The XMM-SSC survey of hard-spectrum XMM–Newton sources – I. Optically bright sources

Abstract

1 INTRODUCTION

Deep surveys with Chandra and XMM–Newton have now resolved the majority of the extragalactic X-ray background (XRB) at <5 keV into point sources (Rosati et al. 2002; Alexander et al. 2003; Bauer et al. 2004; Worsley et al. 2004). At the bright flux limits that were probed with Einstein and ROSAT surveys, the soft X-ray sky is dominated by broad-line active galactic nuclei (BLAGN) with soft X-ray spectra (Maccacaro et al. 1988; Mittaz et al. 1999). At faint fluxes, the bulk of the sources have hard X-ray spectra (Brandt et al. 2001; Giacconi et al. 2001), as required to produce the overall XRB spectral shape. XRB synthesis models predict that these faint, hard sources are intrinsically absorbed AGN (e.g. Setti & Woltjer 1989; Comastri et al. 1995; Gilli, Salvati & Hasinger 2001). These absorbed sources are likely to be the dominant AGN population: they are responsible for the majority of the XRB energy density and probably outnumber unabsorbed AGN by a factor of 4 or more (Fabian & Iwasawa 1999), as is found in optical and X-ray studies of local AGN (Maiolino & Rieke 1995; Risaliti, Maiolino & Salvati 1999). Optical/infrared imaging and spectroscopy of the faint, hard-spectrum sources, while consistent with the absorbed AGN hypothesis, reveals a heterogeneous population. Many of these sources are narrow-line AGN (Seyfert type 2 galaxies and type 2 QSOs), others show no emission lines whatsoever, while some are optically unobscured, BLAGN (e.g. Mainieri et al. 2002; Barger et al. 2003; Szokoly et al. 2004; Page et al. 2006b). This variety suggests that there is a large range of gas-to-dust ratios within the absorbed population.

There are a variety of important issues regarding the absorbed population which remain to be addressed. First, the relationship between the optical spectroscopic characteristics and the level of X-ray absorption is not fully understood (e.g. Maiolino et al. 2001; Page, Mittaz & Carrera 2001; Boller et al. 2003; Barcons, Carrera & Ceballos 2003a; Caccianiga et al. 2004; Carrera, Page & Mittaz 2004). Nor is it established whether the fraction of absorbed AGN depends on redshift and/or luminosity: several recent studies suggest that the fraction of absorbed AGN decreases with luminosity and/or redshift (e.g. Barger et al. 2003; Cowie et al. 2003; Steffen et al. 2003; Ueda et al. 2003; La Franca et al. 2005), while other studies suggest that the absorbed fraction is independent of luminosity (e.g. Treister et al. 2004; Dwelly et al. 2005; Dwelly & Page 2006; Tozzi et al. 2006). Indeed, this issue has been debated since at least the early 1980s (e.g. Lawrence & Elvis 1982; Mushotzky 1982; Maccacaro, Gioia & Stocke 1984). Another particularly important question is whether the absorbed AGN population is wholly related to the unabsorbed population by AGN geometric unification schemes (Antonucci 1993). BLAGN with significant X-ray absorption but little or no optical/ultraviolet (UV) obscuration (e.g. Barr et al. 1977; Elvis et al. 1994; Akiyama et al. 2000; Page et al. 2001) have proven particularly challenging to fit within such schemes. Alternatively, parts of the absorbed population may be physically distinct from the unabsorbed population, as predicted by some evolutionary models (Sanders et al. 1988; Fabian 1999; Elvis, Marengo & Karovska 2002; Franceschini, Braito & Fadda 2002; Page et al. 2004).

In this paper we present a sample of serendipitous XMM–Newton sources selected to have hard X-ray spectra in the 0.5–4.5 keV energy range. This sample is well suited for the study of the X-ray and optical characteristics of the X-ray/optically bright part of the X-ray-absorbed AGN population. We describe the X-ray and optical selection of the sample in detail in Section 2, and in Sections 3 and 4 we describe the optical spectroscopic observations and the construction of the X-ray spectra. Our results are presented in Section 5 and their implications for the absorbed AGN population are discussed in Section 6. Throughout this paper we assume H₀= 70  km s⁻¹ Mpc⁻¹,  Ω_m= 0.3 and Ω_Λ= 0.7. We define a power-law spectrum such that dN/dE=AE^−Γ where N is the number of photons, E is photon energy, Γ is the photon index and A is the normalization.

2 SAMPLE SELECTION

Thanks to XMM–Newton's exceptional throughput and large field of view (30 arcmin diameter), a serendipitous XMM–Newton catalogue is the obvious resource with which to find bright examples of the absorbed AGN population. Our serendipitous catalogue was constructed by concatenating the source lists from 100 pipeline processed XMM–Newton fields. This catalogue preceded the public release of the 1XMM catalogue, which is about six times larger, (Watson et al. 2003) but contains similar information about each source. The targets of the XMM–Newton observations were excluded from the catalogue. Hard-spectrum sources are heavily outnumbered at bright fluxes by soft-spectrum, unabsorbed AGN, even in the 2–10 keV band. Therefore we have selected our sample from the catalogue using the 2–4.5 versus 0.5–2 keV hardness ratio. We refer to this ratio as ‘HR2’ to follow the convention used in the 1XMM catalogue. HR2 is defined as

1

where CR_2–4.5 is the vignetting-corrected count rate in the 2–4.5 keV range and CR_0.5–2 is the vignetting-corrected count rate in the 0.5–2 keV range. HR2 is calculated independently for each of the three EPIC cameras. The specific criterion used to select the sample is

2

where σ_HR2 is the 68 per cent uncertainty on HR2. The HR2 threshold of -0.3 corresponds approximately to the slope of the XRB in the same energy range, so that only objects with spectra harder than the XRB will be picked up with this criterion. This range of HR2 also corresponds to the most heavily absorbed AGN typically found in XMM–Newton surveys (Caccianiga et al. 2004; Della Ceca et al. 2004), the numbers and redshifts of which place significant constraints on XRB synthesis models (Dwelly et al. 2005). For inclusion in the sample, we require a source to have a detection likelihood (DET_ML in 1XMM) > 7.51 in the 2–4.5 keV band, 2–4.5 keV flux >10⁻¹⁴ erg cm⁻² s⁻¹ Galactic N_H < 5 × 10²⁰  cm⁻²,  |b| > 30°, and satisfy equation (2) in at least one of the three EPIC cameras. The DET_ML threshold of 7.5 corresponds to a probability of being a background fluctuation of 5 × 10⁻⁴, and a spurious source fraction of <1 per cent (Watson et al. 2003). From a total of 9340 sources with DET_ML > 7.5 found in the 100 XMM–Newton fields used here, we obtained a sample of 136 sources which meet the selection criteria for our hard-spectrum sample.

The sources were categorized according to optical magnitude to facilitate our optical identification programme. Those sources with counterparts on SuperCOSMOS digitizations of the Palomar or United Kingdom Schmidt Telescope sky-survey plates within 6 arcsec of the X-ray position will be referred to as the optically bright sample; this amounts to 73 sources, and provides the target list for our spectroscopic identification programme on 4-m telescopes. Those sources without sky-survey counterparts2 but which have counterparts in deeper imaging form the target list for our spectroscopic identification programme on 8–10 m telescopes and are considered the optically faint sample. Our identification programme for the optically faint sample is ongoing. The final category of sources (‘blank-field sources’) are those for which no optical counterpart is found to r∼ 23. Discussion of the optically faint and blank-field samples is deferred to a future paper pending the results of the 8–10 m spectroscopic campaign.

The hardness ratios, HR2 and HR3, for the full sample of hard-spectrum sources are shown in Fig. 1, where HR3 is the 4.5–7.5 versus 2–4.5 keV hardness ratio, defined as

3

The optically bright sources are shown as large dots, while the optically faint and blank-field sources are shown as small dots. The optically faint and optically bright parts of the sample have similar hardness ratio distributions: the two distributions are indistinguishable at the 90 per cent confidence level according to the two-dimensional Kolmogorov–Smirnov test (Fasano & Franceschini 1987). The underlying grey-scale and contours show the hardness ratio distribution for the ∣b∣ > 20° 1XMM catalogue (Watson et al. 2003). The peak in the underlying distribution at (−0.6,  −0.5) corresponds to the typical spectra of unabsorbed BLAGN. While such sources dominate the 1XMM catalogue, (and all medium-depth, flux-limited samples of serendipitous XMM–Newton sources) the distribution shows a significant tail stretching to larger values of HR2. The hard-spectrum sample described in this paper follows the underlying distribution of 1XMM sources well for HR 2 > −0.2, and so is ideal for the study of this hard-spectrum tail of the population.

Figure 1

HR2 and HR3 hardness ratio distribution of the optically bright sources (large dots) and optically faint/blank-field sources (small dots) from the hard-spectrum sample. The underlying grey-scale and contours show the pn-camera hardness ratio distribution of the entire high Galactic latitude (∣b∣ > 20°) 1XMM catalogue, and have a logarithmic scaling (Watson et al. 2003).

Open in new tabDownload slide

The optical magnitudes and X-ray fluxes of the hard-spectrum sources are shown in Fig. 2, and in this figure we also show the interval ∣log₁₀(f_X/f_R)∣ < 1, where f_X and f_R are X-ray flux and optical flux, respectively, as defined by McHardy et al. (2003).3 The majority of hard-spectrum sources lie within this interval, as do most of the X-ray-selected population at this flux level (Stocke et al. 1991).

Figure 2

2–4.5 keV X-ray flux versus optical r-band magnitude for the hard-spectrum sources. The solid line corresponds to the ratio f_X/f_R= 1 and the dashed lines indicate the ±1 dex interval, where f_X/f_R is the ratio of X-ray to optical flux as defined in McHardy et al. (2003). The 2–4.5 keV fluxes (f_2–4.5) were determined from the 2–4.5 keV count rates, assuming a photon index Γ= 1.7.

Open in new tabDownload slide

3 OPTICAL SPECTROSCOPIC OBSERVATIONS

Optical spectra were obtained in two observing runs. The first run took place on the nights of 2003 January 27 and 28 at the William Herschel Telescope (WHT) on La Palma. Observing conditions were variable, but most of the data were taken in conditions of good seeing (∼1 arcsec) and sky transparency. Observations were made with the ISIS dual arm spectrograph using a 1.5-arcsec slit oriented at the parallactic angle, and employing the 5400 Å dichroic. On the blue arm we used the R300B grating, covering the wavelength range from 3000 to 5700 Å at 4.9-Å resolution (measured from arc lines taken through the slit) and on the red arm we used the R158R grating, covering the wavelength range 5000–10 000 Å at 10.1-Å resolution. Spectra of CuNe and CuAr arc lamps were used to calibrate the wavelength scale. Flux calibration, and approximate correction for the telluric absorption bands at ∼6900 and ∼7600 Å were achieved through observations of G191−B2B and BD+33 2642. Data were reduced using standard iraf routines. The reduced, flux-calibrated spectra from the red and blue arms were then merged to form a single spectrum for each target.

The second observing run took place at the New Technology Telescope at La Silla on the nights of the 2003 May 2 and 3 in conditions of variable seeing and sky transparency. Observations were made using the EMMI spectrograph using a 1-arcsec slit at the parallactic angle and grism #3, providing 8-Å resolution and a wavelength range of 3800–9000 Å. Spectra of He and Ar arc lamps were used to calibrate the wavelength scale. Observations of the spectrophotometric standards LTT4364 and LTT2415 were used for relative flux calibration.

In total, spectra of the optical counterparts to 42 hard-spectrum sources were observed. The spectra are shown in Fig. 3. The spectroscopic completeness is 100 per cent for the observed sample, that is, an adequate spectrum yielding a reliable redshift was obtained for every optical counterpart that was observed. Apart from the selection of counterparts from sky-survey plates, we did not bias the spectroscopic observations to bright sources. To demonstrate this, Fig. 4 compares the optical magnitude distributions for the optically bright sources which were and were not observed spectroscopically; the two distributions are indistinguishable according to the Kolmogorov–Smirnov test, which gives the probability that they were both drawn from the same distribution as 46 per cent. Hence our spectroscopically identified sources are a representative and statistically complete sample of the optically bright (r<21), hard-spectrum population.

Open in new tabDownload slide

Open in new tabDownload slide

Open in new tabDownload slide

Open in new tabDownload slide

Open in new tabDownload slide

Figure 3

Optical spectra of hard-spectrum serendipitous sources.

Figure 4

Magnitude distributions for sources with r < 21 which were (upper panel) and were not (lower panel) observed during our optical spectroscopic observations.

Open in new tabDownload slide

4 CONSTRUCTION OF X-RAY SPECTRA

We extracted X-ray spectra for every source using the available XMM–Newton EPIC data. For sources which are present in more than one XMM–Newton observation in the public archive, we have used all the available data to obtain the highest possible signal-to-noise ratio. Table 1 gives the XMM–Newton observations used to construct the X-ray spectrum of each source. The EPIC data were reduced using the sas version 5.4. The XMM–Newton point spread function changes with off-axis angle. Therefore source spectra were extracted from elliptical source regions with sizes and orientations depending on the fluxes and off-axis angles of the sources. Typically, these regions have major axes of ∼18 arcsec and minor axes of ∼10 arcsec. Background spectra were extracted from annular regions, of 2.5 arcmin outer radius, centred on the source, and with all significant X-ray sources excised. Response matrices and the appropriate effective area files were computed using the sas tasks rmfgen and arfgen, respectively. For each source the EPIC spectra were combined to form a single spectrum using the method of Page, Davis & Salvi (2003a). Source spectra were grouped using the heasoft tool grppha to ensure the minimum number of counts per bin given in Table 3 (usually 20) before fitting.

5 RESULTS

5.1 Optical spectra

The optical spectra are shown in Fig. 3, and the identifications, redshifts and emission-line characteristics are given in Table 2. The distribution of offsets between the X-ray sources and optical counterparts is similar to those found in other XMM–Newton surveys of similar depth (e.g. Barcons et al. 2002; Fiore et al 2003), with 90 per cent of optical counterparts lying within 3 arcsec of the X-ray source. Note that two of the sources, XMM J122656.53+013125.2 and XMM J134656.75+580316.65 are also members of the XMM–Newton Bright Serendipitous Survey and are also reported in Della Ceca et al. (2004), while one of the sources, XMM J021939.22−051133.7, is also a member of the XMM–Newton Medium Sensitivity Survey and is also reported in Barcons et al. (2007). We have split the objects into four categories. Objects which have any permitted emission lines with measured full width at half-maximum (FWHM)4 >1000 km s⁻¹ have been classified as BLAGN; this class includes QSOs and type 1 Seyfert galaxies. Galaxies which only show emission lines with an FWHM < 1000 km s⁻¹ have been classified as narrow emission-line galaxies (NELGs). Galaxies in which we do not detect any emission lines have been classified simply as ‘galaxies’. Finally, one object has the spectrum of a Galactic star.

As can be seen in Fig. 3 and Table 2, the sources all have z < 0.8, and majority (31/42) of the sources are classified as NELGs. In all the NELGs we observe at least one of Hα, [O iii] 4954 Å/5007 Å or [O ii] 3727 Å. The three objects classified as galaxies have 0.3 < z < 0.4, similar to the mean redshift for the NELGs, 〈z〉= 0.29 ± 0.03, so their galaxy classification is not due to these emission lines being redshifted outside the observed spectral window. The seven BLAGN, however, have a higher mean redshift than either the NELGs or galaxies, 〈z〉= 0.53 ± 0.08. More than half of the BLAGN are identified by a broad Mg ii 2798 Å line.

Two sources (XMM J122656.53+013125.2 and XMM J140057.10−110120.6) have emission lines with 1000 < FWHM < 2000 km s⁻¹, and therefore might conceivably be narrow-line Seyfert 1 galaxies (Osterbrock & Pogge 1985). In XMM J122656.53+013125.2 the Mg ii line has a measured FWHM of 1600 km s⁻¹, and the other lines all have FWHM < 1000 km s⁻¹. However, the large ratio of [O iii]/Hβ rules out a narrow-line Seyfert 1 galaxy (Shuder & Osterbrock 1981), and suggests instead that the broad-line region is heavily reddened. The Mg ii profile could be a composite of emission from the narrow-line region and a reddened line from the broad-line region. For XMM J140057.10−110120.6 only one emission line is detected (Hα) with a measured FWHM of 1900 km s⁻¹. The red, galaxy-dominated continuum leads us to suspect that the nucleus and broad-line region are also reddened in this object, and that the Hα line profile is likely to be a composite of emission from the narrow-line region and a reddened broad line.

5.2 Basic X-ray spectral modelling

The X-ray spectra are shown in Fig. 5. As a starting point for the spectral modelling we considered an absorbed power-law model, corresponding to an AGN viewed through a screen of cold material. For each object we included a zero-redshift absorber with N_H fixed at the Galactic value found from 21-cm observations (Dickey & Lockman 1990) and a cold absorber at the redshift of the source with N_H as a free parameter in the fit. In fits of this type, particularly in low–moderate signal-to-noise ratio data, the power-law photon index Γ and absorbing column density N_H are somewhat degenerate, and this can lead to large, highly coupled uncertainties on both parameters. Therefore we began our spectral fitting with Γ fixed at a value of 1.9, which is the mean slope found for unobscured BLAGN in XMM–Newton surveys (Mainieri et al. 2002; Piconcelli et al. 2002; Caccianiga et al. 2004; Mateos et al. 2005a; Page et al. 2006b).

Open in new tabDownload slide

Open in new tabDownload slide

Open in new tabDownload slide

Figure 5

X-ray spectra of the sources (data points) together with absorbed Γ= 1.9 power-law models (black stepped lines). Both model and data have been divided by the effective area as a function of energy, and the spectra are displayed as E×F_E so that an unabsorbed power law with Γ= 2 would be a horizontal line. When the absorbed Γ= 1.9 power-law model does not provide an acceptable fit to the data, the best alternative model is shown in grey. For XMM J021908.37−044731.4, XMM J021939.22−051133.7 and XMM J094239.79+465005.3 the grey model includes a mekal component, for XMM J083139.11+524206.2, XMM J133026.09+241356.7, XMM J150339.60+101605.6 and XMM J161736.21+122901.5 the grey model includes an unabsorbed power-law component, for XMM J080625.35+244326.0 the grey model has an ionized absorber, and for XMM J122656.53+013125.2 the grey model has a power-law index of Γ= 1.5.

The results of these fits are given in Table 3; the uncertainties listed in Table 3 and all subsequent tables are 95 per cent for one interesting parameter (Δχ²= 4). We consider a null hypothesis probability of 1 per cent to be an appropriate threshold for rejection of the spectral model, because at this level, and for our sample size, we expect less than one rejection by chance if the form of the model is correct. Of the 42 sources, 33 are therefore acceptably fitted with this model, while nine objects (∼20 per cent) are not acceptably fitted. In seven of the nine cases, the poor χ² is due to an excess of flux in the lowest energy channels with respect to the model (see Fig. 5).

5.3 Modelling the nine sources with complex X-ray spectra

We now examine in more detail the X-ray spectra of the nine sources which were not well fitted with a Γ= 1.9 power law and neutral absorption. Before considering spectral models with an additional soft X-ray component, we investigate the possibility that the poor χ² could be a consequence of our decision to fix Γ= 1.9 in the power-law component. There is certainly a range of photon indices in the unabsorbed AGN population (Mateos et al. 2005a,b; Page et al. 2006b), and the fit with fixed Γ could result in significant residuals for AGN which are outliers in this distribution. Therefore, for the nine objects which had unacceptable χ²/ν, we refit the spectra, this time allowing the photon index of the power law to vary as a fit parameter. The results of these fits are shown in Table 4. Four of the objects are acceptably fitted with this model, but only one of these, XMM J122656.53+013125.2, has a value of Γ consistent with a ‘normal’ AGN spectrum. The other three objects (XMM J161736.21+122901.5, XMM J094239.79+465005.3 and XMM J080625.35+244326.0) for which this fit yields a reasonable χ²/ν have best-fitting photon indices which are exceedingly hard, Γ < 1.2. Such photon indices are well outside the range of photon indices normally observed in unabsorbed AGN (>3.5σ outliers from the distribution, Mateos et al. 2005a), and therefore we consider that the variation in AGN photon indices only provides a plausible explanation for the poor fit of the absorbed Γ= 1.9 power-law model to the X-ray spectrum of XMM J122656.53+013125.2.

For the nine objects not well fitted in Section 5.2, we now consider a model of photoelectric absorption in which the absorber is ionized, rather than cold, as ionized absorption is often observed in nearby Seyfert 1 galaxies (e.g. Reynolds 1997; George et al. 1998; Blustin et al. 2005). For this we use the ‘absori’ model in xspec, in which the ionization state of the absorber is defined by the ionization parameter ξ=L/nr², where L is the ionizing luminosity of the source, n is the number density of the absorber and r is the distance of the absorber from the ionizing radiation. The photon index of the power law was fixed at Γ= 1.9, and the temperature of the absorber was fixed at 3 × 10⁴  K (the fits are not sensitive to this parameter). The results are given in Table 5. This model produces acceptable χ²/ν for 4/9 of the sources, but does not give acceptable fits for the other five objects.

Next, we considered two different models that contain an additional component, which is bright at the soft X-ray end of the spectrum. To minimize the number of free parameters in the fit, we maintain a fixed Γ= 1.9 for the primary power-law component in these fits.

In the first model, we include a second power law with Γ= 1.9, but which is not attenuated by the column of cold gas intrinsic to the object, and represents a component of the primary emission which is scattered into our line of sight, bypassing the absorber. The photon indices of both power laws were frozen at a value of 1.9 in this model. Although such a model is lacking in experimental verification from observations of nearby AGN, it has a long heritage (e.g. Holt et al. 1980; Turner et al. 1997; Franceschini et al. 2003; Caccianiga et al. 2004), and its functional form provides a simple and convenient parametrization of the relative contributions of the primary, absorbed component and the component which provides the excess flux at soft energies. The results of fitting this model are listed in Table 6. All but two objects are fitted acceptably using this model. We have also used this model to determine 95 per cent upper limits for the contribution of any soft component in the sources which are acceptably fitted with a simple absorbed power-law model, by increasing the contribution of the unabsorbed power law until Δχ²= 4 with respect to the model without the unabsorbed power law; these upper limits are included in Table 3.