3. Point sources
3.1. Source searching in the presence of diffuse emission
For M 82, source searching must take into account the presence of a spatially variable high surface brightness diffuse background due to the wind. This affects both the PSPC and the HRI. Failure to incorporate this additional background leads to the detection of large number of low significance sources within the diffuse emission, whilst simply increasing the threshold significance leads to the non-detection of what are clearly real sources in regions free of diffuse emission.
As a result of this we employ an iterative procedure which starts by using a smoothed image (including all sources and diffuse emission) as the background for the source searching procedure. Sources detected at are then excised from the image, out to the enclosed energy radius ( in the HRI, in the PSPC), and the resulting holes interpolated over. This dataset is then smoothed to provide a second approximation to the background. The 80% radius, to which sources are removed, is a compromise between removing so large a radius that the interpolation is unreliable, and too small a radius, leaving significant source contamination in the background. The above procedure of source searching followed by background estimation, was repeated until there was no further change in the list of detected sources. In practice this required three cycles. The results of the method depend on the smoothing scale employed when estimating the background, and and were found to give best results for the HRI and PSPC, respectively. The final combined sourcelist is shown in Table 2.
Table 2. PSPC and HRI detected sources. Positional errors are at 90% confidence, flux errors are 1 . Where a source is detected both in the PSPC and the HRI the HRI determined position is given.
3.2. Source spectra
Individual exposure-corrected spectra centred on the positions given in Table 2 were obtained for each PSPC source from the data cube within a radius corresponding to a 95% enclosed energy fraction at an energy of (an appropriate energy for QSO's which form the majority of background sources). Raymond & Smith (1977) and power law models were fitted to these spectra. Standard fitting is inappropriate, due to the low numbers of counts per bin; we therefore used a maximum-likelihood fit. For each source, the spectrum predicted from the spectral model is added to an estimated background spectrum derived from the background model cube discussed in Sect. 2.1. The resulting total source plus background spectrum is fitted to the observed spectral data using the Cash C-statistic (Cash 1979).
Table 3 gives the results of the spectral fits. As maximum likelihood does not provide a goodness-of-fit measure akin to , it is difficult to assess how good the fits are, except by visual inspection and comparing the fitted parameters with typical values for QSO's and stars. Source 8 (the nucleus) is strong enough that significant systematic discrepancies between the data and the fitted models are apparent, and is dealt with separately in Sect. 3.4.
Table 3. PSPC source properties. Error bounds are 1 . The best-fit parameters for a Raymond-Smith hot plasma model and a power law model are given.
3.3. Comparison between PSPC and HRI sources
Given the presence of several possible point sources within or in close proximity to the diffuse emission, the use of the HRI can clarify whether these are true point sources or not. As discussed in Sect. 2.2the presence of strong diffuse emission complicates source searching, which may lead to the detection of spurious sources within the diffuse emission.
Only five sources were detected independently in both the PSPC and the HRI. We have searched for counterparts to these objects at other wavelengths in the SIMBAD database. One (Source 8) is M 82's nuclear source, another (Source 13) a QSO. The other three do not have any counterparts.
For the remaining PSPC sources not detected in the HRI, we found upper limits for the HRI flux due to any point source within a region defined by the PSPC positional uncertainty. These, together with predicted HRI count rates from the spectral fits to the PSPC sources, are given in Table 4. There is little difference between the predicted fluxes for the power law and Raymond & Smith models. For the sources detected in both PSPC and HRI, the observed HRI count rates agree well with those predicted, except for the nucleus, which is discussed below. In most other cases the upper limits are greater than (i.e. consistent with) the predicted count rates. For the two possible sources within the northern wind, sources 3 & 6, the predicted count rates are higher than the HRI upper limits. This could mean that these are not true point sources, merely bright lumps in the wind. Note, however, that source 6 appears to be significantly cooler than the temperature of the wind emission in region n7, where it is centred. Alternatively, the predicted count rate could be overestimating the flux, due to the addition of diffuse flux along with real source flux. Variability is another possible explanation for the HRI non-detection.
Table 4. HRI count rates and upper limits for the PSPC detected sources. The predicted HRI count rate was calculated as described in the text. Only five PSPC sources are directly detected in the HRI, the remaining sources have upper limits quoted. The range in predicted flux corresponds to using either the power law or Raymond-Smith fit.
3.4. The nucleus
As the HRI observations have shown (Collura et al. 1994; Bregman et al. 1995), the nucleus of M 82 contains a very luminous, variable X-ray source, along with non-uniform diffuse emission. It is little surprise that the single component fits to the nuclear source in the PSPC data (source 8 in Table 2) are of poor quality. Given the large number of counts in the spectrum () we can use standard fitting. The power law from Table 2 has a reduced of 7 with 19 degrees of freedom. An additional problem is the finite radius of within which the spectrum was extracted. Although the 95% enclosed energy radius at is good for almost all sources, the brightness of the nucleus, coupled with its hardness, means that significant flux is scattered outside this radius. At energies above this radius only encloses of the flux. Given the large number of photons involved, this is likely to have a significant effect on any fit.
Fitting the data from within a larger (, giving % enclosed energy at , counts) radius from the nucleus, we achieve a best-fit reduced of 2.08 with 15 degrees of freedom (see Table 5) using a two component soft Raymond & Smith plus harder bremsstrahlung model. The fitted temperature of the hard component is outside ROSAT 's energy range, and so should only be interpreted as being being significantly hotter than 6.2 keV . The spectrum can also be fitted by a two component power law plus Raymond & Smith model, however the fit requires the power law to have a lower column than the hot plasma component. This is not physically sensible if the hard component represents nuclear emission, so we rejected this model in favour of the Raymond & Smith plus bremsstrahlung model.
Table 5. Best-fit parameters (reduced ) for the nuclear source. Luminosities are quoted in the ROSAT band (), for a distance of (Freedman et al. 1994) to M 82. The luminosity escaping M 82 is corrected for absorption in our own galaxy. The intrinsic source luminosities are corrected to zero absorption. This is the lower bound. The temperature is unbounded above this value.
The predicted HRI count rate for the hard bremsstrahlung component derived above is cts , compared to the observed value of cts . The thin, hot component of the wind is predicted to provide very little emission in the ROSAT band (Suchkov et al. 1994), and there are no other strong point sources seen by the HRI. Since the HRI source associated with the nucleus is known to be variable (Collura et al. 1994) by a factor of several within the HRI observations, the difference between PSPC and HRI fluxes is most likely due to such variability.
The first HRI observation (rh600021, see Table 1), is divided into two blocks which are interleaved with the PSPC observation. The short () initial HRI pointing showed the source intensity at a level that of the remaining HRI data, taken days later. The first block of PSPC data falls in this 40 day gap, with the remainder commencing days later. We have searched for variation of the nuclear source within the PSPC observation, which is broken into three main blocks separated by long gaps, but observe no significant variation between the blocks. In the second HRI observation, taken a year later, the nuclear count rate decreases steadily over a period of six days from the previous HRI level to about half that (Collura et al. 1994).
The intrinsic nuclear point source luminosity in the ROSAT band of (Table 5) is significantly higher than the ROSAT HRI estimate of . This estimate assumed a Raymond-Smith model with and . With the higher temperature and absorption column inferred from the PSPC spectral fit, the HRI luminosity would increase, although the inferred PSPC luminosity remains several times higher. This PSPC luminosity corresponds to the Eddington luminosity for a object. The position of this source corresponds (to within ROSAT pointing accuracy) to the position of a strong radio source (41.5+597) which, on the basis of a 100% drop in flux within a year, is unlikely to be a supernova remnant (Muxlow et al. 1994). A high surface brightness complex is seen in the optical (region E in O'Connell et al. 1995) at this position. This is an unusual object deserving more study.
© European Southern Observatory (ESO) 1997
Online publication: June 30, 1998