AN EXTREME X-RAY DISK WIND IN THE BLACK HOLE CANDIDATE IGR J17091−3624

The HEG spectra were fit with a fiducial spectral model including an effective H column density (TBabs), a disk blackbody component, and a power-law component.

The first observation (MJD 55775) is well described by column density of N_H = 9.9 ± 0.1 × 10²¹ cm⁻² and a disk blackbody temperature of 1.3  ±  0.1 keV. The resulting fit gave a χ²/ν = 2657/3156 = 0.84. This spectrum is dominated by the disk blackbody component, typical of the high soft state of BHB. A power-law continuum component is not statistically required. An unabsorbed flux of $F_{2\hbox {--}10\ {\rm keV}}$ = 1.5 ± 0.1 × 10⁻⁹ erg cm⁻² s⁻¹ was measured.

The second observation (MJD 55841) also had a consistent flux, $F_{2\hbox {--}10\ {\rm keV}}$ = 1.9 ± 0.5 × 10⁻⁹ erg cm⁻² s⁻¹. Again, the column density was large, at N_H = 1.22 ± 0.07 × 10²² cm⁻². A power-law photon index of Γ = 1.7^+0.07_−0.09 and a disk blackbody temperature of 2.3 ± 0.3 keV were measured. This disk temperature is high but common in GRS 1915+105 (see, e.g., Vierdayanti et al. 2010). The resulting χ²/ν was 2754/3414 = 0.81.

In the second HEG spectrum, absorption features are noted in the Fe K band (see Figure 1), and these were initially fit with simple Gaussians. The two strongest lines are found at energies of 6.91 ± 0.01 keV and 7.32^+0.02_−0.06. Via an F-test (see Protassov et al. 2002 for some cautions), these lines are significant at the 99.94% and 99.67% confidence levels, respectively. Dividing the flux normalization of each line by its minus-side error suggests that the feature at 6.91 keV is significant at the 4σ level of confidence, while the 7.32 keV line is marginal at a 2σ confidence level.

Zoom In Zoom Out Reset image size

We also modeled the second observation continuum with a Comptonization model (compTT) instead of the disk blackbody and power law. In general, this gave a reasonable fit at χ²/ν = 2884/3414 = 0.84. This model also showed residual absorption features at high energy, which again we modeled with Gaussian functions. Relative to this continuum, the features at 6.91 keV and 7.32 keV are detected at a higher level of significance (6σ).

It is reasonable to associate the line at 6.91 ± 0.01 keV with He-like Fe xxv, which has a rest energy of 6.70 keV (Verner et al. 1996). This translates into a blueshift of 9300⁺⁵⁰⁰₋₄₀₀ km s⁻¹. This feature clearly signals an extreme disk wind in IGR J17091−3624. Typical velocities in X-ray binaries are <1000 km s⁻¹ (Miller et al. 2006a, 2006b). If the feature at 7.32^+0.02_−0.06 keV is real and can be associated with H-like Fe xxvi at 6.97 keV, it would correspond to a blueshift of 14,600⁺²⁵⁰⁰₋₈₀₀. For additional details, see Table 1. Although less likely, the 6.91 keV line could also be associated with a redshift from the H-like Fe xxvi line. The corresponding inflowing velocity would be 2600  ±  400 km s⁻¹. If this is due to gravitational redshift, the corresponding radius would be 1.7^+0.3_−0.2 × 10⁸ cm (60 ± 10 R_Schw).

Table 1. Spectral Modeling Parameters of the Second HEG Observation

Parameter	Model 1	Model 2	Model 3	Model 4
	diskbb + po	(diskbb + po)	comptt	(comptt)
	+ Gauss + Gauss	× Xstar × Xstar	+ Gauss + Gauss	× Xstar × Xstar
NH (1022 cm−2)	1.14 ± 0.06	1.13 ± 0.06	0.475 +0.017−0.018	0.558+0.025−0.028
⋅⋅⋅
Tin (keV)	1.53 ± 0.09	1.51 +0.11−0.09	⋅⋅⋅	⋅⋅⋅
Norm	13.1+3.6−2.5	13.8+4.0−1.7	⋅⋅⋅	⋅⋅⋅
Γ	1.93 +0.15−0.16	1.91 ± 0.17	⋅⋅⋅	⋅⋅⋅
Norm	0.35 +0.07−0.08	0.34 ± 0.08	⋅⋅⋅	⋅⋅⋅
⋅⋅⋅
T0 (keV)	⋅⋅⋅	⋅⋅⋅	0.58 ± 0.01	0.59 ± 0.01
kT (keV)	⋅⋅⋅	⋅⋅⋅	10.5+30−1.7	9.8 ± 0.02
τplasma	⋅⋅⋅	⋅⋅⋅	2.24 ± 0.01	2.28 ± 0.01
Norm	⋅⋅⋅	⋅⋅⋅	0.0584 ± 0.0001	0.062+0.002−0.05
⋅⋅⋅
$E_{\rm Fe\,\mathsc{xxv}}$ (keV)	6.91 ± 0.01	⋅⋅⋅	6.91+0.02−0.01	⋅⋅⋅
FWHM (keV)	0.091+0.022−0.049	⋅⋅⋅	0.13+0.19−0.04	⋅⋅⋅
EW (keV)	0.021+0.005−0.002	⋅⋅⋅	0.040+0.007−0.009	⋅⋅⋅
Norm (10−4)	3.5 +0.8−0.6	⋅⋅⋅	6.0+1.1−1.3	⋅⋅⋅
v (km s−1)	9300+500−400	⋅⋅⋅	9300+400−800	⋅⋅⋅
⋅⋅⋅
$E_{Fe\,\mathsc{xxvi}}$ (keV)	7.32+0.02−0.06	⋅⋅⋅	7.30 ± 0.02	⋅⋅⋅
FWHM (keV)	0.081+0.079−0.027	⋅⋅⋅	0.25 +0.13−0.01	⋅⋅⋅
EW (keV)	0.032 +0.018−0.004	⋅⋅⋅	0.089+0.013−0.014	⋅⋅⋅
Norm (10−4)	3.4+1.9−0.4	⋅⋅⋅	11.8+1.7−1.6	⋅⋅⋅
v (km s−1)	14,600+2500−800	⋅⋅⋅	13,800 ± 800	⋅⋅⋅
⋅⋅⋅
N (1022 cm−2)	⋅⋅⋅	0.47+0.17−0.19	⋅⋅⋅	0.45+0.33−0.17
log ξ(erg cm s−1)	⋅⋅⋅	3.3 +0.2−0.1	⋅⋅⋅	3.4 +0.2−0.1
v (km s−1)	⋅⋅⋅	9600+400−500	⋅⋅⋅	9600 ± 300
⋅⋅⋅
N (1022 cm−2)	⋅⋅⋅	1.66 +1.18−0.83	⋅⋅⋅	1.97+1.26−0.51
log ξ(erg cm s−1)	⋅⋅⋅	3.9+0.5−0.3	⋅⋅⋅	3.7 +0.3−0.1
v (km s−1)	⋅⋅⋅	15,400 ± 400	⋅⋅⋅	15,400 +400−300
⋅⋅⋅
χ2/ν	2725/3408 = 0.80	2731/3408 = 0.80	2793/3408 = 0.82	2761/3408 = 0.81

Notes. The line detections using Gaussian functions as well as more self-consistent, photoionization components created with XSTAR, assuming two different continuum models. TBabs is applied to all the models and the errors are 1σ confidence level.

Download table as:  ASCIITypeset image

The absence of emission lines in the second spectrum of IGR J17091−3624 is notable, but is only suggestive of an equatorial wind. Given that disk winds have only been detected in sources viewed at high inclination angles, and given the similarities between IGR J17091−3624 and GRS 1915+105, it is likely that IGR J17091−3624 is also viewed at a high inclination. However, there is no evidence of eclipses in this source, so inclinations above 70° can be ruled out.

Absorption lines like those detected in the second observation of IGR J17091−3624 are absent in the first observation. Fits to the Fe K band using a local continuum model and narrow Gaussian functions with FWHM values corresponding to those measured in the second observation give 1σ confidence limits of 3 eV or less on lines in the 6.70–7.32 keV band. This is significantly below the equivalent widths measured in the second observation (see Table 1). This may simply be due to a variable absorption geometry in IGR J17091−3624; this has previously been observed in H 1743−322 and GRS 1915+105 (Miller et al. 2006a, 2006b, 2008; Neilsen & Lee 2009).

To get a better physical picture of the absorption in the second observation of IGR 17091−3624, we also fit the data with a grid of self-consistent photoionization models created with XSTAR (Kallman & Bautista 2001). The ionizing luminosity for this model was derived from extrapolating the unabsorbed spectrum from the second observation to 0.0136–30 keV, ensuring coverage above 8.8 keV, which is required to ionize Fe xxv. A distance of 8.5 kpc is first assumed to derive this luminosity (L_ion = 3.5 × 10³⁷ erg s⁻¹), owing to the location of J17091−3624 within the Galactic bulge. However, Altamirano et al. (2011b) also suggest the possibility that this source could be accreting at high Eddington fractions but further away, and a distance of 25 kpc was also adopted in a second XSTAR grid (L_ion = 3.5 × 10³⁸ erg s⁻¹).

The density of the absorbing material was chosen to be log(n) = 12.0. This is a reasonable assumption based on the modeling of similar X-ray binaries: GX 13+1, n = 10¹³ cm⁻³ (Ueda et al. 2004), GRO J1655−40, n = 10¹⁴ cm⁻³ (Miller et al. 2008), and H 1743−322, n = 10¹² cm⁻³ (Miller et al. 2006b). A turbulent velocity of 1000 km s⁻¹ was found to provide the best fit after various trials. A covering factor of 0.5 was chosen as the absence of emission lines suggests an equatorial wind. Finally, the Fe abundance was assumed to be twice the solar value after initial fits; this characterizes the Fe K lines but does not predict absorption lines, e.g., Si, that are not observed.

The initial, lower luminosity grid was fit to the data in XSPEC as a multiplicative model; free parameters included the column density, ionization, and velocity shifts of the absorbing gas (see Table 1 and Figure 2). For the disk blackbody and power-law continuum, an ionization parameter of log ξ = 3.3^+0.2_−0.1 is required, as well as a wind column density of N = 4.7^+1.7_−1.9 × 10²¹ cm⁻². Velocity shifts consistent with simple Gaussian models are found using the XSTAR grid.

Zoom In Zoom Out Reset image size

To fit the putative higher energy absorption, a second outflow component is required. An additional, lower luminosity XSTAR component is significant at the 3σ level, relative to both continua. The wind column density was higher at N = 1.7^+1.2_−0.8 × 10²² cm⁻², and the log ξ = 3.9^+0.5_−0.3. This system is moving even faster at 15,400 ± 400 km s⁻¹ = 0.05c (see Table 1 and Figure 2).

Repeating this analysis, but utilizing the higher luminosity XSTAR grid, we find that the two components are again required. In fact, the values of the column density, ionization, and velocity shifts are nearly identical and well within 1σ of the previous model.

To derive one estimate to the radius where these winds are launched, we can estimate the radius at which the observed velocity equals the escape velocity. This constrains the radius to be at r ≃ 2.9 × 10⁹ cm (970 R_Schw). Using ξ = L/(nr²) and N = nrf, where f is the one-dimensional filling factor, we can then derive the filling factor and density of the region. Assuming the ionizing luminosity is 3.5 × 10³⁷ erg s⁻¹, the resulting filling factor is f ≃ 0.0008, and the density is n ≃ 2 × 10¹⁵ cm⁻³. However, if the luminosity is higher (L_ion = 3.5 × 10³⁸ erg s⁻¹), the filling factor decreases to f ≃ 8 × 10⁻⁵, and the density increases to n ≃ 2 × 10¹⁶ cm⁻³.

These density estimates are quite high when compared to other X-ray binaries (e.g., Ueda et al. 2004; Miller et al. 2006b, 2008). However, we can invert the previous argument and instead derive the filling factor and radius from an assumed density, i.e., n = 10¹² cm⁻³. We find a larger filling factor, (f ≃ 0.04), and radius, (r ≃ 1.3 × 10¹¹ cm, 43,300 R_Schw), if we require a luminosity of 3.5 × 10³⁷ erg s⁻¹. A larger luminosity, i.e., L_ion = 3.5 × 10³⁸ erg s⁻¹, reduces the filling factor, (f ≃ 0.01), but increases the radius (r ≃ 3 × 10¹¹ cm, 100,000 R_Schw). At these radii the escape velocity is much lower than the observed velocity.

Finally, we can estimate the mass outflow rate ($\dot{m}_{{\rm wind}}$) using a modified spherical outflow, which can be approximated as $\dot{m}_{{\rm wind}} \simeq 1.23 m_{p} L_{{\rm ion}} f v \Omega / \xi$. Here, we assume a covering factor Ω/4π = 0.5, and an outflowing velocity of v = 9600 km s⁻¹. A luminosity of L_ion = 3.5 × 10³⁸ erg s⁻¹ and filling factor of f = 8 × 10⁻⁵ gives a lower limit of $\dot{m}_{{\rm wind}} \simeq 3.5 \times 10^{16} (10^{4}/\xi)$ g s⁻¹. However, a much larger outflow rate of $\dot{m}_{{\rm wind}} \simeq 1.7 \times 10^{18} (10^{4}/\xi)$ g s⁻¹ is found, if we assume L_ion = 3.5 × 10³⁷ erg s⁻¹ and filling factor of f = 0.04.

For comparison, $L = \eta \dot{m}_{{\rm acc}} c^{2}$, where η is an efficiency factor typically taken to be 10%. For IGR J17091−3624, $\dot{m}_{{\rm acc}} = 3.8\times 10^{17}$ g s⁻¹. Using log ξ = 3.3 from the disk blackbody and power-law model, we find that the observed portion of the outflow is likely to carry away 0.4–20 times the amount of accreted gas. Unless a geometrical consideration serves to bias our estimates, a high fraction of the available gas may not accrete onto the black hole. This trend is not only seen in BHBs but in Seyferts as well. Blustin et al. (2005) note that more than half of their observed Seyferts show $\dot{M}_{{\rm out}} / \dot{M}_{{\rm acc}} > 0.3$.

The EVLA radio observations at 8.4 GHz were made nearly contemporaneously with their X-ray counterparts. Both radio observations were nearly 2 hr in duration. Neither observation detected a source at the location of IGR J17091−3624. The rms noise level for each observation was 0.02 mJy and 0.07 mJy for the two epochs, respectively. The second observation had extended coverage to 4.8 GHz that also had a non-detection. The rms for this frequency was 0.13 mJy. In contrast, IGR J17091−3624 was detected at the 1–2 mJy level during the low/hard state (Rodriguez et al. 2011). This supports prior findings that the radio jet is absent during the periods when winds are seen in BHBs (Miller et al. 2006b, 2008; Neilsen & Lee 2010).