Spectral evolution of the ultraluminous X-ray sources M82 X-1 and X-2
Murray Brightman, Dominic J. Walton, Yanjun Xu, Hannah P. Earnshaw, Fiona, A. Harrison, Daniel Stern, Didier Barret
DDraft version January 22, 2020
Preprint typeset using L A TEX style emulateapj v. 12/16/11
SPECTRAL EVOLUTION OF THE ULTRALUMINOUS X-RAY SOURCES M82 X-1 AND X-2
Murray Brightman , Dominic J. Walton , Yanjun Xu , Hannah P. Earnshaw , Fiona, A. Harrison , DanielStern , Didier Barret , Cahill Center for Astrophysics, California Institute of Technology, 1216 East California Boulevard, Pasadena, CA 91125, USA Institute of Astronomy, Madingley Road, Cambridge CB3 OHA, UK Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA Universite de Toulouse, UPS-OMP, IRAP, Toulouse, France CNRS, IRAP, 9 Av. colonel Roche, BP 44346, F-31028 Toulouse cedex 4, France
Draft version January 22, 2020
ABSTRACTM82 hosts two well-known Ultraluminous X-ray sources (ULXs). X-1, an intermediate-mass black hole(IMBH) candidate, and X-2, an ultraluminous X-ray pulsar (ULXP). Here we present a broadbandX-ray spectral analysis of both sources based on ten observations made simultaneously with
Chandra and
NuSTAR . Chandra provides the high spatial resolution to resolve the crowded field in the 0.5–8keV band, and
NuSTAR provides the sensitive hard X-ray spectral data, extending the bandpass ofour study above 10 keV. The observations, taken in the period 2015–2016, cover a period of flaringfrom X-1, allowing us to study the spectral evolution of this source with luminosity. During four ofthese observations, X-2 was found to be at a low flux level, allowing an unambiguous view of theemission from X-1. We find that the broadband X-ray emission from X-1 is consistent with that seenin other ULXs observed in detail with
NuSTAR , with a spectrum that includes a broadened disk-likecomponent and a high-energy tail. We find that the luminosity of the disk scales with inner disktemperature as L ∝ T − / contrary to expectations of a standard accretion disk and previous results.These findings rule out a thermal state for sub-Eddington accretion and therefore do not support M82X-1 as an IMBH candidate. We also find evidence that the neutral column density of the materialin the line of sight increases with L X , perhaps due to an increased mass outflow with accretion rate.For X-2, we do not find any significant spectral evolution, but we find the spectral parameters ofthe phase-averaged broadband emission are consistent with the pulsed emission at the highest X-rayluminosities. Keywords: black hole physics – X-rays: binaries – X-rays: individual (M82 X-1) – X-rays: individual(M82 X-2) INTRODUCTION
Ultraluminous X-ray sources (ULXs), first discoveredby the
Einstein X-ray observatory (Giacconi et al. 1979;Fabbiano 1989), are observed as bright sources of X-rayswhich appear to exceed the Eddington limit of the typi-cal 10 M (cid:12) black holes found in our own Galaxy. ULXsare not coincidental with the nuclei of their host galax-ies, so are not be powered by the supermassive blackholes (SMBHs) that power active galactic nuclei (AGN),although ∼
25% of candidate ULXs are likely to be back-ground AGN (e.g. Walton et al. 2011; Sutton et al. 2015;Earnshaw et al. 2019). At first ULXs were promisingintermediate mass black hole (IMBH) candidates, sincelarger black holes can radiate at higher luminosities dueto the Eddington limit scaling with mass (e.g. Colbert& Mushotzky 1999; Miller et al. 2003). But as moreobservations were made, the data did not appear con-sistent with this scenario, and instead, lower-mass blackholes accreting at super-Eddington luminosities were fa-vored for most sources (e.g. Mizuno et al. 1999), with afew IMBH candidates still remaining (e.g. ESO 243-49HLX-1, Farrell et al. 2009).M82 (The ‘cigar galaxy’, or NGC 3034, Watson et al.1984) hosts two well-known ULXs, that since are sepa-rated by only 5 (cid:48)(cid:48) on the sky, were only first resolved by
Chandra (Matsumoto et al. 2001). The history of stud- ies of M82 X-1 (CXOU J095550.2+694047), the bright-est ULX in M82, embodies the above narrative. Ithas long been considered one of the best IMBH candi-dates because of its high luminosity, which can reach ∼ erg s − (e.g. Ptak & Griffiths 1999; Rephaeli & Gru-ber 2002; Kaaret et al. 2006); detection of low-frequencyquasi-periodic oscillations (QPOs) in the power spec-trum (54 mHz, Strohmayer & Mushotzky 2003; Dewan-gan et al. 2006; Mucciarelli et al. 2006), indicative of acompact, unbeamed source; as well as twin-peaked QPOsat 3.3 and 5.1 Hz, which lead to a mass estimate of400 M (cid:12) using scaling laws between the QPOs frequen-cies and mass used for stellar-mass black holes (Pashamet al. 2014). Additionally, Feng & Kaaret (2010) (here-after F10) observed the source with XMM-Newton and
Chandra over the course of a flaring episode and fittedthe spectra with the standard thin accretion disk model.They found that the luminosity of the disk, L , scaledwith inner temperature as L ∝ T which is expectedfrom a thin accretion disk with a constant inner radius.From this they inferred a black hole mass in the range300 − M (cid:12) , assuming that the black hole is rapidlyspinning in order to avoid extreme violations of the Ed-dington limit, therefore adding support to the IMBH sce-nario.The luminosity argument no longer stands however, a r X i v : . [ a s t r o - ph . H E ] J a n Brightman et al. since the ULX NGC 5907 ULX1, which also reaches ∼ erg s − (Sutton et al. 2013; F¨urst et al. 2017),and was once considered an IMBH candidate, was dis-covered to be powered by a neutron star with only 1–2 M (cid:12) (Israel et al. 2017) from the detection of coherentpulsations. The mass measurement from the twin QPOsis ambiguous too, since it is not known where these orig-inate and it is not clear if the scaling relationship usedextends to the IMBH range. In Brightman et al. (2016a),hereafter B16, we presented a combined spectral analysisof X-1, also during a flaring episode, using simultaneousobservations with Chandra , NuSTAR and
Swift . Withbroader band data than Feng & Kaaret (2010), we foundthat the temperature profile as a function of disk radius( T ( r ) ∝ r − p ) is significantly flatter than expected fora standard thin accretion disk as implied by F10, andinstead characteristic of a slim disk that is expected athigh accretion rates. Since only one observation was ana-lyzed, the L ∝ T relationship could not be tested. Nev-ertheless, the mass estimates inferred from the inner diskradius for this model imply a stellar-remnant black hole( M BH =26 +9 − M (cid:12) ) when assuming zero spin, or an IMBH( M BH =125 +45 − M (cid:12) ) when assuming maximal spin.M82 X-2 (CXOM82 J095551.1+694045) is typicallythe second brightest X-ray sources in the galaxy, and wasthe first ultraluminous X-ray pulsar discovered (Bachettiet al. 2014) using NuSTAR (Harrison et al. 2013). Withonly a 5 (cid:48)(cid:48) angular separation from X-1, studying the spec-tral properties of this source in detail has been limited to
Chandra observations. We conducted a study of the spec-tral and temporal properties of M82 X-2 in Brightmanet al. (2016b) finding that the source’s luminosity variesover two orders of magnitude over the range 10 − erg s − . Using timing analyses, we were able to isolatethe pulsed emission from this source with NuSTAR , find-ing that it was well described by a cutoff power-law. Ina follow up study, we found evidence that the variationsin luminosity are modulated on a ∼ ∼ Chandra which took placein 2016. The primary goals of this campaign were toperform a temporal analysis of X-1 and X-2 to search fororbital and super-orbital modulations; to perform spec-troscopic studies of the ULXs; and to study the nature ofthe other binary systems in M82. We report the tempo-ral analysis in Brightman et al. (2019) and here we focuson the second objective, to present the most comprehen-sive X-ray spectral analysis of the two ULXs, M82 X-1and X-2, to date. We combine simultaneous observationswith
Chandra to spatially resolve the two sources fromeach other and the other sources in M82, and simulta-neous observations with
NuSTAR to extend the spectralcoverage up to 79 keV. We also present results from
Swift monitoring observations of M82 which have been ongoingsince 2012. Throughout this paper we assume a distanceto M82 of 3.3 Mpc (Foley et al. 2014) which is inferredfrom the lightcurve of SN2014J. OBSERVATIONS AND DATA REDUCTION
Table 1
Observational dataObservatory Start date Start time ObsID Exposure(UT) (s)
NuSTAR
Chandra
NuSTAR
Chandra
Chandra
NuSTAR
NuSTAR
Chandra
NuSTAR
Chandra
NuSTAR
Chandra
NuSTAR
Chandra
NuSTAR
Chandra
Chandra
NuSTAR
NuSTAR
Chandra
All
Chandra and
NuSTAR observations studied herewere taken simultaneously, or quasi-simultaneously(within 24 hours of each other). Table 1 provides a de-scription of the observational data. The following sec-tions describe the individual observations and data re-duction.
Chandra
Since the angular separation of X-1 and X-2 is only5 (cid:48)(cid:48) , only
Chandra (Weisskopf 1999) can spatially resolvethe emission from these two sources. The majority ofthe
Chandra data analyzed here were taken during 2016(Cycle 17) as part of a Large Program aimed at system-atic monitoring of binaries in M82. The program con-sisted of 12 individual observations taken at ∼ monthlyintervals, but only 8 having simultaneous NuSTAR ob-servations were used in this work. We additionally usetwo observations taken in 2015 which also have a simul-taneous
NuSTAR observation. Full details are listed inTable 1. All 2016 observations were taken with ACIS-Iat the optical axis using a 1/8th sub-array of pixels onchip I1 or I3, depending on the roll angle. The ULXsat the center of M82 were placed 3 (cid:48) .5 off-axis to smearout the PSF enough to reduce the effects of pile-up, butnot so much as to cause significant blending of the PSFs.The sub-array of pixels was used to decrease the readouttime of the detector to 0.4 s, further reducing the effectsof pile-up.We proceeded to extract the
Chandra spectra in thesame way as described in B16, using the ciao (v4.7,CALDB v4.6.5) tool specextract . For the pointsources, spectra were extracted from elliptical regionsdrawn by eye to encompass the shape of the off-axis
Chandra
PSF. For X-1 we used an ellipse with a semi-major axis of 2–3 (cid:48)(cid:48) and a semi-minor axis of 1–2 (cid:48)(cid:48) . ForX-2 the major and minor axes were 2 (cid:48)(cid:48) and 1 (cid:48)(cid:48) respec-tively. A small rectangular region close by was used forbackground subtraction. Figure 1 shows examples of the ciao tool pileup map to give an indi-cation of the level of pileup in each observation. Theoutput, which is in counts per frame, ranges from 0.06 to0.35 at maximum, which occurs at the position of X-1.These numbers correspond to pileup fractions of < < . >
10% for > . > NuSTAR , such as from the fainter point sources and thediffuse emission, we extract spectra from a 49 (cid:48)(cid:48) radius cir-cular region centered on X-1, but with the X-1 and X-2regions as described above masked out. A larger back-ground region external to the galaxy was extracted inorder to assess the background coming from the CosmicX-ray and particle backgrounds.
NuSTAR
The raw
NuSTAR data were reduced using the nus-tardas software package version 1.4.1 and CALDB ver-sion 20150316. The events were cleaned and filteredusing the nupipeline script with standard parameters.The nuproducts task was used to generate the spectraand the corresponding response files. Spectra were ex-tracted from a circular aperture of radius 49 (cid:48)(cid:48) centeredon the peak of the emission. The background spectrawere extracted from a circular region encompassing thesame detector chip as the source, with a radius of 118 (cid:48)(cid:48) ,excluding the source extraction region and avoiding thewings of the PSF as much as possible. Data from bothfocal plane modules (FPMA and FPMB) were used forsimultaneous fitting, without co-adding.
SwiftSwift conducted monitoring of M82 with a typical ca-dence of a few days between 2012–2018. Since this mon-itoring ran contemporaneously with our
Chandra and
NuSTAR observations, the well-sampled lightcurve pro-vides us with context for our study. A total of 113 ob-servations, mostly consisting of obsIDs 00032503099–154and 00092202001–051, have been made of the galaxy overthe 2015–2016 period which we use here to calculate along-term lightcurve.We calculate the fluxes via spectral fitting. We use the heasoft (v 6.16) tool xselect to filter events from a49 (cid:48)(cid:48) radius region centered on the ULXs and to extractthe spectrum. This extraction region encloses all sourcesof X-ray emission in the galaxy. Background events wereextracted from a nearby circular region of the same size.We group the spectra with a minimum of one count perbin using the heasoft tool grppha . We conduct spec-tral fitting in the range 0.2–10 keV. We fit the spectrawith a simple power-law subjected to absorption intrin-sic to M82 at z = 0 . zwabs*powerlaw in xspec )with the Cash statistic (Cash 1979) which uses a Pois-son likelihood function and is hence most suitable for lownumbers of counts per bin. From this model we calculatethe observed flux in the 0.5–8 keV range, equivalent tothe Chandra band. The lightcurve is presented in Figure2.
X-2 X-3 X-4 X-1 X-5 X-6 X-7 E N
Figure 1.
Chandra image of M82 from obsID 17678 showing ex-amples of the various extraction regions used, including the back-ground for
Chandra analysis shown as a small rectangular region.Small ellipses are used for X-1 and X-2, whereas a large circlewith a radius of 49 (cid:48)(cid:48) is used to extract events from the rest of thegalaxy and the
NuSTAR and
Swift /XRT events. The brightestpoint-sources within this region are labelled, however X-1 and X-2dominate. North is up, east is left, indicated by the arrows in theupper right corner, which are 10 (cid:48)(cid:48) long.
Figure 2.
Swift /XRT. The highly variable X-ray emissionis caused by X-1. The times of the simultaneous
Chandra and
NuSTAR observations we use here to study the spectral evolutionof X-1 are shown with dashed lines.
The
Swift lightcurve shows that our observations tookplace during a period of flaring activity from M82, whichwe found to be due to increased activity from X-1 (B16).The first
Chandra / NuSTAR observations took place be-fore the increase in activity, and the remaining observa-tions tracked the activity over the next two years. SPECTRAL ANALYSIS
Brightman et al.
Figure 3.
Chandra spectra of X-1 (red), X-2 (green) and thediffuse emission (purple), and the
NuSTAR
FPMA spectra of allsources (blue) for each of the ten observational epochs listed inTable 1. The spectra have been heavily rebinned for clarity.
All
Chandra and
NuSTAR spectra were grouped witha minimum of 20 counts per bin using the heasoft tool grppha . Spectral fitting was carried out using xspec v12.8.2 (Arnaud 1996) and the χ statistic was used forspectral fitting to background subtracted spectra. Alluncertainties quoted are 90%. We present the Chandra spectra of X-1, X-2 and the additional X-ray emissionfrom M82, and the
NuSTAR spectra of the entire galaxyfor each of the 10 observational epochs in Figure 3.While the presence of X-2 in a bright state introducedsome ambiguity to the results on the spectral analysis ofX-1 in B16, in four observations here,
Chandra obsIDs18062, 18065, 18069 and 18070, X-2 was observed in alow state, allowing an unambiguous view of X-1 for thefirst time using
NuSTAR . For these four observations,we neglect the emission from X-2 in our spectral fits,whereas for the rest of the observations we included it. We proceed to conduct a spectral analysis for all 10observations, where we fitted the spectra in the sameway as described in B16. A cross-calibration constantwas applied to each spectral data set to allow for abso-lute differences in normalization, and allowed to vary by ±
10% (Madsen et al. 2015). For the diffuse emission fromM82 we use a combination of three absorbed zwabs*apec models with the temperatures and abundances fixed tothe values found in B16. We allow the normalizations,both with respect to B16 and to each other, to vary hereto account for small differences in the detector responses.For the spectrum of X-1, we use the zwabs*diskpbb model, where zwabs is a redshifted neutral absorptioncomponent and diskpbb is a model representing emis-sion from a multicolour accretion disk, with a variableradial temperature profile. This model combination wasfound to best represent the emission from X-1 with re-gards to other disk models in B16. Additionally, sinceX-1 is a bright source, and despite measures taken toreduce pileup, the source still suffers from pileup. Weaccount for this in spectral fitting using the pileup con-volution model, with frame time set to 0.4 s (Davis 2001).In the top panel of Figure 4 we show the residuals tothis set of models which reveal strong residuals above10 keV, especially in obsID 18062, which is reminiscentof the hard tail seen in the
NuSTAR spectra of severalother ULXs such as Holmberg II X-1 (e.g. Walton et al.2015), including those already identified as neutron staraccretors such as NGC 7793 P13 (Walton et al. 2018b)and NGC 5907 ULX (F¨urst et al. 2017; Walton et al.2018a). This was also seen in our analysis in B16, butsince X-2 was bright during that observation, there wasambiguity regarding its origin. Here it is clear that itoriginates from X-1.We test two models to fit this component, simpl , whichdescribes the power-law emission from the Comptoniza-tion of a disk spectrum (Steiner et al. 2009), and a cutoffpl model used to model the pulsed emission fromULXPs (Brightman et al. 2016b; Walton et al. 2018b,a).In Walton et al. (2018a), where no pulsations from asource had been detected, as is the case for M82 X-1,Γ was fixed at 0.5 for the cutoffpl model which is theaverage for the pulsed emission from ULXPs. Thereforewe do the same when using the cutoffpl model and fixΓ = 0 .
5. The energy of the cut off was allowed to vary.We find that the simpl model provides a better descrip-tion of the data, with ∆ χ =70–200 for the addition oftwo free parameters from simpl and ∆ χ =3–90 for theaddition of two free parameters from cutoffpl as shownin the middle and bottom panels of Figure 4. We do notfind a significant improvement for the cutoffpl model ifwe allow Γ to vary. Therefore we use the simpl model tofit the high energy emission from X-1 for the full dataset.We proceed to fit the Chandra and
NuSTAR spectra of all 10 observational epochs with the pileup*zwabs*diskpbb*simpl model combination todescribe the emission from X-1. For the six observa-tions where X-2 is bright, we model the emission fromthis source with an absorbed cut-off power-law model, zwabs*cutoffpl , which was used in B16 to model thepulsed emission. Figure 5 presents the unfolded spec-tra with the different model components shown. Table2 presents the best-fit spectral parameters for X-1, andTable 3 presents the best-fit spectral parameters for X-2. Figure 4.
Spectral residuals for a fit to the 18062 dataset onM82 X-1 where X-2 is off and a clear view of the source is seen.A prominent hard excess is seen with
NuSTAR when fitted with a pileup*zwabs*diskpbb model (top), which we account for with a simpl model (middle) and cutoffpl model (bottom).
In order to calculate the intrinsic luminosity of eachsource, we use the cflux model component in xspec placed after the absorption component with the normal-ization of the main model fixed. We calculate the intrin-sic flux over the range 0.5–30 keV, and present these val-ues with their uncertainties in Table 2 and Table 3. Wecalculate the intrinsic luminosities over these ranges as-suming a distance to M82 of 3.3 Mpc (Foley et al. 2014). SPECTRAL EVOLUTION OF M82 X-1
Our main goal is to explore the spectral evolution ofM82 X-1. For each pair of spectral parameters, we com-pute Spearman’s rank correlation, using the idl tool r correlate.pro , to assess the presence of any corre-lation and its significance. This assesses how well the re-lationship between two variables can be described usinga monotonic function. The results from this test, beingthe rank correlation coefficient, ρ , and the two-sided sig-nificance of its deviation from zero, p , are presented inTable 4.The significance is a value in the interval [0.0, 1.0] and asmall value indicates a significant correlation. The com-monly used threshold to judge that a correlation is sig-nificant is p < .
05. We find that this criterion is metfor the following pairs of parameters: L X and N H , L X and T in , L X and log norm, T in and log norm, L X and f scatt , T in and f scatt , and Γ and f scatt . Figure 5.
Chandra spectra of X-1 (red), X-2 (green) and thediffuse emission (purple), and the
NuSTAR
FPMA spectra of allsources (blue) unfolded through the spectral responses with theassumed spectral models.
However, for many parameters, especially those of the simpl model, the uncertainties are large, which the cor-relation test does not account for. First, in order to ac-count for this, we conduct 1000 Monte-Carlo simulationswhere for each pair of parameters we randomly draw avalue from the 90% confidence interval. We calculatefrom how many of the 1000 simulations do we find a cor-relation that has a p value which is less than 0.05. For allcorrelations involving parameters of the simpl model, wefind that in less than 25% of the simulations, a p valueless than 0.05 is recovered. Therefore we do not have con-fidence in these correlations being real due to the largeuncertainties in the parameters.Furthermore, for the correlations where the uncertain-ties in the parameters are smaller, some of these corre-lations can be expected due to degeneracies between pa-rameters. In order to determine if these correlations are Brightman et al.
Table 2
Spectral fitting results for X-1
Chandra N H T in p log norm Γ f scatt χ /DoF F X L X ObsID (10 cm − ) (keV) (10 − erg cm − s − ) (10 erg s − )17578 0.98 +0 . − . +2 . − . +0 . − . -2.6 +2 . − . +2 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -0.8 +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -1.8 +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -3.7 +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +1 . − . +0 . − . -3.3 +0 . − . +2 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -1.6 +0 . − . +1 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -1.4 +0 . − . +1 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -2.7 +1 . − . +1 . − . +0 . − . +0 . − . +0 . − . +0 . − . +1 . − . +0 . − . -2.7 +0 . − . +1 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . -2.5 +0 . − . +1 . − . +0 . − . +0 . − . +0 . − . Note . — Best-fit parameters for the diskpbb model fitted to X-1. Fluxes and luminosities are given in the 0.5 −
30 keV range, and arecorrected for absorption and pileup. Luminosities are calculated assuming a distance of 3.3 Mpc to M82.
Table 3
Spectral fitting results for X-2
Chandra N H Γ E C log norm χ /DoF F X L X ObsID (10 cm − ) (keV) (10 − erg cm − s − ) (10 erg s − )17578 1.7 +2 . − . -1.3 +3 . − . +22 − . -3.6 +0 . − . +0 . − . +0 . − . +1 . − . +0 . − . +77 − -2.7 +0 . − . +0 . − . +0 . − . +0 . − . -1.2 +0 . − . +0 . − . -3.6 +0 . − . +0 . − . +0 . − . +1 . − . -1.3 +1 . − . +0 . − . -3.4 +0 . − . +0 . − . +0 . − . +0 . − . +0 . − . +25 − -3.0 +0 . − . +0 . − . +0 . − . +0 . − . -2.9 +6 . − . +0 . − . -4.0 +0 . − . +0 . − . +0 . − . Note . — Best-fit parameters for the cutoffpl model fitted to X-2. Fluxes and luminosities are given in the 0.5 −
30 keV range, and arecorrected for absorption. Luminosities are calculated assuming a distance of 3.3 Mpc to M82. driven by degeneracies, we explore the two-dimensional χ space around the best fit using the xspec command steppar , and overplot the 3- σ contours on Figure 6. Wedo this only for one observation, Chandra obsID 18069,since it is computationally expensive, but this shouldbe adequate to reveal any spectral degeneracies. Thisdataset is a typical observation where X-2 is at low fluxes,pileup for X-1 is low, and the measured parameters arein the middle of the distributions.The contours show that there is slight degeneracy be-tween L X and N H , but that does not appear to be strongenough to induce the correlation. There is no apparentdegeneracy between L X and T in . A correlation between L X and log norm is expected and a clear degeneracybetween T in and log norm is seen.Since the background is higher in the NuSTAR datathan in the
Chandra data, and signal to noise lower, weinvestigated whether binning the
NuSTAR spectra withmore counts would affect our results. We grouped the
NuSTAR spectra with a minimum of 60 counts per bin,rather than 20, and refit the spectra. We looked at thetemperature of the diskpbb component, which is mostsensitive to the NuSTAR data and the one of the keyparameters involved in our results. We found the aver-age difference in temperature between the stronger bin-ning and the weaker one to be -0.02 keV, which is muchsmaller than the typical uncertainty on the parameter.We therefore conclude that the spectral binning does not affect our results.Taking into account the uncertainties in parametersand degeneracies between them, we can only say withconfidence that there is a correlation and therefore phys-ical link between the neutral column density and the X-ray luminosity, and the the inner disk temperature andX-ray luminosity.We test the apparent dependence of N H on L X furtherby fixing N H at the approximate mean value of 1 . × cm − in all fits and note the difference in χ . Inall cases χ is worse or the same as when N H is a freeparameter. For Chandra obsID 18062 L X is high andthe N H measured is particularly high at 1 . × cm − .When N H is fixed at the mean value, the difference in χ is 80. For obsID 18068 L X is low and N H = 9 × cm − .When fixed at 1 . × cm − , the χ increases by 150,supporting the finding that N H does indeed depend on L X .The anti-correlation between L X and T in that we findhere is in contrast with the correlation found betweenthese two parameters for M82 X-1 by Feng & Kaaret(2010), where the apparent L X ∝ T relationship leadthe authors to conclude that the source was observedin the thermal dominant state. In order to determinethe exponent on the relationship we have found, we takethe logarithm of both L X and T in , such that log L X = α log T in + β and conduct a linear regression analy-sis. We use the idl tool linmix err.pro which takes L X = ( − . ± . T in + (41 . ± . σ . Therefore the exponentis − . ± .
00, which is a > σ difference from the re-sults from F10, where the exponent was implied to be4, although the uncertainty in this parameter was notlisted. If we run the same linear regression analysis onthe results presented in F10, we find that the relationshipis essentially unconstrained.The key differences between our analysis and that ofF10 are that they did not have accompanying NuSTAR data, and so their analysis was restricted to the narrowbandpass of 0.7–7 keV. The spectral models used are alsodifferent, whereby F10 used the diskbb model, which isrelated to diskpbb that we use when p is fixed at 0.75.However, as we showed in B16, and confirm here, thedata are inconsistent with the p = 0 .
75 that describesa geometrically thin accretion disk. Furthermore, with-out
NuSTAR data, the high energy excess that we detectand fit with simpl was not observable by F10. Finally,the uncertainties related to degeneracies in the pileup model are reduced when data from an instrument with-out pileup such as
NuSTAR is used.We investigate what effect pileup has on our resultsby exploring the dependence of L X , N H and T in on the pileup model parameter, α . We find that none of theseparameters show any dependence on α , leading us toconclude that this model component does not drive ourresults. Furthermore, as mentioned earlier, for three ob-servations, odsID 17678, 18065 and 18067, the pileupfraction for X-1 is greater than 10%, and therefore the pileup model used in the spectral fitting may not be ableto reliably account for this effect. We test the dependenceof our results on these observations by removing themfrom our analysis. In doing so, we still find a significantcorrelation between L X and N H , and indeed the signif-icance increases, but for L X and T in the correlation isno longer significant. A fit with a linear relationshipreveals log L X = ( − . ± . log T in +(41 . ± . and therefore we can only rule out a L X ∝ T de-pendency at ∼ σ . Finally, we note that some of our spectral fitshave large χ relative to the number of degreesof freedom (DoF), indicating an unacceptable fit.This is likely due to the fits being quite complex,with many DoFs and data sets being fitted si-multaneously. However, we note the presence ofa potential excess of counts in the NuSTAR dataat 3–4 keV that has been found in bright X-raybinaries and is a known calibration issue. Wefind that if we ignore data from
NuSTAR below5 keV, the fits improve and the result of this isto systematically reduce the temperature of the diskpbb component. This reduction is consistentwith the uncertainties on this component, how-ever, and does not alter our result that there isan anti-correlation between L X and T in since theeffect is systematic across all observations. These findings therefore rule out a thermal state forsub-Eddington accretion and therefore do not supportM82 X-1 as an IMBH candidate. NEW INSIGHTS INTO M82 X-2
Table 4
Spearman’s rank correlation results for X-1 L X N H T in p log norm Γ N H in -0.76 -0.560.011 0.090p -0.33 -0.62 0.030.347 0.054 0.934log norm 0.79 0.62 -0.99 -0.100.006 0.054 0.000 0.777Γ 0.48 0.58 -0.22 -0.36 0.280.162 0.082 0.533 0.310 0.425 f scatt Note . — For each pair of parameters we list the rank correlationcoefficient (top) and the two-sided significance of its deviation fromzero (bottom).
We conduct the same analysis for X-1 on X-2, wherethe spectral parameters are plotted against each otherin Figure 7 and the correlation analysis is shown in Ta-ble 5. Here our analysis suggests correlations between N H and Γ, E C and Γ, N H and log norm, and Γ andlog norm. However, the χ contours show that it islikely that strong degeneracies between these parametersdrive these apparent correlations (Fig 7). We thereforedon’t find evidence for any significant spectral evolutionin X-2.While the emission from X-2 is difficult to disentan-gle from the other sources of emission in M82, we wereable to isolate the pulsed emission in the NuSTAR bandfrom this source in Brightman et al. (2016b). We foundthat the pulsed emission is best fit by a power-law with ahigh-energy cut-off, where Γ = 0 . ± . E C = 14 +5 − keV. In Figure 7 we show the parameters of the pulsedemission as a separate data point for comparison. Wesee that the values for Γ and E C from our broadbandfits are consistent with the pulsed emission when X-2 isat its highest luminosities, L X > erg s − , indicatingthat at these times the pulsations are most likely to bedetected. Bachetti et al. (2019) have recently detectedpulsations again from a NuSTAR observation taken on2016-09-10, which unfortunately did not have any simul-taneous
Chandra observations, and so we did not includeit in our analysis here. DISCUSSION AND IMPLICATIONS
M82 X-1 as an intermediate-mass black holecandidate
M82 X-1 has been claimed to be an intermediate-massblack hole candidate based on its high X-ray luminosity,twin QPOs, and L X ∝ T scaling, all of which put themass of the black hole at ∼ M (cid:12) . Here we find that wecan rule out a L X ∝ T scaling. This scaling relationshipwas expected from a standard accretion disk which existsat moderate accretion rates, and allows an estimate of theblack hole mass from measurements of the inner edge ofthe accretion disk. Without these pieces of evidence, thestatus of M82 X-1 as an IMBH accretor is less certain. M82 X-1 as a super-Eddington accretingstellar-remnant
Brightman et al.
Figure 6.
The relationship between the spectral parameters of the zwabs*simpl*diskpbb model for X-1 where a significant correlationwas found. Red data points indicate observations where X-2 is at low fluxes and thus the view of the emission from X-1 is unambiguous.We also overplot the 3- σ χ contour in blue, shifted for clarify, from Chandra obsID 18069 to demonstrate if the correlation is driven bydegeneracy between the parameters, which appears to be the case for T in and log norm. Figure 7.
The relationship between the spectral parameters of the zwabs*cutoffpl model for X-2 where evidence for a correlation hasbeen found. Red data points indicate the best-fit parameters of the pulsed emission. We also overplot the 3- σ χ contour in blue, shiftedfor clarify, from Chandra obsID 18063 to demonstrate if the correlation is driven by degeneracy between the parameters, which appears tobe the case for all parameter pairs
Table 5
Spearman’s rank correlation results for X-2 L X N H E C Γ N H C norm 0.71 1.00 0.77 0.830.111 0.000 0.072 0.042 Note . — For each pair of parameters we list the rank correlationcoefficient (top) and the two-sided significance of its deviation fromzero (bottom).
The X-ray properties of M82 X-1 may be explained byit harboring a super-Eddington accreting stellar remnantblack hole or neutron star. The spectral shape of M82 X-1, consisting of a broadened disk with a high energy tailis very similar to all other ULXs with high-quality broad-band spectral data from
NuSTAR (e.g. Walton et al.2018a). This sample includes the known neutron staraccretors NGC 5907 ULX1 and NGC 7793 P13, and atfirst glance their spectral shapes are not dissimilar fromthe rest of the sample. This implies that these ULXs,including M82 X-1, are also super-Eddington accretors,although it is still not known whether they are powered by neutron stars or black holes.One popular model to explain the spectral evolution ofULXs is that they are stellar-remnant black holes accret-ing at super-Eddington rates. In this model a powerfulwind is radiatively driven from the accretion disk (e.g.Poutanen et al. 2007), as recently revealed through thedetection of highly ionized material in the high resolu-tion X-ray spectra of NGC 1313 X-1 (Pinto et al. 2016)among others (Pinto et al. 2017; Kosec et al. 2018) andthe detection of blueshifted iron-K absorption (Waltonet al. 2016). Regarding the link between N H and L X asseen in M82 X-1, if this source is a stellar-mass blackhole accreting at super-Eddington rates, as the mass ac-cretion rate increases an increase in the X-ray luminosityfollows, which drives further outflow of material from thesystem. Depending on the line of sight, this can causean increase in the line of sight absorption. However, thismaterial is expected to be highly ionized at small radii.Middleton et al. (2015) also found for NGC 1313 X-1 thatthe neutral column density anti-correlates with spectralhardness, suggesting that at large radii, there is a cool,neutral component of the outflow, where the column den-sity is linked to mass loss via increase mass accretion rate,as predicted by Poutanen et al. (2007).A negative relationship between L X and T in has beenobserved in other ULXs. For example Luangtip et al.(2016) studied the spectral evolution of Holmberg IXX-1, finding that the peak of the spectrum decreaseswith luminosity, suggesting an anti-correlation between L X and T (see also Walton et al. (2017)). Kajava &Poutanen (2009) also explored the relationship between L X and T in a sample of ULXs with XMM-Newton and
Chandra observations. While they found that for sev-eral sources fitted with a multicolor disk model followthe L X ∝ T scaling, sources with a disk plus power-lawspectral shape show a negative L X ∝ T − . SUMMARY AND CONCLUSIONS
We have conducted a comprehensive investigation intothe spectral evolution of the ultraluminous X-ray sourcesM82 X-1 and X-2 using ten simultaneous
Chandra and
NuSTAR observations. The
Chandra data allowed usto spatially resolve the sources, separated by only 5 (cid:48)(cid:48) ,while the
NuSTAR data allowed a broadband X-ray spec-tral analysis. We found that for X-1, the luminosity ofthe disk scales with the inner temperature as L ∝ T − / ,which is contrary to previous findings of a L ∝ T scalingthat supported a standard accretion disk powering thesystem. We furthermore find evidence that the neutralcolumn density of the material in the line of sight in-creases with L X , perhaps due to an increased mass out-flow with accretion rate. For X-2, we do not find anysignificant spectral evolution, but we can constrain thespectral parameters, and find the broadband emission tobe consistent with the pulsed emission at the highest X-ray luminosities. ACKNOWLEDGEMENTS
This research has made use of data obtained with
NuS-TAR , a project led by the California Institute of Tech-nology, managed by the Jet Propulsion Laboratory, andfunded by NASA. We thank the
NuSTAR
Operations,Software and Calibration teams for support with theexecution and analysis of these observations. This re-search has made use of the
NuSTAR
Data Analysis Soft-ware (NuSTARDAS) jointly developed by the ASI Sci-ence Data Center (ASDC, Italy) and the California In-stitute of Technology (USA). Support for this work wasprovided by the National Aeronautics and Space Admin-istration through
Chandra
Award Number GO6-17080Xissued by the
Chandra
X-ray Center, which is operated bythe Smithsonian Astrophysical Observatory for and onbehalf of the National Aeronautics Space Administrationunder contract NAS8-03060. This research has made useof software provided by the Chandra X-ray Center (CXC)in the application package ciao . We also acknowledgethe use of public data from the
Swift data archive. DJWacknowledges support from an STFC Ernest RutherfordFellowship.
Facilities: Chandra (ACIS),
NuSTAR , Swift (XRT)REFERENCES(XRT)REFERENCES