SUZAKU OBSERVES WEAK FLARES FROM IGR J17391−3021 REPRESENTING A COMMON LOW-ACTIVITY STATE IN THIS SUPERGIANT FAST X-RAY TRANSIENT

Supergiant fast X-ray transients (SFXTs; Smith et al. 2004; Sguera et al. 2005, 2006; Negueruela et al. 2006b) form a new subclass within the group of high-mass X-ray binaries (HMXBs). Previously, HMXBs were cataloged (e.g., Liu et al. 2000, and references therein) into one of either two groups: the transient sources with B-emission line stellar companions (BeXBs); or the persistent, but variable, HMXBs with supergiant O- or B-type companions (SGXBs). The SFXTs share properties from both groups: like BeXBs, they are characterized by outbursts of short (hours) to long (days) duration whereby the peak intensity is several orders of magnitude greater than during quiescence; and like SGXBs, they are systems in which a compact object (usually a neutron star (NS)) is paired with a supergiant OB star. Their study is therefore important because SFXTs could represent an evolutionary link between BeXBs and SGXBs (e.g., Negueruela et al. 2008; Chaty 2010; Bodaghee et al. 2010; Liu et al. 2010).

XTE J1739−302 (Smith et al. 1998) was the first hard X-ray transient to display behavior that was identified as characteristic of the class of SFXTs (Negueruela et al. 2006b). It was listed as AX J1739.1−3020 in the ASCA catalog of X-ray sources in the Galactic center region (Sakano et al. 2002). Five years after its discovery, INTEGRAL detected a transient source named IGR J17391−3021 which, because of the positional coincidence and similar emission behavior, was shown to be the hard X-ray (18–50 keV) counterpart to the RXTE source (Sunyaev et al. 2003; Smith et al. 2003). A refined X-ray position was obtained with Chandra (Smith et al. 2006), which led to the identification of USNO-B1.0 0596–0585865 (= 2MASS J17391155−3020380), a supergiant star of spectral type O8 Iab located ∼2.7 kpc away, as the optical/IR counterpart (Negueruela et al. 2006a; Chaty et al. 2008; Rahoui et al. 2008).

This source has been detected numerous times in outburst and in quiescence thanks to monitoring campaigns with INTEGRAL (Lutovinov et al. 2005; Sguera et al. 2005; Türler et al. 2007; Blay et al. 2008; Chenevez et al. 2008; Romano et al. 2009b; Ducci et al. 2010) and with Swift (Romano et al. 2008a, 2008b, 2009a, 2010; Sidoli et al. 2008, 2009a, 2009b). Most of the outbursts are short (∼0.5 hr) and weak (∼10 cps in 20–40 keV as seen with INTEGRAL), while a few flares last longer than an hour (sometimes days as seen in the 0.5–10 keV band with Swift) and are rather intense (∼60 cps or ∼400 mCrab with INTEGRAL; Blay et al. 2008). The photoelectric absorption is variable from outburst to outburst (Smith et al. 2006), reaching up to (13⁺⁴₋₃) × 10²² cm⁻² during the rising portion of certain flaring episodes (Sidoli et al. 2009a, 2009b). The dynamic range between quiescence and peak outburst is ∼10⁴ (in't Zand 2005). Based on multiple observations of IGR J17391−3021 gathered over the course of a year with Swift, Romano et al. (2009a) showed that bright outbursts account for only 3%–5% of the entire observation set, while the source spends 39% ± 5% of the time in an inactive state near quiescence. Quiescence, i.e., a flux in the 0.5–10 keV range below the Swift detection limit of 1.6 × 10⁻¹² erg cm⁻² s⁻¹ for a ∼1 ks exposure, is a very rare state (Romano et al. 2009a).

What triggers the outbursts is still an open question. Several mechanisms have been proposed to explain the peculiar X-ray emission behavior of SFXTs. The outbursts could be due to one or a combination of the following effects: the intermittent accretion of clumpy wind material by a compact object in a larger orbital radius than in classical SGXBs; passage of the compact object in an eccentric orbit through anisotropic winds emanating from some supergiant stars; and the dynamics of wind accretion onto highly magnetized NSs including gating effects, photo-ionization, transient disks, and Rayleigh–Taylor instabilities (e.g., in't Zand 2005; Walter & Zurita Heras 2007; Sidoli et al. 2007; Negueruela et al. 2008; Bozzo et al. 2008; Ducci et al. 2009, and references therein).

While the identity of the stellar companion is known, the nature of the compact object has yet to be conclusively determined. The spectral properties of IGR J17391−3021, i.e., its hard power-law shape and possible spectral cutoff around 13 keV (Sidoli et al. 2009a), suggest an NS primary. However, coherent pulsations or cyclotron absorption lines that would firmly establish the presence of an NS have not been detected thus far (Smith et al. 1998; Sguera et al. 2005; Blay et al. 2008). Recently, the orbital period of the system was shown to be 51.47(2) day with hard X-ray outbursts (≳20 keV) and quiescent epochs coexisting at periastron suggesting clumpy, anisotropic winds (Drave et al. 2010).

In 2008 February, Suzaku observed IGR J17391−3021 for 37 ks of effective exposure time. The data from this observation are described in Section 2. Results from timing and spectral analyses are reported in Sections 3 and 4, respectively. We discuss our findings and present our conclusions in Sections 5 and 6.

The Suzaku space telescope features two co-aligned instruments working in tandem to observe the hard X-ray sky (0.2–600 keV): the X-ray Imaging Spectrometer (XIS; Koyama et al. 2007) and the Hard X-ray Detector (HXD, Takahashi et al. 2007). Currently, XIS consists of two front-illuminated CCD cameras (XIS0 and XIS3), and one that is back-illuminated (XIS1).

Observation ID 402066010 (PI: J.A. Tomsick) targeted IGR J17391−3021 on 2008 February 22 and 23 (MJD 54518.495–54519.396) for 72.271 ks. After removing bad events during data reduction (see below), and observation gaps due to the occultation of the satellite by the Earth, the net exposure time that remained was 36.546 ks. The XIS-nominal pointing mode was used with the quarter-window option in order to gain a higher time resolution.

Data reduction employed v.6.9 of the HEAsoft software package following the procedures described in the Suzaku ABC Guide v.2. The task xispi converted the unfiltered XIS event files to pulse-invariant channels, which were then input into xisrepro to return cleaned event files. We extracted the source spectrum from each XIS detector using a circular region of 120'' radius centered on the position given by Chandra (see Figure 1). The background was extracted from a pair of 240'' wide squares positioned on a source-free region of the detector. Based on these source and background extraction regions, we generated response matrices with xisrmfgen and xissimarfgen.

Zoom In Zoom Out Reset image size

To improve photon statistics, we merged the source spectral files for the front-illuminated CCDs (XIS0 and XIS3), and we did the same for the background. Each bin in the XIS spectra contained a minimum of 150 counts. Source and background light curves were generated in 0.5–10 keV (XIS) with a time resolution of 2 s. Timing data from the three XIS detectors were combined into single light curves for the source and for the background.

A tuned (v.2.0) background file and the common GTI were used during spectral extraction of HXD–PIN. Dead-time corrections were performed with hxddtcor. The cosmic X-ray background was modeled using the flat-field response file from 2008 January 29 and its contribution was accounted for in the PIN background spectrum. Since IGR J17391−3021 was very weak during this observation, PIN recorded few high-energy photons. Furthermore, IGR J17391−3021 is located close to the Galactic center so diffuse emission from the Galactic Ridge can influence the high-energy spectrum of IGR J17391−3021 (Yuasa et al. 2008). Applying the best-fitting parameters for the ridge emission from Valinia & Marshall (1998) to the PIN spectrum, we find that the ridge contributes over 50% of the PIN photon counts. Hence, the statistics being insufficient for timing and spectral analyses, observations with PIN are not discussed any further.

Figure 2 presents the light curve of IGR J17391−3021 in the 0.5–10 keV band from XIS. In this energy range, the average source count rate (CR; combining XIS 0, 1, and 3) for the entire observation is 0.251 ± 0.005 cps. The best-fitting model to the (absorbed) spectrum of the time-averaged Suzaku observation (see Section 4) yields a coefficient of 2.0 × 10⁻¹¹ erg cm⁻² ct⁻¹ which we can use to convert instrumental CRs to units of 10⁻¹¹ erg cm⁻² s⁻¹. Thus, the average flux is (5.0 ± 0.1) × 10⁻¹² erg cm⁻² s⁻¹.

Zoom In Zoom Out Reset image size

Based on Figure 2, we can identify two distinct emission states from IGR J17391−3021 in this observation. During the initial 33 ks, IGR J17391−3021 is in a quiescent state in which the average source CR is 0.100 ± 0.009 cps ((2.0 ± 0.2) × 10⁻¹² erg cm⁻² s⁻¹). As with previous observations of IGR J17391−3021 when it was at or near quiescence (Sidoli et al. 2008; Romano et al. 2009a, 2010; Bozzo et al. 2010), our Suzaku observation shows that even in this state, IGR J17391−3021 varies on timescales of a few hundred seconds.

Around 33 ks after the start of the observation (corresponding to MJD 54518.92), the source enters a low state in which the mean CR of 0.358 ± 0.008 cps ((7.2 ± 0.2) × 10⁻¹² erg cm⁻² s⁻¹) is a factor 3–4 that of the flux in the quiescent state that preceded it. The observations prior to and after MJD 54518.92 are henceforth referred to in this analysis as "quiescent" and "low" states, respectively (see Table 1). In addition, three flares lasting ∼2 ks each are revealed in the light curve at t ∼ 40, 54, and 70 ks. The peak flux of 1.3 ± 0.2 cps ((2.6 ± 0.4) × 10⁻¹¹ erg cm⁻² s⁻¹) is an order of magnitude larger than the flux during the quiescent state, but the maxima are still much lower than the typical high-state outbursts expected and seen in such objects. We note that the peaks are nearly equidistant in time being separated by intervals of approximately 15 ks.

The light curve was divided into two pieces corresponding to a soft (S) band between 0.5 and 2 keV, and a hard band (H) with energies between 2 and 10 keV. The hardness ratio can then be defined as (H − S)/(H + S) where a negative value indicates a soft spectrum and a positive value denotes a hard spectrum. Figure 2 displays no obvious correlation between the hardness ratio and the source intensity. Binning the light curve into longer time segments, and plotting the hardness ratio versus the source intensity, shows only a marginal correlation (but within the errors) between the two parameters at large intensities.

As part of a monitoring campaign focusing on SFXTs, Swift targeted IGR J17391−3021 at various times throughout 2008 (Romano et al. 2009a). In particular, there is a ∼900 s long X-Ray Telescope (XRT) pointing (ObsID: 00030987012) several hours prior to our Suzaku observation. The 0.5–10 keV Swift-XRT detections and 3σ upper limits are shown along with the average CR recorded by XIS in Figure 3. The instrumental conversion coefficient for XRT is 1.3 × 10⁻¹⁰ erg cm⁻² ct⁻¹ (Romano et al. 2009a).

Zoom In Zoom Out Reset image size

The time-averaged spectrum of IGR J17391−3021 gathered in the 0.5–10 keV range by XIS is presented in Figure 4. We fit an absorbed power law, employing the abundances of Wilms et al. (2000) and the photo-ionization cross sections of Balucinska-Church & McCammon (1992), to the XIS spectra which yields Γ= 1.4 ± 0.1 and N_H= (3.6 ± 0.4)× 10²² cm⁻² (errors are provided at the 90% confidence level). These are compatible with the values derived by Sidoli et al. (2008) for out-of-outburst epochs in a similar energy range with Swift-XRT, with an average luminosity that is slightly higher at 4.8 × 10³³ erg s⁻¹ compared with 3.9 × 10³³ erg s⁻¹ for Sidoli et al. (2008). Both studies assume a source distance of 2.7 kpc (Rahoui et al. 2008). The model fits the data well (χ²_ν = 0.82 for 66 degrees of freedom (dof)).

Zoom In Zoom Out Reset image size

Residuals remain at low energies (≲1 keV) which could hint at the presence of soft excess emission. We know that this source exhibits a soft excess since this emission was detected with XMM-Newton (Bozzo et al. 2010). To account for this feature, we added a blackbody whose temperature was fixed to the best-fitting value from XMM-Newton (kT = 1.2 keV). This reduces the χ²_ν to 0.55 for 65 dof (F-test probability of 5×10⁻⁵), and N_H =(1.71^+0.91_−0.26) × 10²² cm⁻², but the photon index is poorly constrained (Γ = 0.2^+0.3_−1.4).

We split the spectra into two segments corresponding to the quiescent and low states described in Section 3 (i.e., before and after MJD 54518.92). Table 2, which lists the spectral parameters of IGR J17391−3021 for each epoch, shows that there is a clear distinction in column density, photon index, and luminosity. The N_H is at least twice as high in the low state ((4.1^+0.5_−0.4) × 10²² cm⁻²) as it is in the quiescent phase ((1.0^+0.6_−0.5) × 10²² cm⁻²). The source spectrum is softer (Γ = 1.5 ± 0.1) during the low state compared to the quiescent state (Γ = 1.0 ± 0.3). The unabsorbed 0.5–10 keV luminosity rises from 1.3 when the source is quiescent to 7.4×10³³ erg s⁻¹ when it is in the low state.

Figure 5 illustrates the incompatibility (at >99% confidence) in the absorbing column from quiescence to the low state. It is clear from this figure that the column density is the main parameter that evolves from quiescent to low states. Indeed, for a constant Γ ∼ 1.3, the N_H ranges from ∼1 × 10²² to ∼4 × 10²² cm⁻² between quiescent and low states, respectively. This is different from what was seen by Bozzo et al. (2010) who found that the column density remained steady (within the statistical errors) during the low state while the Γ varied by around 50%.

Zoom In Zoom Out Reset image size

To test whether we overestimated the column density during the low state by not considering a soft excess, we included a blackbody component in the spectral fits to quiescent and low states, and we fixed its temperature to the optimal value from Bozzo et al. (2010; 1.2 keV). The absorbing columns remain inconsistent (at the 90% confidence level) between quiescence (N_H ⩽1.35 × 10²² cm⁻²) and the low state (N_H =(1.99^+0.79_−0.27) × 10²² cm⁻²). When the blackbody is fixed to the lower temperature of 0.13 keV provided by Bozzo et al. (2010), the column densities are compatible within the large error bars between the quiescent (N_H =(2.5^+5.3_−2.0) × 10²² cm⁻²) and low states (N_H =(5.8^+1.4_−1.1) × 10²² cm⁻²), but in this case, the resulting blackbody radius is ≳100 km which is much larger than the expected size of the compact accretor.

For ease of comparison with previous studies, we fit the source spectrum with models used for this source by other authors (Sidoli et al. 2008; Romano et al. 2009a; Bozzo et al. 2010). An absorbed blackbody provides a better fit than the absorbed power law (χ²_ν = 0.71 for 66 dof). The parameters from this model agree with those describing the blackbody-modeled spectrum of IGR J17391−3021 when it was in the low state as seen with Swift (Sidoli et al. 2008; Romano et al. 2009a). The results of this model fit to the various states (i.e., time-averaged observation, quiescence, and low state) are listed in Table 2. Once again, statistically significant differences emerge for the column density (as well as for the radius of the blackbody emission region) for spectra collected in quiescence (N_H ⩽0.22 × 10²² cm⁻²) and during the low state (N_H =(1.4^+0.3_−0.2) × 10²² cm⁻²).

In the low state, a power law with a partial-covering absorber and an additional component to account for Galactic absorption (fixed to 1.3 × 10²² cm⁻² from Kalberla et al. (2005)) yields N_H =(5.1^+1.0_−1.5) × 10²² cm⁻² with a covering fraction of 9%1 ± 4% and Γ = 1.76^+0.14_−0.25 (χ²_ν/dof =0.56/53). In other words, during the low state that we observed, a large percentage of the X-ray emission is shielded by an absorber whose N_H is higher than the expected interstellar value. We could not constrain the value of the partially absorbing column in the quiescent state.

A cutoff power-law model gives a good fit to the time-averaged spectrum (χ²_ν/dof =0.57/65). The power-law index is Γ = −0.2^+0.6_−0.8 with a cutoff at E_cut = 3⁺²₋₁ keV and the column density is N_H =(2.1 ± 0.3) × 10²² cm⁻². These are consistent with the parameters listed in Bozzo et al. (2010). The index and cutoff energy are compatible between the quiescent (Γ = −0.1^+1.3_−1.3, E_cut>2 keV) and low (Γ = −0.3^+0.6_−1.0, E_cut = 3⁺¹₋₁ keV) states. In this case, the upper limit to the column density during quiescence (<1.5 × 10²² cm⁻²) lies within the 90% confidence interval of N_H in the low state ((2.2^+0.7_−1.2) × 10²² cm⁻²).

The spectra were also fit with the Mewe–Kaastra–Liedahl (MEKAL) model employing a dual absorber: one that affects the system's soft X-ray emission from hot diffuse gas in the supergiant star's wind (N_Hsg); and another that impacts the high-energy power-law emission from the compact accretor (N_Hx). The gas temperature was set to 0.15 keV (Bozzo et al. 2010). In the quiescent state, we obtain N_Hsg ⩽ 2.2 × 10²² cm⁻² and N_Hx ⩽ 2.4 × 10²² cm⁻² for Γ fixed to the optimal value of 1.0 from the absorbed power-law fit to the quiescent spectrum (χ²_ν/dof =0.54/52). For the spectrum from the low state, N_Hsg = (5.8^+0.7_−2.0) × 10²² cm⁻² and N_Hx ⩽ 1.4 × 10²² cm⁻² for Γ = 1.8 ± 0.2 (χ²_ν/dof =0.55/65). This suggests that not only is there material around the binary system absorbing X-rays, there is also additional matter located near the compact object.

We sought to determine how the spectral parameters varied with the source intensity. So we created spectra from time intervals in which the CRs were <0.1 cps ("low CR" or LCR), between 0.1 and 0.3 cps ("medium CR" or MCR), and >0.3 cps ("high CR" or HCR). Then, each rate-resolved spectrum was fit with an absorbed power law, an absorbed cutoff power law with E_cut fixed to our best-fitting value (3 keV), and an absorbed blackbody model. Table 3 lists the results of the spectral fits. Regardless of the spectral model, the N_H is significantly smaller during the LCR epochs. The MCR and HCR epochs have spectra that are consistent, and both are statistically incompatible with the spectrum from LCR epochs.

5.1. Weak Flares and Low Activity

5.2. Spectral Evolution

5.3. The Ubiquity of Low Activity

Our Suzaku observation of IGR J17391−3021, when placed in the context of the long-term monitoring campaign of this source by Swift-XRT (Romano et al. 2009a), shows that the average flux that we recorded is consistent with many of the XRT detections at a low flux level (F ∼ (2–10) × 10⁻¹² erg cm⁻² s⁻¹) and is on par with a few of the 3σ upper limits to the CRs of non-detections as measured by XRT (Figure 3).

Weak flares that are just above quiescence, and whose peak fluxes are at the level of the faint detections in the Swift monitoring data, compel us to propose four states in the emission behavior of IGR J17391−3021: the extremely low but variable emission of quiescence (average F_abs ≲ (1–2) × 10⁻¹² erg cm⁻² s⁻¹ in 0.5–10 keV); a low-activity state that potentially includes weak flares such as those seen here and in Bozzo et al. (2010; F ∼ (2–10) × 10⁻¹² erg cm⁻² s⁻¹); a medium state with flares of larger amplitude (F ∼ (2–10) × 10⁻¹¹ erg cm⁻² s⁻¹); and finally, a high state in which the outbursts represent the maximum source luminosity (F ≳ 10⁻⁹ erg cm⁻² s⁻¹). The boundaries are not strict since the peak fluxes in the low state extend up to 3 × 10⁻¹¹ erg cm⁻² s⁻¹, but they provide a useful classification for the various emission states in this source.

If we assume that the Swift detections when the CRs were sufficiently low (i.e., F ∼ (2–10) × 10⁻¹² erg cm⁻² s⁻¹) were due to low states that may or may not include weak flares, then we can define a low-state duty cycle (DC) as follows:

$$\begin{equation} \mathrm{DC} = \frac{ \Sigma T_{\mathrm{low}} }{ \Sigma T_{\mathrm{tot}} } = 60\pm 5\%, \end{equation} \tag{ 1 }$$

Here, ΣT_low is the sum of exposure times of all observations in which IGR J17391−3021 is detected at an average flux within the range of F ∼ (2–10) × 10⁻¹² erg cm⁻² s⁻¹, including the Suzaku and XMM-Newton observations, while ΣT_tot denotes the total exposure from all observations. As expected, the DC is not far from the value of 61% ± 5% provided in Romano et al. (2009a, 2010).

Low-activity states in which the flux is above the "typical" quiescent level, but well below the medium and bright flaring episodes, are therefore quite common in this source. In the field of IGR J17391−3021, the limiting flux for XRT is 1.6 × 10⁻¹² and 1.5 × 10⁻¹³ erg cm⁻² s⁻¹ for exposure times of 1 and 30 ks, respectively (Romano et al. 2009a). Therefore, most Swift upper limits would be detections if the exposures had been equivalent to those of the Suzaku and XMM-Newton observations suggesting that the DC of the low state is much higher.

According to the source ephemeris (Drave et al. 2010), the midpoint of our observation coincides with an orbital phase of 0.37^+0.02_−0.01 whereas the XMM-Newton observation took place at phase 0.8. This is another indication that the low state is common since it is not confined to a specific part of the orbit.

Despite the large DC, weak flares lasting a few hundred to a few thousand seconds can be missed by short monitoring observations. Because their peak flux is so low, detecting a low state requires prolonged observations where the long integration time guarantees that we will achieve the sensitivity necessary to detect such low levels of activity. Given that IGR J17391−3021 is a prototypical member of its class, it would not be surprising to find weak flares from other SFXTs as telescopes continue to devote observing time to such objects. In fact, recent Suzaku and XMM-Newton observations of the SFXT IGR J08408−4503 included similar low states that were punctuated by weak flares (Bozzo et al. 2010; Sidoli et al. 2010).

The main results from our Suzaku observation of the SFXT IGR J17391−3021 can be summarized as follows.

• 1.  
  For most of the observation, IGR J17391−3021 is in what we believe represents the quiescent state. The unabsorbed luminosity in the 0.5–10 keV range is 1.3 × 10³³ erg s⁻¹and as low as 8.7 × 10³² erg s⁻¹ in the flux-limited spectrum. The spectrum from this epoch is well described by an absorbed, hard power law with N_H =(1.0^+0.6_−0.5) × 10²² cm⁻² and Γ = 1.0 ± 0.3, or by an absorbed thermal blackbody with a temperature of 1.57^+0.16_−0.15 keV.
• 2.  
  Around 33 ks into the observation (MJD 54518.92) the source became slightly more active ("low state") culminating in a series of weak flares each lasting ∼3 ks whose peak luminosity is only a factor of ∼5 greater than the quiescent emission that preceded it. During this low state, the N_H reached (4.1^+0.5_−0.4) × 10²² cm⁻² which is a factor of ∼2–4 times that of the expected interstellar value, and which is ∼2–9 times the N_H measured during quiescence. The accretion of small clumps of stellar wind material can be responsible for the weak flares, while the increase in N_H can be explained by the passage of the clump in the line of sight to the X-ray source prior to accretion.
• 3.  
  The average luminosity of the entire observation is 4.8 × 10³³ erg s⁻¹. When placed in the full light curve from the Swift-XRT monitoring campaign by Romano et al. (2009a), we find that this luminosity is on par with the low-intensity XRT detections (as well as a few of the 3σ upper limits from non-detections). Furthermore, we suspect that many of the faintest detections from the XRT contain similar low states (potentially including weak flares) that are just above the quiescent flux. We estimate that the DC of the low state is 60% ± 5%, which makes it the most common state for this source.

As a prototypical member of the class of SFXTs, IGR J17391−3021 holds valuable clues to the accretion processes of these intriguing transients. Continued monitoring of this source by sensitive instruments such as Suzaku should fuel the ongoing debate concerning the mechanisms that are responsible for their peculiar emission behavior.

The authors thank the anonymous referee whose revision of the draft led to significant improvements in the quality of the manuscript. A.B. and J.T. acknowledge Suzaku Guest Observer Grant NNX08AB88G and Chandra Grant G089055X. This research has made use of data obtained from the High Energy Astrophysics Science Archive Research Center (HEASARC) provided by NASA's Goddard Space Flight Center; the SIMBAD database operated at CDS, Strasbourg, France; NASA's Astrophysics Data System Bibliographic Services.

Facilities: Suzaku - Suzaku (ASTRO-EII), Swift - Swift Gamma-Ray Burst Mission.