当前位置:首页 >> >>

XMM-Newton observations of the brightest Ultraluminous X-ray sources

Mon. Not. R. Astron. Soc. 000, 1–16 (2004)

Printed 5 February 2008

A (MN L TEX style ?le v2.2)

XMM-Newton observations of the brightest Ultraluminous X-ray sources
A-M. Stobbart 1 , T.P. Roberts 1 and J. Wilms 2 1
2 Department

X-ray & Observational Astronomy Group, Dept. of Physics & Astronomy, University of Leicester, Leicester LE1 7RH, U.K. of Physics, University of Warwick, Coventry, CV4 7AL

arXiv:astro-ph/0601651v1 27 Jan 2006



We present an analysis of 13 of the best quality Ultraluminous X-ray source (ULX) datasets available from XMM-Newton European Photon Imaging Camera (EPIC) observations. We utilise the high signal-to-noise in these ULX spectra to investigate the best descriptions of their spectral shape in the 0.3–10 keV range. Simple models of an absorbed power-law or multicolour disc blackbody prove inadequate at describing the spectra. Better ?ts are found using a combination of these two components, with both variants of this model - a cool (? 0.2 keV) disc blackbody plus hard power-law continuum, and a soft power-law continuum, dominant at low energies, plus a warm (? 1.7 keV) disc blackbody - providing good ?ts to 8/13 ULX spectra. However, by examining the data above 2 keV, we ?nd evidence for curvature in the majority of datasets (8/13 with at least marginal detections), inconsistent with the dominance of a power-law in this regime. In fact, the most successful empirical description of the spectra proved to be a combination of a cool (? 0.2 keV) classic blackbody spectrum, plus a warm disc blackbody, that ?ts acceptably to 10/13 ULXs. The best overall ?ts are provided by a physically self-consistent accretion disc plus Comptonised corona model (DISKPN + EQPAIR), which ?ts acceptably to 11/13 ULXs. This model provides a physical explanation for the spectral curvature, namely that it originates in an optically-thick corona, though the accretion disc photons seeding this corona still originate in an apparently cool disc. We note similarities between this ?t and models of Galactic black hole binaries at high accretion rates, most notably the model of Done & Kubota (2005). In this scenario the inner-disc and corona become energetically-coupled at high accretion rates, resulting in a cooled accretion disc and optically-thick corona. We conclude that this analysis of the best spectral data for ULXs shows it to be plausible that the majority of the population are high accretion rate stellar-mass (perhaps up to 80-M⊙ ) black holes, though we cannot categorically rule out the presence of larger, ? 1000-M⊙ intermediate-mass black holes (IMBHs) in individual sources with the current X-ray data. Key words: accretion, accretion discs – black hole physics – X-rays: binaries – X-rays: galaxies

1 INTRODUCTION Einstein X-ray observations were the ?rst to reveal point-like, extranuclear sources in some nearby galaxies with luminosities in excess of 1039 erg s?1 (Fabbiano 1989). Subsequently, many of these so-called Ultraluminous X-ray sources (ULXs) have displayed short and long term variability, which suggests they are predominantly accreting objects (see Miller & Colbert 2004 and references therein). However, the observed luminosities of most ULXs exceed the Eddington limit for spherical accretion onto a stellar-mass (?10-M⊙ ) black hole (BH). In fact, their luminosites are intermediate between those of normal stellar mass BH Xray Binaries (BHBs) and Active Galactic Nuclei (AGN). Therec 2004 RAS

fore, the accretion of matter onto intermediate-mass black holes (IMBHs, of ?102 –104 M⊙ ) provide an attractive explanation for ULXs, and could represent the long sought-after ‘missing link’ between stellar mass BHs and the supermassive BHs in the nuclei of galaxies. However, the large populations of ULXs associated with sites of active star formation (e.g., in the Cartwheel galaxy, Gao et al. 2003) demand rather too high formation rates of IMBHs if they are to explain the ULX class as a whole (King 2004). An alternative to accreting IMBHs is that ULXs may be a type of stellar-mass BHB with geometrically (King et al. 2001) or relativistically (K¨ rding, Falcke, & Markoff 2002) beamed emiso sion, such that their intrinsic X-ray luminosity does not exceed the


A-M. Stobbart et al.
BH mass, MXR , from the following equation (cf. Makishima et al. 2000 equations (5)–(8)). MXR = ξκ2 D √ 8.86α cosi fbol M⊙ 4 2σTin (1)

Eddington limit. Another possibility is that they are stellar-mass BHBs that can achieve truly super Eddington luminosities via slim (Ebisawa et al. 2003) or radiation pressure dominated (Begelman 2002) accretion discs. As ULXs are probably the brightest class of X-ray binary fueled by the accretion of matter onto a BH1 , a knowledge of the properties of Galactic BHBs could be vital in interpreting their characteristics. Traditionally the X-ray spectra of BHBs have been ?tted empirically with two components, namely a powerlaw continuum and a multicolour disc blackbody (MCD) component (Mitsuda et al. 1984; Makishima et al. 1986). In the standard picture, the power-law component is thought to represent inverseCompton scattering of thermal photons from the accretion disc by hot electrons in a surrounding corona. As such, the power-law represents the hard tail of the X-ray emission while the MCD component models the soft X-ray emission from the accretion disc. The MCD model itself has been formulated based on the best known model for accretion onto BHs (i.e., the thin accretion disc model, Shakura & Sunyaev 1973). It has long been recognised that Galactic BHBs demonstrate various X-ray spectral states which are de?ned by the balance of these two components (i.e., power-law and MCD) at any one time. The three most familiar X-ray bright states are the low/hard (LH), high/soft (HS, also described as ‘thermal dominated’) and the very high (VH, or ‘steep power-law’) states (see McClintock & Remillard 2003 for further details). At lower mass accretion rates, a BHB usually enters the LH state where their X-ray emission is dominated by a hard power-law component (Γ?1.7), thought to arise from Comptonisation of soft photons by a hot optically thin corona2 . In this state, the disc is either undetected (e.g., Belloni et al. 1999) or appears truncated at a much larger inner radius and hence cooler than the parameters derived for the soft state (Wilms et al. 1999, McClintock et al. 2001). The soft X-ray state is generally seen at a higher luminosity (i.e., the HS state) and is best explained as ?1 keV thermal emission from a multitemperature accretion disc (i.e., modelled with a MCD component). In this state, the spectrum may also display a hard tail that contributes a small percentage of the total ?ux. The VH state is in many cases the most luminous state and is characterised by an unbroken power-law spectrum extending out to a few hundred keV or more. The photon index is typically steeper ( 2.5) than found in the LH state and generally coincides with the onset of strong X-ray quasi-periodic oscillations (QPOs). A MCD component may also be present in the VH state and the EXOSAT era demonstrated that some of the QPOs occur when both disc and power-law components contribute substantial luminosity (van der Klis 1995). The idea of ULXs as analogues to Galactic BHBs in the HS state was supported by ASCA observations, which revealed that their 0.5–10 keV spectra were successfully ?tted with the MCD model with relatively high disc temperatures (1.0–1.8 keV, Makishima et al. 2000). As such, the ULXs were considered to be mass-accreting BHs with the X-ray emission originating in an optically-thick accretion disc. In fact, the use of the MCD model to describe these spectra permits one to obtain an ‘X-ray–estimated’

Although the most likely reservoir of fuel for an ULX is a companion star, others have been suggested, for example the direct accretion of matter from molecular clouds (Krolik 2004). 2 However this is still a topic of debate, with the main alternative for the X-ray power-law emission being synchrotron emission from the radio jet that is associated with this state (e.g., Falcke & Biermann 1995; Markoff, Falcke, & Fender 2001).

Where D is the distance to the X-ray source, which has an inclination i, a full bolometric luminosity (from the MCD model) of fbol and an observed maximum disc colour temperature Tin . In addition σ is the Stefan-Boltzmann constant, κ is the ratio of the colour temperature to the effective temperature (‘spectral hardening factor’), and ξ is a correction factor re?ecting the fact that Tin occurs at a radius somewhat larger than Rin (here, we assume that Rin is at the last stable Keplerian orbit). Makishima et al. (2000) use values of ξ = 0.412 and κ = 1.7, though other work has found different values for the spectral hardening factor (e.g. κ = 2.6 for GRO J1655-40, Schrader & Titarchuk 2003). Finally, α is a positive parameter with α = 1 corresponding to a Schwarzschild BH. However, the masses inferred from the ASCA data and Equation (1) are far too low to be compatible with the large BH masses suggested by their luminosities (assuming Eddington-limited accretion), for standard accretion discs around Schwarzchild BHs. Makishima et al. (2000) suggested that this incompatibility could be explained if the BHs were in the Kerr metric (i.e., rapidly rotating objects), allowing smaller inner disc radii and hence higher disc temperatures. Chandra observations have provided some support for the Makishima et al. (2000) results, with some ULX spectra being consistent with the MCD model (e.g., Roberts et al. 2002). However, Chandra also revealed that some ULX spectra showed a preference for a power-law continuum rather than the MCD model (e.g., Strickland et al. 2001; Roberts et al. 2004; Terashima & Wilson 2004). It has been suggested that this preference for a power-law spectrum could be interpreted in terms of the LH state seen in Galactic BHB candidates, relativistically beamed jets or emission from a Comptonised accretion disc in the VH state. As well as these single component models, ASCA and Chandra spectroscopy have also suggested the presence of two component spectra for some ULXs, comprising a MCD with a power-law component. For example, previous ASCA analyses hinted at evidence for IMBHs, i.e., cool accretion disc components (see below), but these observations were not sensitive enough to statistically require two component modelling (e.g., Colbert & Mushotzky 1999). Similar results have been obtained by Chandra, e.g., ULXs in NGC 5408 (Kaaret et al. 2003) and NGC 6946 (Roberts & Colbert 2003). Conversely, Chandra spectra of the Antennae ULXs (Zezas et al. 2002a; Zezas et al. 2002b) revealed an accretion disc (MCD) component consistent with the high temperature ASCA results (i.e., kTin ?1 keV), together with a hard power-law component (Γ?1.2). It is only recently, using high quality XMM-Newton/EPIC spectroscopy of ULXs, that it has been demonstrated that the addition of a soft thermal disc component to a power-law continuum spectrum provides a strong statistical improvement to the best ?tting models to ULX data (e.g., Miller et al. 2003; Miller, Fabian, & Miller 2004a). These particular observations have provided strong support for the IMBH hypothesis by revealing disc temperatures in these sources up to 10 times lower than commonly measured in stellar mass BHBs, consistent with the expectation for the accretion disc around a ?1000-M⊙ IMBH3 (cf.

It is common for the generic range of masses for IMBHs to be quoted c 2004 RAS, MNRAS 000, 1–16

XMM-Newton observations of the brightest ULXs
Equation (1)). However, in a few cases, XMM-Newton/EPIC observations have revealed a more unusual two-component X-ray spectrum. A detailed analysis of such a source is presented in Stobbart, Roberts, & Warwick (2004). In this case, the ULX is the brightest X-ray source in the nearby (1.78 Mpc) Magellanic-type galaxy NGC 55 and, although its X-ray luminosity only marginally exceeds 1039 erg s?1 , it represents one of the highest quality ULX datasets obtained to date. The initial low ?ux state data were best ?tted with an absorbed power-law continuum (Γ?4), while a subsequent ?ux increase was almost entirely due to an additional contribution at energies > 1 keV, adequately modelled by a MCD component (kTin ?0.9 keV). Whilst this accretion disc component is reasonable for stellarmass BHs, the dominance of the power-law continuum at soft Xray energies is problematic. Such a soft power-law cannot represent Comptonised emission from a hot corona, as one would not expect to see the coronal component extend down below the peak emissivity of the accretion disc, where there would be insuf?cient photons to seed the corona. Alternative sources of seed photons for the corona are unlikely; for example, the incident photon ?ux of the secondary star at the inner regions of the accretion disc is too low to provide the seeding (cf. Roberts et al. 2005). It also seems unlikely that the power-law emission could arise from processes at the base of a jet, as these are typically represented by much harder photon indices than measured here (Γ?1.5–2; Markoff et al. 2005 and references therein). Indeed, with the possible exception of NGC 5408 X-1 (Kaaret et al. 2003), there is no evidence that ULXs do display bright radio jets, though this cannot be excluded by current observations (K¨ rding et al. 2005). The possibility of the soft component o resulting from an out?ow of material from the accretion disc may also be discounted as this would produce a thermal spectrum rather than a power-law continuum. Although this spectral description has not been seen in Galactic systems, a second case has been reported independently for the nearest persistent extragalactic ULX (M33 X-8) by Foschini et al. (2004). The non-standard model provided the best ?t to this ULX with Γ?2.5 and kTin ?1.2 keV. However, this source is also at the low luminosity end of the ULX regime with LX ?2×1039 erg s?1 . A further possible case, in a more luminous ULX, has arisen from the XMM-Newton data analysis of NGC 5204 X-1 (Roberts et al. 2005). In this case the authors show that there is spectral ambiguity between the non-standard ?t (Γ?3.3, kTin ? 2.2 keV) and the IMBH model (Γ?2.0, kTin ? 0.2 keV), with both providing statistically acceptable ?ts to the data. Even more recently, two additional examples of this spectral form have been uncovered in an XMM-Newton survey of ULXs by Feng & Kaaret (2005). Although it is dif?cult to derive a literal physical interpretation from the non-standard model, it does provide an accurate empirical description of ULX spectra in some cases, and as such it has the potential to provide new insights into the nature of these sources. Therefore in this paper we re-evaluate current data in an attempt to determine the best spectral description for the shape of high quality ULX spectra, and ask what consequences this has for the idea
as 20–106 M⊙ . The lower limit comes from a consideration of the measured masses of BHs in our own Galaxy (McClintock & Remillard 2003), and a theoretical limit for the mass of a BH formed from a single massive star (Fryer & Kalogera 2001). However, more recent population synthesis analyses show that BHs of up to ? 80-M⊙ may be formed in young stellar populations (Belczynski et al. 2004). Hence, when we refer to IMBHs in this paper we refer speci?cally to the larger ?1000-M⊙ IMBHs implied by the cool accretion disc measurements. c 2004 RAS, MNRAS 000, 1–16


of ULXs as accreting IMBHs. The paper is structured as follows: Sec. 2 – introduction to the ULX sample; Sec. 3 – details of the observations and data reduction; Sec. 4 – description of the spectral analysis; Sec. 5 – a comment on the luminosity and inner disc temperature relationship of these ULXs; Sec. 6 – a discussion of our results; and ?nally Sec. 7 – our conclusions.

2 THE SAMPLE As our primary goal is to ?nd the best description(s) of the shape of ULX spectra, only the highest quality datasets were chosen. The ULXs were initially selected from the ROSAT catalogues of Roberts & Warwick (2000) and Colbert & Ptak (2002) to provide a list of historically-bright ULXs that are resolved at a spatial resolution similar to XMM-Newton4 . We applied a source selection criteria of observed count rates of > 10 counts ks?1 in the ROSAT HRI camera, combined with > 10 ks of XMM-Newton/EPIC data available in the archive by December 2004, to select ten ULXs with potentially suf?cient counts for very detailed spectral analysis. In addition, we included three more high quality ULX datasets: proprietary data for Holmberg II X-1 (hereafter Ho II X-1), and two sources not quite bright enough in the ROSAT bandpass to be classi?ed as ULXs, namely the ULX in NGC 55 and M33 X-8. Whilst some of the sources in this sample have been observed more than once, we have only selected the longest individual exposure in each case to provide the clearest single view of their spectra. Our ?nal sample of 13 sources from 12 different galaxies is listed in Table 1. The selected ULXs are located at distances of between 800 kpc and 17.8 Mpc, possess XMM-Newton count rates between 0.1 and 8.9 count s?1 and cover the full range of ULX luminosities (?1039 erg s?1 – few ×1040 erg s?1 ). Hereafter we refer to the sources by their names as given in column (1) of Table 1.

3 OBSERVATIONS AND DATA ANALYSIS In this work we have utilised data from the EPIC cameras on board XMM-Newton (Turner et al. 2001; Str¨ der et al. 2001). The u datasets were obtained through the XMM-Newton public data archive (excluding proprietary Ho II X-1 data) and details of the observations are shown in Table 2. The data were processed and reduced using the standard tools of XMM - SAS software v.6.0.0. In some cases the observations were affected by soft proton ?aring for which preliminary cleaning was necessary. For these observations we extracted full ?eld X-ray (0.3–10 keV) light curves and screened for ?aring using Good Time Interval (GTI) ?les based on either a time or count rate criterion. The NGC 55 ULX and NGC 5204 X-1 observations were not affected by ?aring episodes. The NGC 2403 X-1 observation was only affected by ?aring at the end of the exposure (the last ?20 ks), therefore we used a time selection to exclude this ?aring event. For the remaining sources we used a count rate cut-off criterion to produce a GTI ?le. The exact value of the cut-off was allowed to vary from ?eld-to-?eld, to provide the best compromise in each case between excluding high background periods and facilitating the longest available exposure on the ULX. In practise the actual cut-off values varied in the 6.5–17.5 count s?1 (0.3–10 keV) range. In all cases we used only those data recorded when the
4 The ROSAT lists were deemed appropriate as most bright ULXs are persistent and vary by factors of no more than 2–3 in ?ux over a baseline of years, cf. Roberts et al. (2004).


A-M. Stobbart et al.

Table 1. The sample NH (1020 cm?2 ) (5) 1.74 d (Mpc) (6) 1.78 b LX (1039 erg s?1 ) (7) 1.3

Source (1) NGC 55 ULX 1

Alternate names (2) XMMU J001528.9?391319 a NGC 55 6 c Source 7 d – Source 6 g IXO 7 j Source 4 g IXO 8 j Source 21 l IXO 31 j Holmberg IX X-1 n IXO 34 j NGC 3031 10 q H 44 r IXO 39 j NGC 4395 X2 t IXO 53 j X-7 v IXO 65 j IXO 73 j X1 y IXO 77 j HST 3 z U1 aa Source 13 bb IXO 82 j H30 dd H2 ee

R.A. (J2000) (3) 00 15 28.9

DEC (J2000) (4) ?39 13 19.1 a

M33 X-8 2 NGC 1313 X-1 3 NGC 1313 X-2 3 NGC 2403 X-1 4 Ho II X-1 5 M81 X-9 6

01 33 50.9 03 18 20.0 03 18 22.3 07 36 25.5 08 19 29.0 09 57 53.2

+30 39 37.2 e ?66 29 11.0 h ?66 36 03.8 k +65 35 40.0 l +70 42 19.3 m +69 03 48.3 o

5.69 3.96 3.94 4.17 3.41 4.25

0.70 f 3.70 i 3.70 i 4.20 f 4.50 i 3.55 p

1.0 4.7 1.7 2.7 17 12

NGC 3628 X-1 4 NGC 4395 X-1 4 NGC 4559 X-1 4 NGC 4861 ULX 1 NGC 5204 X-1 4

11 20 15.8 12 26 01.5 12 35 51.7 12 59 01.9 13 29 38.6

+13 35 13.6 s +33 31 30.5 u +27 56 04.1 w +34 51 13.5 x +58 25 05.7 z

2.22 1.36 0.82 1.21 1.38

7.70 f 3.60 f 9.70 f 17.80 i 4.80 f

5.2 0.6 9.1 8.8 4.4

M83 ULX 1

13 37 19.8

?29 53 48.9 cc


4.70 i


N OTES : (1) Source designation; (2) Alternative names; (3–4) X-ray source position from XMM-Newton and Chandra data, or position of possible optical counterpart; (5) Galactic absorption column density from the ‘NH’ FTOOLS program (based on the measurements of Dickey & Lockman 1990); (6) Distance to the host galaxy; (7) Observed X-ray luminosity (0.3–10 keV) based on the results of the physically self-consistent modelling (see later). R EFERENCES : 1 This paper, 2 Markert & Rallis (1983), 3 Colbert et al. (1995), 4 Roberts & Warwick (2000), 5 Dewangan et al. (2004), 6 Fabbiano (1988), a Stobbart et al. (2004), b Karachentsev et al. (2003), c Read, Ponman & Strickland (1997), d Schlegel, Barrett & Singh (1997), e Foschini et al. (2004), f Ho, Filippenko, & Sargent (1997), g Schlegel et al. (2000), h Miller et al. (2003), i Tully (1988), j Colbert & Ptak (2002), k Zampieri et al. (2004), l Schlegel & Pannuti (2003), m Kaaret, Ward, & Zezas (2004), n Miller et al. (2004a), o Ramsey et al. (2006), p Paturel et al. (2002), q Radecke (1997), r Immler & Wang (2001), s Strickland et al. (2001), t Lira, Lawrence, & Johnson (2000), u Sourcelist (Obs.Id 5302) via Chandra X-ray Centre: http://cxc.harvard.edu/chaser, v Vogler, Pietsch, & Bertoldi (1997), w Cropper et al. (2004), x Sourcelist (Obs.Id 014115010) via XMM-Newton Science Archive: http://xmm.vilspa.esa.es/external/xmm data acc/xsa/index.shtml, y Liu & Bregman (2005), z Goad et al. (2002), aa Liu, Bregman, & Seitzer (2004), bb Ehle et al. (1998), cc Soria & Wu (2002), dd Immler et al. (1999), ee Trinchieri, Fabbiano, & Paulumbo (1985).

MOS and pn cameras were in simultaneous operation (see Table 2 for the net good exposure times). For each ULX, events were extracted in a circular aperture centred on the source position given in Table 1. The background was taken from a circular region, near to the source in the pn camera and at the same distance from the readout node. The chosen background regions were the same in all three detectors where observations were taken using the full-frame observing mode. However, in three cases the MOS cameras were operated in either the largewindow or small-window mode, so here we used a background region closest to the one used for the pn extraction. The size of the source and background spectral extraction regions are listed in Table 3. For the pn camera the spectra were extracted using event patterns 0–4, which allows ‘single’ and ‘double’ pixel events. We also set ‘FLAG=0’ to exclude all events at the edge of the CCD and

events from bad pixels. For the MOS detectors we used event patterns 0–12 which allows ‘single’, ‘double’, ‘triple’ and ‘quadruple’ pixel events. We employed a less conservative screening criterion for the MOS data by using the ?ag expression # XMMEA EM to exclude hot pixels and events outside of the ?eld of view. The SAS task especget was used to produce source and background spectra for each ULX, together with the appropriate Redistribution Matrix File (RMF) and Ancilliary Response File (ARF). Spectral ?les were grouped to require at least 20 counts bin?1 before ?tting, to ensure Gaussian statistics. Model spectra were ?tted to the data using the HEAsoft Xray spectral ?tting package XSPEC (v.11.3.0). The pn, MOS-1 and MOS-2 spectra for each object were ?tted simultaneously, but we included constant multiplicative factors in each model to allow for calibration differences between the cameras. This value was frozen at unity for the pn data and allowed to vary for the MOS detectors,
c 2004 RAS, MNRAS 000, 1–16

XMM-Newton observations of the brightest ULXs
Table 2. XMM-Newton Observation Log Source (1) NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX Obs.ID (2) 0028740201 0102640101 0106860101 0106860101 0164560901 0200470101 0112521101 0110980101 0142830101 0152170501 0141150101 0142770101 0110910201 Date (3) 2001-11-14 2000-08-04 2000-10-17 2000-10-17 2004-09-12 2004-04-15 2002-04-16 2000-11-27 2003-11-30 2003-05-27 2003-06-14 2003-01-06 2003-01-27 Duration (s) (4) 34025 18672 42769 42769 84600 111999 11935 60745 118900 43290 29600 32999 31722 Net exp. (s) (5) 30410 6850 18490 18490 57063 47260 8440 45260 98900 37660 14590 17048 21130 Rate (ct s?1 ) (6) 2.08 8.55 1.15 0.39 0.47 4.52 3.25 0.20 0.18 0.48 0.11 0.87 0.21 MOS-1 (7) FF SW FF FF FF LW FF FF FF SW FF FF FF MOS-2 (8) FF SW FF FF FF LW FF FF FF SW FF FF FF pn (9) FF FF FF FF FF FF FF FF FF FF FF FF EFF Position (10) on-axis on-axis on-axis off-axisa on-axis on-axis on-axis on-axis on-axis on-axis on-axis on-axis off-axisb Ref (11) 1 2,3,4 4,5,6 4,5,6 4 7 4,5 4,5 4,8 4,9 – 4,10 4


N OTES : (1) Source designation; (2) Observation identi?er; (3) Observation date (yyyy-mm-dd); (4) Observation duration (NB. this does include calibration observations in some cases); (5) Useful exposure after correcting for ?aring episodes and ensuring simultaneous operation of the EPIC cameras; (6) Combined EPIC count rates derived from the combined X-ray light curves (0.3–10 keV); (7–9) Observing mode of each EPIC detector (SW : Small Window, LW : Large Window, FF : Full Frame, EFF : Extended Full Frame); (10) Source position with respect to the centre of the pn ?eld of view: a–?7 ′ offset; b– ?6.5 ′ offset; (11) References for previous analyses of these datasets (although not all references contain a complete analysis of the ULX): 1–Stobbart et al. (2004), 2–Foschini et al. (2004), 3–Pietsch et al. (2004), 4–Feng & Kaaret (2005), 5–Wang et al. (2004), 6–Miller et al. (2003), 7–Goad et al. (2006), 8–Vaughan et al. (2005), 9–Cropper et al. (2004), 10–Roberts et al. (2005).

with the values typically agreeing within 20 per cent. Quoted ?uxes are an average of the three measurements. Spectra were initially ?tted in the 0.3–10 keV band. However, in several cases we found that the data below 0.5 keV did not agree with any of the models we tried, leaving large residuals (particularly with respect to the pn data) to the best ?ts. We note that the datasets with this problem – NGC 55 ULX, NGC 1313 X-1 & X-2 and NGC 3628 X-1 – were all obtained in the years 2000–2001 (in fact M33 X-8 is the only observation from this epoch without these ?tting residuals). At ?rst glance this would appear to be a problem with the calibration at early mission times; however the pn calibration at least has remained remarkably static in orbit (EPIC team, priv. comm.). As we cannot explain this variation we conservatively excluded these data in all EPIC cameras, such that spectral ?tting was restricted to the 0.5–10 keV range in these four cases. Also, in the case of Ho II X-1 there is a possible pn calibration feature that is particularly prominent at low energies (Goad et al. 2006), therefore we followed the method of these authors and excluded the pn data below 0.7 keV to account for this, whilst retaining MOS data down to 0.3 keV. The X-ray spectra were modi?ed for absorption following Balucinska-Church & McCammon (1992), assuming the solar abundances of Anders & Grevesse (1989) (i.e., the WABS component in XSPEC). We used two absorption components, one of which was ?xed for each source to represent the appropriate foreground column density through our Galaxy (Dickey & Lockman 1990 – see Table 1), and the second component was left free to ?t the data to represent additional absorption within the host galaxy and/or intrinsic to the ULX. The errors quoted in this work are at the 90% con?dence level for one interesting parameter. Throughout this analysis we distinguish statistically acceptable ?ts from unacceptable ?ts using a ?xed criterion of Prej < 95 per cent, where Prej is the probability of rejection derived directly from the χ2 statistic for the spectral ?t.
c 2004 RAS, MNRAS 000, 1–16

Table 3. Spectral extraction apertures Extraction radius Source Background (2) (3) 60 35 35 35 45 45 30 35 40 40 15 40 35 75 120 70 70 45 60 45 70 40 80 45 50 50

Source (1) NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

N OTES : (1) Source designation; (2)–(3) Aperture radii in arcseconds

4 SPECTRAL PROPERTIES 4.1 Simple models The X-ray spectra appeared relatively smooth and featureless in each case, hence we began by ?tting simple continuum models to the data. We tried the two models most often used in the past as single-component descriptions of ULX spectra, namely a powerlaw continuum model and the canonical MCD model. The details of these ?ts are given in Table 4 and the quality of the X-ray spectra is illustrated using the power-law ?ts in Figure 1. Due to the excellent spectral quality, we were able to reject the power-law model in 8 out of the 13 cases at > 95% con?dence. Interestingly, the ?ve ULXs for which a power-law is an adequate description of the data (NGC 1313 X-1 & X-2, M83 ULX, NGC 4861 ULX, NGC 5204 X-1) include the three poorest quality datasets. Even more notably,


A-M. Stobbart et al.

Figure 1. EPIC pn count rate spectra and ?χ residuals for a simple power-law continuum model ?t for each of the ULXs in the sample.

c 2004 RAS, MNRAS 000, 1–16

XMM-Newton observations of the brightest ULXs


Figure 1 – continued

the MCD model does not provide an acceptable ?t to any of the spectra.

4.2 Power-law + MCD model As simple spectral models were inadequate, we proceeded to ?t the data with the two component model generally employed for BHB systems, namely the combination of a power-law plus a MCD component. More speci?cally, we began by testing the standard IMBH model, i.e., a cool accretion disc plus power-law continuum model. This spectral description improved on the simple models by providing acceptable ?ts to 8 sources (NGC 1313 X-1 & X-2, M81 X-9, NGC 3628 X-1, NGC 4559 X-1, NGC 4861 ULX, NGC 5204 X-1, M83 ULX), with 0.1 keV < kTin < 0.3 keV and 1.6 < Γ < 2.5
c 2004 RAS, MNRAS 000, 1–16

(Table 5). As with previously published work, these disc temperatures are broadly consistent with IMBHs of around ?1000-M⊙ in size, though the power-law slopes are puzzlingly shallow for what are supposed HS (or VH state) sources (see Roberts et al. 2005 for further discussion). In all cases the 0.3–10 keV ?ux is dominated by the power-law component. We then attempted to ?t the X-ray spectra with the nonstandard model i.e., with the power-law component dominant at soft energies. This also provided statistically-acceptable ?ts to 8 ULXs (M33 X-8, NGC 1313 X-2, NGC 2403 X-1, NGC 3628 X1, NGC 4559 X-1, NGC 4861 ULX, NGC 5204 X-1, M83 ULX), with disc temperatures of 1.2 keV < kTin < 2.2 keV (excluding M 83 for which the temperature is unconstrained below ?4 keV) and 2.5 < Γ < 4.5 (except NGC 3628 X-1, for which Γ is un-


A-M. Stobbart et al.

Table 4. Single component spectral ?ts
WA * WA * PO



Γb 3.38±0.03 2.28±0.02 1.84+0.04 ?0.03 2.27+0.08 ?0.07 2.40±0.03 2.76±0.02 1.89±0.03 1.45±0.05 4.37+0.12 ?0.11 2.38+0.05 ?0.04 2.49+0.17 ?0.15 2.10+0.05 ?0.04 +0.11 2.51?0.10 kTin d 0.60 1.08±0.01 1.32 0.93±0.04 1.04±0.02 0.59 1.42 2.27+0.13 ?0.12 0.32±0.01 0.69 0.64 0.66 0.64±0.03

AP c 3.09±0.08 6.02±0.12 0.55±0.02 0.31±0.02 0.65±0.02 3.23±0.04 1.92+0.06 ?0.05 0.10±0.01 0.09±0.01 0.27±0.01 0.09±0.01 0.39+0.02 ?0.01 0.14±0.01 AMCD e 1.62 0.62+0.02 ?0.01 0.04 0.05±0.01 0.06±0.01 2.74 0.09 (1.38+0.28 ) × 10?3 ?0.24 0.81+0.16 ?0.13 0.15 0.06 0.33 0.11±0.02

χ2 /dof 1128.9/830 1790.1/1115 723.0/675 261.7/257 1130.2/851 1690.6/1314 938.1/864 459.8/406 540.8/368 752/599 94.0/74 572.3/529 199.2/207 χ2 /dof 1796.3/830 1538.7/1115 1409.6/675 443.3/257 923.7/851 8646.7/131 1758.6/864 466.6/406 693.0/368 1651.4/599 166.9/74 1650.1/529 310.7/207

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

3.50±0.08 1.94±0.06 1.05±0.11 1.68±0.02 3.78±0.13 1.53±0.04 1.57±0.08 2.77±0.03 2.73+0.17 ?0.16 1.15±0.08 1.24+0.31 ?0.30 0.50±0.08 +0.21 1.22?0.20 NH

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

0.37 0 0 < 0.02 1.20+0.08 ?0.07 0 0 1.04±0.02 0.13±0.09 0 0 0 < 0.01

N OTES : Models are abbreviated to XSPEC syntax: WA–absorption components for the Galactic value and external absorption; PO–power-law continuum; DISKBB–MCD. a External absorption column (1021 atoms cm?2 ), b PO photon index, c PO normalisation (10?3 photon cm?2 s?1 keV?1 at 1 keV ), d Inner disc temperature (keV). e DISKBB normalisation (((Rin /km)/(D/10 kpc))2 cos Θ; where Rin –inner disc radius, D–distance to source, Θ–inclination angle of the disc). Where the reduced χ2 value (i.e. 2, errors are not shown. Statistically acceptable ?ts are highlighted in bold.
χ2 ) dof


In total, the power-law + MCD combination could not be rejected at > 95 per cent con?dence for 10/13 sources in the sample. Two of these sources were unambiguously best ?tted with the IMBH model (NGC 1313 X-1, M81 X-9) while two more were best ?tted by the non-standard model (M33 X-8, NGC 2403 X-1). Statistically acceptable ?ts existed for both models for the remaining six sources. Three sources in this sample (NGC 55 ULX, Ho II X-1 and NGC 4395) rejected both models at > 95 percent con?dence, although a comparison of the ?ts show that the spectrum of the NGC 55 ULX appeared to ?t much better to the non-standard model, whilst Ho II X-1 and NGC 4395 X-1 much preferred the IMBH model. This sample of high quality ULX datasets demonstrates spectral ambiguity for six sources (NGC 1313 X-2, NGC 3628 X-1, NGC 4559 X-1, NGC 4861 ULX, NGC 5204 X-1, M83 ULX). As Roberts et al. (2005) suggest, a key discriminator between the models could be a test for spectral curvature at high energies (> 2 keV). Curvature is not expected in the IMBH model where the

constrained below ?1.2) (Table 5). The balance in 0.3–10 keV ?ux between the two components was much more varied for this model, though most sources showed a relatively equitable balance (ratios of no more than 3:1 in either direction, and ?1:1 in ?ve cases).

power-law component is dominant above 2 keV, whereas sources best-?tted by the non-standard model have the curved MCD component dominant at these hard X-ray energies. We therefore attempted to ?t the high energy (2–10 keV) ULX spectra with a broken power-law model. For comparison we also ?tted the 2–10 keV ULX spectra with a single power-law component, as shown in Table 6. We did not include absorption in the ?t in either case, as single- and two-component ?ts generally limit absorption to < 4 × 1021 atoms cm?2 , which should not strongly affect the ? data above 2 keV. When we do include absorption in these ?ts we actually measure far higher columns in several cases, which is physically unrealistic (and may be a result of the absorption compensating for intrinsic curvature). The results of the broken powerlaw spectral ?ts, together with the statistical probability of the ?t improvement over the single power-law ?ts (using the F-test) are shown in Table 7. To demonstrate the validity of this method, the two sources with unambiguous non-standard ?ts (M33 X-8 and NGC 2403 X1), plus NGC 55 ULX which clearly prefers this model, show unacceptable power-law ?ts that are made acceptable by the inclusion of a break. In these cases the improvement is highly statistically signi?cant (> 9σ improvement over a simple power-law ?t accordc 2004 RAS, MNRAS 000, 1–16

XMM-Newton observations of the brightest ULXs
Table 5. Two component spectral ?ts
WA * WA *( PO + DISKBB )1




Γb 3.38±0.03 2.52+0.02 ?0.01 1.77±0.06 2.11+0.15 ?0.17 2.68+0.02 ?0.05 2.63±0.03 1.81+0.05 ?0.04 1.58+0.07 ?0.04 3.72+0.18 ?0.27 2.23+0.07 ?0.05 2.24+0.23 ?0.24 1.91+0.07 ?0.08 2.47+0.14 ?0.17 Γb 4.31+0.19 ?0.27 2.49+0.18 ?0.14 – 2.93+0.75 ?0.47 4.05+0.70 ?0.75 3.45+0.11 ?0.10 – < 1.18 – 4.47+0.37 ?0.34 4.36+1.34 ?1.08 3.33+0.34 ?0.32 2.75+0.94 ?0.45 kT h 0.20±0.01 0.27±0.02 0.25+0.02 ?0.01 0.27+0.04 ?0.02 0.21+0.03 ?0.02 0.23±0.002 0.27±0.01 0.60+0.05 ?0.06 0.17±0.01 0.17±0.01 0.20±0.03 0.20±0.01 0.21+0.02 ?0.03

AP c 3.09+0.07 ?0.08 8.51+0.20 ?0.12 0.51±0.04 0.25+0.05 ?0.06 0.98±0.03 2.81±0.09 1.74+0.14 ?0.10 0.12±0.01 0.05±0.01 0.24+0.03 ?0.01 0.07±0.02 0.31±0.03 0.14+0.02 ?0.03 AP c 3.29+0.21 ?0.27 3.15+0.31 ?0.32 – 0.36+0.09 ?0.05 0.55+0.20 ?0.15 3.78+0.11 ?0.10 – < 0.04 – 0.41+0.05 ?0.04 0.13+0.08 ?0.04 0.44+0.05 ?0.03 0.14±0.02 AB i 3.35+0.33 ?0.31 3.09+0.31 ?0.30 0.86+0.11 ?0.10 0.41+0.07 ?0.06 0.38+0.14 ?0.09 4.37+0.06 ?0.07 2.56±0.16 0.16+0.04 ?0.06 0.15±0.02 0.61+0.11 ?0.08 0.15+0.09 ?0.04 0.67±0.03 0.16+0.05 ?0.02

kTin d < 0.16 0.09±0.002 0.19±0.02 0.27+0.10 ?0.09 0.09+0.002 ?0.004 0.19±0.01 0.20±0.03 0.08±0.01 0.18±0.02 0.14±0.01 0.18+0.08 ?0.04 0.22+0.04 ?0.03 [? 0.2] k kTin d 0.86±0.03 1.18±0.06 – 2.65+1.06 ?0.78 1.16+0.04 ?0.05 1.79+0.10 ?0.09 – 1.40±0.21 – 1.61+0.13 ?0.12 1.60+0.60 ?0.32 2.23+0.32 ?0.26 < 4.03 kTin d 0.81±0.01 1.26+0.02 ?0.03 2.20±0.10 1.71+0.15 ?0.16 1.12+0.04 ?0.03 1.28±0.02 2.21+0.12 ?0.11 4.32+0.74 ?0.87 0.61±0.05 1.33+0.07 ?0.06 1.37+0.20 ?0.08 1.69+0.10 ?0.09 1.10+0.13 ?0.05

AMCD e (×102 ) < 9.71 × 104 3312+1694 ?1055 0.81+1.39 ?0.55 0.03+0.27 ?0.02 527+644 ?224 211+1.23 ?0.78 1.05+2.78 ?0.61 127+402 ?96 0.11+0.09 ?0.05 +5.64 2.74?0.96 0.13+0.03 ?0.12 0.12+0.21 ?0.07 < 0.98 AMCD e (×10?3 )
+35.51 170.47?40.63 +53.06 264.55?43.13 – <2.19 +10.14 36.93?7.95 10.33+3.33 ?2.70 – 4.68+1.65 ?2.31 – 4.04+1.64 ?1.59 1.09+1.34 ?0.87 1.96+1.17 ?0.87 <4.53

χ2 /dof 1128.8/828 1527.0/1113 664.5/673 255.0/255 1028.5/849 1453.1/1312 877.2/862 424.6/404 492.2/366 597.2/597 84.4/72 529.9/527 198.6/205 χ2 /dof 957.6/828 1187.0/1113 – 255.9/255 847.7/849 1508.8/1312 – 439.5/404 – 637.0/597 85.6/72 525.5/527 198.1/205 χ2 /dof 884.1/828 1216.7/1113 666.5/673 257.1/255 830.0/849 1677.9/1312 922.3/862 429.4/404 464.5/366 604.9/597 77.7/72 527.8/527 193.2/205

fXPO f 99.9 92.7 90.9 91.7 93.6 89.7 94.0 98.1 71.6 81.5 82.4 88.2 96.6 fXPO f 67.6 48.8 – 64.4 27.3 74.5 – 53.4 – 47.3 51.9 50.3 82.9 fXBB j 36.4 11.7 17.7 27.7 8.7 41.4 17.5 17.7 65.5 30.6 34.1 28.0 27.8

fXMCD g 0.1 7.3 9.1 8.3 6.4 10.3 6.0 1.9 28.4 18.5 17.6 11.8 3.4 fXMCD g 32.4 51.2 – 35.6 72.7 25.5 – 46.6 – 52.7 48.1 49.7 17.1 fXMCD g 63.6 88.3 82.3 72.3 91.3 58.6 82.5 82.3 34.5 69.4 65.9 72.0 72.2

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX
WA * WA *( PO + DISKBB )2

3.50+0.42 ?0.09 4.43+0.25 ?0.06 2.37+0.52 ?0.48 1.80+0.76 ?0.42 7.18+0.04 ?0.05 1.74+0.10 ?0.09 2.17+0.43 ?0.23 4.26+0.77 ?0.70 1.96+0.26 ?0.40 2.33+0.70 ?0.20 1.75+1.13 ?0.69 0.68+0.22 ?0.18 1.28+0.54 ?0.37 NH

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX
WA * WA *( BB + DISKBB )

4.32+0.24 ?0.34 1.42+0.21 ?0.18 – 2.52+0.97 ?0.62 4.77+1.01 ?0.99 2.34+0.13 ?0.12 – 1.43+0.47 ?0.26 – 3.35+0.46 ?0.41 3.33+1.75 ?1.30 1.64+0.37 ?0.33 1.37+0.92 ?0.45 NH

AMCD e (×10?3 ) 343+53 ?47 303+37 ?16 5.07+0.31 ?0.16 4.02+0.49 ?1.36 46.63±3.42 77.69+6.44 ?6.05 15.93+3.34 ?2.83 0.13+0.17 ?0.07 17.66+9.56 ?6.26 9.94+2.29 ?1.82 2.38+1.89 ?1.13 7.37+1.69 ?1.40 9.28+2.76 ?3.55

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

1.43+0.16 ?0.17 0.16+0.10 ?0.09 0.60+0.14 ?0.25 < 0.54 2.02±0.30 < 0.01 0.64+0.14 ?0.10 0.71+0.28 ?0.24 0.34+0.19 ?0.17 0.80+0.17 ?0.11 0.46+0.05 ?0.03 < 0.03 < 0.37

N OTES : Models are abbreviated to XSPEC syntax: WA, PO and DISKBB as before, BB–blackbody continuum. Model ?tted with a cool1 or hot2 DISKBB component. a External absorption column (1021 atoms cm?2 ), b PO photon index, c PO normalisation (10?3 photon cm?2 s?1 keV?1 at 1 keV), d inner disc temperature (keV). e DISKBB normalisation as before (((Rin /km)/(D/10 kpc))2 cos Θ), f fraction of the total ?ux (0.3–10 keV) in the PO component, g fraction of the total ?ux (0.3–10 keV) in the DISKBB component, h blackbody temperature (keV), i blackbody normalisation, (10?5 L39 /D2 ; where L39 –source luminosity in 1039 erg s?1 , D10 –source distance in 10 kpc). j Fraction of the total ?ux (0.3–10 keV) in the 10 BB component. k Unconstrained at the 90 per cent con?dence level. In the three cases in the central portion of the table where no ?t is shown, the 2 -minimisation always found the minimum describing the IMBH model. Statistically acceptable ?ts are highlighted in bold. (NB. In cases where the χ spectral ?tting was not performed over the 0.3–10 keV energy range, we created a dummy response matrix in XSPEC (see Sec. 4.5), to determine the fraction of the total ?ux in each spectral component over 0.3–10 keV, consistent with the other datasets).

c 2004 RAS, MNRAS 000, 1–16


A-M. Stobbart et al.

ing to the F-test). However, these were the clearest-cut cases. Of the ambiguous spectra, three out of six showed evidence for curvature. This was marginal in the case of M83 ULX (> 2σ level), but more signi?cant for NGC 4559 X-1 and NGC 5204 X-1 (> 3σ), with the latter case having an unacceptable power-law ?t. Most interestingly, though, one of the IMBH-?t sources (NGC 1313 X-1) shows evidence for curvature (> 3σ improvement), whilst Ho II X1 also shows a signi?cant break (> 4σ improvement). In total 8/13 sources show at least marginal evidence for curvature at the high energy end of their XMM-Newton spectrum. This is strongly suggestive that the majority of sources are not dominated by a powerlaw continuum at these energies, such as one might expect to see if the X-ray emission arises from the optically thin, hot corona assumed in the IMBH model.
Figure 2. EPIC pn count rate spectra and ?χ residuals for the MEKAL + PO model ?t to NGC 4395 X-1. Also shown are the individual additive model components: solid line–PO, dotted line–MEKAL.

4.3 Dual thermal models In the above analysis we determine that there is (at least marginal) evidence of curvature above 2 keV in the majority of our sources, arguing that a power-law continuum is not an adequate description of this emission, while the hotter accretion disc present in the alternate model may be more appropriate. The situation below 2 keV is less clear cut. Though the alternate model uses a power-law continuum to describe the soft component, this is modi?ed to appear curved by absorption. As this power-law emission is dif?cult to understand physically we have therefore explored a new variation of the alternate model in which the soft component is also intrinsically curved (i.e., it does not rely solely on absorption to produce the observed curvature). In this model we ?tted a cool blackbody (BB) continuum to the soft component, and a MCD to the hard component. The physical motivation for the soft component comes from the suggestion that optically-thick out?owing winds from BHs accreting at or above the Eddington limit could explain ultrasoft components in ULXs (King & Pounds 2003). Such winds would appear as BB continua with temperatures of ?0.1–0.3 keV. One might then observe emission from both the accretion disc and the wind given a favourable accretion disc geometry and viewing angle (A. King, priv. comm.)5 . The results of applying this new model to the data are shown in Table 5. It proved to be the most successful empirical description of the data, providing statistically acceptable ?ts to 10 out of the 13 sources, with 0.15 keV < kT < 0.3 keV and 0.8 keV < kTin < 2.2 keV (excluding NGC 3628 X-1 for which the temperatures are much higher than shown in the other sources, i.e., kT ?0.6 keV and kTin ?4.3 keV). This model provided the only acceptable empirical ?t to the ULX in NGC 55, and all six ambiguous sources are adequately described by this model. Curiously, this is also true for both IMBH sources (though it does provide a notably worse ?t than the IMBH model for M 81 X-9), and for one of the two non-standard model sources (NGC 2403 X-1). However, this model could not provide statistically acceptable ?ts to the other non-standard source (M 33 X-8) or to the ?nal two ULXs (Ho II X1 & NGC 4395 X-1, which also do not ?t well with the power-law + MCD combination).

4.4 Why do Ho II X-1 and NGC 4395 X-1 not conform? As shown in the previous sections, statistically acceptable ?ts for Ho II X-1 and NGC 4395 X-1 cannot be found using the chosen two-component models. Interestingly, these are two of the softest sources in the sample which means that their X-ray spectra are relatively dominated by < 1 keV emission, where it is most sensitive to absorption characteristics (cf. Fig. 1). The analysis of an RGS spectrum of Ho II X-1 by Goad et al. (2006) revealed that its Xray emission is subject to absorption by a medium with a sub-solar oxygen abundance. The mis-modelling of absorption, in combination with the exceptionally high signal to noise EPIC spectrum of Ho II X-1, may be responsible for the lack of a good ?t to this ULX in our analysis. However, even after correcting for the subsolar abundance absorption acting on Ho II X-1, Goad et al. (2006) still did not ?nd acceptable ?ts using the canonical power-law + MCD model. Instead, a statistically acceptable solution to the spectrum was found through using a more physical model, namely the DISKPN + COMPTT combination. We explore the utility of a similar model in describing our whole dataset in the next section. In the case of NGC 4395 X-1, a further inspection of the residuals to a simple power-law ?t does suggest some structure, notably including a smooth ‘hump’ at ?1 keV, which is unlike the featureless X-ray spectra of the other ULXs. Therefore we tried a ?t including a MEKAL component (nominally representing the emission spectrum from hot, collisionally-ionised gas) in addition to a power-law continuum to model this spectrum. The result is shown in Table 8. This model does indeed provide the best (and only statistically accceptable) ?t to NGC 4395 X-1 with χ2 /dof = 397.5/366. The model ?tted to the pn count rate spectrum is shown in Fig. 2, as well as the individual additive model components. From this ?gure, one can clearly see how the MEKAL component effectively models the emission hump at ? 1 keV. Using a ?xed solar-abundance absorber, we ?nd the temperature of the MEKAL component is ?0.75 keV, while the photon index is quite steep at Γ ? 4. MEKAL components have been reported in the spectral ?ts of a small number of ULXs. In fact, two other ULXs within this sample have been described thus in previous analyses, namely Ho II X1 based on joint ROSAT - ASCA ?ts (Miyaji, Lehmann & Hasinger 2001), and NGC 4559 X-1 based on one of two Chandra observations (Roberts et al. 2004). We have investigated whether such MEKAL + power-law models can be applied to the current data for these sources, and show the results in Table 8. This model is again
c 2004 RAS, MNRAS 000, 1–16


Miller et al. (2004a) argue that cool, optically-thick out?ows cannot explain the soft excess in ULXs as it would be impossible to form powerful enough shocks to produce a luminous non-thermal power-law continuum component in such systems. If the hard component is not a power-law, but instead originates directly from the accretion disc, then the requirement for shocks to be present is removed and their argument is circumvented.

XMM-Newton observations of the brightest ULXs
Table 6. Power-law spectral ?ts (2–10 keV)
WA * WA * PO


Γa 3.58+0.06 ?0.07 2.59±0.04 1.70+0.05 ?0.07 2.19+0.12 ?0.16 2.63+4.95 ?0.07 2.61±0.03 1.78±0.05 1.50+0.06 ?0.08 4.48+0.31 ?0.45 2.28+0.08 ?0.10 2.50±0.38 1.89±0.09 2.67+0.21 ?0.28

AP b 3.57±0.25 8.90+0.41 ?0.39 0.44+0.03 ?0.04 0.27+0.04 ?0.05 0.82+0.05 ?0.06 2.65±0.10 1.59±0.10 0.10±0.01 0.12±0.04 0.24+0.02 ?0.03 0.09+0.06 ?0.04 0.29+0.04 ?0.03 0.17±0.05

χ2 /dof 459.3/326 663.1/548 263.7/259 60.2/70 467.3/340 805.7/824 360.4/355 219.5/211 33.1/28 171.3/155 12.3/8 160.8/133 45.5/39

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

N OTES : Models are abbreviated to XSPEC syntax: PO as before. a PO photon index, b PO normalisation (10?3 photon cm?2 s?1 keV?1 at 1 keV). Statistically acceptable ?ts are highlighted in bold.

Table 7. Broken power-law spectral ?ts (2–10 keV)

Γ1 a 3.08±0.11 2.17±0.09 1.55+0.10 ?0.09 – 2.07+0.10 ?0.11 2.55±0.04 – – 2.77+1.31 ?2.58 2.07+0.12 ?0.14 – +0.10 1.65?0.15 <1.79

Ebreak b 3.90+0.20 ?0.18 3.94+0.26 ?0.21 4.93+1.37 ?0.53 – 4.00+0.17 ?0.15 5.31+0.43 ?0.56 – – 2.45+0.93 ?0.23 4.77+0.65 ?0.41 – +0.46 4.92?0.41 2.41+0.31 ?0.15

Γ2 c 5.34+0.53 ?0.41 3.32+0.19 ?0.15 2.16+0.66 ?0.25 – 4.05+0.33 ?0.28 3.08+0.25 ?0.22 – – 4.92+0.77 ?0.60 3.14+0.62 ?0.43 – +0.65 2.99?0.45 3.11+0.46 ?0.40

ABP d 2.32+0.26 ?0.24 6.07+0.56 ?0.52 0.37+0.05 ?0.04 – 0.49±0.05 2.50+0.11 ?0.12 – – <0.09 0.20±0.03 – 0.23+0.03 ?0.04 < 0.08

χ2 /dof 341.0/324 516.1/546 250.9/257 – 325.4/338 786.1/822 – – 30.0/26 157.7/153 – 138.3/131 37.5/37

?χ2e 118.3 147.0 12.9 – 141.9 19.6 – – 3.1 13.6 – 22.5 8.0

1-P(F-test)f >99.9 >99.9 99.8 – >99.9 >99.9 – – 72.1 99.8 – >99.9 97.2

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

N OTES : Models are abbreviated to XSPEC syntax:BKNPO–broken power-law model. a Photon index below the break energy, b break point for the energy (keV), c photon index above the break energy, d BKNPO normalisation (10?3 photon cm?2 s?1 keV?1 at 1 keV), e χ2 improvement over a single PO ?t, for two extra degrees of freedom, f statistical probability (per cent) of the ?t improvement over a single PO ?t. We only show the results where we could constrain ?ts that showed the requisite behaviour, i.e. Γ2 > Γ1 and Ebreak constrained in the 2–10 keV range.. Statistically acceptable ?ts are highlighted in bold.

Table 8. MEKAL + power-law spectral ?ts
WA * WA *( MEKAL + PO )



kT b 0.66±0.03 0.75±0.04 0.25±0.02

AM c 15.73+1.86 ?1.52 1.32±0.15 7.23+1.56 ?1.54

Γd 2.67±0.02 3.88+0.08 ?0.09 2.25+0.05 ?0.04

AP e 2.90±0.05 5.38+0.46 ?0.14 0.24±0.01

χ2 /dof 1463.9/1312 397.5/366 612.1/597

fXPO f 96.7 84.4 92.8

fXM g 3.3 15.6 7.2

Ho II X-1 NGC 4395 X-1 NGC 4559 X-1

1.42±0.04 1.73±0.13 1.26+0.10 ?0.11

N OTES : Models are abbreviated to XSPEC syntax: WA and PO as before, MEKAL–thermal plasma model. a External absorption column (1021 atoms cm?2 ), b plasma temperature (keV), c MEKAL normalisation ((10?19 /(4π[D(1 + z)]2 )) ne nH dV , where D is the distance to the source (cm), ne is the electron density (cm?3 ), nH is the hydrogen density (cm?3 )), d PO photon index, e PO normalisation (10?5 photon cm?2 s?1 keV?1 at 1 keV), f fraction of the total ?ux (0.3–10 keV) in the PO component, g fraction of the total ?ux (0.3–10 keV) in the MEKAL component. Statistically acceptable ?ts are highlighted in bold. (NB For Ho II X-1 we created a dummy response matrix over the 0.3–10 keV energy range, to determine the fractional ?ux in each spectral component (see Sec 4.5)).

c 2004 RAS, MNRAS 000, 1–16


A-M. Stobbart et al.
as the internal mechanisms, and it provides the dominant in?uence on the spectral shape. The source geometry assumed is either spherical or a disc-corona (slab) geometry. Low energy (UV or X-ray) thermal photons from the accretion disc are assumed to be emitted uniformly inside the source region for the spherical models and enter from the base of the corona in the slab geometry. We take the seed photons to have an accretion disc spectrum as described with the DISKPN model of Gierli? ski et al. (1999), with the inner n edge of the disc at 6GM/c2 . This model assumes a proper general relativistic potential and has a characteristic temperature kTbb . To allow for a patchy Compton corona, we add a second DISKPN component to the model, with the temperature coupled to that of the EQPAIR seed photons. This model ultimately proved the best description of the ULX spectra in our sample, with statistically acceptable ?ts to 11/13 datasets, all with similar or improved goodness of ?ts compared to the best empirical modelling, for only one extra degree of freedom. Only Ho II X-1 and NGC 4395 X-1 were rejected at > 95 per cent con?dence (see Table 9), although in the case of Ho II X-1 it does provide clearly the best ?t of all that were attempted to the data, and is only marginally rejected (Prej ≈ 97.7 per cent). Again the ?t to NGC 4395 X-1 is poor, for reasons explained in the previous section. The ?ts provide a uniformly low measurement of the disc temperature across the sample, in the range 0.07 < kTmax < 0.29. Taken in isolation, this might be interpreted as strong support for the presence of IMBHs in ULXs. However, many of the ?ts show a second remakable characteristic, which is that the optical depth of the coronae appear to be very high, ranging from τ ? 8 for Ho II X-1 up to depths well in excess of 30 (NGC 55 ULX, NGC 2403 X1)7 . Importantly, this result gives a physical explanation for the curvature noted to be present in the spectra in previous sections. This second characteristic appears irreconcilable with the main assumption behind the IMBH model, which is that they operate as simple scaled-up BHBs (see also Goad et al. 2006), as such sources typically do not possess very optically-thick coronae. This result must therefore strongly challenge the IMBH interpretation for the nine ULXs (including Ho II X-1) well-described by this model. However, three ULXs - NGC 1313 X-1, M81 X-9 and (possibly) NGC 3628 X-1 - may still possess optically-thin coronae, and as such remain viable IMBH candidates. As this spectral model provides at least the same goodness of ?t as the empirical models in all but one case, we used these ?ts to determine the observed X-ray ?ux. Due to uncertainties in the low energy spectrum of some ULXs, the spectral ?tting was restricted to 0.5–10 keV (and in the case of Ho II X-1 the pn data was restricted to 0.7–10 keV) instead of the 0.3–10 keV range adopted for the other sources. Therefore to determine the absorbed X-ray ?ux of each ULX over the same energy range, we created a dummy response matrix in XSPEC over the 0.3–10 keV energy range for those cases with calibration uncertainties. This dummy response temporarily supersedes the response matrix used in the spectral ?tting of those sources, allowing us to examine the behaviour of the DISKPN + EQPAIR model over this energy range. The three EPIC cameras are consistent to within 20 per cent in 12/13 cases, therefore we derived an absorbed ?ux (0.3–10 keV) for each of these ULXs based on an average of the three measurements. We con-

rejected for Ho II X-1, though it is notable that the quality of the ?t is very close to that of the best empirical description of the source (cool MCD plus power-law). In this case the temperature of the MEKAL is again relatively high at ? 0.66 keV, but the contribution of this component is minimal (? 3 per cent of the 0.3–10 keV ?ux)6 . This is both far hotter and far fainter than the ? 0.3 keV thermal plasma that composed 20 - 30 per cent of the 0.5–2 keV emission modelled by Miyaji et al. (2001). On the other hand, the temperature of the best-?tting MEKAL to NGC 4559 X-1 is lower, at ? 0.25 keV, but it does constitute part of a statistically-acceptable ?t to the data. This temperature is similar to the ? 0.18 keV plasma inferred by Roberts et al. (2004). We note that this latter result just adds to the spectral ambiguity already seen in the case of this ULX. Finally, if NGC 4395 X-1 really does possess an X-ray lineemitting component, what is its physical origin? Several ideas have been put forward to explain such a component in the spectrum of an ULX. Miyaji et al. (2001) suggest the presence of a young supernova remnant coincident with Ho II X-1, a theme expanded on by Feng & Kaaret (2005). Indeed, these authors independently con?rm the presence of the MEKAL component in the spectrum of NGC 4395 X-1, and ?nd two further ULXs with similar characteristics. Alternately, as Roberts et al. (2004) ?nd the MEKAL component in NGC 4559 X-1 to switch on between two Chandra observations separated by ? 5 months, they suggest that the plasma may originate in the collision of a jet or out?ow from the ULX with a denser medium in the close proximity of the system. A second alternative is offered by Terashima & Wilson (2004), who suggest that emission lines in the spectrum of an ULX in M51 may originate from a photoionised stellar wind, similar to what is seen in some high-mass X-ray binaries in our own Galaxy. A ?nal point of interest is that a broad ? 1 keV feature, similar to that driving the MEKAL ?t in NGC 4395 X-1, has recently been seen in the XMMNewton spectrum of GRS 1915+105 by Martocchia et al. (2005) (although see their paper for caveats). One intriguing possibility is that this feature could originate in a disc wind. Clearly ULXs possessing MEKAL components in their spectra are an interesting subject in their own right, and should be the subject of future attention.

4.5 Physical models The success of the COMPTT model in describing the spectra of some ULXs (e.g. Miller et al. 2003, Goad et al. 2006) shows that a Comptonised spectrum is in principle another viable alternative. The COMPTT model however, is not fully self-consistent e.g., it allows spectra to have temperatures higher than expected when Compton cooling is taken into account. We therefore attempted to ?t a more physical model to the ULX spectra, namely the EQPAIR model (Coppi 1999), which allows thermal and non thermal electron distributions. In our modelling, we assume a purely thermal electron distribution, the temperature of which is computed selfconsistently by balancing heating and cooling (the latter of which is mainly due to Compton cooling). A key parameter of this model is the ratio lh /ls , where lh and ls represent the compactness of the electrons and the compactness of the seed photon distribution respectively. This ratio depends on the geometry of the source as well


The contribution of the MEKAL is so small that its parameterisation is relatively insensitive to changes in metallicity. For example, setting a low abundance (as found by Miyaji et al. 2001) leaves the temperature of the MEKAL unchanged at ? 0.66 keV.

7 Indeed, the optical depth for four sources hit the arti?cial upper limit of τ = 100 during spectral ?tting, therefore we quote values as > ‘lower limit’ in these cases.

c 2004 RAS, MNRAS 000, 1–16

XMM-Newton observations of the brightest ULXs
Table 9.
DISKPN + EQPAIR spectral


?ts NH


kTmax b 0.24±0.01 0.08+0.01 ?0.07 0.21+0.03 ?0.02 0.26+0.09 ?0.08 0.27+0.05 ?0.04 0.20+0.02 ?0.01 0.23+0.01 ?0.03 0.07+0.02 ?0.01 0.20±0.01 0.16+0.01 ?0.02 0.24±0.06 0.29+0.03 ?0.05 0.23+0.08 ?0.07

AD c 1.37+0.50 ?0.36 < 4.37 × 10?10 < 0.44 < 0.27 0.11+0.21 ?0.06 2.01+0.27 ?0.19 < 2.08 167+972 ?112 0.15+0.08 ?0.05 +3.57 2.21?0.76 0.06+0.22 ?0.03 0.09+0.09 ?0.05 0.05+0.21 ?0.04

lh /ls d 4.54+1.90 ?2.44 3.22+0.48 ?0.21 2.22+2.48 ?0.46 1.83+0.95 ?0.40 4.50+7.39 ?1.83 1.01±0.03 3.25+2.09 ?0.47 +12.6 26.9?9.4 5.97+0.34 ?3.17 2.23+0.51 ?0.71 5.30+8.65 ?3.18 4.32+3.27 ?1.25 1.96+7.63 ?0.85

τP e > 37.2 18.7+0.6 ?0.5 0.2+0.4 ?0.1 +10.7 9.2?7.7 > 33.3 8.5+0.6 ?0.7 0.5+0.3 ?0.1 < 7.4 > 30.4 13.9+3.4 ?0.6 > 15.8 +57.0 26.3?5.6 > 15.8

AE f 0.96+3.94 ?0.24 1083+21859 ?78 9.99+8.72 ?2.59 0.72+2.80 ?0.34 0.36+1.21 ?0.31 26.7+4.4 ?8.5 +16.2 18.9?7.0 +21.1 47.8?11.2 0.03+0.03 ?0.01 4.12+1.49 ?1.45 0.06+0.55 ?0.04 0.25+0.60 ?0.13 0.32+3.98 ?0.19

χ2 /dof 889.3/827 1188.0/1112 658.7/672 254.1/254 831.7/848 1414.8/1311 873.8/861 424.9/403 467.0/365 577.5/596 78.7/71 505.8/526 193.2/204

NGC 55 ULX M33 X-8 NGC 1313 X-1 NGC 1313 X-2 NGC 2403 X-1 Ho II X-1 M81 X-9 NGC 3628 X-1 NGC 4395 X-1 NGC 4559 X-1 NGC 4861 ULX NGC 5204 X-1 M83 ULX

2.12+0.12 ?0.11 1.03+0.09 ?0.13 2.09+0.21 ?0.39 1.44+0.30 ?0.54 2.53+0.23 ?0.28 1.11+0.05 ?0.07 1.90+0.14 ?0.19 4.28+0.36 ?0.42 1.00+0.12 ?0.08 1.97+0.15 ?0.08 1.09+0.81 ?0.42 0.24±0.08 0.65+0.34 ?0.33

N OTES : Models are abbreviated to XSPEC syntax: WA as before, DISKPN–accretion disc model, EQPAIR–Comptonisation model, External absorption column (1021 atoms cm?2 ), b maximum temperature in the accretion disc (keV), c DISKPN normalisation (10?3 (M 2 cos(i))/(D 2 β 4 ); where M–central mass (M⊙ ), D–distance to the source (kpc), i–inclination angle of the disc, β–colour/effective temperature ratio), d ratio between the compactness of the electrons and the compactness of the seed photon distribution, e optical depth, f EQ PAIR normalisation (corresponding to the disc component) i.e., (fc M 2 cos(i))/(D 2 β 4 ), where fc is the covering factor. Statistically acceptable ?ts are highlighted in bold.

verted these ?uxes to an observed X-ray luminosity for each source in the 0.3–10 keV band (assuming the appropriate distance) and tabulate the results in Table 18 .

5 A COMMENT ON THE RELATIONSHIP BETWEEN X-RAY LUMINOSITY AND INNER-DISC TEMPERATURE IN ULXS. Perhaps the most visually striking evidence for ULXs containing IMBHs was presented by Miller, Fabian, & Miller (2004b) (their Figures 1 & 2). These authors selected a sample of ULXs with published estimates of LX > 1040 erg s?1 , that require a soft excess component (at least at the 3σ con?dence level) in the low energy part of a two component X-ray spectrum. They then compared the inferred disc temperatures (taken from the soft component) and unabsorbed luminosities (0.5–10 keV) of their ULXs to those of a number of Galactic BHBs, and found that these ULXs and stellar mass BHBs occupy distinct regions of a LX – kT diagram. More speci?cally, the ULXs are more luminous but have cooler thermal disc components than standard stellar-mass BHBs, consistent with the ULXs harbouring IMBHs. Following on from this work, we have reproduced LX – kT diagrams, based on the inferred unabsorbed 0.5–10 keV X-ray luminosities for ULXs in this sample (Fig. 3). However, we do this for two cases: (i) disc temperatures measured from the IMBH (cool disc + hard power-law) ?ts; and (ii) disc temperatures taken from the dual thermal (cool blackbody + warm disc) ?ts. In both cases we only show sources where the spectral ?ts are statistically acceptable, and the ?t parameters are well constrained. For this rea8

son, we excluded the values from the IMBH ?t to M83 ULX, as the disc temperature is unconstrained in this case. Additionally, NGC 3628 X-1 does not appear on either plot as its unusual disc temperatures place it outside the range displayed in both cases. In each case the luminosity was derived from an average of the unabsorbed ?ux (0.5–10 keV) from each detector. The errors on the disc temperature are 90 per cent con?dence errors, likewise the luminosity errors are based on the 90 per cent con?dence errors in the average measured ?ux, as calculated by XSPEC. Finally, we illustrate the regions of parameter space occupied by the BHBs and ULXs in Miller et al. (2004b) by ellipses, with the ellipse representing IMBH-candidate ULXs at the upper-left. The two LX – kT diagrams in Fig. 3 tell very different stories. As would be expected, our IMBH model results reproduce those of Miller et al. (2004b), i.e., the ULXs occupy a very different region of parameter space to the stellar-mass BHBs, consistent with larger black holes and hence more luminous, cooler accretion discs. However, the dual thermal model disc temperatures suggest an alternative interpretation. In this case, the ULXs appear to be a direct, high luminosity extension of the BHB class, and follow the LX ∝ T 4 trend one would expect from standard accretion discs. Obviously one cannot conclude which is the correct interpretation on the basis of these plots. However, it does demonstrate very clearly that it is the choice of empirical model used to determine the characteristics of ULX spectra that governs whether one concludes that some ULXs contain IMBHs, or whether they are more similar to stellar-mass BHBs.

Although the power-law + MEKAL spectral model statistically provided the best ?t to NGC 4395 X-1, the measured source ?ux derived from this model and the DISKPN + EQPAIR model were in close agreement. For consistency, we quote the value from the DISKPN + EQPAIR model. However, as the ULX lay on a hot column in the pn data we simply use the average of the two MOS ?ux measurements to estimate its luminosity. c 2004 RAS, MNRAS 000, 1–16

In this work we have analysed the X-ray spectra of a small sample of 13 ULXs, that constituted the highest quality (i.e. most photonrich) datasets in the XMM-Newton data holdings as of December 2004. We demonstrate that the superior collecting area of XMMNewton over its rivals leads to better spectral de?nition, primarily through the rejection of simple single-component spectral models


A-M. Stobbart et al.
ous spectra, we accumulated no more than ? 18000 EPIC counts (pn and MOS combined). In each spectrum that we resolved this simple ambiguity, or excluded both cases as unacceptable ?ts, we had > 20000 EPIC counts (excluding NGC 4395 X-1, as this had both an atypical spectrum and an underestimated pn count rate due to its location on a hot column). Even then, the introduction of a model composed of two thermal components adds an extra layer of ambiguity to many of the datasets. This suggests that one requires very photon-rich datasets to make progress in descriptive empirical XMM-Newton ULX spectroscopy, equivalent to at least 100-ks of ?are-free data for a 0.2 count s?1 on-axis ULX observed by the EPIC cameras, and spectral results from poorer quality data should be regarded with caution. Issues of spectral ambiguity aside, perhaps our most interesting result is the probable detection of curvature in the high-energy part of the XMM-Newton EPIC spectrum in more than half of the sample. Though this has previously been detected in individual ULXs (Stobbart et al. 2004; Foschini et al. 2004), or as minority populations in ULX samples (Feng & Kaaret 2005), this is the ?rst suggestion that the majority of ULXs, across the whole range of ULX luminosity, might appear thus. Indeed, we demonstrate that the most successful empirical modelling of our sample is achieved using a model in which the high-energy portion of the spectrum is described by the MCD model, with a soft excess modelled as a cool blackbody emitter. This obviously adds to the challenges in modelling ULXs as IMBHs. Indeed, though 8/13 ULXs present acceptable ?ts to the simple IMBH model of a cool disc plus a hard power-law continuum, we are again faced with the problem noted by Roberts et al. (2005). Namely we detect a very dominant power-law (> 80 per cent of the observed 0.3–10 keV ?ux) that appears too hard for what should be a HS (or perhaps VH) source, if we are to believe that the inner-disc temperature can provide an estimate of the BH mass (1.6 < Γ < 2.5 for the IMBH candidates in our sample, compared to approximately 2.1 < Γ < 4.8 for the HS, and Γ > 2.5 for the VH state - McClintock & Remillard 2003). Curvature above 2 keV compounds these problems further, as the IMBH model assumes that whilst the bigger black hole results in a cooler accretion disc, the corona remains similar to that observed in Galactic BHBs, i.e. optically-thin and hence modelled with a power-law. In fact, there is no reason to suppose that increasing the mass of the BH should alter the state of the corona - timing measurements at least appear to scale linearly with BH mass for given spectral states, implying the properties of the states themselves are invariant with mass (cf. Done & Gierli? ski 2005). But the detection of curvature implies n that either the state of the coronae is changing as the BH becomes more massive, or a different physical process is responsible for the emission above 2 keV. The dual thermal model uses a MCD to model this curvature, with inner disc temperatures in the range 1.1–2.2 keV. As Fig 3 shows, this provides a natural explanation of ULXs as a simple extension of the behaviour of Galactic BHBs, with more luminous and slightly hotter discs. In this case, assuming one can exceed the Eddington limit, there is no obvious requirement for IMBHs. However, even this simple empirical model has limitations; for instance, in order to see both the inner regions of the accretion disc and an optically-thick out?ow simultaneously, one requires a very speci?c geometry and viewing angle. One may therefore have to consider further explanations for the origin of the soft excess component; we note that one interesting suggestion, that has recently been found to work in some AGN X-ray spectra, is that the soft excess could arise in relativistically-blurred atomic feac 2004 RAS, MNRAS 000, 1–16

Figure 3. Unabsorbed X-ray luminosity (0.5–10 keV) plotted against accretion disc temperatures inferred from the X-ray spectral ?ts (as per Miller et al. 2004b). The ellipses represent the regions occupied by ULXs (small ellipse) and Galactic BHCs (large ellipse) presented in Figure 2 of Miller et al. (2004b). The values of the data points displayed in these ?gures are derived from the IMBH spectral ?ts (top panel) or the dual thermal model ?ts (bottom panel), for sources with statistically acceptable and well constrained ?ts for that model. (N OTE : IMBH data points neglect M83 ULX, which has an unconstrained temperature, and NGC 3628 X-1 which has a lower temperature than the other sources [?80 eV]; dual thermal data points also neglect NGC 3628 X-1 which has a much higher temperature than the other sources [?4 keV])

(power-law, MCD) that have adequately described previous spectra of many of these ULXs. However, we are still photon-limited in many of these spectra, to the extent that in six cases we are unable to tell whether a combination of a MCD plus a power-law uniquely ?ts the data with the MCD component at the low or the high energy end of the X-ray spectrum. In each of these six ambigu-

XMM-Newton observations of the brightest ULXs
tures found in the re?ection spectrum of a photoionised accretion disc (Crummy et al. 2006). A further concern for this model is that similar spectral shapes have not been commonly found in Galactic BHBs, with the exception of slight soft excesses (modelled by the extension of the power-law component below the MCD) seen in the spectra in LMC X-1 and LMC X-3 (Haardt et al. 2001). One explanation could be that soft excesses, below the MCD component in Galactic BHB spectra, are not seen due to a combination of high absorption columns and lack of detector sensitivity at low X-ray energies. However, the columns to a signi?cant minority of BHBs are not much in excess of the ? 1021 cm?2 columns inferred for our ULXs (McClintock & Remillard 2003), and many have now been observed by XMM-Newton and Chandra without such soft excesses being reported. Perhaps this is simply telling us that this spectral shape is unique to the very high accretion rates required if ULXs contain stellar-mass BHs.


The model that did best in terms of providing acceptable ?ts to the data was also the only physically self-consistent model we used in the analysis, namely the DISKPN + EQPAIR spectral model. The best ?ts provided by this model show, without exception, that the accretion disc photons seeding the corona are cool. In fact they are cool enough, at < 0.3 keV, to be consistent with the accretion disc around an IMBH. However, this model also provides a physical explanation for the curvature above 2 keV: unlike in conventional BHBs, it originates in an optically-thick corona. This combination of a cool disc and optically-thick corona has already been observed in Ho II X-1; Goad et al. (2006) note the cosmetic similarities between this spectral model and the model of Zhang et al. (2000) of a three-layered atmospheric structure in the accretion discs around BHBs. More speci?cally, their model includes a warm layer (kT ?1–1.5 keV, τ ?10) between the cool optically-thick accretion disc (kT ?0.2–0.5 keV) and the hot optically-thin corona (kT ?100 keV, τ ?1) of BHBs, which is responsible for the dominant component below ?10 keV (see also Nayakshin & Melia 1997, Misra et al. 1998). The warm layer appears relatively stable and hence unconnected to the hot corona, which can be highly dynamic and even disappear completely (Misra et al. 1998). In this picture, the cool disc seeds the warm optically thick scattering medium, and as such may explain both components seen in our modelling.

This model has been successful in describing the X-ray spectrum of GRS 1915+105 (Zhang et al. 2000), and so Goad et al. (2006) speculate that this modelling, combined with an upper limit on the mass of the BH in Ho II X-1 of ? 100-M⊙ derived from timing properties, implies that Ho II X-1 behaves in an analogous manner to GRS 1915+105 in its χ-class. We speculate, by extension, that many of the ULXs in our sample that are well ?tted by this spectral model could also be GRS 1915+105 analogues, i.e. stellar-mass BHs accreting at around and in excess of the Eddington limit. We note that if individual BHs with mass up to 80-M⊙ can form from stellar processes as suggested by Belczynski et al. (2004), and these larger stellar-mass BHs exist in ULXs, then the factors by which the Eddington limit are exceeded need not be large for even the brighter ULXs (factors ? 2 ? 3 only). One piece of additional, encouraging evidence in this respect is that, similarly to Ho II X-1 and GRS 1915+105 in its analogous χ-class of behaviour, the light curves of the ULXs in this sample show little or no intrinsic variability on timescales of minutes to hours, with the
c 2004 RAS, MNRAS 000, 1–16

fractional variability (in excess of counting noise) limited to < 10 ? per cent in all cases9 . Recently a second, very plausible explanation for our physically self-consistent spectral modelling has come to light. Done & Kubota (2005) describe spectral modelling of the Galactic BHB XTE J1550-564 in its high-luminosity VH state, using a model in which the energetics of the inner regions of the accretion disc are coupled to a surrounding corona. This results in a cooler apparent disc temperature, as the corona drains energy from the inner disc, and an optically-thick corona, that are both part of the accretion ?ow. Done & Kubota (2005) note that this may provide an explanation for the low disc temperatures observed in ULXs not reliant upon the presence of an IMBH. As we also provide evidence that the corona itself is indeed optically-thick in such sources, this model must constitute a very serious physical alternative to IMBHs for the majority of ULXs. However, even here there are caveats, for instance Feng & Kaaret (2005) note that optically-thick coronae should show deep Fe K absorption edges, unless the accreting material has a very low metallicity. We do not detect such features in our analysis. Our empirical spectral ?tting has posed new challenges for the IMBH model ?ts to ULX spectra, and our physical modelling describes scenarios in which the bulk of ULXs could be stellar-mass BHs accreting at around the Eddington limit. This is perhaps not surprising, as many strands of recent evidence have pointed away from an IMBH model for most ULXs. In fact, excepting the somewhat unique case for M82 X-1 as an IMBH, possibly formed in the dense MGG-11 cluster or captured as the nucleus of an accreted dwarf galaxy (e.g. Strohmayer & Mushotzky 2003; Portegies Zwart et al. 2004; King & Dehnen 2005; Mucciarelli et al. 2006), and the cool disc detections (shown to be somewhat ambiguous in this paper), observational results have tended to argue in the opposite sense. For example, the probable breaking of the Eddington limit seen in some - perhaps most (see Jonker & Nelemans 2004) - Galactic BHBs, and especially GRS 1915+105, argues that we cannot exclude stellar-mass BHs from producing ULXs on this trivial basis (McClintock & Remillard 2003). Also, the shape of the universal X-ray Luminosity Function (XLF) for high-mass X-ray binaries, derived by Grimm et al. (2003), is somewhat puzzling if IMBHs constitute a signi?cant part of the ULX population. In particular, why does the XLF appear to cut-off at ? 2 × 1040 erg s?1 ?10 Surely no such cut-off would be present if ? 1000-M⊙ IMBHs constitute a large fraction of the ULX population (we are certainly unaware of any other source population that cuts off at ? 0.1 Eddington rate). We therefore conclude that, on current evidence, it is unlikely that accreting IMBHs constitute a large proportion of the total ULX population.

7 CONCLUSIONS We have conducted a detailed examination of the X-ray spectral shapes in a sample of the highest quality XMM-Newton EPIC ULX
Excepting the NGC 55 ULX, which shows prominent dips in its XMMNewton light curve (Stobbart et al. 2004). 10 A near identical cut-off is found from a large sample of ULXs studied by Swartz et al. (2004). Independent support for this cut-off comes from the empirical LX ? star formation rate relationship of Grimm et al. (2003), which can only have its linear form above ? 10M⊙ yr?1 if the cut-off is real (see also Gilfanov et al. 2004).


A-M. Stobbart et al.
Ehle M., Pietsch W., Beck R., Klein U., 1998, A&A, 329, 39 Fabbiano G., 1988, ApJ, 325, 544 Fabbiano G., 1989, ARA&A, 27, 87 Falcke H., Biermann P. L., 1995, A&A, 293, 665 Feng H., Kaaret P., 2005, ApJ, 633, 1052 Foschini, L., Rodriguez, J., Fuchs, Y., Ho, L. C., Dadina, M., Di Cocco, G., Courvoisier, T. J.-L., & Malaguti, G. 2004, A&A, 416, 529 Fryer C.L., Kalogera V., 2001, ApJ, 554, 548 Gao Y., Wang Q. D., Appleton P. N., Lucas R. A., 2003, ApJ, 596, L171 Gierli? ski M., Zdziarski A. A., Poutanen J., Coppi P. S., Ebisawa n K., Johnson W. N., 1999, MNRAS, 309, 496 Gilfanov M., Grimm H.-J., Sunyaev R., 2004, NuPhS, 132, 369 Goad M. R., Roberts T. P., Knigge C., Lira P., 2002, MNRAS, 335, L67 Goad M. R., Roberts T. P., Reeves J., Uttley P., 2006, MNRAS, 365, 191 Grimm H.-J., Gilfanov M., Sunyaev R., 2003, MNRAS, 339, 793 Haardt F., et al., 2001, ApJS, 133, 187 Ho L. C., Filippenko A. V., Sargent W. L. W., 1997, ApJS, 112, 315 Immler S., Vogler A., Ehle M., Pietsch W., 1999, A&A, 352, 415 Immler S., Wang Q. D., 2001, ApJ, 554, 202 Jonker P.G., Nelemans G., 2004, MNRAS, 354, 355 Kaaret P., Corbel S., Prestwich A. H., Zezas A., 2003, Science, 299, 365 Kaaret P., Ward M. J., Zezas A., 2004, MNRAS, 351, L83 Karachentsev I. D. et al., 2003, A&A, 404, 93 King A. R., 2004, MNRAS, 347, L18 King A. R., Davies M. B., Ward M. J., Fabbiano G., Elvis M., 2001, ApJ, 552, L109 King A.R., Dehnen W., 2005, MNRAS, 357, 275 King A. R., Pounds K. A., 2003, MNRAS, 345, 657 K¨ rding E., Falcke H., Markoff S., 2002, A&A, 382, L13 o K¨ rding E., Colbert E., Falcke H., 2005, A&A, 436, 427 o Krolik J. H., 2004, ApJ, 615, 383 Lira P., Lawrence A., Johnson R. A., 2000, MNRAS, 319, 17 Liu J., Bregman J. N., Seitzer P., 2004, ApJ, 602, 249 Liu J., Bregman J. N., 2005, ApJS, 157, 59 Makishima K., Maejima Y., Mitsuda K., Bradt H. V., Remillard R. A., Tuohy I. R., Hoshi R., Nakagawa M., 1986, ApJ, 308, 635 Makishima K., et al., 2000, ApJ, 535, 632 Markert T. H., Rallis A. D., 1983, ApJ, 275, 571 Markoff S., Falcke H., Fender R., 2001, A&A, 372, L25 Markoff S., Nowak M.A., Wilms J., 2005, ApJ, 635, 1203 Martocchia A., Matt G., Belloni T., Feroci M., Karas V., Ponti G., 2005, A&A, in press (astro-ph/0510651) McClintock J. E., et al., 2001, ApJ, 555, 477 McClintock J. E., Remillard R. A., 2003, in Lewin W. H. G., van der Klis M., eds, Compact Stellar X-ray Sources. Cambridge Univ. Press, Cambridge, in press (astro-ph/0306213) Miller J. M., Fabbiano G., Miller M. C., Fabian A. C., 2003, ApJ, 585, L37 Miller J. M., Fabian A. C., Miller M. C., 2004a, ApJ, 607, 931 Miller J. M., Fabian A. C., Miller M. C., 2004b, ApJ, 614, L117 Miller M. C., Colbert E. J. M., 2004, IJMPD, 13, 1 Misra R., Chitnis V. R., Melia F., 1998, ApJ, 495, 407 Mitsuda K., et al., 1984, PASJ, 36, 741 Miyaji T., Lehmann I., Hasinger G., 2001, AJ, 121, 3041 Mucciarelli P., Casella P., Belloni T., Zampieri L., Ranalli P., 2006, MNRAS, 365, 1123
c 2004 RAS, MNRAS 000, 1–16

datasets available to us. Most notably, more than half of the ULXs show at least marginal evidence for curvature in their 2–10 keV spectra, which is somewhat unexpected if they are to be interpreted as the accreting ? 1000-M⊙ IMBHs suggested by modelling the soft spectral components as accretion discs. Physical modelling shows that this curvature is likely to originate in optically-thick coronae, which in turn leads to interpretations of the ULXs in terms of high accretion-rate stellar-mass (or slightly larger) BHs operating at around the Eddington limit. However, while we conclude that it is likely that the general ULX population does not have a large contribution from IMBHs, we obviously cannot rule out the possibility that some ULXs do possess IMBHs. Perhaps the best candidate on the basis of our spectral ?tting is M81 X-9, which is well ?tted by cool disc plus power-law/optically-thin corona models, and does not show explicit curvature in its 2–10 keV spectrum, though even this ULX may be ?tted using a hot (? 2.2 keV) accretion disc plus soft excess model. Clearly, it is dif?cult to ?nd unique solutions for these sources even with high quality XMM-Newton EPIC data. Ultimately, we may perhaps have to wait for radial velocity measurements from the optical counterpart of an ULX, leading to dynamical mass measurements of the compact accretor, before we have conclusive evidence whether any individual ULX does harbour an IMBH.

ACKNOWLEDGMENTS We thank an anonymous referee for their help in improving this paper. AMS and TPR gratefully acknowledge funding from PPARC. We thank Mike Goad for allowing us access to the Ho II X1 data while still proprietary, and for providing helpful comments on this manuscript. This work is based on observations obtained with XMM-Newton an ESA Science Mission with instruments and contributions directly funded by ESA member states and the USA (National Aeronautics and Space Administration).

REFERENCES Anders E., Grevesse N., 1989, GeCoA, 53, 197 Balucinska-Church M., McCammon D., 1992, ApJ, 400, 699 Begelman M. C., 2002, ApJ, 568, L97 Belczynski K., Sadowski A., Rasio F.A., 2004, ApJ, 611, 1068 Belloni T., M? ndez M., van der Klis M., Lewin W. H. G., Dieters e S., 1999, ApJ, 519, L159 Colbert E. J. M., Mushotzky R. F., 1999, ApJ, 519, 89 Colbert E. J. M., Ptak A. F., 2002, ApJS, 143, 25 Colbert E. J. M., Petre R., Schlegel E. M., Ryder S. D., 1995, ApJ, 446, 177 Coppi P. S., 1999, ASPC, 161, 375 Cropper M., Soria R., Mushotzky R. F., Wu K., Markwardt C. B., Pakull M., 2004, MNRAS, 349, 39 Crummy J., Fabian A.C., Gallo L., Ross R.R., 2006, MNRAS, 365, 1067 Dewangan G. C., Miyaji T., Grif?ths R. E., Lehmann I., 2004, ApJ, 608, L57 Dickey J. M., Lockman F. J., 1990, ARA&A, 28, 215 Done C., Gierli? ski M., 2005, MNRAS, 364, 208 n Done C., Kubota A., 2005, MNRAS, submitted (astro-ph/0511030) ˙ Ebisawa K., Zycki P., Kubota A., Mizuno T., Watarai K., 2003, ApJ, 597, 780

XMM-Newton observations of the brightest ULXs
Nayakshin S., Melia F., 1997, ApJ, 490, L13 Paturel G., Theureau G., Fouqu? P., Terry J. N., Musella I., e Ekholm T., 2002, A&A, 383, 398 Pietsch W., Mochejska B. J., Misanovic Z., Haberl F., Ehle M., Trinchieri G., 2004, A&A, 413, 879 Portegies Zwart S.F., Baumgardt H., Hut P., Makino J., McMillan S.L.W., 2004, Nature, 428, 724 Radecke H.-D., 1997, A&A, 319, 18 Ramsey C.J., Williams R.M., Gruendl R.A., Chen C.-H.R., Chu Y.-H., Wang Q.D., 2006, ApJ, in press (astro-ph/0511540) Read A. M., Ponman T. J., Strickland D. K., 1997, MNRAS, 286, 626 Roberts T. P., Warwick R. S., 2000, MNRAS, 315, 98 Roberts T. P., Warwick R. S., Ward M. J., Murray S. S., 2002, MNRAS, 337, 677 Roberts T. P., Colbert E. J. M., 2003, MNRAS, 341, L49 Roberts T. P., Warwick R. S., Ward M. J., Goad M. R., 2004, MNRAS, 349, 1193 Roberts T. P., Warwick R. S., Ward M. J., Goad M. R., Jenkins L. P., 2005, MNRAS, 357, 1363 Schlegel E. M., Petre R., Colbert E. J. M., Miller S., 2000, AJ, 120, 2373 Schlegel E. M., Barrett P., Singh K. P., 1997, AJ, 113, 1296 Schlegel E. M., Pannuti T. G., 2003, AJ, 125, 3025 Shakura N. I., Sunyaev R. A., 1973, A&A, 24, 337 Shrader C.R., Titarchuk L., 2003, ApJ, 598, 168 Soria R., Wu K., 2002, A&A, 384, 99 Stobbart A.-M., Roberts T. P., Warwick R. S., 2004, MNRAS, 351, 1063 Strickland D. K., Colbert E. J. M., Heckman T. M., Weaver K. A., Dahlem M., Stevens I. R., 2001, ApJ, 560, 707 Strohmayer T.E., Mushotzky R.F., 2003, ApJ, 568, L61 Str¨ der L. et al., 2001, A&A, 365, L18 u Swartz D.A., Ghosh K.K., Tennant A.F., Wu K., 2004, ApJS, 154, 519 Terashima Y., Wilson A. S., 2004, ApJ, 601, 735 Trinchieri G., Fabbiano G., Paulumbo G. G. C., 1985, ApJ, 290, 96 Tully R. B., 1988, Nearby Galaxies Catalog (Cambridge University Press, Cambridge) Turner M. J. L. et al., 2001, A&A, 365, L27 van der Klis M., 1995, in X-ray binaries, ed. W. H. G. Lewin, J. van Paradijs, & E. P. J. van den Heuvel (Cambridge: Cambridge University Press), 252 Vaughan S., Iwasawa K., Fabian A. C.,Hayashida K., 2005, MNRAS, 356, 524 Vogler A., Pietsch W., Bertoldi F., 1997, A&A, 318, 768 Wang Q. D., Yao Y., Fukui W., Zhang S. N., Williams R., 2004, ApJ, 609, 113 Wilms J., Nowak M. A., Dove J. B., Fender R. P., di Matteo T., 1999, ApJ, 522, 460 Zampieri L., Mucciarelli P., Falomo R., Kaaret P., Di Stefano R., Turolla R., Chieregato M., Treves A., 2004, ApJ, 603, 523 Zhang S. N., Cui W., Chen W., Yao Y., Zhang X., Sun X., Wu X., Xu H., 2000, Sci, 287, 1239 Zezas A., Fabbiano G., Rots A. H., Murray S. S., 2002a, ApJS, 142, 239 Zezas A., Fabbiano G., Rots A. H., Murray S. S., 2002b, ApJ, 577, 710


c 2004 RAS, MNRAS 000, 1–16


All rights reserved Powered by 甜梦文库 9512.net

copyright ©right 2010-2021。