An XMM–Newton spectral survey of 12 μm selected galaxies – I. X-ray data Free

Abstract

We present an X-ray spectral analysis of 126 galaxies of the 12 μm galaxy sample. By studying this sample at X-ray wavelengths, we aim to determine the intrinsic power, continuum shape and obscuration level in these sources. We improve upon previous works by the use of superior data in the form of higher signal-to-noise ratio spectra, finer spectral resolution and a broader bandpass from XMM–Newton. We pay particular attention to Compton thick active galactic nucleus (AGN) with the help of new spectral fitting models that we have produced, which are based on Monte Carlo simulations of X-ray radiative transfer, using both a spherical and torus geometry, and taking into account Compton scattering and iron fluorescence. We use this data to show that with a torus geometry, unobscured sightlines can achieve a maximum equivalent width of the Fe Kα line of ∼150 eV, originally shown by Ghisellini et al. In order for this to be exceeded, the line of sight must be obscured with N_H > 10²³ cm⁻², as we show for one case, NGC 3690. We also calculate flux suppression factors from the simulated data, the main conclusion from which is that for N_H≥ 10²⁵ cm⁻², the X-ray flux is suppressed by a factor of at least 10 in all X-ray bands and at all redshifts, revealing the biases present against these extremely heavily obscured systems inherent in all X-ray surveys. Furthermore, we confirm previous results from Murphy & Yaqoob that show that the reflection fraction determined from slab geometries is underestimated with respect to toroidal geometries. For the 12 μm selected galaxies, we investigate the distribution of X-ray power-law indices, finding that the mean (〈Γ〉= 1.90^+0.05_−0.07 and σ_Γ= 0.31^+0.05_−0.05) is consistent with previous works, and that the distribution of Γ for obscured and unobscured sources is consistent with the source populations being the same, in general support of unification schemes. We determine a Compton thick fraction for the X-ray AGN in our sample to be 18 ± 5 per cent which is higher than the hard X-ray (>10 keV) selected samples. Finally we find that the obscured fraction for our sample is a strong function of X-ray luminosity, peaking at a luminosity of ∼10^42.43 erg s⁻¹.

1 INTRODUCTION

Studying active galactic nucleus (AGN) at X-ray wavelengths allows an assessment of the intrinsic source properties, such as the measurement of the continuum and the source power, due to the ability of X-rays to penetrate and measure large columns of obscuring material. This makes X-ray observations essential for testing AGN unification schemes (Awaki et al. 1991) which invoke obscuration to explain the perceived difference between Seyfert 1s and Seyfert 2s (Antonucci 1993; Urry & Padovani 1995). The N_H and Γ distributions for AGN are also an important ingredients for synthesis models of the cosmic X-ray background (XRB; Gilli, Comastri & Hasinger 2007).

Most of our knowledge of the X-ray properties of local AGN comes from X-ray selected samples, such as the Piccinotti et al. (1982) sample derived from HEAO 1 observations (e.g. Turner & Pounds 1989; Nandra & Pounds 1994). X-ray selection of AGN is in principle very effective as in this band the nucleus predominantly outshines the host galaxy, and can penetrate obscuration. The hard X-ray telescopes aboard INTEGRAL and Swift and their large sky coverage have allowed for the compilation of hard X-ray selected AGN catalogues. Beckmann et al. (2006) and Beckmann et al. (2009) have presented catalogues for INTEGRAL selected AGN, which total 199 after 5 yr of observations, and Tueller et al. (2008) and Tueller et al. (2010) have presented 229 Swift/Burst Alert Telescope (BAT) selected Seyfert galaxies after 22 months of observations. However, the discovery from these surveys seems to be that the fraction of Compton thick AGN detected is smaller than expected from XRB studies (Gilli et al. 2007). This point to the empirical fact that even the X-ray band is biased against the most obscured, Compton thick, AGN. X-ray selection at energies greater than 10 keV is less affected by Compton thick obscuration than 2–10 keV selection though.

An alternative to X-ray AGN selection is mid-infrared (MIR) selection where the primary radiation of the AGN is re-emitted after having been reprocessed by hot dust. The extended IRAS 12 μm galaxy sample (12MGS; Rush, Malkan & Spinoglio 1993, hereafter RMS) is a sample of 893 MIR selected local galaxies which contains a high fraction of AGN (13 per cent at the time of publication; RMS). The sample is taken from the IRAS Faint Source Catalogue, version 2 (FSC-2) and imposes a flux limit of 0.22 Jy, including only sources with a rising flux density from 12 to 60 μm (to exclude stars) and with a galactic latitude of |b| ≥ 25°. Being selected in the MIR this sample is also relatively unbiased against absorption, low-luminosity systems and ‘hidden AGN’. Spinoglio & Malkan (1989) showed that a wide variety of AGN types emit a constant fraction of their bolometric luminosities at this wavelength, and furthermore shown to be true for star-forming galaxies as well by Spinoglio et al. (1995). The 12MGS should be therefore also representative of the true number of different active galaxy types in relation to each other.

It is possible, though, that even at 12 μm, the AGN emission may be suppressed in the most heavily obscured nuclei. Pier & Krolik (1992) present radiative transfer modelling for the IR emission from dust tori for varying torus orientations. The most significant result of this study was that throughout the IR, the emission in the edge-on direction is weaker with respect to the face-on direction, and at 12 μm, the difference can be up to several orders of magnitude for optically thick tori. These models suggest that AGN selection at 12 μm is biased against heavily obscured nuclei. However, recently this was tested observationally by Horst et al. (2008). They used high-resolution 12.3 μm imaging of a sample of local Seyfert nuclei, with X-ray data to explore the ratio of the MIR to intrinsic X-ray nuclear luminosities in order to detect any difference between the obscured nuclei (Seyfert 2s) and unobscured nuclei (Seyfert 1s). As the X-ray luminosities have been corrected for absorption, L_X should represent the intrinsic emission from the nucleus. For smooth tori modelled by Pier & Krolik (1992), obscured systems should then present lower L_MIR/L_X ratios than unobscured systems due to the suppressed MIR flux seen in these systems. A difference of an order of magnitude for a difference in N_H of between ∼10²⁰ and ∼10²⁴ cm⁻² is predicted by the torus models, and should be measurable. However, this is not seen observationally, as Horst et al. (2008) find no significant difference between Seyfert 1s and Seyfert 2s, or even Compton thick nuclei. They argue from this that the torus may not be smooth, and instead clumpy, where the torus is seen to be optically thin, but the clumps are optically thick, though with a small volume filling factor. Most importantly for this work though, it suggests that 12 μm selection is in fact not heavily biased against obscuration. This therefore makes the 12MGS the ideal parent sample for study of obscuration and unification using X-ray observations and is as such the subject of this paper.

The 12MGS has previously been studied at X-ray wavelengths by Barcons et al. (1995) who used hard X-ray observations by HEAO 1, and at soft X-rays by Rush et al. (1996) with ROSAT data. Barcons et al. (1995) use the X-ray and IR data sets to show that most of the local X-ray emissivity originates from Seyfert galaxies and that most of the local 2–10 keV X-ray luminosity function between 10⁴² and 10⁴⁶ erg s⁻¹ can be accounted for by 12 μm emitting AGN. They also test the unification scheme and come to the conclusion that it must be modified to include the reduction of the covering fraction of the torus with increasing source luminosity, an observation first suggested by Lawrence (1991) which has come to be known as the ‘receding torus model’. Rush et al. (1996) conducted a spectral analysis of the 0.1–2.4 keV ROSAT data. For spectra with enough counts, they fit absorbed power laws, finding a median soft X-ray spectral index of 2.3 for all Seyfert types.

In this paper, we present an X-ray spectral analysis of the galaxies of the 12MGS with an observation by XMM–Newton, which consists of a heterogeneous mix of Seyferts, low-ionization nuclear emission-line region galaxies (LINERs) and star-forming galaxies. We focus on the X-ray spectral properties of the sources without regard for their optical spectroscopic types and use X-ray luminosity to select unambiguous AGN. In a companion paper though, we continue our investigation into this sample combining the X-ray data from this paper with optical spectroscopic and IR data. In Section 2 we present work on Monte Carlo simulations of X-ray radiative transport which we use in our spectral analysis. Section 3 describes our spectral fitting methods, and Section 4 gives our spectral fitting results. Finally we discuss all of these results in Section 5 and present our conclusions in Section 6.

2 NEW X-RAY SPECTRAL MODELS FOR HEAVILY OBSCURED SOURCES

In our X-ray spectral analysis of the 12MGS, we pay particular attention to whether sources are Compton thick, as the Compton thick fraction is particularly crucial for the XRB models and knowledge of the accretion history of the Universe. Compton thick AGN are most commonly identified in the 2–10 keV X-ray band by the measurement of a flat spectrum, where Γ < 1.0, and the equivalent width (EW) of the Fe Kα line has been measured to be high (EW > 1 keV; Matt, Brandt & Fabian 1996). This is typical of a reflection spectrum, produced by X-ray scattering from an optically thick surface, such as the accretion disc or wall of the torus in AGN. It is indicative of heavy obscuration of the central source as the reflection is dominating over the transmission of the intrinsic power law through the obscuring material.

In some cases, the transmission of the intrinsic power law can be detected in the 2–10 keV band for these heavily obscured sources. However, due to the extreme suppression of flux in this band at these high column densities, this is very difficult. Also at the high optical depths in these Compton thick sources the calculations for modelling the transmission spectrum become non-linear due to multiple scatterings. Thus, models describing simple attenuation of the spectrum by absorption and scattering become invalid.

Therefore, accurate models that correctly account for Compton scattering, especially multiple scatterings, are necessary when fitting the spectra of these sources. Currently, the best established model within xspec for correctly describing X-ray reprocessing by photoelectric absorption and Compton scattering is plcabs. This model describes the X-ray transmission from an isotropic source at the centre of a uniform spherical distribution of matter. It is valid up to energies between 10 and 18.5 keV and up to column densities of 5 × 10²⁴ cm⁻² (Yaqoob 1997). Unfortunately plcabs has a limited range of validity and does not include line emission, modelling the continuum only. Recently, two works by Ikeda, Awaki & Terashima (2009) and Murphy & Yaqoob (2009) have produced improved Monte Carlo models based on a toroidal obscuring medium similar to calculations by Ghisellini, Haardt & Matt (1994) which include Fe line emission.

Here we have carried out our own Monte Carlo simulations based mainly on the methods of Nandra & George (1994, hereafter NG94), but with a point source of X-rays located at the centre of both a spherical and toroidal distribution of gas emitting a power-law spectrum. These simulations follow the propagation of X-ray photons which interact with the medium via Compton scattering or photoelectric absorption, which can then result in fluorescent emission. The type of interaction is determined by the energy-dependent interaction cross-sections. We calculate the Compton scattering cross-section described in Rybicki & Lightman (1986) and calculate the photoelectric absorption cross-sections using data from Verner et al. (1996), using abundances from Anders & Grevesse (1989) and fluorescent yields from Bambynek et al. (1972). The medium is assumed to be of constant density and there is no dependence of the line of sight column density on the inclination angle of the observer, with the exception of the unobscured sightlines in the torus geometry which have a zero column density. Fig. 1 illustrates the geometry used in the torus simulations, where the shading only serves to differentiate the different opening angles used, and does not correspond to density.

The models importantly also include line emission, most significantly Fe Kα (6.4 keV), but also Fe Kβ (7.06 keV) and Kα emission from several other elements described in Table 1. As mentioned before, the EW of the Fe line is strongly dependent on absorption, as is the shape of the line’s Compton shoulder. The Compton shoulder feature is seen on the red side of the line, and is a result of Compton down scattering of the first-order line and depends on the scattering optical depth. In these models, the flux of the line and its spectral shape are inextricably linked to N_H and therefore it provides additional constraints on this parameter when spectral fitting. Yaqoob & Murphy (2010) present a detailed examination of the Compton shoulder’s properties, showing that it is also dependent on spectral shape and geometry of the absorber, and hence contains valuable information given that it is well resolved.

The weights of the emergent packets are then binned at the end in the same way. This technique has the advantage that it is possible to vary the number of quanta released depending on how detailed the spectrum needs to be at that energy. It is especially useful for increasing the signal-to-noise ratio of the output spectrum around the Fe K-shell regime (6–8 keV). Valuable computational time may also be saved by decreasing the signal-to-noise ratio in the featureless parts of the spectrum, above 10 keV for example. We also made an improvement to the original method by introducing variable elemental abundances (with respect to the abundance of hydrogen). The abundance of iron is a separate parameter also, as the modelling of the Fe lines is particularly sensitive to this.

We have constructed table models for use within xspec, which we call trans for the spherical distribution of matter, and torus for the toroidal distribution of matter. The parameters of the models are the line of sight column density of the neutral material (10²⁰≤N_H≤ 10²⁶ cm⁻²) and the photon index of the intrinsic power law (1 ≤Γ≤ 3). For trans the abundance of iron (0.1 solar ≤ Fe_abund≤ 10 solar) and the total elemental abundance (0.1 solar ≤ abund. ≤ 10 solar) are included as parameters, whereas for torus, the opening angle of the torus (25≤θ_tor≤⁠) and the inclination angle of the torus (⁠≤θ_i≤⁠) are included. Both models have validity in an energy range of 0.1 to 320 keV.

2.1 FE Kα LINE EQUIVALENT WIDTH CALCULATIONS

We also use the results from our Monte Carlo simulations to make predictions of the EW of the Fe Kα line for both geometries given a set of model parameters. In the simulations the photon packets produced by a simulated Fe K-shell fluorescence are flagged and summed as they exit the simulation. In this way, we can calculate the EW of the line produced, by the sum of Fe Kα photon flux divided by the continuum level at 6.4 keV and the width of the energy bin at this energy. The calculation represents the EW of the entire line, including the Compton shoulder.

Fig. 2 shows how the EW varies with N_H, Γ and the iron and elemental abundances for the spherical geometry. The EW of the line increases linearly versus N_H up to τ∼ 0.1, where the relationship increases faster than linearly beyond that. This is due to the suppression of the continuum at 6.4 keV at high column densities, against which the EW is measured, but possibly also due to the greater number of photons at and above the 7.1 keV Fe K shell threshold having been down-scattered from higher energies. Increasing the iron abundance will as expected increase the EW of the line due to the greater number of iron atoms per hydrogen atom to produce the transition. The abundance of the other elements also affects the EW of the Fe line, though not as significantly. The EW is seen to increase from low elemental abundance to ∼0.3 times solar abundance. This is also expected as an increase in the elemental abundances will cause some continuum suppression at 6.4 keV, leading to an increase in the EW of the Fe line. This trend flattens off for N_H < 10²³ as the elemental abundance increases beyond solar though, but for higher N_H values, the EW decreases above ∼0.3 times solar abundance. Again though, at these extreme N_H values the continuum against which the EW is measured is very suppressed so it is difficult to know if this behaviour is real.

Fig. 3 shows how the EW of the Fe line changes with N_H in the torus model for different torus parameters. The left-hand plot shows this relationship for equatorial sightlines, where the different lines show different opening angles. The smaller the opening angle, and hence the larger solid angle of the torus, the larger the EW of the Fe line. The EW reaches a maximum here of 4, 2.1 and 1.6 keV for torus opening angles of θ_tor = 30°, 60° and 75°. The middle panel shows the same relationship, but for the polar sightline. The EWs here are systematically lower due to the greater, unabsorbed continuum. The EWs reach a maximum of 150, 96 and 63 eV for 30°, 60° and 75° opening angles. Because of this dependence on column density and the opening angle of the torus, it may be possible to make inferences about the circumnuclear matter distribution in AGN from the EW of the Fe line, even for unobscured sources. For example, unobscured sources with an EW greater than ∼150 eV must have a very narrow opening in the torus, or else supersolar iron abundance. Alternatively, the source may not be unobscured at all. It may be very heavily obscured with the intrinsic emission extremely suppressed below 10 keV, and only a weak scattered or galactic component visible. We use these data later when determining the Compton thick nature of our sources. Finally the right-hand panel in this figure shows the variation of EW for a fixed torus opening angle of 60°, and varying viewing angles, and shows that for unobscured sources, the viewing angle does not affect the EW of the Fe line greatly.

2.2 X-ray flux suppression

We also use the results from our new models to calculate expected flux suppression factors for different X-ray bands, as a function of absorption and for different redshifts. This information is useful when estimating the N_H, given only an observed flux and a predicted intrinsic flux, or when considering the biases against heavily obscured sources in X-ray selected AGN samples. We do this for both spherical and toroidal distributions of matter with solar abundances, Γ = 2 intrinsic power laws and for the case of the torus, a 60° opening angle and 90° viewing angle. Fig. 4 gives the flux suppression factor at different column densities and in different bands. The flux suppression factor is calculated as the intrinsic, unabsorbed energy flux in a particular band divided by the flux observed in that band.

The attenuation of the intrinsic emission by the absorbing material is evident as the N_H increases. For the spherical case, at a redshift of zero and for energy bands above 10 keV, the effect of this attenuation is negligible (<2) for N_H up to ∼4 × 10²⁴ cm⁻², however, at N_H = 10²⁵ cm⁻², the flux in these bands is suppressed by a factor of 10. This relationship is roughly independent of redshift for these bands. For the 2–10 keV band, locally, the flux is suppressed by a factor of 10 at N_H=6 × 10²³ cm⁻². As redshift increases, the N_H at which the flux suppression factor reaches 10 increases due to the photoelectric absorption cut-off being redshifted through the band. The N_H at which the flux suppression factor is 10 for redshifts one, two and three is 3 × 10²⁴, 7 × 10²⁴ and 10²⁵ cm⁻², respectively, for the 2–10 keV band.

Flux suppression factors calculated from the torus geometry behave similarly to the spherical case until ∼4 × 10²⁴ cm⁻² where they flatten off, rather than continuing to increase as they do for the spherical geometry. This is due to reflected X-ray flux from the torus which is present regardless of the level of obscuration seen directly through the torus. The flux suppression factor flattens off at ∼100 for the 2–10 keV band at z = 0, however, this improves to ∼30, ∼15 and ∼12 for z = 1, 2 and 3 due to this band observing higher energy rest-frame X-rays which are less effected by photoelectric absorption that dominate in the 2–10 keV band at rest-frame energies. For the >10 keV bands, the flux suppression factor levels off at ∼10 at all redshifts. At z = 3, flux suppression in the 2–10 keV band is less than that seen in the 20–100 keV (INTEGRAL) and 14–195 keV (Swift/BAT) bands. This is most likely due to the Compton hump, produced by the ‘down-scattering’ of higher energy X-rays and peaks at rest frame ∼30 keV. At z = 3 it appears directly in the 2–10 keV band, whereas the higher energy bands only view the tail of the hump.

For both geometries, all redshifts and X-ray bands, flux suppression is >10 at N_H = 10²⁵ cm⁻². This shows that for sources with extreme obscuration, even X-rays are strongly attenuated.

3 X-RAY SPECTRAL ANALYSIS

The X-ray spectra of galaxies and AGN are very complicated and as such it is necessary to utilize high signal-to-noise ratio spectra with good spectral resolution and a wide bandpass in order to correctly determine the source properties. These have been provided by XMM–Newton better than ever before, so in using data obtained from this observatory, we aim to gain new and improved insights into the X-ray spectral properties of the 12MGS. Rush et al. (1996) studied the soft X-ray spectra of the 12MGS from ROSAT All-Sky Survey observations, but in doing so will have missed any moderate to heavily obscured emission, and furthermore, the signal-to-noise ratio and resolution of ROSAT was incomparable to that of XMM–Newton. Barcons et al. (1995) used hard X-ray data from HEAO 1 and will therefore have been sensitive to obscured sources. However, their data also suffered from poor signal-to-noise ratio and low spectral resolution, meaning they could only determine the source intensity with no meaningful investigation of the absorption or continuum slope.

We develop a systematic approach to the spectral fitting of our sample in order to determine the intrinsic X-ray continuum slope, Γ, the neutral absorption column density, N_H, and the 2–10 keV intrinsic source power, L_X. Characterizing Γ is necessary for insights into the primary X-ray generation process in AGN, thought to be the Comptonization of accretion disc photons by a hot corona (Sunyaev & Titarchuk 1980; Haardt & Maraschi 1991). This intrinsic source continuum parameter is also useful for comparing different source populations, such as Seyfert 1s and Seyfert 2s, to determine if their central engines are the same, as predicted by unified schemes. Determining the N_H distribution in AGN is a key ingredient to XRB synthesis models, and measuring the intrinsic power allows for comparison to the power at other wavelengths. We will discuss most of these issues in a subsequent publication however, and we focus on an accurate determination of these parameters here.

3.1 Sample selection and data reduction

The sample of observations presented here consists of all European Photon Imaging Camera (EPIC-pn) XMM–Newton observations of the extended 12MGS (RMS) that were made public in the XMM–Newton data archive as of 2008 December, including sources which were observed serendipitously. In the case of multiple observations of the same source, we use the longest observation available. The data were reduced homogeneously and systematically using the method described by Nandra et al. (2007) using sas v7.0 tasks. This involves extraction of source events from a circular region, radius 35 arcsecs, centred on the source position, and extraction of background events from rectangular regions, ensuring no contamination from interloping sources. We search for pileup effects in the observations by comparing the distribution of single and double pixel events, which are sensitive to these effects, to model distributions. Pileup effects are reduced if and when detected, by using only CCD pattern 0 (single) events (Ballet 1999). Background flares are filtered out by calculating the level of the background count rate at which the excess variance is determined to be zero. A background count rate of twice this level is used to filter the observation of flares. The subsequent background-subtracted spectra extracted are then binned with a minimum of 20 counts per bin, appropriate for χ² fitting. We include for analysis only those sources for which a meaningful spectrum has been produced by this process. This excludes those observations for which the flaring of the background has rendered the observation useless, where the source has not been detected above the background, or the detection is at low significance. We determine a meaningful spectrum to be one where the normalization of a simple power-law model can be constrained to be greater than zero (Δχ² > 2.71) with a fit to the 0.2–10 keV spectrum. After these considerations, the sample consists of 126 observations. The names, positions and redshifts of these galaxies along with the details of the XMM–Newton observations are given in Table 2. This set of XMM–Newton observations is quite varied, with exposures varying from hundreds of seconds to tens of kiloseconds (Fig. 5), with differing observation modes and filters used. We do however, treat each spectral fit the same, regardless of the number and distribution of counts in the spectrum (see Fig. 5 for distribution of counts in these observations and the observed flux distribution).

3.2 Spectral fitting method: 2.5–10 keV

We use xspec v11.3 to carry out the spectral fitting of the 12 μm selected galaxies. As each spectrum has been grouped with a minimum of 20 counts per bin, the errors on the data points can be considered as Gaussian, and therefore χ² fitting can be used to find the best-fitting parameters for a given model.

Factors complicating the determination of the intrinsic parameters and the N_H are the soft excess components, ubiquitous in the soft X-ray spectra of AGN, which may arise from the accretion disc, thermal emission, scattered intrinsic emission and/or unresolved emission lines. For this reason, where possible, we determine the intrinsic source parameters from a fit to the 2.5–10 keV spectra for these galaxies, where the soft excess components are negligible. The following is a description of the spectral fitting methods we use.

We begin by fitting the 2.5–10 keV spectrum with a simple power-law model (powerlaw in xspec). We then systematically include further model components in order to fit the more complex features that the spectrum presents. We note the change in the χ² value of the fit after the addition of each added component and the change in the number of degrees of freedom (dof). For a model combination to be allowed, we stipulate that Δχ²≥ 4.0 for each dof lost in adding the additional component (based loosely on 95 per cent confidence levels quantified using an F test). The combination of model components which meets these criteria and has the smallest reduced χ² (χ²_r=χ²/dof) is presented as our best systematic fit for that spectrum.

We use the following models in combination to acquire a fit to the 2.5–10 keV spectrum.

• Neutral absorption model, zwabs (Morrison & McCammon 1983), for absorption intrinsic to the source.

• Neutral reflection from Compton thick material, pexmon (Nandra et al. 2007), which is derived from the pexrav model of Magdziarz & Zdziarski (1995) and includes self-consistent Fe Kα, with its Compton shoulder, and Fe Kβ line emission (George & Fabian 1991; Matt 2002).

• Our new model describing a heavily absorbed power law, trans, based on Monte Carlo calculations, with a spherical geometry, which take into account Compton scattering and includes line emission (as described in Section 2). We use this model in addition to the zwabs model as it is more likely to pick out a heavily obscured source due to the inclusion of line emission.

• Additional line emission, modelled by a Gaussian.

The pexmon and trans models include Fe Kα line emission, produced by cold distant matter, and therefore, if a narrow emission feature exists at 6.4 keV, these models should be preferentially included in the fit. We also check to see if any emission is present from ionized iron by adding narrow Gaussians at 6.7 and 6.96 which correspond to Kα emission from He-like and H-like iron, respectively. We also add a third Gaussian component with energy constrained to be 6.2 < E_line < 6.6 keV and with a free width parameter, to check for broad Fe Kα. This represents ΔE = 0.2 keV from the rest-frame energy of neutral Fe Kα so as to avoid fitting ionized Fe lines at higher energy. In reality, broad Fe Kα lines have been found with a lower energy than the energy we permit (e.g. Nandra et al. 2007), but we use this conservative range as broad Fe Kα lines are not the focus of this investigation. We keep these Gaussian components if the normalization of each line is constrained to be greater than zero (Δχ² > 2.71).

We keep the power-law indices for the pexmon and trans components equal to the primary power law for each fit as it is presumed that each component in this band has the same origin. For spectra where the power-law index, Γ, is badly constrained we fix the power-law index to Γ = 1.9, the canonical value for Seyfert galaxies (Nandra & Pounds 1994), which make up the majority of our sample, but also roughly appropriate for galaxies (Ptak et al. 1999). We define fits where Γ cannot be constrained within ΔΓ≤ 0.7 to be unconstrained as such an uncertainty introduces a large overall uncertainty into the fit and other fit parameters as well as the source luminosity.

We fit the 2.5–10 keV spectra initially as this band is least affected by galactic and other complex soft X-ray emission, and is therefore best suited for determining the intrinsic parameters of the nuclear source. Furthermore, the higher signal-to-noise ratio at soft energies often results in spectral fits being driven by the soft features. For example, strong soft emission may cause the power-law index to steepen and as a result the measured N_H may increase to compensate for this. However, the lower signal-to-noise ratio in the hard band can sometimes lead to unconstrained fits. We do not use fits where the normalization of the power law is not constrained to be greater than zero and instead use the full band fit method, described below. For fits that produce a low Γ, a hidden transmission or reflection component is likely to be present and not recognized by the 2.5–10 keV fit. Where Γ < 1.4, this presents a scenario where the photon index lies greater than 3σ away from the mean measured by Nandra & Pounds (1994). In this case we also refer to the full band fit, where a reflection or heavily obscured transmission component is often then detected.

3.3 Spectral fitting method: 0.2–10 keV

Ignoring data below 2.5 keV leaves us insensitive to neutral columns less than ∼10²² cm⁻². For satisfactory hard band fits, we then freeze the hard band fit and continue fitting the 0.2–10 keV spectrum with these parameters frozen, with the exception of the N_H parameter. The following models are then systematically added, again requiring that Δχ²≥ 4.0 for each dof lost in adding the additional component. Again, the combination of model components which meets these criteria and has the smallest reduced χ² (χ²_r=χ²/dof) is adopted as our best systematic fit for the 0.2–10 spectrum.

• Neutral absorption (wabs) associated with the Galaxy, frozen to the value given in Table 3 and applied to the whole spectrum. This component is assumed to be present regardless of the Δχ².

• Absorption applied to the whole spectrum and intrinsic to the host galaxy, be it neutral or ionized (see below). If no neutral absorption is detected in the hard band, we also check to see if a neutral absorption component is required for the primary power law in the soft band (zwabs).

• A warm absorber [cwa18, based on XSTAR (Kallman et al. 2004); see Nandra et al. (2007) and references therein for description] to model cases where ionized absorption features are present in the spectrum. If evidence for neutral absorption is present in the hard band fit, we also check whether an improved fit results by adding a warm absorber on top of this or indeed, replacing this with a warm absorber.

• A thermal plasma component (apec; Smith et al. 2001). The soft X-ray spectra of obscured AGN have been shown to consist partly of emission lines photoionized by the AGN (Guainazzi & Bianchi 2007). Models such as apec are often used to account for this emission when these lines are not resolved. When using this model, the abundances are initially frozen to the solar values, however, once the best fit is obtained, we check if thawing the abundances produces a better fit.

• A second power law, where the power-law index Γ₂ is constrained such that Γ₂=Γ₁. This component is used to fit any ‘scattered power law’ often seen in type 2 Seyfert spectra (Turner et al. 1997). It can only be seen when moderate or high absorption is occurring to the primary power law, and therefore we only add it if absorption is detected in the hard band fit.

• We also check to see whether Γ₂≥Γ₁ produces a better fit.

3.3.1 Modelling the soft excess seen in Seyfert 1 like spectra

Modelling the soft excess seen in quasar and Seyfert 1 like spectra is complicated, especially as we do not fully understand the nature of this feature. The models used to fit the soft excess are additive (often a blackbody or Comptonized component), thus accounting for flux in the soft X-ray band. However, if an absorption component is used in conjunction, adding the soft excess component can lead to an overestimation of the N_H parameter due to ‘balancing’ of emission and absorption. Thus, we apply strict criteria in order to include a soft excess model in the spectral fit.

This soft excess is thought to originate from close to the central engine, and thus should be an excess above the intrinsic, unabsorbed power-law spectrum. To test for the soft excess, we extrapolate our fit of the 2.5–10 keV spectrum into the soft band, removing any absorption components. Only then, if there is emission in excess of this extrapolated fit in the soft band, do we include soft excess components in the soft band fit. To detect excess emission in the soft band, we use a ‘soft ratio’, defined as the data-to-model ratio at 0.4 keV, done similarly by Sobolewska & Done (2007). An energy of 0.4 keV is used to avoid atomic emission from oxygen at ∼0.7 keV. We choose a soft ratio lower cut-off of 1.2 (20 per cent), which is high enough to exclude uncertainties in the spectral calibration (see below). We also require this ratio to be rising towards softer energies, as emission from thermal plasmas can cause an excess at 0.4 keV, but that peaks at higher energy. Fig. 6 gives two examples where this soft ratio is greater than 20 per cent, but one shows a classical soft excess which rises from 1 to 0.5 keV (3C 273), whereas the other shows an excess due to the emission from a thermal plasma, which falls from 1 to 0.5 keV (Messier 100). If the soft ratio exceeds 1.2 and the ratio is rising from 1 to 0.5 keV, we then proceed to include soft excess model components in the spectral fit. The model components we use to fit the soft excess are either a blackbody (bbody) or a Comptonized (compTT) component, whichever provides the best fit.

3.3.2 Bad spectral fits

We define fits which give χ²_r > 2.0 as ‘bad’. Although this definition is arbitrary, it is useful for identifying cases where the models used are not a good representation of the spectrum. For the 2.5–10 keV fits, only NGC 1068, NGC 1365 and M82 have bad fits. These sources have notoriously complex spectra, which are high signal-to-noise ratio in this case, with several emission and absorption features. We do not attempt to fit these additional complexities, however, and present only the basic hard band parameters for these fits. No errors are quoted because the standard formalism does not apply unless the model fits the data (Lampton, Margon & Bowyer 1976).

When we fit the full band spectrum, we sometimes find that keeping the hard band parameters fixed will cause a bad full band fit, so we then thaw these parameters for an improvement. After this procedure, 17/126 fits are still bad. These bad fits usually result from high signal-to-noise ratio spectra which reveal highly structured soft X-ray features which are too complex for our models. However, the nature of the soft X-ray emission is not the focus of this investigation, as we seek mainly the intrinsic hard X-ray power of the source, along with any neutral absorption present. In these cases, we do not present the results from the full band fit, rather only the hard band fit. These are denoted by an ^(H) in the spectral fit table. Furthermore, we note that the calibration uncertainties on XMM–Newton are at the level of ∼5 per cent (Kirsch et al. 2004). If the ratio of the data to the model is less than this in the soft band, the bad fit may be partially due to calibration uncertainties. In this case we present the soft band fit results, despite a bad χ². We present the parameters Γ and the power-law normalization from the hard band fits here as these fits had ‘good’χ²_r values and thus the uncertainties could be estimated. These are denoted by a ^(D) in the spectral fit table. 3C 120 is an example of a source with a bad spectral fit, but a data/model ratio of less than 5 per cent, which we show as an example in Fig. 8. Our procedure should not result in grossly inaccurate estimates of Γ, L_X or N_H in such cases.

For cases where the hard band fit is unconstrained (i.e. the power-law normalization is not constrained to be greater than zero) or the hard band fit is bad (χ²_r > 2.0), then we cannot use the method described above, whereby the hard band fit is extended into the soft band. In these cases, we carry out simultaneous full, 0.2–10 keV band fits with all model components. Although this is less desirable, it is useful in the case where the hard band fit is unsatisfactory and for cases where the hard band measures a flat power-law index. These are denoted by an ^(F) in the spectral fit table.

3.4 Intrinsic power and reflection fraction

We calculate the intrinsic power emitted in the 2–10 keV band for each galaxy based on the absorption corrected flux in the 2–10 keV band from the spectral fits to the primary power law. We define the primary power law as the power-law component with the greatest normalization, i.e. the intrinsic flux at 1 keV. This may include the trans component, if this is present, which will typically model a heavily obscured hard component visible only at hard energies. Standard cosmological parameters are assumed throughout (Ω_m = 0.3, Λ = 0.7, H₀ = 70 km s⁻¹ Mpc⁻¹). In the top panel of Fig. 7, the intrinsic, absorption corrected 2–10 keV luminosity of each galaxy is plotted against its redshift. This figure illustrates that the 12MGS is a low-redshift galaxy sample, with a wide range of X-ray luminosities from 10³⁸ to 10⁴⁴ erg s⁻¹ (though with one very luminous source, 3C 273 at ∼10⁴⁶ erg s⁻¹). The lower panel shows the size of the correction made to the observed luminosity against the intrinsic luminosity, which can be up to a factor of ∼50.

For reflection-dominated spectra, we derive the intrinsic power of the source from the intrinsic power law required to produce the reflection component. The pexmon model calculates the reflection from a semi-infinite slab, subtending 2π sr on the sky. The estimate of the intrinsic power in the reflection-dominated sources is uncertain as it depends on the precise geometry of the reflector. This was shown by Murphy & Yaqoob (2009) who modelled reflection from a toroidal distribution of matter, finding that the reflection fraction can be underestimated in this way by a factor of ∼6. We also investigate the reflection fraction and intrinsic luminosity derived using the pexmon model and compare these to those derived from our own torus model.

3.5 What can we determine using the torus spectral model?

For spectra in which we find reflection is strong (R_pexmon > 1, where R_pexmon is the ratio of the normalization of the pexmon component to the normalization of the primary power law), we fit the same spectrum with model spectra from the torus simulations with two aims. First, as mentioned above, we aim to investigate the reflection fraction and intrinsic source luminosity as determined using the torus model with respect to the pexmon slab model, and secondly, we investigate if it is possible to determine and constrain the torus geometric parameters, θ_tor and θ_i from the 2–10 keV spectra. For determining the reflection fraction, we fix θ_tor = 60°, leaving θ_i free to vary. We then determine if the inclination angle can be constrained. Following this, we fix θ_i = 90°, and allow θ_tor to vary, to see if this can be constrained.

3.6 X-ray AGN in our sample

The main goal of this work is to ascertain the properties of the primary emission, as well as the absorption characteristics of our sample. We do this initially for our whole sample, and we then go on to do so for AGN only. We define a set of unambiguous X-ray AGN as galaxies with an observed 2–10 keV luminosity greater than 10⁴² erg s⁻¹ as these sources are unambiguously powered by the accretion of material on to a supermassive black hole. No local pure star-forming galaxy has ever presented a 2–10 keV luminosity above this limit. One of the most X-ray luminous star-forming galaxies is NGC 3256, which presents a 2–10 keV luminosity of 2.5 × 10⁴¹ erg s⁻¹, but shows no evidence for AGN activity (Moran, Lehnert & Helfand 1999). Therefore, galaxies with an observed 2–10 keV luminosity greater than 10⁴² erg s⁻¹ are very likely to host an AGN. However, this will not select all AGN in this sample, so the remaining sources cannot be thought of as non-AGN. In this way 60/126 = 48 per cent of the sources in our sample are unambiguous AGN taking into account their X-ray luminosities only.

4 SPECTRAL FITTING RESULTS

4.1 Spectral fit components and parameters

From our spectral fitting method, we have determined a best-fitting model for each source, which includes all the absorption and soft excess components, giving us an accurate estimate of Γ, N_H and the intrinsic power accounting for the complexities in the spectra. For 126 spectra, we find that in 62 of them, we can constrain Γ on the primary power law to ΔΓ < 0.7 (we fix Γ = 1.9 otherwise). Furthermore, 96 spectral fits require an absorption component, 39 of which warm absorption has been detected. A thermal plasma component is ubiquitous in the fits as well, needed to fit 93 spectra. Fe Kα emission is detected in 71 spectra, where 36 of these are broad lines (σ > 0.1 keV). A secondary power-law component is required in 44 of our fits, and reflection is also required in 44 of the fits.

The spectral parameters of all fits are listed in Tables 3–8. Table 3 presents the basic spectral fit parameters such as the fit statistics, the parameters of the primary power law and the thermal plasma component, where present, and the observed 2–10 keV flux and intrinsic 2–10 keV luminosity. Table 4 gives details of any secondary power-law component required in the fit; Table 5 gives details of the sources which have a soft excess detected and of the models used to fit this component; Table 6 presents details of the reflection component needed in each fit, including data obtained from the torus model; Table 7 presents the parameters of the warm absorber model if ionized absorption has been detected in the spectrum and Table 8 gives details of the line emission detected in the Fe K regime. Included on all fit parameters in which a satisfactory fit has been acquired are the 90 per cent confidence errors for one interesting parameter (Δχ² = 2.706).

A selection of example spectra is presented in Fig. 8. These are the spectra of 3C 120, NGC 3079, NGC 4593, NGC 5170, UGC 08850 and NGC 7479, chosen to display a range in signal-to-noise ratio as well as model components. 3C 120 and NGC 4593 have a soft excess present in their spectra as determined by the ratio of the data at 0.4 keV to the extrapolated 2.5–10 keV power law, and fitted by the compTT model. Reflection from neutral material is included in the fit, where in the case of NGC 4593, this component accounts for the Fe Kα emission seen. 3C 120 requires an additional Gaussian component to model what seems to be broad Fe Kα emission. These fits also include a thermal plasma component, however, it is possible that in these cases this model component is not representative of real thermal plasma emission as this component is not likely to be detected above the emission from the AGN, and is rather being used to fit the soft complexities of the spectra. A warm absorber has also been detected in the spectrum of NGC 4593.

The spectrum of NGC 3079 is typical of a reflection-dominated spectrum (R_pexmon = 16.3) with an intense Fe Kα component (940 eV) and is certainly Compton thick. UGC 08850 presents a transmission-dominated spectrum, which is well fitted by our new model trans measuring N_H = 3.8 × 10²³ cm⁻², plus a scattered power law. In the spectrum of NGC 7479 we have detected a hidden transmission component (N_H = 2 × 10²⁴ cm⁻²) beneath a reflection dominated spectrum, though it is not well constrained. The dominance of the reflection component and high EW of the Fe Kα line (768 eV) imply a high column density in support the measured N_H in the transmission component.

The spectrum of NGC 5170 has a low signal-to-noise ratio and in this case we have had to fix the power-law index to 1.9. The only added model component here is low-level neutral absorption.

In Fig. 9, we present the 2.5–10 keV spectrum of NGC 1068, which is an example of where the 2.5–10 keV fit is bad. This is a high signal-to-noise ratio spectrum which reveals multiple discrete features, possibly due to absorption and emission, or even calibration uncertainties.

4.2 Properties of the primary emission

The primary power law is found to have a mean power-law index, 〈Γ〉 = 1.90^+0.05_−0.07 for the whole sample, and an intrinsic spread of σ = 0.31^+0.05_−0.05. We use the maximum likelihood method of Maccacaro et al. (1988), which takes into account the measurement errors on Γ, to determine both the mean and the intrinsic standard deviation of this distribution.

We also investigate the Γ distributions for obscured and unobscured sources, which is an important test for AGN unification. We do this for the entire sample, and also for unambiguous AGN with L_X > 10⁴² erg s⁻¹. In Fig. 10, we show how the distribution of Γ differs for obscured sources in these two cases. We again use the maximum likelihood method of Maccacaro et al. (1988) to determine both the mean and the intrinsic standard deviation for each of these distributions. We find that for all unobscured sources, 〈Γ〉 = 1.90^+0.09_−0.09 and σ = 0.33^+0.07_−0.06 and for all obscured sources, 〈Γ〉 = 1.97^+0.14_−0.12 and σ = 0.44^+0.12_−0.09. This is statistically consistent with the intrinsic power-law indices of these two populations being the same. We also find that for unobscured AGN, 〈Γ〉 = 1.84^+0.12_−0.14 and σ = 0.32^+0.11_−0.08 and for obscured AGN, 〈Γ〉 = 1.90^+0.16_−0.14 and σ = 0.38^+0.13_−0.07. This is again statistically consistent with the intrinsic power-law indices and their intrinsic spreads being the same for both obscured and unobscured AGN. A Kolmogorov–Smirnov (K–S) test confirms that the probability that these distributions are drawn from different parent samples is low (42 per cent) for AGN.

Furthermore, we find no dependence of Γ on X-ray luminosity. We show this in Fig. 11 in the range of 10³⁸ to 10⁴⁵ erg s⁻¹, where the scatter of Γ versus X-ray luminosity is plotted, as well as the mean values for each logarithmically spaced luminosity bins. The vertical error bars show one standard deviation spreads in the values.

4.2.1 Soft excess modelling

We find that a total of 15 spectra in our sample require a soft excess model component in their spectral fit. We have determined this from two soft ratio criteria, that the soft ratio must be >1.2 and that is must be rising from 1 to 0.5 keV. 3C 120 has a soft excess which does not fit our criteria, as it does not have a data/model ratio greater than 20 per cent at 0.4 keV. However, this source has a clear soft excess at 0.7 keV not well modelled by thermal plasma emission. We find that a ratio of 3.2 at 0.4 keV is recovered when an additional intrinsic absorption component is used. In Fig. 12, we plot the distribution of the soft ratio, which ranges from ∼1.3 to 7.8 for our sample, though with only one spectrum with a ratio greater than 4. Sobolewska & Done (2007) note that this soft ratio ranges from ∼2 to 3 for their sample of quasars. The details of these fits are presented in Table 5. All but three of the fits prefer a Comptonized component over a blackbody component.

4.3 Obscuration and reflection

4.3.1 Compton thin sources

We find that 28/126 (22 per cent) of our spectra show no evidence for any absorption at all, whereas 77 (61 per cent) spectra present Compton thin (neutral) absorption, showing that absorption is commonplace in the X-ray spectra of all galaxies.

4.3.2 Compton thick sources

As mentioned previously, for 0.2–10 keV X-ray spectra, a direct measurement of a column density of Compton thick source is difficult due to the extreme suppression of the X-ray flux in this band. We have, however, successfully used our new model, trans, to directly measure the N_H to be greater than 1.5 × 10²⁴ cm⁻² in four sources (NGC 1320, NGC 1667, NGC 4968 and NGC 7479). The inclusion of a self-consistently modelled Fe Kα line is likely to have provided greater constraint on the N_H than models which include this component separately.

In our analysis, we call sources with a reflection-dominated spectrum Compton thick, of which there are 12. We define a reflection-dominated spectrum to be where R_pexmon > 8. The R_pexmon > 8 criterion is arbitrary and chosen to be low enough to include known Compton thick AGN such as NGC 1068. Fig. 13 shows the EW of the Fe Kα line against N_H for our sample for reflection-dominated sources (green), mildly reflection-dominated sources (blue, 1 < R < 8) and sources with weak or no reflection (red, R < 1). It can be seen from this that reflection-dominated sources also have high Fe line EWs. We plot only narrow line EWs here, which are most likely to originate in cold distant obscuring matter, unless no narrow line has been measured, in which case we plot the broad line EW, if one has been measured.

There are two sources, however, that have high Fe line EWs which are constrained to be >300 keV, but have a measured N_H < 10²² cm⁻². Our Monte Carlo simulations have shown that this is not possible for either a spherical or toroidal matter distribution, and so we investigate them further. These sources are NGC 3690 and IRAS F13349+2438. NGC 3690 is the merging system ARP 299, which was shown by Beppo-SAX to host a Compton thick AGN (Della Ceca et al. 2002), confirming our assertion. We therefore add it to our sample of Compton thick AGN. Longinotti et al. (2003) examine the Fe Kα features in the X-ray spectrum of IRAS F13349+2438, a narrow-line Seyfert 1, finding them to be broad and complex. This line is therefore unlikely to originate from distant Compton thick obscuring matter. This shows that Fe Kα EW diagnostics for obscuration are only suitable when we can be sure that the line originates in cold, distant matter such as the torus.

For sources where their Compton thick nature has been inferred by the EW and the reflection fraction, we can only put a lower limit on the N_H of 1.5 × 10²⁴ cm⁻², but in reality, this could be higher and requires data above 10 keV to constrain. We find a total of 16 Compton thick sources in our sample, the details of which are given in Table 9. The Compton thick fraction for all 126 galaxies is 16/126 (=13 ± 3 per cent), where the errors have been calculated for a binomial distribution (⁠⁠, where p is the proportion and N the total number of sources). The Compton thick fraction for unambiguous AGN (L_X > 10⁴² erg s⁻¹) is 11/60 (=18 ± 5 per cent).

4.3.3 N_Hdistribution and obscured fraction

In Fig. 14, we present the N_H distribution for the whole sample, and overplotted is the N_H distribution of those with L_X > 10⁴² erg s⁻¹. The main difference in the two distributions is the strong peak of unobscured sources at low X-ray luminosities and the greater proportion of heavily absorbed X-ray luminous AGN. The low luminosity, unabsorbed sources may consist of unobscured star-forming galaxies, LINERs or lower luminosity AGN, but we will investigate this further in Paper II. The N_H distribution of unambiguous AGN shows a smaller unobscured peak, but with a second obscured peak at N_H = 10^23-24.

Fig. 15 plots N_H versus L_X and also the obscured fraction, defined as the ratio of the number of sources with N_H > 10²² cm⁻² to the total number of sources in each luminosity bin. There is significant variation in the obscured fraction with X-ray luminosity, particularly above and below 10⁴² erg s⁻¹ where the ratio declines steeply from a peak.

4.3.4 Reflection-dominated sources

For sources where the X-ray spectrum includes a strong reflection component, we have also fitted our new torus model, first to investigate the reflection fraction and intrinsic source luminosities as determined using this model with respect to the pexmon slab model. Secondly, we investigate if it is possible to determine the torus geometric parameters, θ_tor and θ_i from the 2–10 keV spectra. In Fig. 16, we plot the distributions of the ratio of the reflection fraction as determined by the torus model to the reflection fraction determined by the pexmon model and the ratio of the intrinsic luminosities as determined by each model. We find that for most sources, both the reflection fraction and intrinsic source luminosity as determined using the torus model are greater than that determined by the pexmon model. The means and medians of these ratios are 3.08 and 2.70 for R_torus/R_pexmon and 2.62 and 1.99 for LX_torus/LX_pexmon. We also find that it is not possible to constrain the inclination angle or opening angle of the torus with these data below 10 keV, as neither parameter can be constrained, even with the other parameter fixed.

5 DISCUSSION

5.1 X-ray transmission in heavily obscured sources

We have made predictions of the EW of the Fe line for different model parameters and for two matter geometries, a sphere and a torus. The predictions we make for the torus are useful for comparing observed EWs and making inferences from these about the matter geometry and parameters in the observed system. For example, we find that a maximum EW of 150 eV is achievable when considering a torus geometry for unobscured sightlines with a typical Γ (2.0) and solar abundances. To achieve greater EWs than this, the line of sight must be obscured with N_H > 10²³ cm⁻². Therefore, if an EW higher than this is measured for an unabsorbed spectrum, it may cast doubts on the reliability of the measured line of sight N_H. We showed this for NGC 3690, which shows a Fe Kα EW of 913⁺³⁸⁰₋₃₄₆ eV, but a measured N_H of only 6.14^+5.57_−6.14× 10²⁰ cm⁻² which is clearly at odds with our model predictions. Data above 10 keV from Beppo-SAX show, however, that this source is indeed heavily obscured with N_H > 1.5 × 10²⁴ cm⁻² (Della Ceca et al. 2002), in full support of our model predictions. Our Fe Kα EW predictions also agree well with predictions of the same nature as made by Ghisellini et al. (1994), Ikeda et al. (2009) and Murphy & Yaqoob (2009).

We also present flux suppression factors for both spherical and toroidal geometries as a function of N_H for different bands and redshifts. These are particularly relevant for AGN surveys that use the X-ray band as they indicate the effectiveness of each band at detecting heavily obscured systems. It is known that even the hardest X-ray energies are affected by extreme obscuration. Here we quantify the flux suppression in these bands, finding that locally, flux suppression reaches a factor of 10 at N_H=10²⁵ cm⁻² for the >10 keV bands. This may explain the fewer than expected local Compton thick AGN detected in hard X-ray surveys conducted by Swift/BAT and INTEGRAL. For the >10 keV X-ray bands, the flux suppression factor is 10 at ∼10²⁵ cm⁻² for all redshifts. This is because at these energies, Compton scattering is the dominant interaction and source of attenuation in the spectrum. The cross-section of this interaction is roughly constant as a function of energy and therefore the effect of redshift is negligible. Flux suppression in the 2–10 keV band reaches 10 at 6 × 10²³ cm⁻² but this improves as redshift increases, becoming ∼10²⁵ cm⁻² at z = 3. For the torus geometry, the flux suppression factors behave similarly to the spherical geometry up to ∼4 × 10²⁴ cm⁻². Here, the flux suppression factors level off due to reflected flux from the inner part of the torus, which remains constant regardless of the line of sight N_H. In all cases though, an absorbing column of N_H=10²⁵ leaves all X-ray bands severely insensitive to detection of X-ray sources.

Using model spectral data from the torus simulations, we also investigated the difference in the reflection fractions and intrinsic luminosities as determined by using these new torus data, to that derived from fits with the slab model, pexmon. Murphy & Yaqoob (2009) find that the reflection fraction is underestimated by a factor of ∼6 when they compare their torus data to slab data. We also find that the reflection fraction is systematically higher when using the torus model data, by a factor of 3 on average. Though this is not as high as the underestimation presented by Murphy & Yaqoob (2009), the discrepancy is probably due to differing torus geometries, as they use a cylindrical torus geometry which gives varying N_H as a function of inclination angle, whereas we use a spherical torus with constant N_H. We also find though that the intrinsic luminosity derived from the slab models are underestimated with respect to the torus model by a factor of 2–3. Additionally, using the torus model, we find that we are unable to constrain the torus geometrical parameters, the opening angle or the inclination angle using the XMM–Newton spectral data. Ikeda et al. (2009) also find that it was not possible to constrain these parameters with their model, even with spectral data above 10 keV with Suzaku.

5.2 Properties of the primary X-ray emission from 12 μm selected galaxies

We have presented the properties of the primary X-ray emission of our sample of 12 μm selected galaxies. As our sample is a heterogeneous mix of galaxies, these results reflect the properties of all X-ray sources. We will present results on individual optical classes in Paper II. We do however define a subset of unambiguous AGN using their observed 2–10 keV luminosity though this will of course not include AGN with low X-ray luminosity which may be due to significant absorption or being of intrinsic low luminosity.

We find that the mean power-law index is 〈Γ〉 = 1.90^+0.05_−0.07 for the whole sample, with an intrinsic spread of σ = 0.31^+0.05_−0.05. The mean is consistent with previous works on predominantly X-ray selected samples. For example Nandra & Pounds (1994) who find 〈Γ〉 = 1.95 ± 0.05 and σ_p = 0.15 ± 0.04 for their sample of Seyfert galaxies observed with Ginga. We do find a larger intrinsic spread in the indices though than Nandra & Pounds (1994). Our result is also consistent with the results from harder X-ray selected samples (e.g. 1.93 for all Seyferts in the INTEGRAL catalogue of Beckmann et al. 2009).

Comparing the distribution of power-law indices for obscured sources and unobscured sources is a key test of the unification scheme, which states that the central engine is the same for both, the difference being the orientation of the system to the observer (Antonucci 1993; Urry & Padovani 1995). If a systematic difference is found in these distributions, it would suggest that this is not simply the case. Comparing the average X-ray spectra of Seyfert galaxies with Ginga and OSSE data, Zdziarski et al. (1995) originally found a systematic difference between the Γ distributions of Seyfert 1s and Seyfert 2s, however, more recent works have tended to find that the distributions are indeed consistent with each other, generally due to the inclusion of more complex spectral models, such as Compton reflection components (e.g. Dadina 2008). We investigated these distributions for obscured and unobscured sources, as well as obscured and unobscured unambiguous AGN. Our analysis showed that the means of these distributions in both cases are not significantly different (the means lie within 1σ of each other) and a K–S test supports the same conclusion, implying that the X-ray generation mechanism in both obscured and unobscured AGN is the same, lending support to AGN unification. The fact that the intrinsic dispersions are similar also supports this conclusion. Beckmann et al. (2009) carried out the same analysis for obscured and unobscured sources on INTEGRAL data, which should be less affected by absorption effects, and also find no significant differences between their obscured and unobscured sources. Despite all of this, however, we do notice that the Γ distribution of obscured AGN is skewed towards low Γ values. We discount the possibility of unidentified reflection in these sources, but results presented by Malizia et al. (2003) suggest an alternative cause. They also find a harder distribution of Γ values for Seyfert 2s with respect to Seyfert 1s from their analysis of Beppo-SAX data, but reconcile this difference by including more complex absorption models, such as partial covering scenarios, which also improve their fits. There have been several instances where absorption in AGN has been shown to be more complex than a single smooth obscurer, for example NGC 1365, which shows rapid variability in the line of sight obscuration (Risaliti et al. 2009) and further examples by Ricci et al. (2010). As we do not use partial covering models, this may be the reason for the skewed distribution towards low Γ values in our obscured AGN.

We have also investigated any correlation between the power-law index and the intrinsic X-ray luminosity (Fig. 11). Our results do not support a correlation between these two quantities however. This was also found by George et al. (2000) in their sample of local AGN using ASCA data. At higher redshifts, however, there have been reports of a correlation between Γ and L_X. For example Dai et al. (2004) find an anticorrelation between the two for their minisurvey of 1.7 < z < 4 sources observed with XMM–Newton and Chandra in the luminosity range of 10^43-45 erg s⁻¹. This finding was later supported by Saez et al. (2008) for the Chandra deep fields who use a larger range in luminosity (10^42-45 erg s⁻¹) and redshift (0.1–4) with 173 AGN. Our luminosity range is 10^38-44 erg s⁻¹, which is somewhat lower than the higher redshift studies, though the local sample of George et al. (2000) ranges from 10^42-46 erg s⁻¹, with no correlation. This may suggest an evolutionary effect, or a selection effect in the X-ray samples. Our sample does not suffer from such effects, as it is MIR selected.

5.3 Properties of X-ray obscuration in 12 μm selected galaxies

5.3.1 Compton thick sources

We find that 16/126 (=13 ± 3 per cent) of the whole sample of sources studied in this sample are Compton thick. For X-ray bright AGN (L_X > 10⁴² erg s⁻¹), there are 11/60 (=18 ± 5 per cent) Compton thick sources. This fraction is important for synthesis models of the XRB and the accretion history of the Universe. For X-ray samples selected above 10 keV, the Compton thick fraction has been constrained to be less than that predicted by the XRB. Beckmann et al. (2009) found that only 4 ± 2 per cent of the AGN in their sample are Compton thick, whereas the XRB predicts a fraction close to 15 per cent at the corresponding limiting flux. Treister, Urry & Virani (2009) claim that the observed differences between observation and models is due to degeneracies in the parameters of the models, and that only direct observations of Compton thick AGN can determine their relative number. However, when accounting for the biases present in hard X-ray selection, Burlon et al. (2010) recover a 20 per cent Compton thick fraction in the Swift/BAT survey, which is consistent with the XRB and our results.

The Compton thick fraction that we measure in 12 μm selected sources is almost 3σ greater than that measured in hard X-ray selected sources, suggesting that MIR selection is more efficient than hard X-ray selection in finding Compton thick AGN. This is explained by the suppression of X-ray flux in the >10 keV bands at high column densities, which we have shown to be a factor of >10 at N_H=10²⁵ cm⁻² using our Monte Carlo X-ray transmission results.

The XRB synthesis models of Gilli et al. (2007) allow the prediction of the number counts expected from an X-ray survey given the set of parameters of that survey.1 As this sample is not X-ray flux limited, the comparison is not straightforward. However, the Compton thick fraction expected for a survey with our redshift range (0 < z < 0.1) and 2–10 keV luminosity range (10⁴² < L_X < 10⁴⁴, a lower limit of L_X > 10⁴² as we are only considering AGN) lies between ∼2 and ∼20 per cent for X-ray limiting fluxes between ∼10⁻¹³ and ∼10⁻¹⁴ erg cm⁻² s⁻¹, in between which our ‘effective’ X-ray flux limit lies (see Fig. 5). Thus our results would seem roughly consistent with the XRB models.

From the four Compton thick AGN which we manage to measure the N_H directly with trans, in three of these cases, a prominent Fe line aids in the determination of the N_H. However, in the fourth source, NGC 1667, the spectrum does not present a prominent Fe line. Here the model is required to fit a hard excess seen in spectrum, but the N_H is not well constrained (N_H = 271⁺⁵⁷³⁰₋₂₅₁). This is a low signal-to-noise ratio spectrum, and should be considered a marginal Compton thick source. This technique of fitting hard excesses in Brightman & Nandra (2008), where we show that it can be an effective method of identifying heavily obscured nuclei (e.g. the Chandra observation of NGC 4501). In the absence of high signal-to-noise ratio data around the Fe Kα line, or data above 10 keV, this method is the next best thing for picking out Compton thick candidates from X-ray spectral data alone. This may then motivate longer exposures to gain higher signal-to-noise ratio data, or observations of broader banded telescopes such as Suzaku.

We also infer the Compton thick nature of 12 sources from their reflection-dominated spectra, which we define as being those with R_pexmon > 8. Though this definition is arbitrary, it is well supported in most sources by high Fe Kα EWs or a directly measured N_H > 10²⁴ cm⁻². Lowering this criterion to e.g. R_pexmon > 1 would include two further sources, NGC 4631 and NGC 4666, without prior indications of being heavily absorbed. However, these sources do not have well constrained Fe Kα EWs and hence do not support a lower R_pexmon criterion. If a higher value was chosen, e.g. R_pexmon > 20, this would exclude two sources, NGC 1068 and 2MASX J15504152. However, NGC 1068 is well known to be Compton thick, and as such motivated the criterion in the first place. We therefore conclude that the R_pexmon > 8 criterion is best suited to selecting Compton thick AGN.

5.3.2 N_Hdistribution and obscured fraction

The N_H distribution of AGN has a peak at low column densities, which consists of all unobscured sources. For obscured sources, the distribution peaks in the N_H = 10^23-24 cm⁻² bin. We find that 29/60 (=48 ± 6 per cent) of X-ray luminous AGN are heavily obscured (N_H≥ 10²³ cm⁻²). Recent studies into the distribution of absorbing columns, such as that by Tueller et al. (2008), reveal that for their sample of Swift/BAT 14–195 keV selected AGN, 31 ± 5 per cent are heavily obscured, which is a greater than 3σ difference. Their sample does however extend to higher X-ray luminosities, which will include a greater fraction of unobscured sources due to the dependence they find of obscuration on luminosity. We also intend to investigate the N_H distribution for the AGN in our sample, but using optical line emission to define the AGN in Paper II. It is likely here that AGN definition based on X-ray luminosity alone will bias a sample. Those sources which do not meet the luminosity criterion, but have N_H > 10²³ cm⁻² are also likely to be AGN.

Our analysis of the dependence of obscuration on luminosity shows that obscuration depends heavily on the intrinsic power of the X-ray source. The data show that obscuration peaks at L_X = 10⁴² erg s⁻¹, and declines steeply both above and below this. Several previous studies have reported the decline in the obscured fraction above 10⁴² erg s⁻¹ in X-ray selected samples (Ueda et al. 2003; La Franca et al. 2005; Akylas et al. 2006; Tueller et al. 2008; Beckmann et al. 2009), which we confirm here in our MIR selected sample. These results are in support of the ‘receding torus’ model of Lawrence (1991) and Simpson (2005) which describe the decrease in the covering fraction of the torus in AGN with source luminosity, due to sublimation of dust. The decline that we show below 10⁴² erg s⁻¹ may be caused in part by the inclusion of non-AGN powered sources which emit at these lower luminosities. However, recently, Burlon et al. (2010) find compelling evidence for a decrease in the obscured fraction at low X-ray luminosities from analysis of the X-ray luminosity functions of obscured and unobscured AGN from the Swift/BAT survey. Akylas & Georgantopoulos (2009) also evidence for a lower incidence of obscuration at low X-ray luminosities in their optically selected sample. Both sets of authors suggest this may be due to the sustenance of torus by accretion-driven winds modelled by Elitzur & Shlosman (2006). Our result must be confirmed though with a non-X-ray luminosity-dependent method of identifying AGNs, which we intend to do in Paper II using optical AGN classifications.

6 CONCLUSIONS

In summary, we have presented an X-ray spectral analysis of the 126 galaxies with XMM–Newton coverage in the 12MGS, which is a relatively unbiased and representative selection of galaxies. We have done so with the help of new X-ray radiative transfer calculations, based on Monte Carlo simulations which take into account photoelectric absorption, Compton scattering and includes fluorescent Fe line emission. We have compiled new X-ray spectral models from these data, trans and torus which we have presented here. These table models are available publicly on the web at http://astro.ic.ac.uk/mbrightman/home.

Our main results from these models are the following.

• Unobscured AGN with a presumed toroidal distribution of matter around it can achieve a maximum Fe Kα EW of ∼150 eV, a useful limit to consider when assessing if a source is truly unobscured or not. From our calculations, we show that in order to exceed this EW, the line of sight must be obscured with N_H > 10²³ cm⁻² (e.g. NGC 3690).

• For N_H=10²⁵ cm⁻², the flux seen in the 10–40, 20–100 keV (INTEGRAL), and 14–195 keV (Swift/BAT) bands is only 10 per cent of the intrinsic flux, which is important for considering the biases present against hard X-ray selected, heavily obscured AGN. At this N_H, 10 per cent or less of the intrinsic flux is seen in any X-ray band at all redshifts, which reveals the bias present in all X-ray surveys against such heavily obscured systems.

• Using spectral models based on slab geometries such as pexrav or pexmon will underestimate the reflection fraction and intrinsic luminosities with respect to toroidal geometries by a factor of 2–3, leading to underestimation of the intrinsic luminosity of the source, as also found by Murphy & Yaqoob (2009).

By combining the positive attributes of 12 μm selection with the penetrative power of X-rays, we have been able to determine the intrinsic X-ray properties of galaxies such as source power, power-law index, and in addition we have assessed the absorption occurring in the sources. We have improved upon previous works on this sample with better signal-to-noise ratio, spectral resolution and larger bandpass by using XMM–Newton data. We have developed a detailed systematic approach to the fitting of the spectral data, with the aim of determining the intrinsic power, Γ and N_H as accurately as possible. We have paid particular attention to the uncovering of Compton thick sources using our new model, trans and the pexmon model. The main conclusions from this study are the following.

• We find the mean primary power-law index to be 〈Γ〉 = 1.90^+0.05_−0.07 for our sample with an intrinsic spread of σ = 0.31^+0.05_−0.05. The index is consistent with previous works, though we find a larger intrinsic dispersion in our results than previously reported.

• We also find that the mean power-law index for obscured and unobscured sources is consistent with the two source populations being the same, supporting unified schemes.

• We find no dependence of Γ on X-ray luminosity as has been previously reported for local AGN.

• We find an 18 per cent Compton thick fraction for the X-ray AGN defined here, which is higher than the hard X-ray selected samples, but roughly consistent with XRB predictions.

• The obscured fraction for our sample is a strong function of X-ray luminosity, peaking at a luminosity of ∼10^42-43 erg s⁻¹. The decline in the obscured fraction at high luminosities, where these sources are unambiguously AGN, is a confirmation of previous works. The decline at lower luminosities has also been suggested but needs further explanation accounting for optical spectroscopic classifications.

We would like to thank the anonymous referee for the careful reading of this manuscript and the constructive criticism provided. MB would like to thank STFC for financial support. This research has made use of data obtained from the High Energy Astrophysics Science Archive Research Center (HEASARC), provided by NASA’s Goddard Space Flight Center. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. We thank those who built and operate XMM–Newton.

REFERENCES