X-ray spectral evolution in M33 X-8
aa r X i v : . [ a s t r o - ph . C O ] J un Mon. Not. R. Astron. Soc. , 1–8 (2009) Printed 17 July 2018 (MN L A TEX style file v2.2)
X-ray spectral evolution in the ultraluminous X-ray sourceM33 X-8
Matthew J. Middleton , Andrew D. Sutton & Timothy P. Roberts Department of Physics, University of Durham, South Road, Durham DH1 3LE, UK
17 July 2018
ABSTRACT
The bright ultraluminous X-ray source (ULX), M33 X-8, has been observed severaltimes by
XMM-Newton , providing us with a rare opportunity to ‘flux bin’ the spec-tral data and search for changes in the average X-ray spectrum with flux level. Theaggregated X-ray spectra appear unlike standard sub-Eddington accretion state spec-tra which, alongside the lack of discernible variability at any energy, argues stronglyagainst conventional two-component, sub-Eddington models. Although the lack ofvariability could be consistent with disc-dominated spectra, sub-Eddington disc mod-els are not sufficiently broad to explain the observed spectra. Fits with a ∼ Eddingtonaccretion rate slim disc model are acceptable, but the fits show that the temperaturedecreases with flux, contrary to expectations, and this is accompanied by the appear-ance of a harder tail to the spectrum. Applying a suitable two-component model revealsthat the disc becomes cooler and less advection dominated as the X-ray flux increases,and this is allied to the emergence of an optically-thick Comptonisation medium. Wepresent a scenario in which this is explained by the onset of a radiatively-driven windfrom the innermost regions of the accretion disc, as M33 X-8 exceeds the Eddingtonlimit. Furthermore, we argue that the direct evolution of this spectrum with increas-ing luminosity (and hence radiation pressure) leads naturally to the two-componentspectra seen in more luminous ULXs.
Key words: accretion, accretion discs – X-rays: binaries – X-rays: individual: M33X-8
Ultra-luminous X-ray sources (ULXs) are point-like objectswith high ( > erg s − ) X-ray luminosities that are not as-sociated with an active galactic nucleus (AGN) or, indeed,the central regions of a host galaxy (see Miller & Colbert2004; Roberts 2007; Gladstone 2011). The nature of theseobjects has been the subject of much speculation, with CCDresolution X-ray spectroscopy playing a major role in ad-vancing our understanding. Although other missions haveplayed an important part (e.g. ASCA detection of possiblestate transitions in ULXs, Kubota et al. 2001), these resultshave predominantly come from the
XMM-Newton mission.Its first major advance was the detection of a soft excess inthe spectra of many ULXs, with a temperature consistentwith that expected for the inner edge of an accretion discaround an intermediate-mass black hole (IMBH; e.g. Milleret al. 2003; Miller, Fabian & Miller 2004). However, laterstudies showed that the second, harder component in thesespectra turns over within the
XMM-Newton bandpass, andso appears much cooler and optically thicker than the cor- responding Comptonisation media in Galactic black holes(Stobbart et al. 2006). This is inconsistent with the identi-fication of a sub-Eddington state for an IMBH, and moreindicative of super-Eddington accretion onto a stellar-massblack hole (Gladstone, Done & Roberts 2009). The appar-ent divergence of the spectra of more luminous ULXs intotwo components (see Fig. 8 of Gladstone et al. 2009) canbe interpreted in terms of the emergence of a radiatively-driven wind at super-Eddington accretion rates, with theoutflowing material thermalising the underlying disc emis-sion to produce the soft spectral component as predicted bye.g. King (2004), Poutanen et al. (2007). The hard compo-nent is then produced within the photospheric radius, withits characteristic optically-thick Comptonisation signatureeither the result of a thick shroud of Comptonising elec-trons around the hot inner disc, or perhaps a change in theopacity of the outer layers of the hot inner accretion discitself (Middleton et al. 2011). Such a model can explain thestartling lack of variability seen in many of these sources(Heil et al. 2009). In those few cases where large amounts of c (cid:13) M. J. Middleton, A. D. Sutton & T. P. Roberts
Table 1.
XMM-Newton observations of M33 X-8.OBSID obs. date useful exposure off-axis angle f x net count rate total counts bin(ks, MOS1) (arcmin) ( × − erg cm − s − ) (ct s − , MOS1)0102640101 2000-08-04 8.2 1.2 1 . → .
84 1.84 30000 Medium0102640301 2000-08-07 5.3 13.9 1 . → .
79 0.69 7500 Medium0102641001 2001-07-08 8.7 10.5 1 . → .
67 0.88 14500 Low0102642001 2001-08-15 10.7 14.1 1 . → .
78 0.61 13000 Medium0102642101 2002-01-25 12.2 10.7 1 . → .
79 0.96 22600 Medium0102642301 2002-01-27 12.2 8.7 1 . → .
61 1.05 24300 Low0141980501 2003-01-22 3.5 1.1 1 . → .
16 1.11 6800 Low0141980601 2003-01-23 12.8 14.1 1 . → .
97 0.63 16600 High0141980401 2003-01-24 7.2 12.7 1 . → .
00 0.77 12200 High0141980801 2003-02-12 12.6 1.2 1 . → .
44 1.33 26800 Low0141980101 2003-07-11 6.1 10.6 1 . → .
54 0.88 10600 Low0141980301 2003-07-25 —–Notes: The observation date is given in year-month-day format. The observed flux, f x , the background-subtracted net count rate andthe total source counts (sum of MOS1 + MOS2) are in the 0.3 – 10 keV range. OBSID 0141980301 had no MOS observations and wasexcluded. variability can be seen, it is likely that the angle of observa-tion coincides with the edge of the turbulent photosphere,adding extrinsic variability to the X-ray signal (Middletonet al. 2011).Thus it appears that high quality X-ray spectroscopyof ULXs is consistent with many of the characteristics pre-dicted for super-Eddington accretion (e.g. Begelman et al.2006; Poutanen et al. 2007; Mineshige & Ohsuga 2011).However, there is a sub-class of ULXs with luminosities closeto 10 erg s − , that do not show the spectral inflection at ∼ ∼ Eddington through to super- and hyper-Eddington rates.A prime method for investigating this is to observe the evolu-tion of X-ray spectra with luminosity for these objects. Suchstudies have been undertaken for a number of more luminousULXs on the basis of data from
Chandra (e.g. Roberts etal. 2006),
Swift (e.g. Vierdayanti et al. 2010; Kong et al.2010) and
XMM-Newton (e.g. Feng & Kaaret 2006; Feng& Kaaret 2009), yielding important insights including thedemonstration that the temperature-luminosity relation forthe soft excess is indicative of the presence of a wind-drivenphotosphere, rather than a standard accretion disc aroundan IMBH (Kajava & Poutanen 2009). However, spectral evo-lution studies for less luminous ULXs are less numerous, andcan suffer from moderate data quality (e.g. four objects de-scribed by a simple multi-colour disc model in Kajava &Poutanen 2009).An obvious means of increasing our understanding ofthese objects is to study the brightest example, M33 X-8. This is the closest established ULX, long known as thebrightest persistent X-ray source in the Local Group (Longet al. 1981; Gottwald, Pietsch & Hasinger 1987). It is alsoone of only two ULXs (with M82 X-1) whose flux regularlyexceeds 10 − erg cm − s − in the XMM-Newton bandpass, and so has been the subject of many studies (e.g. Dubus,Charles & Long 2004; La Parola et al. 2003; Parmar et al.2001; Takano et al. 1994). It has been observed on severaloccasions with the EPIC instruments on
XMM-Newton (seee.g. Foschini et al. 2004; 2006). However, due to the mod-erate statistical quality of individual observations it has notbeen possible to constrain changes to the X-ray spectrumbetween observations (Weng et al. 2009). Here we presenta flux binned analysis that shows that the X-ray spectrumof M33 X-8 does alter subtly with luminosity, and has a lu-minosity dependence that is consistent with the picture ofULX spectral evolution in the super-Eddington regime.
The EPIC instruments on-board
XMM-Newton observedM33 X-8 on 12 occasions of varying durations during theearly years of the mission (see Table 1 for details on each ofthese) . As the source is often not visible in the PN obser-vations (as it lies off the edge of the chip) we restricted our-selves to using only the MOS data. The data were initiallycleaned using standard procedures. Good time interval fileswere created from the >
10 keV full-field light curves of eachobservation by removing periods with a raised backgroundcount rate, and used to filter subsequent data extractions.Where the source itself appeared in the outer MOSchips we used circular data extraction apertures of 45 arc-second radius for both source and background extractions,and in the three cases where the source is on-axis (OB-SIDs: 0102640101, 0141980501 and 0141980801 respectively)we instead use 30 arcsecond radius regions. We proceededto filter the data using sas v10 and extracted source andbackground spectra and lightcurves over standard event pat-terns ( ≤
12) and flags (= 0). The background subtracted0.3 – 10 keV count rate is given in Table 1 and suggests New, deep observations were obtained in the summer of 2010that are not public at the time of writing.c (cid:13) , 1–8 -ray spectral evolution in M33 X-8 Figure 1.
Average, re-binned (by a geometrical factor of 1.2), 0.3 – 10 keV PDS with 1 σ error bars. The PDS are extracted fromlightcurves made from the combination of the finessed, background-subtracted, source lightcurves of each flux bin. In practice this hasmeant losing some of the shorter available segments in order to obtain a larger frequency bandpass. In the low flux binned data weuse 401 bins/interval and 4 intervals/frame, in the medium bin, 525 bins/interval and 4 intervals/frame and in the high flux bin, 431bins/interval and 4 intervals/frame. The white noise level is shown as the horizontal dashed line and it is clear that there is no constrainedvariability power density in any bin. This is consistent with a model where the spectrum is dominated by stable disc emission. that there may be slight pile-up in OBSIDs 0141980501and 0141980801 (note that OBSID 0102640101 was taken insmall window mode, and therefore does not suffer from pile-up at these count rates). However, inspection of the patternsusing the sas tool epatplot showed that this is not signifi-cant in the MOS1 and only marginally significant for doublepatterns in the MOS2 observations and as such should notimpact a combined spectrum where the small effect is con-siderably diluted.The crude individual spectra appear to be well de-scribed by a single component with a quasi-thermal shape.To obtain flux estimates for each observation, we modelthe MOS1 and MOS2 spectral data with a simple thermalcontinuum (absorbed nthcomp ) together with the cflux model in xspec , which provides an error range on the fluxbased on the data uncertainty. We achieve good or accept-able fits to each of the individual observation datasets andobtain the absorbed fluxes given in Table 1. From the ob-served distribution of fluxes we break the observations into3 classes: low, medium and high flux levels. Assuming a dis-tance of 817 kpc to M33, consistent with recent X-ray sur-veys of the galaxy (e.g. T¨ullmann et al. 2011), these binscorrespond to observed luminosities of < . × erg s − (low flux bin), (1 . → . × erg s − (medium fluxbin), and > . × erg s − (high flux bin). Usingthe ftool addspec we proceed to co-add the respectivedatasets (spectra and response matrices) and obtain a MOS1and MOS2 dataset for each flux bin.As our selection of flux bin limits is somewhat arbitrary,we repeat the model dependent analyses reported in Section3 on rebinned data, where the highest flux observation ofthe low and medium flux bins was promoted into the nexthighest bin. In all cases we find behaviour consistent withthe results reported below, indicating our results are notsimply an artefact of the flux bins used.The total aggregated counts for each observation aregiven in Table 1 and, whilst the dispersion in the co-addeddata will be affected by the weighted differences betweenthe observations, the data quality overall is very good forthe low and medium flux bins (totals of ∼ ∼ XMM-Newton archive, and sufficient to constrain the spectra and drawconclusions on spectral variability in M33 X-8.
An initial inspection of the data using simple empirical mod-els demonstrated that M33 X-8 does indeed display spectralvariability between the flux binned datasets (which can betrivially confirmed by inspection of both Tables 2 & 3, andFigures 2 & 3). Here, we focus on how the spectra evolvewith luminosity, and analyse this evolution in light of a rangeof physical assumptions and models.
Several authors have used the high luminosity of ULXsto claim the presence of an intermediate-mass black hole( >
500 M ⊙ , e.g. Kaaret et al. 2003; Miller et al. 2003; Miller,Fabian & Miller 2004), which must imply sub-Eddingtonmass accretion rates. In black hole X-ray binaries suchstates are characterised by a two-component model of op-tically thick disc emission (Shakura & Sunyaev 1973) andComptonisation (see the reviews of McClintock & Remil-lard 2006; Done, Gierli´nski & Kubota 2007). We test thisassertion by applying a model containing both of thesecomponents together with neutral absorption and a con-stant to account for the differences between the detectors(in xspec: constant*tbabs*(diskbb+comptt) ) . WhilstTable 2 shows this model provides an acceptable fit to thedata in all three cases, the characteristic temperature of thehigh energy tail is ∼ In these and all subsequent fits a constant component is utilised.It is fixed for MOS1, and the MOS2 value never deviates by morethan ± . × cm − in the directionof M33 (Kalberla et al. 2005) is assumed, and used as the lowerbound in fitting the absorption component.c (cid:13) , 1–8 M. J. Middleton, A. D. Sutton & T. P. Roberts
Table 2.
Best fitting sub-Eddington models.Flux bin Low Medium High tbabs*(diskbb+comptt) N H +0 . − . ± .
006 0.081 ± . kT in (keV) 0.29 +0 . − . ± .
18 0.67 ± . kT comp (keV) 1.40 +0 . − . +0 . − . +0 . − . τ +0 . − . > . > . χ (d.o.f.) 689.8 (655) 654.9 (614) 521.9 (478)Null P 0.17 0.12 0.08 tbabs*kerrbb N H ± .
003 0.072 ± .
003 0.067 +0 . − . a > . > . > . m +0 . − . +51 . − . +53 . − . χ (d.o.f.) 1080.1 (655) 717.3 (614) 570.6 (478)Null P 2.0 × − × − × − Notes: Best-fitting parameters for the two sub-Eddington mod-els used. The units of the foreground column, N H , are 10 cm − . kT in is the temperature of the inner edge of the accretion disc, and kT comp and τ are the temperature and optical depth of the Comp-tonising medium. a is the dimensionless spin parameter and ˙ m isthe ‘effective’ mass accretion rate of the relativistically-smearedaccretion disc (with the latter in units of 10 g s − ), assumingzero torque at the inner boundary. The table also shows the valuesfor χ and the number of degrees of freedom for each model, andthe null hypothesis probability for this model being an acceptablefit to the data. XRB spectra (indicated by the unbroken power-law con-tinua out to >
100 keV, see e.g. McClintock & Remillard2006; although see e.g. Zdziarski et al. 2005 for the ULX-like behaviour of GRS 1915+105). We confirm this physicaldifference by fixing the plasma temperature at 50 keV inour models, and measuring the resulting change in fit qual-ity. In all three cases the fit is poorer with a hot corona, by∆ χ of 134, 42 and 15 respectively for one more degree offreedom. This shows that a cool corona provides a 99.98%improvement according to the F-test for the high flux bin,and a substantially more significant improvement for theother bins, compelling evidence that the corona does notappear similar to that in standard sub-Eddington states.Additionally, the Compton tail in a sub-Eddington stateis highly variable on short timescales providing an unam-biguous test for such a model. We characterise the variabil-ity using the excess rms of the 3 – 10 keV lightcurve in eachflux bin (binned to 250 s), i.e. the variability above the Pois-son (white) noise level of the lightcurve (see Edelson et al.2002). We find that there is no constrained variability above3 keV with 3 σ upper limits of < <
8% and <
11% for thethree flux bins respectively. This strongly argues against atwo-component sub-Eddington model for the data. Indeed,we note that the combination of a hot disc (in 2/3 fits), and acool, optically thick and invariant corona, demonstrates thatthis ULX cannot be described by the cool disc plus power-law continuum toy model previously used to infer IMBHs inULXs.If the data were to be described solely by disc emis-sion then we would not expect there to be any variabilityon anything other than the longest timescales (Wilkinson & Uttley 2009). As we have substantial evidence againstthe two-component sub-Eddington model and are now in-terested in the variability behaviour of the emission as awhole, we extract the Fourier-frequency dependent powerdensity spectra (PDS) over the full energy bandpass. Wenget al. (2009) report the lack of high frequency variabilityin the individual observations. Here however, we can ex-tend this to longer timescales by selecting the shortest seg-ment of continuous lightcurve and taking integer number ofintervals of this length across the remaining observations.This can require certain finessing to obtain the maximumamount of available data and broadest frequency bandpass(as this goes from 1/[longest available individual segmentlength] to 1/[2 × bin size]). In Figure 1 we present the av-erage power density spectrum (PDS - extracted using the ftool powspec ) for each flux bin, normalised to rms unitsfollowing geometrical re-binning with white noise included.Quite clearly there is no constrained variability over any ofthe available frequency bins, consistent with the source hav-ing suppressed red noise as seen in a handful of ULXs (Heilet al. 2009), and the emission originating in the accretiondisc.However, a thin disc is a poor description of the spec-tral data as the Wien tail is not broad enough to fit thehigh energy emission, leading to poor fits ( χ of 1441.5/658,931.4/617 and 650.7/481 for an absorbed multi-colour discblackbody spectrum – diskbb in xspec – fit to each fluxbin respectively). Disc emission that has been smearedby relativistic effects is considerably broader and so wealso fitted the spectral data with the xspec model con-stant*tbabs*kerrbb . The data and best fitting modelsare shown in Figure 2, and the fit parameters are detailedin Table 2. Even in this extremely relativistically broadenedcase we still find the models poorly describe the data, withobvious excesses above the model at high energies. It wouldtherefore appear that none of the sub-Eddington models candescribe both the broad shape of the spectra and also thelack of high energy variability. Examples of Eddington accretion rates (i.e. accretion ratesclosely approximating those expected at the Eddington limitfor an object) have been inferred in the outburst of severalGalactic X-ray binaries, for example V404 Cyg ( ˙Zycki, Done& Smith 1999), V4641 Sgr (Revnivtsev et al. 2002) and theneutron star system Cir X-1 (Done & Gierli´nski 2003). How-ever, ULXs are observed to be persistently luminous, whichwould suggest a closer analogy to a long-lived outburst sys-tem such as GRS 1915+105 (see Remillard & McClintock2006), although this source is also known to undergo dra-matic state changes on short timescales (e.g. Belloni et al.2000; Middleton et al. 2006).Taking a theoretical approach, we would expect that,as the Eddington limit is neared, the properties of the flowthrough the disc would change as advection becomes moreimportant as a method for removing energy from the flow(Mineshige et al. 2000; Abramowicz et al. 1988). We wouldalso expect that, where the flow is highly illuminated, ma-terial would be driven off the disc surface by radiation pres-sure, with the massive outflow forming a photosphere abovethe disc (see Poutanen et al. 2007). This should modify the c (cid:13) , 1–8 -ray spectral evolution in M33 X-8 Figure 2.
Flux binned X-ray data for the low, medium and high datasets (MOS1 in black, MOS2 in red) folded with the best-fittingsmeared disc model in blue ( tbabs*kerrbb ). The residuals to the best fit are shown below the best-fitting model and demonstrate that,with even a highly smeared disc, the broad shape of the data is not well matched.
Table 3.
Best fitting Eddington models.Flux bin Low Medium High tbabs*diskpbb N H +0 . − . +0 . − . +0 . − . kT in (keV) 1.90 +0 . − . +0 . − . +0 . − . p ± .
01 0.56 ± .
01 0.56 +0 . − . χ (d.o.f.) 699.10 (657) 667.5 (616) 532.1 (480)Null P 0.12 0.07 0.05 tbabs*(diskpbb+comptt) N H ± .
010 0.115 +0 . − . ± . kT in (keV) 1.22 ± .
04 0.93 ± .
02 0.70 +0 . − . p ± .
01 0.62 ± .
01 0.68 +0 . − . kT comp (keV) < .
76 1.67 +0 . − . +0 . − . τ < .
85 9.63 +3 . − . +2 . − . χ (d.o.f.) 691.0 (654) 651.4 (613) 521.6 (477)Null P 0.15 0.14 0.08Notes: Best-fitting parameters for the Eddington and/or super-Eddington models. Here p is the dimensionless index of the radialtemperature dependence of an advection dominated slim disc (seetext), and the other variables are as per Table 2. shape of the continuum emission via absorption, emissionand scattering, producing a fully thermalised, blackbody-like component. Here however, the spectral shape of M33X-8 is unlikely to be described solely by a photosphere cov-ering a hot disc, as the peak temperatures of the observedspectra are implausibly high for a pure photosphere (c.f.King 2004). We therefore arrive at the model of an advection-dominated Eddington ‘slim’ disc (Mineshige et al. 2000).Several authors (Foschini et al. 2006; Weng et al. 2009) haveshown this model to be a reasonable description of the mod-erate signal-to-noise data of individual observations, with areported requirement for an additional soft power-law con-tinuum to high energies in some cases. Here however, wefit the flux binned data with an absorbed p -free disc model( constant*tbabs*diskpbb in xspec ), where the index ofthe temperature profile p is a free parameter ( T ∝ R − p for disc temperature T and radius R ). Such a model hasbeen used as evidence for slim disc spectra in ULXs, as apredicted change in the profile from p = 0 .
75 for standarddiscs to p = 0 . p to increase significantly with flux ei-ther. We can see from Table 3 that the temperature dropsby more than 0.3 keV yet the p value increases significantly. c (cid:13) , 1–8 M. J. Middleton, A. D. Sutton & T. P. Roberts
Figure 3.
The ratio of the data in each flux bin to the best fittingslim disc model ( tbabs*diskpbb ) for the lowest flux bin. Thepanels show the residuals from: a) the low flux bin; b) the mediumflux bin; and c) the high flux bin. The MOS1 data is shown inblack, and MOS2 in red. It is clear that, besides the expectedincrease in flux, the medium and high flux binned datasets differin shape with an apparent deviation in the continuum above ∼ Thus the change here is contrary to what we would expectfor an advection-dominated Eddington flow. However, thedecrease in temperature of the peak of the emission could beexplained by the inclusion of a flux-dependent cooling mech-anism. In particular, cooling could occur via the launchingof a wind, and Comptonisation in the optically thick plasmaof the outflowing material. In Eddington/super-Eddingtonflows, the launching of material is dominated by the radia-tion pressure (as the effective gravity goes as 1-L/L
Edd ) andso one would expect to see its effect increase with luminosity.Hence, in terms of the spectral evolution, we might expectto see both the disc cool, and an optically-thick Comptoni-sation medium emerge, with increased luminosity. We illus-trate what the data shows in Figure 3, where we show thespectral residuals for each dataset compared to the best fit-ting model for the absorbed p -free disc model in the lowestflux bin (cf. also Table 3). This shows that, besides the ex-pected change in the flux level, we also see changes in theshape of the continuum with flux. In particular, a compo-nent with turnover in the 2 – 3 keV range appears in the medium flux bin spectrum, and appears more pronouncedin the high flux bin.We attempt to model this by introducing a Comptontail to the model (using the comptt model, such that weemployed a constant*tbabs*(diskpbb+comptt) modelin xspec ) with the seed photons drawn from the soft compo-nent in order to provide a physically limiting case and allowthe spectral parameters to be constrained. In the case of themedium flux binned dataset the fit is notably improved bythis additional component (∆ χ of 16 for 3 d.o.f.s), and itis also marginally improved in the high flux binned dataset(∆ χ > of 11 for 3 d.o.f.s). Although we would expect theslim disc emission to dominate at lower luminosities we alsofit this model to the low flux binned data. We obtain onlya slightly improved fit (∆ χ > of 8 for 3 d.o.f.s) and, as ex-pected, the disc dominates the spectrum with only a smallcontribution to the higher energy emission made via Comp-tonisation. In order to realistically constrain the propertiesof the best-fitting two component model, we freeze each best-fitting component in turn and determine 90% error limits onthe model parameters. The resulting fit parameters and er-rors are presented in Table 3. They suggest that, as the lumi-nosity of M33 X-8 increases, the soft component decreases intemperature and becomes less advection-dominated whilstthe hard component becomes stronger and cooler. A plot ofthe evolution of the best-fitting models is shown in Figure4. As a consistency check we apply these best-fitting modelsto each of the individual observations of M33 X-8 in eachbin and, in every case, find a good or acceptable fit to thedata. Assuming that the population of ULXs contain stellar massblack holes rather than IMBHs, their spectral evolutionshould provide the much sought-after explanation for thechanging structure of the accretion flow at and above theEddington limit. This requires repeat observations of brightULXs at different luminosities. Whilst the relatively largeamounts of data from M33 X-8 have been analysed by otherauthors (e.g. Weng et al. 2009, La Parola et al. 2003, etc), ithas not previously been flux binned in an overt attempt toidentify spectral evolution. Using this method, we have beenable to test models for the spectral behaviour of this source.We argue strongly against a sub-Eddington two-componentdescription, particularly one describing ULXs as IMBHs,based on the shape of and lack of variability in the harderspectral component, that results from Comptonisation. Theaverage PDS of each flux bin further constrained the source’sproperties and showed that, across the full energy bandpass,there is no constrained variability above the white noise atany frequency. Whilst this could imply that the spectrumis dominated by stable emission from a disc, we show thatneither a standard nor a relativistically broadened thin (sub-Eddington) disc is a suitable description of the data.Given the source luminosity of ∼ × erg s − , Ed-dington mass accretion rates are easily satisfied by a blackhole mass of ∼ ⊙ , consistent with the masses of Galacticblack holes. We find that an Eddington slim disc providesa good description of the data with constrained differencesin the temperatures and radial emission profiles of the low c (cid:13) , 1–8 -ray spectral evolution in M33 X-8 Figure 4.
Best-fitting super-Eddington, two-component models to each of the flux bins. The blue line shows the total spectrum, with thegreen and red lines showing the disc and Comptonised spectral components respectively. Although the spectra are broad and can sufferfrom model degeneracies, the evolution of a slim disc spectrum alone from low to medium fluxes appears unphysical, and so this modelpresents the most likely physical description of the flux binned data. The changing properties of the disc and relative contributions fromthe two components is consistent with the increased flux leading to greater wind production, which in turn leads to increased coolingof the disc and hence an increasingly less advection-dominated solution. This is naturally accompanied by the emerging optically-thickcoronal component as more material is driven out from the disc. and medium flux binned data. However, these differences areinconsistent with the expected characteristics of disc accre-tion, and instead imply the presence of an additional coolingmechanism. Subtle changes in the spectra of the medium andhigh flux binned states, compared to the low flux state, arewell modelled by including a Comptonised component.M33 X-8, then, turns out to be a crucial object for fur-thering our understanding of accretion at Eddington andsuper-Eddington rates. At its lower fluxes it appears welldescribed by a slim disc model alone; however as its flux in-creases, an optically-thick Comptonisation media becomesmore important, in effect ‘stretching’ the disc-like spectrumby cooling the disc component, and simultaneously provid-ing a harder, upscattered, spectral component. This is cru-cial because it appears to reflect the initial stages of the di-vergence of the ULX spectra we see at higher luminosities,where the spectra are well-modelled by a cool, disc-like com-ponent and a harder, optically-thick Comptonisation com-ponent (Gladstone et al. 2009). If so, it is likely the resultof the emergence of the expected radiatively-driven wind.We sketch a toy model of a possible physical scenarioto explain both this initial wind emergence phase, and thehigher luminosity photosphere phase in Figure 5. In the toppanel, the accretion rate is only just entering the Eddingtonregime. Here, the soft component is the outer, advection-dominated slim disc, which extends down to some radius(given as R2 in Figure 5). Within this radius the accretionflow is highly illuminated and we see the emergence of out-flowing winds down to some radius close to the innermoststable circular orbit (ISCO) beyond which we expect theflow to become highly turbulent (possibly producing the ex-pected, as yet unobserved, high energy, optically thin com-ponent). The unbound plasma in the wind Comptonises theunderlying hotter disc photons, producing the hard compo-nent in the spectrum. As the luminosity increases, windsmay be driven from further out, whereas closer in, the windmay become increasingly mass loaded as bound material at lower luminosities may now be lifted from the ‘surface’ of thedisc. This further cools the underlying accretion flow leadingto a cooler peak in the hard spectral component. In addition,the soft component is now cut off at a larger radius and sopeaks at a lower temperature as well as being less advectiondominated (due to being further out in the gravitational po-tential of the BH). This scenario qualitatively matches theimplied behaviour of M33 X-8. The key remaining issue isthen how this scenario relates to the spectra seen for moreluminous ULXs. We emphasise that in this toy model, asthe X-ray luminosity increases we would expect to see thewind being driven from increasingly further out from theISCO. Eventually this would leave little or no disc emissionbeyond the outer launching radius, as the very massive windwould extend over most of the hot (X-ray emitting) regionsof the disc and fully thermalise the underlying disc photons,producing the soft excess seen in ULX spectra. However, wewould also expect to see ‘bare’ disc emission from within theinner launching radius of the wind (the photospheric radius)where the local gravity is so low that the radiation pressurehas ‘blown’ the excess material away, leaving an approximatehot thin disc. The thermal Comptonisation spectrum thenoriginates either in a corona of hot electrons tightly boundto the upper layers of the accretion disc; or perhaps is thespectrum of the accretion disc itself, as its opacity wouldalter at the high temperatures within the photospheric ra-dius. This scenario is shown in the bottom panel of Fig. 5,and can explain the spectral properties of the more luminousULXs seen to have clear two-component spectra (see Fig. 8Gladstone et al. 2009, also Middleton et al. 2011). In thesecases the wind/photosphere emission is thermally decoupledfrom the inner disc emission and so the relative amounts ofhard/soft emission are most likely degenerate in inclinationangle and mass accretion rate.This model can also consistently explain the associatedvariability properties of ULXs as in all cases the spectralcomponents in the X-ray bandpass are stable (although we c (cid:13) , 1–8 M. J. Middleton, A. D. Sutton & T. P. Roberts
Figure 5.
Our toy model for ULX disc behaviour as the luminosity reaches (top panel) and far exceeds (bottom panel) its Eddingtonluminosity. The top panel shows how, at or near to Eddington, we expect the majority of the disc to be advection-dominated (slim disc:red), but as the flow is highly illuminated further in (shown as within the radius R2), we expect material to be lifted from the surfaceof the disc by radiation pressure (from the inner photospheric radius, R1 close to the inner stable circular orbit, ISCO) which cools theunderlying flow. This outflowing, hot material Comptonises the hot disc photons, producing the hard component in the spectrum, whilstthe soft component originates in the slim disc emission from beyond the outer photospheric radius (R2). As the mass accretion rate andluminosity increases we would expect winds to be driven from further out in the disc, and increased mass loading of the wind closer in.This produces a cooler hot component and a cooler disc component (as the inner edge of the ’naked’ disc emission is now also furtherout), as is seen in M33 X-8. The bottom panel then shows what we think may be happening in those ULXs where the luminosity hasincreased further, to be substantially super-Eddington. Here, the radiation pressure is so intense that all the excess material within R2has been driven off leaving a bare, hot, thin disc. The outer launching radius now extends to cover all of the remaining X-ray emittingdisc, producing the soft cool component in the spectrum in a thick photosphere, whilst the hard component is from the hot bare disc.We suggest that the varying amounts of each component evident in different ULX spectra (c.f. Gladstone et al. 2009) may be due todegeneracies in inclination and mass accretion rate between different ULXs. predict that at higher energies there is likely to be an op-tically thin, highly variable component as seen in Galacticblack hole binaries). However, if our line-of-sight interceptsthe launching region of the photosphere at high luminositieswhere the wind is in the outer rather than inner disc, thenwe expect this extrinsic variability to lead to the hard com-ponent being highly variable (e.g. NGC 5408 X-1: Middletonet al. 2011).Hence, by examining how the X-ray spectral and tim-ing properties of the nearest ULX vary with source flux,we have shown that it is consistent with the picture of asource accreting at the threshold of the super-Eddingtonregime. As its luminosity increases, it begins to betray thesignature of a possible outflow from its central regions. Thismay be the emergence of the outflow that appears to domi-nate the characteristics of ULXs at higher luminosities. Thiswork shows the power of considering both spectral and tim-ing data, using the highest quality datasets available from
XMM-Newton . Clearly, if we are to develop a deeper un-derstanding of how ULXs work, and further investigate thispicture of objects dominated by a radiatively-driven wind,obtaining an increased number of similar datasets must bea priority for this and future missions.
We thank the anonymous referee for their useful sugges-tions. MM and TR thank STFC for support in the form ofa standard grant, and AS similarly thanks STFC for sup-port via a PhD studentship. This work is based on observa-tions obtained with
XMM-Newton , an ESA science missionwith instruments and contributions directly funded by ESAMember States and NASA.
REFERENCES
Abramowicz M. A., Czerny B., Lasota J. P., Szuszkiewicz E.,1988, ApJ, 332, 646Begelman M. C., King A. R., Pringle J. E., 2006, MNRAS, 370,399Belloni T., Klein-Wolt M., M´endez M., van der Klis M., vanParadijs J., 2000, A&A, 355, 271Done C., Gierli´nski M., 2003, MNRAS, 342, 1041Done C., Gierli´nski M., Kubota A., 2007, A&ARv, 15, 1Dubus G., Charles P. A., Long K. S., 2004, A&A, 425, 95Edelson R., Turner T. J., Pounds K., Vaughan S., Markowitz A.,Marshall H., Dobbie P., Warwick R., 2002, ApJ, 568, 610Feng H., Kaaret P., 2006, ApJ, 650, L75Feng H., Kaaret P., 2009, ApJ, 696, 1712Foschini L., Rodriguez J., Fuchs Y., Ho L. C., Dadina M., DiCocco G., Courvoisier T. J.-L., Malaguti G., 2004, A&A, 416,529 c (cid:13) , 1–8 -ray spectral evolution in M33 X-8 Foschini L., et al., 2006, AdSpR, 38, 1378Gladstone J. C., 2011, arXiv, arXiv:1101.5387Gladstone J. C., Roberts T. P., Done C., 2009, MNRAS, 397,1836Gottwald M., Pietsch W., Hasinger G., 1987, A&A, 175, 45Heil L. M., Vaughan S., Roberts T. P., 2009, MNRAS, 397, 1061Kaaret P., Corbel S., Prestwich A. H., Zezas A., 2003, Sci, 299,365Kajava J. J. E., Poutanen J., 2009, MNRAS, 398, 1450Kalberla P. M. W., Burton W. B., Hartmann D., Arnal E. M.,Bajaja E., Morras R., P¨oppel W. G. L., 2005, A&A, 440, 775King A. R., 2004, NuPhS, 132, 376Kong A. K. H., Yang Y. J., Yen T.-C., Feng H., Kaaret P., 2010,ApJ, 722, 1816Kubota A., Mizuno T., Makishima K., Fukazawa Y., Kotoku J.,Ohnishi T., Tashiro M., 2001, ApJ, 547, L119La Parola V., Damiani F., Fabbiano G., Peres G., 2003, ApJ, 583,758La Parola V., Peres G., Fabbiano G., Kim D. W., Bocchino F.,2001, ApJ, 556, 47Long K. S., Dodorico S., Charles P. A., Dopita M. A., 1981, ApJ,246, L61McClintock & Remillard, 2006, In
Compact Stellar X-raySources , ed. WHG Lewin, M van der Klis, pp. 157. Cam-bridge: Cambridge University PressMiddleton M., Done C., Gierli´nski M., Davis S. W., 2006, MN-RAS, 373, 1004Middleton M. J., Roberts T. P., Done C., Jackson F. E., 2011,MNRAS, 411, 644Miller M. C., Colbert E. J. M., 2004, IJMPD, 13, 1Miller J. M., Fabbiano G., Miller M. C., Fabian A. C., 2003, ApJ,585, L37Miller J. M., Fabian A. C., Miller M. C., 2004, ApJ, 607, 931Mineshige S., Kawaguchi T., Takeuchi M., Hayashida K., 2000,PASJ, 52, 499Mineshige S., Ohsuga K., 2011, AN, 332, 402Parmar A. N., et al., 2001, A&A, 368, 420Poutanen J., Lipunova G., Fabrika S., Butkevich A. G., Abol-masov P., 2007, MNRAS, 377, 1187Remillard R. A., McClintock J. E., 2006, ARA&A, 44, 49Revnivtsev M., Gilfanov M., Churazov E., Sunyaev R., 2002,A&A, 391, 1013Roberts T. P., 2007, Ap&SS, 311, 203Roberts T. P., Kilgard R. E., Warwick R. S., Goad M. R., WardM. J., 2006, MNRAS, 371, 1877Shakura N. I., Sunyaev R. A., 1973, A&A, 24, 337Stobbart A.-M., Roberts T. P., Wilms J., 2006, MNRAS, 368, 397Takano M., Mitsuda K., Fukazawa Y., Nagase F., 1994, ApJ, 436,L47T¨ullmann R., et al., 2011, ApJS, 193, 31Vierdayanti K., Done C., Roberts T. P., Mineshige S., 2010, MN-RAS, 403, 1206Vierdayanti K., Mineshige S., Ebisawa K., Kawaguchi T., 2006,PASJ, 58, 915Weng S.-S., Wang J.-X., Gu W.-M., Lu J.-F., 2009, PASJ, 61,1287Wilkinson T., Uttley P., 2009, MNRAS, 397, 666Zdziarski A. A., Gierli´nski M., Rao A. R., Vadawale S. V.,Miko lajewska J., 2005, MNRAS, 360, 825˙Zycki P. T., Done C., Smith D. A., 1999, MNRAS, 309, 561c (cid:13)000