A spectroscopic survey of X-ray-selected AGNs in the northern XMM-XXL field

Abstract

This paper presents a survey of X-ray-selected active galactic nuclei (AGNs) with optical spectroscopic follow-up in a ∼ 18 deg² area of the equatorial XMM-XXL north field. A sample of 8445 point-like X-ray sources detected by XMM–Newton above a limiting flux of |$F_{\rm 0.5{\rm -}10\, keV} > 10^{-15} \rm \,erg\, cm^{-2}\, s^{-1}$| was matched to optical (Sloan Digital Sky Survey, SDSS) and infrared (IR; WISE) counterparts. We followed up 3042 sources brighter than r = 22.5 mag with the SDSS Baryon Oscillation Spectroscopic Survey (BOSS) spectrograph. The spectra yielded a reliable redshift measurement for 2578 AGNs in the redshift range z = 0.02–5.0, with 0.5-2 keV luminosities ranging from 10³⁹-10⁴⁶ erg s^{− 1}. This is currently the largest published spectroscopic sample of X-ray-selected AGNs in a contiguous area. The BOSS spectra of AGN candidates show a distribution of optical line widths which is clearly bimodal, allowing an efficient separation between broad- and narrow-emission line AGNs. The former dominate our sample (70 per cent) due to the relatively bright X-ray flux limit and the optical BOSS magnitude limit. We classify the narrow-emission line objects (22 per cent of the full sample) using standard optical emission line diagnostics: the majority have line ratios indicating the dominant source of ionization is the AGN. A small number (8 per cent of the full sample) exhibit the typical narrow line ratios of star-forming galaxies, or only have absorption lines in their spectra. We term the latter two classes ‘elusive’ AGN, which would not be easy to identify correctly without their X-ray emission. We also compare X-ray (XMM–Newton), optical colour (SDSS) and and IR (WISE) AGN selections in this field. X-ray observations reveal, by far, the largest number of AGN. The overlap between the selections, which is a strong function of the imaging depth in a given band, is also remarkably small. We show using spectral stacking that a large fraction of the X-ray AGNs would not be selectable via optical or IR colours due to host galaxy contamination. A substantial fraction of AGN may therefore be missed by these longer wavelength selection methods.

1 INTRODUCTION

A fundamental question in current astrophysics research on active galactic nuclei (AGNs) is the origin and evolution of supermassive black holes (SMBHs) as well as their interaction with their host galaxy. A large body of evidence suggests the presence of SMBHs in nearly all local spheroids (e.g. Magorrian et al. 1998; Kormendy & Ho 2013 and references therein) and argues that their growth occurs predominantly through accretion processes (e.g. Soltan 1982; Marconi et al. 2004). In order to study the evolution and properties of SMBH, one should approach them by a preferably unbiased data set of AGNs. The main limiting factor in AGN selection is the diversity of their spectral energy distributions (SEDs). This depends on the presence of dust and gas clouds along the line of sight, the contrast between accretion-luminosity and stellar emission from the host galaxy, and the detailed physics of the accretion process itself. Many methods have been developed to account for these factors using different wavelength bands and both photometric as well as spectroscopic selection criteria, but they are always, by construction, subject to selection effects.

The accretion processes in SMBHs are highly complex and emit at different wavelengths. In the optical/UV spectra of AGNs, there is the characteristic ‘blue bump’ which is supposed to originate from a thermally radiating accretion disc. This feature allows AGN to be selected via their blue colours, and has been extensively used (see Boyle et al. 2000; Richards et al. 2009; Ross et al. 2012 and references therein). This selection has its limitations because the galaxy dilution as well as the optical obscuration suppress the blue bump. It is therefore only applied to optically point-like sources.

X-rays directly trace highly energetic processes occurring in the neighbourhood of the SMBH (Brandt & Hasinger 2005; Brandt & Alexander 2015). Thus, X-ray selection is highly efficient and provides a clean separation from the host galaxies which typically emit at lower fluxes. However, the most heavily obscured, Compton-thick AGNs are opaque even to X-rays. In this case, infrared (IR) AGN selection methods, tracing the reprocessed nuclear emission, benefit from the clearly distinct SED shapes of AGNs and galaxy components in this wavelength range. Therefore, IR is able to detect heavily obscured AGNs (Stern et al. 2005, 2012; Lacy et al. 2007; Assef et al. 2013; Messias et al. 2014).

Another very well-known AGN selection criterion is the optical narrow-emission line diagnostic (Baldwin, Phillips & Terlevich 1981; Ho, Filippenko & Sargent 1997; Kauffmann et al. 2003; Kewley et al. 2006; Lamareille 2010). It uses the fact that ionizing radiation from the AGN is harder than that of stars, which has an effect on the measured emission line ratios. However, this method is claimed to suffer from (i) contamination from the host galaxy which can potential swamp AGN signatures and (ii) the fact that the method selects a large fraction of Low Ionization Narrow Emission Region galaxies (LINERS) whose nature is still debated (e.g. Yan & Blanton 2012).

Broad-emission lines (⁠|$\rm FWHM>1000\,\rm km\,s^{-1}$|⁠) are unique spectral features which identify optical unobscured AGNs, where the observer has a direct view on the fast moving clouds close to the black hole. Purely board line selected samples of AGNs have been used for example in the VVDS and zCOSMOS surveys (e.g. Gavignaud et al. 2006; Trump et al. 2009).

It is essential to understand the selection function of the different methods for an unbiased census of AGNs across cosmic time. Different selection methods may not only depend on obscuration or host galaxy contamination, but on other fundamental parameters of the active black hole, such as Eddington ratio (i.e. the ratio between observed accretion rate and the Eddington rate), instantaneous accretion luminosity, black hole mass, etc.

Several works (Hickox et al. 2009; Yan et al. 2011; Donley et al. 2012; Mendez et al. 2013) combined and compared multiple selection methods, such as X-ray, UV, optical and IR. These studies have been performed in areas benefiting from synergies of deep X-ray coverage, multiwavelength observations and spectroscopic follow-up (Brandt & Alexander 2015), e.g. the COSMOS and CDFS field (Alexander et al. 2003; Brusa et al. 2010; Luo et al. 2010; Civano et al. 2012) and Lockman Hole (Mateos et al. 2005). However, most area surveys are limited to few |$\deg ^{2}$| or less. Larger area surveys, such as Boötes (Hickox et al. 2009), achieve better statistics for rare objects, e.g. luminous AGNs, and improve clustering studies, but have been difficult to carry out, mainly due to the small field of view of X-ray imaging telescopes.

In this work, we present one of the largest spectroscopic surveys of X-ray-selected AGN (XMM–Newton) in a homogenous and contiguous area with the same instrument (BOSS spectrograph of the Baryon Oscillation Spectroscopic Survey). This data set allows for a more detailed analysis of properties of X-ray-selected AGNs, enables a comparison of selection methods for AGNs based on different wavelengths, and provides a forecast for the AGN population in the extended ROentgen Survey with an Imaging Telescope Array (eROSITA) survey and its follow-up programme SPIDERS (Spectroscopic Identification of eROSITA Sources; Merloni et al. 2012; Predehl et al. 2014, Clerc et al., in preparation; Dwelly et al., in preparation). We chose the XMM-XXL north field (PI M. Pierre), which is an extension of the former XMM-LSS survey (Pierre et al. 2004; Chiappetti et al. 2013; Clerc et al. 2014) and was conceived to study the large-scale structure in the Universe by mapping a well-defined statistical sample of galaxy clusters. It provides access to large subsamples of AGNs when selected by luminosity, morphology or obscuration. Furthermore, many multiwavelength data sets in optical (e.g. CFHTLS, SDSS), IR (e.g. UKIDSS, WISE, VIDEO), UV (e.g. GALEX) and radio (e.g. VLA) bands are available in this field. Our reduction of the XMM-XXL area ( ∼ 22 deg²) yields in a flux limit of |$F_{0.5{\rm -}2\,\rm {keV}} \sim 3.5\times 10^{-15} \,\rm {erg\,cm^{-2}\,s^{-1}}$| for 50 per cent of the area.

The paper is organized as follows: Section 2 introduces the multiwavelength data sets from XMM–Newton, SDSS, BOSS and WISE, along with the cross-matching and the spectroscopic target selection. Section 3 presents the redshift determination, as well as the classification of the BOSS spectroscopic follow-up data. An overview of the spectroscopic redshift and luminosity properties of the X-ray-selected AGNs is given in Section 4. In Section 5, we compare optical and IR AGN selection criteria and evaluate them using our sample of X-ray-selected AGNs. We discuss the survey analysis in Section 6 and provide a forecast for eROSITA and SPIDERS in Section 7. Finally, we summarize the results in Section 8. In this paper, we use the J2000 standard epoch and adopt a cosmology with H₀ = 70  km s^{− 1} Mpc^{− 1}, Ω_M = 0.27 and Ω_Λ = 0.73. The optical magnitudes from Sloan Digital Sky Survey (SDSS) are given in the AB system and IR magnitudes from WISE are given in the Vega system.

2 DATA SETS AND SPECTROSCOPIC TARGET SELECTION IN XMM-XXL NORTH

In this section, we introduce the multiwavelength data sets in the XMM-XXL north field, the cross-matching of the optical and IR counterparts, and the spectroscopic target selection of our X-ray-selected AGNs. The full catalogue is presented in Appendix E.

2.1 Introduction of imaging data sets

2.1.1 XMM-XXL survey and X-ray source catalogue

The XMM–Newton XXL survey (XMM-XXL, PI Pierre) is a medium-depth (10 ks per pointing) X-ray survey that covers a total area of 50 deg² split into two fields equal in size. In this paper, we focus on the equatorial sub-region of the XMM-XXL (XMM-XXL north), which overlaps with the SDSS-DR8 imaging survey (Aihara et al. 2011). The XMM-XXL north observations were distributed around the area of the original 11 deg² XMM-LSS survey (Clerc et al. 2014) and therefore build upon and extend that sample.

The X-ray data used in this paper are primarily from the XMM-XXL and XMM-LSS surveys. We also include however, any additional XMM–Newton pointings that are contiguous to the area covered by those programmes, such as the XMM–Newton observations of the Subaru/XMM–Newton Deep Survey (SXDS; Ueda 2008). The data reduction, source detection and sensitivity map construction for the X-ray catalogue follow the methods described by Georgakakis & Nandra (2011). Specific detail on the reduction of the XMM–Newton observations in the XMM-XXL north field and X-ray spectroscopic properties are presented in Liu et al. (2015, submitted). In brief, the X-ray data reduction is carried out using the XMM Science Analysis System (sas) version 12. Our survey comprises the XMM-XXL data observed prior to 2012 January 23. At that date, the XMM-XXL programme was partially complete, which results in the inhomogeneous X-ray coverage shown in Fig. 1. In the following, we always refer to this coverage as ‘XMM-XXL north’ area. The catalogue contains in total 8445 sources including 8016 sources with detections in the soft band (F_0.5-2keV), 4802 sources with detections in the hard band (F_2-10 keV) and 8309 sources with detections in the full band (F_0.5-10 keV). Within these three bands, ∼47, ∼50 and ∼45 per cent of the sources have an optical counterpart in the r-band (r < 22.5 mag), respectively. For the total area of our survey with spectroscopic follow-up (18 deg²), the observations reach a flux limit of |$F_{0.5{\rm -}10\,\rm keV} > 7.50 \times 10^{-15} \rm \,erg\, cm^{-2}\, s^{-1}$| for 10 per cent of the area and a flux limit of |$F_{0.5{\rm -}10\,\rm keV}> 1.27 \times 10^{-14} \rm \,erg\, cm^{-2}\, s^{-1}$| for 50 per cent of the area.

2.1.2 Optical source catalogue of SDSS

The XMM-XXL north area is covered by the optical imaging of the third programme of the Sloan Digital Sky Survey (SDSS-III, Eisenstein et al. 2011). This is an optical survey extending over ∼ 14 555 deg² at the ground based 2.5-metre telescope at the Apache Point Observatory, New Mexico (Gunn et al. 2006). The five broad-bands (average wavelength indicated) u [3551 Å], g [4686 Å], r [6165 Å], i [7481 Å], z [8931 Å] (Fukugita et al. 1996) have the following (AB) magnitude limits: 22.0, 22.2, 22.2, 21.3 and 20.5 mag (corresponding to 95 per cent completeness for point sources). We retrieved the SDSS imaging data from the DR8 (Aihara et al. 2011) and obtained 538 508 sources in the XMM-XXL north. In this work, we use the flux measurements from SDSS: psfmag (for point-like sources, |$\rm {type}=6$|⁠) and modelmag (for extended objects, |$\rm {type}=3$|⁠). Both magnitudes are not corrected for extinction. Furthermore, we do not apply any cuts to correct for e.g. blending and moving objects.

2.1.3 IR source catalogue of WISE

The Wide-field Infrared Survey Explorer (WISE) observed the entire sky using a four channel imager (mean wavelength indicated): W1 [3.4 μm], W2 [4.6 μm], W3 [12 μm] and W4 [22 μm] (Wright et al. 2010). For the IR AGN selection criteria in Section 5.2, we used the mpro magnitudes for both point-like and extended objects from the allWISE data release (Wright et al. 2010; Mainzer et al. 2011). We excluded diffraction spikes, persistent sources, haloes, optical ghosts and blended sources (⁠|$\rm {cc\_flag}=0$| in the W1 and W2 band, |$\rm {NB}\le 2$|⁠) following Stern et al. (2012). In the XMM-XXL north, we obtain in total 334 697 IR sources which yielded 321 352 sources after cleaning for photometric failures.

2.2 Cross-matching of data sets

The X-ray sources from the XMM-XXL north survey have been separately cross-matched to the SDSS catalogue and to the allWISE catalogue. This allows an independent comparison of AGN selection criteria with the full multiwavelength data set. The cross-matching was performed using the likelihood ratio method as presented in Georgakakis & Nandra (2011). We apply a lower likelihood ratio limit for the matched catalogues, which allows for an estimation of the spurious identification rate. Any cuts for e.g. photometric errors or flux have been applied after the matching process. Fig. 1 demonstrates the full band X-ray flux histogram for the point-like X-ray sources, their associated counterparts and the followed-up spectra (as described in Section 2.3.2).

5294 X-ray sources have an optical counterpart in the SDSS DR8 (Aihara et al. 2011) inside a maximal search radius of 4 arcsec. Out of those, 4075 sources with a likelihood ratio LR_{XMM, SDSS} > 1.5 are selected as targets for the spectroscopic follow-up with BOSS. The spurious identification rate for these objects is estimated to be about 7 per cent. Fig. 2 shows the r_model magnitude distribution of both optically extended and point-like XMM-SDSS cross-matched sources. The distribution has its maximum at r_model ∼ 22 mag. The plot also shows the 3042 sources which have been targeted by BOSS in our observation programme.

5414 X-ray sources have an IR counterpart in the WISE imaging data base within an maximal matching distance of 5 arcsec. We apply a likelihood ratio threshold of LR_{XMM, WISE} > 0.5, corresponding to a spurious identification rate 5 per cent and retrieve 4844 sources, including 4811 with good WISE photometry. In Fig. 3, we show the W2-band distribution and the histogram of all XMM-WISE cross-matched objects. The distribution has its maximum at W2 ∼ 16 mag. In addition, we indicate the 2474 sources, which have spectroscopic BOSS follow-up, too.

The angular distance between the optical and IR counterpart of X-ray sources gives the probability to be the counterpart of the same source. We assume an upper bound distance of 1 arcsec, which corresponds to the radius of a BOSS fibre and is a very conservative limit for non-blended WISE sources. Out of the 3305 X-ray sources with both SDSS and WISE cross-matches, 13 per cent have counterparts which are separated by more than 1 arcsec. Such a fraction decreases to 4 per cent for distances larger than 2 arcsec. This number is consistent with expectations, given the predicted numbers of spurious associations we estimate from the likelihood ratio thresholds for the IR and optical matches to the X-ray sources (see above).

2.3 Spectroscopic data set

2.3.1 BOSS spectroscopic survey

We performed optical spectroscopic follow-up of the XMM-SDSS matched sources with the BOSS spectrograph of SDSS (Smee et al. 2013). It is a multiple fibre spectrograph using a standard plate of 7 deg² which can host 1000 optical fibres of 2 arcsec diameter and is typically exposed for 4500 s. The covered wavelength range spans λ = 360-1040 nm with an average resolution of R = 2000 and a redshift accuracy of 65 km s^{− 1}. The BOSS spectroscopic targets are limited in their r-band magnitude range. We apply a magnitude cut of 15.0 < r < 22.5 mag. The bright limit is necessary to avoid cross-talk effects in the optical spectrograph, while the faint one is introduced to ensure realistic chances of successfully obtaining a spectrum for strong emission line objects in a typical BOSS exposure.

The fibre allocation of one plate is split into fibres for the dedicated science programme and fibres which are assigned to standard calibration stars and sky calibration targets (n_star ≈ 20 and n_calib ≈ 80) as well as repeated observations of other BOSS programmes. The fibre tiling procedure of the BOSS plates starts with the determination of the largest possible subset of targets which do not collide with each other. In a second step, the remaining fibres are optimally distributed to minimize fibre collisions (Blanton et al. 2003). We can conclude that deselection from the initial target list was only due to random fibre collisions.

2.3.2 Spectroscopic target selection

The spectroscopic follow-up of the 4075 matched XMM-SDSS AGN comes from three different programmes, including two dedicated BOSS ancillary programmes (published as a part of DR12, Alam et al. 2015) and former BOSS DR10 observations (Ahn et al. 2012) in the same region. In the following, we explain the target selection and refer for technical details to Appendix A.

First BOSS ancillary programme. The first group of spectra has been retrieved from the BOSS ancillary programme by PI P. Green and A. Merloni. It was devised to test target selection algorithms and strategies for both the SPIDERS and Time Domain Spectroscopic Survey (TDSS) survey components of the SDSS-IV. TDSS (Morganson et al. 2015) is a follow-up programme of time variable objects (e.g. from PAN-STARRS1), with either single as well as multi-epoch spectroscopy. The dedicated BOSS targeting plate covered 7 deg² including both the XMM-LSS as well as Pan-STARRS Medium Deep Field MD01. In the area, we selected 1159 point-like X-ray sources at a flux limit of |$F_{0.5{\rm -}2 \,\rm {keV}} > 6\times 10^{-15} \,\rm {erg\,cm^{-2}\,s^{-1}}$| and 11 point-like X-ray sources at a flux limit of |$F_{2{\rm -}10 \,\rm {keV}} > 6 \times 10^{-14} \,\rm {erg\,cm^{-2}\,s^{-1}}$|⁠. These flux limits were chosen to correspond to the planned eROSITA limits in the deep exposed ecliptic pole regions. Applying an r-band cut of |$17.0 \,\rm {mag} < r_{\rm psf} < 22.5 \,\rm {mag}$| and likelihood ratio of LR_{XMM, SDSS} > 1.5, there were 795 BOSS targets, out of which 499 sources have been observed. In Fig. 4, we show the position of the BOSS spectra and the ancillary plate in the northern XMM-XXL area.

Second BOSS ancillary programme. The second group of spectra comes from the larger ancillary programme comprising four BOSS plates led by A. Georgakakis. This programme was fully dedicated to the follow-up of point-like X-ray sources covering nearly the entire XMM-XXL north area. Due to the larger amount of available fibres, we selected all 8445 extracted X-ray point-like sources in the 18  deg² XMM-XXL north reaching to a flux of |$F_{\rm 0.5{\rm -}10\, keV} \approx 10^{-15} \rm \,erg\, cm^{-2}\, s^{-1}$| and we applied 15.0 < r_psf/model < 22.5 mag (see Fig. 2) for optically unresolved sources (⁠|$\rm {TYPE}=6$|⁠: psf-magnitude) and optically resolved sources (⁠|$\rm {TYPE}=3$|⁠: model-magnitude). Within the 18 deg², our data set contains 3885 XMM-SDSS matched sources within the corresponding r-band threshold and with LR_{XMM, SDSS} > 1.5. The footprint of the four plates (see Fig. 4) comprises 3461 XMM-SDSS matched sources, out of which 2357 have been followed-up by BOSS.

DR10 spectra. Some of the X-ray sources from the second ancillary programme in the same field had already been independently targeted as part of the BOSS large-scale structure survey programme as LRG (luminous red galaxies) or candidates for high-redshift QSOs. Thus, in order to increase the spectroscopic completeness of our sample within 15.0 < r_psf/model < 22.5 mag, we included this third group of 401 public available spectra from 14 different plates within BOSS-DR10 (Ahn et al. 2012).

Summarizing, we selected 8445 X-ray sources and matched 3885 SDSS counterparts by applying the r-band criteria 15.0 < r_psf/model < 22.5 mag and likelihood ratio LR_{XMM, SDSS} > 1.5. From the BOSS observations, we obtained 3302 spectra. This number reduces to 3042 unique objects after correction for 260 objects with multiple spectra (257 double observations, 3 triple observations) by prioritizing the spectra with the higher signal-to-noise ratio (S/N ratio). 2474 of the BOSS spectra have a WISE counterpart with good photometry.

3 SPECTROSCOPIC REDSHIFT AND CLASSIFICATION OF AGNs

In this section, we present the processing of optical spectra in the BOSS pipeline and discuss the determination of spectroscopic redshifts by visual inspection, as well as objective classification of our X-ray sources based on their spectroscopic properties.

3.1 BOSS pipeline products

The BOSS pipeline processes the observational raw data in two steps. In the first pass, the spec2d pipeline converts the two-dimensional CCD data into one-dimensional spectra, introduces calibrations in wavelength and flux, and combines the red and blue spectral halves. In the second pass, the spec1d pipeline automatically analyses the features of the one-dimensional spectra and assigns a redshift and a classification with a χ²-minimization method (Bolton et al. 2012; Stoughton et al. 2002). The relevant information about redshift and emission line features of the spectra are stored in the pipeline products spZall file and spZline file (DR12, v5_7_0, Bolton et al. 2012).

For the determination of the redshift, the pipeline fits templates of galaxies (−0.01 < z < 1.0) and quasars (0.0033 < z < 7.0) with a linear combination of four principal components. The stars have dedicated standard templates. After the fitting process, the BOSS pipeline assigns a quality flag (ZWARNING) to each redshift |$\rm {Z\_BOSS}$|⁠. The optimal flag is |$\rm {ZWARNING}=0$|⁠, which is assigned to 80 per cent of our spectra. Redshift identifications with |$\rm {ZWARNING}>0$| may be caused by two best-fitting templates with similar χ², outlying points from best-fitting model or a minimal χ² at the redshift edge.

The line features of every spectrum are provided by the spZline data model of the BOSS pipeline. We employ these data without any additional line fitting procedure. The line fit is performed for 31 emission lines with a single Gaussian on top of the continuum subtracted spectrum. The redshift is re-fit non-linearly after the initial guess of the main redshift analysis and fixed for all lines beside Lyα. Blueshifts of individual lines are not accounted for. The line widths are calculated as a strength-weighted average of dedicated emission line groups, independent of the broad or narrow line character:

• Balmer lines: Hα, Hβ, Hδ, Hγ, Hϵ

• Lyman lines: Lyα

• Nitrogen line: N v

• Other lines: [Ar iii], [S ii], [N ii], [O i], [S iii], HeI, He ii, [O iii], [Ne iii], [O ii], Mg ii, C iii], C iv.

The fluxes of [O iii] 5007 and [O iii] 4959 as well as [N ii] 6583 and [N ii] 6548 are imposed with a ratio of 3:1. The average width of group (iv) lines is less representative of the true width. It includes both narrow and broad lines, and therefore underestimates the individual broad lines widths and overestimates the individual narrow lines widths. In our work, we focus on nine emission lines of interest (λ in vacuum wavelength): C iv (1549.48 Å), C iii] (1908.73 Å), Mg ii (2800.32 Å), [O ii]-doublet (3727.09 Å) and (3729.88 Å), Hβ (4862.68 Å), [O iii] (5008.24 Å), Hα (6564.61 Å) and N ii (6585.27 Å). There are three emission line parameters from spZline, which are important for the classification:

• the emission line flux A_gauss (LINEAREA) and the error ΔA_gauss (LINEAREA_ERR),

• the full width at half-maximum FWHM derived from |$\rm FWHM = \sigma \times (2 \sqrt{2\ln {2}})$|⁠, where σ is the Gaussian width in km s^{− 1} (LINESIGMA) and

• the equivalent width EW (LINEEW).

We impose a significance threshold of A_gauss/ΔA_gauss > 3 for all emission lines.

3.2 Redshift determination

The redshifts for the BOSS-observed s are taken from the spZall file. Typically, all QSO observed by BOSS are candidates for baryonic oscillations studies and therefore pass a visual inspection (Pâris et al. 2012, 2014). We expect the sources of our pilot study to be different from the majority of the standard BOSS-LRG and -QSO targets which are selected based on optical criteria. They will include e.g. narrow-emission line AGNs, host galaxy dominated AGNs and AGNs with less steep power laws. For this reason, we started a visual screening of our data set. We evaluate the redshift provided by the pipeline Z_BOSS by visual inspection and assign both a new redshift Z and confidence parameter Z_CONF, as presented in Appendix B. The following redshift variables are provided in the catalogue:

• |$\rm {Z\_BOSS}$|⁠: the redshift provided by the BOSS pipeline before visual inspection;

• |$\rm {Z}$|⁠: the adapted redshift after visual inspection;

• |$\rm {ZERR}$|⁠: the error of Z;

• |$\rm {Z\_CONF}$|⁠: the redshift confidence after visual inspection:

•   3 – reliable pipeline redshift,

•   2 – not robust pipeline redshift,

•   1 – bad spectrum,

•   30 – reliable visual redshift and pipeline failure,

•   20 – not robust visual redshift and pipeline failure,

• |$\rm {STAR/BLLAC}$|⁠: star or BL Lac flag.

After the visual inspection, 2525 sources have a reliable initial redshift assigned by BOSS pipeline, which corresponds to 83 per cent of redshift success. Additional 53 sources get a different reliable redshift, out of which 8 have only visual redshifts because they cannot not be correctly fitted by the pipeline (⁠|$\rm {Z\_CONF} = 30$|⁠). For the classification of the spectra in the following sections, we only use the 2570 BOSS spectra with |$\rm {Z\_CONF}= 3$|⁠, because they have the complete data products from the BOSS pipeline.

3.3 Spectroscopic classification

In this section, we present the classification rules for our X-ray-selected sources based on spectroscopic properties. We first introduce the bimodal FWHM distribution of AGN emission lines to separate the population of broad line and narrow line emitters. In a second step, we determine the ionization source of the narrow line emitters with the help of optical emission line diagnostic diagrams. The flow chart in Fig. 5 visualizes the steps of the classification process.

3.3.1 Line width bimodality of broad and narrow-emission line emitters

Our large sample of uniformly selected X-ray AGN allows for a systematic analysis of the FWHM of emission lines originating in the different regions of the AGNs. For the classification process, we evaluate the properties of potentially broad lines Hβ, Mg ii, C iii] and CIV. As described in 3.1, the widths for Hβ and Mg ii correspond to the strength-weighted averaged of their groups (i) and (iv).

In Fig. 6, we show the FWHM distribution of all significantly detected Hβ and Mg ii emission lines. Both distributions show a clear bimodal shape, which suggests the presence of two physically distinct populations. In detail, the minima of the distributions are at |$\rm FWHM_{(\rm H\,\beta )}=985\,\rm km\,s^{-1}$| and |$\rm FWHM_{\rm Mg\,\small {II}}=957\,\rm km\,s^{-1}$|⁠. The high-FWHM sources are associated with broad line region (BLR) emission and the low-FWHM sources are associated with narrow line region (NLR) emission. In the following, we will use the FWHM threshold of |$\rm FWHM=1000 \,\rm km\,s^{-1}$| to separate BLR and NLR emitters.

For our classification, we analyse the FHWM of the Hβ, Mg ii, C iii] and CIV lines in sequence. The first significant emission line, which has an FWHM larger than 1000 km s^{− 1} defines an object to be a BLR emitter. For z > 1.14, Hβ is redshifted out of the BOSS wavelength range. In this case, the width of Mg ii, C iii] or CIV are solely used for the classification. We find that at these redshifts, no more NLR emitters with significant emission lines are present in the sample. Therefore, the averaged FWHM for the group (iv) lines is not underestimated by strong narrow-emission line contribution.

In the following sections, we will refer to the BLR emitters as ‘unobscured’ (referring to type 1 AGN) from the optical point of view and call them ‘Broad Line AGN [of type] 1’: BLAGN1. Their optical spectra show the blue/UV continuum from the accretion disc and the characteristic Hβ, Mg ii, CIV and C iii] broad-emission lines from the high-velocity clouds close to the black hole.

The NLR emitters with |$\rm FWHM_{(\rm H\,\beta )}<1000\,\rm km\,s^{-1}$| can have different ionization sources, which will be discussed in the following section.

3.3.2 Ionization source diagnostics of narrow-emission line emitters

The ionizing radiation that excites the NLR (⁠|$\rm FWHM < 1000 \rm \,km\,s^{-1}$|⁠) of X-ray-detected AGN candidates can originate from different regions in the galaxy: the accretion disc in the nuclear region (AGN), star formation in the host galaxy (SFG) or both. The narrow line emitters powered by an AGN typically have an orientation which prevents a direct look into the accretion disc (Antonucci 1993; Urry & Padovani 1995). The BLR clouds close to the black hole are obscured and only narrow lines from distant clouds in low-density environments are visible in the spectrum. The excitation of narrow-emission lines by star formation typically happens through photoionization, e.g. in OB stars.

As is customary, we derive the dominating source of ionization for our sample of narrow line emitters between 0 < z < 1.08 by comparing the emission line ratios [O iii]/Hβ versus [N ii]/Hα (BPT-N ii-diagram, Meléndez et al. 2014) or [O iii]/Hβ versus [O ii]/Hβ (Blue-O ii-diagram, Lamareille 2010), as presented in Appendix C. AGNs have a stronger ionization continuum than SFG and typically reside in the top-right corner of the diagrams, whereas SFG reside in the bottom-left corner. In total, our data set contains 633 narrow line emitters out of which 349 sources can be classified by the diagnostic lines diagrams. There are 271 ‘AGN’ (BPT-N ii-diagram: 244, Blue-O ii-diagram: 27) which form the group of optically ‘obscured’ AGN (referring to type 2 AGN) and, in this work, are named ‘Narrow Line AGN [of type] 2: NLAGN2. Furthermore, there are 78 ‘SFG’ (BPT-N ii-diagram: 40, Blue-O ii-diagram: 38) which will be referred to as ‘elusive’ AGN with star formation: eAGN-SFG. They do not show any AGN-driven emission lines in their optical spectra, but their X-ray luminosities are characteristic of an active accretion process in the centre.

The ionization origin of the remaining 284 narrow line emitters cannot be reliably determined, because the required narrow-emission lines are below the significance limit or the redshift of the objects is too high (z > 1.08). The low significance of the lines is caused by very strong host galaxy continuum contribution or very low S/N ratio spectra. We classify this group as NLAGN2 candidates: NLAGN2cand and refer to their specific properties in Section 4.2.

Our data set also comprises objects whose spectra do not have any significant emission lines at all. Similarly, to the eAGN-SFG, they indeed have characteristic X-ray luminosities for AGN, but only show the features of passive galaxies in their optical spectra. We classify them as ‘elusive’ absorption line galaxies: eAGN-ALG. For their selection, we apply a significance threshold of A_gauss/ΔA_gauss = 3 to every emission line in the spectrum. Below this significance threshold, there is a large group of sources which are not eAGN-ALG. These are sources whose spectra mainly suffer from very low S/N ratio and do not show any stellar continuum or absorption lines at all. It is possible to derive a secure redshift by visual inspection, but our classification method is not able to automatically separate this group of objects. Therefore, we apply an additional conservative visual inspection to all of these objects. After the inspection, we sort them into two groups: 93 ‘eAGN-ALG’ and 57 ‘not classifiable’ spectra: NOC.

3.4 Final classified sample

Summarizing our final sample with 2570 spectra of optical counterparts to X-ray sources (⁠|$\rm {Z\_CONF}=3$|⁠) with emission line information, we classified:

• 1787 BLAGN1 (70 per cent),

• 271 NLAGN2 (11 per cent),

• 284 NLAGN2cand (11 per cent),

• 78 eAGN-SFG (3 per cent),

• 93 eAGN-ALG (4 per cent) and

• 57 not classifiable spectra (2 per cent).

The percentages relate to the AGN selection only. We point out that our data set includes also 85 X-ray-detected stars and 2 X-ray-detected BL Lac.

4 CLASS PROPERTIES

In the following paragraphs, we describe in more detail the redshift, X-ray and optical spectra properties of each class. We complement the analysis with spectral stacks, which are performed using the idl script run_composite written by Min-Su Shin (2009 January). In this process, the contributing spectra are normalized to a chosen wavelength and corrected for Galactic extinction as well as the BOSS spectroscopic resolution.

4.1 BLAGN1

The BLAGN1 span a large redshift range of 0.06 < z < 5.01 and a luminosity range of |$4.4 \times 10^{41} < L_{\rm 0.5{\rm -}2\,keV} < 3.7\times 10^{45} \, \rm {erg\,s^{-1}}$| (see Fig. 8). The upper redshift limit is strongly related to the faint SDSS r-band magnitude limit for the BOSS targets. This limit hampers the observation of objects at z > 4.5. At higher redshift, the Lyα absorption from the intergalactic medium strongly suppresses the spectroscopic features in the observed r-band. Our sample contains one exception at z = 5.01, which is caused by a Ly β-line in the r-band (r_psf = 20.92 mag).

In the two top panels of Fig. 9, we show the spectral stack of all BLAGN1 in the redshift ranges of 0 < z < 1 and 1 < z < 2, respectively. For comparison, we overplot the SDSS DR4 QSO template (Vanden Berk et al. 2001). The BLAGN1 show clear broad line features for the emission lines Mg ii, Hβ and Hα. At low redshifts, the spectral stack of the BLAGN1 tends to be redder and flatter than the SDSS QSO template. This is caused by the contribution of the host galaxy, indicated by the prominent Ca-doublet absorption and the continuum at λ > 4000 Å. At higher redshifts, the stack approaches the QSO template because the host contribution is outshone by the more luminous nuclear emission. The stack is still redder than the template, because our BLAGN1 have different optical colours than the QSO selection criteria and include mildly extinguished objects (see Section 5.1).

4.2 NLAGN2 and NLAGN2cand

The NLAGN2 are detected at a relatively low-redshift range of 0.04 < z < 0.91 and a luminosity range of |$1.9\times 10^{40} < L_{0.5{\rm -}2\,\rm keV} < 1.9\times 10^{44}\, \rm {erg\,s^{-1}}$| (see Fig. 8). The upper redshift limit of NLAGN2 is hampered by the optical line diagnostic diagrams. Within the redshift range of 0.5 < z < 0.9, the NLAGN2 magnitude distribution reaches r_model = 22.5 mag and the dominating host contribution becomes too faint for detection in SDSS images. Additionally, the fraction of obscured AGN is known to decrease at higher luminosities (Ueda et al. 2003; Hasinger 2008; Merloni et al. 2014). Thus, the high-redshift and high-luminosity objects in our sample tend to be more dominated by BLAGN1.

In the central panel of Fig. 9, we show the spectral stack of all NLAGN2. We chose the redshift range of 0 < z < 1 which corresponds to the upper redshift bound of the optical emission line diagrams. The stack displays the narrow AGN emission lines occurring for [O ii], Hβ, [O iii], Hα and N ii. Furthermore, there are strong host galaxy features, such as a clear Ca-doublet with the 4000 Å break and the stellar continuum in the red part of the spectrum.

We now introduce the group of NLAGN2cands for narrow line objects whose emission origin cannot be determined by the BPT-N ii/Blue-O iidiagrams. We find that these sources either have spectra with strong host contribution, low S/N ratio or they are narrow line emitters at z > 1.08. They span a redshift range of 0.04 < z < 2.66 and luminosity range of |$2.4\times 10^{40} < L_{0.5{\rm -}2\,\rm keV} < 1.2\times 10^{45}\,\rm {erg\,s^{-1}}$| (see Fig. 8). Their spectral stack (0 < z < 1) in Fig. 9 reveals the continuum features of pure NLAGN2 but also the slightly broadened lines of BLAGN1. Furthermore, the optical SDSS morphology of 94 per cent of NLAGN2cand is extended and their optical colours differ from QSO (see Section 5.1). Overall, it suggests that the majority of the sources are NLAGN2, which is reflected in their naming. Deeper exposed optical spectra or IR spectra are needed to properly classify these objects. The three outstanding NLAGN2cand at 2.42 < z < 2.66 have low S/N ratio spectra and show broad-emission line features below the required significance threshold. They are probably candidates for faint BLAGN1.

4.3 eAGN-ALG and eAGN-SFG

The group of elusive AGN consists of sources whose spectra are consistent with either absorption line galaxies (ALGs) or star-forming galaxies (SFGs). The eAGN-ALG reside within a redshift range of 0.14 < z < 0.96 and a luminosity range of |$1.4\times 10^{41} < L_{0.5{\rm -}2\,\rm keV} < 2.0\times 10^{44}\, \rm {erg\,s^{-1}}$| (see Fig. 8). Their median soft X-ray luminosity is |$L_{0.5{\rm -}2\,\rm keV} = 3.43 \times 10^{42}\, \rm {erg\,s^{-1}}$|⁠. The eAGN-SFG span a redshift range of 0.02 < z < 1.01 and a luminosity range of |$2.1\times 10^{39} < L_{0.5{\rm -}2\,\rm keV} < 9.1\times 10^{43}\, \rm {erg\,s^{-1}}$|⁠. Their median soft X-ray luminosity is at |$L_{0.5{\rm -}2\,\rm keV} = 2.89 \times 10^{41}\, \rm {erg\,s^{-1}}$| (see Fig. 8). The counterparts of the eAGN have the same distribution in r-band, S/N ratio, likelihood ratio, separation distance and number of counterparts as the entire population. Therefore, they should have the same probability distribution of counterpart association as the other objects.

The spectral stack of eAGN-ALG in the bottom panel of Fig. 9 only shows absorption line features and is very similar to the typical SDSS DR4 early-type galaxy templates (Yip et al. 2004). On the other hand, the eAGN-SFG reveal the stellar emission lines, e.g. [O ii] and [O iii] and resemble the SDSS DR4 template for late type galaxies (Yip et al. 2004).

What is the engine of elusive AGN? In order to determine whether these objects really host an AGNs, we study their soft X-ray luminosities. We can assume that ALG have a maximal stellar mass of ∼10¹²M_⊙, based on the massive BOSS LRG sample of Maraston et al. (2013). According to Gilfanov (2004), this corresponds to an X-ray luminosity of |$L_{0.5{\rm -}2\rm \, keV}\le 10^{41}\,\rm erg\,s^{-1}$| caused by low-mass X-ray binaries in the galaxy. The soft X-ray luminosity of all 78 eAGN-ALG in our sample is above 10⁴¹ erg s^{− 1}. Therefore, they can be considered as hosts of AGNs and are not dominated by luminous massive X-ray binaries.

Focusing on the SFG population, their overall X-ray emission – apart from the nuclear AGNs – consists of the collective emission from stellar remnants in a galaxy. The upper limit on the X-ray luminosity can be associated with the maximum of the total star formation rate (Nandra et al. 2002; Grimm-J., Gilfanov & Sunyaev 2003; Ranalli et al. 2003). Typically, the value of |$L_{0.5{\rm -}2\rm \,keV}\sim 10^{42}\rm \, erg\,s^{-1}$| is considered as a threshold above which only AGNs should be mainly responsible for the observed emission. Our sample of eAGN-SFG consists of 49 objects above this luminosity limit and 29 objects below this luminosity limit, corresponding to a ratio of 0.59 ± 0.40. Comparing the number fraction to securely classified NLAGN2 in the same redshift and luminosity range, we obtain 0.72 ± 0.18. The consistency of both ratios might be an indication that the majority of our eAGN-SFG with |$L_{0.5-2\rm \,keV}<10^{42}\rm \, erg\,s^{-1}$| in our sample really host a comparably weak AGN.

5 ASSESSMENT OF OPTICAL AND MID-IR AGN COLOUR SELECTION TECHNIQUES

One of the expected strengths of the X-ray AGN selection is the ability to identify accreting black holes which are either obscured, or heavily contaminated by stellar emission processes at optical and NIR wavelengths. Indeed, our classification work presented in the previous section clearly demonstrates that among X-ray-selected AGNs, we find a variety of objects with a quite diverse set of optical properties. This significant fraction of AGNs would otherwise not have been selected by other selection criteria.

In order to further evaluate how X-ray selection of AGNs compares with photometric optical and mid-IR colour selection, we analyse in this section the optical and IR colour properties of our sample of X-ray-selected AGNs and we compare it to the overall population of XDQSO (Section 5.1) and WISE (Section 5.2) colour-selected AGNs in the XMM-XXL north.

5.1 Optical: XDQSO targeting algorithm

For the optical selection of AGNs, we refer to the XDQSO algorithm (Bovy et al. 2011), which is a probabilistic selection technique developed for the efficient QSO selection on the basis of broad-band optical imaging and point-like morphology. It has been employed as a target selection method for the measurement of baryon acoustic features with quasars within the eBOSS programme (Myers et al. 2015). The selection is limited to optically point-like objects (⁠|$\rm {TYPE}=6$|⁠), with a dereddened i-band magnitude of 17.75 < i < 22.45 mag and at least one primary detection with u_psf < 22.5 mag, g_psf < 22.5 mag, r_psf < 22.5 mag, i_psf < 22.0 mag or z_psf < 21.5 mag. We only choose sources with a good BOSS quasar targeting flag (⁠|$\rm {GOOD}=0$|⁠, Bovy et al. 2011) excluding the photometric blending, moving sources and interpolation effects. The algorithm uses the density estimation in flux space and assigns probabilities P(QSO, z) of any SDSS point source to be a QSO in dedicated redshift ranges. In this work, we refer to the quasar probability P_sum(QSO) as the sum of the low-redshift (z < 2.2), mid-redshift (2.2 ≤ z ≤ 3.5) and high-redshift (z > 3.5) probabilities. As a threshold for the AGN selection, we define the arbitrary probability of P_sum(QSO) > 0.5. The algorithm has been trained using a star sample from SDSS Stripe 82 and a quasar sample from the SDSS DR7 Quasar catalogue.

5.1.1 X-ray detection fraction

In the 18 deg² coverage area of XMM–Newton in the XMM-XXL north, the XDQSO catalogue contains 49 172 sources with assigned P(QSO, z) and |$\rm {GOOD}=0$|⁠. There are 1617 XDQSO sources which can be associated with the XMM-SDSS catalogue (LR_{XMM, SDSS} > 1.5). Considering a QSO probability of P_sum(QSO) > 0.5, the entire XMM-XXL north comprises 2408 XDQSO sources out of which 1159 are also X-ray selected.

Assuming P_sum(QSO) > 0.5 and |$F_{0.5{\rm -}10\,\rm keV}>10^{-15}\,\rm erg cm^{-2} \, s^{-1}$|⁠, ∼48 per cent of all XDQSO-selected sources will be picked up in X-ray at faint optical magnitudes (r ≤ 22.5 mag) and up to ∼72 per cent at bright r magnitudes (r ≤ 18.5 mag).

5.1.2 Morphological and spectroscopic properties of XDQSO-selected AGN

In the following section, we analyse the XDQSO properties of the X-ray sources with BOSS follow-up: 1203 XDQSO sources have reliable BOSS spectra, out of which 902 have P_sum(QSO) > 0.5. As shown in Table 1, these objects only comprise of optically point-like objects and are dominated by BLAGN1. Only few are classified as NLAGN2cand, NLAGN2 and eAGN. There are 196 point-like BLAGN1 which are not XDQSO selected due to their photometric properties (⁠|$\rm {GOOD}>0$|⁠). 27 of them are too faint or bright for the initial magnitude thresholds of the XDQSO selection (17.75 < i_dered < 22.45 mag). By construction, the selection misses all optically extended objects.

In Fig. 11, we perform spectral stacks of X-ray-selected XDQSO sources with different QSO probabilities and soft X-ray luminosities. We chose the redshift range of 0 < z < 2 which covers a representative sample of the entire population. The stack includes the four following groups of X-ray and XDQSO-selected targets:

• P_sum(QSO) > 0.5, |$L_{0.5{\rm -}2\,\rm keV}>10^{44}\rm \, erg\,s^{-1}$| (n = 381),

• P_sum(QSO) > 0.5, |$L_{0.5{\rm -}2\,\rm keV}<10^{44}\rm \, erg\,s^{-1}$| (n = 310),

• P_sum(QSO) < 0.5, |$L_{0.5{\rm -}2\,\rm keV}>10^{44}\rm \, erg\,s^{-1}$| (n = 70),

• P_sum(QSO) < 0.5, |$L_{0.5{\rm -}2\,\rm keV}<10^{44}\rm \, erg\,s^{-1}$| (n = 134).

The groups of high QSO probability spectra from (a) and (b) show clear broad line features. The low QSO probability spectra from (c) and (d) have a less steep power law and a more prominent galaxy contamination at redder wavelengths. Apparently, a more dominant host galaxy continuum affects the optical colour properties of the AGN which leads to a lower QSO probability. The X-ray luminosity for the same QSO probability range has only a very weak impact on the red part of the spectrum, because the host contribution mostly affects P_sum(QSO).

Summarizing, in the XMM-XXL, 48 per cent of the XDQSO-selected AGN (P_sum(QSO) > 0.5) are also X-ray selected and 13 per cent of all X-ray sources have XDQSO counterparts. In order to perform a full comparison of both AGN selections, we are missing a representative spectroscopic sample for only XDQSO-selected objects without X-ray detection. But we confirm that the XDQSO targeting algorithm selects optically point-like BLAGN1 showing an unobscured and blue optical spectrum with only weak host galaxy contribution. Among the XDQSO-selected AGN with X-ray detection, the typical X-ray luminosity is |$L_{0.5{\rm -}2\,\rm keV}> 10^{42}\rm \, erg\,s^{-1}$| and redshift range is 0 < z < 5.

5.2 IR: WISE colour selection

5.2.1 X-ray detection fraction

In Fig. 12, the Stern et al. (2012) (dashed line) and the Assef et al. (2013) (solid lines) selections do not differ much in their number densities up to W2 ≤ 15 mag. At fainter magnitude, the Assef threshold allows the selection of further AGNs. Assuming a soft X-ray flux of |$F_{0.5{\rm -}10\,\rm keV}> 10^{-15}\,\rm erg\, cm^{-2}\, s^{-1}$| and the faint W2 limit of the WISE selections, ∼44 per cent (W2 ≤ 17 mag) of all Assef and ∼56 per cent (W2 ≤ 15 mag) of all Stern-selected AGN are detected in X-ray, too. For brighter magnitudes (W2 ≤ 13 mag), the fraction increases up to ∼67 per cent for the Assef and Stern selection.

5.2.2 IR and spectroscopic properties of WISE-selected AGNs

The overall population of X-ray sources with IR counterpart comprises 4811 sources, out of which 2117 also have reliable BOSS classifications. In the top panels of Fig. 13, we plot these sources in the W1 − W2 colour and W2 − W3 colour diagram and compare the source locations with characteristic source regions from Wright et al. (2010). The classifications of our BOSS sources are consistent: the BLAGN1 mainly reside in the QSO and Seyfert area, the NLAGN2 extend from the spiral region in the Seyfert region. The eAGN-SFG live in the starburst region extending to the Seyfert region. The eAGN-ALG are in the spiral region, which might be caused by the contribution of the AGN emission in the W2 and W3 band. The stars selected via X-rays fall into the common WISE star region but also spread beyond.

In the bottom-left panel of Fig. 13, we indicate the WISE selections from Stern et al. (2012) and Assef et al. (2013) in the W1 − W2 colour and W2 magnitude diagram. The plot comprises the density contours of all WISE sources in the XMM-XXL north area and we overplot the X-ray sources with WISE counterparts and the BOSS observed spectra. The Stern et al. (2012) criterion selects 390 BOSS observed X-ray sources and the Assef et al. (2013) criterion selects 585 targets. The number statistics and classifications of the WISE and X-ray-selected sources are listed in Table 2. The selected sources mainly consist of BLAGN1 (89 per cent for both selections), NLAGN2/NLAGN2cand and few eAGN. The lower-right panel demonstrates the W1 − W2 colour and redshift dependency of X-ray-selected Assef et al. (2013) AGN with spectroscopic information. Both selections are limited to a redshift range of z < 3 due to the overall SED shape of AGNs in the IR rest frame (see Stern et al. 2012).

In Fig. 14, we show the spectral stacks of X-ray-selected Assef et al. (2013) AGNs at different luminosities in the redshift range 0 < z < 2. We chose the average luminosity of |$L_{0.5{\rm -}2\,\rm keV}= 10^{44}\rm \, erg\,s^{-1}$| as a threshold for the high- and low-luminosity populations. This redshift range comprises a representative population from the BLAGN1 and both NLAGN2 and eAGN classes. The four groups of X-ray-selected spectra are

• Assef selected: |$L_{0.5{\rm -}2\,\rm keV}>10^{44}\rm \, erg\,s^{-1}$| (n = 266),

• Assef selected: |$L_{0.5{\rm -}2\,\rm keV}<10^{44}\rm \, erg\,s^{-1}$| (n = 251),

• Not Assef selected: |$L_{0.5{\rm -}2\,\rm keV}>10^{44}\rm \, erg\,s^{-1}$| (n = 249),

• Not Assef selected: |$L_{0.5{\rm -}2\,\rm keV}<10^{44}\rm \, erg\,s^{-1}$| (n = 1010).

The high luminosity X-ray and Assef et al. (2013) selected spectra from (a) comprise only BLAGN1 which show clear broad line features and nearly no host galaxy contribution. The stack of low luminosity sources from (b) shows emission lines which are less broad and a redder continuum indicating host galaxy contribution from NLAGN2 and host galaxy dominated BLAGN1. The stacks from (c) and (d) are X-ray sources not selected by Assef et al. (2013). They show stronger host galaxy contribution than the Assef et al. (2013) selected sources in the same luminosity range. The stack from (d) includes the large majority of host-dominated AGNs.

Summarizing, 44 per cent of the Assef et al. (2013) selected AGNs are also selected by X-ray and 11 per cent of all X-ray-selected AGNs are also IR AGNs. Our comparison of X-ray selection criteria with Assef et al. (2013) and Stern et al. (2012) is biased because the spectroscopic sample does not comprise only IR selected AGNs and is limited in the r-band. But applying the two IR AGN selection criteria to our X-ray-selected sample, they access a population of AGNs with very weak host contribution and red W1 − W2 colours. This group includes mainly BLAGN1 and few NLAGN2 in the redshift range of 0 < z < 3 and has a large luminosity range of |$4\times 10^{40}<L_{0.5{\rm -}2\rm \,keV}<2\times 10^{45}\rm \, erg\,s^{-1}$|⁠.

6 DISCUSSION

6.1 Understanding the AGN population in the XMM-XXL north

The identification and classification of X-ray sources has been already performed in many subfields of the XMM-XXL north region. Tajer et al. (2007) and Polletta et al. (2007) used a sample of 136 X-ray point like sources at a flux of |$F_{2{\rm -}10 \,\rm {keV}} >10^{-14} \,\rm {erg\,cm^{-2}\,s^{-1}}$| in a 1 deg² area and obtained a reliable photometric redshift and classification for 107 sources with optical and IR photometry. Garcet et al. (2007) used 612 point-like X-ray sources at a limiting flux of |$F_{2-10 \,\rm {keV}} > 8\times 10^{-15} \,\rm {erg\,cm^{-2}\,s^{-1}}$| in an area of 3 deg² and associated 99 objects followed up by optical spectroscopy (2dF, VIMOS, SALT) with secure redshift and classification (R < 22 mag). In the same field, Stalin et al. (2010) selected 829 objects at |$F_{0.5{\rm -}2 \,\rm {keV}} > 10^{-15} \,\rm {erg\,cm^{-2}\,s^{-1}}$| and obtained a secure identification for 487 sources with optical spectroscopy at AAT (g′ < 22 mag).

Compared to all these programmes, our campaign stands out for the sheer number of spectra available from the same instrument, under stable, uniform observing conditions, and for the use of a well tested, well understood and robust data analysis pipeline. In the XMM-XXL north, we obtained 3042 unique BOSS spectra and reached a spectroscopic completeness of 32 per cent for all X-ray-selected point sources at |$F_{0.5{\rm -}10 \,\rm {keV}} > 10^{-15} \,\rm {erg\,cm^{-2}\,s^{-1}}$| and 78 per cent for all optical cross-matched sources with 17 < r < 22.5 mag. In order to increase the number of reliable spectra, we performed a visual inspection for the spectra with critical redshift estimates of the BOSS pipeline. We find that 85 per cent of the X-ray-selected and BOSS observed targets have a reliable redshift identification after the visual inspection. Our classification of X-ray-selected and optically followed-up spectra is based on the optical emission line features of the sources. The algorithm presented in this work classifies 2513. For 2 per cent of all spectra with reliable redshift, the emission line information are not sufficient for a classification.

The principal element of our classification algorithm is the bimodal distribution of the emission line FWHM, separating NLR emitters and broad line region (BLR) emitters at an FHWM-threshold of |$\rm FWHM\sim 1000\rm \,km\,s^{-1}$|⁠. The bimodality can be explained by the distinct locations of NLR and BLR. Similar distributions have also been shown in the paper of Hao et al. (2005) with a threshold of |$\rm FWHM\sim 1200\rm \,km\,s^{-1}$| based on a sample of low-redshift AGNs with subtracted galaxy continuum. The authors fit each individual emission line with multiple gaussians, whereas in this work we use averaged widths. The major group of X-ray-selected objects in the XMM-XXL north are the BLR emitters: BLAGN1. The group of NLR emitters includes objects whose X-ray emission originates either from an obscured AGN, the star-forming region or both. They are selected based on their position in the BPT-N ii and Blue-O ii diagram. The NLAGN2 are powered by a central AGN and form the second large group with 11 per cent of all classified X-ray-selected objects. Our data set comprises a group of equal size with NLAGN2 candidates (11 per cent) which do not have enough significant emission lines or lie at a redshift of z > 1.08 and cannot be projected in the BPT-N ii and Blue-O ii diagram. Their spectral stack as well as optical colours and morphology indicate that the majority of the sources are NLAGN2. Deeper exposed optical spectra or IR spectra are needed to properly classify these objects. For the majority of BLAGN1 and NLAGN2, the optical classification is also confirmed by their X-ray spectra. But there are known exceptions of e.g. optically type 1 AGN with obscured X-ray spectra (e.g. Brusa et al. 2003; Merloni et al. 2014) or optically type 2 AGN with unobscured X-ray spectra. These objects require more detailed X-ray spectral analysis (see Liu et al. 2015, submitted).

Our sample contains 6 per cent elusive AGN (eAGN) living in SFGs (eAGN-SFG) and absorption line galaxies (eAGN-ALG). They do not stand out in any cross-matching related variable compared to the entire population. The fraction of elusive AGNs is comparable to the COSMOS sample (Brusa et al. 2010, 4 per cent, considering |$L_{0.5{\rm -}2\,\rm keV} > 10^{42}\, \rm {erg\,s^{-1}}$|⁠). Elusive AGN have been referred to as optically ‘dull’ AGN or XBONGS by Comastri et al. (2002), Maiolino et al. (2003), Caccianiga et al. (2007), Trump et al. (2009) and Pons & Watson (2014). The absence of AGN signatures in their optical spectra is still a matter of debate. According to Georgantopoulos & Georgakakis (2005), it is caused by dilution effects of a strong host galaxy component. Comastri et al. (2002) argues for a strong obscuration of the nuclear source preventing optical AGN emission. Yuan & Narayan (2004) suggest that the XBONGS have truncated disc close to the black hole causing the absence of the characteristic AGN emission line. Extreme obscuration of the NLR by AGN-fueling dusty spirals (Kraemer et al. 2011), or AGN flickering effects (Schawinski et al. 2015) have also been suggested. Based on their soft X-ray luminosity, we can confirm that the eAGN-ALG host an AGN. The group of eAGN-SFG partly falls below the upper luminosity threshold of SFG (⁠|$L_{0.5{\rm -}2\,\,\rm keV}=10^{42}\,\rm erg\,s^{-1}$|⁠), but the number fraction is consistent with NLAGN2 in the same redshift and luminosity range. Liu et al. (2015, submitted) analyse the X-ray properties of the elusive AGNs. 75 per cent of them have N_H < 21.5 cm^{− 2} indicating the presence of a low luminosity type 1 AGN, whose optical emission is probably fully dominated by stellar emission processes (Davies et al. 2015).

Narrow Line Seyfert 1 AGN (NLSey1) are also part of our X-ray-selected AGN. As shown by e.g. Osterbrock & Pogge (1985), Boller, Brandt & Fink (1996), this population of optically unobscured AGN typically has FWHM for Hβ in the range of |$500 < \rm FWHM_{\rm H\,\beta } < 1500\,\rm km\,s^{-1}$|⁠, very steep X-ray spectra, strong optical Fe ii emission and strong variability. The NLSey1 contribute to our BLAGN1 group, but probably also affect narrow-emission line objects with |$\rm FWHM_{\rm H\,\beta } < 1000\,\rm km\,s^{-1}$|⁠. Castelló-Mor et al. (2012) find that NLSey1 are classified as SFG by the optical line diagnostic diagram due to their smaller Hβ EW. The NLAGN2 of our data set are securely identified as AGNs because of their projection in the AGN region of the optical line diagnostic diagrams. But the group of NLAGN2cand and eAGN-SFG could potentially be contaminated by NLSey1. According to Castelló-Mor et al. (2012), the NLSey1 which are projected in the SFG region have high X-ray luminosities |$L_{2{\rm -}10\,\,\rm keV}>10^{42}\,\rm erg \,s^{-1}$| and large X-ray to optical flux ratios (log F_X/F_opt > 0.1). This applies to a subset of our eAGN-SFG as shown in Fig. 7. From the stack of the NLAGN2 and eAGN-ALG, however, we see that the majority of contributing spectra have very strong host galaxy continua (see bottom panel of Fig. 9). This indicates that our classification procedure for objects with |$\rm FWHM_{\rm H\,\beta } < 1000\,\rm km\,s^{-1}$| mainly selects optically obscured AGNs and is only weakly affected by NLSey1. A dedicated study of the NLSey1 contribution in our sample would require further FeII complex and X-ray spectra analysis, which is not in the subject of this paper.

The X-ray-selected data set also contains Galactic stars. We can estimate the star fraction of the entire X-ray data set, assuming a fraction of 3 per cent for all XMM-SDSS cross-matched objects with r > 15 and a fraction of 100 per cent for all objects with r < 15. This results in a maximal star fraction of ∼7 per cent.

6.2 Uniqueness of X-ray AGN selection

Considering current and future large area surveys in different wavelength bands, the number of observable AGN populations is going to increase dramatically. Many studies underline the strong impact of selections on the characteristic features of AGNs. In this work, we have access to a very large area of 18 deg², covered by multiwavelength data, allowing for comparison of three AGN selection criteria in X-ray, optical and IR, with high statistic reliability. We find interesting trends regarding the optical morphology, the X-ray luminosity, the redshift and host galaxy contribution.

In Fig. 15, we illustrate the relative sizes of the differently selected AGN samples and their reliability fractions in the XMM-XXL. In addition, we show the subsets of sources with X-ray fluxes above the expected limit of the first and final eROSITA all sky coverage (eRASS:1 and eRASS:8, respectively), as explained in Section 7 below. The WISE and SDSS catalogues have been matched via their source positions to be within a distance of 1 arcsec. The associated X-ray sources belong to the likelihood ratio match catalogue from Section 2.2.

Considering the limiting fluxes of SDSS, WISE and XMM–Newton in the XMM-XXL north, the combination of all AGN selection criteria picks up about 600 AGN deg^{− 2}. Their common subgroup includes only ∼23 AGN deg^{− 2} out of which 99 per cent are optically point-like BLAGN1 with very weak host galaxy components. Applying a less conservative limit for the positional distance between SDSS and WISE sources, the number of objects in the common subgroup with XMM–Newton increases from 415 (1 arcsec), to 430 (2 arcsec) and finally 434 (∼ 8.5 arcsec). The total intersection of only SDSS and WISE comprises 596 sources for 1 arcsec and 623 sources for 2 arcsec.

Each of the three AGN selection methods also retrieves an exclusive population of AGNs. At the full depth of |$F_{\rm 0.5{\rm -}10\, keV} > 10^{-15} \rm \,erg\, cm^{-2}\, s^{-1}$|⁠, ∼380 AGN deg^{− 2} are only selected in X-rays, corresponding to 80 per cent of the total X-ray selection. XDQSO and IR selections have both access to nearly ∼60 exclusive AGN deg^{− 2} each, which corresponds to ∼50 per cent of their full selections.

The last two selections preferentially pick up AGNs which are outshining their hosts at high optical or IR luminosities, respectively (Bahcall et al. 1997). The XDQSO selection (Bovy et al. 2011) is trained for typical optically point-like BOSS-QSO and is highly sensitive within the available depth of optical SDSS data. This is crucial e.g. to reach a high number of QSO in the baryonic oscillations programme of eBOSS (Dawson et al. 2013). Our spectroscopic sample does not include BOSS spectra that are exclusively XDQSO selected, but these sources are most likely to be optically point-like low X-ray luminosity BLAGN1. The IR selection from Assef et al. (2013) is limited to the redshift range of z < 3 and the majority of selected sources are classified as BLAGN1 having red W1 − W2 colours. The exclusively IR selected AGN are expected to be low X-ray luminosity AGN (both BLAGN1 and NLAGN2) with a weak host galaxy contribution, or heavily obscured AGNs (both BLAGN1 and NLAGN2).

Our study confirms that X-rays are less sensitive to dilution effects in the host galaxy, optical morphology, obscuring material and star formation, as first shown by Nandra et al. (2007) and Bundy et al. (2008). In our sample, they pick up a broad variety of AGNs, such as typical QSO, optically unobscured AGNs with strong host galaxies, optically obscured AGN and elusive AGN in a large luminosity and redshift range.

Based on our results, we expect that the selection method has a strong effect on statistical properties such as clustering and dark halo mass of AGN populations. As also shown by Merloni (2015), at any given AGN luminosity, X-rays select systems accreting at lower Eddington rates λ. These objects reside in more massive hosts with lower contrast of host galaxy to nuclear emission, and are missed by any other selection method. This is in line with the study of Mendez et al. (2015) focusing on the clustering properties of X-ray, radio, and IR selected AGN from the PRIMUS and DEEP2 surveys. The authors find that X-ray-selected AGNs cluster more than IR galaxies and reside in more massive dark matter haloes. This is not the case if the stellar mass, specific star formation rate and redshift distributions of the selected galaxies are matched to the same control sample.

7 eROSITA AND SPIDERS FORECAST

In light of the upcoming all sky X-ray survey by eROSITA on-board the SRG mission (Merloni et al. 2012; Predehl et al. 2014), our study forms an outstanding data set to study the characteristics of the X-ray AGN population which will be discovered in large numbers. Comparing to the ROSAT all-sky survey (Voges et al. 1999), the X-ray-selected AGN density over the whole sky is going to increase from ∼ 2 deg^{− 1} to ∼ 90 deg^{− 1}. For an angular resolution averaged across the field of view of about 26 arcmin (half energy width, Burwitz et al. 2014), the predicted flux limits of eROSITA are

• – |$F_{0.5-2 \,\rm {keV}}>4.0\times 10^{-14}\rm \,erg\,s^{-1}\,cm^{-2}$| (eRASS:1, 0.5 yr),

• – |$F_{0.5-2 \,\rm {keV}}>2.5\times 10^{-14}\rm \,erg\,s^{-1}\,cm^{-2}$| (eRASS:2, 1 yr),

• – |$F_{0.5-2 \,\rm {keV}}>1.5\times 10^{-14}\rm \,erg\,s^{-1}\,cm^{-2}$| (eRASS:4, 2 yr),

• – |$F_{0.5-2 \,\rm {keV}}>9.8\times 10^{-15}\rm \,erg\,s^{-1}\,cm^{-2}$| (eRASS:8, 4 yr).

The scanning strategy is still a subject of debate, but the XMM-XXL survey reaches a depth comparable to the deepest planned eROSITA exposure (⁠|$F_{0.5{\rm -}2 \,\rm {keV}}\sim 4.0\times 10^{-15}\rm \,erg\,s^{-1}\,cm^{-2}$|⁠) near the ecliptic poles covering a solid angle of ≳ 500 deg².

The eROSITA X-ray catalogue of AGNs is going to be the target of the spectroscopic follow-up programme SPIDERS. This programme belongs to the eBOSS survey (Dawson et al., 2015) in the SDSS-IV to follow-up ROSAT, XMM–Newton, and eventually eROSITA sources with the optical BOSS spectrograph (see http://www.sdss.org). It aims to be one of the largest optical spectroscopic follow-up surveys of X-ray-selected AGNs. In the following, we will provide both a scientific as well as technical forecast for eROSITA and SPIDERS based on our spectroscopic XMM-XXL data set.

7.1 Spectroscopic completeness

In order to predict the spectroscopic completeness of SPIDERS (and other future eROSITA follow-up programmes), we first calculate, for the XMM-XXL survey, the number densities N of:

• all X-ray sources, and of those with

• matched SDSS counterparts (LR_XMM,SDSS > 1.5),

• r-band within 15 < r < 22.5 mag,

• BOSS follow-up, or

• reliable redshift (⁠|$\rm {Z\_CONF}=3$| and |$\rm {Z\_CONF}=30$|⁠).

In Fig. 16, we show the spectroscopic completeness of our data set. The fractions correspond to the number of sources with reliable redshifts over the total number of X-ray sources (yellow line), sources with matched SDSS counterparts (green line) and sources within the r-band limits (blue line). We indicate the fractions at different X-ray fluxes and highlight the eROSITA survey depths. After the correction for fibre collision (dashed lines), the spectroscopic completeness for all BOSS targets (15 < r < 22.5 mag) ranges from 85 per cent at XMM-XXL depth to 96 per cent for eROSITA depth. The spectroscopic completeness for all X-ray sources with optical counterparts (LR_XMM,SDSS > 1.5) drops at |$F_{0.5{\rm -}2\,\,\rm keV}\gtrsim 4\times 10^{-14}\,\rm erg\,s^{-1}$|⁠. This is due to the counterpart distribution, which shifts towards brighter optical magnitudes with shallow X-ray fluxes (see Fig. 2) and reduces the number of BOSS targets within the r-band threshold (r ≤ 15-17 mag). The overall spectroscopic completeness for all X-ray sources at deep X-ray flux limits reflects both the sensitivity limits of SDSS for the optical counterparts and the faint r-band threshold of the BOSS spectrograph.

The optical classification of the AGNs also correlates with the X-ray flux. For the XMM-XXL depth, our AGN sample comprises 73 per cent BLAGN1, 20 per cent NLAGN2/NLAGN2cand and 6 per cent eAGN. For the eRASS:8, these ratios change to 80 per cent BLAGN1, 16 per cent NLAGN2/NLAGN2cand and 4 per cent eAGN.

7.2 X-ray detection fraction of XDQSO and WISE-selected AGNs

Following the explained procedures from Sections 5.1 and 5.2, we derived the number densities and X-ray detection fraction of XDQSO (see Fig. 17) and and WISE (see Fig. 18) for an X-ray-selected sample at eROSITA fluxes.

8 CONCLUSIONS

We present and publicly release one of the largest contiguous catalogues of X-ray-selected and spectroscopically observed AGNs to date. It provides a unique data set to study the optical properties of X-ray-selected AGNs and also serves as a pilot study for the eROSITA follow-up programme SPIDERS in SDSS-IV.

The survey contains 8445 point-like X-ray sources covering an area of ∼ 18 deg² in the XXM-XXL north area reaching down to a flux of |$F_{0.5-10\,\,\rm keV}>10^{-15}\,\rm erg\,cm^{-2}\,s^{-1}$|⁠. They have been cross-matched to their SDSS and WISE counterparts via the maximum-likelihood ratio method. The BOSS spectrograph followed up 3042 sources within 15 < r_{SDSS psf/model} < 22.5 mag and after visual inspection, we obtained 2578 sources with reliable redshifts. The sample covers a redshift range of 0 < z < 5 and a luminosity range of |$2\times 10^{39}<L_{0.5{\rm -}2\rm \,keV}<4\times 10^{45}\rm \, erg\,s^{-1}$|⁠.

For the study of the AGN properties of our X-ray-selected sample, we introduce a spectral classification method which is based on optical emission line properties provided by the BOSS pipeline. We analysed properties, such as the line widths of AGN-induced and star-formation-induced emission lines and finally derived the following classes for a subset of 2570 spectra: 1787 BLAGN1, 271 NLAGN2, 284 NLAGN2cand, 78 eAGN-SFG, 93 eAGN-ALG, 85 stars and 2 BL Lac. 57 sources cannot be classified, because of missing emission line information. We compared the X-ray AGN selection to common AGN selection using X-ray, optical (Bovy et al. 2011), and IR colours (Stern et al. 2012; Assef et al. 2013).

In the following, we conclude with the main scientific outcomes of the analysis of X-ray-selected AGNs in the XMM-XXL north.

• The bimodal FWHM-distribution of optical AGN-induced emission lines widths (e.g. Hβ and Mg ii) clearly separates the population of X-ray-selected AGNs into BLR emitters and NLR emitters. The minimum of the FWHM distribution is at |$\rm FWHM\sim 1000\rm \,km\,s^{-1}$|⁠.

• The X-ray selection probes a wide variety of AGNs with respect to the obscuring material along the line of sight and the contribution of the passive or active host galaxy. It allows for a selection of particular classes, such as optically unobscured BLAGN1 with strong host galaxy contribution, optically obscured NLAGN2, and optically elusive AGNs. Because of the r-band magnitude limits for the optical spectroscopy, we are biased against optically faint sources, which mainly affects optically obscured NLAGN2.

• Applied to our X-ray-selected sample, we find that the optical AGN selection via the XDQSO targeting algorithm (Bovy et al. 2011) is, by construction, biased towards optically point-like sources and selects BLAGN1 with weak host features in the entire redshift range of the sample. The WISE colour AGN selections from Assef et al. (2013) and Stern et al. (2012) applied to our sample preferentially selects BLAGN1 at z < 3 with weak hosts and red W1 − W2 colours.

• In the coming years, multi-object optical spectrographs with characteristics similar to BOSS or with higher performance (e.g. 4MOST, de Jong et al. 2014; DESI, http://desi.lbl.gov) will be able to provide highly complete, and very efficient follow-up programmes for the upcoming eROSITA all-sky X-ray surveys, bringing the study of AGN populations to an unprecedented level of statistical accuracy.

Funding for SDSS-III has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, and the U.S. Department of Energy Office of Science. The SDSS-III web site is http://www.sdss3.org. SDSS-III is managed by the Astrophysical Research Consortium for the Participating Institutions of the SDSS-III Collaboration including the University of Arizona, the Brazilian Participation Group, Brookhaven National Laboratory, Carnegie Mellon University, University of Florida, the French Participation Group, the German Participation Group, Harvard University, the Instituto de Astrofisica de Canarias, the Michigan State/Notre Dame/JINA Participation Group, Johns Hopkins University, Lawrence Berkeley National Laboratory, Max Planck Institute for Astrophysics, Max Planck Institute for Extraterrestrial Physics, New Mexico State University, New York University, Ohio State University, Pennsylvania State University, University of Portsmouth, Princeton University, the Spanish Participation Group, University of Tokyo, University of Utah, Vanderbilt University, University of Virginia, University of Washington and Yale University. MB acknowledges support from the FP7 grant ‘eEASy’ (CIG321913).

We thank Adam Bolton (University of Utah), Damien Coffey (MPE), Jeremy Sanders (MPE), Alina Streblyanska (IAC), Michael di Pompeo (Dartmouth College) and Adam Myers (University of Wyoming) for their support. Furthermore, we express our gratitude towards anonymous reviewers from MNRAS for their constructive comments.

REFERENCES

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Appendix E. Catalogue.

(Supplementary Data).

Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

APPENDIX A: SPECTROSCOPIC TARGET SELECTION

The spectroscopic follow-up data come from two BOSS ancillary programmes comprising seven BOSS plates and furthermore the released DR10 catalogue. We indicate the source numbers of every spectroscopic targets selection step in Fig. A1, provide additional technical information for the different programmes and list the plate centre coordinates of the dedicated ancillary programmes in the Table A1.

The first SDSS ancillary project is named ‘TDSS/SPIDERS/eBOSS Pilot Survey’, the BOSS targeting primary programme is called ‘TDSS/SPIDERS/BOSS Pilot Survey’ and the Ancillary Bit Numbers are 23, 24 and 25. The BOSS spectrograph observed the dedicated plate 6369 in 2012 October without the use of washers.

The second SDSS ancillary project is named ‘Follow-up spectroscopy of wide-area XMM fields’. The observations of the plates 7235, 7236, 7237, 7238took place in 2013 November and 2014 January without the use of washers, resulting in 2357 spectra. The BOSS targeting programme is called ‘Wide-Area XMM fields’ and the Ancillary Bit Numbers are 32 and 33.

The added sources with BOSS spectroscopy from the DR10 are located both outside and inside the ancillary plates footprint as indicated in Fig. 4. There are 30 spectra from the plate 6369 of the TDSS targets which did not meet the flux threshold from the first ancillary programme, but are within the flux criteria of the second ancillary programme.

APPENDIX B: VISUAL REDSHIFT DETERMINATION

In order to determine critical parameters which are correlated to redshift failures of the BOSS pipeline, we first checked all spectra of the first ancillary programme (Plate Number: 6369). Assuming this data set to be a good representative for all our targets, we applied this knowledge to the second ancillary programme and only inspected the subset of spectra most likely to be problematic. In total, we inspected ∼1200 spectra. The remaining spectra were part of the visual inspection process of the BOSS QSO group and are published in Paris et al. (in preparation).

We evaluate the redshift provided by the pipeline Z_BOSS by visual inspection and assign both a new redshift Z and confidence parameter Z_CONF (see Section 3.2). The spectra, whose visual inspection redshift coincide with the BOSS pipeline redshifts, keep their initial redshifts |$\rm {Z}=\rm {Z\_BOSS}$|⁠. For the spectra with wrong BOSS pipeline redshifts, we rerun the BOSS pipeline (spreduce1d.pro) in a small redshift range suggested by the visual inspection and assign a new redshift |$\rm {Z}\ne \rm {Z\_BOSS}$|⁠. The redshift confidence parameter Z_CONF contains information about the reliability of the visual redshift assignment. There are flags for ‘reliable’ and ‘not robust’ redshifts. Furthermore, we assign a flag for ‘bad spectra’ where no redshift can be obtained by visual inspection. For a subsample of spectra, the pipeline fails to determine a correct redshift as suggested by the visual inspection. In this case, we assign the flags ‘reliable visual redshift and pipeline failure’ or ‘not robust visual redshift and pipeline failure’. During the visual inspection, we mark all stars and BL Lac with the STAR/BLLAC parameter. Their precise redshift determination is not included in our visual evaluation, because stars and BL Lacs are not in the focus of this study. In Table B1, we give an overview about the redshifts obtained from the BOSS pipeline after visual inspection.

In the following, we want to highlight the critical spectral characteristics which lead to a redshift failure by the BOSS pipeline.

These characteristics are

• low-redshift quality: |$\rm {ZWARNING}>0$| (583 spectra with 293 failures),

• large redshift error: |$\rm {ZERR\_BOSS}>0.01$| (56 spectra with 45 failures),

• low S/N ratio: SN_MEDIAN_ALL <1.6 (837 spectra with 272 failures),

• very high redshifts: |$\rm {Z\_BOSS}>4$| (49 spectra with 39 failures),

• very low redshifts: |$\rm {Z\_BOSS}<0.05$| (131 spectra with 37 failures).

  The items (ii)–(v) are often associated with |$\rm {ZWARNING}>0$|⁠. The redshifts failures at high redshifts (iv) are often caused by emission line confusion of Lyα, Mg ii, CIV, C iii] and Hα. We have to point out that problematic sources can also be caused by external influences, such as:

• blending of a neighbour object (1 out of 423 spectra from plate 6369)

• instrumental fibre problems (4 out of 423 spectra from plate 6369).

These features are not correlated with the S/N ratio or the redshift quality flag. The influence of neighbour objects can be reduced by cross-checking with the optical images. There are some fibres with instrumental problems resulting in spectra which are missing correct spectroscopic information in parts of the observed wavelength range. These spectra provide redshift information for the objects, but classification is not possible. For our catalogue, we will exclude these objects.

The group of spectra with correct and robust BOSS redshifts is defined by |$\rm {ZWARNING}=0$|⁠, SN_MEDIAN_ALL >1.6 and |$0.05<\rm {Z\_BOSS}<4$|⁠, and we obtained an extremely high reliability of 99 per cent for the BOSS pipeline results. Therefore, within this parameter space, we did not perform any visual inspection for the spectra of the secondary ancillary programme, nor for all added spectra from the DR10.

APPENDIX C: OPTICAL LINE DIAGNOSTIC DIAGRAMS

BPT-Nii-diagram. For spectra with significant Hβ, [O iii], Hα and [N ii], we follow the BPT-N ii selection criteria (Baldwin et al. 1981; Kauffmann et al. 2003; Kewley et al. 2006; Meléndez et al. 2014) with the ratios [O iii]/Hβ versus [N ii]/Hα. These ratios are less sensitive to reddening because of the small wavelength separation of the relevant emission lines. In addition, we use EW which removes the direct reddening dependence. The BPT-N ii-diagram only classifies objects at a redshift of 0 < z < 0.54, because Hα and [N ii] are shifted out of the BOSS wavelength range for higher redshifts.

Blue-Oii-diagram. The [O ii]/Hβ emission line ratio diagram identifies objects with significant [O ii]-doublet (EW of λ = 3726.032, 3728.815 Å) at higher redshifts (0 < z < 1.08) (Lamareille et al. 2004; Lamareille 2010). For the sake of simplicity, we refer to the [O ii] doublet by the name of ‘[O ii]’. The emission lines Hβ and [O ii] have a very large separation in wavelength space and the reddening influences the emission lines as well as the underlying continuum differently. Therefore, even the EW is effected by reddening. For the separation of AGN and SFG, we choose the demarcation line of Lamareille (2010):

APPENDIX D: eROSITA

In Table D1, we provide the number densities and fibre collision correction for the X-ray-selected AGNs in the XMM-XXL north at different flux depths, including eROSITA depths.