Astron. Astrophys. 331, 925-933 (1998)

3. Results

3.1. Temporal behaviour

The light curve of RX J0947.0+4721, shown in Fig. 4, contains data from all OBIs. The largest amplitude of variability between two unobscured detections is a factor of . However, this is not the maximum amplitude observed. Since there is no technical reason for the non-detection in OBI 700165-1, a genuine change of a factor must have occured.

Fig. 4. Light curve of RX J0947.0+4721. Broad band count rate is plotted against time in spacecraft seconds. Each entry represents one OBI. Filled diamonds: good OBIs, open triangles: obscured OBIs, arrows: upper limits for non-detections, dots: background.

Besides these large variations, the count rate has been seen to drop by a factor of 2.7 in 25 hours as well as a factor of 1.6 in 30 hours. There are also less significant indications of more rapid variability. Unfortunately, some of the relevant OBIs are obscured. The count rates derived from these data can be underestimated, so that the variations are of questionable significance.

It has to be checked whether the changes are pure intensity variations or spectral variations as well. Since the OBIs contain generally too few hard photons to perform a meaningful spectral fit, the hardness ratio is used instead, which is defined as

with B, H and S being the counts in the broad, hard (0.4-1.0 keV) and soft (0.1-0.4 keV) band, respectively. The error in HR is then

There could be some dependence of the hardness ratio on the position in the FOV: at large off-axis angles, parts of the soft PSF might fall outside the detector and the source appears harder. The available data did not show any correlation between HR and , regardless of whether the OBIs are partially shadowed or not. Separate unweighted averages for obscured and unobscured OBIs yield and , respectively, the overall average is .

Fig. 5 shows HR vs. count rate. Three OBIs have been left out, the two upper limits and 800102-7, where no hard source counts were detected. No trend of HR with can be seen. A fit to all data (39 OBIs) gives with and a likelihood that this value of will be exceeded by chance of . Fits to the good (14) and obscured (25) OBIs separately give , , and , , , respectively. The values for constant HR are 9.50, 4.14 and 4.67 for all, good, and obscured OBIs. In all cases, the slopes are compatible with zero, and the differences in are too small to favour the straight line fit. The observed changes are therefore considered to be pure intensity variations.

Fig. 5. Hardness ratio HR versus broad band count rate. OBIs without valid HR are left out. Open symbols: obscured data; solid line: average HR; dotted line: straight line fit to all data

3.2. Spectral behaviour

Since no changes in HR could be detected, all data were merged into one spectrum () before conducting spectral fits. The spectrum was binned with roughly the same relative error in each bin below 1.0 keV, yielding 14 bins. Above 1.0 keV, only source photons are detected; these are filled into one single bin with a relative error of 14%. Merging, binning, and fitting was done using the EXSAS software package (Zimmermann et al. 1993); no additional errors were added.

First, several single-component models were applied to the 0.1-1.0 keV range. Table 3 lists the results of the fits, and Fig. 6a-c shows

Table 3. This Table lists the results of fitting several models to RX J0947.0+4721. Fluxes are observed values, kT is given in the quasar's rest frame.

Fig. 6a-c. contour plots (, and confidence level) for the single component fits presented in Table 3. The indicated spectral parameter is plotted against (in ). a single power law; b thermal bremsstrahlung; c blackbody.

the error ellipses. A single power law with cold absorption gives an unacceptable fit when is fixed at the Galactic value. With free, the value obtained for the column density is more than twice the Galactic value. Thermal bremsstrahlung with fixed Galactic absorption yields a rather poor fit with and . Treating as a free parameter improves the fit, but again results in . An absorbed blackbody model gives acceptable fits for both fixed and free. The Galactic is higher than the best fit value, but still within the confidence interval.

When the high energy tail (1.0-2.4 keV) is included, both thermal models underestimate the flux above 1 keV significantly. A single power law fits the data, but again the fitted is more than twice the Galactic value.

We then tried a two-component fit of the total ROSAT band with Galactic absorption, the soft component modeled by a blackbody and the hard by a power law. It might be thought that the hard component is mainly a result of contamination by RX J0947.1+4721 which is at least partly inside the extraction area of RX J0947.0+4721. To check this, we fixed power law index and flux at the values estimated for that source. The fit gives unacceptable results by underestimating the high energy tail of the spectrum, thus hinting at a noticeable hard component of RX J0947.0+4721 itself. We decided to fix the power law index at the 'canonical' value and fit only the two normalizations and the blackbody temperature because the statistics above 1 keV are too poor to fit as well. Only of the source photons are detected above 1 keV, and if these are filled into more than one spectral bin, the relative error per bin is above 25%, too high to give reasonable constraints on . The fit result is presented in Fig. 7, and the parameters are listed in Table 3.

Fig. 7. Fit of an absorbed blackbody plus power law model to the average spectrum of RX J0947.0+4721. The dotted and dashed lines mark the blackbody and power law components, respectively.

With this model, the average flux is if all data are averaged, and if only the unobscured OBIs are used. As was the case for the hardness ratios, no significant change in the spectral parameters can be seen when only the unobscured OBIs are used for the fits. The contribution of RX J0947.1+4721 to the overall averaged flux will be about at worst (Sect. 2.1).

With H₀ and q₀ as before, the fitted flux corresponds to a good (overall) average luminosity () in the QSO's rest frame (0.15-3.70 keV). The detected maximum and minimum luminosities, in this model, are and , respectively, and the upper limit corresponds to . We applied no K-correction to avoid including parts of the spectrum which we did not observe, and to avoid a shift of the band onto the Rayleigh-Jeans part of the blackbody component of the spectrum. The values for and have to be taken with caution, because they are strongly model dependent.

So far, no indication for spectral variability could be found. As a further test, we created `high count rate' and `low count rate' spectra which were binned as described before, and then divided (fig. 8). The separating count rate, , was chosen to give roughly the same number of counts in both spectra. Spectral changes should be visible as a deviation of the ratio from a constant.

Fig. 8. Ratio of `high count rate' to `low count rate' spectra (good data). Solid line: average value, dotted lines: average standard deviation.

No such deviation can be seen. The data are consistent with a constant value ( 13 d.o.f.; ). If all data are used, or if a different binning is applied, the results are similar. Separate fits to high and low spectra give spectral parameters in good agreement with the ones in Table 3. Again, the conclusion is that no significant spectral changes are seen.

© European Southern Observatory (ESO) 1998

Online publication: March 3, 1998
helpdesk.link@springer.de