© ESO, 2010

1. Introduction

Accreting binaries often show Fe line emission in their X-ray spectra. The ability of Chandra, XMM-Newton, Suzaku  and Swift to obtain medium- to high-resolution spectra covering the Fe K energy band has opened a new era in the observations of stellar-mass black holes (BHs) and NS binaries. The large effective area of these observatories is crucial for detecting Fe line emission, while the high resolution enables us to distinguish between narrow and intrinsically broad features. Broad Fe lines are also present in active galactic nuclei (Tanaka et al. 1995; Nandra et al. 1997), though studies of large samples of sources show that the average fraction of broad Fe lines is never higher than 40% (Guainazzi et al. 2006; Nandra et al. 2007).

The origin of broad Fe lines has been extensively discussed (e.g. Sunyaev & Titarchuk 1980; Fabian et al. 1989; Done et al. 2007; Ross & Fabian 2007; Titarchuk et al. 2009). However, even after the advent of the current powerful X-ray observatories, the exact determination of the line width and the mechanisms responsible for it are still controversial. Undoubtedly, the most exciting possibility is that such lines originate in the disc close to the BH event horizon or to the NS by fluorescent emission following illumination (and photoinisation) of the accretion disc by an external source of X-rays (Reynolds & Nowak 2003; Fabian & Miniutti 2005; Matt 2006). In this model, the combination of relativistic Doppler effects arising from the high orbital velocities and gravitational effects of the strong gravitational field in the vicinity of the compact object smear the reflected features. In this case, detailed X-ray spectroscopy of Fe line features can be used to study Doppler and gravitational redshifts, thereby providing key information on the location and kinematics of the material in the vicinity of the compact object. Most interesting of all is the potential for establishing BH spin using relativistic Fe Kα lines. The spin value is constrained mainly by the lower boundary of the broad line, which depends on the inner boundary of the disc emission (identified with the marginally stable orbit) where the gravitational redshift is maximal. Therefore, lines that are strongly skewed toward lower energies can indicate black hole spin (Miller et al. 2009; Hiemstra et al. 2009; Reis et al. 2009).

The recent claim of broad skewed Fe lines from NS binaries has opened the exciting possibility of determining an upper limit to the radius of the NS, the most difficult parameter to obtain to constrain the equation of state of NSs (Bhattacharyya & Strohmayer 2007). The advantage of using Fe K emission lines as a probe of the NS radius is that they only require short observations to clearly reveal the relativistic lines, and do not require any knowledge of the distance to the object.

An alternative location for the line-emitting region could be the inner part of the so-called accretion disc corona (ADC), formed by evaporation of the outer layers of the disc illuminated by the emission of the central object (e.g., White & Holt 1982). In the corona, where the gas is highly ionised, the Fe line emission is likely to come from recombination onto hydrogenic and helium-like Fe. In the region below the transition region, the Fe will be ionised at most a few times and fluorescence will be the dominant process (Kallman & White 1989). In the corona, Compton scattering will broaden and shift the centroid of the line. The line broadening and shift are due to Doppler and/or recoil effects, depending on the temperature of the plasma compared to the energy of the photons (Pozdnyakov et al. 1979; Sunyaev & Titarchuk 1980). The final line width is determined by the simultaneous effects of blending, Compton scattering, and rotation. For a value of f  ~  0.1 (f being the fraction of the incident X-ray flux, which is assumed to penetrate the base of the hot medium or corona), Kallman & White (1989) calculated a centroid energy of 6.6 − 6.7 keV for the Fe line and a width of  ~ 0.6 keV at small radii (R ≲ 10⁸ cm). For f  ~  1, the width can be as much as  ~ 1 keV (FWHM). Alternatively, the widths of  ~ 1 keV can also be attained via emission from a photoionised gas with large Thomson depth (τ ≳ 3). Such large Compton depths could arise if the corona is heated by some mechanism other than X-rays, or if the line of sight of the observer through the corona were very different from that of the compact X-ray source (Kallman & White 1989).

A third scenario for the formation of broad Fe lines in LMXBs could be one where extensive red wings form by recoil of line photons in an optically thick medium expanding, or converging, at relativistic velocities (Titarchuk et al. 2003; Laming & Titarchuk 2004; Laurent & Titarchuk 2007). It is assumed that the wide-open wind is launched at some disc radius where the local radiation force presumably exceeds the local disc gravity. The wind should be illuminated by the emission of X-rays formed in the innermost part of the source. The Kα line is generated in a narrow wind shell, and the line profile is formed in the partly ionised wind when Kα line photons are scattered off the diverging flow (wind) electrons (Titarchuk et al. 2009). The red-skewed part of the spectrum is formed by photons undergoing multiple scatterings, while the primary peak is formed by photons escaping directly to the observer.

The line properties of an ADC are in general expected to be different from those of an accretion disc (Brandt & Matt 1994). In the former case, the equivalent width (EW) is nearly independent of inclination except at very high angles (Vrtilek et al. 1993), while the dependence on inclination in the case of an accretion disc is significant (Brandt & Matt 1994). Therefore Fe lines appear to be a powerful tool for distinguishing between an ADC, an accretion disc, or any other origin of the Fe line in LMXBs. Proper modelling of the Fe line is mandatory to determine the unique origin of the line and, in the case of relativistic broadened Fe lines, to derive the spin of the BH or an upper limit to the radius of the NS.

White et al. (1986) performed a systematic analysis of six NS LMXBs with EXOSAT and found line emission in 83% of the objects. The lines were broad with FWHM of 0.8 − 1.3 keV and had centroid energies of 6.6 − 6.9 keV. Furthermore, the line properties did not show any obvious correlation with luminosity. This led White et al. (1986) to conclude that the only plausible broadening mechanism was Comptonisation in a cloud with a Thomson depth of a few and an electron temperature close to 7 keV. Hirano et al. (1987) analysed ten NS LMXBs observed with Tenma and found line emission from six of them. The line energy had a weighted average of 6.66  ±  0.05 keV with EWs in the range 20 − 60 eV. A significant line width of 0.55 keV was only obtained for one source, though a width of  ~ 1 keV could not be rejected for the other cases. They attributed the line emission to recombination processes in the accretion-disc corona and explained its width as Compton scattering in the region with an optical depth of 1 and an average temperature  ~ 1 keV. Asai et al. (2000) extended the analysis of the Fe K emission lines to twenty NS LMXBs observed with ASCA. They detected significant Fe lines from about half of the sources. The average properties of the lines were a line centre of 6.56 keV and a finite width of  ~ 0.5 keV (FWHM) in six of the sources. They also found a large scatter in the EW of the detected lines, ranging between 10 and 170 eV. They concluded that the Fe K lines are likely produced through the radiative recombination of photoionised plasma and explained the line width as a combination of line blending, Doppler broadening, and Compton scatterings.

The picture outlined by the analyses based on EXOSAT, Tenma and ASCA data changed with recent analyses of observations of NS LMXB observations by XMM-Newton, Chandra  and Suzaku. Several skewed Fe lines have been claimed, and their origin attributed to relativistic effects (e.g., Bhattacharyya & Strohmayer 2007; Papitto et al. 2009; Di Salvo et al. 2009; D’Aì et al. 2009) or to Compton scattering in an optically thick medium expanding, or converging, at relativistic velocities (e.g., Laurent & Titarchuk 2007; Titarchuk et al. 2009). However, only some of the NS LMXBs have shown such asymmetric lines, while others show symmetric lines (e.g., Cackett et al. 2008). A systematic analysis of the Fe K lines is at this stage fundamental for establishing why some sources exhibit skewed lines while others show symmetric lines or why lines are only present in some of the sources. Having an answer to these questions would help clarify the currently controversial origin of the lines.

With this aim we performed a systematic analysis of all the NS LMXBs observed by XMM-Newton since the beginning of the mission and publicly available up to September 30, 2009. We excluded from this sample observations of dipping or ADC sources, i.e. with known inclinations above 70°, since their analyses at the Fe band are complicated by strong absorption in the line of sight. We chose XMM-Newton to perform this study for the following reasons. Firstly, its high effective area both below and above the Fe K band allows a good determination of the shape and width of the Fe line as well as the continuum. Secondly, XMM-Newton has provided most of the detections and well-determined profiles of broad Fe lines. Finally, after 10 years of operation, XMM-Newton has an ample archive of public data which can be used to determine class properties in the least biased way.

In total we analysed 26 observations of 16 sources from the XMM-Newton archive. We restricted our analysis to sources observed in the EPIC pn timing mode. This mode is specially suited to observation of bright sources. Therefore, by selecting all the observations performed in this mode we obtained a sample of spectra with the best possible statistics.

In what follows we first present the properties of the selected sample and our analysis method. Special attention has been paid to eliminate pile-up effects and to the treatment of background. Then, we performed spectral fitting accounting for the excess emission at the Fe K energy band. We finally discuss the implications of our analysis regarding the characteristics of the lines and their possible origin. In an appendix we compare the characteristics of the lines in this work with previous analyses of the same observations and discuss the reason for any discrepancies that were found.

2. Observations and data reduction

The XMM-Newton Observatory (Jansen et al. 2001) includes three 1500 cm² X-ray telescopes, each with an EPIC (0.1 − 15 keV) at the focus. Two of the EPIC imaging spectrometers use MOS CCDs (Turner et al. 2001), and one uses pn CCDs (Strüder et al. 2001). The RGSs (0.35 − 2.5 keV, Den Herder et al. 2001) are located behind two of the telescopes. In addition, there is a co-aligned 30 cm diameter Optical/UV Monitor telescope (OM, Mason et al. 2001) that provides simultaneously coverage with the X-ray instruments. Data products were reduced using the Science Analysis Software (SAS) version 9.0. The EPIC MOS cameras were not used in most of the observations analysed in this paper because their telemetry was allocated to the EPIC pn camera to avoid Full Scientific Buffer in the latter. Since the pn has an effective area  ~ 5 times higher at 7 keV than the MOS CCDs and the latter were not available for most of the observations, we present only the analysis of the EPIC pn data here. We did not analyse the RGS data since they do not cover the Fe K energy band in which we are interested.

Table 1 is a summary of the XMM-Newton observations. The EPIC pn was used in timing mode for all the observations. In this mode only one CCD chip is operated, and the data are collapsed into a one-dimensional row (4$\mathrm{\prime ̣}$4) and read out at high speed, with the second dimension replaced by timing information. This allows a time resolution of 30 μs. We used the SAS task epfast on the event files to correct for a charge transfer inefficiency (CTI) effect seen in the EPIC pn timing mode when high count rates are present¹. Ancillary response files were generated using the SAS task arfgen following the recommendations of the XMM-Newton SAS User Guide² for piled-up observations in timing mode whenever applicable. Response matrices were generated using the SAS task rmfgen. We extracted one EPIC pn spectrum per observation, not taking any intra-observational variability into account (see Sect. 3.1). Bursts were excluded for the calculation of the total energy spectra whenever present.

Light curves were generated with the SAS task epiclccorr, which accounts for time-dependent corrections within a exposure, such as dead time and GTIs.

2.1. Pile-up treatment

Since the average count rate in the EPIC pn was close to, or above, the 800 counts s^-1 level at which pile-up effects may become significant for at least eight observations in the sample, we investigated the presence of pile-up in detail before extracting the spectra. We used the SAS task epatplot, which utilises the relative ratios of single- and double-pixel events deviating from the standard values as a diagnostic tool in case of significant pile-up in the pn camera timing mode. We found that the spectra of twelve observations from eight sources were affected by pile-up (see Table 1).

Then, we extracted several spectra selecting single and double timing mode events (patterns 0 to 4) and different spatial regions for each of the piled-up observations. For 4U 1705 − 44 (Obs 0551270201), 4U 1636 − 536, 4U 1735 − 44, and GX 9 + 9 (Obs 0090340601), the source coordinates fell into the centre of the central column of the CCD. Therefore, source events were first extracted from a 64′′ (15 columns) wide box centred on the source position (Region 1). Then, we excluded the neighbouring 1, 3, 5, and 7 columns from the centre of Region 1 (Regions 2 − 5, respectively) and extracted one spectrum for each of the defined regions. For GX 340 + 0, GX 349 + 2, Ser X − 1, and GX 9 + 9 (Obs 0090340101) the source fell between two columns in the CCD. Therefore, source events were first extracted from a 68′′ (16 columns) wide box centred on the source position (Region 1), and Regions 2 − 5 were defined by excluding 2, 4, 6, and 8 columns respectively from the centre of Region 1. This was done to preserve the best symmetry when excluding piled-up events. Table 1 lists the number of columns that had to be extracted from each source, in order to obtain spectra free of pile-up. It is evident that pile-up already starts to be important at a count rate of  ≳ 450 s^-1. Pile-up depends on the spectral shape and the time variation of the source in a complex way. Therefore, although the average count rate of an observation generally indicates whether the effects of pile-up are important, it is also of utmost importance to carefully inspect the epatplot to evaluate the PSF radius at which the relative ratios of single- and double-pixel events do not deviate from the standard values in the full energy band.

As an example of the use of epatplot to determine the amount of pile-up, we show the epatplots in a highly piled-up source, GX 349 + 2, and in a dim source free of pile-up, 4U 1728 − 34, in Appendix A.

2.2. Background treatment

In the EPIC pn timing mode, there is no source-free background region, since the PSF of the telescope extends further than the central CCD boundaries. The central CCD has a field of view of in the pn. In timing mode, the largest column is the one in which the data are collapsed into one-dimensional row. Therefore, the maximum angle for background extraction is 2′, compared to 5′ for imaging modes. In our sample, sixteen out of the nineteen observations (for which source emission is detected) are very bright, with total count rates  ≳ 200 s^-1 (see Table 1). Therefore, the spectra from these sources will not be significantly modified by the “real” background, which contributes less than 1% to the total count rate in most of the bandwidth. Conversely, subtracting the background extracted from the outer columns of the central CCD will modify the source spectrum, since the PSF is energy dependent and the source photons scattered to the outer columns do not show the same energy dependence as the photons focused on the inner columns.

For the reasons mentioned above, in this work we chose not to subtract the “background” extracted from the outer regions of the central CCD. This is an appropriate method for all the sources where the “real” background is negligible compared to the source count rate at all energies. In contrast, for sources with N_H  ≳  1  ×    10²² atom cm^-2, the background is expected to contribute significantly at energies  ≲ 1 keV, where most of the source photons are absorbed in the interstellar medium. Therefore for these sources, it is mandatory to remove the bins where the background is significant before spectral analysis.

With this aim, we extracted the background from the outer regions of the central CCD of “blank fields”, where we took as examples of “blank fields” the observations of Aql X − 1 and 4U 1608 − 52 listed in Table 1 for which the source was not significantly detected. Then, we inspected the spectra of sources with N_H  ≳  0.5  ×    10²² atom cm^-2, and compared the “blank field” background with the background extracted from the observation for which the source was being analysed. We scaled the former so that it was comparable to the latter at energies where we did not expect a significant flux from the source. We estimated such energy as the one at which simulated spectra with different values of N_H started to flatten as a consequence of being dominated by background events. Then we removed energy bins where we expect a contribution from the background to the total count rate of more than 1%. For 4U 1705 − 44, 4U 1728 − 34, and GX 340 + 0, we removed bins below 1.5, 1.8, and 2.2 keV, respectively. We used all the other spectra from 0.7 keV up to 10 keV.

To illustrate the effects of background subtraction in the spectral fitting and to evaluate the validity of our method, we chose the weakest and brightest sources of our sample, 4U 1705 − 44 (Obs 0402300201) and GX 349 + 2. We examined the differences in the spectra after (1) not subtracting any background from the source spectrum (Spec 1); (2) subtracting the background extracted from a blank field (Spec 2); and (3) subtracting the background extracted from the outer columns of the central CCD during the observation (Spec 3). Figure 1 (left panels) shows the ratio of Spec 1 to its best-fit continuum model (black) and the difference from Specs 2 and 3 with respect to the same model. The red and black residuals are consistent within the errors at all energies for GX 349 + 2 (lower panels) and above  ~1 keV for 4U 1705 − 44 (upper panels). The deviation below  ~1 keV for 4U 1705 − 44 is expected, since at such energies the background photons make a significant contribution to the total spectrum. In contrast, we observe an energy-dependent discrepancy of Spec 3 with respect to Specs 1 and 2, as expected from the energy dependence of the PSF. These results are consistent with the shape of the background shown in Fig. 1 (right panels). The “background” extracted from the outer columns of the CCD is clearly contaminated by the source at all energies (compared to the source spectrum), while the “background” extracted from a blank field shows the shape expected when compared to Fig. 35 of the XMM-Newton Users Handbook.

3. Light curves and spectral fitting

3.1. Light curves

Figure 2 shows EPIC pn light curves and hardness ratios of all the XMM-Newton observations analysed in this paper with a binning of 64 s. The hardness ratio is obtained by dividing the number of counts in the 3 − 10 keV energy range by those in the 0.7 − 3 keV energy range, except for 4U 1705 − 44, 4U 1728 − 34 and GX 340 + 0, for which the soft band is 1.5 − 3 keV, 1.8 − 3 keV and 2.2 − 3 keV, respectively. The light curves of observations with count rates  ≳ 200 counts s^-1 suffer from regular telemetry drops that are seen when the light curves are plotted with a high time resolution, e.g. 1 s. The SAS task epiclccorr accounts for such gaps, as well as for dead time with a different origin, by taking the count rate in adjacent frames into account (see Sect. 2).

The average count rate changed by more than 2 orders of magnitude among the studied sources. Observations of the same source taken within months showed strong variability in some cases (e.g. 4U 1636 − 536) and a steady level in others (e.g. Ser X − 1).

Strong variability of  ≳ 30% within one observation is only present in the light curves of GX 340 + 0 and GX 349 + 2. The variabilities within such observations were studied by D’Aì et al. (2009) and Iaria et al. (2009), who find that the spectral fits to different intervals within the observations gave consistent results for the parameters of the Fe line.

3.2. X-ray spectra

We rebinned the EPIC pn spectra to over-sample the FWHM of the energy resolution by a factor of 3 and to have a minimum of 25 counts per bin, to allow the use of the χ² statistic. To account for systematic effects in the EPIC pn timing mode, we added a 2% uncertainty quadratically to each pn spectral bin. We performed spectral analysis using XSPEC (Arnaud 1996) version 12.3.1. We used the photo-electric cross-sections of Wilms et al. (2000) to account for absorption by neutral gas with solar abundances (the so-called tbabs XSPEC model). Spectral uncertainties are given at 90% confidence (Δχ² = 2.71), and upper limits at 95% confidence.

We first fitted the EPIC pn spectra with a model consisting of a blackbody and a disc blackbody, both modified by photo-electric absorption from neutral material (model tbabs*(bbodyrad+diskbb) in XSPEC). The fits were unacceptable in most of the cases, mainly owing to the broad emission feature at  ~ 6 − 7 keV and a significant excess in the pn spectrum below  ~ 1.5 keV. This excess consisted of a “Gaussian-like” component centred at  ~ 1 keV. This feature has been previously modelled in a number of sources either as an emission line or as an edge, and its nature is unclear (e.g., Sidoli et al. 2001; Boirin & Parmar 2003; Boirin et al. 2004, 2005). If the feature has an astrophysical origin, its energy is consistent with Ne x or a blend of Fe xx-Fe xxiv emission. When the feature is edge-like, its energy is consistent with O viii. We therefore fitted such a feature with a Gaussian component or with an edge whenever present, and discuss its origin in Sect. 5. We fitted the excess at  ~ 6 − 7 keV with either a Gaussian or a laor component (see below). Finally, we added two Gaussian absorption/emission features at  ~ 1.84 keV and  ~ 2.28 keV. Such features probably stem from an incorrect modelling of the Si and Au absorption in the CCD detectors by the EPIC pn calibration, so are not discussed further.

Summarising, our final model consisted of disc blackbody and blackbody components, one Gaussian emission feature at  ~ 1 keV (or absorption edge at  ~ 0.87 keV), and one emission feature at  ~ 6.5 keV (modelled with Gaussian or laor), both modified by photo-electric absorption from neutral material, and two narrow Gaussian features at  ~ 1.84 and  ~ 2.28 keV to account for the calibration deficiencies at these energies (Models 1a and 2a, see Table 2).

The fits with Model 1a were acceptable for fifteen out of the nineteen observations for which we extracted a spectrum, with between 0.7 and 1.2 for a number of degrees of freedom (d.o.f.) between 161 and 221. The parameters of the best fit with this model are given in Table 3, and the residuals of the fits are shown in Fig. C.1 (see Appendix C).

Substituting the blackbody or disc blackbody components in Model 1a by a power law component resulted in fits with similar or worse for all the observations shown in Table 3 (see Table 6). Substituting the blackbody component by a cutoff power law in Model 1a resulted in fits with similar . This is expected, since the blackbody in Model 1a accounts for the emission of the boundary layer and could therefore be equally well fitted by a cutoff power law or a comptt component, representing saturated Comptonisation. Finally, substituting the disc blackbody component by a cutoff power law in Model 1a resulted in fits with similar , but with unrealistic values for the index and the cutoff energy of the cutoff power law component. For example, for the bright source GX 349 + 2, we obtained an index of  − 0.2 ± 0.4 and an energy for the cutoff of 1.0 keV. Similarly, for the dim source 4U 1728 − 34, we obtained an energy for the cutoff of 0.51  ±  0.02 keV, and the index was unconstrained. Fixing the index of the cutoff power law to a more realistic value of 1 (Hasinger et al. 1990) yielded fits with significantly worse .

We compared the for the different continua before adding the emission line at  ~ 6.5 keV to avoid the EW of the line being affected by the possible deficiencies of the fit to the continuum. Whenever two different continua yielded a similar before including the Fe line, e.g. using a disc blackbody in combination with a blackbody or cutoff power law components, the breadth of the line was similar in both fits.

For one of the remaining four observations, corresponding to XTE J1807 − 294, we had already obtained an acceptable of 1.20 for 224 d.o.f. when fitting the continuum with Model 1a. However, a better fit was obtained substituting the blackbody or disc blackbody components by a power law (Models 1b or 1c, see Table 2), with a final of 1.00 (Model 1b) and 1.02 (Model 1c) for 224 d.o.f. None of the fits showed significant residual emission at  ~ 1 or  ~ 6.5 keV. We calculated an upper limit of 17 eV for the EW of a potential Fe line with an energy of 6.6 keV and a width of 0.3 keV in this observation. For the other AMXP in our sample, SAX J1808.4 − 3658, fitting the spectrum with Models 1a or 1b gave significantly worse results ( of 20.7 and 2.2 for 224 d.o.f., respectively) than with Model 1c ( of 1.1 for 224 d.o.f.). Since XTE J1807 − 294 was equally well fitted with Models 1b and 1c and SAX J1808.4 − 3658 was best-fit with Model 1c, we chose the latter to model both sources, thus allowing a better comparison of the parameters. For SAX J1808.4 − 3658, we show two fits, corresponding to 2 different spectra: one extracted using the full PSF and one after removal of the 2 central columns of the PSF (marked with ** in Tables 4, 5). The reason is that it was difficult to decide based on the epatplot whether the original spectrum had some residual pile-up. The second spectrum (**) was completely free of pile-up. We show the parameters of the best fit for XTE J1807 − 294 and SAX J1808.4 − 3658 in Table 4.

Similarly, for 4U 0614 + 09, we obtained a better of 1.19 (220) with Model 1c, compared to 2.25 (220) with Model 1a and 1.21 (220) with Model 1b. We show the parameters of the best fit for this source in Table 4.

Finally, for Obs 0303250201 of 4U 1636 − 536 we needed three continuum components, blackbody, disc blackbody, and power law, in order to obtain an acceptable . Fits of this observation with a simultaneous RXTE observation also require three components and show a power law extending well above 10 keV (Pandel et al. 2008). We show the parameters of the best fit for this observation in Table 4.

Next, we examined the asymmetry of the line and evaluated to what extent the Fe emission was better approximated by a model including relativistic effects. For this, we substituted the second Gaussian component at  ~ 6.5 keV (gau₂) in Models 1a, 1b, and 1c by a laor component and re-fitted all the spectra with the new models (Models 2a, 2b, and 2c, see Table 2). We show the parameters of the best fit with these models in Tables 3, 4 and the residuals of the fits in Fig. C.1. The fits with Models 2a, 2b, and 2c were again acceptable, with between 0.8 and 1.2 for a number of d.o.f. between 161 and 221, i.e. similar to the obtained when a simple Gaussian model was used to fit the lines.

We found that fits to continua different to the one selected that yielded significantly higher values of had a clear influence on the line breadth for all the sources analysed in this work.

Narrow absorption features were not visible in any of the spectra, as expected for sources that are not at high inclination. Similarly, adding an absorption edge to the best-fit models shown in Table 3 did not significantly improve the goodness of the fits for any source with the exception of GX 349 + 2. For this source, the improved from 1.15 (216) to 1.07 (214) after adding an edge at  ~ 9.2 ± 0.1 keV, which we attribute to absorption from Fe xxvi. The parameters of the Fe line did not change after including this additional edge. This edge was previously detected in a BeppoSAX observation of the source (Iaria et al. 2004).

For sources with significant emission at  ~ 6.5 keV, the parameters of the continuum did not change significantly when fitting the former with a Gaussian or a laor component, except for 4U 1728 − 34. Previous analyses of this source based on Chandra and RXTE or BeppoSAX data (D’Aì et al. 2006) have shown equally good fits when modelling the emission at  ~ 6.5 keV with a broad line or with two absorption edges. Therefore, we regard the XMM-Newton fits from 4U 1728 − 34 in this work with caution and discuss them in the framework of the general properties of the Fe lines in the following sections. Figure 3 shows the ratio between the data and the best-fit model at the Fe line energy band for all the sources with significant emission when the Fe line is not included.

4. Properties of the sample

4.1. Fe line emission

Examination of the spectral fits (Tables 3 to 5 and Figs. C.1 − C.2) shows that we are able to successfully model the emission in the Fe K band with a simple Gaussian component. Fits of the excess in the Fe K band with the more complicated laor component do not improve the goodness of the fit significantly for any of the observations presented here (see Table 6).

Considering the whole sample, the Fe line has a weighted average energy of 6.67  ±  0.02 keV, a width of 0.33  ±  0.02 keV, and an EW of 42  ±  3 eV. The statistical distribution of these values is shown in Fig. 4. The width and the EW have a well-defined “Gaussian-like” distribution around the weighted average value. The outliers of the histograms in Fig. 4 have values with large errors, so they do not contribute to the weighted average significantly. The energy distribution peaks at  ~ 6.7 keV, in agreement with the value of the weighted average. This is consistent with emission from Fe xxv. Clearly, the distribution has values consistent with emission from highly ionised species of Fe, from Fe xxii to Fe xxvi, and it never shows a value consistent with neutral Fe.

To find out the reason for the large errors in the lines with an EW above 30 eV, we plotted the EW of the lines as a function of the total number of counts in the spectrum (see Fig. 5). The lines that show large errors are not associated to the spectra with fewer counts, indicating that the detection of the lines within our sample is not limited by statistics. Therefore, we examined in detail the three cases for which the errors on the parameters of the line are very large (corresponding to 4U 0614 + 09, 4U 1636 − 536 Obs 0303250201, and 4U 1728 − 34) to determine the reason for such large errors. As indicated in Sect. 3.2, fits to 4U 1728 − 34 spectra of Chandra and RXTE or BeppoSAX (D’Aì et al. 2006) show equally good fits when modelling the emission at  ~ 6.5 keV with a broad line or with two absorption edges. Including one edge in the XMM-Newton spectrum at 7.6 keV and no Fe line yields a of 1.32 (178). Although the is higher than when including the Fe line, the residuals at the Fe band are reduced considerably. The fits to the XMM-Newton spectrum have significantly different continuum parameters when fitting the residuals at the Fe band with Models 1a or 1b (see Table 3), in contrast to all other observations. This indicates that after including the line, the parameters of the fit change significantly so the line fit may not be realistic. Interestingly, 4U 0614 + 09 and 4U 1728 − 34 show the lowest luminosities of the full sample. Therefore, it is plausible that such observations have a significantly different, more complex spectrum compared to higher luminosity observations. For comparison, the MOS spectra of the observation of 4U 0614 + 09 analysed in this work were previously fitted with an absorbed power law and an emission feature at 0.65 keV (Méndez et al. 2002) and did not require a line at  ~ 6.5 keV. In the case of 4U 1636 − 536 Obs 0303250201, although the luminosity is higher than for other sources in our sample fitted with the “standard” Model 1a/1b, we needed three continuum components to obtain an acceptable fit. This again points to a complex spectrum. Figure 6 shows the residuals of the fit to the continuum before including the Fe line for the three observations discussed. The residuals at the Fe energy band are very different than those shown in Fig. 3, indicating that the lines fitted for these observations may not be realistic and that such fits may need, e.g., the inclusion of absorption edges. For these reasons we regard the properties of these lines with caution.

It is interesting to compare the properties of the Fe Kα line with general properties of the sources such as luminosity or temperatures of the blackbody and disc blackbody components, with the aim of constraining the origin of the line emission region. For example, we may expect a correlation between the source accretion rate and the breadth of the line, if the latter is caused by relativistic effects. Figure 7 shows the properties of the Fe Kα line (width and EW) as a function of the luminosity calculated in Table 7.

We do not find any clear correlation between the energy centroid, the width, or the EW of the Fe line with the source luminosity (see Fig. 7). The source luminosities range over 2 orders of magnitude, covering the full range of Eddington ratios from 0.003 to 1.26, while the line centroid, width, and EW do not show any systematic trend. Similary, we do not find any clear correlation between the EW of the Fe line and the temperature or flux of the blackbody or disc blackbody components.

4.2. Continuum emission

We examined the continuum properties of the sample for all the sources shown in Table 3. Figure 8 shows the variation in the blackbody luminosity in the band 0.5–30 keV as a function of the total luminosity in the same band. In this figure, we excluded the two AMSP XTE J1807 − 294 and SAX J1808.4 − 3658, 4U 0614 + 09 and Obs 0303250201 of 4U 1636 − 536. The reason was that these observations were fitted with a continuum different than the standard one of blackbody and disc blackbody components and therefore indicate different properties. We further excluded the two dim observations corresponding to 4U 1705 − 44 (Obs 0402300201) and 4U 1728 − 34, since for such dim sources, Compton scattering is expected to play an important role so was not included in our model.

The blackbody luminosity is very near the value of 50% of total luminosity expected from simple energy considerations for all the sources of the sample. Interestingly, the temperature of the blackbody component, kT_bb, is nearly independent of the global luminosity at luminosities  ≳ 2 × 10³⁶erg s^-1 (Fig. 8). Ignoring the point with the highest temperature, since it has associated a significantly larger error than the other points, kT_bb falls by  ~ 45% while the total luminosity increases by more than two orders of magnitude and the blackbody radius by more than a factor of 40.

5. Discussion

We performed a systematic analysis of 26 XMM-Newton observations corresponding to all the NS LMXBs observed with EPIC pn timing mode and publicly available up until September 30, 2009 to establish the characteristics of the Fe K line emission in these objects.

In seven observations we did not detect the source significantly. For the remaining nineteen observations, we extracted one spectrum per observation. We paid special attention to the effects of pile-up and background subtraction.

5.1. Fe line emission

We detected Fe Kα line emission in 80% of the observations where the source was significantly detected. The energy of the line has values between 6.53 and 6.97 keV, consistent with highly ionised Fe from Fe xxii to Fe xxvi. The width has values between 0.17 and 1.15 keV and the EW between 17 and 190 eV. Only in four cases out of the fifteen lines reported in this work did we find a width near or above 1 keV and an EW above 100 eV. Interestingly, there is a gap in the distributions of σ and EW between these high values and the rest of the sample. Lines with high values of σ or EW also have associated large errors. Vaughan & Uttley (2008) interpret a similar finding in the context of detection of narrow lines from AGN as the lines with large EWs as the most likely false detections. In the cases studied here, the lines with widths near 1 keV are also in the limit of detectability of 3σ, and the large errors most likely point to an inappropriate modelling of the continuum for 4U 0614 + 09, 4U 1728 − 34, and 4U 1636 − 536. Considering the whole sample, the Fe line has a weighted average energy of 6.67  ±  0.02 keV, a width (σ) of 0.33  ±  0.02 keV, and an EW of 42  ±  3 eV.

Recently, broad skewed Fe K emission lines from the disc have been reported from LMXBs containing a NS (e.g. Bhattacharyya & Strohmayer 2007; Iaria et al. 2009; D’Aì et al. 2009). All the XMM-Newton observations where a broad skewed Fe line has been reported were included in our sample (see Sect. 4). However, in contrast to previous analyses, we do not need to invoke relativistic effects to explain its width, which could be due to mechanisms such as Compton broadening. The lines are equally well fitted with the relativistic laor component or with a simpler Gaussian component. Furthermore, the profiles shown in Fig. 3 do not show any asymmetric shape, as expected if the lines are emitted close to the NS and shaped by relativistic effects. The line profile is instead symmetric, similar to the one found in dipping sources (Díaz Trigo et al. 2006). The major difference between the analysis presented in this work and previous works is a careful treatment of pile-up effects, common in the observations of bright LMXBs, and a different modelling of the continuum in some cases, which has a strong effect on the EW of the lines.

As shown by Brandt & Matt (1994), in a scenario where the Fe line is generated at the inner disc, we would expect the centroid and the width of the Fe line to decrease as the accretion rate diminishes and the disc recedes. The behaviour of the EW is somewhat more complex, because the line’s EW is expected to first increase with ξ up to a certain value where it starts to decrease (see Fig. 5b from Brandt & Matt 1994). This comes from the emission properties of Fe atoms as a function of the ionisation parameter (Matt et al. 1994). When the most abundant ions are Fe xxiv–Fe xxvi, line emission is more intense than in the neutral case, owing to the greater fluorescent yield and the smaller photoabsorption cross-section at the line energy. However, if ξ increases further, the ion distribution becomes dominated by fully stripped atoms, and the line emission decreases. In the observations presented in this paper, we do not observe a correlation between the centroid and the width of the Fe line with luminosity, in agreement with a similar systematic analysis based on ASCA observations (Asai et al. 2000). In contrast, the three cases for which a large width and EW are measured occur at relatively low luminosities and are actually associated to large errors. This seems to be at odds with the behaviour predicted by Brandt & Matt (1994) for lines generated at the disc and suggests instead that the physical condition of the line-emitting region is rather similar among the LMXBs.

An interpretation of the line in the context of disc emission would imply that the line is produced far from the neutron star to explain its symmetric shape. However, in such a case, it is difficult to justify the variations in the line as the disc inner radius changes.

Alternatively, the origin of the line broadening could be Compton scattering in a corona. Based on a systematic analysis of twenty NS LMXBs with ASCA, Asai et al. (2000) conclude that the Fe lines were likely produced through the radiative recombination of photoionised plasma, and the line width is a result of the combination of line blending, Doppler broadening, and Compton scattering. Combining such effects, Kallman & White (1989) estimated that EWs of up to 100 eV were attainable from an ADC using standard parameters. This agrees with the values obtained in this work.

Finally, the line could originate in a partially ionised wind as a result of illumination by the central source continuum photons and broadened by electron downscattering in the wind environment (e.g., Laurent & Titarchuk 2007; Titarchuk et al. 2009). The line profiles obtained via such a mechanism are again asymmetric, similar to the ones produced by relativistic effects. Since the line profiles obtained here are highly symmetric, we do not attribute the line origin to the outflow based on these observations.

5.2. Continuum emission

The X-ray continua for 80% of the observations were well fitted by a model consisting of a blackbody and disc-blackbody components absorbed by neutral material. For 15% of the observations, a model consisting of a blackbody and power law components absorbed by neutral material was preferred to the combination of blackbody and disc-blackbody components. For one observation, a fit with three components, namely, blackbody, disc-blackbody, and power law, was required to get an acceptable .

For the fits with blackbody and disc-blackbody components, the temperature of the blackbody component had values between 1.5 and 2.8 keV, except for 4U 1728 − 34, for which a temperature of 3.8 keV was found. The temperature of the disc blackbody component had lower values between 0.6 and 1.3 keV, except for 4U 1728 − 34, for which a temperature of 1.9  ±  0.3 keV was found. This is consistent with the interpretation of the blackbody component as the boundary layer between the neutron star and the inner disc, which shows smaller radii and higher temperatures. Based on an ASCA survey, Church & Balucińska-Church (2001) found a broad dependence of the blackbody luminosity on the total luminosity, not previously known for LMXBs in general; but while they found that the blackbody luminosity was falling more rapidly than the total luminosity as the mass accretion rate decreases, we find a very tight correlation where the blackbody luminosity is very near the value of 50% of total luminosity at all luminosities. This may be due to the different continuum model used, blackbody and cutoff power law components (Church & Balucińska-Church 2001) versus blackbody, and disc blackbody components (this work). Similar to Church & Balucińska-Church (2001), we find that the major change in blackbody emission results primarily from changes in the emitting area, not the temperature. The independence of the temperature of this component on the global value of Eddington ratio lends support to the theoretical suggestion that the boundary layer is radiation-pressure-supported (Inogamov & Sunyaev 1999; Revnivtsev & Gilfanov 2006).

For the sources fitted with blackbody and power law components, the temperature of the blackbody component had values between 0.6 and 2.2 keV and the index of the power law was between 1.8 and 2.6.

We detected an excess of emission at  ~ 1 keV in all the observations where we could use the spectrum down to 0.7 keV, except for the two accreting millisecond pulsars XTE J1807 − 294 and SAX J1808.4 − 3658. We modelled such emission with a Gaussian component centred between 1.0 and 1.15 keV in 12 observations and as an edge between 0.84 and 0.86 keV in 2 observations. Provided the feature has an astrophysical origin, appearing always at the similar energy points to line emission, since a soft component due to a, say, blackbody emission is expected to change its temperature for different sources. Its energy is consistent with emission from highly ionised Fe, from Fe xx to Fe xxiv, and/or Ne x. We did not find any clear correlation between the width of the line emission at  ~ 1 keV and the Fe Kα band. However, the fact that the same feature is not observed in simultaneous EPIC and RGS exposures (e.g. Boirin et al. 2005) indicates that the feature is likely blended emission from various ions and additionally broadened by the same mechanism as the Fe Kα line.

In this work, we show that effects of pile-up can significantly affect the width, hence the physical interpretation, of the Fe line in NS X-ray binaries. Furthermore, parallel works on XMM-Newton and Suzaku observations of the black hole X-ray binary GX 339-4 demonstrate that, taking pile-up effects into account, the breadth of the line is consistent with a truncation of the disc (Done & Díaz Trigo 2010) or with a black hole with moderate spin (Yamada et al. 2009), in contrast to previous claims (e.g. Miller et al. 2006, 2008). Therefore, we urge caution in using piled-up data for detailed spectral analysis.

Note Added in proof. While this paper was under revision, a second analysis of the two XMM-Newton observations of 4U 1705 − 44 analysed here was reported (D’Aì et al. 2010). Their analysis of the first observation of this source is consistent with the results reported in this paper. However, the analysis of the second observation is consistent with the first analysis by Di Salvo et al. (2009) discussed in Appendix B.4 and consequently inconsistent with the results reported in this paper. Therefore, our conclusions remain unchanged.

Online material

Appendix A: Examples of pile-up removal

Figures A.1 and A.2 show the effects of pile-up in the epatplot in the bright source GX 349 + 2 and the dim source 4U 1728 − 34, respectively. For GX 349 + 2 we observe a significant energy-dependent deviation between the data and the model  ≳ 4 keV when the full PSF region is used (Fig. A.1, upper left). As we remove the inner columns from the PSF the effects of pile-up are mitigated. After removal of 6 columns, there is still a small deviation above 9 keV. After removal of 8 columns, there is not an energy-dependent deviation anymore (Fig. A.1, right panel). For 4U 1728 − 34 we do not observe any energy-dependent deviation between the data and the model when the full PSF region is used (Fig. A.2, left panel). Similarly, when we excise the core of the PSF, the pattern ratios do not change with respect to the case where the core of the PSF is included. This indicates that pile-up is not present in this source.

Appendix B: Comparison with previous results

Several sources analysed in this paper have been previously published by other authors (e.g. Bhattacharyya & Strohmayer 2007; Iaria et al. 2009; D’Aì et al. 2009). In what follows, we compare the properties of the Fe Kα line found in previous analyses with the ones inferred in this paper. For most of the sources we obtained significantly different parameters for the Fe line when fitted with a laor component, compared to previous analyses (we discuss the discrepancies in detail below).

Therefore, to obtain a quantitative measurement of the difference between the Fe line obtained by other authors and in this work we performed fits to the spectra with the best-fit model shown in Sect. 3.2 but fixing all the parameters of the laor component to the values obtained in previous analyses except the normalisation, which was left free. The of these fits is shown in Table B.1. For comparison, we show in the same table the obtained when the line is fitted both with a Gaussian and a laor component and leaving all the parameters free, as in Tables 3 − 5. We compared the results in this work with the results obtained by Cackett et al. (2010) only in Table B.1 and not in the sections that follow, since their work is being refereed at the moment of acceptance of this paper, and the parameters may still change in their final version.

We found that the of the fits to the lines with the parameters of the literature yielded higher values of for eight of the observations analysed in this work (see Table B.1). For the three observations of 4U 1636 − 536, the difference in is not statistically significant. Indeed the lines fitted in this work have a width σ near 1 keV, similar to previous analyses. However, two of the three lines are below the level of 3σ detectability significance in this work, and the third one is only slightly above such limit (see Sect. 4). Also in the case of 4U 1636 − 536 Obs 0500350401, we obtained a lower, though not significant, when using the parameters of the line from previous analyses. This is because we did not allow any inclination higher than 70° in our analysis, since the source shows neither dips nor eclipses. In contrast, the value of the inclination in the laor component in previous analyses had a higher value of  > 81° (e.g., Pandel et al. 2008; Cackett et al. 2010). For the observations of Ser X − 1, SAX J1808.4 − 3658, and GX 340 + 0, the increase in the when we use the parameters of the lines in the literature is not statistically significant, since the is already well below 1 in the fits in this work. However, since the is already below 1 for fits with the simple Gaussian model, there is no reason to consider a laor component for these fits.

B.1. Ser X − 1

Bhattacharyya & Strohmayer (2007) first report a skewed Fe Kα line with a moderately extended red wing in an NS LMXB based on the three XMM-Newton observations of Ser X − 1 that we re-analysed in this paper. They fitted the line with a laor component and find EWs for the lines between 86 and 105 eV, inner radii between 4.04 and 16.19 r_g, and an energy centroid of 6.4 keV. On average we found smaller EWs, between 57  ±  18 and 63  ±  12 eV, and larger energy centroids, between 6.60  ±  0.1 and 6.79 keV, when fitting the line with the same component and EWs between 50 and 52 and an average energy centroid of 6.60  ±  0.04 keV when fitting the line with a Gaussian component. We also inferred a smaller average inclination for the source from the laor fit: between  < 25 and 28  ±  9°, compared to between 39.7 and 50.2° found by Bhattacharyya & Strohmayer (2007). The inner radius was not well constrained when fitting the line with a laor component, with a value of 13.3 r_g for the best-constrained observation (Obs 0084020401). In our analysis the fit with the laor component was not statistically preferred to the one with the Gaussian component.

We explain the discrepancies in the results as mainly due to two differences in the analysis. First, while Bhattacharyya & Strohmayer (2007) consider pile-up to be insignificant for these observations, we found that the epatplot shows a significant degree of pile-up before we remove the central 4 columns of the PSF. Figure B.1 shows the difference in the spectra extracted when including the core of the PSF, as done by Bhattacharyya & Strohmayer (2007), and excluding the 4 central columns to remove pile-up effects, as in this paper. The pile-up effects show up as an excess of photons from as low as 3 keV and becomes especially prominent above 5 keV in the red spectrum. This, combined with a improper continuum modelling (see below), may create an excess of photons below 6 keV, which could be misinterpreted as a red wing of the Fe line.

The second difference in the analysis is precisely that we used a different continuum model to fit the spectrum. As shown in Table 6 we got a significantly worse using a combination of diskbb and comptt components to fit the continuum, as done by Bhattacharyya & Strohmayer (2007), when compared to diskbb and bbody components used in this paper.

Therefore, for this particular source, the main contribution to the broad Fe line detected in the first analysis by Bhattacharyya & Strohmayer (2007) is likely related to residuals from the fit to the continuum, showing that modelling the spectra with different continua may significantly modify the parameters of the Fe line.

B.2. GX 340 + 0

D’Aì et al. (2009) report a broad asymmetric emission line in the XMM-Newton spectrum of GX 340 + 0. Owing to the variability of the source as revealed by the light curve (see Fig. 2), they divided the observation into five segments and analysed each segment separately. They modelled the spectral continuum with diskbb and bbody components and find a significant improvement in the fit when using the diskline component to model the Fe Kα emission compared to a Gaussian component. The line had an average energy of 6.69  ±  0.02 keV, inner radius of 13 ± 3 r_g, and EW of 41  ±  3 eV. In spite of using the same continuum to model the spectrum, we obtained equally good fits when modelling the Fe line with a Gaussian or with a laor component. The line had an average energy of 6.82 keV, an inner radius  > 16 r_g, and a smaller EW of 15  ±  5 eV. We inferred an average inclination for the source of  < 34°, compared to 35 ± 1° found by D’Aì et al. (2009). When fitted with a Gaussian profile, the line had an energy of 6.76  ±  0.02 keV and width of 0.24  ±  0.02 keV (D’Aì et al. 2009), consistent with our values of 6.72  ±  0.06 keV and width of 0.17 keV.

The main difference bewteen our analysis and theirs is again the treatment of pile-up. While they only considered the initial  ~ 7 ks of the observation (interval 5) to be affected by pile-up and excised the 2 central columns to remove its effects, we found that the full observation was strongly affected by pile-up so removed the central 8 columns. This is consistent with the high count rate of the source, with peaks up to  ≳ 1100 counts s^-1and a minimum of 650 counts s^-1 along the observation, as well as the hardness of the spectrum (see Sect. 2.1). The light curve in this paper (see Fig. 2) shows a count rate twice as high as the one shown in Fig. 1 from D’Aì et al. (2009). A potential explanation of this difference in the light curves could be that D’Aì et al. (2009) averaged intervals of real data with those for which no data were recorded, because of telemetry saturation.

B.3. GX 349 + 2

Iaria et al. (2009) report several emission features, consistent with the transitions of L-shell Fe xxii-Fe xxiii, S xvi, Ar xviii, Ca xix, and highly ionised Fe in the XMM-Newton spectrum of GX 349 + 2. While the first four features can be fitted equally well with Gaussian features or the relativistic diskline component, they find that the Fe Kα feature is better fitted using the diskline component at 6.76 keV or two diskline components at 6.7 and 6.97 keV. The line had an energy of 6.80 ± 0.02 keV, width (σ) of 0.28 keV, and EW of 49 eV when modelled with a Gaussian component. The same line had an energy of 6.76 ± 0.02 keV, inner radius 6.2, and EW of 61 ± 9 eV when modelled with a diskline component.

In spite of using the same continuum to model the spectrum, we obtained equally good fits when modelling the Fe line with a Gaussian or with a laor component and a significantly larger EW in any of the fits, compared to Iaria et al. (2009). In this paper, the line has an energy of 6.72 ± 0.06 keV, width (σ) of 0.32 keV and EW of 82 eV when modelled with a Gaussian component and energy of 6.94 keV, inner radius of 11, and EW of 96 eV when modelled with a laor component. We obtained a significantly different inclination for the source, 17  ±  9°, compared to 41.4° (Iaria et al. 2009). Apart from the inclination and the EW, the line parameters were not significantly different within the errors. We attribute the differences in some properties of the line to the PSF extraction regions used. While Iaria et al. (2009) exclude only the 2 central pixels of the PSF before spectral extraction, their Fig. 4 shows an excess of double events and a deficit of single events above  ~ 7.5 keV compared to the predicted model. This is a clear indication of residual pile-up in their spectra (see XMM-Newton Users Handbook). In contrast, we found that rejection of the inner 8 columns of the PSF was necessary in order to completely remove the pile-up effects.

We found a lower temperature for the diskbb and bbody components, 0.92  ±  0.03 and 1.57  ±  0.02 keV, compared to 1.05 and 1.792 keV (Iaria et al. 2009). Thus the spectrum extracted in this paper is significantly softer compared to the analysis by Iaria et al. (2009). Again, this is expected if pile-up effects have not been completely removed from the spectra in their analysis.

B.4. 4U 1705 − 44

4U 1705 − 44 was observed twice by XMM-Newton with the EPIC pn camera in timing mode. The first observation had a relatively low flux, 2.5 × 10^-10erg cm^-2 s^-1, and was at least one order of magnitude below the typical count rate where pile-up effects become important. In 2008 XMM-Newton observed 4U 1705 − 44 at a significantly higher flux, 642 counts s^-1 compared to 28 counts s^-1 in 2006 (see Table 1). Di Salvo et al. (2009) report four emission features and one absorption edge in the spectral residuals after modelling the continuum with a bbody and a comptt components, which were identified with emission of S xvi, Ar xviii, Ca xix, and Fe xxv, and (redshifted) absorption from Fe xxv. The lines were broad with σ between 120 and 260 eV. They find a significant improvement in the fit when the Fe xxv feature is modelled with a diskline component compared to a Gaussian component and reported values of 6.66  ±  0.01 keV, 14  ±  2 r_g, 39  ±  1 and 56  ±  2 eV for the energy, inner radius, inclination, and EW for the fit with a diskline component. The absorption edge appeared to be smeared (width of 0.7 keV) and redshifted with an energy of 8.3  ±  0.1 keV with respect to the rest frame energy of 8.83 keV. We found values of 6.45  ±  0.05 keV,  < 9 r_g,  < 34°, and 135 eV for the energy, inner radius, inclination and EW for the fit with a laor component, i.e. significantly different to those reported by Di Salvo et al. (2009). When fitted with a Gaussian component, the centroid was 6.56  ±  0.05 keV and the EW 92 eV.

The difference between both analyses is again related to both the pile-up treatment and the modelling of the continuum. We found that it was necessary to excise the 7 central columns of the PSF to completely remove pile-up effects, while Di Salvo et al. (2009) did not remove any column. In addition, Di Salvo et al. (2009) used a continuum of bbody and comptt to fit the spectrum, while we used bbody and diskbb.

B.5. SAX J1808.4 − 3658

SAX J1808.4 − 3658 was observed by XMM-Newton in 2008 and its spectrum modelled with a combination of diskbb, blackbody, and power law components (Papitto et al. 2009; Patruno et al. 2009; Cackett et al. 2009). The residual emission at the Fe Kα band is modelled with a diskline profile in all the analyses, although Papitto et al. (2009) obtained only a slight improvement in the fit when using a diskline profile compared to a Gaussian profile. The parameters of the diskline component were in general consistent in the three analyses, with some exceptions such as the inclination derived for the source, probably because of small differences in the analyses such as including the RGS in the simultaneous fit or not or considering slightly different energy bands. For example, Papitto et al. (2009) (Cackett et al. 2009) used the 1.4 − 11 keV (1.2 − 11 keV) band for spectral fitting, since they found large residuals below 1.4 (1.2) keV, respectively.

As explained in Sect. 3.2 we extracted two different spectra for this observation. Spectrum 1 was extracted including the core of the PSF, as done by the authors above. Despite using the same extraction region, we found a significantly smaller line, with an EW of 51 eV when fitting the line with a laor component and 23 eV in the fits with a Gaussian component. In contrast, in the previous analyses with a laor component, the EW was found to be as large as 121 eV (Papitto et al. 2009), 118  ±  10 (Cackett et al. 2009), and 97.7  ±  31.4 eV (Patruno et al. 2009).

This may be related to the different continuum used to fit the spectrum. While all the previous analyses of the XMM-Newton observation of SAX J1808.4 − 3658 required 3 continuum components to achieve an acceptable fit, we already obtain a of 0.98 (223) with a continuum consisting of just blackbody and power-law components.

B.6. 4U 1636 − 536

Pandel et al. (2008) find evidence for relativistic lines from Fe in different ionisation states in the three XMM-Newton observations of 4U 1636 − 536 presented here. They report EWs of 215, 98, and 140 eV for the Fe Kα line present in Obs 0303250201, 0500350301, and 0500350401, respectively, which they modelled with a diskline component. In contrast, we found smaller EWs of 130  ±  14, 36, and 6  ±  3 eV when fitting the line with a laor component and of 210  ±  129, 28  ±  19, and 59 eV when using a Gaussian component instead. When fitting the line with a Gaussian component, the lines from Obs 0303250201 and 0500350301 are below the 3σ significance level so should not be considered as detections. Pandel et al. (2008) regard the high inclination,  > 81°, obtained for Obs 0303250201 and 0500350401 as unrealistic (given the values of the inclination of 36 − 74° determined for this source Casares et al. (2006)) and interpreted it as an indication that the excess at the Fe Kα band was a blend of at least two lines. In contrast, we found only a high inclination for Obs 0303250201 and an upper value of 70° (as used for all the fits in this sample) already gave an acceptable fit with a of 0.76 (216) for this observation. We obtained equally good fits when modelling the Fe line with a Gaussian or with a laor component.

We attribute the differences in some properties of the line to the PSF extraction regions used. Pandel et al. (2008) consider that none of the three XMM-Newton observations suffer from pile-up effects based on the count rate limit for the pn timing mode. In contrast we found that rejection of the inner 1(3) columns of the PSF was necessary in order to remove the pile-up effects completely for Obs 0500350301 (0500350401). The residual pile-up effects in their spectra are most likely responsible for the harder spectrum. They report a difference of their spectra with respect to the simultaneous RXTE PCA spectra both of flux and slope. The PCA spectra show a flux excess of  ~30% at 3 keV and  ~10 − 15% at 10 keV with respect to their EPIC pn spectra. Qualitatively, this is what we expect if the XMM-Newton spectra are affected by pile-up: photons will be lost so that in average the flux is lower and in addition soft photons will be counted as hard photons, so that in average the PCA will show a softer spectrum.

Appendix C: Spectra

The best fits of each individual spectrum, together with the residuals of the fits, are shown in Fig. C.1. Figure C.2 shows the unfolded spectra for each observation.

---

Acknowledgments

Based on observations obtained with XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA member states and the USA (NASA). We thank the anonymous referee for helpful comments. We also thank M. Guainazzi and E. Kuulkers for a careful reading of this manuscript and the suggestions that helped to improve it. We acknowledge support from the Faculty of the European Space Astronomy Centre (ESAC).

References

All Tables

All Figures