Long Term Wind-Driven X-Ray Spectral Variability of NGC 1365 with Swift
aa r X i v : . [ a s t r o - ph . H E ] M a r Mon. Not. R. Astron. Soc. , 1–9 (2014) Printed 12 March 2018 (MN L A TEX style file v2.2)
Long Term Wind-Driven X-Ray Spectral Variability ofNGC 1365 with
Swift
S. D. Connolly ⋆ , I.M. McHardy , T. Dwelly , University of Southampton, Highfield, Southampton, SO17 1BJ, UK Max-Planck-Institut f¨ur Extraterrestrische Physik, Giessenbachstrasse 1, 85748, Garching, DE
Accepted 2014 March 17. Received 2014 March 17; in original form 2014 January 06
ABSTRACT
We present long-term (months-years) X-ray spectral variability of the Seyfert 1.8galaxy NGC 1365 as observed by
Swift , which provides well sampled observationsover a much longer timescale (6 years) and a much larger flux range than is affordedby other observatories. At very low luminosities the spectrum is very soft, becomingrapidly harder as the luminosity increases and then, above a particular luminosity,softening again. At a given flux level, the scatter in hardness ratio is not very large,meaning that the spectral shape is largely determined by the luminosity. The spectrawere therefore summed in luminosity bins and fitted with a variety of models. Thebest fitting model consists of two power laws, one unabsorbed and another, moreluminous, which is absorbed. In this model, we find a range of intrinsic 0.5-10.0 keVluminosities of approximately 1 . − . − , and a very large range of absorbingcolumns, of approximately 10 − cm − . Interestingly, we find that the absorbingcolumn decreases with increasing luminosity, but that this result is not due to changesin ionisation. We suggest that these observations might be interpreted in terms of awind model in which the launch radius varies as a function of ionising flux and disctemperature and therefore moves out with increasing accretion rate, i.e. increasingX-ray luminosity. Thus, depending on the inclination angle of the disc relative tothe observer, the absorbing column may decrease as the accretion rate goes up. Theweaker, unabsorbed, component may be a scattered component from the wind. Key words:
X-rays: galaxies galaxies: active galaxies: nuclei - galaxies: individualNGC 1365 galaxies: Seyfert
X-ray spectral observations have shown that variability inthe column of absorbing material between the X-ray sourceand the observer is present in a number of Seyfert galaxies(Risaliti et al. 2002). The detection of variable absorptionon a timescales of hours has indicated that the absorbingmaterial must be close to the nucleus, at a distance similarto that of the Broad Emission Line Region (e.g. Lamer et al.2003; Elvis et al. 2004; Puccetti et al. 2007), with claimsthat complete occultations by Broad Line Region cloudshave been observed on timescales of days (Risaliti 2007A).NGC 1365 is a nearby Seyfert 1.8 galaxy(Maiolino & Rieke 1995) which displays a large amount ofX-ray spectral variability (Risaliti et al. 2009) on timescalesof hours to years (Brenneman et al. 2013). These variationshave been interpreted as the spectrum changing from being ⋆ E-mail: [email protected] ‘transmission dominated’ to ‘reflection dominated’. Whenthe spectrum is ‘transmission dominated’ (e.g. Risaliti et al.2000) the absorbing material is Compton thin and thetransmitted component dominates the spectrum; whenthe spectrum is ‘reflection dominated’ (e.g. Iyomoyo et al.1997) the absorbing material is Compton thick, meaningthe majority of direct emission is absorbed and reflectedemission dominates the spectrum (Risaliti et al. 2007B;Matt et al. 2003).A number of absorption and emission lines have beenseen in the spectrum. A strong Fe fluorescence emission lineis present at 6.4 keV, together with a group of Fe absorptionlines between 6.7 and 8.3 keV, attributed to FeXXV andFeXXVI K α and K β transitions. The measured velocitiesof these lines has lead to speculation that they originatefrom a highly-ionised, high-velocity outflow from NGC 1365(Risaliti et al. 2005A).Although there have been many previous X-ray spectralstudies of NGC 1365, these studies have all concentrated c (cid:13) S.D. Connolly, I.M. McHardy and T. Dwelly either on detailed analysis of a single epoch spectrum or onanalysis of a small number of spectra taken over a relativelyshort timescale (hours or days). By contrast, here we study190
Swift spectra taken over a period of six years. Whilst
Swift does not provide spectral resolution as high as thatused in most previous short-time X-ray spectral studies, e.g.with
XMM-Newton or Suzaku , the
Swift data cover a muchlonger time period and a far greater flux range. The
Swift data therefore allow a proper investigation of flux-relatedspectral variability and of long term spectral variations, overa much larger dynamic range than in previous studies.The spectrum of NGC 1365, as with most AGN,has previously been modelled using a power law com-ponent, with an intrinsic spectral index, Γ. It is notknown whether Γ varies or not during changes in X-ray luminosity. Many groups (e.g. Miller, Turner & Reeves2008; Turner et al. 2007; Fabian et al. 2005; Pounds et al.2004) assume that there is no change. Observations inthe 2 −
10 keV band generally do show some varia-tion, although the changes with luminosity are not large(e.g. Sobolewska & Papadakis 2009; Zdiarski, Lubi & Smith1999). Sobolewska & Papadakis (2009), for example, whosimply fit a power law to the 2-10 keV spectra, find thatthe observed Γ varies as 2.7 ˙m . over similar time scales tothat of our data. In reality, however, these measurements ofΓ are, of course, depend on other parameters which were notincluded in the fits, such as the reflection component and anyabsorption. If the variation in the observed spectral indexis interpreted in terms of the sum of a variable, steep spec-trum component and a relatively constant reflection compo-nent with a hard spectrum, the intrinsic spectral index canremain constant; in this case, when the flux of the variablesteep spectrum component is low, the hard spectrum com-ponent dominates, causing the observed spectral index tochange (e.g. Guainazzi & Antonelli 1999; Uttley et al. 1999;Ponti et al. 2006; Fabian et al. 2003). Furthermore, whereobservations with a large spectral range have been made,allowing good definition of the primary continuum slope,the observed variation of Γ with luminosity has not beenlarge (e.g. 0.1 in NGC 4151, Lubi´nski et al. 2010), 0.2 inNGC 4507, Braito et al. 2012).Theoretical Comptonisation modelling (e.g.Beloborodov 1999; Coppi 1992) shows that the pho-ton index can depend on the ratio of L diss to L s (whereL diss is the power dissipated in the corona during variationsand L s is the input soft photon luminosity) to a low power(-0.1 for AGN). Unless there are very large variationsin these parameters, the intrinsic spectral index shouldtherefore not change by more than a few tenths. Thus,although it is possible that a small change in spectral indexmay occur over the flux range sampled by our observations,the large changes in Γ required by the pivoting power lawmodels are assumed to be unlikely. In many of our models,including the model we deem most accurate, we thereforeassume Γ to be constant, as this is likely to be a reasonableapproximation.The variability of the spectrum of NGC 1365 has previ-ously been modelled using a partial covering model, in whicha varying fraction of the X-ray source is obscured by absorb-ing material (e.g. Risaliti et al. 2009). This model has beenfound to fit the data for individual, short-timescale eventsand is therefore also tested here. Observations of NGC 1365 using the
Swift satellite havebeen carried out as part of a number of different pro-grammes, the combined data from which has been used inthis study. The observations were performed using the
Swift
XRT in ‘photon counting mode’, between 21 July 2006 and17 March 2013. A total of 293 spectra from individual
Swift ‘visits’, or exposures, were used, with a total of more than220 kiloseconds of exposure time. Individual exposure timesranged from <
10 seconds to > ∼
165 kiloseconds of expo-sure time and over 25000 photon counts. The raw data forall
Swift XRT observations of NGC 1365 were downloadedfrom the HEASARC archive .The XRT data were reduced using an automaticpipeline, fully described in Fabian et al. (2012) and previ-ously used in e.g. Cameron et al. (2012). In each case, thereduction used the most recent version of the standard Swift
XRTPIPELINE software (versions 0.12.4 - 0.12.6). The XS-ELECT tool was used to extract spectra and lightcurves,using flux-dependent source and background extraction re-gions which were chosen such as to reduce the backgroundcontamination at faint fluxes, and to mitigate the effects ofpile-up at high fluxes. The sensitivity of the XRT is not uni-form over the field of view, due to vignetting and the pres-ence of bad pixels and columns on the CCD; the
Swift
XR-TEXPOMAP and XRTMKARF tools were therefore usedto generate an exposure map (including vignetting and badpixels) and an ancillary response file (ARF) for each visit,in order to correct for these effects. The relevant redistribu-tion matrix file (RMF) from the
Swift calibration databasewas also supplied in each case. The local X-ray backgroundwas estimated and subtracted from the instrumental countrates, using the area-scaled count rate measured in a back-ground annulus region. The observed XRT count rates werecarefully corrected for the fraction of counts lost due to badpixels and columns, vignetting effects, and the finite extrac-tion aperture (including regions excised in order to mitigatepileup effects).Fig. 1 shows a 0 . − . Swift data over the six-year period of observation. The largeflux range is readily apparent on a range of timescales.
Chandra images show that the nucleus of NGC 1365 isembedded within a region of extended low surface brightnessemission, of radius approximately 15 arcsec (see Wang et al.2009). The region used to determine the
Swift backgroundcontribution lies outside this small region of extended emis-sion. Thus, a contribution from the constant extended emis-sion within the
Swift
PSF will remain in our
Swift spectra.To determine what that contribution is, we examined
Chan-dra spectra of the extended emission region. The
Chandra http://heasarc.gsfc.nasa.gov/cgi-bin/W3Browse/swift.plc (cid:13) , 1–9 ind-Driven X-Ray Spectral Variability of NGC 1365 . . . . . . . . . . . . . . . . . . MJD - 50,000 C oun t R a t e ( c oun t ss − ) Figure 1.
The
Swift
X-ray lightcurve of NGC 1365, with a broken axis where data were not taken. The section on the right shows theperiod during which more intensive
Swift monitoring was taking place as a result of SN2012fr.OBSIDs MJD range N obs N visits T tot (s) Cnts tot Table 1.
Summary of
Swift observations used in this work. N obs ,N visits and T tot are the values remaining after unusable data hasbeen excluded. data used were taken in December 2002, using the ACIS-S instrument (OBSID 3554). Four spectra were extractedfrom circular regions, of size similar to the Swift
PSF, lyingclose to the nucleus. The spectra were extracted from thethe event files in the primary (reduced) data set, using theCIAO 4.6 tool ‘specextract’. Extraction regions of radius 2arcsecs were used in each case.The background spectra were taken from within thesame region from which the
Swift background spectra weretaken, such that the resultant spectra did not contain thebackground which had been subtracted from the
Swift spec-tra. Circular extraction regions of radius 2 arcsecs were alsoused.
Plots of the hardness ratio against the hard count rate, andthe hard count rate against the soft count rate are shown inFig. 2. Whereas most previous measurements of the photonindex, Γ, concentrate on the 2 . − . . − . . − . . − . − Soft Cnt RateHard Cnt Rate + Soft Cnt Rate (1) Fig. 2 shows the spectrum to be extremely soft atvery low fluxes, but to become hard very rapidly withwith increasing flux. Beyond this sharp rise, still at a rel-atively low flux, the hardness decreases again more grad-ually with increasing flux, as often seen in Seyfert galaxieswithin the 2 . − . The 190 separate time-resolved XRT spectra of NGC 1365were divided into 11 flux bins and combined using the
HEADAS tool ‘addspec’ (see Fig. 3 for a sample of thesesummed spectra). The bins were chosen such that eachbinned spectrum had both a minimum of 2000 total countsand a minimum width of 0.025 counts s − across the totalrange of hard count rates. This binning method ensured thateach summed spectrum would possess a sufficient signal tonoise ratio for accurate spectral fitting, and that the fluxbins were roughly evenly spaced across the hard flux rangecovered by the spectra. The energy channels of each of thesesummed spectra were then grouped, using the HEADAS tool‘grppha’, such that each group contained a minimum of 15counts.A variety of models were fitted to the spectra, using the
XSPEC
Chandra spectrawere fitted. These spectra were composed of a minimum of1300 counts. The spectral energy channels were grouped in c (cid:13) , 1–9 S.D. Connolly, I.M. McHardy and T. Dwelly
Model Xspec description Spectralindex Absorbingcolumn Ionisation Γ χ Red
DoF1 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Fixed Free Free 1.92 (fixed) 1.39 13292 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Fixed Free Tied 1.92 (fixed) 1.42 13393 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Fixed Free 0 1.92 (fixed) 1.44 13404 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Fixed Tied Free 1.92 (fixed) 2.28 13395 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Tied Free Free 1.47 (tied) 1.21 13286 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Free Free Free 1.22 - 2.09 1.17 13187 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Free Free Tied 1.19 - 2.09 1.20 13288 Absorbed & unabsorbed power laws powerlaw + absori × powerlaw Free Tied Free 0.30 - 1.78 1.86 13289 Single, absorbed power law absori × powerlaw Free Free Free 0.61 -1.81 1.36 132810 Single, absorbed power law absori × powerlaw Free Free Tied 0.58 -1.82 1.63 132811 Single, absorbed power law absori × powerlaw Free Tied Free 0 -1.91 1.42 131812 Compton scattering (powerlaw+absori × powerlaw) × cabs Fixed Free Tied 1.92 (fixed) 1.42 133913 Partial covering, fraction tied pcfabs × powerlaw Fixed Tied n/a 1.92 (fixed) 2.57 135114 Partial covering, fraction tied pcfabs × powerlaw Fixed Free n/a 1.92 (fixed) 1.45 135115 Partial covering, fraction free pcfabs × powerlaw Fixed Free n/a 1.92 (fixed) 1.45 1341 Table 2.
Summary of the main components of each model fitted to the average spectra, showing the parameters which were fixed, tiedor left free in each case, and the value or range of values for the spectral index, Γ, which was fixed or best fitting. The reduced χ valueand number of degrees of freedom (DoF) of the best fit with each model is also shown. Each model also contains the fixed componentsof the best fit model to the diffuse gas around the nucleus (tbabs ∗ apec) and the entire model is multiplied by the wabs model set to thegalactic absorbing column.Model Γ N H n A n U ξ T F1 1 .
92 9 . +1 . − . . +8 . − . . +0 . − . . +5 . − . . +0 . †− . − .
92 6 . +0 . − . . +9 . − . . +0 . − . . +0 . − . . +4 . − . − .
92 5 . +0 . − . . +8 . − . . +0 . − . − −
4* 1 .
92 5 .
44 84 . .
63 0 .
017 10 . − . +0 . − . . +1 . − . . +4 . − . . +0 . − . . +12 . − . . +0 . †− . − . +0 . − . . +2 . − . . +10 . − . . +0 . − . . +41 . − . . +0 . †− . − . +0 . − . . +0 . − . . +9 . − . . +0 . − . . +0 . − . . +6 . − . − . +0 . − . . +0 . − . . +7 . − . . +0 . − . . +65 . − . . +7 . − . − . +0 . − . . +0 . − . . +0 . − . − . +55 . − . . +0 . − . −
10 0 . +0 . − . . +0 . − . . +1 . − . − . +23 . − . . +0 . − . −
11 0 . +0 . − . . +0 . − . . +0 . − . − . +334 . − . . +0 . − . −
12 1 .
92 6 . +0 . − . . +8 . − . . +0 . − . . +0 . − . . +9 . − . −
13* 1 .
92 4 .
51 73 . − − − . .
92 7 . +0 . − . . +9 . − . − − − . +0 . − .
15 1 .
92 7 . +0 . − . . +5 . − . − − − . +0 . − . Table 3.
Typical parameter values for each model described in Table 2, taken from the fit to the spectrum of the central flux-bin. 90%errors are given where a parameter was free or tied. Errors for models with an asterisk (*) are not given, as the reduced χ of the fitwas >
2, preventing their calculation in
XSPEC . As the absorber temperature in models 5 and 6 reached the upper limit of the absori model, the positive errors on these values (indicated with † ) are given as zero. These values are, however, unconstrained in the positivedirection. The parameters are as follows: Γ - spectral index, N H - absorbing column (10 cm − ), n A - normalisation of the absorbedpower law (10 − keV − cm − s − at 1 keV), n U - normalisation of the unabsorbed power law (10 − keV − cm − s − at 1 keV), ξ -ionisation state (L − /N e R - see Done et al. 1992), T - absorber temperature (10 K), F - covering fraction. the same way as the
Swift spectra, such that each groupcontained a minimum of 15 counts.As found previously by Wang et al. (2009), the best fit-ting model is the apec model for collisionally-ionised diffusegas combined with the tbabs model for absorption by gasand dust. These models were used together with the wabs model, to account for the known galactic absorption alongthe line of sight (1 . × cm − ). A simultaneous fit to 4regions, allowing N H to vary, but with all other parameterstied, gave a reduced χ of 0.96, with 197 degrees of free-dom. The absorbing column varied a little between regions.However, as a single value is required to fit the average con-tribution to the Swift nuclear spectrum, we fitted all regions with the same N H , giving a reduced χ of 1.46, with 200 de-grees of freedom. In this model, the best fitting parameterswere: an absorbing column of 6 . − . . × cm − in ad-dition to galactic absorption, a gas temperature of 0 . − . . keV, a metal abundance of 0 . − . . and a normalisationof the apec model of 5 . − . . × − . The components ofthis model, with each of the parameters fixed at the bestfit values, were included in all of the subsequent fits of the Swift data.When modelling the resultant XRT background-subtracted nucleus, it was discovered that a single, unab-sorbed power law of photon index ∼ .
92 (as found byRisaliti et al. 2013), fitted the lowest flux observations very c (cid:13) , 1–9 ind-Driven X-Ray Spectral Variability of NGC 1365 . . . . . Hard (2.0-10.0 keV) Count Rate ( s − ) − . − . . . . H a r dne ss R a t i o ( H - S / H + S ) . . . . . Hard (2.0-10.0 keV) Count Rate ( s − ) . . . . . . S o ft ( . - . k e V ) C oun t R a t e ( s − ) Figure 2.
Top:
Plot of the hard count rate against hardness ratioof NGC 1365. The scatter around the mean at a given count rateis low.
Bottom:
Plot of the hard count rate against soft count rateof NGC 1365. In both plots, the data are binned such that eachbin contains a minimum of 5 data points. Errors in the hard countrate are the standard deviations of the distribution of points ineach bin. The unbinned data are shown lightly behind the binneddata. The typical fractional errors on the unbinned data are 0.16,0.22 and 0.38 for the soft counts, hard counts and hardness ratiorespectively. well, except for a small excess at higher energies ( > ∼ . NuStar . It was possible to obtain equally goodfits at all fluxes using a single absorbed pivoting power lawand better fits using two pivoting power laws of the samespectral index (see models 6-11 in Table 2). However, thebest fitting models of this kind required a very large range ofΓ (from ∼ . > ∼ . absori model for an ionised photoelectric ab-sorber. The absorber temperature was tied in all fits. Aredshift of 5 . × − (Lavaux & Hudson 2011) and aniron abundance of 2.8 times solar abundance, as found byRisaliti et al. (2009), were used in all fits. Galactic ab-sorption of 1 . × cm − was also included in eachmodel, using the wabs model for electromagnetic absorption(Dickey & Lockman 1990).The effects of Compton Scattering on the fits weretested by adding the cabs model in XSPEC (see model 12in Table 2). The inclusion of this model was found to havevery little effect on the best fit parameters, or the goodnessof fit. This result is not unexpected, as the highest measuredabsorbing column is 1 . × cm − , which is below the col-umn required to be Compton thick (1 . × cm − , seeMalizia et al. 2009).In model 3 of Table 2 the absorbing column is neutral,i.e. the ionisation state is fixed at zero. The best fit in thismodel is almost the same as those in which the ionisationis free (models 1 and 2). Thus, although the value of theabsorbing column has a large effect on our fits, the ionisationstate is not well constrained.Finally, a set of partial covering models were fitted tothe data, as this type of model has previously been found tobe successful in modelling changes over short timescales (e.g.Risaliti et al. 2009) (see models 13-15 in Table 2). The model( pcfabs in XSPEC ) consists of a neutral absorber which cov-ers a fraction of the X-ray source, resulting in a spectrumcomposed of an absorbed and an unabsorbed component.We found that the data could not be well fit by a modelin which the absorbing column is constant between spectralfits, with only the covering fraction changing (model 13). Ifwe allow the absorbing column to vary, we obtain the samequality of fit whether we allow the covering fraction to vary(model 14) or not (model 15), as, when the fraction is al-lowed to vary, the same value is derived at each flux level.Model 14 is physically very similar to our two componentmodel (model 2). c (cid:13) , 1–9 S.D. Connolly, I.M. McHardy and T. Dwelly . . Energy (keV) − − − − F l u x ( P ho t on sc m − s − k e V − ) Figure 3.
A sample of the set of unfolded, summed spectra pro-duced by combining spectra in the same flux range. Alternatespectra are excluded, to prevent the plot being crowded. Thespectra are simultaneously fitted with the best fitting model, con-sisting of two power laws, one of which is absorbed by a partiallyionised absorbing column and one of which is not. Both powerlaws have a fixed spectral index of 1.92, as found by Risaliti et al.(2013). The ionisation state of the absorbing material is tied, butthe absorbing column is allowed to vary between individual fits,producing the variation seen between spectra from different fluxlevels. The flux values are calculated by
XSPEC using the model.The data are binned for clarity. . . Energy (keV) − − − − F l u x ( P ho t on sc m − s − k e V − ) Figure 4.
The sample of best fitting models shown in Fig. 3( solid lines ), together with the components of each model - theunabsorbed power law and absorbed power law ( dashed lines ),and the spectrum of the underlying diffuse emission ( dotted line ). In all fits described here, we fix the underlying power lawspectral index at the value given by Risaliti et al. (2013).Allowing Γ to vary does give better fits but, as discussedabove, the required range of Γ is extremely large and almostcertainly unphysical.In the two-component models with a fixed Γ, leaving theionisation state to vary, but tying the absorbing column, wasfound to be insufficient to account for the degree of variation observed in the spectra, at any fixed absorbing column, forboth a pivoting power law and the two-component model,as can be seen from the χ values.Tying the ionisation state whilst leaving the absorbingcolumn free to vary, however, gives good fits to the data. Sig-nificantly, the χ value is very similar to that of the modelin which both the ionisation state and absorbing column areboth left free to vary, as the ionisation varies little in the bestfit model. Whilst the ionisation state of the absorbing ma-terial undoubtedly changes, the data show both that largechanges are not required to give a good fit, and that largechanges in the absorbing column are essential to accountfor the spectral variation observed regardless. Changes inionisation alone cannot account for these spectral changesobserved. The two-component model with one absorbed andone unabsorbed power law, in which the absorbing column isleft free to vary between spectra whilst the ionisation stateis kept constant, is therefore the simplest model with whichthe spectral variability seen in the data can be described.In this model, there was a large range of absorbingcolumns, from 1 . × to 10 cm − . The temperatureof the absorber was 9 . × K and the ionisation state ofthe absorber was 0.11. The normalisations of the two powerlaws varied by a factor of 3, giving a range of intrinsic lumi-nosities of approximately 1.1 to 3.5 ergs s − for the 0.5-10.0keV band.Fig. 3 shows a sample of the flux-binned spectra fit-ted with this model. Fig. 4 shows how the two power lawcomponents vary as the flux changes. The plot shows thatthe absorbing column of the absorbed power law decreasesas the normalisation of the power law (i.e. the flux beforeabsorption) increases. These two parameters are plotted inFig. 5. There is a strong decrease in the absorbing columnas the normalisation increases (Spearman ranked correlationcoefficient ρ = 0 . +0 . − . ), but the data are not of sufficientquality to precisely determ ne the form of the relationshipbetween these parameters.In Fig. 6 the normalisation of the unabsorbed power lawis plotted against that of the absorbed power law. The nor-malisations can be fitted well with a linear model, showingthem to be correlated (r = 0.91). These observations con-firm that there is a real change in the underlying luminosityof the source and that the observed flux changes are notjust due to changes in absorption. The 15 −
150 keV
BAT lightcurve of NGC 1365, which undergoes far less absorptionthan the 0.5-10 keV band, confirms this intrinsic variability,showing variation of approximately a factor of 4.As a consistency check, the hard count rate and hard-ness ratio from each model fitted to the summed data areplotted over the original data as in Fig. 2 (Fig. 7). Theseplots show that the summed spectra also follow the trendsshown by the individual spectra, reinforcing the assumptionthat the spectrum is similar at a given flux level, indepen-dent of time.A two component spectral model consisting of an unab-sorbed scattered component and a more luminous absorbedcomponent has previously been used to describe the singleepoch spectrum of NGC 4945 (Done et al. 1996). Broadlysimilar two-component spectra are also reported elsewhere(e.g. NGC4507, Braito et al. 2012). Variations of hardnessratio versus count rate similar to those shown here, in Fig.2, have also been seen in the spectra of X-ray binary systems c (cid:13) , 1–9 ind-Driven X-Ray Spectral Variability of NGC 1365 keV − cm − s − at keV )24681020406080100 A b s o r b i ng C o l u m n ( N H c m − ) Figure 5.
Log plot of the normalisation parameter of the ab-sorbed power law against the column of the absorbing materialin the model described above (model 2 in Table 2). The normal-isation of the power law is equivalent to the flux of this compo-nent before absorption, and therefore the X-ray luminosity of thesource. keV − cm − s − at keV ) 1e-20123456 N o r m a li s a t i ono f U nb s o r bed P o w e r La w ( k e V − c m − s − a t k e V ) Figure 6.
Plot of the normalisation parameter of the absorbedpower law against the normalisation parameter of the unabsorbedpower law in the model described above (model 2 in Table 2).These normalisations are equivalent to the unabsorbed flux fromeach component; the correlation between them implies the twocomponents originate from the same source. The data are fittedwith a linear model going through the origin, with a reduced χ of 1.50 for 10 degrees of freedom. (Fig. 1 in Kuulkers et al. 1998) from RXTE data (2-13 keV).Although Kuulkers et al. (1998) do not discuss flux-binnedspectra as seen in our Fig. 3, they also favour a similar twocomponent model. These studies show that a two componentmodel broadly similar to that which we describe here can,not infrequently, be used to parameterise the X-ray spectraof both AGN and X-ray binary sources. However, the maincontribution of this paper is to show, for the first time, thatthe complete, and very large, range of spectral variability . . . . . . Hard (2.0-10.0 keV) Count Rate ( cm − s − ) − . − . . . . H a r dne ss ( H - S / H + S ) Figure 7.
Plot of the hardness against the hard count rate ofeach of the 190 observed spectra ( small circles ), together withthat of the model spectra described above (model 2 in Table 2)when fitted to each of the 11 summed, flux binned spectra ( largecircles ). displayed by at least one AGN can be explained by system-atic variation of the absorber, with the absorption varyinginversely with luminosity. Below we discuss a possible modelto explain this behaviour. For the best-fitting two-component model described above(model 2 in Table 2), Fig. 5 shows the column of absorb-ing material to be inversely related to the normalisation pa-rameter of the absorbed power law, i.e. to the flux prior toabsorption. As this component dominates the unabsorbedluminosity of the source, the absorbing column is inverselyproportional to the source luminosity. Whilst one might ini-tially assume any reduction in absorption with increasingflux to be due to increased ionisation, we have shown thatmodels involving varying ionisation alone do not fit the data;models allowing variation of ionisation and absorbing col-umn simultaneously also require an inverse relationship be-tween the absorbing column and the unabsorbed luminosityand do not require the ionisation to vary significantly. Fitsto spectra of NGC 4151 by Lubi´nski et al. (2010) at differentflux levels show a similar reduction in the absorbing columnwith increasing flux, implying that this relationship is notunique to NGC 1365. A physical mechanism is required toexplain this relationship and a possible solution lies in anX-ray wind of absorbing material rising from the accretiondisc.
X-ray absorption is often attributed to outflows, in particu-lar to a disc wind (see e.g. Kaastra et al. 2000, Blustin et al.2005, Tombesi et al. 2013). In the AGN model proposed by c (cid:13) , 1–9 S.D. Connolly, I.M. McHardy and T. Dwelly
Elvis (2000), absorbing material arises from a narrow rangeof accretion disc radii in a biconical ‘wind’. Nicastro (2000)shows that an X-ray absorbing wind could originate froma narrow boundary region between the radiation pressure-dominated and gas pressure-dominated regions of the accre-tion disc. In these models, a rise in accretion rate, whichwill give rise to an increase in X-ray luminosity, naturallyleads to an increase in the radii from which the wind arisesthrough at least two mechanisms. Firstly, a higher accretionrate leads to an increase in disc temperature and hence anincrease in the radii from which the wind arises. Secondly,a higher accretion rate leads to an increased ionising UVflux from the inner disc. When potential wind material atthe disc surface is fully ionised, it is only subject to Comp-ton scattering; the much more powerful line-driving force(Proga, Stone & Drew 1999) is no longer applicable, hencethis material would never be driven off as a wind. Thus, if theinclination is such that the observer views the X-ray sourcethrough the inner edge of the wind, an increase in accretionrate would move the wind outwards such that the observerwould now be viewing the source through a lower absorbingcolumn. An additional geometric factor which would reduceabsorbing column is that, in the Elvis (2000) model, an in-crease in X-ray luminosity causes the opening angle of thewind to increase due to the increased radiation pressure.The above affects will thus naturally lead to the inverse re-lationship between absorbing column and luminosity whichwe see. Moving the absorbing wind to larger radii also meansthat the line of sight would go through the less dense, morehighly ionised front part of the wind, as we see in our fits.The low ionisation states found in our models suggestthat the material involved is not associated with the highly-ionised, high-velocity outflows suggested by Risaliti et al.(2005A), which is too ionised to cause the observed X-rayabsorption. We therefore suggest that the absorbing mate-rial lies in a region which is further out than this highlyionised material, as part of a stratified wind (Tombesi et al.2013, Elvis 2000).Depending on which mechanism dominates in pushingthe wind to larger radii with increasing accretion rate, we ex-pect different lags between the change in absorbing columnand the change in luminosity which could lead to hysteresisand scatter in the X-ray luminosity / hardness relationships.If the outward movement of the wind launch radii iscaused by a change in disc temperature due to inwardlypropagating accretion rate fluctuations, then the X-ray lu-minosity will lag by the viscous travel time of those fluctu-ations to either the inner edge of the disc where the seedphotons are mainly produced and/or to the X-ray emittingcorona itself. For typical wind launch radii of a few hundredR g (Higginbottom et al. 2013), this timescale would be oforder weeks to months.If the outward movement of the wind is dominated byan increase in ionising UV photons from the inner edge ofthe disc, which also dominate the X-ray seed photon flux,then variations in the X-ray luminosity will lead changesin the absorbing column slightly, by the difference betweenthe light travel time from the UV region to the wind andto the corona respectively. This difference is likely to besmall (hours). Alternatively, if the wind moves outwards inresponse to increased UV flux, but the X-ray luminosity risesin response to increased accretion rate rather than increased seed photon flux, then the X-ray luminosity will lag by thedifference between the viscous propagation time from theUV to X-ray emitting regions and the light travel time fromthe UV emitting to wind launch radii. This difference is alsolikely to be small (1-2 days).Our observations do not show a great deal of scatter inthe hardness-count rate diagram, indicating that any lagsare short. We therefore favour a mechanism by which thewind launch radii are pushed outwards mainly by a rise inionising UV flux, rather than by a rise in local disc temper-ature. Regardless of mechanism by which these outflows aredriven, it is clear from our data that a geometrical responseto changes in luminosity is necessary in the absorber in orderto explain the variability in the spectra.Finally, we note that, in all models, the wind is ex-pected to be clumpy, meaning short timescale variations ofthe absorbing column, independent of the unabsorbed lu-minosity, should also occur, which would also add scatterto the relationship between X-ray luminosity and absorb-ing column. The spectral variations reported previously onshort timescales are probably mainly the result of fittingvariations induced as clumps pass over the line of sight. Swift
X-ray observations of NGC 1365 over a period of 6years show a large amount of spectral variability. These vari-ations are best explained by a two-component model con-sisting of an unabsorbed power law and a more luminousabsorbed power law; for both components, the spectral in-dex was fixed. The normalisations of the two power lawsvary together. The absorbing column of the absorbed powerlaw varies inversely with its luminosity, an effect which isnot simply due to increased ionisation. This effect can besimply explained by viewing through the edge of a windwhose launch radius varies inversely with increasing accre-tion rate. The unabsorbed power law could be explained, asin the standard Elvis (2000) wind model, as the scatteredcomponent from the far side of the wind.
ACKNOWLEDGMENTS
SDC thanks the STFC for support under a studentshipand IMcH thanks the STFC for support via grantST/G003084/1. We thank Christian Knigge and JamesMatthews for useful discussions.
REFERENCES
Arnaud, K.A., 1996, Astronomical Data Analysis Softwareand Systems V, 101, 17Beloborodov, A., 1999, ASPC, 161, 295.Blustin, A. J., Page, M. J., Fuerst, S. V., Branduardi-Raymont, G., Ashton, C. E., 2005, A&A, 431, 111Braito, V., Ballo, L., Reeves, J. N., Risaliti, G., Ptak, A.,Turner, T. J., 2012, MNRAS, 428, 2516Brenneman, L.W., Risaliti, G., Elvis, M., Nardini, E., 2013MNRAS, 429, 2662Cameron, D. T., McHardy, I., Dwelly, T., Breedt, E., Utt-ley, P., Lira, P., Arevalo, P., 2012, MNRAS, 422, 902 c (cid:13) , 1–9 ind-Driven X-Ray Spectral Variability of NGC 1365 Childress, M. J., Scalzo, R. A., Sim, S. A., Tucker, B. E.,Yuan, F., Schmidt, B. P., Cenko, S. B., Silverman, J. M.,Contreras, C., Hsiao, E. Y., Phillips, M., Morrell, N., Jha,S. W., McCully, C., Filippenko, A. V., Anderson, J. P.,Benetti, S., Bufano, F., de Jaeger, T., Forster, F., Gal-Yam, A., Le Guillou, L., Maguire, K., Maund, J., Mazzali,P. A., Pignata, G., Smartt, S., Spyromilio, J., Sullivan,M., Taddia, F., Valenti, S., Bayliss, D. D. R., Bessell, M.,Blanc, G. A., Carson, D. J., Clubb, K. I., de Burgh-Day,C., Desjardins, T. D., Fang, J. J., Fox, O. D., Gates, E.L., Ho, I.-T., Keller, S., Kelly, P. L., Lidman, C., Loar-ing, N. S., Mould, J. R., Owers, M., Ozbilgen, S., Pei, L.,Pickering, T., Pracy, M. B., Rich, J. A., Schaefer, B. E.,Scott, N., Stritzinger, M., Vogt, F. P. A., Zhou, G., 2013,ApJ, 770, 29Coppi, S., 1992, MNRAS, 258, 657Dickey, J. M., Lockman, F. J., 1990, ARAA, 28, 215Done, C., Mulchaey, J.S., Mushotzky, R.F., Arnaud, K.A.,1992, ApJ, 395, 275Done, C., Madejski, G.M., Smith, D.A., 1996, ApJ, 463, 63Elvis, M., 2000, ApJ, 545, 63Elvis, M., Risaliti, G., Nicastro, F., Miller, J. M., Fiore, F.,Puccetti, S., 2004, ApJ, 615, 25Fabian, A.C., Vaughan, S., 2003, MNRAS, 340, 28.Fabian, A.C., Miniutti, G., Iwasawa, K., Ross, R. R., 2005,MNRAS, 361, 795Fabian, A.C., Zoghbi, A., Ross, R. R., Uttley, P., Gallo, L.C., Brandt, W. N., Blustin, A. J., Boller, T., Caballero-Garcia, M. D., Larsson, J., Miller, J. M., Miniutti, G.,Ponti, G., Reis, R. C., Reynolds, C. S., Tanaka, Y., Young,A. J., 2009, Nature, 459, 540Fabian, A. C., Zoghbi, A., Wilkins, D., Dwelly, T., Utt-ley, P., Schartel, N., Miniutti, G., Gallo, L., Grupe, D.,Komossa, S., Santos-Lle, M., 2012, MNRAS, 419, 116Guainazzi, M., Antonelli, L. A., 1999, MNRAS, 304, 15Higginbottom, N., Knigge, C., Long, K. S., Sim, S.A., Matthews, J. H., 2013, MNRAS, advanced access:arXiv:1308.5973Iyomoyo, N., Makishima, K, Fukazawa, Y., Tashiro, M.Ishisaki, Y., 1997, PASJ, 49, 425Kaastra, J. S., Mewe, R., Liedahl, D. A., Komossa, S.,Brinkman, A. C., 2000, A&A, 354L, 83Klotz, A., Normand, J., Conseil, E., 2012, CBET, 3276, 1Kulkers, E., Wijnands, R., Belloni, T., Mendez, M., vanDer Klis, M., van Paradijs, J., ApJ, 494, 753Lamer, G., McHardy, I.M., Uttley, P., Jahoda, K., 2003,MNRAS, 338, 323Lavaux, G., Hudson, M.J., 2011, MNRAS, 416, 2840Lubi´nski, P., Zdziarski, A. A., Walter, R., Paltani, S., Beck-mann, V., Soldi, S., Ferrigno, C., Courvoisier, T. J.-L.,2010, MNRAS, 408, 1851Maiolino, R., Rieke, G. H., 1995, ApJ, 454, 95Malizia, A., Stephen, J. B., Bassani, L., Bird, A. J.,Panessa, F., Ubertini, P., 2009, MNRAS, 399, 944.Matt, G., Guainazzi, M., Maiolino, R., 2003, MNRAS, 342,422Markowitz, A., Reeves, J. N., Miniutti, G., Serlemitsos,P., Kunieda, H., Yaqoob, T., Fabian, A. C., Fukazawa,Y., Mushotzky, R., Okajima, T., Gallo, L. C., Awaki, H.,Griffiths, R. E., 2008, PASJ, 60, 277Miller, L., Turner, T., Reeves, J., 2008, A&A, 483, 437Nicastro, F., 2000, ApJ, 530, 65 Papadakis, I. E., Sobolewska, M., Arevalo, P., Markowitz,A., McHardy, I. M., Miller, L., Reeves, J. N., Turner, T.J.,2009, A&A, 494, 905Ponti, G., Miniutti, G., Cappi, M., Maraschi, L., Fabian,A. C., Iwasawa, K., 2006, MNRAS, 368, 903Pounds, K., Reeves, J., Page, K., O’Brien, P.T., 2004, ApJ,616, 696Proga, D., Stone, J. M., Drew, J. E., 1999, MNRAS, 310,476Puccetti, S., Fiore, F., Risaliti, G., Capalbi, M., Elvis, M.,Nicastro, F., 2007, MNRAS, 377, 607Risaliti, G., 2007,ASPC, 373, 458Risaliti, G., Maiolino, R., Bassani, L., A&A, 365, 33Risaliti, G., Elvis, M., Nicastro, F., 2002, ApJ, 571, 234Risaliti, G.,Bianchi, S., Matt, G., Baldi, A., Elvis, M., Fab-biano, G., Zezas, A., 2005, ApJ,630, 129Risaliti, G., Elvis, M., Fabbiano, G., Baldi, A., Zezas, A.,2005, ApJ, 623, 93Risaliti, G., Elvis, M., Fabbiano, Baldi, A.,Zezas, A., Sal-vati, M., 2007, ApJ, 659, 111Risaliti, G., Miniutti, G., Evlis, M., Fabbiano, G., Salvati,M., Baldi, A., Braito, V., Bianchi, S., Matt, G., Reeves,J., Soria, R., Zezas, A., 2009, ApJ,696,160Risaliti, G., Harrison, F.A., Madsen, K.K., Walton, D.J.,Boggs, S.E., Christensen, F.E., Craig, W.W., Grefen-stette, B.W., Hailey, C.J., Nardini, E., Stern, D., Zhang,W.W., 2013, Nature, 494, 449Shakura, N. I.,Sunyaev, R. A., A&A, 24, 337Sobolewska, M. A., Papadakis, I. E., 2009, MNRAS, 399,1597Tombesi, F., Cappi, M., Reeves, J., Nemmen, R.S., Braito,V., Gaspari, M., Reynolds, C. S., 2013, MNRAS, 430, 1102Turner, T., Miller, L., Reeves, J. N., Kraemer, S. B., 2007,A&A, 475, 121Treves, A., Maraschi, L., Abramowicz, M., 1988, PASP,100, 427Uttley, P., McHardy, I. M., Papadakis, I. E., Guainazzi, M.,Fruscione, A., 1999, MNRAS, 307, 6Wang, J., Fabbiano, G., Evlis, M., Risaliti, G., 2009, ApJ,294, 718Zdziarski, A. A., Lubi, P., Smith, D. A., MNRAS, 303, 11 c (cid:13)000