THE BLACK HOLE SPIN AND SOFT X-RAY EXCESS OF THE LUMINOUS SEYFERT GALAXY FAIRALL 9

2.1.1. Suzaku

A summary of the observations discussed in this paper is presented in Table 1. All Suzaku data reduction was performed within Heasoft v6.11. The version 20110608 of the relevant Suzaku CALDB files were used. The Suzaku data were reduced following the standard procedure described in the "Suzaku ABC Guide."⁷

Table 1. Overview of Observations and Exposures

Observatory	Instrument	Date	ObsID	Exposure
				(ks)
Suzaku B	XISa	2010 May 19	705063010	191
	PIN			162
Suzaku A	XISa	2007 Jun 7	702043010	139
	PIN			127
XMM-Newton B	EPIC-pn	2009 Dec 9	0605800401	91
XMM-Newton A	EPIC-pn	2000 Jul 5	0101040201	16

Note. ^aDenotes that XIS0/XIS1/XIS3 were used. All Suzaku data were obtained after the failure of XIS2.

Download table as:  ASCIITypeset image

For the Suzaku X-ray Imaging Spectrometer (XIS; Koyama et al. 2007), all data were taken in the full window mode. The XIS data files were first reprocessed as recommended by the ABC Guide. For all operating XIS detectors (XIS0, XIS1, XIS3), the source extraction region was chosen to be a circle of radius of ∼4 arcmin and the background was taken from a region on the same chip. Individual spectra and response files for each detector and data mode were created using the tasks xisrmfgen and xissimarfgen. Finally, the spectra and response files of all data modes for a given detector were summed, weighting by exposure time, to yield one spectrum for each XIS. Light curves were also created from the reprocessed event files using xselect.

The hard X-ray detector (HXD; Takahashi et al. 2007) data were first reprocessed as recommended by the ABC Guide, and then filtered using xselect with the standard criteria. Only the PIN data are considered in this work as GSO is not suitable for sources with the count rate as low as Fairall 9. For the PIN, the appropriate "tuned" background files were downloaded from the High Energy Astrophysics Science Archive Research Center (HEASARC). We then extracted source and background spectra from the "cleaned" event files using hxdpinxbpi.

In addition to the reduction steps outlined above, the 2010 May Fairall 9 Suzaku data had to be treated with special caution as it is affected by a variety of calibration/pointing problems. As one can see in Figure 1, the pointing was unstable and the standard attitude solution was inaccurate during this observation, with the satellite wobbling back and forth between two positions separated by 1 arcmin. Although it is known that Suzaku is suffering from an attitude control problem and therefore a mild form of this is already accounted for in the calibration files delivered with the data, it appeared to be too strong to be properly removed with the standard correction (Suzaku memo 2010-05). The best results are achieved by creating a new, corrected attitude file using the tool aeattcor.sl⁸ written by Mike Nowak; a detailed description of this tool can be found in Nowak et al. (2011).

Zoom In Zoom Out Reset image size

Once corrected for, the wobble leaves three remaining imprints on the spectra. First, since the wobble puts some of the source photons into a region of the image plane where the high-energy vignetting function is more poorly known, there are some deviations between the different XIS spectra above 8–9 keV. We choose to use XIS data up to 10 keV but must be cautious about interpreting spectral residuals at the highest XIS energies. Second, as is well known, the XIS spectra are contaminated by absorption from a hydrocarbon layer residing on the optical blocking filter. This contamination is monitored by the Suzaku team and corrected for in the construction of the effective area file. However, the contamination has spatial dependence and, for XIS0, it appears that the wobble during this observation has placed substantial numbers of source photons on parts of the image plane in which the contamination is poorly known. As a result, the XIS0 spectrum deviates from the other XIS spectra below 2 keV. This deviation is significant (reaching values as high as 50%) and forces us to ignore XIS0 data below 2.3 keV. Lastly, due to the vignetting issue, there is no reason to suspect that the standard XIS/PIN cross normalization factor of 1.18 (nominal HXD aimpoint; Suzaku memo 2008-06) is appropriate for this data set—the suppression of XIS flux by the additional vignetting would increase this cross normalization. We must allow this cross normalization to be a free parameter and, indeed, our best-fit models suggest values of 1.3.

2.1.2. XMM-Newton

Unless explicitly stated otherwise, all XIS/EPIC-pn spectra are binned to an S/N of 10. The considered energy ranges for fitting are 0.7–1.5 keV and 2.3–10 keV for Suzaku-XIS, with the energies around 2 keV being excluded because of known calibration issues around the mirror (gold) and detector (silicon) edges. The only exception to this is the XIS0 spectrum for the newest Fairall 9 Suzaku pointing; as already mentioned in Section 2.1.1 this spectrum suffers from uncorrected contamination at soft energies and, therefore, we ignore the region of this spectrum below 2.3 keV. EPIC-pn spectra are used in the 0.5–10 keV band. For all Suzaku pointings, PIN spectra were binned to S/N of 5 and are used in the 16–35 keV band. As our spectral models include a convolution with a relativistic transfer function, which requires an evaluation of the underlying model outside of the energy range covered by the data, we extended all energy grids to energies far beyond the upper energy limit given by the highest data bin considered in fitting. The observations considered in this work span 10 years (Table 1), and hence give an idea of the long-term variability of the object. None of these observations show significant short-term (intra-observation) variability. Therefore, our spectral analysis considers only pointing averaged spectra.

As an initial exploration of these data, we fit a simple absorbed power law to the 2.3–4.5 keV part of the spectrum; Figure 2 shows the ratio of the data in the 0.5–10 keV band to this simple model for all four data sets. All data sets show a prominent narrow, fluorescent iron Kα line (6.4 keV rest frame) as well as the blend of the Kβ/Fe xxvi lines (7.07 keV/6.96 keV rest frame). The spectrum shows clear concavity, revealing the presence of a soft excess and, possibly, a Compton reflection. The two narrow fluorescent iron lines together with the reflection hump are clear evidence for cold reflection in the source, either from the torus or from the outer parts of the accretion disk. None of the spectra show any signs of warm absorption, as already discussed in Section 1. Despite these similarities, the spectral shape varies significantly throughout the years, with the XMM-Newton pointings (2000, 2009) catching the source in a lower flux state. This variability is even more apparent when plotting the unfolded⁹ spectra (Figure 3). As the S/N in the short (2000) XMM-Newton pointing is significantly lower than the other data sets, we do not consider this pointing in the remaining part of this work.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Building upon previous works (Schmoll et al. 2009; Emmanoulopoulos et al. 2011), we construct a multi-component spectral model to describe our spectra. The underlying primary continuum is described by a power law with photon index Γ. We use the model pexmon (Nandra et al. 2007) to describe the cold reflection of this power law from distant material. In particular, pexmon self-consistently models the strength of the narrow Fe Kα, Fe Kβ, and Ni Kα lines, the Compton shoulder of the Fe Kα line, and the Compton reflection continuum. The strength of the distant reflection is characterized by the usual reflection fraction R normalized such that R = 1 corresponds to a reflector that subtends half of the sky as seen from the X-ray source. As some parameters for pexmon cannot be determined from our fits we fix them to certain values; we fix the inclination to 60° (most probable value for an isotropic distribution) and abundances to solar values. The high-energy cutoff of the continuum power law is fixed to 300 keV.

To describe the ionized reflection associated with the inner accretion disk, we used a modified version of reflionx developed by Ross & Fabian (2005). Starting from the publicly available version, we redefined the normalization parameter (norm → norm/ξ) such as to statistically decouple the true flux normalization from the ionization parameter ξ. This definition results in a more rapid and robust convergence of the spectral fit. For the ionized reflection model, the iron abundance and ionization parameter are allowed to vary freely. The photon index of the irradiating continuum, however, is tied to the index for the continuum underlying the cold reflection. The ionized reflection component is then relativistically blurred using the model relconv (Dauser et al. 2010); this naturally gives rise to a relativistic iron line, blurred Compton reflection hump, and a soft excess. The radial dependence of the emissivity of the reflection component is assumed to have a broken power-law form, breaking from $r^{-q_1}$ to $r^{-q_2}$ at a radius R_break. The inner edge of the X-ray reflection regime is taken to be at the ISCO (Reynolds & Fabian 2008), and the outer edge was fixed to 400 R_g. Provided that q₂ > 2, the relativistic blurring kernel is only weakly dependent upon this outer radius. The accretion disk inclination i and the black hole spin a were left as free parameters. In addition, photoelectric Galactic absorption is modeled with TBnew¹⁰, a newer version of TBabs (Wilms et al. 2000), with cross sections set to vern and abundances set to wilm.

A careful examination of the residuals shown in Figure 2, especially for observation Suzaku B, reveals a line-like feature between cold-FeKα and cold-FeKβ lines. While some or all of this emission may be associated with the blue peak of the broad iron line, another possibility is that we are seeing FeKα line emission from Fe xxv (which produces a complex at 6.7 keV). It therefore seems possible that the spectral model discussed above is missing an emission line system producing Fe xxv and possibly Fe xxvi line emission. Hence we include an additional photoionized emission component (Figure 4); physically, this may be an accretion disk wind (warm absorber) which is out of our line of sight. The photoemission is parameterized by the ionization parameter and norm. It is modeled using a table model calculated using xstar2xspec from the XSTAR model photemis,¹¹ assuming irradiation by a power law L_ε∝ε^α with α = −1 and a number density of 10¹⁰ cm⁻³. At the ionization stages relevant for this work (log ξ > 3.6) there is no dependence of the spectral shape on column density for the model, its value was therefore kept fixed at 10²² cm⁻³. We note that the inclusion of this emission line component does not preclude the possibility that the residuals at 6.7 keV are from the blue peak of the iron line, but it does allow the model to explore any degeneracy created by the superposition of the broad iron line and the narrow ionized line emission.

Zoom In Zoom Out Reset image size

When applying this spectral model to the data, a cross-calibration constant was introduced for Suzaku between XIS and PIN (and in case of the 2010 Suzaku pointing, also between the XISs). The constant was fixed to the values given in the Suzaku ABC Guide, 1.16 for XIS aimpoint and 1.18 for HXD aimpoint. An exception to this is observation Suzaku B; as mentioned above, due to the pointing issues, there is no expectation that the fiducial XIS/PIN cross normalization should be appropriate and hence it was left as a free parameter.

We begin with this model as our base model and, in Section 3.2, we fit this model to each data set individually. After noting problematic issues associated with these fits, we perform simultaneous fits to data from different observations of Fairall 9, under the assumption that the most fundamental parameters of the system (spin, accretion disk inclination, and iron abundance) do not change—this so-called multi-epoch fitting is presented in Section 3.3 and, as we shall see, yields insights into the spectral properties of this AGN.

Throughout this work, the spectral analysis were performed with the Interactive Spectral Interpretation System¹² (ISIS Version 1.6.0-7; Houck & Denicola 2000) and the newest XSPEC 12.0 models are used (Arnaud 1996). All uncertainties are quoted at the 90% confidence level for one interesting parameter (Δχ² = 2.7).

We begin by conducting independent fits of our base spectral model to the three remaining observations under consideration, XMM-Newton B, Suzaku A, and Suzaku B. Our base model provides a good description of the data for three data sets, with reduced-χ² values between 1.00 and 1.04 (Table 2). A sample spectrum with residuals is shown in Figure 5 illustrating the good quality of the fit.

Zoom In Zoom Out Reset image size

We argue below that these fits are incomplete and hence that the inferred parameters are thus not necessarily correct. However, it is instructive to compare our results from these fits (Table 2) with previously published works. For Suzaku A, our derived parameters are similar to those found by Schmoll et al. (2009) with the exception of black hole spin, disk inclination, and emissivity index. Schmoll et al. (2009) apply the constraint that the inner emissivity index q₁ should not exceed 5; if we impose the same constraint, we recover a very similar intermediate spin and inclination constraint. However, our base model allows the emissivity index to be much steeper (formally pegging at q₁ = 10) and we find, as a result, that the inferred black hole spin tends to high values. For Suzaku B, our fits are in agreement with those of Patrick et al. (2011), who find that spin is unconstrained by this data set. For XMM-Newton B, we find a spin parameter and inclination that is strongly discrepant with those derived in Emmanoulopoulos et al. (2011). We attribute this difference to the inclusion of a photoemission component, a newer absorption model (TBnew versus wabs), and a different model construction.

However, just looking at the results of our own uniform analysis, we find important and informative inconsistencies. Comparing the fit parameters derived from the different observations, we find that the black hole spin, the disk inclination, and the iron abundance appear to have significantly different values between the three data sets (Table 2). Naturally, we would expect these physical quantities to be constant on the time spanned by these observations.

As we show below, a multi-epoch analysis reveals the need for a new continuum component to describe the soft excess. Foreshadowing that discussion, we note that adding such a soft excess component to the model for each individual data set results in unconstrained spin parameters and emissivity indices, i.e., the individual data sets do not possess the S/N to determine spin and inclination in the more complex models; there is a significant tradeoff between the parameters of the soft component and the spin and inclination for each individual data set. This also implies that the extreme values of spin and q₁ found in a fit of the base model to Suzaku A may arise due to a model which is (possibly falsely) attempting to fit a very smooth soft excess with an ionized reflection spectrum, necessitating extreme broadening.

The inconsistencies resulting from independently fitting our base model to the individual pointings lead us to the use of multi-epoch fitting. In multi-epoch fitting, we assume that the source spectrum is always composed of the same principal physical components and that, when physical considerations demand, parameters are forced to have the same value for all epochs. In our case, we demand that the black hole spin parameter, the inclination of the disk, the iron abundance, and the strength of the photoionized emission component (believed to originate from an extended AGN wind) must have common values across the fits to all of the data sets. In all cases, these parameters are believed to remain fixed for any given AGN over human timescales. Aside from these additional constraints on the parameters, the model setup is the same as for the fitting of the individual pointings. We allow the slope/normalization of the primary power law, as well as the normalization, emissivity profile, and ionization state of the disk reflection, to float freely between data sets—these parameters depend upon the structure/geometry of the disk corona that can plausibly change on timescales that are (much) shorter than the inter-pointing spacing. Fitting in this manner, we find that the broken power-law emissivity profile is ill constrained and a simple power-law emissivity profile is sufficient to describe the data very well. Furthermore, to reduce the number of spectral bins in each data set, we co-added the front-illuminated XIS data (XIS0 and XIS3); this eases the significant computational expenses of multi-epoch fitting.

Statistically, fitting this model to the full multi-epoch data set produces a good fit, with reduced χ² of 1.06. The best-fit parameters and errors are listed in Table 3. The best-fitting spin is a little higher (a = 0.71^+0.08_{− 0.09}) and the inclination a little lower (i = 37⁺⁴_{− 2}°) than found in the previous study of Schmoll et al. (2009) but, given the error bars, there is no strong discrepancy. The ionization parameter either takes rather low values (XMM-Newton, Suzaku A) or very high values (Suzaku B). This is directly tied to the fact that, apart from the iron line itself, there are no other strong features in the soft X-ray band that the reflection models are locking onto in the spectrum. One aspect of this fit that gives us pause is the high PIN/XIS cross normalization component, C_XISf-PIN ≈ 1.74. While we have already acknowledged that the pointing errors would render invalid the standard value of C_XISf-PIN = 1.18, this measured value would imply a ∼30% suppression of the net XIS count rate below the fiducial value, a significantly larger depression than is reasonably expected from the pointing problems. A more likely possibility is that the spectrum has curvature in the concave sense so that the true PIN-band flux is higher than predicted by our base spectral model.

Partly motivated by these considerations, but also guided by the known phenomenology of AGNs, we explore the hypothesis that we are missing another component that can contribute to the soft excess (above and beyond any soft excess emission from the blurred uniformly ionized reflection, that models the broad iron line). A common phenomenological model for the soft excess is a simple blackbody component. Adding a blackbody component to the spectral model does not yield improved fits—none of the data sets show the strong curvature representative of the Wien tail of a blackbody. However, significant improvements in the goodness of fit are achieved using a soft excess model consisting of a "warm" thermal Comptonization component. In detail, we add a thermal Comptonization component described by the comptt model (Titarchuk 1994) with a seed photon temperature of 40 eV (representative of the expected thermal emission from the AGN disk), an optical depth τ, and an electron temperature kT.

When applied to the multi-epoch data, the inclusion of this soft excess component leads to an improvement in the goodness of fit by Δχ² = 198 for nine additional parameters. The best-fit parameters and corresponding errors can be found in Table 4. As compared with the base model above, the best-fit spin has slightly decreased to a = 0.52^+0.19_{− 0.15} and the best-fit inclination has increased to i = 48⁺⁶_{− 2}°. While the change in spin between the two models is within the error bars, the inclination change is significant and brings the results into line with the conclusions of Schmoll et al. (2009). We find that the ionization state of the inner accretion disk is significantly higher at lower flux states. The biggest surprise is that, once we include the explicit soft excess component, the iron abundance of the accretion disk is pushed toward the maximum value tabulated in the model, Z = 10 Z_☉. The formal 90% lower limit on iron abundance is Z > 8.2 Z_☉. From a modeling perspective, such a high iron abundance means simply that the iron line is strong in comparison to the Compton hump.

However, another possibility is that the reflection from the inner accretion disk is characterized by multiple ionization components (Nardini et al. 2011) and a highly blurred, high-ionization reflection component can account for the soft excess in a similar fashion than the additional Comptonization component. To explore this model, we refit the multi-epoch data with two disk reflection components, each of which has its own ionization parameter and emissivity index. The only other difference from the multi-epoch fit above is that the iron abundance is assumed to be the same for all reflection components included in the model. Such a model leads to a fit solution of almost comparable quality to that employing the Comptonization-based soft excess component (Table 5). In order to describe the smooth soft excess, one of the disk reflection components adopts a high-ionization state (ξ_a ∼ 1000) and a very steep emissivity profile (q_a > 8) and the black hole spin parameter becomes large (a > 0.96). At the same time, much less broadening/blurring is needed in order to describe the broad iron line, and so the other disk reflection component adopts a low ionization (ξ_b ∼ 1–10) and a shallow emissivity profile (q ≲ 2.2). This solution avoids the extreme iron abundance, with an inferred iron abundance of 0.68–0.85 Z_☉.