Discovery of a Highly Variable Dipping Ultraluminous X-ray source in M94
Dacheng Lin, Jimmy A. Irwin, Natalie A. Webb, Didier Barret, Ronald A. Remillard
aa r X i v : . [ a s t r o - ph . H E ] N ov Draft version September 10, 2018
Preprint typeset using L A TEX style emulateapj v. 5/2/11
DISCOVERY OF A HIGHLY VARIABLE DIPPING ULTRALUMINOUS X-RAY SOURCE IN M94
Dacheng Lin , Jimmy A. Irwin , Natalie A. Webb , Didier Barret , and Ronald A. Remillard Draft version September 10, 2018
ABSTRACTWe report the discovery of a new ultraluminous X-ray source (ULX) 2XMM J125048.6+410743within the spiral galaxy M94. The source has been observed by
ROSAT , Chandra , and
XMM-Newton on several occasions, exhibiting as a highly variable persistent source or a recurrent transient with aflux variation factor of & ∼ L X ∼ × erg s − (0.2–10 keV, absorbed). In the brightest observation, the source is similar to typical low-luminosity ULXs, with the spectrum showing a high-energy cutoff but harder than that from a stan-dard accretion disk. There are also sporadical short dips, accompanied by spectral softening. In afainter observation with L X ∼ × erg s − , the source appears softer and is probably in the thermalstate seen in Galactic black-hole X-ray binaries (BHBs). In an even fainter observation ( L X ∼ × erg s − ), the spectrum is harder again, and the source might be in the steep-powerlaw state or thehard state of BHBs. In this observation, the light curve might exhibit ∼ HST images. In terms of the colors and the luminosity, the counterpart is probably a G8 supergiant ora compact red globular cluster containing ∼ × K dwarfs, with some possible weak UV excessthat might be ascribed to accretion activity. Thus our source is a candidate stellar-mass BHB with asupergiant companion or with a dwarf companion residing in a globular cluster. Our study supportsthat some low-luminosity ULXs are supercritically accreting stellar-mass BHBs.
Subject headings: accretion, accretion disks — black hole physics — X-rays: binaries — X-rays:individual:2XMM J125048.6+410743 INTRODUCTION
Many bright X-ray sources have been discovered innearby galaxies. Those with luminosities L X exceed-ing 10 erg s − , the Eddington limit for a black-hole X-ray binary (BHBs) with a black-hole (BH) massof M BH ∼ M ⊙ , are often referred to as ultralu-minous X-ray sources (ULXs, for a recent review seeFeng & Soria 2011). While a very small number of themappear as hyperluminous X-ray sources (HLXs, espe-cially ESO 243-49 HLX-1) with L X > erg s − and are strong candidates for intermediate-mass BHs(IMBHs) with M BH ∼ –10 M ⊙ (Farrell et al. 2009;Jonker et al. 2010; Servillat et al. 2011; Sutton et al.2012; Godet et al. 2012), most ULXs have L X < erg s − . The nature of these ’low-luminosity’ ULXs isstill under strong debate. They could be IMBHs (e.g.,Colbert & Mushotzky 1999), super-Eddington low-massBHs (e.g., Komossa & Schulz 1998), beamed emissionfrom geometrically thick disks at high accretion rates(King et al. 2001) or from relativistic jets (Okada et al.1998), or combination of these effects (Poutanen et al.2007; King 2009).Due to their large distances, there are no con-fident dynamical mass measurements for ULXs yet(Feng & Soria 2011), unlike Galactic BHBs, whose dy- Department of Physics and Astronomy, University of Al-abama, Box 870324, Tuscaloosa, AL 35487, USA, email:[email protected] CNRS, IRAP, 9 avenue du Colonel Roche, BP 44346, F-31028 Toulouse Cedex 4, France Universit´e de Toulouse, UPS-OMP, IRAP, Toulouse, France MIT Kavli Institute for Astrophysics and Space Research,MIT, 70 Vassar Street, Cambridge, MA 02139-4307, USA namical masses of the BHs are constrained to below 20 M ⊙ (Remillard & McClintock 2006). The spectral andtiming properties of Galactic BHBs have been well stud-ied, and they are known to exhibit three characteristicX-ray spectral states: the thermal state has a dominantthermal disk and weak fast variability, typically at lu-minosities above a few percent of the Eddington limit;the hard state has a dominant hard powerlaw (PL, withphoton index Γ PL typically within 1.4–2.1) and strongfast variability, generally seen at low luminosities; andthe steep-PL state has a powerful PL with Γ PL ∼ . erg s − , their X-ray spectra ap-peared harder than those from a standard geometricallythin optically thick disk. This is often seen in low-luminosity ULXs, and one possible explanation is Comp-ton up-scattering of disc photons in a wind or photo-sphere (Middleton et al. 2012, 2013). The above simi-larity led Middleton et al. (2012, 2013) to conclude thatthese two transient ULXs in M31 are members of low-luminosity ULXs, though the latter are mostly persistentwith small long-term variability (Feng & Soria 2011).However, these two ULXs are also similar to BHBs withstellar-mass BHs in other aspects, such as the outburstprofile and the coupled X-ray and radio behavior. Thus Lin et al. TABLE 1The X-ray Observation Log
Obs. ID Date Detector OAA T (ks) r src S/N(1) (2) (3) (4) (5) (6) (7)
XMM-Newton :0094360601(X1) 2002-05-23 pn/M1/M2 2.3 ′ ′′ ′ ′′ ′ ′′ Chandra : 808(C1) 2000-05-13 ACIS-S3 0.8 ′
47 1.5 ′′ ′
24 1.5 ′′ ROSAT :RP600050N00(R1) 1991-06-05 PSPCB 1.2 ′
91 20 ′′ ′
112 7 ′′ ′
27 7 ′′ Note . — Columns: (1) the observation ID with our designationgiven in parentheses, (2) the observation start date, (3) the detector,(4) the off-axis angle, (5) the exposures of data used in final analysis,(6) radius of the source extraction region, (7) the signal to noise ratioof the source, combining all detectors.
Middleton et al. (2012, 2013) concluded that many low-luminosity ULXs are stellar-mass BHBs.In Lin et al. (2012b), we classified 4330 sources fromthe 2XMMi-DR3 catalog (Watson et al. 2009). Inthis project, many sources showing interesting behav-ior but poorly studied in the literature were also dis-coverd. We are devoting a series of papers (e.g.,Lin et al. 2011, 2013a,b) to present the properties ofthese sources in detail. Here we continue our studyand concentrate on a highly variable ULX 2XMMJ125048.6+410743 in the spiral galaxy M94 (NGC 4736),which shows some evidence of dips, periodic oscil-lations, and a state transition, and has a possibleoptical counterpart from the
Hubble Space Telescope ( HST ) images. The source, whose 1 σ positional er-ror is 0 . ′′
35 per coordinate from the 2XMMi-DR3 cat-alog, is only 0 . ′′
26 away from CXO J125048.6+410742(RA=12h50m48.605s, Dec=+41d07 ′ ′′ , and the 95%error is 0 . ′′
29) in the
Chandra
Source Catalog (CSC, re-lease 1.1, Evans et al. 2010). Thus we conclude thatthey are the same source. It should also be the off-nuclear source detected in a
ROSAT /PSPCB observa-tion in Cui et al. (1997), because they coincide within 5 ′′ in position and there are no other bright sources within20 ′′ from Chandra observations. We designate the sourceas ULX2 hereafter, considering that there is a differentULX (U30 in Berghea et al. 2008), which is hardly vari-able and has a flux comparable to the peak value ofour source. In Section 2, we describe the analysis ofX-ray and optical observations. In Section 3, we presentthe results. In Section 4, we discuss the nature of thesource and the cause of its short-term X-ray variability.The conclusions of our studies are given in Section 5.Throughout the paper, we assume a source distance of5.2 Mpc (Tonry et al. 2001). DATA ANALYSIS
X-ray Observations
M94 has been observed by many X-ray observatories.However, because ULX2 is faint and is only 1 ′ awayfrom the nucleus around which there are strong dif-fuse emission and many bright point sources, includingthe low-ionization-nuclear-emission-region (LINER) nu-cleus, we only studied the eight observations (Table 1) from XMM-Newton , Chandra and
ROSAT , thanks totheir relatively large effective areas and high angularresolutions. We designate their observations as X1-X3,C1-C2, and R1-R3, respectively (refer to Table 2). Inthe
XMM-Newton observations, all the three EuropeanPhoton Imaging Cameras, i.e., pn, MOS1 (or M1), andMOS2 (or M2) (Jansen et al. 2001; Str¨uder et al. 2001;Turner et al. 2001) were active. We used SAS 12.0.1 andthe calibration files of 2013 January for reprocessing theX-ray event files and follow-up analysis. The data instrong background flare intervals are excluded follow-ing the SAS thread for the filtering against high back-grounds. The event selection criteria followed the de-fault values in the pipeline (see Table 5 in Watson et al.2009). That is, for pn spectra, we used events with PAT-TERN ≤ ≤
12 and (FLAG & 0xfffffeff) = 0. The two
Chan-dra observations used the imaging array of the AXAFCCD Imaging Spectrometer (ACIS; Bautz et al. 1998).ULX2 falls in the back-illuminated chip S3 and the front-illuminated chip I2 in observations C1 and C2, respec-tively. We analyzed the data with the CIAO (version4.5) package and the latest calibration (CALDB 4.5.5.1).The
ROSAT (Truemper 1982) observations were reducedwith FTOOLS 6.12.We extracted the source emission from a circular re-gion with the radius for each observation given in Table 1.The background in the XMM-Newton and ROSAT obser-vations strongly depends on the distance to the nucleusdue to the presence of bright sources near the nucleusand the relatively large point spread functions (PSFs) ofthese two observatories, and we estimated it using fourcircular regions with the same radius and the same dis-tance to the nucleus as the source region. The
Chan-dra observations have no such problem thanks to theirsupreme angular resolutions, and we used a single circu-lar region of 10 ′′ radius near the source to estimate thebackground. We extracted both the source and back-ground spectra/light curves from the source and back-ground regions, respectively. The response files wereconstructed for the spectral fitting. We note that wetook special care to construct the response file for the ROSAT /PSPCB observation R1. The problem is that weused a small source extraction region (a radius of 20 ′′ ) inorder to minimize the contamination from bright sourcesnear the nucleus, leading to significant PSF loss, which isenergy dependent. The response file (we used the defaulton-axis version in the HEASARC archive) does not takeinto account the PSF loss, and there is no FTOOLS toolto correct for this. We estimate the PSF loss using theROSAT observation RP700130N00 of Mkn 501 (it usedthe same gain epoch and has a similar (low) column den-sity and photon index ( ∼ χ statistic for spectral fits.All spectral models used include the absorption describedby the WABS model in XSPEC, with the lower limit of TABLE 2The
HST /WFPC2 Observation Log
Exposure ID Date Chip & Filter T S / N VEGA Mag(s) Counterpart G8I K3V(1) (2) (3) (4) (5) (6) (7) (8)u96r060[1 | ± | ± | ± ± | ± Note . — Columns: (1) the exposure ID of the data set, (2) the observationdate, (3) the chip and filter, (4) the total exposure, (5) the signal to noiseratio, (6) the VEGA mag of the optical counterpart to ULX2, (7–8) magni-tudes estimated based on a G8I stellar spectrum and a K3V stellar spectrum,respectively. N H set to be the Galactic line-of-sight value of 10 cm − (Kalberla et al. 2005).To help to identify the optical counterpart (Sec-tion 2.2), we used X-ray sources within 2 ′ from the nu-cleus to improve the relative astrometry between Chan-dra and
HST . To have good relative positions for theseX-ray sources, we carried out the source detection usingthe CIAO task wavdetect (Freeman et al. 2002). Weused only the
Chandra observation C1 so that all sourcesof interest are in a single CCD, which is not the case forobservation C2. We performed the detection on the 0.3–8 keV image, which was binned at a subpixel resolution(1 / HST Observations
The sky region around ULX2 was observed by theWide Field and Planetary Camera 2 (WFPC2) aboardthe
HST several times (Table 2). We aligned their driz-zled images by matching the point sources to those in theF555W image, resulting in the rms deviations < ′′ inall cases. To map the X-ray positions onto the HST im-ages, we assumed that CXO J125053.0+410713 in theCSC and the UV source at the center of the
HST imagesare the nuclear source and coincident with each otherin position (see, e.g., Constantin & Seth 2012). Becausethe nuclear UV source is stronger and less subject tostellar contamination at shorter wavelength and thus ap-pears more clear in the F336W image than in the otherimages, we used the F336W image to determine its po-sition with the astrolib IDL procedure cntrd . We foundthat it happens to be only 0 . ′′
02 away from the position ofCXO J125053.0+410713 determined by us (Section 2.1).After correcting for this small relative astrometry, wefound an optical point source that is only 0 . ′′ . ′′
1, supporting our astrometry correc-tion. We performed the photometry on the c0f imageswith the HSTPHOT 1.1 package (Dolphin 2000), whichoutputs aperture-corrected VEGA magnitudes. RESULTS
Long-term X-ray Variability
Table 1 lists the signal to noise ratio (S/N) of ULX2 ineach observation. The source was significantly detectedin observations X3, C1, C2, and R1 (S/N ≥ Fig. 1.—
The long-term luminosity curve (0.2–10 keV, absorbed),with
ROSAT (red crosses),
XMM-Newton (green triangles), and
Chandra (blue squares) observations. We note that the third
ROSAT observation (R3), with L x = 0 +0 . × erg s − , hasonly the error bar seen and is very close to the second one (R2,both in 1994 December). R3 (S/N ≤ L X ∼ × erg s − in observation X3, sup-porting ULX2 as a low-luminosity ULX. This luminosityis a factor of ∼
100 of that of observation X1 (we did notuse observations R2 and R3 for comparison because theirluminosities have larger uncertainties). Considering thatobservations R2 and R3 are close in time and that X1and X2 are close in time too, ULX2 was observed in es-sentially six epochs and detected in four (or in five if thedetections in X1 and X2 are really from ULX2) of themover nearly two decades. Thus the source has a largeduty cycle ( ∼
70% or more) and is probably a highlyvariable persistent source or a recurrent transient withprobably at least three outbursts in the last two decades(see Figure 1). Considering that observations X3 and C2are about one year apart, the outburst should last morethan one year if they are in the same outburst.
Short-term X-ray Variability
The left panels of Figure 2 show the light curves ofthe four observations when ULX2 was significantly de-tected, in order of decreasing luminosity from the top to Lin et al.
Fig. 2.—
The light curves (left panels) and unfolded spectra (right panels) of the four observations (the luminosity decreases from thetop to the bottom panels) when ULX2 was significantly detected. The notation in the left panels includes the observatory, the observationID, the instrument, the light curve bin size, and the energy band used, and that in the right panels is the spectral model used. The totalmodel is shown as a black solid line, the MCD component as a red dotted line, the nthComp component as a blue dashed line, and the PLcomponent as a green dot-dashed line. the bottom panels. The brightest observation X3 seemsto experience dipping behavior sporadically (15 ks and25 ks into the observation), with duration a few hun-dred seconds. To understand the spectral properties inthe dipping periods, we extracted the spectrum in thedipping period when the pn 1-10 keV count rate is < − (refer to Figure 2). The spectrum turns out tobe soft. When we fitted it with a multi-color disk (MCD,diskbb in XSPEC), we obtained the inner disk temper-ature kT MCD = 0 . ± .
06 keV (the column density is N H = 1 +5 × cm − , and the reduced χ is relativelyhigh (2.0) for 6 degrees of freedom (d.o.f.)). In com-parison, the total spectrum gives kT MCD = 0 . ± . ± ± . ± .
08 using 0.2–1 keV events and 0 . ± . ∼ σ , 6 σ , and 7 σ , respec-tively. We also extracted the spectra in the bright andfaint intervals, depending on whether the count rate islarger or smaller than 0.005 counts s − . When we fittedthem with a PL, we obtained their photon indices to beΓ PL =2.5 ± ± N H at6 × cm − obtained from the fit to the total spectrum(Section 3.3) because of the low counting statistics of thespectra). We also calculated the hardness ratio, defined Fig. 3.—
The
HST /WFPC2 images around ULX2. The top panel is false-colored using the F814W (red), F555W (green) and F336W(blue) images. The bottom panels, one for each of the five filters, are zoomed in on ULX2. The green circle, with a radius of 0 . ′′
39 (the95% systematic error of
Chandra (Rots & Budav´ari 2011)), is centered at the X-ray position, indicating the presence of a possible opticalcounterpart to ULX2. as the count rate in 1–8 keV divided by that in 0.3–1 keVand obtained the values of 0.71 ± ± Spectral Modeling
The brightest observation X3 has the best qualityand allows for relatively detailed modeling. FollowingStobbart et al. (2006), we fitted the 2–10 keV spectrumwith a PL and a broken PL (both unabsorbed) and foundthat the latter fit (the total χ is 98.2 for 118 d.o.f.)showed a significant (5 σ ) improvement over the formerone (the total χ is 125.2 for 120 d.o.f.), indicating thepresence of spectral curvature at several keV (the inferredbreak energy is 4 . ± . χ of 1.31 (560 d.o.f., Table 3). The fit with theMCD model is better (the reduced χ is 1.15), but thereare still clear residuals at high energies, as is often seenin the spectral fits to ULXs and was explained as dueto the advection effect and/or the Compton scatteringof disk photons in an optically thick wind, photosphereor corona in many studies, such as Gladstone et al.(2009) and Middleton et al. (2012, 2013). Similar tothese studies, we tested two-component models to in- clude a Comptonization component, i.e. MCD+PL andMCD+CompTT (CompTT (Titarchuk 1994) is availablein XSPEC). The fit with the MCD+PL model is muchbetter than that with the MCD model, decreasing the to-tal χ by 114. The MCD+CompTT model, with the seedphotons tied to the inner disk temperature, decreasesthe total χ further, by 32 compared with the MCD+PLmodel for one more d.o.f. The MCD+CompTT modelinferred a population of cool (1.0 keV) Comptonizingelectrons with a high optical depth ( τ ∼ kT MCD = 0 . +0 . − . keV (90% error) or kT MCD = 0 . +0 . − . keV (2 σ error) (here we report twoerror bars because we found that χ varied slowly for arelatively large range of kT MCD ).We also tested the Comptonization model nthComp(in XSPEC, ˙Zycki et al. 1999; Zdziarski et al. 1996;Lightman & Zdziarski 1987), whose seed photons canbe assumed to have the MCD shape. This modelwas used in the fits to very bright (around the Ed-dington limit) spectra from the ρ state of the BHBGRS 1915+105 by Neilsen et al. (2011) and those fromthe luminous neutron-star (NS) low-mass X-ray binary(LMXB) GX 17+2 by Lin et al. (2012a) as a substitu-tion of the SIMPL model (Steiner et al. 2009) to ac-count for the high-energy cutoff in the spectra. The Lin et al. TABLE 3Spectral fit results
Obs ID model N H Other Parameters χ ν ( ν )a L abs L unabs(1022 cm −
2) (1038 erg s − · · · ± ± · · · ± ± . . − .
01 ΓPL=2 . . − . N PL=22+1 − × − . . . − . . . − . . . kT MCD=0 . . − . N MCD=5 . . − . × − . . . − . . . − . . . − . kT MCD=0 . . − . N MCD=3 . . − . × −
2, ΓPL=2 . . − . N PL=6+2 − × − . . . − . . . − . . . kT MCD=0 . . − . N MCD=0 . . − .
23, ΓnthComp=1 . . − . kT e , nthComp=1 . . − . N nthComp=9+7 − × − . . . − . . . − . . . kT MCD=0 . . − . N MCD=1 . . − . τ compTT=11+1 − kT e , compTT=1 . . − . N compTT=2 . . − . × − . . . − . . . − . . .
07 ΓPL=2 . . − . N PL=8+2 − × − . . . − . . . − . . . kT MCD=0 . . − . N MCD=3 . . − . × − . . . − . . . − . . . kT MCD=0 . . − . N MCD=2 . . − . × − . . . − . . . − . . . − .
11 ΓPL=2 . . − . N PL=6+3 − × − . . . − . . . − . . . kT MCD=0 . . − . N MCD=0 . . − .
13 1 . . . − . . . − . . .
01 ΓPL=2 . . − . N PL=9 . . − . × − . . . − . . . − . · · · ± ± · · · ± ± Note . — The fits were carried out only on observations X3, C1, C2, and R1. The 0.2-10 keV absorbed ( L abs) and unabsorbed ( L unabs) luminsosities are given. Forobservations X1, X2, R2, and R3, they were estimated based on the PL fit to observation C1. The energy bands of the fits are 0.2–10 keV for X3, 0.3–8 keV for C1and C2, and 0.2–2.4 keV for R1. All errors are at a 90%-confidence level. fitting results with MCD+nthComp model are given inTable 3 and the unfolded spectrum is shown in Fig-ure 2. The MCD normalization is not well constrained,with the 90%-confidence lower limit reaching zero, whichwas also seen in Neilsen et al. (2011) and Lin et al.(2012a). Following Lin et al. (2012a), we roughlyestimated the pre-Comptonization disk normalization N disk , pre − Compton =( R in , km /D ) cos i , where R in , km is the inner disk radius, D the source distance, and i the disk inclination, by adding the photons Comptonscattered (the nthComp component) and those unscat-tered (the MCD component), assuming that the photonsare conserved. We obtained N disk , pre − Compton =0 . +0 . − . .The optical depth of the corona, which is not an ex-plicit parameter of the model but can be derived usingEquation A1 in Zdziarski et al. (1996), is ∼
22, which ishigh, similar to that inferred from the MCD+CompTTmodel. Therefore we have made the simple assumptionthat while the optically thick corona can have seed pho-tons from the disk it does not strongly affect the thermaldisk emission in turn in both the MCD+CompTT andMCD+nthComp models.The other spectra have much lower quality, and weonly test the single-component models, i.e., a MCD anda PL. The unfolded spectra are plotted in Figure 2.The
ROSAT /PSPCB observation R1 can be fitted witheither model. We caution that the fit for this obser-vation is limited by the narrow energy band coverage(0.2–2.4 keV) and calibration uncertainty. The
Chan-dra observation C2 can also be fitted with either model,but the PL model requires a relatively strong absorption( N H =0.24 × cm − ) and a steep PL (Γ PL =2.9). Ob-servation C1, the one with (quasi-)periodic flares (Fig-ure 2), can be fitted with a PL, but not well with theMCD model (the reduced χ is 1.9). The inferred pho-ton index is Γ PL = 2 . +0 . − . .We did not carry out spectral fits to the other fourobservations (R2, R3, X1, and X2) when the source wasnot significantly detected and subject to strong contami-nation from bright sources near the nucleus. Instead, weestimated their source luminosities assuming the best- fitting PL model from observation C1, since C1 was theclosest in flux to these observations (Table 3). We notethat for observations R2 and R3, in which the source wasnot detected, we calculated the 90% confidence intervalsusing Bayesian statistics (Kraft et al. 1991). The Optical Counterpart
The candidate optical counterpart to ULX2 fortu-nately appears on the outskirts of a star-forming regiondevoid of bright sources.
HSTPhot indicates this counter-part as point-like in all filters. Its VEGA magnitudes aregiven in Table 2. The counterpart seems relatively red.It was the most significantly detected in the F555W andF814W filters. Further considering that these two filtersare less subject to possible emission from accretion activ-ity than the other filters, we used the SYNPHOT packageto compare the color of these two filters with the stellarspectra in Pickles (1998), assuming the Galactic extinc-tion of E (B − V)=0.02 (Schlegel et al. 1998). We foundthat the counterpart has the F555W − F814W color (1.21mag) the closest to those of G8I (1.09 mag) and K3V(1.16 mag) stars among supergiants and dwarfs in Pickles(1998), respectively. We fitted the F450W, F555W, andF814W fluxes of the counterpart with these two stellarspectra by minimizing the total χ , with the normaliza-tion of the spectra as a free parameter. The correspond-ing apparent magnitudes of the best-fitting spectra arelisted in Table 2. The fit residuals are . H α filter F656N, the deviation is slightlylarger, with our source brighter than the fits by 0.3–0.4mag (2–3 σ ). In the UV filter F336W, the excess is muchlarger, by 1.5 and 0.7 mag compared with the fits witha G8I star and a K3V star, respectively, though in thisfilter the source was detected only at 6 σ . The apparentvisual magnitude of our source is V =22.22, correspond-ing to an absolute magnitude of − . ∼ × K3Vstars. In the above, we have neglected the possible effectof binary evolution on the color of the donor star. DISCUSSION
ULX2 as a Super-Eddington Accreting Stellar-massBHB
ULX2 was fortunately captured at different flux levels,showing clear spectral evolution. With similar spectralmodels tested, we can easily compare ULX2 with thesecond ULX in M31 (XMMU J004243.6+412519) stud-ied by Middleton et al. (2013). The brightest observa-tion X3 of ULX2 is very similar to the brightest
XMM-Newton observations XMM3 and XMM4 of XMMUJ004243.6+412519. All these observations have L X ∼ erg s − and are better fitted by the MCD+CompTTmodel than by a PL, a MCD, or their combination.The fits with the MCD+CompTT model to these ob-servations all inferred a relatively cool ( < ∼ τ ∼
10) corona. XMMUJ004243.6+412519 is probably a stellar-mass BHB with M BH ∼ M ⊙ , based on the joint radio/X-ray behav-ior and the observation of a disk-dominated state at avery low luminosity, supporting that the above spectraare probably characteristic of accretion at the Edding-ton limit (Middleton et al. 2013). Then ULX2 couldbe a stellar-mass BHB as well, with an accretion ratearound the Eddington limit in observation X3. Manylow-luminosity ULXs show similar spectra and couldalso be explained as supercritically accreting stellar-massBHBs (Middleton et al. 2013).If ULX2 is really a stellar-mass BHB, we would expectit to behave similar to typical Galactic BHBs when it isat well sub-Eddington luminosities. In the fainter obser-vation C2 ( L X ∼ . × erg s − ), the spectrum issofter than observation X3 and can be described with astandard thermal accretion disk, thus consistent with be-ing in the thermal state of Galactic BHBs. The thermalstate in Galactic BHBs tends to occur above ∼
3% of theEddington limit (Dunn et al. 2010). Assuming observa-tion C2 to be in this state and using its 0.2–10 keV unab-sorbed luminosity from the MCD model, we constrainedthe mass of the BH in ULX2 to be . M ⊙ , support-ing the identification as a stellar-mass BHB. We notethat for this observation we cannot rule out a steep PLmodel with relatively strong absorption from the fit. Inthe even fainter observation C1 ( L X ∼ × erg s − ),the spectrum becomes harder again and can be describedwith a PL with Γ PL ∼ . ± .
5. Such a photon index isoften seen in the steep-PL state of Galactic BHBs, andthe source might be in such a state in this observation,but considering its large uncertainty, we cannot rule outthe source being in the hard state instead. In any case,we might have observed a state transition often seen inBHBs. We clearly need higher quality data for more de-tailed comparison.Therefore transient/highly variable ULXs like ULX2and XMMU J004243.6+412519 serve a link between per-sistent low-luminosity ULXs and classical stellar-massBHBs. This is reminiscent of the first known transientZ source XTE J1701-462 linking Z and atoll sources,the two main classes of weakly magnetized NS LMXBs(Lin et al. 2009b; Homan et al. 2010). Here we brieflydiscuss the NS LMXB case to gain some insights intothe physics involved in the BHB case. Z sources reachthe Eddington limit or above and are mostly persistentwith variability factors of only a few, while atolls have lu- minosities typically less than 50% of the Eddington limitand tend to be transient or highly variable. The proofthat Z sources accrete at near or super-Eddington limitand is the same class of object as atolls but at differentaccretion rates based on XTE J1701-462 is very convinc-ing in several ways (Lin et al. 2009b; Homan et al. 2010).First, Z and atoll sources have well-known distinct tim-ing and spectral properties, and XTE J1701-462 showedZ-source properties at high luminosities and then atoll-source ones during the decay in its 2006-2007 outburst.Secondly, NS LMXBs have a unique way to infer the Ed-dington luminosity, i.e. radius expansion Type-1 X-raybursts, which were detected in XTE J1701-462, and theluminosities when XTE J1701-462 behaved as a Z sourcewere indeed near or above the Eddington luminosity in-ferred from the bursts (Lin et al. 2009a). Finally, thespectral fitting by Lin et al. (2009b) also gathered someevidence that the Eddington limit was reached in the Z-source stage, especially the result that the disk in theZ-source lower vertex had a relatively constant tempera-ture and an increasing inner disk radius with increasingluminosity, deviating from the L ∝ T trend observedin the atoll-source stage. The increase in the inner diskradius with luminosity is expected because as the massaccretion rate into the disk increases, more and more in-ner part of the disk reaches the local Eddington limit,leading to increasing radial advection flow and/or massoutflow (Ohsuga & Mineshige 2007).Therefore studies of NS LMXBs not only demonstratethat near or super-Eddington accretion is possible butalso assure us that objects with different outbursting be-haviors and luminosities can belong to the same classand have overall spectral/timing properties mainly de-termined by the accretion rate. All this supports the pos-sibility that most low-luminosity ULXs are in fact nearor super-Eddington accreting stellar-mass BHBs. Galac-tic BHBs are known to show many similarities to atolls,such as sub-Eddington luminosity, transient behavior,broad-band noise, and possibly also disk spectral evo-lution (van der Klis 2006; Lin et al. 2007), though thereare also some differences, such as hotter thermal spec-tra and stronger millisecond variability in atolls, whichcould be reasonably ascribed to emission from the impactof materials onto the NS surface. Then there is a ques-tion of whether ULXs that are supercritically accretingstellar-mass BHBs also show some properties observedin Z sources. One possible interesting similarity is theirgenerally small long-term variability. Classical Z sources(Sco X-1, GX 17+2, GX 349+2, GX 340+0, GX 5-1, andCyg X-2), all with a low-mass companion, are persistentwith long-term variation factors of only a few (only two Zsources (XTE J1701-462 and IGR J17480-2446), discov-ered recently, are transients). Most known ULXs are alsopersistent. Lin et al. (2012b) found only 15 with a long-term variation factor >
10 in their 100 ULXs, which isin contrast with the general transient behavior of Galac-tic BHBs, especially those with a low-mass companion.This could be because most of these ULXs might have ahigh-mass companion, but from Z sources we cannot ruleout that the long-time activity might be related to highaccretion rates. It would also be worthwhile to searchfor similarity of the disk behavior between ULXs and Zsources. Neilsen et al. (2011) had reported similar diskevolution in the BHB GRS 1915+105 in a very bright Lin et al.state, i.e., the ρ state, to that seen in XTE J1701-462in the Z-source stage. A lot of work is still needed inthe future to establish the possible connection betweenZ sources and low-luminosity ULXs. Cause of Short-term X-ray Variability
The observations of dips from ULX2 make it oneof the dipping sources, which include about 20 Galac-tic X-ray binaries and a few ULXs, such as the ULXin NGC 55 (Stobbart et al. 2004) and NGC 5408 X-1(Pasham & Strohmayer 2013; Gris´e et al. 2013). Longdips lasting for 10%–30% of the orbital phase are mostlyfound in NS LMXBs and can be explained as due to ab-sorption by a bulge on the edge of the accretion diskat the point where the gas stream impacts the disk(e.g., White & Swank 1982). Short dips less than afew hundred seconds (thus typically <
1% of the or-bital period) are commonly observed in two BHBs GROJ1655-40 (with a low-mass companion) and Cyg X-1(with a supergiant companion). They show orbital phasedependence and could be due to absorption in accre-tion streams from the companion (Kuulkers et al. 2000;Ba luci´nska-Church et al. 2000; Feng & Cui 2002). It isnot clear whether the dipping ULXs in NGC 55 andNGC 5408 also show orbital phase dependence, but theymight have similar origin. Dips in ULX2 are short andare thus probably also due to absorption in accretionstreams. The spectrum becomes soft in the dips in ob-servation X3, which could be because a hot componentis obscured and/or because the absorbing matter is par-tially ionized. In terms of the MCD+nthComp modelthat we used to fit observation X3, the dipping spec-trum that we created from this observation could be veryroughly accounted for with either complete obscurationof the nthComp component (the reduced χ is 3.7 for7 d.o.f.) or with the presence of absorbing matter witha column density of 6 . × cm − and ionization pa-rameter of log ξ = 2 . χ is 4.3 for 7 d.o.f.,using the ionized absorption model zxipcf in XSPEC).We do not have enough dips to search for periodicity,which could otherwise provide information of the orbitalperiod. In any case, the dips often imply that the systemis at a high inclination. Detection of ULXs at high incli-nations is important, because it poses a problem to usethe beaming effect to explain the ultraluminous natureof these sources.The possible large (quasi-)periodic X-ray modula-tions/flares in observation C1 is also a special propertyof ULX2. Similar modulations were also observed insome ULXs such as CXOU J141312.3-652013 in Circinus( ∼ ∼ ∼ L X ∼ erg s − while thosesources have L X > erg s − when such large mod-ulations were observed. Various explanations for suchmodulations were discussed by those studies, includingthe binary eclipse, modulations of the accretion rate dueto some instability, and variability at the base of the jet.Because ULX2 might have a high inclination consideringthe detection of dips, if the flares are periodic, the binaryeclipsing explanation seems plausible, with the ∼ Nature of the Optical Counterpart
Another interesting aspect of ULX2 is our identifi-cation of its point-like red optical counterpart candi-date. Some ULXs have been reported to have opticalcounterpart candidates (Tao et al. 2011; Gladstone et al.2013, e.g.,). Most of them appear blue and are probablycontaminated by accretion activity, while some have aclear red component like our source, such as IC 342 X-1(Feng & Kaaret 2008) and M81-ULS1 (Liu & Di Stefano2008). For M81-ULS1, there is also an extra blue com-ponent which was shown by Liu & Di Stefano (2008) tobe probably from disk emission, while the red componentcould be an asymptotic giant branch star. There is somepossible UV excess in ULX2, and the most likely expla-nation is the disk emission too. We note that our sourceis highly variable while the
HST observations are not si-multaneous with each other or with X-ray observations.Therefore, one explanation for why the counterpart toULX2 appears red while those to other ULXs are mostlyblue is that the
HST observations of ULX2 were madewhen it was not X-ray ultraluminous, though the alter-native explanation that ULX2 resides in an old cluster isalso possible. CONCLUSIONS
We have shown many intriguing properties of ULX2.It is either a highly variable persistent source or a re-current transient, with a long-term variation factor ofat least ∼