Astron. Astrophys. 343, 23-32 (1999)

3. X-ray data analysis

NGC 3923 was observed by two Narrow Field Instruments on board the BeppoSAX satellite (Boella et al. 1997a): the Low Energy Concentrator Spectrometer (LECS) and the Medium Energy Concentrator Spectrometer (MECS). The journal of the observation is given in Table 2. LECS and MECS are made of grazing incidence telescopes with position sensitive gas scintillation proportional counters in their focal planes. The MECS, which consists of three equal units, has a field of view of 56 arcmin diameter, works in the range 1.5-10 keV, with an energy resolution of %, and moderate spatial resolution of arcmin FWHM at 6 keV (Boella et al. 1997b). The total effective area of the MECS is comparable to that of the 2 GIS units on board ASCA . The LECS is sensitive also at softer energies (over 0.5-10 keV), has a field of view of 37 arcmin diameter, an energy resolution a factor of better than that of the ROSAT PSPC, but an effective area much lower (between a factor of 6 and 2 lower, going from 0.3 to 1.5 keV; Parmar et al. 1997).

Table 2. Observation log
Notes:
^a On-source net exposure time. The LECS exposure time is considerably reduced with respect to the MECS one, because the LECS can operate only when the spacecraft is not illuminated by the Sun.
^b Background subtracted source count rates, with photon counting statistics errors, within the regions specified in the last column.
^c The extraction region is a circle for the LECS data and an ellipse of semi-axes a and b for the MECS data.

The cleaned and linearized data have been retrieved from the BeppoSAX Science Data Center archive, and later reduced and analysed using the standard software (XSELECT v1.3, FTOOLS v4.0, IRAF-PROS v2.5, and XSPEC v10.0). For the MECS, the event file made by merging the data of the 3 MECS units, properly equalized, has been used.

3.1. Spatial analysis

In Fig. 1 a contour plot of the MECS image is shown. The counts in this image belong to pulse invariant gain-corrected spectral channels 29-218 (1.7-10 keV). The data have been smoothed with a gaussian of . An extended source is clearly visible, of extension larger than the PSF, and covering at least 8 optical effective radii. The X-ray emission shows roughly the same position angle of the optical emission. The background subtracted radial profile obtained from the MECS data over 1.7-10 keV is plotted in Fig. 2.

Fig. 1. The MECS image of NGC 3923, smoothed with a gaussian of . Contours plotted correspond 27, 30, 33, 39, 45, 55, 70, 85, 95% of the peak intensity. The ellipse has the same position angle of the optical image, and indicates the extent of the X-ray image that has been used to extract the spectrum (see Sects. 3.1 and 3.2). The numbers 1, 2, 3, 4 mark respectively the positions of the foreground/background sources 1, 2, NWa and MWb found by BC in the ROSAT image of NGC 3923.

Fig. 2. Background subtracted azimuthally averaged surface brightness profile obtained from the MECS data over keV). Also plotted is the PSF of the MECS for keV (dashed line), and the convolution of the optical profile with the MECS PSFs calculated at the source photon energies (solid line; see Sect. 4.1). One pixel size is .

The X-ray source extent needs to be estimated precisely, in order to determine the size of the region from which to extract the counts for the spectral analysis, and the estimate of the flux. We consider the source boundary to be the point at which the total (i.e., source+background) X-ray surface brightness flattens onto the background level. The background is estimated from blank fields event files ², accumulated on five different pointings of empty fields, and using extraction regions corresponding in size and position to those of the source. In addition to the inspection of various surface brightness profiles (e.g., the azimuthally averaged one, and also those along the major and minor axes), the precise source extent was finally established also through the requirement that within it the S/N of the background-subtracted counts keeps at a high value, not much lower than that of the central regions of the galaxy. This produced an extraction region of an ellipse for the MECS data, of semimajor axis of and semiminor axis of (Table 2). The PSF of the LECS is a strong function of energy, and it is broader than that of the MECS below 1 keV, while it is similar to that of the MECS above 2 keV.

A radius of is suggested to encircle all the photons of soft sources (http://www.sdc.asi.it/software/cookbook). This radius turned out to be optimal in our case, satisfying both the requirement of being the radius at which the brightness profile flattens onto the background level, and that of a high S/N. These different extraction regions for the LECS and the MECS images encircle the same fraction of the source photons, in a given band common to the two instruments, as verified later during the spectral analysis.

Analyzing the ROSAT PSPC image of NGC 3923, BC detected a few foreground/background sources; those falling within the extraction region are displayed in Fig. 1. The nature of these sources is unknown. Sources called "1", "NWa", and "NWb" by BC cannot be distinguished in the MECS image, while this image looks quite aligned to the North-South due to source "2". By deriving the net counts in each one of four quadrants, obtained by dividing the ellipse in Fig. 1 with its major and minor axes, it turns out that the quadrant comprising sources "1","2", and "NWb" has roughly 6020 net counts more than the average of the others. This is a contribution of % to the MECS net counts. The quadrant with source "NWa" shows no significant net counts enhancement. By inspecting the surface brightness profile in a stripe across source "2", a contribution of net counts is estimated from this source; this is likely to be responsible for the jump at a radius of in the azimuthally averaged surface brightness profile of Fig. 2. Further implications coming from the presence of these sources for the spectral analysis are discussed in Sect. 4.

3.2. Spectral analysis

Within the source regions determined as described above, we obtained the net count rates given in Table 2, and finally extracted the LECS and MECS spectra. Spectral channels covering the energy ranges 0.3-10 keV, and 1.7-10 keV respectively for the LECS and MECS have been used. The original channels have been grouped into larger bins, adequately filled for applicability of the statistic to assess the goodness of a fit (Cash 1979). The data have been compared to models, convolved with the instrumental and mirror responses, using the minimization method. The spectral response matrices released in September 1997 have been used in the fitting process. We fitted the models simultaneously to the LECS and MECS spectral data.

Given the discussion of Sect. 1, the basic models used for the fits are the bremsstrahlung model, and two models describing the thermal emission of an optically thin hot plasma, both from continuum and lines: the standard Raymond-Smith model (hereafter R-S, Raymond & Smith 1977), and the MEKAL model. The latter is a modification of the original MEKA model (Kaastra & Mewe 1993) where the Fe-L shell transitions have been recalculated recently according to the prescriptions of Liedahl et al. (1995). The abundance ratios of the heavy elements cannot be constrained by the modeling, because of the poor statistics. So, we assume that the abundance ratios are solar, as accurately determined by Matsushita et al. (1998) for the ISM of the X-ray bright galaxy NGC 4636, using a very long ASCA exposure. The solar abundances are those given by Feldman (1992; e.g., the abundance of iron relative to hydrogen is 3.24 by number, also known as meteoritic abundance). The models are modified by photoelectric absorption of the X-ray photons due to intervening cold gas along the line of sight, of column density ; we take absorption cross sections according to Baluciska-Church & McCammon (1992). The results of the spectral analysis are presented in Tables 3 and 4.

Table 3. One-component spectral fits
Notes:
^a When in this column cm^-2, it has been kept at this value (the Galactic one) during the fit. When is a free parameter of the fit, it is constrained to lie in the interval cm^-2, and its "best fit" value always turns out to be the lower boundary of the interval.
^b Number of degrees of freedom in the fit.
^c In this case kT in the third column gives the maximum temperature from which the gas is cooling.
The ranges of values between parentheses indicate the 90% confidence range of variation for one interesting parameter.

Table 4. Two-component spectral fits

3.2.1. One-component models

We first tried to fit one-component models to the data (Table 3). The probabilities of exceeding the reduced are quite low (they range from 0.1 to 0.2) but the fits are formally acceptable. At the best fit tends to be lower than the Galactic value; the fits are not worsened significantly when keeping fixed at the Galactic value. All the one-component models give fits of comparable quality, and a temperature around 3 keV.

The abundance at the best fit is extremely low, both for the MEKAL and the R-S models, but it is just constrained to be by the data. So, we have also performed some fits with the abundance fixed at the solar value. These findings are in partial agreement with those obtained from ASCA -SIS data over (0.5-5) keV by BF: when keeping fixed at the Galactic value, the single-component MEKAL fit gives them an abundance close to that found by us, but a temperature of only keV. This could reflect the different extraction regions used for the spectrum, if the emission is much softer at the center; or could be the result of different spectral sensitivities between the BeppoSAX -MECS and the ASCA -SIS, from which the best data come in the cases of the two satellites. The spectral models derived from ASCA data are mostly constrained by the SIS data near 1 keV (BF), while those derived from BeppoSAX are mostly constrained by the MECS data at energies above that region.

3.2.2. Multi-component models

We then tried to fit multi-temperature models, to check whether the quality of the fits can be improved. Since the R-S model gives results substantially equal to those given by the MEKAL model, for these data, we give in the following only the results obtained by using the more updated MEKAL model.

First we tried the fit with a cooling flow model, in which the cooling is isobaric, and the emissivity is as described in Johnstone et al. (1992); this model is made by the superposition of many thermal components, each described by a MEKAL model at a certain temperature. The results are given in Table 3; the fits are improved (the associated probability is 0.4-0.5). The upper temperature from which the gas cools is very high, keV; this is also the temperature at which the emission measure peaks (the emission-weighted temperature is somewhat lower). This modeling reveals that hard emission is present at a significant level, so that the distribution of the temperatures is forced to extend to high values ("high" for the temperatures expected in a galaxy cooling flow). The upper temperature values found here are higher than that given by BF ( keV) for a fit with the same model, with fixed at the Galactic value, and resulting from the fit. The reasons for this discrepancy are likely the same given above for the one-component case.

In agreement with BF we find that the abundance at the best fit (0.6 ) is raised with respect to that obtained from the one component MEKAL fit.

Then we tried the coupling of two thermal components (MEKAL+bremsstrahlung, and MEKAL+MEKAL) with the same (Table 4). The quality of the fits is much improved now, and the probability of exceeding is very large (). As before, allowing to be free does not significantly improve the fits; when Z is fixed at 1 , at the best fit is reasonably close to the Galactic value. In Fig. 3 we show the LECS and MECS data together with the two-component model which gives the best fit. More in detail, the results obtained by fitting with two thermal components are as follows:

a) MEKAL + bremsstrahlung models: the first turns out to describe a soft component, of temperature 0.4 keV, and the second describes a hard component of temperature   between 6.2 and 7.6 keV.   is higher for a lower metallicity of the soft component. The abundance of the soft component is again very subsolar at the best fit (0.16 ), and not constrained by the data.

b) Two MEKAL components: the results are similar. The abundance of the hard component turns out to be practically zero at the best fit; that of the soft component is also very low (0.1 ), but again Z is just constrained to be . At the best fit the temperatures are   around 0.4 keV and   around 5 or 8 keV, depending on the abundance, fixed at the solar value or allowed to vary freely respectively. As before the temperature of the hard component is higher for a lower metallicity, but the effect is more pronounced now. In this kind of fit BF find temperatures close to ours (=0.55 keV, and =4.2 keV), but ( at 90% confidence), with kept at the Galactic value ³.

Fig. 3. The LECS and MECS spectral data with the best fit MEKAL + bremsstrahlung model with solar abundance and fixed at the Galactic value, shown with thinner lines, and the residuals from the fit.

Summarizing, all the two-component models give   around 0.4 keV, and   around 6-8 keV, except when the hard component is described by a MEKAL model with solar abundance (then keV). The latter case is also the one giving the worst fit among all the two-component models (). Heavy element abundances at the best fit tend to be very subsolar, except for the cooling flow model. The effect of imposing an abundance higher than that at the best fit is to produce a lower temperature of the hard component. This effect is particularly pronounced when fitting with two components (respectively   decreases of to keV, using the bremsstrahlung and the MEKAL model for the hard component).

3.3. X-ray fluxes and luminosities

In Table 4 the absorbed fluxes in the (0.5-4.5) and (0.5-10) keV bands are given; unabsorbed fluxes are about 30% higher for the soft component, and 5 to 9% higher for the hard one, depending on the band. An average best fit value of the total absorbed flux keV) is erg cm^-2 s^{- 1}, and keV) erg cm^-2 s^{- 1}. The ratio between the absorbed fluxes in the soft and in the hard components in the (0.5-4.5) keV band is , close to the value of 0.99 found by BF in the (0.5-5) keV band. This ratio actually is when Z is kept at (and keV), and when ( keV). An average value for the ratio between the absorbed fluxes in the soft and in the hard component in the (0.5-10) keV band is . Again it is lower (i.e., ) when Z is kept at , and higher () when Z is very subsolar. In the (0.5-4.5) keV band, the soft component unabsorbed flux is 52% of the total, while the hard component amounts to % of the total (0.5-10) keV unabsorbed flux.

Adopting the distance in Table 1, the fluxes can be converted into the following values of the luminosities in the (0.5-4.5) keV band: an average value of the soft component absorbed luminosity is erg s^-1 (unabsorbed erg s^-1), and of the hard component is erg s^-1 (unabsorbed erg s^-1).

© European Southern Observatory (ESO) 1999

Online publication: March 1, 1999
helpdesk.link@springer.de