BROADBAND X-RAY SPECTRA OF THE ULTRALUMINOUS X-RAY SOURCE HOLMBERG IX X-1 OBSERVED WITH NuSTAR, XMM-NEWTON, AND SUZAKU

Ultraluminous X-ray sources (ULXs) are off-nuclear point sources with X-ray luminosities L_X > 10³⁹ erg s⁻¹, in excess of the Eddington limit for the typical 10 M_☉ stellar-remnant black holes observed in Galactic black hole binaries (BHBs; e.g., Orosz 2003). Multi-wavelength observations have largely excluded strong anisotropic emission as a means of substantially skewing luminosity estimates (e.g., Berghea et al. 2010; Moon et al. 2011), thus these high luminosities require either the presence of larger black holes, either stellar remnant black holes more massive than observed in our own Galaxy (e.g., Zampieri & Roberts 2009) or perhaps even the long postulated "intermediate mass" black holes (IMBHs: 10² ≲ M_BH ≲ 10⁵ M_☉, e.g., Miller et al. 2004; Strohmayer & Mushotzky 2009), or exotic super-Eddington modes of accretion (e.g., Poutanen et al. 2007; Finke & Böttcher 2007). The majority of ULXs only radiate marginally in excess of 10³⁹ erg s⁻¹ (Walton et al. 2011b; Swartz et al. 2011), and are likely to simply represent a high luminosity extension of the stellar mass BHB population (Middleton et al. 2013; Liu et al. 2013). Of particular interest are the rare population of extreme ULXs with X-ray luminosities L_X > 10⁴⁰ erg s⁻¹ (e.g., Farrell et al. 2009; Walton et al. 2011b, 2013a; Jonker et al. 2012; Sutton et al. 2012). The extreme luminosities displayed by these sources make them the best candidates for hosting more massive black holes. For recent reviews on the observational status and the potential nature of ULXs, see Roberts (2007) and Feng & Soria (2011).

Previous studies have established that the 0.3–10.0 keV X-ray spectra of these extreme ULXs typically show evidence for two separate continuum components (e.g., Miller et al. 2003; Vierdayanti et al. 2010; Miller et al. 2013b), one dominating at softer (≲2 keV) and the other at harder X-rays (≳2 keV), potentially analogous to the disk–corona accretion components inferred for sub-Eddington BHBs (see Remillard & McClintock 2006 for a review). However, studies focusing on the highest signal-to-noise data have found that the harder component generally shows evidence of subtle curvature in the ∼3–10 keV bandpass (Stobbart et al. 2006; Gladstone et al. 2009; Walton et al. 2011a), which is not observed in the coronal emission of standard sub-Eddington accretion states. A number of interpretations have since been proposed for this spectral structure, which can broadly be grouped into those that invoke thermal processes for the harder component (e.g., emission from a hot accretion disk; Gladstone et al. 2009; Middleton et al. 2011b; Sutton et al. 2013), which generally invoke super-Eddington accretion, and those that invoke nonthermal processes (e.g., a combination of a power-law continuum and relativistic reflection from the inner disk; Caballero-García & Fabian 2010), which may still involve intermediate mass black holes (IMBHs; M_BH ∼10^{2 − 5} M_☉). As demonstrated in Walton et al. (2011a), these different model families predict substantially different spectra above 10 keV, in the bandpass only readily accessible with the focusing hard X-ray telescopes aboard the recently launched Nuclear Spectroscopic Telescope Array (NuSTAR; Harrison et al. 2013).

Holmberg IX X-1 is one of the best studied extreme ULXs, which although known to vary in flux (e.g., Kong et al. 2010; Vierdayanti et al. 2010) is one of the few sources (within ∼5 Mpc) to persistently radiate at L_X > 10⁴⁰ erg s⁻¹. Early XMM-Newton observations revealed the possible presence of a very cool accretion disk (Miller et al. 2003), which may evolve in a fashion similar to the L∝T⁴ relation expected for simple blackbody radiation (Miller et al. 2013b), and indicate the presence of a massive black hole. However, as with other extreme ULXs for which high quality data are available, Holmberg IX X-1 shows evidence for high energy spectral curvature (e.g., Stobbart et al. 2006; Gladstone et al. 2009; Walton et al. 2013b). Holmberg IX X-1 is also one of the brightest ULXs in the iron K α bandpass, and sensitive searches have been made for absorption features that would be indicative of the massive outflows ubiquitously predicted by simulations of super-Eddington accretion (e.g., Ohsuga & Mineshige 2011; Dotan & Shaviv 2011). No features are detected, with limits that require that any undetected features to be weaker than the iron absorption features resulting from the outflows in a number of sub-Eddington Galactic BHBs (Walton et al. 2012, 2013b).

Reconciling these various results into a coherent picture regarding the nature of Holmberg IX X-1 remains challenging. In order to shed further light onto the nature of the accretion in this source, we undertook a series of observations with NuSTAR in order to determine the nature of the high energy emission. These were coordinated with Suzaku and/or XMM-Newton, providing the first high quality broadband (∼0.3–30.0 keV) X-ray spectra of Holmberg IX X-1. The paper is structured as follows. Section 2 describes our data reduction procedure, and Sections 3 and 4 describe the analysis performed. We discuss our results in Section 5 and summarize our conclusions in Section 6.

Holmberg IX X-1 was observed by each of the NuSTAR, XMM-Newton, and Suzaku X-ray observatories on multiple occasions during 2012. The observations used in this work are summarized in Table 1. Here, we outline our data reduction for these observations.

2.1. NuSTAR

2.2. Suzaku

2.2.1. XIS Detectors

The first NuSTAR observation of Holmberg IX X-1 was performed simultaneously with a portion of our recent long integration (Walton et al. 2013b) with the Suzaku observatory (Mitsuda et al. 2007). Figure 1 highlights the relative coordination of all the observations considered in this work. The data reduction procedure for the XIS detectors (Koyama et al. 2007) for the full Suzaku data set, including observations taken in 2012 April, has already been described in Walton et al. (2013b), following the procedure recommended in the Suzaku data reduction guide.¹¹ Here we only consider the Suzaku data taken either simultaneously or contemporaneously with NuSTAR in October, and we re-reduce the spectra following the same procedure with the latest XIS calibration files (released 2013 September), which substantially improves the agreement between the front- and back-illuminated XIS detectors at low energies. In this work, we model the XIS data over the 0.7–10.0 keV energy range (unless stated otherwise), excluding the 1.6–2.1 keV band throughout, owing to remaining calibration uncertainties associated with the instrumental silicon K edge. We also rebin the XIS spectra to have a minimum of 50 counts per energy bin.

Zoom In Zoom Out Reset image size

2.2.2. HXD PIN

Owing to the combination of the systematic uncertainty in the background model for the Suzaku PIN detector (Takahashi et al. 2007)—equivalent to ≳25% of the "source" flux for the weak detection of the Holmberg IX field (see, e.g., discussion in Walton et al. 2013c)—and the source confusion resulting from its lack of imaging capability—the variable nucleus of M81 (e.g., Markoff et al. 2008; Miller et al. 2010) also falls in the PIN field-of-view—the data obtained with this detector unfortunately cannot be used to constrain the high energy (E > 10 keV) properties of Holmberg IX X-1 (Walton et al. 2013b). Therefore, we do not consider the PIN data here, and focus instead on the high energy data provided by NuSTAR.

2.3. XMM-Newton

During these observations, Holmberg IX X-1 displayed some flux variability, both between epochs and within a single epoch (Figure 1). We therefore test to see whether this flux variability is accompanied by strong spectral variability in order to determine how to best extract spectra. In Figure 2 we show the XMM-Newton spectra obtained from each of the three observations associated with each epoch. Despite the mild flux variability, the spectra obtained within each epoch show good consistency. However, we do see differences in the broadband spectra when comparing epoch 1 and epoch 2. Given the lack of short-term spectral variability, we do not require strict simultaneity between NuSTAR, Suzaku and/or XMM-Newton. This allows us to maintain the highest S/N possible in the NuSTAR data, and we combine the available data from each of the missions into average spectra for each epoch, which we analyze simultaneously. However, we keep our analysis of epochs 1 and 2 separate.

Zoom In Zoom Out Reset image size

We investigate the broadband spectral properties of Holmberg IX X-1 utilizing a similar suite of models to those applied in recent broadband spectral studies of other ULXs (Walton et al. 2013a; Bachetti et al. 2013; Rana et al. 2014). Throughout this work, spectral modeling is performed with XSPEC v12.8.0 (Arnaud 1996), and quoted uncertainties on spectral parameters are the 90% confidence limits for a single parameter of interest, unless stated otherwise. Neutral absorption is treated with TBNEW,¹³ the latest version of the TBABS absorption code (Wilms et al. 2000), with the appropriate solar abundances, and the absorption cross-sections of Verner et al. (1996). All models include Galactic absorption with a column of N_{H; Gal} = 5.54 × 10²⁰ cm⁻² (Kalberla et al. 2005), in addition to an intrinsic neutral absorber at the redshift of Holmberg IX (z = 0.000153) with a column that is free to vary (unless stated otherwise).

3.1. Epoch 1

3.1.1. Cross-Calibration

We focus initially on the data obtained during the first epoch, and first investigate the agreement between the three missions over their common 3–10 keV bandpass, taking this epoch to be representative. We first apply an unabsorbed power-law model to this energy range for each of the three observatories, allowing multiplicative constants to float between the spectra obtained from the various detectors (we choose the constant for NuSTAR FPMA to be unity). Although this model formally provides a good fit, with $\chi ^{2}_{\nu }$ = 1984/1928 and $\Gamma _{\rm 3-10\, keV} = 1.81^{+0.01}_{-0.02}$, systematic curvature in the residuals can be seen across this bandpass (see Figure 3, top panel). Parameterizing the data with a curved continuum instead, using a simple unabsorbed bremsstrahlung model, provides an excellent fit (Figure 3, bottom panel), with $\chi ^{2}_{\nu }$ = 1870/1928 and T = 11.0 ± 0.3 keV, an improvement of Δχ² = 114 (for no additional degrees of freedom) over the power-law continuum. Allowing the temperatures to vary independently for each of the different missions does not significantly improve the fit ($\chi ^{2}_{\nu }$ = 1861/1926), and the temperatures obtained all agree within 2σ or better: $T_{\rm XMM} = 12.7^{+1.8}_{-1.3}$ keV, T_Suzaku = 11.1 ± 0.4 keV and T_NuSTAR = 10.4 ± 0.5 keV. Therefore, we conclude the spectra obtained with XMM-Newton, Suzaku, and NuSTAR show good agreement (see also Walton et al. 2013a; K. K. Madsen et al. 2014; in preparation). The fluxes obtained for each of the different detectors all agree with that of NuSTAR FPMA to better than ∼10%.

Zoom In Zoom Out Reset image size

Furthermore, the preference for the bremsstrahlung continuum over the power law confirms the presence of curvature in the observed 3–10 keV spectrum, similar to that inferred from earlier XMM-Newton data (Stobbart et al. 2006; Gladstone et al. 2009), and from the full Suzaku data set (including additional data obtained earlier in 2012; Walton et al. 2013b). The absorption column toward Holmberg IX X-1 is N_H ≲ 2 × 10²¹ cm⁻² (e.g., Miller et al. 2013b), which is not sufficient to significantly influence the spectrum in the 3–10 keV bandpass. Thus, the observed curvature must be intrinsic to the 3–10 keV continuum.

3.1.2. Broadband Continuum Modelling

We now analyze the full broadband XMM-Newton + Suzaku + NuSTAR spectrum obtained from the first epoch, applying a suite of continuum models in order to examine the nature of the broadband emission from Holmberg IX X-1. Figure 4 (top panel) shows the spectra from epoch 1 unfolded through a model that simply consists of a constant. We also show in Figure 4 (bottom panel) the data/model ratio to the power-law continuum initially applied to the 3–10 keV bandpass (as outlined in Section 3.1.1), then extrapolated across the full 0.3–30.0 keV spectrum considered here.

Zoom In Zoom Out Reset image size

The NuSTAR data clearly show that the curvature displayed in the 3–10 keV bandpass extends to higher energy, and genuinely represents a spectral cutoff, similar to NGC 1313 X-1 (Bachetti et al. 2013) and IC 342 X-1 (Rana et al. 2014). This is contrary to the basic expectation for the interpretation in which the 3–10 keV curvature is produced by relativistic disk reflection with the accretion disk illuminated by a standard sub-Eddington power-law-like corona (Caballero-García & Fabian 2010), as the Compton hump should cause the spectrum to turn back up at higher energies (≳10 keV; Walton et al. 2011a). This model provides a good fit to the data below 10 keV ($\chi ^{2}_{\nu }$ = 3419/3360), utilizing the REFLIONX code (Ross & Fabian 2005) for the reflected emission and the RELCONV kernel (Dauser et al. 2010) to account for the relativistic effects inherent to the inner accretion flow around a black hole. However, when applying this model to the full 0.3–30.0 keV bandpass we see that when the curvature is modeled as being due to iron emission, the data are significantly overpredicted at the highest energies (Figure 5), and the resulting broadband fit is poor ($\chi ^{2}_{\nu }$ = 6016/3516). We therefore proceed by considering models that invoke a thermal origin for the high energy curvature.

Zoom In Zoom Out Reset image size

Simple multi-color blackbody accretion disk models (e.g., DISKBB: Mitsuda et al. 1984, DISKPN: Gierliński et al. 1999), assuming a geometrically thin, optically thick disk as per Shakura & Sunyaev (1973), also fail to fit the broadband data (DISKBB: $\chi ^{2}_{\nu }$ = 22682/3523; DISKPN: $\chi ^{2}_{\nu }$ = 20303/3522). Although the high energy data do show a cutoff similar to a thermal spectrum, the overall profile is too broad to be explained by such simple models. However, at very high- or super-Eddington rates the accretion disk may differ substantially from the simple Shakura & Sunyaev (1973) thin disk profile owing to the increased effects of radiation pressure and advection (Abramowicz et al. 1988). This can result in shallower radial temperature profiles and the appearance of a broader, less-peaked emission profile from the disk. Allowing the radial temperature profile of the disk (p) to vary as an additional free parameter using the DISKPBB model (Mineshige et al. 1994) does substantially improve the fit ($\chi ^{2}_{\nu }$ = 3913/3522), and the inferred temperature profile is shallower than expected for a thin disk (i.e., p < 3/4); the results are presented in Table 2. However, significant structure remains in the residuals (see Figure 5), and the spectrum does appear to require at least two separate continuum components.

Table 2. Best Fit Parameters Obtained for the Variety of Continuum Models Applied to the Broadband Data Available for Holmberg IX X-1

Model	NH; int	kTin, 1	kTe or kTin, 2a	p or τ	Γ	fscat	χ2/DoF
	(1021 cm−2)	(keV)	(keV)			(%)
Epoch 1
DISKPBB	$1.27^{+0.03}_{-0.04}$	$4.87^{+0.13}_{-0.12}$	⋅⋅⋅	0.542 ± 0.001	⋅⋅⋅	⋅⋅⋅	3913/3522
DISKBB+COMPTT	$1.5^{+0.2}_{-0.3}$	0.15 ± 0.01	3.1 ± 0.1	6.3 ± 0.1	⋅⋅⋅	⋅⋅⋅	3651/3520
DISKBB+SIMPL⊗COMPTT	$1.4^{+0.1}_{-0.2}$	0.23 ± 0.04	$2.4^{+0.3}_{-0.4}$	$7.3^{+0.4}_{-0.5}$	>2.4	>16	3581/3518
DISKBB+DISKPBB	1.7 ± 0.1	0.26 ± 0.02	4.2 ± 0.1	$0.560^{+0.004}_{-0.003}$	⋅⋅⋅	⋅⋅⋅	3606/3520
DISKBB+SIMPL⊗DISKPBB	1.6 ± 0.1	$0.27^{+0.01}_{-0.02}$	3.8 ± 0.2	0.564 ± 0.004	$1.63^{+0.09}_{-0.08}$	3 ± 1	3584/3518
Epoch 2
DISKPBB	1.67 ± 0.06	2.56 ± 0.03	⋅⋅⋅	0.587 ± 0.004	⋅⋅⋅	⋅⋅⋅	2166/1584
DISKBB+COMPTT	0.59 ± 0.04	1.17 ± 0.06	2.7 ± 0.1	$7.1^{+0.7}_{-0.5}$	⋅⋅⋅	⋅⋅⋅	1949/1582
DISKBB+SIMPL⊗COMPTT	$1.3^{+0.2}_{-0.1}$	0.3 ± 0.1	$1.11^{+0.06}_{-0.04}$	11.3 ± 0.6	3.6 ± 0.1	>81	1695/1580
DISKBB+DISKPBB	1.7 ± 0.1	1.79 ± 0.04	$5.4^{+0.5}_{-0.4}$	0.514 ± 0.007	⋅⋅⋅	⋅⋅⋅	1693/1582
DISKBB+SIMPL⊗DISKPBB	$1.5^{+0.3}_{-0.1}$	$1.63^{+0.07}_{-0.06}$	$3.5^{+0.6}_{-0.3}$	$0.55^{+0.01}_{-0.02}$	<1.65	$2.3^{+0.9}_{-0.6}$	1686/1580

Note. ^a This column gives the temperature of the hotter continuum component, where relevant, i.e., kT_e for models including a COMPTT component, or the inner temperature kT_in of the DISKPBB component for models invoking both DISKBB and DISKPBB.

Download table as:  ASCIITypeset image

We therefore model the broadband spectrum from epoch 1 with the DISKBB+COMPTT combination frequently used to parameterize the spectra from bright ULXs below 10 keV (e.g., Gladstone et al. 2009; Middleton et al. 2011b; Walton et al. 2011a). This model is used to represent a disk–corona accretion geometry, in which hot coronal electrons Compton up-scatter seed photons from the (cooler) accretion disk into a second hard emission component, modeled here with COMPTT (Titarchuk 1994). For simplicity, we require the temperature of the seed photons to be that of the accretion disk component. Although this may not be appropriate if this emission genuinely arises from an optically thick corona (as frequently inferred for ULXs with this model), which may obscure our view of the true inner disk (e.g., Gladstone et al. 2009; Pintore et al. 2014), it is sufficient for our purposes as relaxing this assumption does not improve the fit, nor change the parameter values obtained. The COMPTT model allows both the temperature and the optical depth of the Comptonizing region to be fit as free parameters. This does offer a significant improvement over the simpler DISKPBB model, with $\chi ^{2}_{\nu }$ = 3651/3520, and the results obtained are fairly similar to previous applications of this model to ULX spectra (see Table 2), i.e., the electron plasma is inferred to be fairly cool (T_e ∼ 3 keV) and optically thick (τ ∼ 6) owing to the observed spectral curvature in the ∼3–10 keV bandpass, and the seed photon/disk temperature is very cool (T_in ∼ 0.15 keV). However, this model leaves a significant excess at the highest energies probed by NuSTAR (see Figure 5), requiring a further high energy continuum component.

Although the DISKBB+COMPTT model formally describes a disk–corona scenario, when applied to ULX spectra the COMPTT component has frequently been interpreted as representing emission from a distorted high-Eddington accretion disk (e.g., Middleton et al. 2011b) rather than standard coronal emission, owing to the large optical depth typically inferred, and the unusually low electron temperatures compared to those observed from standard sub-Eddington coronae (e.g., Gierliński et al. 1999; Miller et al. 2013a; Tomsick et al. 2014; Natalucci et al. 2014; Brenneman et al. 2014). In this scenario, it is therefore plausible that the high energy excess observed represents the true emission from the Comptonizing corona. The addition of a power-law tail to the COMPTT component included in this model, using the SIMPL convolution model (Steiner et al. 2009; limiting the photon index to 1.5 ⩽ Γ ⩽ 4, broadly encompassing the range observed from Galactic binaries), provides a further significant improvement of Δχ² = 70 (for two additional free parameters) to the fit, and resolves the high energy excess (see Figure 5). Even if poorly constrained owing to a strong degeneracy with the scattered fraction f_scat (which serves as the SIMPL "normalization"; see Table 2), the photon index is very steep, Γ > 2.4. We stress that even though this model still utilizes COMPTT to model the ∼3–10 keV emission, the necessity for a second Comptonizing region at even higher energies strongly suggests that this emission does arise from a hot, super-Eddington disk, which the COMPTT model merely has the flexibility to mimic, rather than physically being associated with an optically thick corona.

Finally, if the high energy emission is dominated by a hot accretion disk, describing this emission with a COMPTT component with the seed photon temperature linked to the soft thermal component might not be correct. We therefore also test other models composed of two thermal components in order to test whether the hard excess observed with the DISKBB+COMPTT combination is simply a consequence of that particular model. A model consisting of two DISKBB components provides a poor fit to the data ($\chi ^{2}_{\nu }$ = 4228/3521), leaving a strong excess in the NuSTAR data above 10 keV and substantial residual structure at lower energies. Replacing the second (hotter) DISKBB with a DISKPBB component, with the radial temperature profile free to vary again provides a significant improvement ($\chi ^{2}_{\nu }$ = 3606/3520). However, there is still evidence for an excess remaining at the highest energies (see Figure 5); the addition of SIMPL here gives an improvement of Δχ² = 22 (for two additional free parameters). Although the best fit photon index of the power-law tail is rather hard in this case, we note that there is a local minimum of similar statistical quality ($\chi ^{2}_{\nu }$ = 3588/3518) with a steep photon index (Γ > 3.1). Furthermore, even in the model with the hard photon index, the hard power-law continuum only begins to dominate the model outside of the bandpass probed. Within the NuSTAR bandpass, the effect is still to append a steep tail onto the DISKPBB component, similar to the case with the steep photon index, as the model begins to curve up into the hard portion of this continuum. It seems that the marginal preference for a hard photon index arises from the subtle difference this model has on the curvature at lower energies, rather than the slope of the observed continuum above ∼15 keV, which Figure 4 shows is steep.

3.2. Epoch 2

During the second epoch, Holmberg IX X-1 was substantially brighter than during the first epoch (see Table 3). Although there is no evidence for substantial spectral variability during either of the two epochs, the spectra from the two epochs do exhibit marked differences, as shown in Figure 6. The strongest variability can clearly be seen at ∼3 keV, and the 3–10 keV spectrum now appears to be more peaked than during epoch 1. In contrast, however, there is remarkably little variability in the spectrum at the highest (≳15 keV) and the lowest (≲1 keV) energies. One of the thermal components required in epoch 1 appears to have evolved substantially both in terms of its temperature and its flux, while the other appears to have remained relatively stable.

Zoom In Zoom Out Reset image size

We therefore apply the thermal models considered for epoch 1 to these data, in order to investigate which of these can adequately reproduce the epoch 2 spectrum. Even though the spectrum now appears to be primarily dominated by one of the thermal components, simple accretion disk models still fail to adequately reproduce the data; an absorbed DISKBB model gives a very poor fit, with $\chi ^{2}_{\nu }$ = 4727/1585. Allowing the radial temperature profile to vary again substantially improves the fit ($\chi ^{2}_{\nu }$ = 2166/1584), but the data still show a strong excess over the model at the highest energies, and at least two continuum components still appear to be required for this epoch (see Figure 7).

Zoom In Zoom Out Reset image size

We therefore fit the DISKBB+COMPTT model considered previously. While the fit is improved over the DISKPBB model, it is still relatively poor ($\chi ^{2}_{\nu }$ = 1949/1582), and the high energy excess persists, even though in this model it is the DISKBB component that shifts up in temperature to try and account for the peak of the emission (at ∼4 keV; Figure 6), while the best fit temperature and the optical depth for the COMPTT component are similar to the first epoch (see Table 2). Allowing again for a further high energy power-law tail to the COMPTT component with SIMPL substantially improves the fit ($\chi ^{2}_{\nu }$ = 1695/1580), and accounts for the high energy excess. However, the best fit results are very different to those obtained with the simpler DISKBB+COMPTT model. Now it is the DISKBB component that appears to have remained fairly constant, while the COMPTT component has decreased its temperature and increased its flux to account for the ∼4 keV peak. Other aspects have remained similar to the first epoch though, and the COMPTT component is again inferred to be optically thick (τ ∼ 11).

Finally, we also reconsider the DISKBB+DISKPBB model applied previously to the first epoch. This model provides a similar quality fit to the DISKBB+COMPTT+SIMPL combination ($\chi ^{2}_{\nu }$ = 1693/1582), but here it is the DISKPBB component which primarily accounts for the high energy emission, while the temperature of the DISKBB component has increased substantially to account for the ∼4 keV spectral peak (again, see Table 2). In contrast to the first epoch, adding a high energy power-law tail to the DISKPBB component only provides a marginal improvement to the fit ($\chi ^{2}_{\nu }$ = 1686/1580). Furthermore, we note that as with the first epoch, even though the best fit photon index is rather hard, this portion of the continuum again only dominates outside of the observed bandpass, and the observed continuum at the highest energies probed is still actually very steep, as is clear from Figure 6.

We also investigate the short-term variability behavior during these epochs. In order to crudely assess how the variability may evolve with energy, we produce light curves over the energy ranges 3–10 and 10–30 keV for each of the NuSTAR observations, and estimate the fractional excess variability (F_var; Edelson et al. 2002; Vaughan et al. 2003) for each energy band for the two epochs. In order to ensure the same timescales are probed during each epoch, we split the light curves into ∼70 ks segments, calculate F_var for each and compute the weighted average. We use 70 ks segments as this is roughly the duration of the last part of the second observation, and it divides the first observation neatly into three segments. As the central part of observation 2 does not span sufficient duration, these data are not considered here, and we only utilize the first 70 ks of the first part. In order to ensure the variability above 10 keV is not dominated by Poisson noise, we again use time bins of 5 ks. The results are shown in Figure 8. During the first epoch, mild variability is observed, which seems to be constant with energy. However, during the second epoch, while the variability below 10 keV is broadly consistent with the level displayed during the first, we see some evidence for enhanced variability above 10 keV, both in comparison to the variability below 10 keV during this epoch, and the variability observed during the first epoch.

Zoom In Zoom Out Reset image size

We have presented an analysis of the broadband X-ray spectrum of the extreme ULX Holmberg IX X-1, observed twice during 2012, by NuSTAR, XMM-Newton, and Suzaku. NuSTAR has provided the first high quality hard X-ray (E > 10 keV) spectra of this remarkable source to date. During both epochs the hard X-ray emission is weak in comparison to the soft X-ray (E < 10 keV) emission, as demonstrated by the fluxes presented in Table 3; the flux above 10 keV represents at most ∼25% of the full 0.3–30.0 keV flux observed during either epoch. The observed 0.3–30.0 keV X-ray luminosities from the two epochs (assuming isotropy and before any absorption corrections) are L_{X, 1} = (1.90 ± 0.03) × 10⁴⁰ erg s⁻¹ and L_{X, 2} = (3.35 ± 0.03) × 10⁴⁰ erg s⁻¹, for a distance to Holmberg IX of 3.55 Mpc (Paturel et al. 2002). Correcting for the absorption column inferred from our spectral analysis¹⁴ (Table 2), these correspond to intrinsic 0.3–30.0 keV luminosities of L_{int, 1} = (2.21 ± 0.05) × 10⁴⁰ erg s⁻¹ and L_{int, 2} = (3.91 ± 0.08) × 10⁴⁰ erg s⁻¹.

The NuSTAR data confirm that the curvature observed previously in the 3–10 keV bandpass (e.g., Stobbart et al. 2006; Gladstone et al. 2009; Walton et al. 2013b) is a genuine spectral cutoff, as is suggested by the weak INTEGRAL detection presented by Sazonov et al. (2013). This is also similar to the results observed from other extreme (L_X ≳ 10⁴⁰ erg s⁻¹) ULXs observed by NuSTAR to date, e.g., Circinus ULX5 (Walton et al. 2013a), NGC 1313 X-1 (Bachetti et al. 2013), and IC 342 X-1 (Rana et al. 2014). Indeed, the broadband spectrum of Holmberg IX X-1 from the first epoch is remarkably similar to the latter two sources. Neither epoch is consistent with emission from the standard low/hard state commonly exhibited by Galactic BHBs (see Remillard & McClintock 2006), as would broadly have been expected if Holmberg IX X-1 were an IMBH accreting at substantially sub-Eddington accretion rates, under the assumption that the accretion geometry is independent of black hole mass. It therefore seems likely that we are viewing an unusual high-Eddington phase of accretion, the physics of which we are only just beginning to probe.

During each epoch, the spectra appear to be best described with a combination of two thermal components, potentially with a power-law-like tail present at the highest energies observed, although the contribution of this latter component is somewhat model dependent. In both epochs, one of these components is consistent with being a standard thin accretion disk (Shakura & Sunyaev 1973), although the second (hotter) thermal component deviates from this independent of the inclusion of a high energy tail. The hotter component can be modeled with either an optically thick thermal Comptonization model (which strongly requires the additional high energy power-law-like tail), or a multi-color blackbody disk model in which the radial temperature profile deviates from that expected from a simple thin disk. If an additional power-law-like emission component truly is present at the highest energies probed, it generally appears to be very steep (see Figure 9), potentially similar to the very high/steep power-law state exhibited by Galactic BHBs at relatively high accretion rates (Remillard & McClintock 2006).

Zoom In Zoom Out Reset image size

Two component thermal models could plausibly represent a variety of scenarios. The cooler DISKBB could be standard disk emission (see also Miller et al. 2013b) from the outer accretion flow, while the hotter, nonstandard thermal component could arise from the inner disk where the effects of radiation pressure are the most prominent, and drive the disk structure away from the standard thin disk scenario (e.g., Abramowicz et al. 1988). Alternatively, the soft thermal component could be emission from the photosphere of a massive optically thick outflow (e.g., Middleton et al. 2011a; Sutton et al. 2013), which may display some similarity to standard disk emission, with the hotter emission again arising in the distorted inner disk, broadly similar to the picture for high/super-Eddington accretion proposed by Poutanen et al. (2007); see also Dotan & Shaviv (2011) and Ohsuga & Mineshige (2011).

Alternatively, Dexter & Quataert (2012) proposed a framework in which the surface of the disk is inhomogeneous, resulting in surface temperature fluctuations and thus deviations in the relative contribution of the emission at different temperatures from the simple thin disk approximation. Such inhomogeneities could potentially result in a highly distorted thermal spectrum, and might possibly be able to simultaneously explain both of the thermal components required to model the observed spectra, as suggested in the context of ULXs by Miller et al. (2013b, 2014). As discussed in Miller et al. (2014), such inhomogeneities could arise naturally through the "photon-bubble" instability (e.g., Gammie 1998; Begelman 2002), which may play an important role in high-Eddington accretion flows.

Finally, we stress again that while the hotter component has also previously been interpreted as an optically thick corona (e.g., Gladstone et al. 2009; Pintore et al. 2014), the necessity for an additional Comptonizing continuum at higher energies with such a model renders such a physical interpretation highly unlikely, and favors the hot, inner accretion disk scenarios outlined above.

5.1. Spectral Variability

We have presented results from the coordinated broadband X-ray observations of the extreme ULX Holmberg IX X-1 performed by NuSTAR, XMM-Newton and Suzaku in late 2012. The NuSTAR detections provide the first high-quality spectra of Holmberg IX X-1 at hard X-rays to date, extending our X-ray coverage up to ∼30–35 keV. Observations were undertaken during two epochs, between which Holmberg IX X-1 exhibited strong spectral variability. Neither epoch is consistent with emission from the standard low/hard accretion state, as would have been expected if Holmberg IX X-1 harbors an IMBH accreting at substantially sub-Eddington rates; the NuSTAR data confirm in each case that the curvature observed previously in the 3–10 keV bandpass is a true spectral cutoff. Instead, the spectrum appears to be dominated by two optically thick thermal components, likely associated with a distorted accretion disk, with a nonthermal tail also potentially detected at the highest energies probed. The data allow for either of the two thermal components to dominate the spectral evolution, although we find that both scenarios require highly nonstandard behavior for behavior for thermal accretion disk emission. Further broadband observations covering a broader range of fluxes will be required to determine which component is truly dominating the observed evolution.

The authors would like to thank the referee for providing useful feedback, which helped to improve the manuscript. This research has made use of data obtained with the NuSTAR mission, a project led by the California Institute of Technology (Caltech), managed by the Jet Propulsion Laboratory (JPL) and funded by NASA, XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA Member States and NASA, and Suzaku, a collaborative mission between the space agencies of Japan (JAXA) and the USA (NASA). We thank the NuSTAR Operations, Software, and Calibration teams for support with the execution and analysis of these observations. This research was supported under NASA grant No. NNG08FD60C, and has made use of the NuSTAR Data Analysis Software (NUSTARDAS), jointly developed by the ASI Science Data Center (ASDC, Italy) and Caltech (USA). We also made use of the NASA/IPAC Extragalactic Database (NED), which is operated by JPL, Caltech, under contract with NASA. Many of the figures included in this work have been produced with the Veusz plotting package: http://home.gna.org/veusz, written and maintained by Jeremy Sanders. DB and MB are grateful to the Centre National d'Etudes Spatiales (CNES) for funding their activities.

Facilites: NuSTAR - The NuSTAR (Nuclear Spectroscopic Telescope Array) mission, XMM - Newton X-Ray Multimirror Mission satellite, Suzaku - Suzaku (ASTRO-EII)