Spectral variability

With data of this quality, it is obviously interesting to search for spectral variability on the shortest timescales possible. Experimentation shows that an adequate spectrum requires an exposure of 10ksec of data. In this section, we analyze spectral variability across eleven uniformly spaced 10ksec segments of our observation. The median ``live=time'' for each of these segments is about 7ksec.

Figure 6: The filled and open circles show two representative difference spectra (for the 80-90ksec and 100-110ksec segments, respectively) ratioed against the best-fitting power-law model. The 10-20keV segment has been used as our low-state spectrum when forming these difference spectra, although very similar results are obtained when we use the 20-30keV data instead. Also shown is the time-average spectrum ratioed against the best-fitting power-law model (open squares). See Section 4 for a detailed discussion.

Of particular interest, of course, is any variability of the iron line profile. We begin our investigation of iron line variability by examining ``difference spectra'', following the work of Fabian et al. (2002) and Fabian & Vaughan (2003). In detail, we isolate and examine the variable part of the X-ray spectrum by subtracting the lowest flux spectrum from the other spectra. Since we are primarily interested in iron line variability, we restrict our attention to the 2-10keV region of the EPIC-pn spectrum. Figure 6 shows two representative difference spectra (for the 80-90ksec and 100-110ksec segments), using the 10-20ksec segment of data as our representative lowest-state spectrum; the broad spectral feature that we interpret as reflection from a relativistic disk can be seen in both of these difference spectra. In fact, 8 of the 10 difference spectra show evidence for the very broad disk feature, with the remaining two spectra being too noisy to draw any conclusions. Furthermore, the narrow iron line does not appear in the difference spectra. In other words, the narrow iron line has a constant absolute flux, as expected if it originates from distant material.

Figure 7: Result of fitting a simple absorbed power-law plus laor component to the 10ksec segments of data. Panel (a) shows the absolute intensity of the laor component as a function of 2-10keV flux. Note the apparent correlation of line intensity with continuum flux. Panel (b) shows the equivalent width of the laor component as a function of 2-10keV flux. It can be seen that these data are consistent with a constant equivalent width. Error bars are shown at the 90% level for one significant parameter ($\Delta \chi ^2=2.71$).

Although they are rather noisy, the difference spectra suggest that (apart from the narrow iron line) the 2-10keV EPIC-pn spectrum maintains the same overall shape during large changes in source flux once small variations in the underlying power-law index have been taken into account. In other words, the difference spectra imply that the broadened reflection features maintain a constant equivalent width relative to the underlying continuum, as opposed to a constant absolute intensity.

To investigate this further, we fit each spectrum with a power-law modified by Galactic absorption plus a simple broad iron line described by the laor model in XSPEC. These fits are much simpler than those discussed in Section 3 since we do not include the reflected X-ray continuum. We fix all parameters of the laor component apart from its intensity at the values derived by fitting this model to the time-averaged spectrum; $r_{\rm in}=1.24r_{\mathrm g}$, $r_{\rm out}=400r_{\mathrm g}$, $\beta=6.4$, $i=48^\circ$. Of course, the spectral feature of interest is a combination of both a broad iron line and the rather complex ionized reflection continuum -- however, the simple broad line model allows us to measure a robust intensity for this feature as a whole. As shown in Fig. 7, there is indeed a correlation between the 2-10keV continuum flux and the intensity of the broad disk feature such as to keep an approximately constant equivalent width.

Both the difference spectra and the direct spectral fitting shows that the equivalent width, not the absolute intensity, of the broad disk feature remains approximately constant throughout this observation. This is the behaviour expected within the disk reflection paradigm, but is at odds with the findings of previous investigations. We discuss this discrepancy and its possible resolution in Section 5.

Figure 8: 2-10keV spectra from the 10ksec segments of data (open squares) ratioed against a simple power-law. The photon index of the power-law has been fixed to that derived from fitting to the time-averaged spectrum, $\Gamma =1.80$. No significant changes in the profile of the disk-reflection component is found.

It is also interesting to search for stochastic changes in the iron line profiles that are uncorrelated with the overall source flux. In Fig. 8 we show both the instantaneous 2-10keV spectrum and the time-averaged 2-10keV spectrum, ratioed against a simple power-law modified by Galactic absorption. While there are hints of numerous features popping in and out of the line profile (especially at $\sim 4{\rm\thinspace keV}$), they are not statistically significant at the 90 per cent level and hence will not be discussed further. In fact, we find no gross changes in the velocity profile of the X-ray reflection features.