger of three individual progenitor galaxies, or whether it is a clumpy galaxy. Clumpy galaxies have been observed in deep Hubble observations (e.g., Moustakas et al. 2004; Elmegreen & Elmegreen 2005; Elmegreen et al. 2009) while mergers of multiple galaxies are rare, at least in the local universe (two orders of magnitude less frequent than binary mergers; see Darg et al. 2011). Neither the ACS optical image nor the WFC3 IR image shows clear signs of tidal tails, merger debris, or close companions, so even with the higher frequency of mergers at high redshift it seems more likely that the Triple AGN is a clumpy galaxy.

There are only two cases of Triple AGN systems reported in the literature (Djorgovski et al. 2007 and Liu et al. 2011). These multiple AGN systems are identified as merging systems. Each progenitor is thought to have brought in an already existing black hole, and the double AGN represents the stage of the merger prior to the final coalescence into a single central black hole. The properties of both the cases are different from the object presented here: the z ∼ 0 Triple AGN contains massive >10⁸ M_☉ black holes radiating at significantly sub-Eddington rates, while the z ∼ 0 triple quasar has high black hole masses and high luminosity.

Clumpy or irregular galaxies are seen in deep HST fields, rise in abundance with redshift, and make up >50% of galaxies at z > 1 (Moustakas et al. 2004; Elmegreen & Elmegreen 2005; Conselice et al. 2005; Elmegreen et al. 2009) while the fraction of merging galaxies remains below ∼10% (Bell et al. 2006; Lotz et al. 2008). The clumps in this common type of star-forming galaxy are thought to be fragmenting gas disks where individual regions become self-gravitating, star-forming clumps (e.g., Noguchi 1999; Immeli et al. 2004; Bournaud et al. 2007, 2011; Elmegreen et al. 2008a, 2008b).

Spectroscopic studies of high-redshift star-forming galaxies reveal that many of them are turbulent, gas-rich disks with clumps rapidly forming without the need for mergers (Genzel et al. 2008; Förster Schreiber et al. 2009, 2011; Daddi et al. 2010). Theorists have reproduced such gas-rich disk galaxies with the clumps emerging by fragmentation as sub-clumps become self-gravitating (e.g., Noguchi 1999; Immeli et al. 2004; Bournaud et al. 2007, 2011; Elmegreen et al. 2008a, 2008b). If the clumps are seeded with black holes, Elmegreen et al. (2008b) have shown that as the stellar clumps merge into a bulge, the black holes similarly coalesced into a supermassive black hole at the center, preserving the local M–σ relation (Ferrarese & Merritt 2000; Gebhardt et al. 2000).

As dense, gas-rich, collapsing regions of space, they are viable birth places for the direct formation of supermassive black holes and our discovery of the z = 1.35 Triple AGN shows that—if the host really is a clumpy galaxy—they can host growing black holes inside their constituent clumps.