Free Access
Issue
A&A
Volume 569, September 2014
Article Number A71
Number of page(s) 15
Section Extragalactic astronomy
DOI https://doi.org/10.1051/0004-6361/201424129
Published online 25 September 2014

© ESO, 2014

1. Introduction

The hard X-ray surveys (2–10 keV) performed with the Chandra1 and XMM-Newton2 missions provide the most unbiased census of the accretion history in the Universe because they can penetrate large amounts of dust and gas. The deep 4 Ms Chandra survey reached a surface density of ~20 000 active galactic nuclei (AGNs) deg-2 (e.g., Xue et al. 2011). Most X-ray AGNs are obscured by high column densities (NH), typically above 1022 cm-2 (e.g., Tozzi et al. 2006; Akylas et al. 2006). In contrast, optical quasi-stellar object (QSOs) surveys yield surface densities lower by about two orders of magnitude because they are prone to obscuration, although [OIII] selection diminishes the effect of this bias (Bongiorno et al. 2010). Significant improvements have been made in mid-IR surveys, which now reach similar surface densities as X-ray surveys (e.g., Brown et al. 2006; Delvecchio et al. 2014) because they can easily detect luminous QSOs. Nevertheless, they can hardly separate obscured and, in general, low-luminosity AGNs from star-forming galaxies where the host-galaxy colors dominate the spectral energy distribution (e.g., Barmby et al. 2006; Georgantopoulos et al. 2008). However, even the very efficient hard X-ray surveys fail to detect a fraction of highly obscured AGNs (NH> 1023 cm-2), and in particular, AGNs with column densities above 1.5 × 1024 cm-2. The latter are the Compton-thick (CT) AGNs, where the primary mechanism for X-ray attenuation is Compton scattering on electrons instead of photoelectric absorption, which is the primary absorption mechanism at lower column densities.

Several authors have suggested that the number of CT sources can be constrained by using the spectrum of the extragalactic X-ray light, the X-ray cosmic background (XCB). CT AGNs are a basic ingredient of XCB synthesis models (e.g., Gilli et al. 2007; Treister et al. 2009; Ballantyne et al. 2011; Akylas et al. 2012) because they are needed to reproduce the broad peak at 20–30 keV observed in the XCB (Frontera et al. 2007; Moretti et al. 2009). However, different authors reported very different CT AGNs fractions, the exact intrinsic fraction remaining uncertain by at least a factor of two, ranging from about 10% of the total AGNs population up to 35%. This is most likely caused by the strong parameter degeneracy on XCB synthesis models, as shown in Akylas et al. (2012). Owing to ultra-hard X-ray surveys above 10 keV performed with Swift3 and INTEGRAL4, CT AGNs are commonly observed in the local Universe and represent 4–20% of local active galaxies at energies 15–200 keV down to a flux limit of 10-11 erg cm-2 s-1 (see Burlon et al. 2011, and references therein). NuSTAR5 is currently extending these searches to about two orders of magnitude deeper (Alexander et al. 2013; Lansbury et al. 2014).

Still, the identification of CT AGNs in the commonly used 2–0 keV band is difficult. Attempts to identify CT AGNs have been made primarily in deep X-ray surveys with Chandra and XMM-Newton missions. These efforts include those of Tozzi et al. (2006), Comastri et al. (2011) and Georgantopoulos et al. (2013), all in the Chandra deep field south (CDFS), the region of the sky with the deepest X-ray observations both in Chandra and XMM-Newton. An alternative approach is to cover a large area of the sky albeit at a relatively bright flux limit, and at lower redshifts. According to the XCB population synthesis models of e.g., Akylas et al. (2012), the fraction of CT sources among all AGNs increases by a factor of three as the flux limit, in the 2–10 keV band, decreases from 2 × 10-14erg cm-2 s-1, the effective flux limit of the 120 deg2 serendipitous XMM/SDSS survey (Georgakakis & Nandra 2011), to 5 × 10-17erg cm-2 s-1, the flux limit of the 0.12 deg2, 4Ms survey in the CDFS. This means that the XMM/SDSS, composed of ~40 000 X-ray sources detected over an area of ~122 deg2, contains a factor of a couple of hundred more CT AGNs than the CDFS. Given that Brightman & Ueda (2012) reported 40 CT AGNs, or CT candidates, in the 4Ms CDFS survey, this implies that there are about 8000 CT AGNs within the XMM/SDSS survey.

Here, we present a fully automated selection technique of highly obscured (NH> 1023 cm-2), and CT AGNs (NH> 1024 cm-2). The most reliable way of identifying highly obscured AGNs and CT AGNs from X-ray surveys is to manually fit their X-ray spectra to derive the actual intrinsic column density. However, in large X-ray samples, this method is extremely time consuming, and less reliable color-selection techniques are often used. We developed a highly efficient selection technique (efficiency ~80% in selecting highly obscured AGNs), which makes use of automated spectral fits to pinpoint this type of sources, and can be applied to large X-ray surveys. To develop this technique, we used X-ray spectral data from the XMM-Newton serendipitous source catalog (Watson et al. 2009), and applied automated X-ray spectral fits implemented for the XMM-Newton spectral-fit database (Corral et al. 2014). The test sample of AGNs used to develop this technique is composed of more than 1000 AGNs with spectroscopic redshifts extracted from the XMM/SDSS-DR7 (Sloan Digital Sky Survey Data release 7) cross-correlation presented in Georgakakis & Nandra (2011).

2. XMM-Newton serendipitous source catalog

The XMM-Newton serendipitous source catalog is the largest catalog of X-ray sources built to date (Watson et al. 2009). In its latest release, the 3XMM-Newton data release 4 (hereafter 3XMM-DR4) contains photometric information for more than 500 000 source detections corresponding to ~370 000 unique sources. As part of the catalog construction, time series and spectra were also extracted if the source counts collected in the European Photon Imaging Camera (EPIC) were >100 (more than 120 000 detections). As a result, 3XMM-DR4 contains X-ray source and background spectra as well as ancillary matrices for more than ~85 000 individual sources. The redistribution matrices used in this work are the canned matrices provided by the XMM-Newton Science Operations Centre (SOC). Note that the EPIC camera is composed of three detectors: one pn and two MOS cameras6. The count limit of 100 counts adopted in spectral extraction during the construction of the 3XMM-DR4 catalog corresponds to the addition of source (background-subtracted) counts in the three detectors7.

3. XMM-Newton spectral-fit database

The XMM-Newton spectral-fit database is an European Space Agency (ESA) funded project aimed at the construction of a catalog of automated spectral-fitting results corresponding to all sources for which spectral data are available within the 3XMM-DR4. The main goal is to provide the astronomical community with a tool to construct large and representative samples of X-ray sources according to their spectral properties, rather than to their photometric ones.

The resulting spectral-fit database contains one row per source and observation, listing source information, spectral-fit output parameters and errors, as well as fluxes and additional information about the goodness of fit for every model applied8.

3.1. Automated spectral fitting

The spectral-fit database is constructed by using automated spectral fits. The software used to perform the spectral fits is XSPEC v12.7 (see Arnaud 1996), the standard package for X-ray spectral analysis. 3XMM-DR4 source spectra are grouped to one count per bin, and Cash statistics, implemented as C-stat in XSPEC, are used to fit the data. This statistic was selected instead of the more commonly used χ2 statistics to optimize the spectral fitting for low-quality spectra. Grouping to one count per bin in combination with C-stat has been proven to work very well for low-count spectra down to 40 net (background-subtracted) counts (Krumpe et al. 2008). All available instruments and exposures for a single observation of a source are fitted together. All parameters for different instruments are tied together except for a relative normalization, which accounts for the differences between different flux calibrations for different EPIC instruments (MOS1, MOS2, and pn), which it is left free to vary. The distribution of ratios between the normalizations for the different instruments is shown in Fig. 1. The plotted values correspond to the sample used in this work, described in Sect. 4.1, and they were obtained from an absorbed power-law model fit.

thumbnail Fig. 1

Distribution of normalization ratios for the three EPIC instruments: MOS2-to-MOS1 ratio (empty histogram); pn-to-MOS1 ratio (line shaded histogram); and pn-to-MOS2 ratio (filled histogram).

Open with DEXTER

A lower limit on the number of counts in each individual spectrum was imposed to ensure a minimum quality on the spectral fits. As a result, not all 3XMM-DR4 spectra, but only spectra corresponding to a single EPIC instrument, with more than 50 source counts in the full band (0.5–10 keV) were used in the spectral fits. Parameter errors were computed and reported in the database at the 90% confidence level (ΔC = 2.706 in one interesting parameter). The final XMM-Newton spectral-fit database contains spectral-fitting results for 114 000 detections corresponding to ~77 000 unique sources.

Three simple (absorbed power-law, absorbed thermal, and absorbed black-body models) and three more complex (absorbed double power-law, absorbed thermal plus power-law, and absorbed black-body plus power-law) models were implemented within the spectral-fitting pipeline according to the number of counts in the X-ray spectra. Simple models were applied to sources in 3XMM-DR4 with more than 50 counts, and more complex models only to sources with more than 500 counts. These models were selected and optimized to reproduce the most commonly observed X-ray spectral shapes among different astronomical sources. Unlike the spectral fits used in this work (see Sect. 4.2), only X-ray data were used to construct the full spectral-fit database, that is, no information about the source type or its redshift was used during the automated spectral fits.

C-stat statistics lacks an estimate of the goodness of fit. To provide a proxy of the fit quality in the spectral-fit database, goodness of fit was estimated by using the XSPEC command goodness. This command performs a number of simulations and returns the fraction of the simulations that results in a better fit statistic. Therefore, for high return values of this command, a spectral fit with an N% goodness value can be rejected at the N% confidence level. The reduced χ2 value, obtained by using C-stat as the fitting statistic, is also included in the database.

Sometimes, the automated fitting process is unable to constrain all the variable parameters during the error computation. In these cases, spectral parameters that cannot be constrained are fixed during the spectral fits. The fixed values of the parameters depend on the data quality, the complexity of the model, and the energy band in which the spectral fit is being performed. In the case of simple models, the parameter that cannot be constrained is fixed to the value obtained by fitting a model that only includes the corresponding model component in the energy band that encompasses the maximum contribution of that component. For example, if the power-law photon index cannot be constrained in the case of the absorbed power-law model, its value is fixed to that obtained by fitting a power-law model without absorption in the hard (2–10 keV) band. The input parameter values for complex models, which are also the values the parameters are fixed to if they cannot be constrained, are the ones obtained from the spectral-fitting results of the simple models.

A complete description of the XMM-Newton spectral-fit database and the automated spectral-fitting pipeline will be presented in a forthcoming paper (Corral et al., in prep.).

4. Automated selection of highly absorbed candidates

4.1. Test sample

thumbnail Fig. 2

Intrinsic 2–10 keV luminosity versus redshift for the full XMM/SDSS spectroscopic sample (gray dots), the sample of the 81 highly absorbed candidates based on the automated spectral-fit selection (circles), and the 28 confirmed highly absorbed sources (filled circles, see text).

Open with DEXTER

To test our automated selection technique, a sample of AGNs was extracted from the XMM/SDSS cross-correlation presented in Georgakakis & Nandra (2011). The initial sample is composed of 1015 sources sources, detected in the hard band (2–8 keV), for which spectral data within the 3XMM-DR4 catalog with more than 50 counts in at least one EPIC instrument, and spectroscopic redshifts within SDSS-DR7 were available. Figure 2 presents the intrinsic luminosities (in the 2–10 keV energy band) as a function of redshift for this sample. Observed fluxes (in 2–10 keV) and total collected counts per source (in 0.5–10 keV) distributions are shown in Figs. 3 and 4, respectively.

4.2. Automated spectral-fit models

thumbnail Fig. 3

Observed 2–10 keV flux distribution (in cgs units) for the full XMM/SDSS spectroscopic sample (filled histogram), the sample of highly absorbed candidates (empty histogram), and the highly absorbed sources (line-shaded histogram, see text).

Open with DEXTER

The spectral models implemented for the spectral-fit database were modified to include the effects of redshift and Galactic absorption. Spectral-fitting results corresponding to all sources within the XMM/SDSS-DR7 cross-correlation detected in the hard (2–8 keV) band (~14 000 sources) can be accessed at the spectral database web page. Although many sources within the 3XMM-DR4 catalog have been observed multiple times, the spectral-fitting pipeline has been designed to fit each observation separately. Therefore, automated spectral-fitting results are available for each individual observation of each source.

For the purpose of this work, four new models were implemented by adding a narrow Gaussian emission line to the absorbed power-law model and the three complex models. This line is intended to represent Fe Kα emission line, the most commonly observed emission line in AGNs X-ray spectra. For spectra with fewer than 100 counts above 2 keV, the line energy and width were fixed to 6.4 keV, and 0.1 keV, respectively. For a larger number of counts, the central energy was allowed to vary within the 6.3–6.9 keV range. Models including this emission line were considered as complex models, and as such, they were only applied to sources with more than 500 counts in at least one observation.

4.3. Selection of highly absorbed candidates

The automated spectral-fitting pipe-line, given the limited number of spectral models that have been implemented, and the lack of goodness of fit, was not designed to decide which model was the best-fit model among those tried for each observation, but to obtain the best possible representation of the spectral shape. Nevertheless, acceptable fits were found for ~80% of the sources in the full XMM-Newton spectral-fit database, in terms of goodness values. This is not often the case for highly absorbed sources. This type of AGNs usually displays complex X-ray spectra that are only poorly fitted by using the rather simple models used for the automated fits. However, the automated fitting results can be used as a proxy of the actual spectral shape, even if the model is an unacceptable fit, and that information can be used to select highly absorbed sources.

To obtain the most reliable selection technique, four different automated selection criteria were explored that accommodate the spectral characteristics most often shown by highly absorbed AGNs. In all cases, and to be able to apply our method to all the sources in the sample with enough number of counts, automated spectral-fitting results were used regardless of the goodness of fit for the model under consideration.

thumbnail Fig. 4

Distribution of total counts per source in the 0.5–10 keV band for the full XMM/SDSS spectroscopic sample (filled histogram), the sample of highly absorbed candidates (empty histogram), and the highly absorbed sources (line-shaded histogram, see text).

Open with DEXTER

A source was considered a highly absorbed candidate if its automated spectral-fitting results fulfilled any of the following criteria (at least in one observation, for sources with multiple observations):

  • 1.

    FLATH sample (67 sources), flat spectrum in the hard (2–10 keV) rest-frame band: We selected sources with a measured photon index in the 2–10 rest-frame band <1.4 at the 90% confidence level, excluding absorption. This photon index was derived during the automated spectral fits as part of the absorbed power-law model.

  • 2.

    FLATA sample (33 sources), flat spectrum in the total (0.5–10 keV) band: We selected sources with a measured photon index <1.4 at the 90% confidence level from the absorbed power-law fit.

  • 3.

    HABS sample (32 sources), highly absorbed sources (intrinsic NH> 5 × 1023 cm-2) from the absorbed power-law fit. We did not take into account the errors in this case since, as pointed out before, highly absorbed sources are usually not acceptably fitted by a simple absorbed power-law model, but we can use a high column density value as an indication of actual heavy absorption or a complex spectrum. Therefore, it is important to remember that this does not mean that the source is actually highly absorbed, since we are not taking the goodness of this fit into account.

  • 4.

    HEW sample (16 sources), large equivalent width sources: We selected spectra for which the best-fit model (the one with the lowest value of goodness) includes a line with equivalent width (EW) > 500 eV at the 90% confidence level. Note that this selection criterion was only applied to sources with more than 500 counts collected in their X-ray spectra, 515 sources in the full sample.

With the exception of the Fe Kα emission line, all the other selection criteria were chosen so as to pinpoint highly obscured sources by using a very simple spectral model (an absorbed power law, wabs*zwabs*pow in XSPEC notation), and therefore, they can be applied to all sources within our sample with the only limitation that the number of counts is >50 in at least one detector.

Since the four samples of highly absorbed candidates have sources in common (see Fig. 5), the final sample is composed of 81 sources. Information about the observations available in 3XMM-DR4 for these sources is listed in Table 4. For sources with multiple observations, all available observations are listed, even if the automated spectral-fitting results classified the source as a highly absorbed candidate in just one of the observations.

thumbnail Fig. 5

Venn diagram for the four samples of highly absorbed candidates. Numbers between parenthesis correspond to confirmed highly absorbed AGNs according to manual spectral fits (see Sect. 5).

Open with DEXTER

5. Manual testing of the automated selection criteria

After the automated selection, all the available observations within 3XMM-DR4 of the 81 highly absorbed candidates were manually analyzed to check if the sources were in fact highly absorbed (NH> 1023 cm-2). To check for source variability between different observations, the values of the photon index, intrinsic absorption and fluxes were compared by fitting each observation separately. Spectra for the same source and EPIC instrument corresponding to different observations were merged if the parameter values and fluxes were consistent within errors at the 90% confidence level.

The models applied during the manual fits are more complex than the rather simple ones used in the construction of the database. For example, to model the soft emission in highly oscured AGNs, we tied together both photon indices when fitting a double power-law model, whereas they are allowed to have different values in the database. In this way, we can separate a scattering component, the soft power-law with the same photon index as the hard one, from an additional thermal component(s), if present. This component(s) can become important in highly obscured AGNs at low luminosities because the contribution of the host galaxy to the soft band becomes more and more significant.

The fit statistic used was C-stat, and the models applied during the manual fits are listed below. Since targets were not excluded from our parent sample of 1015 AGNs, many of the 81 candidates have been previously analyzed in detail and the results published in the literature, by using more complex models than those we use here. We restricted our spectral fits to our limited set of models to keep our analysis consistent. A source was considered as highly absorbed after the manual fits if the resulting intrinsic column density was higher than 1023 cm-2 at the 90% confidence level.

  • PL (wabs*zwabs*pow): photoelectrically absorbed power law. The absorption components wabs and zwabs (also applicable to the following models) correspond to absorption fixed to the Galactic column density at the source coordinates, and absorption shifted to the source redshift, respectively.

  • PL+L (wabs*zwabs*(pow+zgaus)): same as PL, plus a narrow emission line whose central energy is fixed to 6.4 keV rest-frame, and its width to 0.1 keV.

  • WAPL (wabs*absori*pow): same as PL, but in this case the redshift absorber is ionized, usually called a warm absorber.

  • 2WAPL (wabs*absori*absori*pow): same as WAPL, but including an additional ionized absorber.

  • 2WAPL+L (wabs*absori*absori*(pow+zgaus)): same as 2WAPL, plus a narrow emission line defined as in PL+L.

  • WAPL+L (wabs*absori*(pow+zgaus)): same as WAPL, plus a narrow emission line defined as in PL+L.

  • PL+T (wabs*(zwabs*pow+mekal)): same as PL, plus a thermal component.

  • PL+T+L (wabs*(zwabs*(pow+zgaus)+mekal)): same as PL+T, plus a narrow emission line defined as in PL+L.

  • WAPL+T+L (wabs*(absori*(pow+zgaus)+mekal)): same as PL+T+L, but substituting the neutral absorber by an ionized one.

  • PL+R+L (wabs*(zwabs*(pow+zgaus)+pexrav)): same as PL+R, plus a narrow emission line defined as in PL+L.

  • PL+T+R+L (wabs*(zwabs*(pow+zgaus)+pexrav+mekal)): same as PL+R+L, plus a thermal component.

  • 2PL (wabs*(zwabs*pow+pow)): double power-law model in which the photon indices of both power-law components are tied to the same value. The scattered fraction, that is, the fraction of the intrinsic (power law) emission that is scattered into our line of sight by the absorbing material, is estimated by obtaining the ratio between the normalizations of both power laws.

  • PCPL (wabs*zpcfabs*pow): this is in fact functionally the same model as 2PL, but we used this one instead if the scattering fraction for the 2PL model was >10%, which indicates that the soft power-law component is not scattered emission, but transmitted emission.

  • PCPL+L (wabs*zpcfabs*(pow+zgaus)): same as PCPL, plus a narrow emission line defined as in PL+L.

  • PCPL+T (wabs*zpcfabs*(pow+mekal)): same as PCPL, plus a thermal component.

  • 2PL+L (wabs*(zwabs*(pow+zgaus)+pow)): same as 2PL, plus a narrow emission line defined as in PL+L.

  • 2PL+T (wabs*(zwabs*pow+mekal+pow)): same as 2PL, plus a thermal component.

  • 2PL+T+L (wabs*(zwabs*(pow+zgaus)+mekal+pow)): same as 2PL+T, plus a narrow emission line defined as in PL+L.

  • 2PL+2T+L (wabs*(zwabs*(pow+zgaus)+mekal+mekal+pow)): same as 2PL+T+L, plus an additional thermal component.

Of the 81 highly absorbed candidates, 28 sources display large amounts of absorption in their X-ray spectra. The number of sources that are best-fitted by each of the models described above are listed in Table 1. The manual spectral-fitting results, corresponding to the 28 highly absorbed AGNs, are shown in Table 2. In some cases, the addition of ionized absorption (modeled as absori in XSPEC) is necessary to obtain an acceptable fit. These sources are not considered as highly absorbed in this analysis regardless of the column density of the ionized absorber. In addition, if the resulting scattering fraction in a double power-law model is 10%, we assumed that the model represents partial covering absorption, which means that the soft power-law would correspond to transmitted emission instead of scattered emission. Therefore, these sources were not considered as highly absorbed AGNs either, again regardless of the column density of the partial covering absorber. The manually derived column density distribution for the 81 candidates is plotted in Fig. 6, and the column density values versus the Fe Kα line EW values (in the 47 cases in which the line was detected) are plotted in Fig. 7. We find a similar result as reported in Fukazawa et al. (2011). By using Suzaku9 data, the authors reported a positive correlation between the line EW and the measured column density, but only for high column densities.

Table 1

Number of AGNs manually best-fitted by each model.

thumbnail Fig. 6

Column density distribution from the manual analysis for the 81 highly absorbed candidates: the line-shaded histogram correspond to the 21 CT plus near-CT AGNs, filled histogram to the 7 highly absorbed but not CT or near-CT AGNs, and the empty histogram to the remaining AGNs.

Open with DEXTER

thumbnail Fig. 7

Column density versus Fe Kα EW for the candidates for which the emission line is detected. Filled circles correspond to the confirmed highly absorbed AGNs, empty circles to the remaining AGNs.

Open with DEXTER

We considered a source as a CT candidate if the value of NH was consistent with being higher than 1024 cm-2, and/or the value of the Fe Kα line EW was higher than 500 eV. These sources, 15 CT candidates, are marked in boldface in Table 2, and their spectra are plotted in Fig. 8. Only in two cases (3XMMJ131104.6+272806, and 3XMMJ093551.5+612111) were we able to measure a column density higher than 1024 cm-2. In four cases (3XMMJ091804.2+514113, 3XMMJ093952.7+355358, 3XMMJ140700.3+282714, and 3XMMJ215649.5-074531), we estimate that the column density is likely higher than 1024 cm-2 because the X-ray spectrum is reflection dominated, meaning that there is no sign of direct emission. In five cases, the measured column density is lower than 1024 cm-2, but consistent with this value at 90% confidence level. In the remaining four cases, the upper limit at 90% confidence for the column density is lower than 1024 cm-2, but these sources display an Fe Kα line with an EW> 500 eV, therefore we also considered them as a CT candidates. A high value of the Fe Kα line EW, along with a very flat hard spectrum, is a characteristic of reflection-dominated spectra. We cannot exclude that the sources displaying high EW values are in fact reflection dominated. However, in all cases, the hard photon indices are either too steep or become steeper if they are left free to vary (in the cases in which the photon index is fixed to 1.9 to constrain the intrinsic absorption), which suggests that the hard emission is direct emission. We finally classified six additional sources as near-CT AGNs. These six AGNs do not fulfill our CT criteria, but their measured column densities are very high, NH ≳ 5 × 1023 cm-2.

Table 2

Manual spectral-fitting results: highly absorbed sources with NH> 1023 cm-2.

thumbnail Fig. 8

Unfolded spectra and model, and data to model ratio for the Compton-thick candidates.

Open with DEXTER

6. Discussion

6.1. Automated selection reliability

The automated selection method presented in this paper can be applied to X-ray spectral data down to 50 source counts. Given the relatively high flux limit of our test sample (typical fluxes are FX(2–10 keV) ~10-13 erg cm-2 s-1), most of our absorbed sources lie at low redshifts (z ⟩ = 0.15), with the exception of 3XMM141546.2+112943 at z = 2.56, which is a lensed QSO, the cloverleaf quasar H1413+117. A much more detailed analysis of this source is presented in Chartas et al. (2007), by using XMM-Newton and Chandra data, and in which ionized absorption and disk reflection are included in the spectral fit. Since the only limitation to the applicability of this method seems to be the number of collected counts in the X-ray spectra, it could be applied to deeper surveys, and thus, to the selection of highly obscured AGNs at higher redshifts.

The fraction of near-CT plus CT candidates (21 AGNs with column densities 5 × 1023 cm-2) in the FLATH, FLATA, HABS, and HEW samples is 31% (12/67), 12% (4/33), 47% (15/32), and 44% (7/16), respectively. All the sources with manually computed column densities higher than 1023 cm-2, including the near-CT, and CT candidates, belong to the FLATH sample. The best selection criterion is a combination of different automated selection criteria. For all sources, regardless of the spectral quality, the best selection criterion is to belong to both the FLATH and the HABS samples (FLATH+HABS subsample), 80% of these sources are absorbed by column densities higher than 1023 cm-2. For sources with more than 500 counts, that is, sources for which a selection according to the emission line EW is possible, the best selection criterion is to belong to the HEW and the FLATH samples (FLATH+HEW subsample). Again, 80% of these sources are absorbed by column densities higher than 1023 cm-2 according to the manual fits. Out of the 28 highly absorbed AGNs, 23 belong to either or both of these subsamples.

The sources within the automated selected samples that after the manual spectral fit were found to be not highly absorbed, display one or more of the following spectral characteristics:

  • A reflection component, with spectral parameters consistent with those of unabsorbed type 1 AGNs, which produces a flat photon index. This feature, combined with a low number of counts at high energies, can also mimic a high EW line.

  • Significant amount of absorption, but lower than 1023 cm-2 (~several times 1022), totally or partially covering the central source.

  • Significant amount of ionized absorption.

  • Only in some cases, we found that a low number of counts plus an extremely complex spectral shape classified the source in the HABS sample. The lower number of counts prevented us from using automated complex fits, and the simple power-law fit is an extremely poor fit and returns a high value for the column density.

6.2. Compton-thick candidates and previous results

We searched the literature for previous classifications of our 28 highly absorbed sources and found a very good agreement for most cases. When previously reported intrinsic column density or Fe Kα EW values were available, the source was classified according to our CT candidate definition. In three cases, a previous estimate or measurement of the column density could not be found. Out of the 15 sources that we found to be consistent with being a CT AGNs, 12 were previously reported as CT AGNs based on different techniques (see Table 3). Our classification disagrees with previous results in only three cases:

  • 3XMMJ091804.2+514113: while we classified this source as a reflection-dominated CT candidate, an NH value of ~ 6 × 1022 is reported in Georgalakis et al. (2006; although consistent with being highly absorbed within errors). The same XMM-Newton observation was used in both analyses but because there are very few spectral counts, the differences in the spectral-fitting results might be due to the different spectral extraction. Nevertheless, our spectral data display an extremely flat shape that is not consistent with mild absorption.

  • 3XMMJ093952.7+355358: if we fit the same best-fit model as in Hardcastle et al. (2006), that is, a double power-law plus an emission line, both best-fit NH values are consistent within errors, but theirs is lower, probably because the second power-law photon index was left free to vary, which resulted in a very flat slope (Γ ~ 0.5). Using this model, the source would be also classified as a CT candidate in the current work based solely on the Fe Kα line EW, but we were unable to find the relevant value in Hardcastle et al. (2006). Given the extremely flat hard spectrum, we classify this source as a reflection-dominated CT AGNs in this work.

  • 3XMMJ122546.7+123943: as for the previous source, the differences between our classification and that reported in Burlon et al. (2011) is caused by the EW of the iron line, which is not reported in Burlon et al. (2011). Moreover, this is a highly variable AGNs, and a strong Fe Kα line is only found in one out of the three observations used in this work.

We also searched the literature for CT AGNs within our parent sample of 1015 AGNs that could have been missed as such by our automated method. Only one AGNs from the parent sample is included within the local CT AGNs reported in (Brightman & Nandra 2011), NGC 3690, which was classified as a CT AGNs because of a high EW Fe Kα emission line. The automated analysis detected an Fe Kα emission line with an EW> 500 eV, but consistent with being lower than that value at the 90% confidence level. Therefore, it was not flagged as a highly absorbed candidate by our selection criteria. Of the sources in out parent sample that are classified as CT AGNs within the sample of type 2 Seyfert galaxies extracted from the SDSS presented in LaMassa et al. (2009), our classification agrees in all but two cases: SDSS J080359.20+234520.4, and SDSS J112301.31+470308.6. In both cases, the automated fits return very flat photon indices in the hard band, but because of the very low number of counts in that band, consistent with being larger than 1.4 at the 90% confidence level. As a result, these two sources are not flagged as candidates by our selection criteria. It is important to note that these two sources are classified as CT AGNs in LaMassa et al. (2009) based on their X-ray luminosity to optical, and mid-IR luminosity ratios, and not from their X-ray spectral analysis. Finally, we cross-correlated our parent sample with the CT AGNs within the two type 2 QSO samples that were also extracted from the SDSS and reported in Vignali et al. (2010), and Jia et al. (2013). Our classification agrees with that in those works except in one case: SDSS J091345.48+405628.2. Nevertheless, as pointed out in Jia et al. (2013), its X-ray spectrum is complex and dominated by soft emission, and in addition, different X-ray spectral analysis of this source, both using XMM-Newton and Chandra data, have been published reporting a non-CT classification in some cases.

Table 3

Candidate Compton-thick AGNs.

6.3. Hardness ratios versus automated spectral-fit selection

Rest-frame hardness ratios (or X-ray colors) have been proposed by several other studies as an alternative to manual spectral fitting for the selection of highly absorbed sources. The downside of these methods is that to obtain rest-frame colors (or fluxes) from X-ray count rates, a spectral model has to be assumed. This could strongly decrease the accuracy of the selection technique, especially if the assumed spectral model is a poor representation of the actual spectral shape. To compare our proposed technique with color-selection techniques, X-ray colors were computed for our sample following the two different X-ray color selection techniques presented in Brightman & Nandra (2012) and Iwasawa et al. (2012).

Brightman & Nandra (2012) presented an X-ray color selection calibrated by using rest-frame fluxes derived from best-fit models and manual spectral fits. To this end, they used XMM-Newton spectra for a sample of 126 local AGNs, extracted from a parent sample selected in the mid-IR, for which they carried out a detailed X-ray spectral analysis. Brightman & Nandra (2012) defined two X-ray colors, HR1 and HR2, based on rest-frame fluxes computed in the rest frame bands: 1–2 keV (band 1), 2–4 keV (band 2), and 4–16 keV (band 3), as follows: (1)First, we applied this method to our sample of 81 highly absorbed candidates by using the best-fit model obtained from the manual fits. The results are plotted in Fig. 9 in the top-left panel. The dashed line corresponds to the proposed dividing line between highly absorbed sources (NH> 1023 cm-2) and mildly absorbed or unabsorbed sources in Brightman & Nandra (2012). The solid line corresponds to the wedge defined in that work to contain all their CT AGNs. We found similar results, that is, all highly absorbed sources in our sample but one lie above the dashed line. Nevertheless, we found a higher number of contaminants with lower NH of ~20%, while a value of only 7% was reported in Brightman & Nandra (2012). Moreover, not all our CT candidates fall within the solid wedge.

Brightman & Nandra (2012) also proposed that their selection technique could be applied to X-ray colors derived by using observed count rates, assuming a simple power-law model (Γ = 1.4), and using the HEASARC portable interactive multimission software (PIMMS) to derive the rest-frame fluxes. However, when we applied that method to obtain the X-ray colors (Fig. 9: top right panel), the contamination of the highly absorbed region by sources with lower NH increases dramatically. As a comparison, we derived the X-ray colors by using the best-fit model from the automated spectral fits (Fig. 9: bottom left panel). We found a lower contamination of the highly absorbed AGNs area by unabsorbed AGNs, but still quite high.

thumbnail Fig. 9

X-ray colors as in Brightman & Nandra (2012). Different symbols correspond to column densities from manual spectral fits: CT candidate (filled squares), near-CT (filled triangles), log (NH) > 23 (empty squares), 22 < log (NH) < 23 (filled circles), and log (NH) < 22 (empty circles). Top-left: X-ray colors derived from manual spectral analysis. Top-right: colors obtained by using HEASARC PIMMS (see text for details). Bottom-left: colors obtained by using automated spectral fits. Bottom-right: colors obtained by using automated spectral fits for the sources not belonging to the sample of highly absorbed candidates.

Open with DEXTER

thumbnail Fig. 10

X-ray colors as in Iwasawa et al. (2012). Different symbols correspond to column densities from manual spectral fits: CT candidate (filled squares), near-CT (filled triangles), log (NH) > 23 (empty squares), 22 < log (NH) < 23 (filled circles), and log (NH) < 22 (empty circles). Top-left: X-ray colors derived from manual spectral analysis. Top-right: colors obtained by using HEASARC PIMMS (see text for details). Bottom-left: colors obtained by using automated spectral fits. Bottom-right: colors obtained by using automated spectral fits for the sources not belonging to the sample of highly absorbed candidates.

Open with DEXTER

As a final step, we derived X-ray colors by using automated fits for the sources in the full sample that were not flagged as highly absorbed candidates by the automated selection (Fig. 9: bottom right panel) to check whether we might be missing a significant number of highly absorbed AGNs. We found that only 10% of the not selected sources lie above the dashed line. In Brightman & Nandra (2012), most of the low NH contaminants of that region were sources with a complex spectrum. In our case, since manual fit results are not available for the full sample, we checked for the best-fit automated model for these sources. In 95% of the cases, the preferred model was in fact a complex one. Therefore, these sources are probably not highly absorbed sources missed by our selection criteria, but sources showing a complex X-ray spectrum.

Iwasawa et al. (2012) presented a different method based on observed count rates instead of rest-frame fluxes, and designed to efficiently select highly absorbed sources at redshifts higher than 1.7. The sample used to test this method was composed of 47 AGNs detected at high significance in the XMM-CDFS, and with either photometric or spectroscopic redshifts available. They also compared their results with those obtained by using high-energy data (>10 keV) for a very small sample of local and well-known AGNs. The method presented in Iwasawa et al. (2012) is based on the use of observed count rates in the following rest-frame energy bands: 3–5 keV (band s), 5–9 keV (band m), and 9–20 keV (band h). They defined two colors as s/m and h/m. Since most of our highly absorbed candidates are at low redshift, these three energy bands are not covered by most of our XMM-Newton data. To be able to apply a method as similar to that in Iwasawa et al. (2012) as possible, we used rest-frame fluxes instead of count rates. As for the X-ray colors in Brightman & Nandra (2012), a model has to be assumed to obtain the s/m and h/m colors in this case.

We performed the same comparison as for the colors in Brightman & Nandra (2012). The results are plotted in Fig. 10. First, we used the best-fit model from the manual fits to recover the X-ray colors (Fig. 10: top left panel). The dashed lines limit the different regions according to the expected column densities. Regions U (unabsorbed), M (modestly absorbed), A (absorbed), and V (very absorbed) correspond to sources with typical log NH lower than 22, 22.7, 23.4, and 23.8, respectively. We drew a tentative limit (solid line) that separates sources with NH> 1023 cm-2 in our sample. This method very efficiently separates moderate or unabsorbed AGNs from highly absorbed AGNs. However, as for the Brightman & Nandra (2012), and the automated selection criteria, we cannot separate highly absorbed sources from CT candidates. We repeated the same exercise by using PIMMS (assuming a power-law model with photon index equal to 1.4, Fig. 10: top right panel), and by using the automated spectral-fitting results (Fig. 10: bottom left panel). The contamination of low NH sources is similar in both cases. In the automated fits, highly obscured AGNs occupy a broader region of the plot, while for PIMMS colors, the highly obscured region is contaminated by unabsorbed AGNs. The method of Iwasawa et al. (2012) seems to separate AGNs with different amounts of intrinsic absorption much better than that presented in Brightman & Nandra (2012).

Finally, we applied this method to the full sample by using the best-fit model from the automated fits (Fig. 10: bottom right panel). All the sources not classified as highly absorbed candidates by our automated criteria fall outside the highly absorbed AGNs region.

7. Conclusions

We have derived X-ray spectral fits for very many 3XMM-DR4 sources (~77 000) that contain more than 50 photons per detector (Corral et al. 2014). Here, we used a subsample of 1000 AGNs in the common SDSS and 3XMM area (covering 120 deg2) with spectroscopic redshifts available. We searched for highly obscured AGNs by applying an automated selection technique based on an automated X-ray spectral analysis. In particular, the selection was based on the presence of a) flat spectra with a photon index lower than 1.4 at the 90% confidence level in the 2–10 keV rest-frame spectra; b) flat spectra with a photon index lower than 1.4 at the 90% confidence level in the 0.5–10 keV observed spectra; c) an absorption turnover, indicative of a high rest-frame column density; or d) an Fe Kα line with a large equivalent width (>500 eV). We found 81 candidate highly obscured sources. Subsequent detailed manual spectral fits revealed that 28 are heavily obscured with a column density of NH> 1023 cm-2. Of these 28 sources, six are near-CT AGNs with a column density of NH ~ 5 × 1023 cm-2. Finally, 15 are candidate CT AGNs on the basis of either a high column density, consistent within the 90% confidence level with 1024 cm-2, or a large equivalent width (>500 eV) of the Fe Kα line.

Our automated method is very efficient in selecting highly absorbed AGNs (NH> 1023 cm-2), with a successful rate of 80%:

  • For low-quality spectra, and by using only results from a simple absorbed power-law fit, 80% (20 out of 25 AGNs) of the sources whose X-ray spectra were flagged as flat in the 2–10 rest-frame band and for which the automatically derived column density was higher than 5 × 1023 cm-2 were highly absorbed, as tested by using manual spectral fits.

  • For medium- to high-quality spectra, 80% (7 out of 9 AGNs) of the sources with an automatically detected high EW Fe Kα line, plus a flat continuum in the 2–10 keV rest frame band were highly absorbed, as tested by using manual spectral fits.

We compared our results with rest-frame color CT AGNs selection techniques developed by Brightman & Nandra (2012)

and Iwasawa et al. (2012). The method of Iwasawa et al. (2012), modified by using a spectral model to obtain rest-frame fluxes, was the best for separating highly absorbed (NH> 1023 cm-2) from moderately to unabsorbed sources.


7

For a detailed description of the catalog and the spectral extraction see http://xmmssc-www.star.le.ac.uk/Catalogue/3XMM-DR4/

8

A detailed description of the database and the different spectral models applied is presented on the project web page: http://xraygroup.astro.noa.gr/Webpage-prodec/index.html

Acknowledgments

Based on observations obtained with XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA Member States and NASA. A. Corral acknowledges financial support by the European Space Agency (ESA) under the PRODEX program. P. Ranalli and E. Koulouridis acknowledge financial support from the “Support to Postdoctoral Researchers” projects PE9-3493 and PE9-1145 which are jointly funded by the European Union and the Greek Government in the framework of the programme Education and lifelong learning. G. Mountrichas and G. Lanzuisi acknowledge financial support from the THALES project 383549, which is jointly funded by the European Union and the Greek Government in the framework of the programme Education and lifelong learning. We thank the referee for providing constructive comments and suggestions that helped to improve this paper.

References

Online material

Table 4

Sample of highly absorbed candidates.

All Tables

Table 1

Number of AGNs manually best-fitted by each model.

Table 2

Manual spectral-fitting results: highly absorbed sources with NH> 1023 cm-2.

Table 3

Candidate Compton-thick AGNs.

Table 4

Sample of highly absorbed candidates.

All Figures

thumbnail Fig. 1

Distribution of normalization ratios for the three EPIC instruments: MOS2-to-MOS1 ratio (empty histogram); pn-to-MOS1 ratio (line shaded histogram); and pn-to-MOS2 ratio (filled histogram).

Open with DEXTER
In the text
thumbnail Fig. 2

Intrinsic 2–10 keV luminosity versus redshift for the full XMM/SDSS spectroscopic sample (gray dots), the sample of the 81 highly absorbed candidates based on the automated spectral-fit selection (circles), and the 28 confirmed highly absorbed sources (filled circles, see text).

Open with DEXTER
In the text
thumbnail Fig. 3

Observed 2–10 keV flux distribution (in cgs units) for the full XMM/SDSS spectroscopic sample (filled histogram), the sample of highly absorbed candidates (empty histogram), and the highly absorbed sources (line-shaded histogram, see text).

Open with DEXTER
In the text
thumbnail Fig. 4

Distribution of total counts per source in the 0.5–10 keV band for the full XMM/SDSS spectroscopic sample (filled histogram), the sample of highly absorbed candidates (empty histogram), and the highly absorbed sources (line-shaded histogram, see text).

Open with DEXTER
In the text
thumbnail Fig. 5

Venn diagram for the four samples of highly absorbed candidates. Numbers between parenthesis correspond to confirmed highly absorbed AGNs according to manual spectral fits (see Sect. 5).

Open with DEXTER
In the text
thumbnail Fig. 6

Column density distribution from the manual analysis for the 81 highly absorbed candidates: the line-shaded histogram correspond to the 21 CT plus near-CT AGNs, filled histogram to the 7 highly absorbed but not CT or near-CT AGNs, and the empty histogram to the remaining AGNs.

Open with DEXTER
In the text
thumbnail Fig. 7

Column density versus Fe Kα EW for the candidates for which the emission line is detected. Filled circles correspond to the confirmed highly absorbed AGNs, empty circles to the remaining AGNs.

Open with DEXTER
In the text
thumbnail Fig. 8

Unfolded spectra and model, and data to model ratio for the Compton-thick candidates.

Open with DEXTER
In the text
thumbnail Fig. 9

X-ray colors as in Brightman & Nandra (2012). Different symbols correspond to column densities from manual spectral fits: CT candidate (filled squares), near-CT (filled triangles), log (NH) > 23 (empty squares), 22 < log (NH) < 23 (filled circles), and log (NH) < 22 (empty circles). Top-left: X-ray colors derived from manual spectral analysis. Top-right: colors obtained by using HEASARC PIMMS (see text for details). Bottom-left: colors obtained by using automated spectral fits. Bottom-right: colors obtained by using automated spectral fits for the sources not belonging to the sample of highly absorbed candidates.

Open with DEXTER
In the text
thumbnail Fig. 10

X-ray colors as in Iwasawa et al. (2012). Different symbols correspond to column densities from manual spectral fits: CT candidate (filled squares), near-CT (filled triangles), log (NH) > 23 (empty squares), 22 < log (NH) < 23 (filled circles), and log (NH) < 22 (empty circles). Top-left: X-ray colors derived from manual spectral analysis. Top-right: colors obtained by using HEASARC PIMMS (see text for details). Bottom-left: colors obtained by using automated spectral fits. Bottom-right: colors obtained by using automated spectral fits for the sources not belonging to the sample of highly absorbed candidates.

Open with DEXTER
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.