X-ray Variability and Hardness of ESO 243-49 HLX-1: Clear Evidence for Spectral State Transitions
Mathieu Servillat, Sean A. Farrell, Dacheng Lin, Olivier Godet, Didier Barret, Natalie A. Webb
aa r X i v : . [ a s t r o - ph . H E ] A ug Draft version September 27, 2018
Preprint typeset using L A TEX style emulateapj v. 11/10/09
X-RAY VARIABILITY AND HARDNESS OF ESO 243–49 HLX-1:CLEAR EVIDENCE FOR SPECTRAL STATE TRANSITIONS
Mathieu Servillat , Sean A. Farrell , Dacheng Lin , Olivier Godet ,Didier Barret , Natalie A. Webb Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, MS-67, Cambridge, MA 02138 – [email protected] Department of Physics and Astronomy, University of Leicester, University Road, Leicester, LE1 7RH, UK Sydney Institute for Astronomy, School of Physics A29, The University of Sydney, NSW 2006, Australia Universit´e de Toulouse; Universit´e Paul Sabatier – Observatoire Midi-Pyr´en´ees,Institut de Recherche en Astrophysique et Plan´etologie (IRAP), Toulouse, France and Centre National de la Recherche Scientifique; IRAP; 9 Avenue du colonel Roche, BP 44346, F-31028 Toulouse cedex 4, France
Draft version September 27, 2018
ABSTRACTThe ultra-luminous X-ray (ULX) source ESO 243–49 HLX-1, which reaches a maximum luminosityof 10 erg s − (0.2–10 keV), currently provides the strongest evidence for the existence of intermediatemass black holes. To study the spectral variability of the source, we conduct an ongoing monitoringcampaign with the Swift
X-ray Telescope, which now spans more than two years. We found thatHLX-1 showed two fast rise and exponential decay (FRED) type outbursts in the
Swift
XRT light-curve with increases in the count rate of a factor ∼
40 separated by 375 ±
13 days. We obtainednew
XMM-Newton and
Chandra dedicated pointings that were triggered at the lowest and highestluminosities, respectively. From spectral fitting, the unabsorbed luminosities ranged from 1 . × to1 . × erg s − . We confirm here the detection of spectral state transitions from HLX-1 reminiscentof Galactic black hole binaries: at high luminosities, the X-ray spectrum showed a thermal statedominated by a disk component with temperatures of 0.26 keV at most, and at low luminosities thespectrum is dominated by a hard power law with a photon index in the range 1.4–2.1, consistentwith a hard state. The source was also observed in a state consistent with the steep power law state,with a photon index of ∼ .
5. In the thermal state, the luminosity of the disk component appears toscale with the fourth power of the inner disk temperature which supports the presence of an opticallythick, geometrically thin accretion disk. The low fractional variability (rms of 9 ± > M ⊙ can be derived, and therelatively low disk temperature in the thermal state also suggests the presence of an intermediatemass black hole of a few 10 M ⊙ . Subject headings:
X-rays: individual (ESO 243–49 HLX-1) – X-rays: binaries – black holes – accretion,accretion disks INTRODUCTION
At present, only two families of black holeshave convincing observational evidence: stellar-massblack holes detected in some X-ray binaries (e.g.Remillard & McClintock 2006) and super-massive blackholes (SMBHs, 10 − M ⊙ ) which are ubiquitous inthe center of galaxies (Kormendy & Richstone 1995),sometimes revealing themselves as active galactic nu-clei (AGN). Remarkably, the existence of intermediatemass black holes (IMBHs, 10 − M ⊙ ) remains to beproven. Such objects may have been ejected follow-ing an interaction with the central SMBH and may re-main in the halo of galaxies (Micic et al. 2011). Theymay be found in globular clusters, and the strongestcase is the massive cluster G1 in M31, where dynami-cal, X-ray, and radio studies are consistent with a massof 2 × M ⊙ (Gebhardt et al. 2002; Kong et al. 2010).However, radio observations of Galactic globular clus-ters currently provide only upper limits on the mass (e.g. Maccarone & Servillat 2008; Lu & Kong 2011). IMBHsmay also be the nuclei of satellite galaxies captured dur-ing hierarchical merging (King & Dehnen 2005). Finally,they may be the engine of some ultra-luminous X-raysources (ULXs, e.g. Miller et al. 2004).A ULX is a non-nuclear extragalactic X-ray source thathas an X-ray luminosity that exceeds the Eddington lu-minosity — 1 . × ( M BH /M ⊙ ) erg s − — for a stellar-mass black hole ( ∼ M ⊙ ), supposing isotropic emission(see Roberts 2007, for a review). Different solutions havebeen proposed for the high luminosity problem of ULXs:(1) they are stellar-mass black holes emitting up to a fac-tor of 10 above their Eddington limit (Begelman 2002)in a new ”ultra-luminous” state with super-Eddingtonaccretion rates (e.g. Gladstone et al. 2009), (2) they arestellar-mass black holes with geometric (e.g. King 2009)or relativistic (K¨ording et al. 2002) beaming, or (3) theobjects are IMBHs (e.g. Miller et al. 2004). It could alsobe a combination of these possibilities. For the mostluminous ULXs above 10 erg s − , the mass argument Servillat et al.seems to be the main one that can convincingly explainthe extreme luminosities.One of the keys to further constraining the natureof the black hole in such sources is the study oftheir spectral and timing variability. Indeed, Galac-tic black hole binaries (GBHBs) containing stellar-mass black holes have been commonly observed to un-dergo transitions between different spectral states (e.g.Tananbaum et al. 1972; Kubota & Makishima 2004, seeRemillard & McClintock 2006; Done et al. 2007; Belloni2010 for reviews). In the thermal state, the emission isdominated by an optically thick, geometrically thin ac-cretion disk component (Shakura & Sunyaev 1973), withtemperatures of ∼ L disk , inner temperature T in and mass of the blackhole M BH are theoretically related: L disk ∝ T for agiven M BH , and T in ∝ M − / (Shakura & Sunyaev 1973;Makishima et al. 2000). In the hard state, the accre-tion disk appears to be fainter and cooler, and may betruncated at a large radius. The physical condition ofmaterial within this radius remains uncertain. Investi-gations of GBHBs in the hard state suggest that bothsynchrotron and Compton components contribute to thebroadband, power law like spectrum with typical photonindices of 1 . < Γ < . > .
4, e.g. Remillard & McClintock 2006).Consideration of simple properties such as X-ray spec-tral hardness and fractional variability have led to abetter understanding of the physical processes in ac-tion (e.g. Belloni 2010). In a hardness-intensity diagram(HID), the source shows a hysteresis curve between thethermal state and the hard state (Miyamoto et al. 1995;Maccarone & Coppi 2003). The root mean square (rms)variability amplitude is generally low in the thermal state( < ∼ kT in = 1 . . L X = 2–8 × erg s − , Feng & Kaaret 2010), and M82X37.8+54 (0 . . L Xmax = 4 . × erg s − ,Jin et al. 2010) . Those sources could harbor a fast spin-ning black hole of 200–800 M ⊙ and < M ⊙ , respec-tively. ULXs with kT in ∼ . ∼ ∼ erg s − (Farrell et al.2009). The distance, and thus the high X-ray luminosityof HLX-1, have been firmly confirmed through the de-tection of an H α line at a redshift consistent with thatof its host galaxy (Wiersema et al. 2010). The H α linehas been found in the spectrum of the optical counter-part which falls inside the Chandra error circle of thesource (0 . ′′ Swift
X-ray Telescope (XRT) in August 2009 showeda hardening of the source at low luminosities which wasinterpreted as the first evidence for a change to the hardspectral state (Godet et al. 2009).Our aim in this paper is to measure the spectral hard-ness and the variability of HLX-1 at different luminositystates, using all the X-ray data available from current ob-servatories in order to test similarities with GBHBs orother ULXs. For this purpose, we obtained new XMM-Newton and
Chandra dedicated pointings that were trig-gered at the lowest and highest luminosities, respectively.The X-ray datasets are described in Section 2. Spectralanalysis of firstly the high luminosity states (Section 3)and then the low luminosity state (Section 4) follows. Aglobal analysis is given in Section 5 and the final Sec-tion 6 discusses the properties of HLX-1 compared toGBHBs and other ULXs. DATA AND LONG TERM VARIABILITY
Swift
XRT data
HLX-1 has been regularly monitored by the
Swift
XRTsince October 2008. Since August 2009, observations oc-curred on average every ∼ Swift
XRT pipeline version 0.12.4.We used the grade 0-12 events, giving slightly higher ef-fective area at higher energies than grade 0 events. Thebackground extraction region, chosen to be close to thesource extraction region, is the same for all epochs. No XMM-Newton sources are present inside the background Except two
Chandra
HRC-I observations (Webb et al. 2010)as this instrument does not provide spectral information. http://heasarc.gsfc.nasa.gov/docs/swift/analysis/ tate transitions of ESO 243–49 HLX-1 3 Time [MJD] S w i f t X R T C o un t R a t e ( . − k e V ) S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 ←XMM1 XMM2 XMM3 Chandra
Figure 1.
Swift
XRT light curve of HLX-1. The epochs that were grouped to generate spectra are shaded with labels S1 to S10. Thedates of the
XMM-Newton and
Chandra observations are represented by vertical lines.
Table 1
XMM-Newton observations of HLX-1Obs. name ObsID Date MJD Exp. time [ks] Instrument Mode Source region Background regionXMM1 0560180901 2004-09-23 53271 21 MOS1&2, pn full-frame 22.5 ′′ ′′ –63.4 ′′ XMM2 0560180901 2008-11-28 54798 50 MOS1&2 full-frame 27 ′′ ′′ –76 ′′ pn small window 27 ′′ × ′′ XMM3 0655510201 2010-05-14 55330 100 MOS1&2 full-frame 14 ′′ ′′ –47 ′′ pn small window 14 ′′ × ′′ extraction region.We generated a light curve from all the Swift point-ings with a binning of a minimum of 50 counts per binfor the energy band 0.3-10 keV (Figure 1) using the webinterface (Evans et al. 2007, 2009). There is a recurrencein the light curve of 375 ±
13 days between the two in-creases that may be periodic, but this will have to beconfirmed by further observations. If the behavior is in-deed periodic, the next maximum would occur in Au-gust/September 2011. The cause of this variability hasbeen investigated in more details by Lasota et al. (2011).We combined the data into different epochs labelledS1 to S10 in Figure 1, and extracted a spectrum foreach epoch. A 20-pixels (47 . ′′
2) radius circle was usedto extract the source and the background spectra us-ing
XSELECT v2.4a. The ancillary files were createdwith
XRTMKARF v0.5.6 and exposure maps gener-ated with
XRTEXPOMAP v0.2.7. The response fileswxpc0to12s6 20070901v011.rmf was used in the spec-tral fitting process. We used a binning of a minimum of20 counts per bin for each spectrum except the S2 andS6 spectra in order to use the χ statistic within Xspec12.6 (Arnaud 1996). For the S2 and S6 spectra, we usedthe Cash statistic (Cash 1979) due to the lower numberof counts in these spectra. XMM-Newton
EPIC data
The field of ESO 243–49 and HLX-1 was observed threetimes with
XMM-Newton (Table 1). HLX-1 was detectedserendipitously at an off-axis angle of 9 . ′
29 in XMM1, andXMM2 was obtained to show the spectral variability ofHLX-1 (Farrell et al. 2009). Following a transition to avery low flux state, we triggered the XMM3 observation.The data were processed using the
XMM-Newton
Sci-ence Analysis System (SAS) v10.0 software with themost recent calibration files as of 2010 December 13. Theobservation data files (ODF) were reduced using the ep-proc and emproc tasks to produce event lists. Singleevent light curves with energies exceeding 10 keV weregenerated for each camera in each observation in orderto identify periods of high background related to softproton flares. In the XMM1 and XMM2 observationsthe background levels were low in each camera with noflaring events, and so no good time interval (GTI) filter-ing was applied. However, during the XMM3 observa-tion significant flaring events were present at the startand end of the exposure, with an additional small short-duration flare occurring ∼
16 ks into the observation. Forthe XMM3 data we therefore generated GTI files for the http://xmm.esa.int/sas/ Servillat et al. −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0.010 -2 -1 C o un t R a e ( . − k e V ) XMM1XMM2 XMM3 pn −0.5 −0.4 −0.3 −0.2 −0.1 0.0 0.1 Hardness Ra io -3 -2 C o un t R a t e ( . − k e V ) XMM1XMM2 XMM3XMM1
MOS1MOS2
Figure 2.
Hardness intensity diagram for HLX-1 based on
XMM-Newton count rates (counts s − ). Hardness ratios are HR =(B2 − B1) / (B1+B2) with B1: 0.3 − −
10 keV. Thediscrepency between MOS1 and 2 in XMM-1 was already identifiedin Farrell et al. (2009). spectral extraction using the SAS task tabgtigen and cut-off count rates of 0.3 counts s − and 0.1 counts s − in theflare background light curves for the MOS and pn cam-eras respectively. This filtering resulted in net exposuretimes of ∼
98 ks and ∼
99 ks (for pn and MOS respec-tively). The same event filtering criteria as used for theproduction of pipeline products for the 2XMM cataloguewere used to produce light curves and spectra for eachcamera in each observation (e.g. single to double eventsfor the pn and single to quadruple events for the MOS,Watson et al. 2009).Source and background spectra were extracted for eachcamera for each observation, with response and ancillaryfiles generated in turn using the tasks rmfgen and arfgen .The source extraction region radii (reported in Table 1)were chosen so as to optimize the signal-to-noise based onthe detected count rates and off-axis position of HLX-1in each of the observations (V. Braito, private communi-cation). The background extraction regions were chosento be three times the area of the source extraction re-gions, so as to provide a robust estimate of the averagebackground level at the position of HLX-1. The spec-tra were grouped to have at least 20 counts per bin toprovide sufficient statistics for spectral analyses using χ statistic.Source light curves were extracted using events in theenergy range of 0.3–2 keV for the pn camera of all obser-vations. They were binned at the frame time, which is73.4 ms for XMM1 and is 5.9 ms for XMM2 and XMM3.Figure 2 shows the count rates for each camera as afunction of a hardness ratio. We see a discrepancy be-tween the XMM1 MOS1 and MOS2 hardness ratios, withthe MOS2 data showing a deficit in counts in the 0.3–1 keV energy band compared to both the MOS1 and pn data . We thus excluded energies below 0.5 keVfor MOS2. Despite this inconsistency between the MOShardness ratios in the XMM1 data, a clear trend is ob-served indicating significant spectral hardening at lowerluminosities, which is most clearly shown by the pn data(Figure 2, top panel). Chandra
ACIS-S data
We obtained a 10 ks observation with the AXAF CCDImaging Spectrometer (ACIS) onboard
Chandra on 2010September 6 (ObsID 13122, MJD 55445) in the director’sdiscretionary time program after the source showed a re-brightening in X-rays (Godet et al. 2010). In order tolimit pile-up, this observation was performed with theACIS-S3 chip only and using a 1/4 chip subarray with aframe time of 0.8 s, leading to 5% dead time (0.041 s).The ACIS-S3 chip was chosen for its higher sensitivity tosofter X-ray photons .The data were processed using CIAO to apply the latestcalibration files. We extracted the source spectrum using psextract with an extraction circle of radius 2 . ′′ ′′ and 50 ′′ respec-tively (thus excluding the host galaxy). We then ran thetasks mkacisrmf and mkarf to obtain the instrument re-sponse files as recommended for ACIS-S data. The spec-trum was binned with a minimum of 15 counts, and wediscarded bins below 0.3 keV where the response of theinstrument is not calibrated. HLX-1 AT HIGH LUMINOSITIES
Re-analysis of XMM1 and XMM2 spectra
Spectral fits of the XMM2 data with the relativisticmodel BHSPEC have been performed by Davis et al.(2011), and a more detailed analysis of all spectra withphysically motivated accretion disk models used for GB-HBs and ULXs is presented by Godet et al. (2011, inpreparation). We used here simple models to fit the X-ray spectra, e.g. an absorption on the line of sight, apower law model ( pow ), and a multi-temperature blackbody model ( diskbb , Makishima et al. 1986). We usedthe wabs absorption model (Morrison & McCammon1983). The Galactic absorption along the line of sightof ESO 243–49 is 1 . × atom cm − (Kalberla et al.2005).For XMM1, we fitted simultaneously the MOS1, MOS2and pn spectra using Xspec and ignored channels withenergy lower than 0.3 keV as well as channels tagged asbad. The spectra are well represented by a simple ab-sorbed steep power law of index 3 . +0 . − . giving a reduced χ of 0.93. The addition of a diskbb component improvedthe χ by 3.4. To test the significance of this compo-nent we used the Bayesian posterior predictive proba-bility values method (e.g. Protassov et al. 2002), as it This discrepancy had been noted earlier by Farrell et al. (2009),leading them to exclude energies below 0.5 keV for MOS2. http://cxc.harvard.edu/proposer/POG, Proposers’ Observa-tory Guide http://cxc.harvard.edu/ciao/ http://cxc.harvard.edu/ciao/threads/createL2 tate transitions of ESO 243–49 HLX-1 5 Table 2
Parameters of best fits for HLX-1 X-ray spectraObs N H20 kT in Γ χ /dof F [0 . −
10 keV] unabs. L [0 . −
10 keV] unabs. L disk (1) (2) (3) (4) (5) (6) (7) (8)Steep power law stateXMM1 10 ± · · · . +0 . − . . ± . . ± . · · · XMM1 5 ± . +0 . − . . +0 . − . . ± . . ± . . +0 . − . Thermal state
Chandra a +4 − . ± . · · · . ± . . ± . . ± . ± . ± .
01 2 . ± . . ± . . ± . . ± . . ± .
02 2 . +0 . − . · · · · · · · · · S3 3 0 . ± . · · · . ± . . ± . . ± . . ± . · · · . ± . . ± . . ± . +6 − · · · . +0 . − . . +0 . − . . +0 . − . · · · XMM3 62 +40 − . ± .
02 2 . +0 . − . . +0 . − . ∼ ∼ b · · · . ± . . ± .
03 0 . ± . · · · XMM3 b . ± .
04 1 . ± . . ± .
04 0 . ± .
04 0 . ± . · · · . ± . . ± . . ± . · · · Note . — Columns: (1) Observations; (2) Absorption on the line of sight, when no error, it is fixed to the bestestimate of 3 ± × atom cm − for XMM2; (3) Temperature of the diskbb model in keV and 90% error; (4)Photon index of the power law component and 90% error; (5) χ of the fit and degrees of freedom; (6) Absorbed fluxin the 0.2–10 keV energy range for the model, with 1 σ error in 10 − erg s − cm − ; (7) Unabsorbed luminosity in the0.2–10 keV energy range assuming a distance of 95 Mpc, with 1 σ error in 10 erg s − ; (8) Unabsorbed bolometricluminosity of the disk component assuming a distance of 95 Mpc, with 1 σ error in 10 erg s − . a The
Chandra spectrum was fitted with a pile-up model with a frame time of 0.8 s, alpha=0.99 and psffrac=0.95,which was then removed to estimate fluxes. b We added a mekal component at a fixed redshift of z = 0 . kT = 0 . ± .
09 keV. This component was removed before estimatingthe flux and luminosity of HLX-1, assuming it is due to the host galaxy. was already done by Farrell et al. (2009). We generated3000 simulated set of spectra with the Xspec fakeit com-mand using the absorbed power law model and the re-sponse files corresponding to the XMM1 spectra. Wethen computed the distribution of the χ improvementof the fit when adding a diskbb component. The χ im-provement of 3.4 for XMM1 translates to a significanceof 73% for the diskbb component, insufficient to claim itis real (Farrell et al. 2009 derived a significance of 70%with a similar method).For the XMM2 spectra, we included channels with en-ergy higher than 0.2 keV due to the higher signal to noiseat low energies. The spectra are poorly fitted with a sim-ple absorbed power law mode ( χ of 577.7 and 348 de-grees of freedom). When adding a diskbb component thereduced χ significantly dropped to an acceptable valueof 1.14, indicating that this model better represents thespectra.For XMM1 and XMM2, we thus obtained best fits inagreement with Farrell et al. (2009). We show the corre-sponding folded spectra in Figure 3. The calibration wasimproved in the low energy range with respect to thatused by Farrell et al. (2009) which lead to smaller errorbars in this work. We thus present new estimates of theparameters in Table 2. Chandra spectrum and pile-up
The spectrum was fitted in Xspec with an absorbed diskbb model which gave a reduced χ of 1.43 for 49degrees of freedom. A clear hard excess appeared inthe residuals that could correspond to the effect of pile-up (Davis 2001). We indeed estimated a possible level of pile-up of 5% with the Chandra
Proposal PlanningToolkit based on PIMMS v4.2. We thus added a pileup component (Davis 2001) to the model and obtained a sat-isfactory fit with a single absorbed diskbb model with areduced χ of 1.18 (see Figure 4 and Table 2). We let theparameter alpha free and found a value of 1 . +0 . − . . Thisparameter is related to the probability of events beingretained as a good grade after filtering and we would ex-pect a value between 0.5 and 0.7. We note that the valueof the diskbb temperature is not significantly affected bythis parameter. Large contributions to the χ seem tocome from features around 0.6 keV (emission) and 1.1and 1.3 keV (absorption). However, given the low num-ber of degrees of freedom, there is a large uncertainty onthe expectation of the reduced χ being 1 ( σ of 0.2 on theestimate of the reduced χ , see e.g. Andrae 2010), so wecannot claim that these features are real. We thereforeconsider the continuum well fitted by the diskbb model.We found that an additional power law component witha fixed photon index of 2 would contribute to at most10% of the 0.2–10 keV absorbed flux, indicating that thedisk component is dominant in the spectrum.The Chandra observation was simulated to better char-acterize the emission of HLX-1. We used the ray tracingtool ChaRT dedicated to
Chandra (Carter et al. 2003)and generated an image with MARX for the ACIS-S de-tector (Wise et al. 1997). The best fit spectral model ofthe source, obtained as explained above, was used as aninput in the energy range 0.3–8 keV. A simulated imageis presented in Figure 5. http://cxc.harvard.edu/toolkit/pimms.jsp Servillat et al. -5 -4 -3 -2 -1 c o un t s s − k e V − XMM1 abs(po)χ /dof=59.7/64 MOS1MOS2pn -5 -4 -3 -2 -1 c o un t s s − k e V − XMM2 abs(po+diskbb)χ /dof=395.6/346 Energy [keV] -6 -5 -4 -3 -2 c o un t s s − k e V − XMM3 abs(po)+abs(mekal)χ /dof=63.6/59 Figure 3.
XMM-Newton folded spectra of HLX-1 and best fitmodels. Data points from the three cameras are reported withtheir error bars and the model is indicated as a dashed line. Dottedlines show components of the model.
Figure 4.
Chandra
ACIS-S folded spectrum of HLX-1. Best fit(top) and residuals (bottom). See Table 2.
In the band 0.3–8 keV, the results from the simulationgave 1294 ±
36 counts. After running the pileup tool inMARX, we obtained 1182 ±
34 expected counts (9 ± ±
34 counts (14 ±
5% loss), which isconsistent within the errors with the MARX simulationwith pile-up effects. The PSF in the observation showsa deficit of counts in the central pixels compared to thesimulation without pile-up. It is however consistent with
Figure 5.
Chandra
ACIS-S images of HLX-1 in the 0.3-8 keV en-ergy band.
Top : data.
Bottom : result of simulation using ChaRTand MARX with pile-up included. Galaxy contours are overlaid(at H-band Vega mag arcsec − of 21, 18.5 and 16). A dashedblack 5 ′′ radius circle is shown around the bulge of the galaxy andsimilar grey circles are placed around the galaxy for comparison.The position of HLX-1 is indicated with its 0 . ′′ the image of the simulation with pile-up applied. We aretherefore confident that the use of the pile-up model inour spectral fitting is justified. Emission from the galaxy
No significant point source is detected inside the galaxycontours except HLX-1. We looked for an excess ofcounts in an extended 5 ′′ radius region encircling thebulge of the galaxy. In the 0.3–8 keV energy band, weexpect 5 ± ± ± ± ± . ± .
10 keV suggesting a soft emission. The emission fromthe bulge of spiral galaxies have previously been mod-eled by a mekal model with temperatures of 0.3 to 0.6keV (e.g. Humphrey et al. 2004). Assuming a mekal model (at the redshift of the galaxy, z = 0 . . × atom cm − ), we find a temperaturek T = 0 . ± .
15 and a flux of ∼ × − erg s − cm − in the 0.2-10 keV range. At the distance of the galaxy,this converts to an unabsorbed X-ray luminosity of ∼ × erg s − . This level of luminosity is consistentwith the integrated X-ray luminosity of spiral galaxies ingeneral (10 –10 erg s − , Fabbiano 1989). This emis-sion might contaminate XMM-Newton and
Swift
HLX-1data at low luminosities given their lower angular reso-tate transitions of ESO 243–49 HLX-1 7 −3−2−10123 χ χ /dof=85.2/60 abs(po) −3−2−10123 χ χ /dof=69.3/58 abs(po+diskbb) −3−2−10123 χ χ /dof=63.6/59 abs(po)+abs(mekal) Energy [keV] −3−2−10123 χ χ /dof=58.0/57 abs( o+diskbb)+abs(mekal) Figure 6.
XMM3 spectrum residuals for different models reportedin Table 2. lution. HLX-1 AT LOW LUMINOSITIES
XMM-Newton spectrum XMM3
We fitted simultaneously the MOS1, MOS2 and pnspectra using Xspec and ignored channels with energylower than 0.3 keV as well as channels tagged as bad. Weused an absorbed power law and obtained a reduced χ of 1.42 with 60 degrees of freedom (Table 2). The high-est contributions to the χ are associated with a possiblesoft excess around 0.7 keV and a hard excess at high en-ergies (Figure 6, first panel). We thus tested two possibleadditive components to the model to account for thosefeatures: (1) a diskbb component to test for the presenceof a thermal disk, and (2) a mekal component whichwould correspond to a possible contamination from thegalaxy bulge to the spectrum (see Section 3.3).When adding a diskbb component to the single ab-sorbed power law model, the reduced χ is loweredto 1.20. However, a marked soft excess appears inthe residuals and the hard excess is still present (Fig-ure 6, second panel). The best fit is found for an N H ∼ . ± . × atom cm − , an order of magnitudehigher than the XMM2 value and a lower temperature(k T in ∼ .
08 keV) than for the
Chandra data in the highstate. The bolometric unabsorbed luminosity of the diskcomponent is thus higher than in the high luminositystate (Section 3).The addition of an abs(mekal) component — at a fixedredshift of z = 0 . χ toan acceptable value of 1.08 and resulted in a lower pho-ton index for the power law (2 . ± . Figure 7.
XMM-Newton
MOS images of HLX-1 in the energyband 0.5–6 keV. The pixel size is 3 . ′′
2. The H-band galaxy contoursare shown as in Figure 5. The
XMM-Newton position error ( < . ′′ Chandra errorcircle (0 . ′′ is thawed, it tends to zero while the power law indexfurther decreases. The residuals are shown in Figure 6(third panel) and the folded spectrum is shown in Fig-ure 3. We found a temperature of kT = 0 . ± .
09 keVconsistent with the expected emission from the galaxy(Section 3.3). Moreover, the flux found for this absorbed mekal component is 3 . ± . × − erg s − cm − whichis comparable to the estimate of the possible galaxy con-tribution found in the Chandra data (Section 3.3). Weare thus confident that some emission from the galaxycontaminates the XMM3 spectrum of HLX-1, at a levelof ∼
17% in flux at most.Finally, we tested the presence of a diskbb componentin this last model. The resulting reduced χ is 1.02, thepower law is harder (1 . ± .
4) and the residuals seem fea-tureless for this fit (Figure 6, fourth panel). We testedthe significance of the diskbb using the Bayesian poste-rior predictive probability values method. In the sameway as in Section 3.1, we generated 3000 simulated setof spectra based on the absorbed power law plus mekal model and the XMM3 response files. We then computedthe distribution of the χ improvement of the fit whenadding a diskbb component. The χ improvement of 5.6for XMM3 translates to a significance of 93% for the diskbb component. We note that a diskbb component isnot needed to adequately describe the spectrum, but thispossibility is intriguing and cannot be ruled out. Astrometry consistency
We aligned all images on the
Chandra image, taking asa reference 3 point-like X-ray sources surrounding HLX-1. This allowed us to reduce the positional error byreplacing the
XMM-Newton
Absolute Measurement Ac-curacy of 4 ′′ by a residual error of less than 1 ′′ after see § XMM-Newton
Users Handbook, http://xmm.esac.esa.int/external/xmm_user_support/documentation/uhb/index.html
Servillat et al.alignment. We still have to take into account the Rela-tive Pointing Error of 2 ′′ and the error on the centroidof the source, which yielded an uncertainty lower thanthe size of the pixels (3 . ′′
2) for all the images presentedin Figure 7. We also note that the relative astrometrywithin all EPIC cameras is accurate to better than 1 . ′′ Chandra data (Sec-tion 3.3), since the bulge diffuse emission should be con-stant. The source is slightly offset from the
Chandra position which falls just outside the 95% error circle. Itis shifted towards the center of the galaxy ESO 243–49,indeed suggesting a contamination from the bulge of thegalaxy. However, the source detection task emldetect didnot split the source into two sources, indicating that thestatistic is not sufficient to claim the presence of a con-taminating source in the imaging data such as the galaxybulge. The PSF full width half maximum (FWHM) is 5 ′′ and 6 ′′ for MOS and pn, respectively, and the half en-ergy width (HEW) is 14 ′′ and 15 ′′ . The distance betweenHLX-1 and the galaxy center is about 8 ′′ so the imagescannot be used as evidence for contamination. HARDNESS-INTENSITY AND HARDNESS-RMSDIAGRAMS
Swift
XRT spectra
We completed the study of the
Chandra and
XMM-Newton spectra with the spectral fitting of grouped spec-tra from
Swift
XRT (Figure 1). We fitted each
Swift
XRT spectrum using Xspec. For all spectra, given thereduced statistical quality, we performed a basic fitting ofthe data using the XMM2 model (i.e. abs(diskbb+pow) ,see Table 2) multiplied by a constant factor in orderto estimate the unabsorbed luminosity. For each spec-trum we estimated the hardness as the ratio of the countrates in the 1–10 and 0.3–1 keV bands. We then usedWebPIMMS to convert those ratios into flux ratios: inthe low and high states (power law with photon index ∼ . ∼ .
0, respectively) we found a conversion fac-tor of 1.5 and 0.9, respectively. We also used the follow-ing model: abs (C1 × diskbb + C2 × pow ) where C1 and C2are two constants. The spectral parameters of the diskbb and power law models were fixed at the values found inXMM2. The goal is to investigate the evolution in thecontribution between the soft and hard components. Allresults are reported in Table 3.For the S3 and S7 spectra (high luminosity states, seeFigure 1), we used a simple absorbed diskbb model andreported the parameters of the fits in Table 2. There isno evidence in these spectra for the need of a hard energycomponent. We note that the temperature for S7 is con-sistent with that derived from the Chandra observationwhich was performed at a similar time. The normaliza-tion is lower for the
Chandra spectrum, but the fluxesfor both spectra remain consistent within the errors.The basic spectral fitting suggests that the source is inthe same state in S1, S4 and S8. We merged the datafrom S1, S4 and S8 that approximately overlap in lumi-nosity assuming a ∼
375 days periodicity of the source. Inorder to get a good fit, it is necessary to add a hard com- http://heasarc.nasa.gov/Tools/w3pimms.html Figure 8.
The PDS of
XMM-Newton observations of HLX-1 usingthe pn camera. The black solid constant lines are the average PDSabove 1 Hz, representing the Poisson level. ponent (power law) to the model. Results are reportedin Table 2. The resulting parameters are consistent withthose found for XMM2 at a similar level of luminosity.Finally, we combined S2 and S6 to increase the signal-to-noise of the
Swift
XRT data in the low luminositystate. We obtained a good fit with an absorbed powerlaw (Γ = 2 . ± .
4, Table 2). We note that the flux ishigher than for the XMM3 data. However, as the S2+S6spectrum is from data obtained at different epochs andintegrated over a long time, the comparison with XMM3is not straightforward and differences in the two spectraare expected. The S2+S6 spectrum may also be affectedby contamination from the galaxy bulge (Section 3.3).
Timing analysis
We calculated the power density spectra (PDS) for thelight curves of HLX-1 from the three
XMM-Newton ob-tate transitions of ESO 243–49 HLX-1 9
Table 3
Fit results for
Swift
XRT spectraName Start End Hardness Count rate Unabs. L X χ /dof C1 C2 χ /dof(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)S1 54763 54784 0 . ± .
04 1 . ± .
06 6 . ± . . ± . . ± . . ± . . ± .
09 0 . ± . ∗ < .
09 0 . ± . ∗ S3 55059 55063 0 . ± .
03 3 . ± .
13 11 ± . / · · · · · · · · · S4 55106 55164 0 . ± .
03 1 . ± .
09 9 . ± . . ± . . ± . . ± .
05 1 . ± .
06 5 . ± . . ± . . ± . . ± . ± .
03 0 . ± . ∗ . +0 . − . ∗ S7 55437 55453 0 . ± .
02 2 . ± .
10 11 . ± . · · · · · · · · · S8 55462 55490 0 . ± .
03 2 . ± .
10 9 . ± . · · · · · · · · · S9 55498 55543 0 . ± .
04 1 . ± .
09 8 . ± . . ± . . ± . . ± .
04 0 . ± . . ± . . ± .
14 1 . ± . Note . — Columns: (1,2,3) Name, start and end day (MJD) of observation as reported in Figure 1; (4) Hardnessis given as the ratio of fluxes F [1 −
10 keV] /F [0 . − − count s − ) in the band 0.3–10 keV; (6) Unabsorbed luminosity in 10 erg s − after fitting the XMM2 model multiplied by a constant; (7) χ and degrees of freedom (dof) for the fit; (8,9,10) Result of the fit with the two XMM2 components multipliedby factors C1 and C2, and goodness of the fit. ∗ Cash statistic was used and the number of PHA bins is indicated instead of dof. servations. The light curves were split into four segmentsof equal lengths, and the PDS were calculated for eachsegment. For each light curve, all four PDS were mergedand averaged by binning in frequency using a logarithmicfactor 1.1, under the condition that each bin contains atleast 20 individual PDS measurements. The errors werecalculated from the sample standard deviation of PDSmeasurements in each bin.The PDS of HLX-1 are shown in Figure 8. The blacksolid lines denote the average PDS above 1 Hz. Thosevalues deviate from the expected Poisson level by lessthan 0.4%. We see that all PDS are flat, showing nosignificant intrinsic source variability, and no QPO. Thefit results using a constant C P are given in Table 4. Toobtain a constraint on the variability, we followed theprocedure adopted by Goad et al. (2006) and Heil et al.(2009). In this procedure, the PDS are fitted with abroken power-law (BPL) model or a Lorentzian modelplus a constant.For the BPL, the indices below and above the breakfrequency ( f b ) were assumed to be − −
2, respec-tively. The value C /f corresponds to the power timesthe frequency below f b from this model. The upper limitsof C /f for two values of f b (10 − and 1 Hz) are shownin Table 4. We found that these limits are relativelyindependent of the assumed f b value, at least withinthe 10 − –1 Hz range. Similarly, it is found to be fairlyconstant across AGN and GBHBs in the range 0.005–0.03 (Papadakis 2004). The upper limits of C /f forXMM1 are consistent with the values generally observedfor AGN/GBHBs in the thermal state. For XMM3, theyare larger and thus not constraining. In XMM2, the up-per limits of C /f are low compared with typical valuesfor AGN and GBHBs, in the range 0.005–0.03.For the Lorentzian model, we assumed the qualityfactor to be 2 as in Goad et al. (2006) and Heil et al.(2009). The upper limits of the integrated power of theLorentzian R are given in Table 4 for two Lorentziancentroid frequencies: 10 − and 1 Hz. They roughly in-crease with the centroid frequency assumed. Comparedwith the typical values of 0.01 for R seen in the GBHBhard state (van der Klis 2006), the upper limits we ob-tained for HLX-1 are high for all three observations. Due to poor statistics, we thus cannot place any firm con-straints on the spectral state using the timing analysis. Combined diagrams
We show the best fit models for all the fitted spec-tra in Figure 9, which illustrates the evolution of thesoft disk emission and the hard tail during spectral statetransitions. For each
Swift , XMM-Newton and
Chandra spectra, we estimated the unabsorbed luminosity in therange 0.2–10 keV for the best fit, and calculated the fluxratio between the bands 0.3–1 and 1–10 keV. Those en-ergy ranges were also used by Godet et al. (2009), so theHID shown in Figure 10 is an updated version of theirFigure 2.We plotted in this diagram values coming from dif-ferent instruments with independent calibrations. Therecould be systematic errors in the different calibrations,leading to a discrepancy of up to 20% in the normal-ization of the flux as it could be observed during simul-taneous observations (e.g. Tsujimoto et al. 2011). Weindeed observed a difference between the
Swift
XRT S7and
Chandra
ACIS-S spectra of HLX-1 (Figure 9) takenover a similar period. However, for these two spectra,hardness ratio and luminosity estimates are consistentwithin this range of errors. Consistent trends are seen inboth the
Swift and
XMM-Newton data, we are thus con-fident that the spectral variations detected in this workare real and significant.We also computed variability rms values and limits inthe 0.0001–0.1 Hz range. They are reported in Table 4and plotted on the HRD in Figure 10. DISCUSSION
Spectral states and transitions
The high luminosity state of HLX-1 is associated witha soft emission with a dominant thermal disk emission,while the low luminosity state is found to be significantlyharder and power-law-like. This is reminiscent of the be-havior of GBHBs (e.g. Remillard & McClintock 2006),and the HID and HRD shown in Figure 10 are con-sistent with similar diagrams for GBHBs (e.g. Belloni2010). HLX-1 therefore clearly showed transitions fromone state to another, confirming the first evidence found0 Servillat et al.
Table 4
Results of the PDS FittingObs C P χ ν ( ν ) C /f (10 − , BPL) R (10 − , BLN) rms(%)10 − Hz 1 Hz 10 − Hz 1 Hz(1) (2) (3) (4) (5) (6) (7) (8)XMM1 46.11 ± < < < < < . ± < < < < . +9 . − . XMM3 898.11 ± < < < < < Note . — Columns: (1) Observations; (2) Best-fitting constant Poisson level in[rms/mean] Hz − ; (3) Reduced χ and degrees of freedom of the fits with a constantPoisson level; (4) Best-fitting C /f assuming a break frequency of 10 − Hz; (5) Best-fitting C /f assuming a break frequency of 1 Hz; (6) Best-fitting R assuming a Lorentzian cen-troid frequency of 10 − Hz; (7) Best-fitting R assuming a Lorentzian centroid frequencyof 1 Hz; (8) Fractional rms variability within 0.0001–0.1 Hz, assuming the Poisson levelto be the average PDS above 1 Hz. All limits are at a 90% confidence level, except forthe fractional rms (column 8), which is 1 σ . Energy [keV] -6 -5 -4 k e V ( P h o t o n s c m − s − k e V − ) S3S7 S2+S6 XMM1 XMM2XMM3Chandra
Figure 9.
Models of X-ray spectra of HLX-1 in different states asobtained from the best fit to the data (see Table 2). For XMM3,the abs(pow)+abs(mekal) model is shown as a dashed line, and themodel without the mekal component as a thick line. For Chandra,the effect of pile-up has been corrected. by Godet et al. (2009). However, fewer luminosity stateshave been observed and therefore the diagrams are notequally sampled than for GBHBs.The high luminosity state of HLX-1 is well describedby an optically thick disk model with temperatures of0.18 to 0.26 keV (Table 2). Moreover, the low variabil-ity level in the XMM2 data ( ∼ ∼ T in ∝ M − / (Makishima et al. 2000). It is thus possible U n a b s . L u m i n o s i t y [ . − k e V ] ( e r g s − ) S1 S2S3S4 S5 S6S7S8 S9 S10 XMM1XMM2 XMM3 XMM3 b Chandra
Swift XRTXMM-NewtonChandra
S1 S2S3S4 S5 S6S7S8 S9S100.2 0.5 1.0 2.0 3.0
Flux [1−10 keV] / Flux [0.3−1 keV] R M S v a r i a b ili t y ( % ) [ . − . H z ] XMM1XMM2 XMM3 XMM3 b Figure 10.
Hardness-intensity (top) and hardness-rms diagrams(bottom) for HLX-1. This plot includes data from
Swift , XMM-Newton and
Chandra . The hardness ratios and luminosities aremodel dependent, and were obtained from the best fit of the spec-tra or using a conversion factor for the
Swift
XRT. XMM3 b cor-responds to the model with the mekal component removed (seeSection 4). We show part of the HID with linear axes as an insetto better follow the evolution of the source in the high state. that the lower temperature is due to the presence of ablack hole with a higher mass — by a factor of ∼
200 to500 — than a typical stellar-mass black hole ( ∼ M ⊙ ),leading to a possible mass of few 10 M ⊙ . This argu-ment has already been used by Miller et al. (2004) for 6bright ULXs ( > erg s − ) to strengthen their classi-fication as IMBH candidates even if the thermal compo-nent was not dominant. In our observations of HLX-1 inthe thermal state, this component is clearly dominant.A complementary study of HLX-1 in the thermal stateby Davis et al. (2011) using the relativistic disk modeltate transitions of ESO 243–49 HLX-1 11 Disk temperature kT in [keV] D i s k un a b s o r b e d l u m i n o s i t y L d i s k [ e r g s − ] S3S7ChandraXMM2XMM3 L d i s k ∝ T i n Figure 11.
Disk bolometric unabsorbed luminosity as a functionof disk temperature kT in (see values in Table 2). The best fit witha fixed L disk ∝ T is indicated with a dashed line. The XMM3point should be taken with caution as the disk component cannotbe confirmed nor ruled out in the X-ray spectrum. BHSPEC provided robust lower and upper limits with3000 < M < × M ⊙ , firmly placing HLX-1 in theIMBH regime. Godet et al. (2011, in preparation) testedadditional models to physically constrain the nature ofHLX-1 and they also found mass estimates in the IMBHrange.We report in Figure 11 the temperature and bolo-metric luminosity of the disk in the thermal statefrom the S3, S7, Chandra , XMM2 observations (seealso Godet et al. 2011, in preparation). All thosemeasurements in the thermal state are consistent withthe correlation between luminosity and temperature ex-pected from a geometrically thin, optically thick accre-tion disk ( L ∝ T , Shakura & Sunyaev 1973). Thissupports the idea of HLX-1 being in a generic ther-mal state as this relation is also observed for GB-HBs in this state (e.g. Kubota & Makishima 2004;Remillard & McClintock 2006). It is however differentto the ULX branch reported e.g. by Soria (2007), wherean anti-correlation between disk luminosity and temper-ature is seen for two ULXs.During the XMM3 observation at low luminosities, thespectrum of HLX-1 is mainly described by a power lawmodel with a photon index that falls in the range 1.4to 2.1 (Table 2), indicating a hard state as observed forGBHBs (we presented preliminary results in Farrell et al.2011). We could not constrain the rms variability of thesource during the XMM3 observation due to low statis-tics. Therefore, a high rms value of ∼ kT in = 0 . ± .
04 keV,with a significance of 93%. Such a disk component wouldbe consistent with the L ∝ T relation reported in Fig-ure 11, as the expected temperature for a disk unab-sorbed luminosity of 4 × erg s − is 0.09 keV. Adisk could thus be either extending down to the ISCO,or maybe truncated as the best fit temperature value islower than the temperature expected from the L ∝ T relation. This would again suggest a resemblance withGBHBs.During the XMM1 observation of HLX-1, the spec-trum is found to be well fitted with a steep power lawof photon index ∼ ∼
10 days or less), followed by an exponential decay (100to 200 days) of the X-ray flux from HLX-1 (FRED-likeoutburst). This could be explained by theoretical mod-els of the hydrogen ionization instability controlling thefast rise (King & Ritter 1998). The cause of this variabil-ity has been investigated in more details by Lasota et al.(2011). They concluded that HLX-1 is unlikely to be ex-plained by a model in which outbursts are triggered bythermal-viscous instabilities in an accretion disc (Lasota2001), and they argue that a more likely explanation isa modulated mass-transfer due to tidal stripping of astar in an eccentric orbit around the IMBH. This wouldstill explain a change in the accretion rate, leading tothe state transitions we observe, and might also explainthe rarity of a source like HLX-1, which shows uniqueproperties for a ULX.
Comparison with other ULXs
It was suggested that most ULXs (few 10 to10 erg s − , with power law spectra) are stellar-masssystems accreting at Eddington ratios of the order of10–30, with mild beaming factors b & . erg s − (Fabrika et al. 2007). In the case of HLX-1 in the higheststate, the Eddington ratio would reach 170 and the beam-ing factor b ∼ . × − (King 2011). Freeland et al.(2006) calculated the broad-band radio–X-ray spectrapredicted by micro-blazar and micro-quasar models forULXs. They argued that a disk and a jet could be presentin the system in a high/hard state close to the Eddingtonluminosity with a high accretion rate and gravitationalenergy released all the way down to the ISCO, leadingto a hard spectrum (Γ=1.4–2.1) at high observed lumi-nosities ( > × erg s − ). They also proposed theexistence of milli-blazars (IMBH and beamed emission)with luminosities > erg s − . The spectrum of HLX-2 Servillat et al.1 in the low luminosity state might be explained by sucha model. However, the transition to a thermal state ateven higher luminosities is not consistent with this pic-ture. The increase in luminosity and softening of thesource (decrease of the peak energy as can be seen inFigure 9) are not consistent with relativistic beamingvariations which would have an opposite effect (e.g. King2009). All this indicates that the spectral variability weobserve for HLX-1 is not due to beaming variations.Gladstone et al. (2009) showed that simple spectralmodels commonly used for the analysis and interpreta-tion of some ULXs (power law continuum and multi-color disk black body models) are inadequate for a smallsample (12 ULXs) of nearby, low luminosity ULXs thathave high quality spectra. Two near ubiquitous featuresare found in the spectra: a soft excess and a rollover inthe spectrum at energies above 3 keV. They suggestedthe existence of a new ultra-luminous state with super-Eddington accretion flows. This would favor the presenceof stellar-mass black holes rather than IMBHs for someULXs. In the spectra of HLX-1, which reaches lumi-nosities two orders of magnitude higher than the averageULX, we found no evidence of such a break above 3 keV.However, the number of counts, even in XMM2, is prob-ably insufficient to detect this feature (Gladstone et al.2009 claim the need of 10 000 counts). Godet et al. (2011,in preparation) discuss in more details the comparisonwith ULXs claimed to be in a ultra-luminous state, butthe variety of X-ray spectra we observe for HLX-1 in thiswork, all in the ultra-luminous range, is clearly uniquefor a ULX. In particular, the thermal state observed athigh luminosities (therefore probably at the highest ac-cretion rate) for HLX-1 is clearly distinct to the proposedultra-luminous state spectrum for some ULXs.Finally, the X-ray observations reported in this workindicate that HLX-1 more resembles GBHBs scaled tohigher luminosities (three orders of magnitude) than anyother category of ULXs. In this picture, the bolometricluminosity (hence the X-ray luminosity) of the source inthe thermal state should be lower than the Eddington lu-minosity for the system, giving a limit on the mass of theblack hole of > M ⊙ . The unabsorbed luminosity inthe hard state is 2.2% of the highest luminosity recordedin this work for which we assume a sub-Eddington lumi-nosity (see Table 2). This level of luminosity is consistentwith the fraction of Eddington luminosity observed forGBHBs in the hard state ( < M ⊙ . CONCLUSION
We observed the bright ULX ESO 243–49 HLX-1 withdifferent X-ray observatories at luminosities ranging from2 . × to 1 . × erg s − . The spectral analy-sis revealed that HLX-1 showed three states: a thermalstate at high luminosity, a hard state at low luminosi-ties, and a steep power law state. These states are typ-ically observed for GBHBs, which have luminosities atleast three orders of magnitude lower than HLX-1. The clear spectral changes we observed are inconsistent withmodels of milli-blazars where the emission is stronglybeamed toward us, and the spectra of HLX-1 does notresemble spectra of ULXs proposed to be in an ultra-luminous state with super-Eddington accretion. Instead,the source is much better explained as being homologousto GBHBs, which have non-beamed and sub-Eddingtonemission. The relatively low temperature of HLX-1 inthe thermal state suggests the presence of an IMBH offew 10 M ⊙ . Finally, in this picture, the source lumi-nosity should be lower than the Eddington luminosity,which converts to a lower limit on the mass of the blackhole of > M ⊙ , placing HLX-1 in the IMBH range.We are grateful for the comments of the referee whichhelped us improve and strengthen the paper. Wethank Harvey Tananbaum and the Chandra team forapproving the Chandra DDT observations. MS thanksJosh Grindlay and Jeff McClintock for helpful discus-sions. MS acknowledges supports from NASA/Chandragrants GO0-11063X, DD0-11050X and NSF grant AST-0909073. SAF acknowledges funding from the UK Sci-ence and Technology Funding Council and the AustralianResearch Council. SAF is the recipient of an AustralianResearch Council Post Doctoral Fellowship, funded bygrant DP110102889. Based on observations from XMM-Newton, an ESA science mission with instruments andcontributions directly funded by ESA Member States andNASA. This research has made use of data obtained fromthe Chandra Data Archive and software provided by theChandra X-ray Center (CXC) in the application pack-ages CIAO, ChIPS, and Sherpa. This work made use ofdata supplied by the UK Swift Science Data Centre atthe University of Leicester. Facilities:
XMM, CXO, Swift (XRT)
REFERENCESAndrae, R. 2010, arXiv, 1009, 2755Arnaud, K. A. 1996, ADASS V, ASP Conference Series, 101, 17Begelman, M. C. 2002, ApJ, 568, L97Belloni, T. M. 2010, In: The Jet Paradigm, Lecture Notes inPhysics, Berlin Springer Verlag, 794, 53, iSBN978-3-540-76936-1Carter, C., Karovska, M., Jerius, D., Glotfelty, K., & Beikman, S.2003, ADASS XII, ASP Conference Series, 295, 477Casella, P., Ponti, G., Patruno, A., Belloni, T., Miniutti, G., &Zampieri, L. 2008, MNRAS, 387, 1707Cash, W. 1979, ApJ, 228, 939Davis, J. E. 2001, ApJ, 562, 575Davis, S. W., Narayan, R., Zhu, Y., Barret, D., Farrell, S. A.,Godet, O., Servillat, M., & Webb, N. A. 2011, arXiv, 1104,2614, accepted for publication in ApJDone, C., Gierli´nski, M., & Kubota, A. 2007, A&A Rev., 15, 1Evans, P. A., et al. 2007, Astronomy and Astrophysics, 469, 379—. 2009, MNRAS, 397, 1177Fabbiano, G. 1989, ARA&A, 27, 87Fabbiano, G., King, A. R., Zezas, A., Ponman, T. J., Rots, A., &Schweizer, F. 2003, ApJ, 591, 843Fabrika, S. N., Abolmasov, P. K., & Karpov, S. 2007, In: BlackHoles from Stars to Galaxies – Across the Range of Masses.Edited by V. Karas and G. Matt. Proceedings of IAUSymposium tate transitions of ESO 243–49 HLX-1 13tate transitions of ESO 243–49 HLX-1 13