© ESO, 2010

1. Introduction

Low mass X-ray binaries (LMXBs) consist of a low-mass donor star (<1 M_⊙) and a compact object that accretes matter through an accretion disk. Most LMXBs are transients, i.e. they undergo week-to-month long outbursts during which the accretion takes place at high rates, giving rise to typical (“persistent”) X-ray luminosity of  ~10³⁶⁻³⁸ erg/s. This is  ≳ 100 times higher than the X-ray luminosity displayed in quiescence (see e.g. Bildsten et al. 1998; Campana et al. 1998; Liu et al. 2007, for reviews). Depending on the peak X-ray luminosity reached during the outburst (L_peak), transient LMXBs are historically classified as “bright transient” (L_peak ≃ 10³⁷⁻³⁸ erg/s), “faint transient” (L_peak ≃ 10³⁶⁻³⁷ erg/s), or “very faint transients” (L_peak ≤ 10³⁶ erg/s, see e.g., Wijnands et al. 2006; Campana 2009). The outbursts of transient LMXBs are usually interpreted in terms of disk instability models (see e.g., Frank et al. 2002). In a number of these sources, an accreting neutron star (NS) as a compact object has been unambiguously identified by the detection of type-I X-ray bursts and/or coherent pulsations. The former are bright (~10³⁸ erg/s) and short (~10−60 s) flares that result from thermonuclear explosions occurring in the material accreted at the NS surface (see e.g. Lewin et al. 1993; Strohmayer & Bildsten 2006, for reviews).

A peculiar subclass of transient LMXBs was discovered about 10 years ago, mainly thanks to the long-term monitoring of the Galactic center carried out with the BeppoSAX wide field cameras (WFC, Heise et al. 1999; Cocchi et al. 2001; Cornelisse et al. 2002a, 2004). At odds with the previously known transient LMXBs, these systems appeared to be characterized by no persistent emission before and after the occurrence of a type-I X-ray burst (Cornelisse et al. 2002a). The upper-limits inferred for their persistent emission were significantly lower than the persistent emission of the bright transients (10³⁵⁻³⁶ erg/s) but not very tight due to the modest sensitivity of the BeppoSAX -WFCs. For this reason, they were collectively termed “burst-only sources”. Only in a few cases, a few ks-long Chandra observations carried out after the discovery of a type-I X-ray burst from some of these sources, were able to reveal the presence of a weak persistent emission at a position consistent with that of the X-ray burst. Typical (quiescent) luminosities derived from these detections were  ~10³²−10³³ erg/s (Cornelisse et al. 2002b).

Owing to their peculiar behavior and the discovery of several new members of this class (see e.g., Chelovekov & Grebenev 2007; Degenaar & Wijnands 2009), the burst-only sources have attracted increasing interest in the past few years (see also Del Santo et al. 2010). Systems undergoing type-I X-ray bursts when subject to very low accretion rates allow us to test models of thermonuclear burning in a regime that is still poorly explored (see e.g., Cooper & Narayan 2007; Peng et al. 2007). Moreover, the low persistent luminosity in quiescence and the spectral properties during the outbursts led to the suggestion that the burst-only sources could be linked to very faint transients and accreting millisecond X-ray pulsars (AMSPs, Wijnands et al. 2006; Campana 2009; Trap et al. 2009).

In this paper, we report on the recent discovery of an intense X-ray activity from the burst-only source Swift J1749.4−2807. We studied the timing and spectral properties of the source during this event by exploiting all the available Swift, INTEGRAL, and RXTE data, and report on the detection of millisecond pulsations in the X-ray emission of this source. The results presented here provide a strong support in favor of the association between burst-only sources and AMSPs. In Sect. 1.1, we describe previous observations of Swift J1749.4−2807, and in Sect. 2 our data analysis and results. Our discussion and conclusions are presented in Sect. 3.

1.1. SWIFT J1749.4−2807

Swift J1749.4−2807 was discovered in 2006 during a bright type-I X-ray burst that was initially recorded as a potential gamma ray burst (GRB060602B, Wijnands 2009). Swift /XRT started to follow-up the evolution of the source from 83 s after the BAT trigger, and monitored the source outburst for the following 8 d. The source X-ray flux decayed in an approximately power-law fashion (with index  ~− 1), fading below the detection threshold of Swift /XRT in less than 10⁶ s. The BAT spectrum extracted at the peak of the type-I X-ray burst could be described by a black-body (BB) model ($\mathit{kT}\mathrm{=}\mathrm{2.}{\mathrm{9}}_{-0.3}^{\mathrm{+}\mathrm{0.4}}$ keV) and provided an upper limit to the source distance of 6.7 ± 1.3 kpc (by assuming that the peak X-ray luminosity of the burst corresponded to the Eddington value, uncertainties at 90% c.l.; Wijnands 2009). During the outburst, the XRT spectrum could be described well by an absorbed power-law model. The measured column density was consistent with being constant throughout the event at a value N_H = 4 × 10²² cm^-2, while the photon index was observed to decrease from $\mathrm{2.}{\mathrm{7}}_{-1.1}^{\mathrm{+}\mathrm{1.5}}$ to 0.5 ± 1.3 (Campana 2009). The estimated X-ray luminosity was  ~10³⁶ erg/s at the peak ot the outburst and  ~10³² erg/s in the latest Swift observation available.

Wijnands (2009) also reported on three serendipitous detections of Swift J1749.4−2807 in archival XMM-Newton observations. The first observation was carried out on 2000 September 23, and the second two were performed on 2006 September 22 and 26. The count-rate of the source in the three XMM-Newton observations was too low to extract any meaningful spectral information. By assuming an absorbed power-law model with Γ = 2 and N_H = 3 × 10²² cm^-2, Wijnands (2009) estimated a 2−10 keV unabsorbed flux of  ~(1−2) × 10^-13 erg/cm²/s. This corresponds to a luminosity of (3−6) × 10³² erg/s (assuming a distance to the source of 7 kpc) and is compatible, to within the errors, with that measured about 6 days after the outburst discovered with Swift.

Swift J1749.4−2807 was detected again in a high X-ray luminosity state on 2010 April 10, during the INTEGRAL Galactic bulge monitoring program (Pavan et al. 2010; Kuulkers et al. 2007). Swift and RXTE target of opportunity observations (ToO) were immediately requested, and these monitored the source outburst for about two weeks. RXTE observations detected coherent pulsations in the X-ray emission of the source at 518 Hz and its second harmonic (Altamirano et al. 2010a; Bozzo et al. 2010). From these data, a preliminary orbital solution was first obtained by Belloni et al. (2010), and then refined by Strohmayer et al. (2010) by using a pulse phase-coherent technique. The latter authors derived an orbital period of 31   740.345 ± 0.04 s (8.8168 h), a projected semi-major axis of asin(i) = 1899.53 ± 0.01 lt-ms, and a barycentric frequency for the second harmonic (i.e., twice the spin frequency) of 1035.840025 Hz  ± 0.4   μHz. The time of the ascending node was 2   455   301.1522672 ± 0.0000014 JD (TDB). This solution implied a mass function of 0.05463 ± 0.00018   M_⊙ and a minimum mass for the companion of 0.475 M_⊙ (assuming a NS of 1.4 M_⊙). The discovery of an X-ray eclipse in the RXTE light curve was reported by Markwardt et al. (2010a); this is the first eclipse observed from an accreting millisecond X-ray pulsar. The most accurate source position to date was provided by Yang et al. (2010) at and 9, with a 90% confinement radius of 1.9, based on Swift /XRT observations.

2. Data analysis and results

2.1. INTEGRAL data

INTEGRAL data were analyzed using the OSA software (v.9) released by the ISDC (Courvoisier et al. 2003). We considered data from both IBIS/ISGRI (Lebrun et al. 2003) and JEM-X2 (Lund et al. 2003) in the 20−40 keV and 3–23 keV energy ranges, respectively. The average fluxes and spectra for JEM-X2 were extracted from the mosaic images, as recommended in the case of weak sources¹. The ISGRI spectrum was extracted using standard procedures. The detail of the INTEGRAL observations is given in Table 1. The source was not detected during the observations carried out on 2010 April 7 and 21−22. Swift J1749.4−2807 was clearly detected (8σ) during the two observations carried out from 2010 April 10 to 13. The simultaneous ISGRI+JEM-X2 spectra could be reasonably well described by a cutoff power-law model of photon index 1.3−1.6 (we fixed the cutoff energy at 20 keV and the absorption column density at 3 × 10²² cm^-2, see below); the estimated flux was 3 × 10^-10 erg/s/cm² (3-20 keV). For these observations, we also extracted an event list from the entire JEM-X2 detector in the 3−20 keV band to search for type-I X-ray bursts. Only one burst was detected: the increase in the source flux at the time of the burst is also confirmed by the higher significance of the source detection in the JEM-X2 image extracted during the duration of the burst (see also Chenevez et al. 2010). Our analysis of the burst parameters is described below. A search for type-I X-ray bursts was also performed in the ISGRI data (we used a list of events selected in the 18-40 keV energy band and with a pixel illumination fraction threshold  >0.75). No statistically significant bursts were detected in these data.

2.1.1. The type-I X-ray burst

In the INTEGRAL observation carried out on 2010 April 13, a type-I X-ray burst was detected by JEM-X2. The light curve of the burst in the 3−20 keV band and with a time resolution of 2 s is shown in Fig. 1. The rise time of the burst was  ~1 s, and the start time² was 2010 April 13 at 16:51:18 (UTC at the satellite location; these values were determined by using the JEM-X2 source event list rebinned to have a time resolution of 0.5 s). The relevant burst parameters are reported in Table 2. Owing to the relatively low signal-to-noise ratio (S/N), a time-resolved spectral analysis of the burst could not be carried out, thus no signature of a possible photospheric radius expansion (PRE; see e.g., Lewin et al. 1993) could be identified. We determined the flux at the peak of the burst by fitting the spectrum of the initial 4 s with a BB model (the N_H was fixed at 3 × 10²² cm^-2, see Table 1). The persistent spectrum extracted from a time interval close to the burst was used as a background in the fit. The best-fit BB temperature and radius at the peak were $\mathit{k}{\mathit{T}}_{\mathrm{bb}\mathit{,}\mathrm{peak}}\mathrm{=}\mathrm{2.}{\mathrm{3}}_{-0.5}^{\mathrm{+}\mathrm{0.7}}$ keV and ${\mathit{R}}_{\mathrm{bb}\mathit{,}\mathrm{peak}}\mathrm{=}\mathrm{4.}{\mathrm{8}}_{-0.3}^{\mathrm{+}\mathrm{1.2}}$ km, respectively (for a source distance of 7 kpc) with a ${\mathit{\chi }}_{\mathrm{red}}^{\mathrm{2}}\mathit{/}\mathrm{d}\mathit{.}\mathrm{o}\mathit{.}\mathrm{f}\mathit{.}\mathrm{=}\mathrm{0.8}\mathit{/}\mathrm{3}$. In Table 2, we report the decay time, τ_lc, of the burst as measured by fitting the observed light curve with an exponential function; this is to be compared with τ, the burst duration that is obtained by dividing the burst fluence by the peak flux (see e.g. Lewin et al. 1993). We note that the two values are in agreement to within the errors. Additional comments on the type-I X-ray burst are given in Sect. 3.

2.2. RXTE data

RXTE /PCA (Jahoda et al. 1996) data analysis and spectral extraction were carried out with the standard tools available in heasoft (V 6.7). Since the source was relatively faint for the instruments on-board RXTE and is located in the Galactic bulge, we paid particular attention to estimate the background as accurately as possible. For this purpose, we excluded from the spectral analysis, performed using the standard2 mode, the data from the PCU0 detector, because the propane veto layer stopped working in May 2000, leading to significantly worse instrumental noise subtraction. Moreover, we considered only data from the upper anode layer to reduce the systematic error in the instrumental noise.

The large contribution from the Galactic diffuse emission to the RXTE/PCA X-ray fluxes, poses a serious problem for data processing. We used the latest observations, performed after 2010 April 20 to estimate the background for the utilized PCU configurations (Table 1). After 2010 April 20, the source flux measured by Swift /XRT (see Table 1) decreased to below 10^-11 erg/cm²/s, indicating that the count-rate in the last two RXTE observations was dominated by the Galactic ridge emission (Valinia et al. 1998). We did not use the HEXTE data, since both units stopped rocking on 2009 December 14 and the background estimation cannot be performed satisfactorily with the available software at the time of writing (the application of the method used for PCA does not give reliable results because of the instrumental variable background induced by high energy particles).

A log of the RXTE observations and a summary of the results of the analysis of these data is provided in Table 1. Here, all the spectra were fit within xspec in the 3−23 keV energy range with an absorbed power-law model, which always provided satisfactory fits. We checked that the relatively large values of the χ² were due to residuals in the background subtraction and not to the presence of additional spectral components.

For the timing analysis, we used the RXTE event data (mode E_125us_64M_0_1s) with 64 energy channels and 122 μs time resolution. To maximize the S/N, we selected the events in all the active PCUs and layers. The barycenter correction was applied using the tool faxbary and the most precise available source position to date (see Sect. 1.1). Pulsations at the spin frequency of the NS and its second harmonic were clearly detected in all the observations by using the Z²-statistics (Buccheri et al. 1983; Markwardt et al. 2002), in agreement with the results reported by Altamirano et al. (2010a) and Bozzo et al. (2010). An orbital solution for Swift J1749.4−2807 was obtained from the timing analysis based on the frequency modulation of the signal (see Fig. 2 and Table 3). It gave results consistent with those reported by Strohmayer et al. (2010), though with larger uncertainties, since they derived a phase-coherent solution. In the following we use their value for the orbital period, projected semi-major axis, barycentric pulse frequency, and epoch of the ascending node.

To study the variations in the amplitude of the fundamental and second harmonic throughout the RXTE monitoring (Bozzo et al. 2010), we used the value of the source spin period reported by Strohmayer et al. (2010) and folded the light curves into time intervals of 100 s to produce in each interval a pulse profile, P_t(φ), in 32 phase bins. We rebinned these pulses adaptively by choosing appropriate time intervals of durations t_stop − t_start that permitted us to obtain a S/N ≳ 50 (we were careful not to combine observations separated by more than  ~1 day). The characteristic amplitude of the first two Fourier components in each rebinned pulse profile was then computed using the equation ${\mathit{A}}_{\mathit{n}}\mathrm{\left(}\mathit{t}\mathrm{\right)}\mathrm{=}\sqrt{{\left({\mathit{I}}_{\mathrm{c}}^{\mathit{n}}\mathrm{\left(}\mathit{t}\mathrm{\right)}\right)}^{\mathrm{2}}\mathrm{+}{\left({\mathit{I}}_{\mathrm{s}}^{\mathit{n}}\mathrm{\left(}\mathit{t}\mathrm{\right)}\right)}^{\mathrm{2}}}\mathit{,}$(1)where n = 1,2 is the Fourier order ${\mathit{I}}_{\mathrm{c}}^{\mathit{n}}\mathrm{\left(}\mathit{t}\mathrm{\right)}\mathrm{=}\mathrm{\int }{\mathit{P}}_{\mathrm{t}}\mathrm{\left(}\mathit{\phi }\mathrm{\right)}\cos \mathrm{\left(}\mathit{n\phi }\mathrm{\right)}\mathrm{d}\mathit{\phi }\hspace{0.17em}\mathit{,}$(2)${\mathit{I}}_{\mathrm{s}}^{\mathit{n}}\mathrm{\left(}\mathit{t}\mathrm{\right)}\mathrm{=}\mathrm{\int }{\mathit{P}}_{\mathrm{t}}\mathrm{\left(}\mathit{\phi }\mathrm{\right)}\sin \mathrm{\left(}\mathit{n\phi }\mathrm{\right)}\mathrm{d}\mathit{\phi }\hspace{0.17em}\mathit{,}$(3)and t = t_start + (t_stop − t_start)/2. This method has the advantage that the statistical uncertainties in A_n(t) can be straightforwardly computed by propagating the errors from the pulse profiles.

From Fig. 3, we note that during the observation 95085-09-01-01 (~10⁵ s after the first RXTE observation) the power of the second harmonic became higher than that of the fundamental, in agreement with the results reported by Bozzo et al. (2010). The relative power of the two components, A₁/A₂, varies with time: the second harmonic was clearly dominant during the early stage, before 2010 April 18 (i.e. t < 3 × 10⁵ s in Fig. 4), while the fundamental became more prominent in the later stages (t > 3 × 10⁵ s in Fig. 4).

To investigate the origin of the behavior of the fundamental and second harmonics with time, we studied the source pulse profile in different energy bands. Because of the relatively low count rate of the source during the last part of the RXTE monitoring (see Table 1), we used for this analysis only the data accumulated prior to 2010 April 17. The pulse profiles of Swift J1749.4−2807 in each observation were extracted in 32 phase bins and corrected for the background by using the technique described above. The results are shown in Fig. 4, where the pulses are phase connected, since the folding reference time is the same. The double peak structure of the pulse profile visible in this figure explains the higher power in the second harmonic with respect to the fundamental spin frequency, and naturally suggests that we are observing at the same time the emission from the two polar caps of the NS with roughly the same intensity. Furthermore, a visual inspection of the pulses extracted in each observation reveals a significant change in the pulse profile with time. In particular, when the second harmonic became dominant (observation 95085-01-01), the two peaks look very similar (see Fig. 4). This might be due to a variation in the accretion flow geometry in the vicinity of the polar caps of the NS, or alternatively to the occultation of at least part of one polar cap by the inner region of the accretion disk (see also Sect. 3). A detailed modelling of the pulse profiles and their changes will be presented elsewhere.

The dependence of the fractional rms (Eq. (2.10) in van der Klis 1988) on energy is shown in Fig. 5. In this case, we used the pulse profiles with 32 phase bins obtained by summing up all the RXTE observations prior to 2010 April 17 (total exposure time  ~17 ks). The fractional rms slightly increases with energy, as in the results reported for other AMSPs (see e.g., Gierlinski & Poutanen 2005; Patruno et al. 2009).

To search for X-ray eclipses (Markwardt et al. 2010a), we extracted the source light curves of each PCU unit with a time resolution of 1 s. These light curves were subtracted for instrumental background (we used the background model available for faint sources) and scaled to the effective area of the full PCA array³. The contribution from the diffuse Galactic emission was subtracted by using the data collected in the observations 95085-09-02-08 and 95085-09-02-10 (see above).

These light curves were then barycentered and folded by using the orbital solution reported by Strohmayer et al. (2010). We found that an eclipse ingress was visible during observation 95085-09-02-11, whereas the eclipse egress was clearly present in observations 95085-09-02-02 and 95085-09-02-04. No other orbital feature was detected. To determine the parameters of the eclipse, we separately fit the light curves extracted from the three observations with the rectangular step function (see e.g., Mangano et al. 2004): $\frac{\mathit{A}}{\mathit{\pi }}\mathrm{[}{\tan }^{-1}\left(\mathit{B}\left(\mathrm{0.5}\mathrm{-}\mathit{C}\mathrm{-}\mathit{\phi }\right)\right)\mathrm{+}{\tan }^{-1}{\left(\mathit{B}\left(\mathit{\phi }\mathrm{-}\mathrm{0.5}\mathrm{-}\mathit{C}\right)\right)}^{\mathrm{]}}\mathrm{+}\mathit{D,}$(4)where φ is the orbital phase, A, B, and D are constants to be determined from the fit, and C is the semi-amplitude of the eclipse in phase units. For large values of B this expression mimics the shape of the light curve of an eclipse with sharp ingress and egress, similar to the one we have in our case (the obscuration of a NS of 10 km moving in a  ~9 h wide orbit around a solar-type star is expected to occur in Δt ~ 10^-6 s). We found that C = (3.450 ± 0.005) × 10^-2, (3.41 ± 0.03) × 10^-2 for the two egresses, and C = (3.56 ± 0.13) × 10^-2 for the ingress (uncertainties are 1σ c.l.). The corresponding durations of the eclipses are 2190 ± 4 s, 2163 ± 17 s and 2260 ± 80 s, which are all compatible with each other within the errors. The duration of the egress is constrained by the first observation to be shorter than 2 s.

The RXTE light curve of Swift J1749.4−2807, folded at the orbital period, is shown in Fig. 6, where we have zoomed around the eclipse. From this figure, we notice that the ingress in the eclipse appeared to be less rapid than the egress, and a residual X-ray flux seemed to be present during the eclipse. We investigated a possible energy dependence of the eclipse ingress profile by extracting source light curves in the 3−5 keV and 6−10 keV energy band and calculating the hardness ratio of these light curves⁴. No evidence of a change in the hardness ratio was found. After taking into account the uncertainties in the removal of the background from the RXTE observations, it is not clear whether the residual X-ray flux visible during the eclipse in Fig. 6 is due to real emission from the source. The RXTE spectrum during the eclipse infers a flux in the 3−20 keV energy band of (1.3 ± 1.1) × 10^-11 erg/cm²/s (the power-law photon index was 2.0$\begin{array}{}\mathrm{+}\mathrm{1.0}\\ -1.5\end{array}$). To investigate the origin of this X-ray emission, we first searched for pulsations in the X-ray emission of Swift J1749.4−2807 during the eclipse. We folded the data from the eclipse in observation ID. 95085-09-02-02, which ended at MJD 55 302.96097 TBD, (effective exposure time 530 s) at the NS spin frequency in 16 phase bins. This profile did not show any statistically significant modulation. To verify this apparent lack of pulsations, we also estimated the power of the pulsed emission during the eclipse. We used as a template the pulse profile in the observation 95085-02-02 excluding the eclipse (energy band 3−20 keV, fractional rms (7.3 ± 0.6)%), and computed its cross-correlation coefficient⁵ with respect to the pulse during the eclipse (exposure 530 s) and during the 530 s following the eclipse. We measured respectively cross-correlation coefficients of −0.16 ± 0.24 and 0.72 ± 0.12.

Given the systematic uncertainties in the background subtraction, providing a clear explanation of the residual emission during the eclipse (if any) with RXTE can be challenging. Much more detailed information can be obtained from the Swift observations, thus we discuss the origin of the residual emission during the eclipse in Sects. 2.3 and 2.4.

2.3. Swift data

We analyzed the Swift /XRT (Gehrels et al. 2004) data collected in both photon counting mode (PC) and window timing mode (WT) using standard procedures (Burrows et al. 2005) and the latest calibration files available. Filtering and screening criteria were applied by using ftools. The barycentric correction was applied to the times of all the event files using the online tool barycorr. We extracted source and background light curves and spectra by selecting event grades of 0−2 and 0−12, respectively, for the WT and PC mode. Exposure maps were created through the xrtexpomap task, and we used the latest spectral redistribution matrices in the heasarc calibration database (v.011). Ancillary response files, accounting for different extraction regions, vignetting, and point spread function (PSF) corrections, were generated by using the xrtmkarf task. We corrected the PC data (where required) for pile-up. A log of the Swift observations is given in Table 1. For each observation in the table, we extracted the spectrum and derived the X-ray flux by fitting an absorbed power-law model. Spectra with sufficiently high count rate were rebinned to collect at least 20 photons per bin, to permit minimum χ² fitting. The spectra extracted in observations with ID. 00031686002 (WT mode) and 00031686006 were characterized by very low count rate; we therefore rebinned these spectra to have at least 5 photons per bin and then fit them by using the C-statistics (Cash 1979). During observation ID. 00031686007, the source was clearly detected in the Swift /XRT image but the exposure was too low to extract a meaningful spectrum. Therefore, we estimated the source count rate of the observation with the sosta program (available within the ftool ximage), and used the resulting count rate within webpimms to derive the X-ray flux (we assumed the same spectral model of the observation ID. 00031686005; see also Bozzo et al. 2009). A similar technique was adopted to estimate a 3σ upper limit to the source X-ray flux during observations ID. 00031686008,9,10, when Swift J1749.4−2807 was not detected. The results from this analysis are reported in Table 1.

The Swift/XRT light curves were folded at the orbital period of the source (see Sect. 2.2) to search for X-ray eclipses. An eclipse occurred during observation ID. 00031686002. A zoom of the light curve of this observation around the eclipse is shown in Fig. 6. This shows a residual X-ray flux during the eclipse (see Sect. 2.4).

We also reanalyzed all the observations carried out with Swift/XRT during the 2006 outburst to search for X-ray eclipses. By using the orbital solution discussed in Sect. 2.2, we noticed that some of these observations were performed when the source was in eclipse. However, the source count rate was far too low to extract a meaningful light curve and spectrum.

2.4. A scattering halo in the direction of Swift J1749.4−2807

To clarify the origin of the residual X-ray emission of Swift J1749.4−2807 during the eclipse, we show in Fig. 7 four images of the source extracted during the Swift observation ID. 00031686002. The two images at the top were accumulated during the time interval in which the X-ray source was eclipsed (total exposure time 840 s) and are in the 0.3−5.0 keV and 5−10 keV energy bands. The source is clearly detected in the lower energy image. The other two images were extracted in the same observation and energy bands, but during the 840 s preceding the eclipse. These figures suggest that the source was not point-like, but slightly extended (by a few arcminutes), especially at lower energies. To support this finding, we compared the radial distribution of photons during the eclipse with the PSF from the calibrations of Swift/XRT (using the tool ximage). We report the results in Fig. 8. The PSF during the eclipse was much flatter than expected for a point source.

This indicates that the residual emission around the source was most likely due to the effect of a scattering halo located halfway to Swift J1749.4−2807 (see e.g., Day & Tennant 1991; Predehl & Schmitt 1995)⁶. According to this interpretation, soft X-ray photons emitted when the source is outside the eclipse are scattered along our line of sight by interstellar dust, reaching us after the obscuration of the X-ray source, as a results of the longer path that they follow. The effect of a dust-scattering halo is most prominent at lower energies, because of the energy dependence of the scattering cross-section, and when the source is obscured, because direct photons from the source are then virtually absent.

We also searched for additional confirmation of our interpretation by comparing the spectra of the source during and outside the eclipse. By using data in observation ID. 00031686002, we found that the spectrum during the eclipse could be described well by an absorbed power-law model with ${\mathit{N}}_{\mathrm{H}}\mathrm{=}\mathrm{\left(}\mathrm{3.}{\mathrm{2}}_{-0.8}^{\mathrm{+}\mathrm{0.9}}\mathrm{\right)}\mathrm{\times }{\mathrm{10}}^{\mathrm{22}}$ cm^-2, and Γ = 3.6 ± 0.6 (C-statistic/d.o.f. = 38.4/44, exposure time 840 s). The average 0.5−10 keV flux in the eclipse was (1.6$\begin{array}{}\mathrm{+}\mathrm{0.1}\\ -1.1\end{array}\mathrm{)}\mathrm{\times }{\mathrm{10}}^{-11}$ erg/cm²/s, with a contribution from the photons in the 5−10 keV energy band lower than 20%. This is compatible (to within the errors) with the flux estimated during the eclipse by using RXTE (see Sect. 2.2).

For comparison, the spectrum of the source extracted outside the eclipse is described well by an absorbed power-law model with N_H = (3.0 ± 0.4) × 10²² cm^-2, and Γ = 2.3 ± 0.2 (${\mathit{\chi }}_{\mathrm{red}}^{\mathrm{2}}$/d.o.f. = 0.99/85, exposure time 2.6 ks). The X-ray flux in this case was (2.6 ± 0.3) × 10^-10 erg/cm²/s, with the photons from the hard energy band (5−10 keV) contributing for more than 45%. In Fig. 9, we also report the light curves of Swift J1749.4−2807 in the observation ID. 00031686002 in two energy bands. The hardness ratio, defined as the ratio of the Swift /XRT count rate in the hard (5−10 keV) to soft (0.3−5 keV) bands versus time, is also shown. Its average value was 0.16 ± 0.02 outside the eclipse and 0.04 ± 0.03 inside the eclipse. We conclude that the spectrum during the eclipse was much softer that that outside the eclipse, and compatible with the softening by a factor of E^-2 expected in case the extended emission is due to a scattering halo (see e.g., Day & Tennant 1991, and references therein).

We note that, for a halo size Θ of a few arcmin radius (see Fig. 7) and an estimated source distance of  ~7 kpc (see Sect. 1.1), the longer path followed by scattered photons translates into a delay (Thompson & Rothschild 2008) $\mathit{\delta t}\mathrm{=}\mathrm{7.6}{\mathit{d}}_{\mathrm{7} \mathrm{kpc}}\mathrm{(}\mathrm{\Theta }\mathit{/}\mathrm{arcsec}{\mathrm{)}}^{\mathrm{2}}\mathit{x}\mathit{/}\mathrm{(}\mathrm{1}\mathrm{-}\mathit{x}\mathrm{)}\mathrm{~}\mathrm{3}\mathrm{\times }{\mathrm{10}}^{\mathrm{4}} \mathrm{s}\mathit{,}$(5)where x is the fractional distance of the halo from the source along the line of sight, d_7   kpc kpc is the source distance in units of 7 kpc, and we have used Θ = 1 arcmin and x = 1/2, as indicative values. The typical delay is thus of few hours, i.e., much longer than the duration of the X-ray eclipse (~2600 s).

In Fig. 10, we report the evolution in the X-ray flux of Swift J1749.4−2807 during the outburst in 2010, derived from the monitoring performed with INTEGRAL, RXTE, and Swift. The source flux decreased by three orders of magnitude in less than 15 days, displaying a quite clear exponential decay. A fit with an exponential function to the Swift data gave an e-folding time of $\mathit{\tau }\mathrm{=}\mathrm{1}{\mathrm{2}}_{-3}^{\mathrm{+}\mathrm{7}}$ days (${\mathit{\chi }}_{\mathrm{red}}^{\mathrm{2}}$/d.o.f. = 15.2/4). Even though the ${\mathit{\chi }}_{\mathrm{red}}^{\mathrm{2}}$ is not formally acceptable, we checked that this is due to the relatively large scatter in the Swift points during the steep decay.

In the right panel of Fig. 10, we also report for comparison the outburst occurred in 2006, observed by Swift /XRT. To compare the flux evolution between the two outbursts more reliably, we plot in this panel also the Swift /XRT observations of the outburst in 2010. In both cases, the times of the observations were scaled to the time of the type-I X-ray burst that was detected during each outburst. From this figure it is apparent that the outburst in 2010 lasted much longer than the one in 2006 (a factor of  ~2), and was characterized by a higher averaged X-ray luminosity (the decrease in the X-ray flux with time was, on average, much slower; see also Sect. 3). In both cases, the decrease of the source X-ray flux throughout the outburst was not smooth. A relatively large scatter between the fluxes measured from different Swift observations, similar to that reported above for the outburst in 2010, is clearly visible also during the outburst in 2006. Wijnands (2009) suggested that this scatter might be caused by some flares occurring during the decay from the outburst.

3. Discussion and conclusions

We have reported on the monitoring of the burst-only source Swift J1749.4−2807, which was discovered to undergo a new X-ray outburst by INTEGRAL on 2010 April 10.

3.1. The outburst decay

The peak outburst luminosity (see Fig. 10) was  ~1.8 × 10³⁶ erg/s (assuming a distance of 7 kpc), only slightly higher than that reported for the outburst that occurred in 2006 (~10³⁶ erg/s). Despite this similarity in peak X-ray luminosity, the evolution in the source X-ray flux during the two events was significantly different. The 2006 outburst displayed a clear power-law-like decay (index  ~− 1, Wijnands 2009), and lasted less than  ~6 days (see also Sect. 1.1). In contrast, the 2010 outburst lasted more that 11 days⁷ and was characterized by an exponential decay (see Sect. 2).

Disk instability models for the outbursts of transient NS LMXBs predict that both linearly and exponentially decaying outbursts can be produced (King 1998). King & Ritter (1998) showed that, when the irradiation of the accretion disk from the X-ray emission of the NS is strong enough (i.e., the peak X-ray luminosity at the onset of the outburst is high), the disk is completely ionized out to its outer edge and the light curve of the outburst follows an exponential decay. In this phase, most of the disc mass can be accreted onto the NS in a viscous timescale. A change in the profile of the outburst is expected once the source has faded below a certain luminosity level. Below this threshold, only part of the accretion disk can be ionized. As a consequence, both the mass accretion rate onto the NS and the outer boundary of the ionized region in the disk begin to decrease with time, leading to a linear (rather than exponential) decay (King 1998; Shahbaz et al. 1998; Powell et al. 2007).

The value of the critical X-ray luminosity at which these changes occur depends on the properties of the system (total mass and orbital period). In the case of Swift J1749.4−2807 , this luminosity is expected to be on the order of 10³⁶ erg/s (Shahbaz et al. 1998), thus comparable to the peak X-ray luminosity estimated during both the events in 2006 and 2010. As the event in 2010 was characterized by an X-ray luminosity slightly higher than that measured at the peak of the outburst in 2006, it is possible that this relatively small difference was sufficient to cause a different behavior during the outburst decay. We note that, even though the peak luminosity of the two outbursts was almost comparable, during the outburst that occurred in 2010 the source remained in a high X-ray luminosity state for a significantly longer time interval than in 2006. This might have caused the ionization of a larger portion of the accretion disk and thus led to an exponential outburst decay. Given the relatively poor number of Swift and RXTE observations during the end of the outburst in 2010, it is not possible to investigate more deeply whether a switch between exponential and linear decay occurred at some point during the steep decay of the 2010 outburst.

The discovery of millisecond pulsations in the X-ray emission of Swift J1749.4−2807 secured the previously suggested association of this source (and other burst-only sources in general) with the class of the accreting AMSPs. This permits a comparison between the two classes. In particular, AMSPs are known to undergo week-to-month-long outbursts that usually exhibit an initial exponential decay, followed by a linear one, in agreement with the prediction of the disk instability model discussed above. An in-depth study of the outburst decay of a number of AMSPs was carried out by Powell et al. (2007). These authors showed that most of these sources display a “knee” in their decay, when the X-ray luminosity falls below the threshold at which the switch between exponential to linear decay is predicted.

This behavior was observed in the AMSPs SAX J1808.4−3658, XTE J1751−305, and XTE J0929−314. Some anomalies in the outburst decay were reported for the AMSPs SAX J1808.4−3658, XTE J1807−294, and IGR J1751−305. During an outburst in 2000, SAX J1808.4−3658 displayed several unexpected “re-flares” about  ~20 days after the initial exponential+linear decay (Wijnands et al. 2006; Campana et al. 2008), whereas the source XTE J1807−294 underwent in 2003 a rather long (≳ 100 day) pure exponential outburst without any clear evidence of a linear phase (see e.g., Falanga et al. 2005; Powell et al. 2007, and references therein). Another AMSP, IGR J1751−305, displayed a few outbursts lasting exceptionally only a few days and reaching a relatively low X-ray luminosity (a few mCrab, see e.g., Falanga et al. 2007; Markwardt et al. 2007). The outbursts observed from Swift J1749.4−2807 in 2006 and 2010 were both significantly shorter than those observed from the other AMSPs, and only the anomalous outbursts detected from IGR J1751−305 resemble those of Swift J1749.4−2807.

Similar dim outbursts have been observed from the very faint X-ray transients (VFXTs, see Sect. 1), and an association between these sources and the burst-only sources was suggested by Campana (2009). Different ways of interpreting the behavior of the VFXTs were reported by King & Wijnands (2006). These authors showed that the VFXTs hosting NS primaries can be relatively well understood if the material accreted on these systems originates from brown dwarfs or planetary companions. However, in the case of Swift J1749.4−2807 such low-mass companion is ruled out by the mass function of the system, which implies a minimum companion mass of  ~0.5 M_⊙ assuming a NS of 1   M_⊙.

3.2. Timing features

In the case of Swift J1749.4−2807, an in-depth study of the properties of the primary and secondary stars is possible for the first time by using the orbital solution and characteristics of the X-ray eclipses. At present, Swift J1749.4−2807 is the only eclipsing AMSPs. An extensive discussion of these issues is beyond the scope of the present work, and will be reported in a separate paper (Campana et al. 2010, in preparation).

In Sect. 2.4, we carried out a careful analysis of the residual X-ray emission of Swift J1749.4−2807 during the eclipse. While our analysis of the RXTE data could not unambiguously identify the origin of this emission, we found in the Swift data strong evidence that residual emission during the eclipse is caused by a scattering halo in the direction of the source. The effect of a dust scattering halo is most prominent when a source is in eclipse and at lower energies (≲3 keV, see Sect. 2.4). Since Swift J1749.4−2807 is the first eclipsing AMSP, it is also the first source in this class for which this effect could be detected.

An additional peculiarity of Swift J1749.4−2807 was revealed by the Fourier analysis of the RXTE data, which showed a signal at the spin frequency of the NS and its second harmonic. This is consistent with the fact that the pulse profiles clearly displayed a double peak (see Fig. 4). This feature has never been observed in another AMSP and is most likely due to emission from both accreting polar caps that sweeps across our line of sight. This geometry is possible because the system is seen nearly edge-on (as confirmed by the detection of X-ray eclipses). Moreover, the study of the pulse profile in different RXTE observations has found that the shape and amplitude of the two peaks changed significantly during the outburst decay. We propose that this behavior is caused by changes in the geometry of the emitting region close to the NS surface and/or changes in the radius and/or thickness of the inner edge of the accretion disk.

3.3. Type I X-ray burst

Our analysis of the data presented here has also detected a type-I X-ray burst in the JEM-X2 light curve. This is the second type-I X-ray burst reported so far from Swift J1749.4−2807. The first burst was discovered with Swift /BAT (Wijnands 2009) and was characterized by a peak flux a factor of  ~2 higher than that measured by JEM-X2 (see also Sect. 1.1). The statistics of the JEM-X2 were of too low quality to perform a time-resolved spectral analysis of the burst, and no evidence of a photospheric radius expansion could be found (see Sect. 2). The observation of a type-I X-ray burst can be used to derive an upper limit to the source distance by assuming that the peak X-ray luminosity of the burst corresponded to the Eddington value L_Edd ≈ 3.8 × 10³⁸ erg/s (as empirically derived by Kuulkers et al. 2003, for a helium burst). However, since the peak flux of the JEM-X2 burst was significantly lower than that observed previously by Swift, the latter already provided the most restrictive upper limit⁸ of 7 kpc on the distance to Swift J1749.4−2807 (see Sect. 1). Using this distance, the persistent unabsorbed 0.1−100 keV flux of the source at the time of the JEM-X2 burst would translate into a bolometric luminosity of L_pers ≈    4.7 × 10³⁶ erg/s, or 1.2% L_Edd. This corresponds to a local accretion rate per unit area of ṁ ≈ 3.2 × 10³ g/s/cm² ≃ 1.2%ṁ_Edd, where we have used the relation L_pers = 4πR²ṁ(GM/R)/(1 + z) (with z = 0.31 the NS gravitational redshift; see e.g., Lewin et al. 1993).

Theoretical models predict that, when 0.01 < ṁ/$\dot{{\mathit{m}}_{\mathrm{edd}}}\mathit{<}\mathrm{0.1}$, H burns stably through the hot CNO cycle and a layer composed of pure He develops underneath the NS surface. This layer can then ignite by means of the 3α process and lead to a pure He burst, with a typical rise time of  ~1 s and a duration of  ~10 s. These quantities, estimated for the burst detected by JEM-X2 (see Sect. 2), are fully compatible with those expected for a pure He burst.

We note that pure He runaways were also reported for the other burst-only source GRS 1741.9−2853 (Trap et al. 2009). The detection of this kind of type-I X-ray bursts from the burst-only sources is particularly intriguing because it would argue against the idea that these sources are the prototypes of the (poorly observed) H-burning bursts with low accretion rates (see Sect. 1 and, e.g. Peng et al. 2007).

Finally, we can check the consistency of the results derived above for the type-I X-ray burst by evaluating the theoretically expected recurrence time of the burst. We first estimate the ignition depth of the burst, y_ign, using the equation E_burst = 4πR²y_ignQ_nuc/(1 + z), where E_burst = 4πd²f_b = 2.0 × 10³⁹ erg (d/7 kpc), f_b is the measured fluence of the burst (see Table 2), and Q_nuc ≈ 1.6 MeV corresponds to the nuclear energy release per nucleon for complete burning of helium to iron group elements (Wallace & Woosley 1981; Fujimoto et al. 1987). We obtained y_ign = 1.4 × 10⁸ g  cm^-2. For the above values of the local accretion rates and ignition depth, the expected recurrence time of a He bursts is about Δt = (y_ign/ṁ)(1 + z) ≃ 0.6 days (independent of the assumed distance). The burst detected by JEM-X2 occurred  ~6.8 d after the first available INTEGRAL observation (see Fig. 10). However, the total effective exposure time on the source was  ~0.9 days, thus compatible with the observation of a single type-I X-ray burst throughout the observational period.

At the time of writing this work, two other papers were submitted describing the same source: Markwardt et al. (2010b) and Altamirano et al. (2010b). These authors focused mostly on the timing analysis of the system and discussed in detail the dynamical constraints on the binary system that could be derived by using the eclipses detected in the RXTE data. In our paper, we have concentrated more on the study and interpretation of the evolution of the timing and spectral properties of the source during the outburst. We checked (where possible) that the results derived in our paper are fully compatible with those reported in the other two papers. We have also discussed for the first time the properties of the eclipse found in the Swift data, and provided an interpretation of the residual flux during the eclipse.

---

Acknowledgments

E.B. and C.F. thank N. Gehrels and the Swift team for their availability and prompt response in carrying out follow-up observations of Swift J1749.4−2807. We thank S. Suchy for support from the HEXTE instrument team and the research groups of IAA-Tübingen and Dr. Remeis-Sternwarte in Bamberg for developing and making available useful scripts to analyze RXTE data. This research has made use of the XRT Data Analysis Software (XRTDAS) developed under the responsibility of the ASI Science Data Center (ASDC), Italy. A.P. acknowledges financial support from the Autonomous Region of Sardinia through a research grant under the program PO Sardegna FSE 2007-2013, L.R. 7/2007 “Promoting scientific research and innovation technology in Sardinia”.

References

All Tables

All Figures