A COMPTON-THICK ACTIVE GALACTIC NUCLEUS AT z ∼ 5 IN THE 4 Ms CHANDRA DEEP FIELD SOUTH

While optically bright quasars are the most spectacular expression of accretion onto supermassive black holes (SMBHs) at galaxy centers, it is widely believed that SMBHs grow most of their mass during obscured phases, in which the detection of the nuclear power becomes challenging (e.g., Fabian 1999). Large amounts of gas and dust are found to hide the majority of active galactic nuclei (AGNs) in the nearby and distant universe, as demonstrated by deep and wide X-ray surveys over different sky fields (see, e.g., Brandt & Hasinger 2005 for a review). From ∼30% to ∼50% of all AGNs are believed to be obscured by extreme gas column densities above N_H = σ⁻¹_T ∼ 10²⁴ cm⁻². These objects are dubbed "Compton thick" and represent the most elusive members of the AGN population. The evidence for an abundant population of local Compton-thick objects is compelling: up to ∼50% of nearby Seyfert 2s contain a Compton-thick nucleus (Risaliti et al. 1999; Akylas & Georgantopoulos 2009); about 50 objects—mostly local—have been certified as "bona fide" Compton-thick AGN by X-ray spectral analysis (Comastri 2004).

Synthesis models of the X-ray background (XRB) suggest that Compton-thick AGNs must be abundant at least up to z ∼ 1 to explain the peak of the XRB at 30 keV (see, e.g., Gilli et al. 2007; Treister et al. 2009, and references therein). A population of distant, Compton-thick AGNs, as abundant as that predicted by XRB synthesis models, is also required to match the SMBH mass function measured in nearby galaxies with that of "relic" SMBHs grown by accretion (e.g., Marconi et al. 2004). In recent years it has been proposed that Compton-thick AGNs represent a key phase of the BH/galaxy coevolution, during which the BH is producing most of its feedback into the host galaxy (e.g., Daddi et al. 2007; Menci et al. 2008), and it has also been suggested that their number density steeply increases with redshift (Treister et al. 2009).

The observation of heavily obscured AGNs at high-z, z ≳ 2–3, remains challenging and it is very difficult to estimate their abundance since they produce only a small fraction of the XRB emission and are thus poorly constrained by synthesis models. Deep X-ray surveys have proven effective in revealing a few "bona fide" Compton-thick AGNs at high-z. For instance, four such objects at 1.53 < z < 3.70 have been discovered in the Chandra/XMM-Deep field South (CDFS; see Norman et al. 2002; Comastri et al. 2011; Feruglio et al. 2011). Other examples of candidate Compton-thick AGN at high-z have been reported (e.g., Tozzi et al. 2006; Polletta et al. 2008), even up to z = 5.8 (Brandt et al. 2001), albeit with poorer X-ray photon statistics. Selection techniques based on the strength of the mid-IR emission with respect to the optical and X-ray emission have also been developed and applied to select large populations of candidate Compton-thick AGNs up to z = 2–3 (Daddi et al. 2007; Alexander et al. 2008a; Fiore et al. 2009; Bauer et al. 2010). Once more, however, the lack of X-ray spectra prevents an unambiguous determination of the absorbing column density, making the measurements by these works largely uncertain.

In this Letter, we report the discovery of a "bona fide" Compton-thick AGN at z = 4.76 in the 4 Ms CDFS. A concordance cosmology with H₀ = 70 km s⁻¹ Mpc⁻¹, Ω_m = 0.27, and Ω_Λ = 0.73 is adopted throughout this Letter.

We searched for V-band dropout objects in the Hubble Space Telescope (HST)/Advanced Camera for Surveys (ACS) v2.0 data of GOODS-South (Giavalisco et al. 2004) associated with X-ray emission in the 4 Ms Chandra image. We used the V₆₀₆-dropout selection criteria from Oesch et al. (2007), which effectively pick sources at 4.7 < z < 5.7.¹¹ Details on the production of the V-dropout catalog are given in Su et al. (2011). Additionally, we required a stellarity parameter (CLASS_STAR) greater than 0.9 in the z₈₅₀ band to choose point-like sources. This led to 21 star-like V₆₀₆-dropouts with z₈₅₀ < 25.5, among which there are four z ∼ 5 galaxies, eleven stars, three lower-redshift galaxies, and one z ∼ 5 AGN, which is the only object detected in X-rays (XID403 in the 4 Ms CDFS catalog of Xue et al. 2011). The remaining two candidates have not been identified spectroscopically. The measured AB magnitudes of XID403 in ACS images are $V_{606}=26.84\pm 0.10, i_{775}=25.21\pm 0.04, \rm{and }\ z_{850}= 25.05\pm 0.04$. The 5σ detection limit in the B₄₃₅ band is 28.4 AB mag.

XID403 ($\alpha _{\rm{J2000}}$ = 03:32:29.29, $\delta _{\rm{J2000}}$ = −27:56:19.5) was recognized as an AGN at z = 4.76 based on FORS-2 spectroscopy (Vanzella et al. 2006). Its optical spectrum exhibits a narrow (FWHM ≲1000 km s⁻¹) Lyα emission line and a broader (FWHM ∼2000 km s⁻¹) NVλ1240 emission line, with an integrated flux similar to Lyα. A more recent spectrum with DEIMOS/Keck confirms both features (Coppin et al. 2009). The spectral energy distribution (SED) of XID403 was published by Coppin et al. (2009). Based on a LABOCA detection at 870 μm, they showed that this source is a bright submillimeter galaxy with star formation rate (SFR) ∼ 1000 M_☉ yr⁻¹. A large reservoir of molecular gas (∼1.6 × 10¹⁰ M_☉) was also identified through CO(2–1) observations (Coppin et al. 2010).

We considered the same optical to mid-IR data sets used by Coppin et al. (2009) and improved on the SED by adding the detection at 1.1 mm by AzTec/ASTE ($f_{1.1\, \rm mm}=3.3\pm 0.5$ mJy; Scott et al. 2010) and the Y, J, and Ks magnitudes from the deep NIR imaging by HAWK-I/VLT ($Y_{\rm{AB}}=24.56\pm 0.12$, $J_{\rm{AB}}=24.37\pm 0.14$, $K_{\rm{AB}}=24.03\pm 0.20$; Castellano et al. 2010). This object is also detected (at ∼3σ) at 1.4 GHz with a peak flux of 22.3 μJy (N. Miller 2010, private communication). Unfortunately, it falls just outside the areas covered by the 16 μm Spitzer/Infrared Spectrograph mosaic (Teplitz et al. 2011) and GOODS-Herschel (PI: D. Elbaz). XID403 is not detected in the 3 Ms XMM image of the CDFS.

A total exposure of ≲4 Ms has been accumulated on the CDFS as a result of 54 individual observations with ACIS-I performed during three different time periods: ∼0.8 Ms in 2000, ∼1 Ms in 2007, and ∼2 Ms in 2010. X-ray data products including event files for each observation and also for the merged data set are publicly available.¹² In this Letter, we use the data products by Xue et al. (2011) who derived X-ray source catalogs from a full reprocessing and astrometric recalibration of the event files. We used CIAO v4.1 and the Funtools package¹³ to perform X-ray aperture photometry at the position of XID403. The separation between the optical and X-ray centroids is ∼04, which is well within the 1σ X-ray source positional uncertainty (∼047). To maximize the signal-to-noise ratio (S/N), we measured the source counts in different bands within a small aperture of 3'' radius, which encloses ∼50% of the point-spread function (PSF) at 1.5 keV at the source location (∼8 arcmin off-axis). We measured 37.0 ± 8.7 net counts in the 0.9–4 keV band, corresponding to a ∼4.2σ detection.¹⁴ We verified that similar results are obtained when using different local background regions. The hardness ratio, defined as HR = (H−S)/(H+S), where S and H are the net counts observed in the 0.5–2 keV and 2–7 keV bands, respectively, is HR = 0.23 ± 0.24 (not corrected for vignetting). This value, for an AGN at z ∼ 5 with a standard intrinsic spectrum (i.e., Γ = 1.8), is highly suggestive of heavy obscuration. For comparison, an AGN at z ∼ 5 with N_H ≲ 10²³cm⁻² is expected to have HR ≲−0.35 at the source position. We extracted the X-ray spectrum in the 0.5–7 keV band using the same 3'' radius aperture and verified that below 0.9 keV and above 4 keV the source emission is indistinguishable from the background. The spectrum and response files were created using the psextract script in CIAO. Since psextract does not account for the PSF fraction when building up the effective area file, we multiplied the 0.5–2 keV and 2–10 keV fluxes as obtained from the spectral fit by a factor of 2 and 2.5, respectively, to recover the full aperture-corrected X-ray fluxes. We found consistent results either using spectral responses extracted from individual observations, or those obtained as an exposure-weighted mean over all individual responses. To double check the reliability of this procedure, we also built spectral response files for one of the Chandra exposures (ObsID = 8594) using the ACIS-Extract software (Broos et al. 2010) which allows proper construction of effective area files at any PSF fraction. Again, consistent results are found when using the ACIS-Extract responses. We analyzed the X-ray spectrum with XSPEC v11.3.2 using the Cash statistic (Cash 1979) to estimate the best-fit parameters. Errors are quoted at 1σ confidence level. We first fitted the data using a power-law spectrum modified by galactic absorption, which returns Γ = −0.64^+1.15_−0.73. We then used the plcabs model (Yaqoob 1997), which follows the propagation of X-ray photons within a uniform, spherical obscuring medium, accounting for both photoelectric absorption and Compton scattering. This model can be used in the case of heavy absorption (up to 5 × 10²⁴ cm⁻²) and up to rest-frame energies of ∼20 keV. When fixing the photon index to Γ = 1.8, we derived a Compton-thick column density of N_H = 1.4^+0.9_−0.5 × 10²⁴ cm⁻² (see Figure 1). If we conservatively assume Γ = 1.0, which is ≳4σ off the average intrinsic AGN value, we still obtain N_H>10²⁴ cm⁻². The measured absorption should be interpreted as a lower limit, since a cold reflection model (pexrav), corresponding to N_H ≳ 10²⁵ cm⁻², provides an equally good fit.¹⁵ Fitting the data with the recent MYTorus model (Murphy & Yaqoob 2009), which accounts for a toroidal distribution of the obscuring matter, again returns N_H>10²⁴ cm⁻². No prominent iron Kα line is observed at 1.1 keV (i.e., 6.4 keV rest frame), but this is not in conflict with the Compton-thick scenario. Indeed, the observed equivalent width (EW) scales with (1 + z)⁻¹, so that EW_rest∼ 1–2 keV, as is typical of Compton-thick AGNs, would translate into EW_obs∼ 170–340 eV. We verified that such a weak line can be easily accommodated in the fit and that only loose upper limits can be derived for EW_rest < 4.3 keV at 90% c.l.). The aperture-corrected X-ray fluxes, as extrapolated from the X-ray fit, are f_0.5–2 = (4.2 ± 1.5) × 10⁻¹⁷ erg cm⁻² s⁻¹ and f_2–10 = (6.8 ± 2.4) × 10⁻¹⁶ erg cm⁻² s⁻¹. The intrinsic (de-absorbed), rest-frame 2–10 keV luminosity is ∼2.5 × 10⁴⁴ erg s⁻¹, which places XID403 at the low end of the X-ray luminosity range for type-2 quasars. Admittedly, the uncertainties in the geometry of the obscuring (and reprocessing) material might substantially affect the derivation of the intrinsic luminosity. However, we note that, if the spectrum were produced by pure reflection and a typical reflection efficiency of ∼2% is assumed (Gilli et al. 2007), the intrinsic luminosity would be even higher.

Zoom In Zoom Out Reset image size

XID403 was detected in the 1 Ms CDFS catalog by Giacconi et al. (2002, XID=618), with f_0.5–2 = (1.6 ± 0.5) × 10⁻¹⁶ erg cm⁻² s⁻¹ and f_2–10 < 7.6 × 10⁻¹⁶ erg cm⁻² s⁻¹. The ∼4 times larger soft X-ray flux is likely due to contamination from high background fluctuations over the larger (8'' radius) extraction region adopted in that catalog. We checked the photometry of the 2000, 2007, and 2010 periods separately using a smaller 3'' radius: no significant source variability is detected in any band. XID403 was below the detection thresholds of the 1 Ms CDFS catalog by Alexander et al. (2003) and 2 Ms CDFS catalogs by Luo et al. (2008).

4.1. SED Fitting

We searched for additional indicators of heavy obscuration by considering data at other wavelengths. We first investigated the observed (i.e., not corrected for absorption), rest-frame 2–10 keV to 6 μm luminosity diagnostic ratio (X/IR; e.g., Alexander et al. 2008a). We derived the X-ray luminosity from the spectral fit and used the ${\sim} 4\,\mu \rm{m}$ rest-frame luminosity (derived from the Spitzer/MIPS data point at 24 μm) as a proxy for the 6 μm luminosity. The X/IR ratio of ∼5 × 10⁻³ would place XID403 in the region populated by Compton-thick AGN (Alexander et al. 2008a). When considering the $F(24\,\mu \rm{m})/F(R)$ versus R − K color diagram elaborated by Fiore et al. (2009), XID403 would fall in "cell E", where a significant fraction (∼25%–30%) of galaxies is found to host a heavily obscured, candidate Compton-thick AGN.

We then built the SED from the X-rays to the radio regime (see Figure 2). As already shown by Coppin et al. (2009), the radio to FIR emission of this source is dominated by dusty star formation, at a rate of ∼1000 M_☉ yr⁻¹. We note that X-ray binaries associated to such a high SFR would produce L^obs_2–10 ∼ (2–5) × 10⁴² erg s⁻¹ (Ranalli et al. 2003; Lehmer et al. 2010), similar to the $\rm{observed}$ value. However, X-ray binaries have on average much softer spectra (Γ ≳ 1.5; Remillard & McClintock 2006) than observed (Γ = −0.64). If absorption is invoked to explain such a spectral hardness, then intrinsic luminosities of ≈10⁴⁴ erg s⁻¹ are derived, which are incompatible with X-ray binary emission.

Zoom In Zoom Out Reset image size

A reasonable match with the MIR to UV data points could be obtained with a reddened QSO template with A_V ∼ 0.7–0.8 (adopting the extinction curve of Gaskell & Benker 2007). When converting the measured optical extinction into an equivalent hydrogen column density by applying the relation valid for the Milky Way interstellar medium (N_H ∼ 1.8 × 10²¹A_V), we find N_H ∼ 1.3 × 10²¹ cm⁻², which is 3 dex smaller than what is estimated from the X-ray spectral fit. A mismatch between the X-ray and the optically estimated column density, in the range of ∼3–100, is observed in local AGN, calling for a number of interpretations (e.g., low dust-to-gas ratio; Maiolino et al. 2001). The mismatch observed in XID403 is ∼1000, which would make this object extreme.

In their SED analysis Coppin et al. (2009) suggest a different, stellar origin for the optical/UV rest-frame emission. Following the parameterization used by Vignali et al. (2009) and Pozzi et al. (2010) for obscured AGNs, we fitted the SED of XID403 with a stellar component, an AGN torus component, and a dusty starburst component. The dusty starburst is responsible for the bulk of the FIR to radio emission, while the AGN torus produces the entire emission at 24 μm (4 μm rest frame). A galaxy template with M_⋆ ∼ 1.2 × 10¹¹ M_☉, A_V ∼ 1, and a ∼1 Gyr old constant SFR nicely fits the optical/UV_rest data. However, a possible problem in interpreting the optical/UV_rest emission as stellar light is that XID403 is point-like in the deep HST/ACS images (CLASS_STAR = 0.99 in i₇₇₅ and z₈₅₀), which would imply a half-light radius of <0.3 kpc. Although very compact morphologies have been observed in a fraction of distant submillimeter galaxies (Ricciardelli et al. 2010), the point-like nature of XID403, coupled to the presence of broad NV emission, might suggest that the optical/UV_rest light has a nuclear origin. In particular, we could be looking at a fraction (∼10%) of nuclear radiation that leaks out without being absorbed or is scattered toward us and thus would be polarized. If true, the effective extinction to the nucleus would be much higher than that estimated by fitting the whole MIR to UV emission with a reddened QSO template, being more in line with the large X-ray column density. This interpretation has already been proposed by Polletta et al. (2008) to explain the relatively blue optical/UV emission and broad line components of two submillimeter galaxies at z ∼ 3.5 hosting heavily obscured AGNs, similar to XID403. Also, although the stellarity parameter is uncertain for the faint K-band detection, the decrease of CLASS_STAR from 0.95 in Y to 0.74 in K may also suggest that the host galaxy contributes significantly only at λ_rest>4000 Å. In Figure 2, we show a possible SED decomposition for XID403 obtained by adding an AGN torus component and a scattering component (corresponding to ∼10% of the AGN intrinsic UV emission¹⁶) to the SED of Arp220. In summary, the full SED analysis shows that XID403 is not a classic, type-2 QSO (i.e., a narrow-line, X-ray obscured AGN whose physical properties can be explained within the standard, geometry-based Unified Model; Norman et al. 2002), but points to a complex physical picture likely related to its active assembly phase.

4.2. Black Hole and Stellar Mass Growth

4.3. Expectations for High-z Compton-thick AGNs

We acknowledge the following supporting agencies and grants: Italian Space Agency (ASI) under the ASI-INAF contracts I/009/10/0 and I/088/06/0 (R.G., C.V., A.C.); NASA through CXC grant SP1-12007A and ADP grant NNX10AC99G (W.N.B., Y.Q.X., B.L.); Chandra archival grant SP89004X (A.P., R.G.); DFG cluster of excellence Origin and Structure of the Universe (P.R.). We thank the referee for a constructive report and N. Miller and M. Mignoli for useful discussions. R.G. gratefully acknowledges hospitality by the Johns Hopkins University where this work started.