On the relativistic iron line and soft excess in the Seyfert 1 galaxy Markarian 335
Paul M. O'Neill, Kirpal Nandra, Massimo Cappi, Anna Lia Longinotti, Stuart A. Sim
aa r X i v : . [ a s t r o - ph ] A ug Mon. Not. R. Astron. Soc. , 1–6 (2007) Printed 22 October 2018 (MN L A TEX style file v2.2)
On the relativistic iron line and soft excess in the Seyfert 1 galaxyMarkarian 335 ⋆ Paul M. O’Neill, , † Kirpal Nandra ‡ , Massimo Cappi , Anna Lia Longinotti andStuart A. Sim Astrophysics Group, Imperial College London, Blackett Laboratory, Prince Consort Road, London SW7 2AZ School of Computing & Mathematics, Charles Sturt University, P.O. Box 588, Wagga Wagga NSW 2678, Australia INAF-IASF Bologna, Via Gobetti 101, I-40129 Bologna, Italy XMM-Newton Science Operation Centre, ESAC, ESA, Apartado 50727, E-28080 Madrid, Spain Max-Planck-Institut fur Astrophysik, 85741 Garching, Germany
Accepted. Received
ABSTRACT
We report on a 133 ks
XMM-Newton observation of the Seyfert 1 galaxy Markarian 335.The 0.4–12 keV spectrum contains an underlying power law continuum, a soft excess below2 keV, and a double-peaked iron emission feature in the 6–7 keV range. We investigate thepossibility that the double-peaked emission might represent the characteristic signature ofthe accretion disc. Detailed investigations show that a moderately broad, accretion disc lineis most likely present, but that the peaks may be owing to narrower components from moredistant material. The peaks at 6.4 and 7 keV can be identified, respectively, with the moleculartorus in active galactic nucleus unification schemes, and very highly ionized, optically thingas filling the torus. The X-ray variability spectra on both long ( ∼ ks) and short ( ∼ ks) timescales disfavour the recent suggestion that the soft excess is an artifact of variable,moderately ionized absorption. Key words: galaxies:individual:Markarian 335 – galaxies:active – galaxies:Seyfert – X-rays:galaxies.
Studying the X-ray emission from active galactc nuclei (AGNs) of-fers the potential of observing relativistic effects. The X-ray emis-sion, being produced deep in the potential well of the putativeblack hole, is predicted to be affected by gravitational redshift andDoppler effects. Emission lines may thus become broadened andskewed, the line profiles yielding information on the inner accre-tion flow such as the geometry, disc emissivity and black hole spin(e.g. Fabian et al. 1989; Stella 1990; Laor 1991). The Doppler ef-fects in particular can give rise to a broad Fe K α line having acharacteristic double-horned structure (Chen et al. 1989).At energies below a few keV, an enhancement of flux abovethe underlying power law is commmonly seen in AGN spectra. Theorigin of this so-called ‘soft excess’ is a long-standing, unsolvedpuzzle in AGN studies (see review by Mushotzky et al. 1993).The possible identification of the soft excess as owing to inverse- ⋆ Based on observations obtained with XMM-Newton, an ESA sciencemission with instruments and contributions directly funded by ESA Mem-ber States and NASA † [email protected] ‡ [email protected] Compton scattering of UV disc photons has been challenged bythe apparent rather constant ‘temperature’ of the excess (e.g.,Gierlinski & Done 2004; Crummy et al. 2006). Emission lines as-sociated with disc reflection (e.g. Crummy et al. 2006), or the influ-ence of a moderately ionized ‘warm’ absorber (Gierlinski & Done2004), have been proposed recently to explain the soft excess.The Seyfert 1 galaxy Markarian 335 ( z = 0 . ) exhibitsboth a soft excess and a broad emission feature in the region ofthe iron line. The soft excess has been observed by various mis-sions, starting with EXOSAT (Pounds et al. 1987). A
ROSAT spec-trum could be modelled equally well as either a double power-law or a power-law with an absorption edge (Turner et al. 1993).The soft excess in an
ASCA spectrum could be modelled as ei-ther a blackbody (Reynolds 1997), additional steep power lawcomponent (George et al. 1998) or reflection from an ionized disc(Ballantyne et al. 2001). Reynolds (1997) found also that the addi-tion of an absorption edge to the underlying hard power-law com-ponent improved the fit. A
BeppoSAX spectrum could be modelledsimilarly to the
ASCA spectrum (Bianchi et al. 2001). A narrowiron line, consistent with arising from neutral material, was de-tected in the
ASCA spectrum with an equivalent width (EW) of ∼ eV (Nandra et al. 1997). When fitted instead with a rela-tivistic line and reflection continuum the EW was ∼ eV. The c (cid:13) P.M. O’Neill et al.
BeppoSAX spectrum could likewise be modelled with a relativisticline having an EW of a few hundred eV. A narrow line at 6.4 keVis also likely to be present in the
BeppoSAX data.
XMM-Newton first observed Mrk 335 in 2000 December for35 ks. Gondoin et al. (2002) found excess emission in the 5–7 keV range above the underlying power-law, which could be de-scribed by relativistically blurred emission from highly ionizediron. The combination of reflection from an unblurred , highly ion-ized disc and Bremstrahlung emission was able broadly to describethe entire 0.3–10 keV range, with the reflection flux contributing ∼ per cent of the soft excess. Crummy et al. (2006) also anal-ysed these data and were able to model the spectrum as relativis-tically blurred ionized disc reflection. Importantly, Gondoin et al.(2002) examined the Reflection Grating Spectrometer (RGS) spec-trum and found no evidence of absorption or emission from ionizedgas. Longinotti et al. (2006) recently reanalysed the European Pho-ton Imaging Camera (EPIC) data from this observation, confirmingthe existence of a broad emission feature and noting an absorptionfeature at ∼ . keV. XMM-Newton re-observed Mrk 335 in 2006 for 133 ks. Thepurpose of these observations was to better characterise the ironline emission. Moreover, being long and nearly uninterrupted, thisobservation offers the best data thus far to study the variability. Wepresent here an initial analysis of the data collected by the EPIC PNinstrument.
XMM-Newton observed Mrk 335 between 2006 January 03 and05, for a duration of 133 ks. We present here an analysis of theEPIC PN data. The observation was conducted in Small Windowmode, thus avoiding photon pile-up, and the Thin filter was used.Source events were extracted from a circular region, centred atthe X-ray centroid, with a radius of 680 detector pixels (34 ′′ ). Back-ground events were extracted from two rectangular regions with acombined area 3.4 times larger than the source region. Backgroundflares were present at the beginning and end of the observation, andthe exclusion of these intervals resulted in a low-background dura-tion of 115 ks. The final usable data train contained gaps amountingto ∼ . per cent of this duration.Source and background light curves were extracted for vari-ous energy bands using time resolutions of 200 and 1000 s, andeach time bin was required to be fully exposed. The fractional root-mean-square (rms) variability spectrum (see, e.g. Vaughan et al.2003) was calculated from the 1000 s light curves, each of whichcomprised 104 bins over the 115 ks duration. The uncertainties inthe rms measurements were determined using Monte Carlo simu-lations. Each simulation involved perturbing the observed countingrates with a Gaussian deviate with a standard deviation equal tothe size of the observed error bar. The fractional rms was then cal-culated from the synthetic light curve. We performed 10,000 suchsimulations, and the standard deviation of the simulated rms valueswas adopted as the 1 σ uncertainty owing only to Poisson noise.As well measuring variability in the conventional manner, we alsocalculated the point-to-point fractional rms spectrum to probes thevariability on relatively short timescales (Edelson et al. 2002).Time-averaged source and background spectra were extractedusing the set of events present in the 1000 s light curves. The The observation identification number is 0306870101. ( c oun t s s − ) I n t en s i t y Time (s)
200 s bins 0.4−12 keV ( pe r c en t ) F r a c t i ona l r m s Energy (keV)1000 s bins
Figure 1. spectral channels were grouped so that: there were no more that 2groups per resolution full-width-at-half-maximum, and each groupin the source spectrum possessed at least 20 counts. The source re-gion spectrum contained ∼ . × counts (0.4–12 keV ), ofwhich 0.14 per cent are expected to be background. The quoted un-certainties in the spectral fits correspond to ∆ χ = 1 , unless statedotherwise. These may underestimate the true uncertainties whenmultiple parameters are fitted (Lampton et al. 1976). In Fig. 1 (top) we present the 0.4–12 keV light curve, using atime-resolution of 200 s, which has a fractional rms variability of . ± . per cent. The rms spectra, calculated using a timeresolution of 1000 s, are shown in Fig. 1 (bottom).The conventional fractional rms spectrum could be satisfac-torily described with a constant of 12.3 per cent ( χ / DOF =34 . / ). The point-to-point spectrum, while still being rather flat,cannot be well described with a constant ( χ / DOF = 72 . / ).Three energy bands of importance are 0.4–0.8, 0.8–2 and 2–12 keV:an ionized absorber with varying opacity, if present, would induceenhanced variability in the 0.8–2 keV band (Gierlinski & Done2006). Considering first the conventional rms spectrum, the val-ues in these three bands are . ± . , . ± . and . ± . per cent, respectively. The uncertainties in these val-ues suggest that any enhancement in the 0.8–2 keV range is con-strained to be less than ∼ . per cent. In the point-to-point rmsspectrum the corresponding rms values are . ± . , . ± . and . ± . per cent, which reveal the absence of a peak in the0.8–2 keV range over short timescales.The fractional rms in the 5.6–7 keV band (i.e., the band con-taining Fe line emission; see below) is . ± . per cent. Thetime-averaged flux in this band is a factor of 1.12 greater than the ‘PI’ channels 380–11995. c (cid:13) , 1–6 he relativistic iron line and soft excess in Markarian 335 D a t a / M ode l Energy (keV)5 6 7 . Figure 2.
Ratio between the observed flux and the best-fitting power-lawfitted over the observed 3–4.5 and 7.5–12 keV energy ranges, showing theiron line and soft excess. The inset shows the double peaked profile. Theenergy scale corresponds to the observed frame and the left- and right-handdotted lines indicate rest energies of 6.4 and 6.97 keV, respectively. Thehighest residual seen at 12 keV contributes only ∼ to the χ of the power-law fit. underlying power-law. If the excess of flux in this band is non-varying, then the rms should be suppressed by the same factor. Theuncertainty in the rms is too large to reach a firm conclusion regard-ing the variability or otherwise of this feature, but it is consistentwith being as variable as the power-law. A power-law modified by Galactic absorption of N H = 3 . × cm − was fitted over the observed frame 3.0–4.5 and 7.5–12 keV energy ranges. The best-fitting power-law had a photon in-dex of Γ = 2 . ± . with an unabsorbed 2–12 keV flux and lu-minosity of . × − ergs cm − s − and . × ergs s − ,respectively. The ratio between the observed flux and best-fittingpower-law is shown in Fig. 2 (top). The presence of a soft excessis clear, reaching a maximum of ∼ times the flux of the extrapo-lated power-law. Double-peaked line emission is also visible in the6–7 keV range. Note that the highest residual seen at 12 keV con-tributes only ∼ to the χ of the power-law fit. Unlike the earlier XMM-Newton observation (Longinotti et al. 2006) there is no clearabsorption feature at ∼ . keV; we leave a detailed investigationfor future work.We initially attempted to parametrize the line complex usingmodels with two or more Gaussians fitted over the 3–12 keV range.A model with two narrow ( σ = 1 eV ) lines yielded χ / DOF =145 . / . We allowed the width of the ∼ . keV line to vary,and the fit improved to χ / DOF = 98 . / , with rest-frameline energies of . ± . keV (broad line) and . ± . keV(narrow line). These energies are consistent with those expected forFe I K α and Fe XXVI Ly α , respectively. The width of the broad linewas σ = 0 . ± .
03 keV , and the EWs of the broad and narrowlines, respectively, were ± and ± eV. We then addeda narrow line and the fit improved to χ / DOF = 95 . / . The The Galactic NH was obtained using the NASA HEASARC ‘nH’ tool,http://heasarc.gsfc.nasa.gov/docs/tools.html . The values H = 70 and λ = 0 . were used in calculating the lumi-nosity. two narrow lines had best-fitting energies of . ± . and . ± . keV, and the corresponding EWs were, respectively, ± and ± eV. The broad line had a best-fitting energy, width and EWof . ± . , . ± . keV and ± eV, respectively.The joint confidence intervals of the 6.4 keV line fluxes indicatethat these are greater than zero with a significance of 91 (narrowline) and > . per cent (broad line). We then replaced the broadGaussian with two narrow lines and the fit worsened to χ / DOF =103 . / , with line energies of . ± . and . ± . keV.A double-peaked profile is a natural characteristic of relativis-tic disc lines, so it may be possible to interpret the entire line profilein this framework, with the 6.4 keV peak representing the red hornand the 7 keV peak the blue horn. We test this explicitly below. To model reflection from a neutral disc we have modified the re-flection continuum model
PEXRAV (Magdziarz & Zdziarski 1995).The enhanced model, which we refer to as
PEXMON , includes theFe K α (including the Compton shoulder), Fe K β and Ni K α emis-sion lines (see Nandra et al. 2007, for more details). The depen-dence of line flux on photon-index and inclination is based onthe results of George & Fabian (1991). In PEXMON the reflectioncontinuum and emission line fluxes are linked, as expected fromphysical models, and the constraints on the ‘reflection fraction’ arelargely from the Fe K α line. To construct a disc line model we con-volved a PEXMON component with the relativistic blurring model
KDBLUR
2, which itself uses the kernel of the
LAOR disc line model(Laor 1991; Fabian et al. 2002). Following Nandra et al. (2007), thedisc emissivity power-law index was fixed to break from a value of0 within the ‘break radius’, r br , to a value of − for larger radii.The disc inner radius was fixed at its minimum value of 1.235 r g ,which corresponds to the innermost stable orbit for a maximallyspinning black hole, and the outer radius was fixed at the maximumpermitted model value of 400 r g . The inclination and r br were freeto vary. With the assumed emissivity, this model well approximatesa point source illuminating a slab, with a peak emissivity at r br anda height of the same order. Note that this emissivity law is non-relativistic, and does not account for enhanced inner disc reflectionowing to light bending effects (e.g. Martocchia et al. 2000).A single neutral disc line was unable to model the entire lineprofile. A large component of line flux is required at the Fe I rest en-ergy of 6.4 keV. This would require the red-horn to be only slightlyshifted in energy via gravitational and Doppler effects. It is not pos-sible to then also produce a blue horn that is significantly shiftedfrom 6.4 keV. Moreover, the blue horn is expected to be more in-tense than the red horn.We therefore attempted to model the profile with a combi-nation of distant, neutral reflection and a disc line. The formercomponent is intended to account primarily for the narrow emis-sion at 6.4 keV and we modelled it using an unblurred PEXMON component with a fixed inclination of 60 degrees. The disc line,on the other hand, is intended to model the blue wing and broadcomponent of the profile. This model yielded a satisfactory fit( χ / DOF = 99 . / ). The break radius was constrained to beless than 70 r g (95 per confidence), with a best-fitting value of 4 r g .While the overall shape of the profile is reproduced, the blue-hornin this model is unable to account for the sharpness of the observedline at 7 keV.We next explored the possibility that there is an Fe XXVI Ly α emission component contributing to the blue wing. This line mightphysically be produced via fluorescence in optically thin plasma. c (cid:13) , 1–6 P.M. O’Neill et al.
We added a narrow ( σ = 1 eV) Gaussian with a fixed energy of6.97 keV to the best-fitting disc line model. The fit was significantlyimproved ( χ / DOF = 89 . / ), with a reduction in χ of 9.8.The break radius and inclination were poorly constrained, with theone-sided 95 per cent confidence intervals ( ∆ χ = 4 . ) restrict-ing the radius and inclination, respectively, to > r g and > de-grees. Indeed, varying r br through values above about ∼
150 r g did not yield any variation in χ . Relative to the power-law con-tinuum, the reflection fraction of the blurred PEXMON componentwas . ± . , which corresponds to the disc subtending a solidangle of . π for a slab geometry. The reflection fraction of the un-blurred PEXMON component and the EW of the Fe
XXVI Ly α linewere . ± . and ± eV, respectively, and the line flux was (4 . ± . × − photons cm − s − .Finally, we removed the PEXMON component representingdistant reflection to investigate whether the profile could be mod-elled with only a relativistic disc line and a narrow Fe
XXVI Ly α line. In this case the best-fitting break radius was constrained to beabove ∼
100 r g , and the reflection fraction increased to . ± . .The fit became significantly worse, with χ increasing by 7.4. Part of the difficulty in reproducing the observed profile with a neu-tral disc line is that, in order to reproduce the sharp blue peak at ∼ keV, a relatively high inclination with minimal gravitational broad-ening is required. However, the red horn in such a model, beingnarrow and at an energy below 6.4 keV, fails to fit the broad ob-served excess around the Fe I rest energy. An ionized disc, on theother hand, can produce Fe lines with rest energies between 6.4 and7 keV. With variable ξ and inclination an ionized disc model might,then, be flexible enough to describe the observed profile.To construct an ionized disc line model we used the REFLION model (Ross & Fabian 2005), which incorporates both line emis-sion with Compton broadening and the reflection continuum. RE - FLION is defined at a limited number of values of ionization pa-rameter, ξ , and Γ , and interpolation between these grid points mayyield deviations from the true model spectrum (2006, R. Ross, priv.comm.). We consider ξ to be a free parameter but keep its valuefixed during minimisation.We constructed a model comprising distant reflection ( PEX - MON ), plus relativistically blurred ionized reflection (
KDBLUR
REFLION ), plus a narrow Fe
XXVI Ly α line. The ionizationparameter was fixed at the minimum permitted value of ξ = 30 ,which yielded χ / DOF = 87 . / . We then increased ξ to itsnext defined value of ξ = 100 , and the best-fitting model yielded χ = 87 . / . When the ionization was increased further, to ξ = 300 , the fit worsened to χ = 99 . / . An ionized discmodel is thus not preferred over a neutral disc. We repeated thesethree fits without a narrow line at 6.97 keV, and confirmed that, asfor the neutral disc model, the inclusion of the Fe XXVI Ly α lineimproved the fit. The broad emission feature in the earlier
XMM-Newton spectrumcould be well-described by either a relativistic disc line (the pre-ferred interpretation) or a neutral partial covering absorber model(Longinotti et al. 2006). One of the goals of obtaining the recent, http://heasarc.gsfc.nasa.gov/xanadu/xspec/models/reflion.html longer observation was to break the degeneracy between these twointerpretations. We have, therefore, attempted to fit the new datawith a partial covering model.We initially fitted the spectrum with a partial covering ab-sorber plus a PEXMON component to represent distant, neutral re-flection, which yielded χ / DOF = 101 . / . The addition of anarrow Fe XXVI Ly α emission line improved the fit ( χ / DOF =96 . / ), with a covering fraction of ± per cent. This model,while yielding a worse fit compared to the neutral disc line, is for-mally acceptable, so we cannot rule out a partial covering inter-pretation. The best-fitting photon index for this model is Γ ∼ . ,compared to Γ ∼ . for the neutral disc line model. The
REFLION model includes the soft X-ray line emission that isproposed by, e.g., Crummy et al. (2006) to explain the soft excess.Detailed modelling of the soft excess is beyond the scope of thisLetter. However, to determine roughly the extent to which the softexcess might be attributed to ionized reflection we examined theratio between the observed spectrum in the 0.4–3 keV range andthe extrapolated best-fitting 3–10 keV ionized disc model.We initially used the best-fitting model with ξ = 30 and SolarFe abundance. The observed flux at 0.4 keV is ∼ . times greaterthan the extrapolated model. Crummy et al. (2006) fitted the earlier XMM-Newton observation with an Fe abundance of 0.7. We there-fore reduced the abundance to 0.5, which is the closest value to 0.7for which
REFLION is defined. This reduces the Fe line flux relativeto the soft flux. We refitted the 3–10 keV spectrum and the 0.4–3 keV residuals were up to a factor of ∼ . above the extrapolatedmodel. Reducing the abundance to the next lowest defined value of0.2 yielded a significantly worse 3–12 keV fit ( ∆ χ = 19 . ).We then increased the ionization parameter to ξ = 300 , withan Fe abundance of 0.5. This increased the model soft flux andreduced the 0.2–3 keV residuals to be up to a factor of only ∼ . above the model. However, the rather smooth shape of the softexcess below 2 keV is not reproduced. An increase in the relativisticblurring could produce a smoother spectrum, but in our simple testthe disc line parameters are constrained by the requirement to fitthe 3–12 keV spectrum. We have conducted an initial analysis of a 133 ks
XMM-Newton ob-servation of the Seyfert 1 galaxy Markarian 335. The time-averagedspectrum in the 3–12 keV range could be described as a power lawcontinuum (
Γ = 2 . ± . ) with a double peaked emission fea-ture in the 6–7 keV range. A strong soft excess is present below ∼ keV. α emission A relativistic disc line alone could not describe the line profile. Amodel comprising a disc line and reflection from neutral, distantmaterial yielded a satisfactory fit. The blue horn in the best-fittingdisc line is not able to describe very well the sharpness of the ob-served peak. Increasing the disc ionization did not improve the fit.However, the addition of Fe
XXVI Ly α emission did significantlyimprove the fit. The iron line profile is, then, likely owing to a su-perposition of distant reflection from neutral material, relativisti-cally blurred reflection from a neutral accretion disc and a narrow c (cid:13) , 1–6 he relativistic iron line and soft excess in Markarian 335 emission line from highly ionized gas. Note that the level of rela-tivistic blurring in our fit is rather moderate. Therefore, our phys-ically motivated emissivity law (see Section 4.2) is preferred overthe often-used unbroken powerlaw.We identify the distant reflection component as originatingfrom the torus in AGN unification schemes. For the torus geometrydefined in Ghisellini et al. (1994), an assumed PEXMON inclinationangle of 60 degrees corresponds to a torus half-opening angle of ∼
30 degrees. The observed
PEXMON component has a reflectionfraction of . ± . , compared to a value of ∼ ∼ degrees yields a reflection fraction compatible withthat observed. The reflection flux depends on the optical depth onthe torus (Ghisellini et al. 1994), and this might also explain thelow observed reflection fraction. The reflection fraction of both thedistant and the disc line components are well within the ranges ob-served in other Seyfert galaxies (e.g. Nandra et al. 2007).The Fe XXVI Ly α emission line, being narrow, must originatefrom optically thin material. An ionized disc, for example, cannotproduce such a narrow line, owing to Compton broadening. Thethin gas might be identified as the hot gas that fills the torus andwhich is responsible for scattering the broad line optical emissioninto the line of sight (Krolik & Kallman 1987; Antonucci & Miller1985). An Fe XXVI absorption edge at 9.29 keV should accompanythe Fe
XXVI Ly α line. For a fluorescent yield of 0.7, the upper limiton the Fe XXVI absorption edge flux in the spectrum, together withthe observed line flux, corresponds to a 95 per cent lower limit onthe covering fraction of ∼
100 per cent. This rather large fractioncan perhaps be explained if the torus does not obscure the hot gason the far side of the nucleus, and there may be a deficit of gas inthe line-of-sight to the central engine.A partial covering interpretation for the line profile was sat-isfactory, yet worse than disc reflection. Partial covering yields asteeper photon index than for the disc line, so high energy data(e.g.,
Suzaku ) will be able to robustly distinguish between thesetwo interpretations (see, e.g. Reynolds et al. 2004).
The soft excess seen in many AGN X-ray spectra is possibly an arti-fact of ionized absorption. The RGS data from the previous
XMM-Newton observation of Mrk 335 (Gondoin et al. 2002), however,revealed no evidence for absorption or emission from ionized gas.We visually inspected the combined fluxed RGS spectrum from thenew data and the only clear feature is an O I absorption line at0.54 keV. However, this lack of lines is possibly owing to strongDoppler smearing (Gierlinski & Done 2004).A moderately ionized absorber can yield a soft excess of thekind observed in Mrk 335, where absorption by O VII , O
VIII andFe-L are important. For lower ξ the soft spectrum would showheavy absorption, whereas for very high ionization no significanteffects on the soft spectrum would be seen at all. For the absorbersproposed to explain soft excesses, Gierlinski & Done (2006) ar-gued that variations in the continuum luminosity will necessarilyinduce variations in ionization, and they predicted excess variabil-ity around 1 keV. Therefore, the presence of an absorber might beinferred by enhanced variability in the 0.8–2 keV range. Mrk 335exhibits no such enhancement, so we cannot invoke a moderatelyionized absorber with varying ξ to explain its soft excess. It is worthnoting that a flat rms spectrum might be produced either in the pres-ence of non-varying absorption or in the context of a partial cover- ing absorber model (see Boller et al. 2002; Tanaka et al. 2004, forthe case of 1H 0707 − XMM-Newton observation of Mrk 335.The purpose of our simple test was to investigate the extent towhich the disc line model could explain the soft excess. Whenwe extrapolated the best-fitting 3–12 keV model down to soft en-ergies we found that disc reflection underpredicts the soft excessflux by up to a factor of ∼ . , even with a reduced Fe abun-dance of 0.5. Increasing the disc ionization reduced the residuals,but the observed soft excess continuum is much smoother thanthe model. A more complicated reflection model could perhapsbe invoked. For example, the soft flux might originate from veryclose to the black hole, in a region that is both ionized and highlysmeared, while the Fe line flux is owing to neutral reflection fromfurther out in the disc. We note in passing that a flat rms spectrummight be associated with ‘regime I’ in the so-called ‘light bend-ing model’ (Miniutti & Fabian 2004). However, the relatively largesource height inferred from our disc line modelling, and the lowreflection flux, is inconsistent with this regime. ACKNOWLEDGMENTS
The authors acknowledge financial support from PPARC (PMO)and the Leverhulme trust (KN). We thank Andy Fabian for the rela-tivistic blurring code and Randy Ross for assistance with
REFLION . REFERENCES
Antonucci R. R. J., Miller J. S., 1985, ApJ, 297, 621Ballantyne D. R., Iwasawa K., Fabian A. C., 2001, MNRAS, 323,506Bianchi S., Matt G., Haardt F., Maraschi L., Nicastro F., PerolaG. C., Petrucci P. O., Piro L., 2001, A&A, 376, 77Boller Th., et al., 2002, MNRAS, 329, L1Chen K., Halpern J. P., Filippenko A. V., 1989, ApJ, 339, 742Crummy J., Fabian A. C., Gallo L., Ross R. R., 2006, MNRAS,365, 1067Done C., Madejski G. M., Mushotzky R. F., Turner T. J., KoyamaK., Kunieda H., 1992, ApJ, 400, 138Edelson R., Turner T. J., Pounds K., Vaughan S., Markowitz A.,Marshall H., Dobbie P., Warwick R., 2002, ApJ, 568, 610Fabian A. C., Rees M. J., Stella L., White N. E., 1989, MNRAS,238, 729Fabian A. C., et al., 2002, MNRAS, 335, L1George I. M., Fabian A. C., 1991, MNRAS, 349, 352George I. M., Turner T. J., Netzer H., Nandra K., Mushotzky R. F.,Yaqoob T., 1998, ApJS, 114, 73Ghisellini G., Haardt F., Matt G., 1994, MNRAS, 267, 743Gierlinski M., Done C., 2004, MNRAS, 349, L7—, 2006, MNRAS, 371, L16Gondoin P., Orr A., Lumb D., Santos-Lleo M., 2002, A&A, 388,74Krolik J. H., Kallman T. R., 1987, ApJ, 320, L5Lampton M., Margon B., Bowyer S., 1976, ApJ, 208, 177Laor A., 1991, ApJ, 376, 90Longinotti A. L., Nandra K., Petrucci P. O., O’Neill P. M., 2004,MNRAS, 355, 929 c (cid:13) , 1–6 P.M. O’Neill et al.
Longinotti A. L., Sim S. A., Nandra K., Cappi M., 2006, MNRAS,in pressMagdziarz P., Zdziarski A., 1995, MNRAS, 273, 837Martocchia A., Karas V., Matt G., 2000, MNRAS, 312, 817Miniutti G., Fabian A. C., 2004, MNRAS, 349, 1435Mushotzky R. F., Done C., Pounds K. A., 1993, ARA&A, 31, 717Nandra K., George I. M., Mushotzky R. F., Turner T. J., YaqoobT., 1997, ApJ, 476, 70Nandra K., O’Neill P. M., George I. M., Reeves J. N., Turner T. J.,2007, MNRAS, in preparationPounds K. A., Stanger V. J., Turner T. J., King A. R., Czerny B.,1987, MNRAS, 224, 443Reynolds C. S., 1997, MNRAS, 286, 513Reynolds C. S., Wilms J., Begelman M. C., Staubert R.,Kendziorra E., 2004, MNRAS, 349, 1153Ross R. R., Fabian A. C., 2005, MNRAS, 358, 211Stella L., 1990, Nat, 344, 747Tanaka Y., Boller Th., Gallo L., Keil R., Ueda Y., 2004, PASJ, 56,L9Turner T. J., George I. M., Mushotzky R. F., 1993, ApJ, 412, 72Vaughan S., Edelson R., Warwick R. S., Uttley P., 2003, MNRAS,345, 1271 c (cid:13)000