CHANDRA IDENTIFICATION OF 26 NEW BLACK HOLE CANDIDATES IN THE CENTRAL REGION OF M31

We obtained accurate positions for our X-ray sources by registering 27 X-ray sources associated with M31 GCs to the M31 Field 5 B-band image provided by Massey et al. (2006). The rms offsets were 011 in R.A. and 009 in decl. (Barnard et al. 2012a); this would be extremely unlikely unless the 27 X-ray sources used for calibration were indeed associated with the GCs.

The X-ray spectra of BH binaries are usually described with two components: a thermal component (often modeled as a multi-temperature disk blackbody), and a power law component to represent unsaturated, inverse-Compton scattering of cool photons on hot electrons (Remillard & McClintock 2006). The hard state is classified by a power law component with photon index (Γ) = 1.4–2.1 and a thermal component that contributes <20% of the 2–20 keV flux (Remillard & McClintock 2006).

Our very first GC BHC (XB045, XBo 45 in Barnard et al. 2008) was associated with the M31 GC B045, named following the Revised Bologna Catalogue v.3.4 (RBC; Galleti et al. 2004, 2006, 2007, 2009). Its ∼17,000 count XMM-Newton/pn spectrum was well described by an absorbed power law with line-of-sight absorption (N_H) = 1.41 ± 0.11 × 10²¹ atom cm⁻², and Γ = 1.45 ± 0.04; χ²/dof = 517/487 (good fit probability = 0.17). Adding a blackbody component improved the fit somewhat, but the thermal contribution to the flux was too small to be constrained; we therefore considered XBo 45 to be in its hard state. Since its unabsorbed 0.3–10 keV luminosity was 2.5 ± 0.2 × 10³⁸ erg s⁻¹, 140% Eddington for a 1.4 M_☉ NS, we considered it a BHC.

The eight remaining GCs BHCs are included in this survey, unlike XBo 45, so we do not describe each spectrum in detail here. However, we note that fitting a two component model to XB144 (XBo 144 in Barnard & Kolb 2009) yielded kT = 0.0082 ± 0.0016 keV, indicating a complete lack of a thermal component in the spectrum. We also note that even though XB082 was best described by Γ = 1.20 ± 0.09, the χ²/dof for that fit was ∼0.9, and we were able to obtain fits where Γ = 1.4 and χ²/dof < 1 (Barnard et al. 2011a).

The unabsorbed 0.3–10 keV luminosities of our GC BHCs exhibiting hard state spectra range over ∼5–45 × 10³⁷ erg s⁻¹ (Barnard & Kolb 2009; Barnard et al. 2011a, 2012b). Comparison with the 0.5–10 keV AGN flux distribution obtained by Georgakakis et al. (2008) yields a 2.4 × 10⁻³⁶ probability that our GC BHCs are coincident AGN. The probability that our brightest GC BHC, XB135, is a coincident AGN is 1.2 × 10⁻⁶. We will present evidence in a separate paper that XB135 may contain the most massive stellar mass BH known to date; it may have been formed by direct collapse of a high mass, metal poor star (R. Barnard et al. 2013, in preparation).

We found that the GCs hosting these very bright X-ray sources were significantly more massive and/or metal rich than the other GCs in M31, agreeing with previous work. However, two GCs had particularly low metallicities; one of these is B135, consistent with the direct collapse formation scenario for XB135 (see Barnard et al. 2012b and references within). Belczynski et al. (2010) have shown that while BH masses are limited to ∼15 M_☉ for solar metallicities, they could theoretically reach ∼80 M_☉ of metallicities ∼0.01 Solar.

High accretion rate NS XBs exhibit multi-component emission that may appear to be hard state spectra in extragalactic X-ray sources, where the spectra have relatively few photons. To compare the emission spectra of our BHCs with NS XB, we use the double thermal (disk blackbody + blackbody) model used by Lin et al. (2007, 2009, 2012) to describe hundreds of RXTE observations of NS XBs in all their varied emission states. We note that the physical interpretation of their model is contradicted for persistent XBs by a substantial body of work; however, Lin et al. (2007, 2009, 2012) have sampled spectra from the full range of NS LMXB emission states in a consistent manner, allowing us to compare our BHCs with NS XBs in a single parameter space.

For a long time, NS XBs were divided into two types, Z-sources (high luminosity) and atoll sources (low luminosity), based on their luminosities and color–color diagrams (CDs): the CDs of Z-sources had three branches (horizontal, normal, and flaring), and evolved along these branches without ever jumping from one branch to the other; the CDs of the atoll sources were more fragmented (Hasinger & van der Klis 1989). Furthermore, the Z-sources were split into those like Cygnus X-2 and those like Scorpius X-1 (Hasinger & van der Klis 1989). It was believed that the differences were due to more than just the accretion rate because the Z-sources varied by a factor of a few when tracing their Z-shaped CDs, while atoll source intensities varied by one to two orders of magnitude (Muno et al. 2002). However, we now know of two transient NS systems that exhibited both types of Z-source behavior before evolving to atoll-source behavior during decay (Homan et al. 2007; Chakraborty et al. 2011).

Lin et al. (2007) examined the spectral evolution of two Galactic X-ray transients, Aql X-1 and 4U 1608−52, over many RXTE observations covering >20 outbursts. They devised a new double-thermal model (disk blackbody + blackbody) to describe an NS transient soft state that is analogous to the BH soft state described by Remillard & McClintock (2006). They have since applied their model to RXTE observations of XTE J1701−462, one of the transients that exhibits Cyg-like and Sco-like Z-source behavior as well as atoll behavior (Lin et al. 2009), and also to the Sco-like Z-source GX 17+2 (Lin et al. 2012). They have applied their model to hundreds of RXTE spectra from Galactic NS binaries, including the full gamut of NS spectral behavior.

They found that their disk blackbody + blackbody was unsuccessful in two situations. First, they found the hard state spectra to be power law dominated, as expected (Lin et al. 2007, 2009). We therefore expect our BHCs to inhabit a separate parameter space than the NS XBs because fitting the double thermal model to hard state spectra will yield unphysical results. Second, they found that Z-sources required a three-component spectrum (disk blackbody + blackbody + power law) on the horizontal branch (Lin et al. 2009, 2012).

Lin et al. (2010) also performed broadband analysis of Suzaku and BeppoSAX observations of the persistently bright Galactic NS XB 4U 1705−44, with energy ranges 0.1–600 keV and 0.1–300 keV, respectively. When using a disk blackbody + blackbody model to the soft state spectra, they found temperatures that were similar to those they observed in the RXTE observations (Lin et al. 2007, 2009, 2012). Fitting an additional Comptonization component, via either a power law or simpl convolution model (following Steiner et al. 2009), resulted only in small changes in the thermal components; this is because the Comptonized component only contributed ≲10% of the total flux in the soft state (see Figure 6 in Lin et al. 2010). Hence, the parametric differences between our BHCs and the NS XBs should not be due to differences in energy bands used in the observations, or the lack of a third (power law) component.

Evidence against the double thermal model indicates an extended corona that contributes a substantial portion of the X-ray flux. This evidence includes the ingress times of periodic absorption dips in the X-ray lightcurves of the high inclination XBs (the dipping sources, see, e.g., Church 2001; Church & Balucińska-Church 2004 and references within), and also broadened emission lines in a Chandra grating spectroscopy of Cyg X-2 (Schulz et al. 2009). Furthermore, compact corona models where the inner disk temperature is tied to the seed photon energy for Comptonization are rejected for ultraluminous X-ray sources in NGC 253 and the confirmed BH+Wolf–Rayet binary IC10 X-1 (Barnard 2010), as well as BHC3 in the steep power law state (Barnard et al. 2011b).

In Barnard et al. (2012b), we applied structure function (SF) analysis to XBs for the first time, following Vagnetti et al. (2011), who created an ensemble SF for AGN. They used a SF to estimate the mean intensity deviation for data separated by time τ:

$$\begin{equation} {\rm SF}\left(\tau \right) \equiv \sqrt{\frac{\pi }{2}\left<|\log f_{\rm X} \left(t+\tau \right) - \log f_{\rm X} \left(t\right)| \right>^2 - \sigma _{\rm n}^2}, \end{equation} \tag{ 1 }$$

where σ_n is the photon noise and f_X is the X-ray flux. They grouped the SF into logarithmic bins with width 0.5; each bin in the range log(τ) = 0.0–3.0 contained more than 100 measurements.

Our sample consisted of 37 X-ray sources associated with objects in the RBC; these were classified by Caldwell et al. (2009) as 30 confirmed GCs, 4 GC candidates, 1 star, and 2 AGN. The SFs of GC XBs with 0.3–10 keV luminosities ∼2–50 × 10³⁶ erg s⁻¹ tended to show significantly more variability than AGN over a wide range of time-scales. The SFs of brighter XBs generally showed comparable or less variability than AGN, despite their high signal to noise; however, their high fluxes made them unlikely AGN. Hence, SFs provide an effective mechanism for distinguishing between XBs and AGN (Barnard et al. 2012b).

In this work, we combine these techniques to search for BHCs in the whole region covered by 152 Chandra observations from our monitoring program; the roll angle is unrestricted, resulting in an approximately circular field of view with radius ∼20'. We surveyed 530 X-ray sources in this region. We also obtain spectra from archival XMM-Newton observations to strengthen the cases for several BHCs.

The 100'' region surrounding M31* is particularly interesting, because Voss & Gilfanov (2007) found an excess of XBs over the radial distribution expected from K-band light (tracing stellar mass); this excess had the distribution expected of dynamically formed XBs (Fabian et al. 1975). Dynamical XB formation requires stellar densities rarely seen outside GCs, and Voss & Gilfanov (2007) proposed that the M31 bulge is sufficiently large and dense to form a significant number of XBs dynamically. However, since the stellar velocities in the M31 bulge are ∼5–10 times higher than in GCs, Voss & Gilfanov (2007) expect only short period binaries to survive, with most of these containing BH accretors. Evidence for BHCs in this region would therefore provide strong support for their theory.

In the next section, we discuss the observations and data analysis; this is followed by our results, then by our discussion.

The central region of M31 has been observed with Chandra on a ∼ monthly basis for the last ∼13 yr in order to monitor transients; we exclude periods when M31 cannot be observed due to satellite constraints (∼ March–April). We have analyzed 98 ACIS observations and 54 HRC observations. We determined the position of each source from a merged 0.3–7.0 keV ACIS image, using the iraf tool imcentroid. The X-ray positions were registered to the Field 5 B-band image of M31 from the Local group Galaxy Survey (LGS; see Massey et al. 2006), using 27 GC X-ray sources and following the procedure described in Barnard et al. (2012a). The rms uncertainties in registration were 011 in R.A. and 009 in decl.

We obtained 0.3–7.0 keV lightcurves and spectra from circular source and background regions for each source. The background region was the same size as the source region and at a similar off-axis angle. The extraction radius varied between sources because larger off-axis angles resulted in larger point spread functions. We used the CIAO v4.5 software suite, with corresponding CALDB to reduce the data, and XSPEC v12.7 to analyze the spectra.

For ACIS observations, we obtained corresponding response matrices and ancillary response files. We initially estimated the conversion from flux to luminosity by assuming a power law emission spectrum with photon index 1.7, with N_H = 7 × 10²⁰ atom cm⁻¹, then determining the unabsorbed 0.3–10 keV luminosity equivalent to 1 count s⁻¹ at the location of the source. After correcting for the exposure, vignetting and background, multiplying the source intensity by this conversion factor gave the source luminosity. Source spectra with >200 net counts were freely fitted. When good spectral fits were found for a particular observation of a source, the parameters of these fits replaced the default parameters when estimating the source luminosity for observations of the source with <200 photons.

For HRC observations, we included only PI channels 48–293, thereby reducing the instrumental background. We used the WebPIMMS tool to find the unabsorbed luminosity equivalent to 1 count s⁻¹, assuming the same emission model as for the ACIS observations with <200 photons. We created a 1 keV exposure map for each observation, and compared the exposure within the source region with that of an identical on-axis region, in order to estimate the necessary exposure correction. We multiplied the background subtracted source intensity by the correction factor to get the 0.3–10 keV luminosity.

We created long term 0.3–10 keV lightcurves for each source, using the luminosities obtained from each observation as described above; all luminosities assume a distance of 780 kpc (Stanek & Garnavich 1998). We only included observations with net source counts ⩾0 after background subtraction. We fitted each long term lightcurve with a line of constant intensity in order to ascertain the source variability.

We note that the ratio of HRC to ACIS luminosity depends strongly on the spectral model and should be 1.0 if the model is correct. Differences between actual and assumed emission spectra during HRC and faint ACIS observations may lead to systematic offsets between luminosities from adjacent HRC and ACIS observations.

We derived SFs from the 0.5–4.5 keV fluxes of each observation of every target by assuming a power law spectrum with the same photon index as for the HRC and faint ACIS observations; an M31 X-ray source with a 0.3–10 keV unabsorbed luminosity of 1.0 × 10³⁷ erg s⁻¹, N_H = 7 × 10²⁰ atom cm⁻², and Γ = 1.7 has a 0.5–4.5 keV flux of 0.80 × 10⁻¹³ erg cm⁻² s⁻¹.

Vagnetti et al. (2011) calculated the noise component from

$$\begin{equation} \sigma_{\rm n}^2 = 2\left<\left(\delta \log f_{\rm X}\right)^2\right> \simeq 2\left(\log e\right)^2 \left< \left(\frac{\delta f_{\rm X}}{f_{\rm X}}\right)^2\right>, \end{equation} \tag{ 2 }$$

assuming that δf_X/f_X = (1/N_phot)^0.5 and N_phot is the number of photons. Our lightcurves are background subtracted and ARF-corrected; furthermore, uncertainties in the luminosities of bright sources include uncertainties in the spectral parameters. As a result, our uncertainties are not simply due to photon counting noise. Hence in our case,

$$\begin{equation} \sigma_{\rm n}^2 \simeq \left(\log e\right)^2 \left< \left[ \frac{\sigma f_{\rm X}\left(t+\tau \right)}{f_{\rm X}\left(t+\tau \right)}\right]^2 + \left[ \frac{\sigma f_{\rm X}\left(t\right)}{f_{\rm X}\left(t\right)} \right]^2\right>, \end{equation} \tag{ 3 }$$

where σf_X is the uncertainty in X-ray flux and often significantly larger than $N_{\rm phot}^{0.5}$. In the simplest case where only one pair has a particular separation, then

$$\begin{equation} \sigma_{\rm n}^2 \simeq \left(\log e\right)^2 \left[ \left(\frac{\sigma f_1}{f_1}\right)^2 + \left(\frac{\sigma f_2}{f_2} \right)^2\right], \end{equation} \tag{ 4 }$$

where f₁ and f₂ are the fluxes of the two observations.

In addition to our Chandra observations, we also obtained spectra for some of our BHCs from archival XMM-Newton observations. Their higher sensitivity and longer exposure times often yielded superior spectra for bright sources. We used only the pn instrument because the MOS detectors are more prone to pile up. We used SAS v10 for the data reduction.

We filtered out background flares by creating a lightcurve with the selection "(PATTERN==0)&&(FLAG==0)&&(PI in [10000:12000])" and 100 s binning, then excluding intervals with >0.4 counts s⁻¹. Source and background spectra were obtained from circular regions with 20''–40'' radius, along with corresponding response files; the filter "(PATTERN<=4)&&(FLAG==0)" was applied along with the good time filter and extraction region.

Our BHC identification process involved three steps. First, we had to establish that the X-ray source was an XB rather than an AGN. Then, we checked the ACIS observations of each XB, to see if it exhibited a high luminosity hard state (Γ ⩽ 2.1 at a 0.3–10 keV unabsorbed luminosity >3 × 10³⁷ erg s⁻¹, i.e., 10% Eddington for a 2 M_☉ NS). Finally, we found the highest quality observation that exhibited a high luminosity hard state, fitted it with the double-thermal model, and compared the results with the known range of parameters exhibited by the NS systems studied by Lin et al. (2007, 2009, 2012).

Likely XBs were identified from their associations with GCs, variable SFs, or relatively high fluxes. There are ∼170 GCs within our field (Peacock et. al. 2011); using the 0.5–10 keV AGN flux distribution (Georgakakis et al. 2008), we expect to find 0.0005 AGN with flux >1.4 × 10⁻¹³ erg cm⁻² s⁻¹ (10³⁷ erg s⁻¹) within 1'' of any GC. Hence, any bright X-ray source associated with a GC is a likely XB. We also identify XBs using SFs that are strikingly more variable than the ensemble AGN SF. Furthermore, only ∼1 AGN deg⁻² is expected with a flux equivalent to ∼5 × 10³⁷ erg s⁻¹; hence, brighter X-ray sources are likely XBs too.

Lin et al. (2007, 2009) found that they were unable to successfully fit hard state spectra with their double thermal model. This is because the disk blackbody component must account for the low-energy flux, resulting in an unphysically low temperature. To demonstrate this, we created a fake hard state spectrum using the xspec command fakeit; we used a power law with Γ = 1.7, with absorption equivalent to 7 × 10²⁰ atom cm⁻². We present the best fit double thermal model for this spectrum in Figure 1, which consists of a 0.67 keV disk blackbody and a 1.76 keV blackbody; the blackbody component contributes ∼80% of the 2–10 keV flux. By contrast, the soft states of NS XBs studied by Lin et al. (2007, 2009, 2012) yielded disk blackbody temperatures ≳1 keV, with the blackbody component contributing less than 50% of the 2–10 keV flux.

Zoom In Zoom Out Reset image size

Since Lin et al. (2007, 2009, 2012) failed to fit hard state spectra with the double thermal model, we expect disk blackbody + blackbody fits to our BHCs to inhabit a different parameter space from the NS binaries if they really are in the hard state. If a BHC exhibits a blackbody temperature <1.5 keV, a disk blackbody temperature <1.0 keV (<1.2 keV for disk blackbody 2–10 keV luminosity >2 × 10³⁷ erg s⁻¹), and/or a blackbody contribution to the total 2–10 keV flux >45% at a 3σ level, then we consider it a strong BHC; otherwise, we label it a plausible BHC. These criteria are drawn from the hundreds of RXTE spectra of Aql X-1, 4U 1608−52, XTE J1701−462, and GX 17+2 analyzed by Lin et al. (2007, 2009, 2012).

Figure 2 shows a detail of the B-band image of M31 Field 5 from the LGS (Massey et al. 2006), superposed with the positions of our BHCs; north is up, east is left. The circle encloses the region within 100'' (∼380 pc) of M31*; this region is enlarged in the top right portion of the figure. BHCs 1 and 8 lie below the southern edge of Field 5.

Zoom In Zoom Out Reset image size

3.1.1. The Central 100'' Region

3.1.2. Further GC Associations

We initially identified most of our BHCs from their long term (∼13 yr) light curves and spectral histories. They exhibited spectra consistent with the hard state (Γ < 2.1) at 0.3–10 keV luminosities >3 × 10³⁷ erg s⁻¹ in at least one observation. For such spectra, we expect the 0.01–1000 keV luminosities to be ≫3 × 10³⁷ erg s⁻¹, our limit for NS hard states following Gladstone et al. (2007).

We present the ∼13 yr, corrected, 0.3–10 keV luminosity lightcurve for each BHC in Figure 3, along with its spectral history. The top panel in each case shows the lightcurve, with ACIS and HRC observations represented by circles and crosses respectively. The bottom panel shows Γ for each ACIS observation: sufficiently bright observations have freely fitted Γ, while faint observations have Γ fixed to the mean value; this approach should be valid because we expect the faint observations to be in the hard state also, with similar a emission spectrum.

Zoom In Zoom Out Reset image size

View extended figure (6 panels)

Figure 4 shows histograms of the number of observation pairs with separation τ for the most frequently observed BHC, BHC13. The histogram with filled circles includes ACIS and HRC observations, while the open circle histogram contains only ACIS observations. The observation pairs were logarithmically binned by τ, with a bin width of 0.2 dex.

Zoom In Zoom Out Reset image size

We present the 0.5–4.5 keV SF for each of our BHCs in Figure 5, with the same scheme as for Figure 4. The ensemble SF for typical AGN obtained by Vagnetti et al. (2011) is represented by a dashed line in each panel: SF(τ) = 0.11τ^{0.10 ± 0.01}. We included an ACIS only SF for each source in addition to an ACIS + HRC SF because faint HRC observations add extremely large noise components for some BHCs, suppressing the variation; BHCs 8 and 35 only show significant excess variability over typical AGN in their ACIS only SFs.

Zoom In Zoom Out Reset image size

View extended figure (6 panels)

We find that 28 out of 35 BHCs exhibit SFs with significantly more variation than the Vagnetti et al. (2011) ensemble AGN SF (>4σ excess in at least one τ channel, >3σ excess in at least two channels, or >2σ excess in at least three channels over the 3σ limit for the AGN SF, assumed to be 0.11τ^0.13). This distinction is important because AGN and XBs often have very similar spectra. The X-ray sources with SFs exhibiting similar or less variation than the ensemble AGN SF (BHCs 2, 5, 7, 21, 23, 24, 25, and 34) had mean 0.3–10 keV fluxes ≳ 5.5 × 10⁻¹³ erg cm⁻² s⁻¹.

We have found 24 X-ray sources in our observations that exhibited mean 0.3–10 keV fluxes ≳ 5.5 × 10⁻¹³ erg cm⁻² s⁻¹ over the ∼13 yr of Chandra monitoring; this includes 8 GC XBs and 1 possible supernova remnant. The 0.5–10 keV flux distribution of AGN found by (Georgakakis et al. 2008) predicts 0.6 AGN at this flux level within our observed region (approximated by a circle with radius ∼20' or ∼4 kpc). Since the observed number of unidentified X-ray sources at these flux levels is larger than the predicted AGN number by a factor ∼25, we conclude that they are likely XBs.

After the initial identification of candidates, we sought the observations that best represented the BHC case for each object; these were the highest quality spectra that appeared to exhibit high luminosity hard states. Table 2 provides the details of these observations. For each BHC we give the observation number and net sources counts. We then give the absorption (N_H), photon index (Γ), χ²/dof, and 0.3–10 keV luminosity for the best fit absorbed power law model. Observations 303–13827 are Chandra ACIS observations, while 0109270101, 0112570101, and 0402560901 are XMM-Newton observations.

We note that the spectrum of BHC 22 rejected simple absorbed power law models. It is well described by an absorbed disk blackbody + power law model with T_in = 1.97$_{-0.3}^{+0.19}$ keV and Γ = 2.16$_{-0.19}^{+0.4}$, with the power law contributing 61% ± 12% of the 8.0 ± 1.1 × 10³⁷ erg s⁻¹ 0.3–10 keV luminosity; it is consistent with either of two canonical BH states: the hard state, or a steep power law state similar to GRS 1915+105 (McClintock & Remillard 2006).

We applied a disk blackbody + blackbody emission model to our spectra; we summarize our findings in Table 3. The disk blackbody components were generally better constrained than the blackbody components, particularly for spectra with relatively few net source counts; this is simply because there are fewer high energy photons than low energy photons and the disk blackbody is cooler than the blackbody. Our best fits yielded blackbody contributions of 65%–99% to the 2–10 keV flux, in stark contrast to the NS systems where the disk blackbody dominates. The blackbody component dominates the 2–10 keV flux most when the disk blackbody temperature is lowest. Four of our BHCs only exhibited apparent high luminosity hard states during 5 ks ACIS observations, and we were unable to constrain double thermal models for the resulting spectra. However, future deep observations may strengthen their BHC cases.

Table 3. Best Fit Parameters for the Blackbody + Disk Blackbody Model

BHC	NH (1022 cm−2)	kTBB	LBB (1037)	kTDBB	LDBB/1037	LBB/LTOT	BH
		(keV)		(keV)
1	0.34 ± 0.07	2.00$_{-0.19}^{+0.6}$	21 ± 2	0.88$_{-0.16}^{+0.4}$	7.6 ± 0.5	0.73 ± 0.09	P
2	0.07f	1.12$_{-0.11}^{+0.4}$	3.0 ± 0.5	0.50$_{-0.10}^{+0.3}$	0.8 ± 0.4	0.8 ± 0.2	P
3	0.07f	1.18$_{-0.08}^{+0.13}$	7.3 ± 0.3	0.49$_{-0.06}^{+0.08}$	0.8 ± 0.3	0.90 ± 0.06	S
4	0.19 ± 0.06	1.50$_{-0.19}^{+1.0}$	3.9 ± 0.8	0.62$_{-0.11}^{+0.2}$	0.8 ± 0.4	0.8 ± 0.2	P
5	0.07f	1.3$_{-0.3}^{+25.0}$	2.0 ± 0.7	0.7$_{-0.2}^{+0.5}$	0.6 ± 0.3	0.8 ± 0.4	P
6	0.07f	1.34$_{-0.05}^{+0.06}$	3.67 ± 0.15	0.510$_{-0.013}^{+0.02}$	0.42 ± 0.05	0.90 ± 0.05	S
7	0.07f	1.47$_{-0.06}^{+0.06}$	4.59 ± 0.13	0.51$_{-0.02}^{+0.03}$	0.54 ± 0.06	0.89 ± 0.04	S
8	...	...	...	...	...	...	P
9	0.07f	0.93$_{-0.06}^{+0.08}$	2.14 ± 0.12	0.38$_{0.04}^{+0.04}$	0.14 ± 0.05	0.94 ± 0.08	S
10	0.07f	2.4$_{-0.9}^{+12.5}$	2.3 ± 0.6	0.86$_{-0.19}^{+0.2}$	0.7 ± 0.3	0.8 ± 0.3	P
11	0.07f	1.11$_{-0.11}^{+0.4}$	3.0 ± 0.5	0.50$_{-0.10}^{+0.3}$	0.8 ± 0.2	0.80 ± 0.18	P
12	0.07f	1.24$_{-0.19}^{+0.7}$	1.74 ± 0.17	0.59$_{-0.08}^{+0.10}$	0.7 ± 0.3	0.72 ± 0.12	S
13	0.07f	1.5$_{-0.3}^{+1.4}$	1.8 ± 0.4	0.67$_{-0.12}^{+0.2}$	0.5 ± 0.3	0.8 ± 0.2	P
14	...	...	...	...	...	...	P
15	0.07f	1.12$_{-0.11}^{+0.4}$	1.6 ± 0.2	0.56$_{-0.11}^{+0.3}$	0.6 ± 0.3	0.74 ± 0.17	P
16	0.07f	2.2$_{-0.5}^{+2.4}$	4.5 ± 0.5	1.03$_{-0.19}^{+0.2}$	2.4 ± 0.8	0.65 ± 0.11	P
17	0.07f	0.96$_{-0.09}^{+0.13}$	1.17 ± 0.09	0.43$_{-0.05}^{+0.06}$	0.17 ± 0.07	0.87 ± 0.10	S
18	0.07f	1.05$_{-0.05}^{+0.08}$	4.44 ± 0.19	0.41$_{-0.04}^{+0.06}$	0.22 ± 0.08	0.95 ± 0.06	S
19	0.26 ± 0.04	1.7$_{-0.3}^{+1.1}$	3.6 ± 0.5	0.83$_{-0.19}^{+0.2}$	1.7 ± 0.5	0.68 ± 0.13	P
20	0.173 ± 0.010	1.61$_{-0.17}^{+0.19}$	18.1 ± 1.7	0.83$_{-0.10}^{+0.12}$	6.9 ± 1.9	0.7 ± 0.2	S
21	0.1 ± 0.05	1.7$_{-0.3}^{+0.8}$	4.1 ± 0.5	0.77$_{-0.13}^{+0.2}$	1.9 ± 0.7	0.68 ± 0.12	P
22	0.07f	1.31$_{-0.05}^{+0.06}$	3.71 ± 0.13	0.55$_{-0.03}^{+0.03}$	0.57 ± 0.08	0.87 ± 0.04	S
23	0.07f	0.99$_{-0.07}^{+0.13}$	2.6 ± 0.5	0.45$_{-0.06}^{+0.13}$	0.25 ± 0.13	0.9 ± 0.2	S
24	0.07f	1.12$_{-0.05}^{+0.06}$	2.12 ± 0.09	0.43$_{-0.02}^{+0.03}$	0.19 ± 0.04	0.92 ± 0.05	S
25	0.07f	1.56$_{-0.08}^{+0.10}$	3.22 ± 0.13	0.61$_{-0.04}^{+0.04}$	0.39 ± 0.07	0.89 ± 0.05	S
26	0.07f	2.3$_{-1.6}^{+20.0}$	2.1 ± 0.6	0.78$_{-0.16}^{+0.2}$	0.5 ± 0.2	0.8 ± 0.3	P
27	0.07f	0.94$_{-0.08}^{+0.3}$	1.56 ± 0.14	0.46$_{-0.08}^{+0.13}$	0.28 ± 0.15	0.85 ± 0.12	S
28	0.07f	1.4$_{-0.3}^{+1.0}$	2.3 ± 0.5	0.63$_{-0.12}^{+0.2}$	0.9 ± 0.4	0.71 ± 0.2	P
29	0.07f	1.00$_{-0.05}^{+0.07}$	16.4 ± 0.9	0.29$_{-0.04}^{+0.06}$	0.22 ± 0.11	0.99 ± 0.08	S
30	...	...	...	...	...	...	P
31	0.07f	1.36$_{-0.04}^{+0.05}$	5.72 ± 0.15	0.45$_{-0.02}^{+0.02}$	0.4 ± 0.05	0.93 ± 0.03	S
32	...	...	...	...	...	...	P
33	0.07f	1.15$_{-0.12}^{+0.3}$	1.87 ± 0.17	0.52$_{-0.08}^{+0.12}$	0.38 ± 0.17	0.83 ± 0.12	S
34	0.07f	1.56$_{-0.11}^{+0.13}$	3.7 ± 0.3	0.63$_{-0.04}^{+0.04}$	0.67 ± 0.12	0.85 ± 0.08	S
35	0.07f	1.44$_{-0.19}^{+1.2}$	2.4 ± 0.4	0.68$_{-0.11}^{+0.18}$	0.8 ± 0.3	0.75 ± 0.17	P

Notes. We provide the absorption, blackbody temperature, 2–10 keV blackbody luminosity, disk blackbody temperature, and 2–10 keV disk blackbody luminosity. Finally, we give the blackbody contribution to the total 2–10 keV luminosity. Uncertainties are quoted at the 1σ level. The last column states whether the BHC identification is strong (S) or plausible (P).

Download table as:  ASCIITypeset image

The 31 BHCs that we did fit with the double thermal model yielded kT_in 0.6σ–28σ outside the NS parameter space; the combined χ² was 212, for 30 dof (probability ∼3 × 10⁻²⁹). Clearly, our BHCs are inconsistent with being soft NS XBs.

In Figures 6 and 7, we plot the 2.0–10 keV luminosity versus temperature for the disk blackbody and blackbody components of the best fit double thermal models for each of our BHCs. We overlay the parameter space inhabited by the NS systems observed by Lin et al. (2007, 2009, 2012) for spectral states where Comptonization is not required in their models, i.e., the atoll soft states and Z-source normal and flaring branches. We note that Lin et al. quote luminosities from broader energy ranges; however, our disk blackbody components make a negligible contribution above 10 keV, while the luminosities of our blackbody components are typically ∼10% higher in the 2.0–20 keV band than the 2.0–10 keV band; hence, our observed 2.0–10 keV luminosity for our BHCs provide a fair comparison with the NS systems observed by Lin et al. (2007, 2009, 2012).

As predicted, our BHCs inhabit a separate region of the disk blackbody parameter space to the NS systems studied by Lin et al. (2007, 2009, 2012), although some sources are consistent with NS values for T_in within 3σ. We also see a correlation between the 2–10 keV luminosity and temperature of the disk blackbody component; this is due to lower disk temperatures resulting in lesser contributions to the 2–10 keV flux. The BHCs are scattered more widely in the blackbody space, but are also systematically offset from the NS systems.

Remembering that Lin et al. (2009, 2012) failed to fit the horizontal branches of XTE J1701−462 and GX 17+2 with their double thermal model, we also modeled our brightest BHCs (L_0.3–10 > 10³⁸ erg s⁻¹) with the disk blackbody + blackbody + Comptonization emission model that they employed for the horizontal branch spectra. Lin et al. (2012) used a cut-off power law to model the Comptonization in GX 17+2 over the 2.9–60 keV band; they required simultaneous fitting of several spectra to constrain the parameters of the cut-off power law, which yielded Γ = 1.40 ± 0.14 and a cut-off energy of 9.9 ± 1.0 keV. We therefore modeled the 0.3–10 keV XMM-Newton pn spectra, or 0.3–7.0 keV Chandra ACIS spectra of our brightest BHCs with a blackbody, disk blackbody, and a power law with Γ fixed to 1.4. Lin et al. (2012) found that the Comptonized component contributed ∼10%–50% of the total flux on the horizontal branch of GX 17+2; we fixed the flux of the power law component in our fits to ∼30% of the total 2–10 keV flux.

We fitted the spectra of BHCs 1, 3, 16, 20, 21, and 31 in this way; the results are summarized in Table 4. We find that this three component model results in lower temperatures for both thermal components than the double thermal model fits to the same spectra. That is, the addition of the power law component drives our BHCs away from the NS parameter space. Increasing the power law contribution to 50% of the flux decreased the temperatures further. Since none of the spectra from our brightest BHCs have three component fits that are consistent with the fits that successfully described the horizontal branch (Lin et al. 2009, 2012), we conclude that none of our BHCs are Z-sources.

Table 4. Best Three-component Fits to the Spectra of our Brightest BHCs

BHC	NH (1022 cm−2)	kTBB	kTDBB
		(keV)	(keV)
1	0.35 f	1.99$_{-0.4}^{+1.95}$	1$_{-0.3}^{+1.1}$
3	0.07 f	1.07$_{-0.11}^{+0.19}$	0.42$_{-0.1}^{+0.15}$
16	0.07 f	1.44$_{-0.3}^{+...}$	0.8$_{-0.3}^{+0.8}$
20	0.197 ± 0.011	1.29$_{-0.13}^{+0.2}$	0.73$_{-0.13}^{+0.19}$
21	0.11 ± 0.07	1.6$_{-0.4}^{+...}$	0.7$_{-0.2}^{+0.4}$
31	0.07 f	1.27$_{-0.08}^{+0.09}$	0.43$_{-0.03}^{+0.03}$

Notes. Best fit blackbody and disk blackbody temperatures when our brightest BHCs are modeled with a blackbody, a disk blackbody, and a Comptonized component represented by a power law with Γ = 1.4 that contributes ∼30% of the 2–10 keV flux. This is for comparison with the spectral modeling conducted by Lin et al. (2012) for the horizontal branch of GX 17+2.

Download table as:  ASCIITypeset image