Compton-thick active galactic nuclei from the 7 Ms observation in the Chandra Deep Field South
aa r X i v : . [ a s t r o - ph . H E ] A ug Astronomy&Astrophysicsmanuscript no. corral˙cdfs˙7ms c (cid:13)
ESO 2019August 27, 2019
Compton-thick active galactic nuclei from the 7 Ms observation inthe Chandra Deep Field South
A. Corral , , I. Georgantopoulos , A. Akylas , and P. Ranalli , Institute for Astronomy, Astrophysics, Space Applications, and Remote Sensing (IAASARS), National Observatory of Athens,15236 Penteli, Greece Instituto de F´ısica de Cantabria (IFCA), CSIC-UC, Avenida de los Castros s / n, 39005, Santander, Spain Lund Observatory, Department of Astronomy and Theoretical Physics, Box 43, SE-22100, Lund, Sweden Combient Mix AB, PO Box 2150, 40313, Gothenburg, SwedenReceived ; accepted
ABSTRACT
We present the X-ray spectroscopic study of the Compton-thick (CT) active galactic nuclei (AGN) population within the
Chandra
Deep Field South (CDF-S) by using the deepest X-ray observation to date, the
Chandra / ordata-quality dependent as possible. To obtain reliable automated spectral fits, we only considered the sources detected in the hard(2-8 keV) band from the CDF-S 2 Ms catalog with either spectroscopic or photometric redshifts available for 259 sources. Instead ofusing our spectral analysis to decide if an AGN is CT, we derived the posterior probability for the column density, and then we usedit to assign a probability of a source being CT. We also tested how the model-dependence of the spectral analysis, and the spectraldata quality, could a ff ect our results by using simulations. We finally derived the number density of CT AGN by taking into accountthe probabilities of our sources being CT and the results from the simulations. Our results are in agreement with X-ray backgroundsynthesis models, which postulate a moderate fraction (25%) of CT objects among the obscured AGN population. Key words.
X-rays: general; X-rays: di ff use background; X-rays: galaxies; galaxies: active
1. Introduction
X-ray surveys are very e ffi cient in detecting active galactic nu-clei (AGN) (Xue et al. 2011; Brandt & Alexander 2015). X-rayscan penetrate large amounts of dust and gas by piercing throughthe obscuring screen that hides the nucleus. Moreover, as X-raysare largely uncontaminated by the host galaxy emission, they canmore readily probe lower luminosity AGN than observations atlonger wavelengths. The most sensitive observations of the X-ray Universe, the 7 Ms survey in the Chandra
Deep Field South(CDF-S), reveal a surface density of a few tens of thousand AGNper square degree (Luo et al. 2017). For comparison purposes,the quasar (QSO) sky density obtained through color surveys(e.g., Croom et al. 2009) is only a couple of hundred per squaredegree.However, even the extremely e ffi cient X-ray surveys havedi ffi culties detecting the most highly obscured AGN. ObscuredAGN are a key ingredient in models for galaxy formation andevolution. Among them, Compton-thick AGN are the most dif-ficult to detect and characterize because of the huge amount ofintervening material obscuring their intrinsic emission. At thesame time, in order to derive a complete census of Compton-thick AGN, and then to determine the possible dependence ofobscuration on their intrinsic properties and their evolution, itis of vital importance to constrain these models. X-ray obser-vations, often complemented with observations in other wave-lengths, are still the best way to carry out the least-biased stud-ies of this type of sources (see Hickox & Alexander 2018 for arecent review). Compton-thick (CT) AGN have column densities (N H )higher than 10 cm − , so that Compton scattering becomesan important contributor to the attenuation of X-rays besidesphotoelectric absorption, which is the main absorption mech-anism at lower column densities. These extreme column den-sities render the CT AGN about two orders of magnitudefainter in the 2-10 keV band while the harder energies above10 keV pass relatively unscathed from the obscuring screen.Therefore, a very e ff ective way to search for the most heav-ily obscured AGN is by using the very hard surveys per-formed by SWIFT / BAT (see Ricci et al. 2015; Akylas et al. 2016)and
NUSTAR (Alexander et al. 2013; Harrison et al. 2016). Analternative route is to use very deep X-ray surveys in thesofter 2-10 keV band that manage to detect highly obscuredAGN at very faint fluxes due to their much higher sensitivity.Previous attempts to detect highly obscured and CT AGN inthis band include the works of Georgantopoulos et al. (2013),Brightman et al. (2014), Buchner et al. (2014), and Liu et al.(2017), all of which use
Chandra and
XMM-Newton observa-tions in the CDF-S.In this paper we exploit the deepest X-ray survey ever ob-tained, the 7 Ms CDF-S. This work di ff ers from previous stud-ies on the X-ray analysis of the CDF-S in that we do not sep-arate CT AGN from non-CT AGN, but we compute the prob-ability of a source being CT and use this probability to deriveour results. Additionally, even though we use 7Ms X-ray spec-tral data, we use the source catalog obtained by Luo et al. (2008)from the 2 Ms Chandra observations. This is because we requireour sources to have a su ffi cient number of counts so that the
1. Corral et al.: CT AGN in the CDF-S characterization of a source as CT is as unambiguous as pos-sible. Given the minimum number of source counts reported inLuo et al. (2008), we expect our spectral data to have at least ∼
30 counts in the hard (2-8 keV) band.
2. Chandra Deep Field South
The CDF-S is the deepest X-ray survey to date, covering anarea of 484.2 arcmin . The most recent catalog of X-ray sourceswithin the CDF-S was produced by using 102 observationswith a total exposure time of ∼ × − erg cm − s − in the hard band (Luo et al. 2008). The hard sam-ple is composed of 265 sources, and either spectroscopic (181sources) or photometric (78 sources) redshifts are available for259 of them (Hsu et al. 2014). The redshift distribution is pre-sented in Fig.1. Fig. 1.
Redshift distribution for full hard sample (solid his-togram), and CT AGN with probabilities >
90% described inSect. 4 (shaded histogram).We reduced all the
Chandra survey data in a uniform man-ner as described in Laird et al. (2009) with the
CIAO data anal-ysis software version 4.8. We used the
SPECEXTRACT script inthe
CIAO package to extract source spectra (with an extractionradius increasing as a function of the o ff -axis angle to enclose90% of the point spread function, PSF, at 1.5 keV), as well as re-sponse and auxiliary matrices, for each individual observation.The spectral data from each observation were then merged tocreate a single source spectrum, response, and auxiliary matri-ces for each source using the FTOOL tasks
MATHPHA , ADDRMF ,and
ADDARF, respectively. Background spectra were extracted infive di ff erent source-free regions for each observation. Then, foreach source, the closest background region among these five wasselected and then merged for each source by taking each sourcedetection (or non-detection) for each observation into account. Since sources near the edges of the field may not be present in allindividual observations because of variations in the aim pointsand roll angles, the final exposure times range between 400 ksand 6.6 Ms. The net (background subtracted) counts distributionin the full (0.5-8 keV) band is presented in Fig.2. Fig. 2.
Net (background subtracted) full (0.5-8 keV) band countsfor full hard sample (solid histogram), and CT AGN with proba-bilities >
90% described in Sect. 4 (shaded histogram). The lastbin in the solid histogram corresponds to the total percentage ofsources with more than 800 counts.
3. Automated spectral analysis
By comparing the di ff erent results obtained throughout theyears for the same sources (Norman et al. 2002; Tozzi et al.2006; Comastri et al. 2011; Georgantopoulos et al. 2013;Brightman et al. 2014; Buchner et al. 2014; Liu et al. 2017), it isclear that CT classification is model and data-quality dependent.Although the sample used in this work is rather small, our aimis to use it to develop an automated spectral selection methodas less model-dependent as possible that may be applied tolarge samples of X-ray spectra spanning a wide range in dataquality, such as the samples from surveys already carried out by Chandra and
XMM-Newton , and also the ones expected to becarried out by the upcoming X-ray missions
SRG / eROSITA and Athena . The main characteristic of our method is that we do notclassify sources as CT or not, but we derive the probability ofa source being CT (Akylas et al. 2016). Therefore in this work,we consider that an AGN is a CT candidate if the resultingprobability of the AGN being CT is larger than zero. We used
Xspec v12.9 (Arnaud 1996) to carry out the spectralanalysis. We selected Cash statistics applied to spectra binned to1 counts / bin, which allow us to obtain reliable spectral resultseven for low count data.Given the data quality among our sample, we selected thefollowing set of models, listed in increasing complexity order, tobe applied to our data: http: // / eROSITA ttps: // / web / athena2. Corral et al.: CT AGN in the CDF-S – An absorbed power-law plus a Gaussian emission line:
Xspec:zwabs*zpo+zgaus . The Gaussian component is in-tended to represent the Fe K α emission line, the most oftenobserved emission line in AGN X-ray spectra. In CT AGN,the equivalent width (EW) of this line is expected to be veryhigh, sometimes over 1 keV. – A double power-law plus a Gaussian line: zwabs*zpo+zpo+zgaus . The unabsorbed power-lawcan represent either soft-scattering of the primary (hard)power-law emission, or transmitted emission in the case of apartial covering absorber. – A modified power-law by a toroidal-shaped absorber plusa soft-scattered component: torus+zpo , where torus cor-respond to the model described in Brightman & Nandra(2011). This model is a more appropriate model for ourhighly absorbed sources since it consistently takes into ac-count photoelectric absorption, fluorescence line emission(Fe K α ), and Compton scattering. The parameters of thismodel are, besides the column density, photon index, andnormalization, the torus opening angle and its inclinationwith respect to the observer. We fixed this angles to 60 and80 degrees, respectively (see the following discussion).Out of the three spectral models described above, the last oneis the more physically-motivated one. However, high data qual-ity is needed to fit both the opening and inclination angles at thesame time. Fixing the opening angle to 60 degrees and the incli-nation angle to 80 degrees is a good approximation in the caseof modeling highly absorbed and CT AGN spectra, but it is notalways a good choice for less absorbed sources. For moderatelyto low absorbed sources, the derived column densities dependmore strongly on the inclination angle than for highly absorbedones (Lanzuisi et al. 2018). Whereas simple models, like the firsttwo, have been proven to be a good representation of the spectralshape up to column densities of 10 cm − (Liu et al. 2017). It isimportant to remember that, up to this point, we are not trying toderive the best-fit model parameters but to characterize the spec-tral shape well enough to be able to pinpoint highly absorbedsources.Therefore, we proceeded with the following two steps: weapplied the first two models to all of our sample, and then weonly applied the torus model to sources displaying high ab-sorption features in their spectra. We looked for these featuresby using the spectral-fitting results of the first two models (seeSect. 3.1 and Corral et al. 2014). Instead of fitting a highly-absorbed-AGN oriented model (thetorus model) to all of our sources, we made use of the auto-mated selection technique for highly obscured AGN presentedin Corral et al. (2014). This is a fast, less model-dependent, andreliable way to pinpoint heavily obscured sources without theneed to apply complex models that could be biased toward cer-tain kinds of AGN. This method uses very simple spectral mod-els (an absorbed power-law: zwabs*zpo+zgaus ; and a doublepower-law: zwabs*zpo+zpo+zgaus ) to select sources as highlyabsorbed candidates by pinpointing signatures of obscuration.To define a region in the best-fit spectral parameters spaceso that all CT sources would be selected, we refined the selec-tion technique presented in Corral et al. (2014) by using simula-tions. We simulated each source and background in our samplefive times by varying the column density from 10 to 10 cm − ,while also preserving their fluxes and redshifts. For column den- sities below 5 × cm − we used the Xspec model plcabs (Yaqoob 1997) to reproduce the absorbed power law, whereasfor CT column densities, we used the mytorus self-consistentmodel (Murphy & Yaqoob 2009). In both cases, an additionalsoft-scattered power-law component was also included with amaximum scattered fraction of 10% with respect to the primaryemission. We then applied our simple models to the simulateddata. In order to down-select all the CT sources, the best-fit pa-rameter for the considered models had to fulfill at least one of thefollowing selection criteria: the measured column density is inthe CT regime ( > cm − ); the power-law photon index is flat( < × cm − ;the power-law photon index is flat ( < α emission line EW is larger than 500 eV; and for sources with thelowest number of counts in the 2-8 keV band, and thus a limitingreliability of the spectral fits, we only require that the power-lawphoton index is flat ( < ≤ H < cm − ) and highly, but not CT, absorbed sources ( ≤ cm − < N H < cm − ). We also used the simulations described in the previous section toquantify how the data quality could limit our results. We appliedthe torus model described in Sect. 3 to all of the simulated data,thus obtaining the values displayed in Fig. 3. There are two clearproblematic regions in this plot labeled FP (false positives) andM (missing CT AGN).The most populated among the problematic regions in Fig. 3is region M, which corresponds to simulated CT AGN that arenot identified as such when fit by the torus model. All these sim-ulations are associated with spectra with less than 100 counts inthe hard (2-8 keV) band. Given the spectral counts distributionin our sample, this implies that we could miss up to 14% of CTcandidates in the case of spectra with less than 100 counts, thatis, we could miss six CT candidates of our actual sample due tolow data quality alone.The second problematic, but much less populated, region inFig. 3 is the FP region, which corresponds to sources misidenti-fied as CT candidates when fit by the torus model. These simula-tions also correspond to spectra with less than 100 counts in thehard band, and they could amount to up to 2% (only one sourcein the actual sample) of misclassified CT candidates.By combining the results from Sect. 3.1 and the ones pre-sented in this section, we can estimate the number of missingand / or misclassified CT candidates in the following analyses(see Sect. 5.2).
4. Bayesian CT probabilities
By applying the selection technique described in Sect. 3.1 to ouractual sample, we ended up with 59 highly absorbed candidates.According to our simulations, most of these sources should be atleast highly absorbed (N H > cm − ), and all of the CT AGNin our sample should be within these 59 candidates.As mentioned at the end of Sect. 3, once we identified thesources most likely to be CT AGN within our sample, we ap-plied a more appropriate model to them in order to obtain our
3. Corral et al.: CT AGN in the CDF-S
Fig. 3.
Simulated column densities (
Xspec:zwabs,plcabs,mytorus ) versus best-fit (recovered) col-umn densities (Brightman & Nandra 2011 torus model). Filled(red) circles correspond to highly absorbed candidates (seeSect. 3.1). FP: false positives. M: missing CT AGN.final spectral results as follows: the torus model described inBrightman & Nandra (2011) and a second power law that ac-counts for soft scattered emission. From the spectral results ofthis model, we confirm that most ( ∼ cm − . In applying this method, we find that 36, outof the initial 59 sources, have a probability higher than zero ofbeing CT (P CT > >
90% (seeFig.5 and Table 1). Taking the probabilities for all of the 36 can-didates into account, the e ff ective number of CT AGN in oursample is 27.It is important to remember that our intention is not to clas-sify sources as CT AGN nor to derive the actual column den-sities of our sources, but to derive the probability of being CT(P CT ). Therefore, the comparison presented in the last column ofTable 1 must only be considered qualitatively and not as a directcomparison (see Sect. 5.1). We tested that by applying di ff er-ent models such as plcabs, mytorus , and / or by adding reflec-tion ( Xspec: pexmon ), whereas changing the resulting valuesfor the best-fit column densities did not significantly a ff ect theresulting integrated probabilities over 10 cm − .
5. Discussion
Brightman et al. (2014); Buchner et al. (2014); Liu et al. (2017)performed systematic studies of CT AGN within the CDF-Sby using
Chandra data, finding nine, eight, and ten candidate CT AGN, respectively. However, their CT classifications arenot consistent with each other in many cases, being that 15AGN is the combined number of candidates from those works.Brightman et al. (2014); Buchner et al. (2014) used 4Ms CDF-Sspectral data but not exactly the same sources, whereas Liu et al.(2017) used 7Ms data but only presented the spectral analysisfor the brighter sources in that catalog.In this work, we derived CT probabilities >
90% for only10 out of the 15 previously reported CT AGN included inBrightman et al. (2014); Buchner et al. (2014); Liu et al. (2017)(see Table 1). For the remaining 5 AGN previously classifiedas CT, we find them only moderately to highly absorbed, ex-cept for source 180 which we find to be near-CT (probability ∼ ff erent results. In the case of sources 138,266, and 312, the di ff erences could be also due to the improvedspectral quality in the 7 Ms data, since these sources are notincluded in the Liu et al. (2017) sample. New CT candidatesare mainly due to lower number of counts in Brightman et al.(2014); Buchner et al. (2014), which used 4Ms data, or becausethe sources are not included in either of the Brightman et al.(2014); Buchner et al. (2014); Liu et al. (2017) samples. We compared our CT number count distribution with the mostrecent cosmic X-ray background (CXB) synthesis models ofAkylas et al. (2012) and Ueda et al. (2014). In our case, weweighted each of our candidate CT by the probability of thembeing CT (see Akylas et al. 2016), so that the number integralnumber count, N( > S), of sources per unit sky area with fluxhigher than S is defined as follows: N ( > S j ) = i = j X i = P CT Ω i , (1)where we summed all the sources with fluxes S i > S j at eachbin, and Ω i represents the sky coverage as a function of flux fromthe 2Ms survey(Luo et al. 2008).Instead of computing the errors in each bin, we estimated aconfidence interval by performing simulations. According to theresults from Sect. 3.2, we could be missing up to 6 CT candi-dates, and one of our candidates could have been wrongly se-lected. We also know that these seven sources have to fulfillthe selection criteria described in Sect. 3.1, that is, they mustbe among the 59 highly absorbed candidates with derived P CT =
0. Besides, their spectra must have less than 100 counts in thehard band. Taking all of this into account, we simulated 10000realizations in which we added up to six sources (simulated fol-lowing the fluxes of the highly absorbed candidates with lessthan 100 counts, and with random P CT ), and randomly removedone of the actual CT candidates. We considered our confidenceinterval to be the region that encompasses 99.7% of our simu-lated number count distributions (gray area in Fig.4).Our results are plotted in Fig. 4. As a comparison, we plot-ted the models of Akylas et al. (2012) for a 15% CT fraction
4. Corral et al.: CT AGN in the CDF-S
Table 1.
Spectral fits results for candidate CT AGN
LID z log N H Γ P1 / P2 Flux log L X log L Xunabs
CT probability PCcm − − erg cm − s − erg s − erg s − (%)(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)96 0.310 24 . + . − . . + . − . . + . − . . + . − . < .
001 1.38 42.3 44.2 98.4 nCT137 1.544 24 . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . . − . . + . − . . + . − . Chandra
ID from the Luo et al. (2010) catalog. (2) Redshift (two decimal and three decimal digits denote photometric andspectroscopic redshifts, respectively). (3) Intrinsic column density. (4) Photon index ( f corresponds to fixed photon index). (5) Ratio betweenthe scattered and the torus components. (6) Observed flux in the 2-10 keV band. (7) Observed luminosity in the 2-10 keV band. (8) Unabsorbedluminosity in the 2-10 keV band.(9) Probability of the source being Compton-thick. (10) Previous classification from Buchner et al. (2014),Brightman et al. (2014), and / or Liu et al. (2017): Compton-thick (CT), near-CT ( N H > × , nCT), not a CT (noCT). and 5% of the reflected emission, for a 25% CT fraction, andthe model of Ueda et al. (2014), which assumes a large ( ∼ ff erences between both models is thatthe Ueda et al. (2014) model has an extra free parameter allow-ing for di ff erential evolution of the CT AGN population relativeto the population of unobscured and mildly obscured AGN. Wealso plotted the results from Lanzuisi et al. (2018), which alsouse Chandra data but from the 3Ms COSMOS-Legacy survey(Civano et al. 2016). In that work, they also derived the prob-ability distribution for the column densities and use them in asimilar way as we did in this work. Our results look more con- sistent with the model proposed in Akylas et al. (2012) for a 25%percentage of CT AGN, a smaller value than the one presentedin Lanzuisi et al. (2018). They report a number density of CTAGN which increases from 30-55% with redshift. Because ofthe relative small size of our sample, we cannot test this evolu-tion with redshift in our case. A very recent work that also makesuse of the CDF-S 7Ms data, although a di ff erent spectral anal-ysis was carried out in that case, reports a moderate fraction ofCT when computing their number counts (Li et al. 2019), whichis consistent with our results. Although smaller fractions and / orCXB models with higher amounts of reflection are also consis-
5. Corral et al.: CT AGN in the CDF-S tent within errors, local studies point to small amounts of reflec-tion (Georgantopoulos & Akylas 2019). −15.50 −15.25 −15.00 −14.75 −14.50 −14.25 −14.00 −13.75 −13.50 log(S ) erg cm −2 s −1 l og ( N C T > S ) d e g − Fig. 4.
Number count distribution of CT AGN (circles) in CDF-S (this work) and for COSMOS-Legacy data in Lanzuisi et al.(2018) (crosses). Solid and dotted lines correspond to the modelpredictions presented in Akylas et al. (2012) for a CT fraction of15% plus 5% reflection, and for a CT fraction of 25%, respec-tively. The dashed line corresponds to the model in Ueda et al.(2014).
6. Conclusions
We used the deepest X-ray observation to date, the 7 Ms
Chandra observation of the CDF-S, to search for CT AGN in thehard (2-8 keV) selected sample presented in Luo et al. (2008),which is based on a shallower 2Ms observation. In this way, wewere able to improve previous spectral analyzes and to betterconstrain the intrinsic column densities of the sources in oursample. Moreover, by making use of simulations, we estimatethat among X-ray spectra with less than 100 counts in the hardband (2-8 keV), X-ray analyses could be missing ∼
14% of theCT AGN just because of the data quality.To optimize the automated spectral analysis and to quan-tify data-quality e ff ects, we applied an automated selection tech-nique to select highly absorbed candidates, and then we applieda Bayesian method to compute the probability of a source beingCT. We find 36 CT candidates with probabilities larger than zero,and 20 with probabilities > Acknowledgements.
We thank the referee for his / her helpful comments andsuggestions.The Chandra data were taken from the Chandra Data Archive atthe Chandra X-ray Center. AC acknowledges funding from ESA under thePRODEX project and financial support through grant AYA2015-64346-C2-1-P (MINECO / FEDER). AC is supported by a “Juan de la Cierva Incorporaci´on” postdoctoral contract from Ministerio de Ciencia, Innovaci´on y Universidades(Spain).
References
Akylas, A., Georgakakis, A., Georgantopoulos, I., Brightman, M., & Nandra, K.2012, A&A, 546, A98Akylas, A., Georgantopoulos, I., Ranalli, P., et al. 2016, A&A, 594, A73Alexander, D. M., Stern, D., Del Moro, A., et al. 2013, ApJ, 773, 125Arnaud, K. A. 1996, in Astronomical Society of the Pacific Conference Series,Vol. 101, Astronomical Data Analysis Software and Systems V, ed. G. H.Jacoby & J. Barnes, 17Brandt, W. N. & Alexander, D. M. 2015, A&A Rev., 23, 1Brightman, M. & Nandra, K. 2011, MNRAS, 413, 1206Brightman, M., Nandra, K., Salvato, M., et al. 2014, MNRAS, 443, 1999Buchner, J., Georgakakis, A., Nandra, K., et al. 2014, A&A, 564, A125Civano, F., Marchesi, S., Comastri, A., et al. 2016, ApJ, 819, 62Comastri, A., Ranalli, P., Iwasawa, K., et al. 2011, A&A, 526, L9Corral, A., Georgantopoulos, I., Watson, M. G., et al. 2014, A&A, 569, A71Croom, S. M., Richards, G. T., Shanks, T., et al. 2009, MNRAS, 399, 1755Georgantopoulos, I. & Akylas, A. 2019, A&A, 621, A28Georgantopoulos, I., Comastri, A., Vignali, C., et al. 2013, A&A, 555, A43Harrison, F. A., Aird, J., Civano, F., et al. 2016, ApJ, 831, 185Hickox, R. C. & Alexander, D. M. 2018, ARA&A, 56, 625Hsu, L.-T., Salvato, M., Nandra, K., et al. 2014, ApJ, 796, 60Laird, E. S., Nandra, K., Georgakakis, A., et al. 2009, ApJS, 180, 102Lanzuisi, G., Civano, F., Marchesi, S., et al. 2018, MNRAS, 480, 2578Li, J., Xue, Y., Sun, M., et al. 2019, ApJ, 877, 5Liu, T., Tozzi, P., Wang, J.-X., et al. 2017, ApJS, 232, 8Luo, B., Bauer, F. E., Brandt, W. N., et al. 2008, ApJS, 179, 19Luo, B., Brandt, W. N., Xue, Y. Q., et al. 2010, ApJS, 187, 560Luo, B., Brandt, W. N., Xue, Y. Q., et al. 2017, ApJS, 228, 2Murphy, K. D. & Yaqoob, T. 2009, MNRAS, 397, 1549Norman, C., Hasinger, G., Giacconi, R., et al. 2002, ApJ, 571, 218Ricci, C., Ueda, Y., Koss, M. J., et al. 2015, ApJ, 815, L13Tozzi, P., Gilli, R., Mainieri, V., et al. 2006, A&A, 451, 457Ueda, Y., Akiyama, M., Hasinger, G., Miyaji, T., & Watson, M. G. 2014, ApJ,786, 104Xue, Y. Q., Luo, B., Brandt, W. N., et al. 2011, ApJS, 195, 10Yaqoob, T. 1997, ApJ, 479, 184 . C o rr a l e t a l . : C T AGN i n t h e C D F - S −8 −7 −6 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −9 −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −7 −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 −8 −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −9 −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 −8 −8 −7 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −9 −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −9 −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D −8 −7 keV (Photons cm −2 s −1 keV −1 ) E n e r gy ( k e V ) L I D F i g . . X -r a yun f o l d e d s p ec t r a f o r ca nd i d a t e C T AGNw it hp r ob a b iliti e s > % ..