Astron. Astrophys. 338, 781-794 (1998)

4. Spectral analysis and constraints on N_H

4.1. Continuum fitting

We used three models to fit the observed continuum.

1. Transmission model (Compton thin). It consists of a power law (similar to that observed in Sy1s) transmitted through an absorbing cold medium.

2. Warm scattering model (Compton thick). This model assumes that the primary power law is completely absorbed along our line of sight by a medium that is thick to Compton scattering. However, the primary radiation is scattered into our line of sight by a warm, highly ionized gas located outside the absorbing medium. As a consequence, this model consists of a power law with no absorption in excess of the Galactic value.

3. Cold reflection model (Compton thick). As in model 2, here the primary radiation is completely absorbed along our line of sight by Compton thick material. However, in this model the primary radiation is Compton reflected into our line of sight by a cold, neutral material in the circumnuclear region, possibly identified with the same torus responsible for the obscuration. In this case, the spectral model consists of the "hump" characteristic of the cold reflection. In the 2-10 keV range, the continuum is much flatter than the primary radiation.

Since the sources in our sample are weak, several of our spectra have low signal-to-noise ratio in the continuum. Therefore, we reduced to the minimum possible the number of free parameters. In particular, we froze the photon index of the continuum to 1.7 in all models. This is the average photon index in Sy1 galaxies and it is thought to arise from a primary power law with photon index 1.9 flattened, in the 2-20 keV range, by a cold reflection component due to reprocessing by the accretion disk (Nandra & Pounds 1994; Nandra et al. 1997).

However, to allow for a comparison with other studies in the literature where is a free parameter, in the Appendix we also report results of transmission model fits with this additional degree of freedom.

In most cases the extrapolation of the continuum model fitted at high energies ( keV) falls short of accounting for the emission in the soft X-ray range ( keV). Therefore, we also introduced a black body component (kT a few 0.1 keV) to account for this "soft excess". The black body is just an analitical form that fits nicely the spectrum below keV, but it does not necessary reflect the real nature of this flux; this will be discussed in Sect. 5.4.

However, only data above 3 keV were used to statistically discriminate between the transmission, warm scattering and cold reflection models, so that the were little affected by the soft excess.

In all models, Galactic photoelectric absorption is included.

4.2. The Fe line

The hard X-ray spectrum of Seyfert galaxies is usually characterized by a fluorescent line at 6.4 keV produced by low ionization iron (i.e. iron with no vacancies in the L shell, and then less ionized than Fe XVII ). Also, in some cases (e.g. NGC 1068, Ueno et al. 1994) recombination/resonant emission from He-like (6.7 keV) and H-like iron (6.96 keV) is observed.

Our fit includes an unresolved gaussian, whose central energy is left free, to account for the iron line. In some cases there are indications that the iron line is resolved; in these cases we tried to fit the line with two gaussians centered at 6.4+6.7 keV (neutral and He-like iron) or at 6.4+6.96 keV (neutral and H-like iron). The two-components fit was considered significant only if the probability improved at a confidence level of at least 90% .

The equivalent width (EW) of the fluorescence 6.4 keV Fe K line provides useful information to constrain the column density that absorbs the continuum and, in some cases, to distinguish between the Compton thick and the Compton thin cases. This fluorescence line is thought to arise from the accretion disk (Lightman & White 1988; Fabian et al. 1989) and, possibly, from the torus as well (Ghisellini et al. 1994), with EW(Fe K) 200-300 eV. Cold absorbing column densities N_H 10²³cm^-2 do not change significantly the observed EW(Fe K), since the photoelectric cutoff occurs at energies lower than 6 keV. The presence of an absorbing medium with N_H 10²³ depresses the continuum beneath the iron lines and, if the line is produced by material more extended than the obscuring medium, the EW increases. Moreover, line photons are produced in the absorber itself, whose EW may be significant provided that the covering factor is large enough. EW(Fe K) 1 keV is characteristic of purely reflected/scattered spectra and, therefore, indicates complete absorption of the primary radiation, and then is used to identify Compton thick candidates. Interpreting EW(Fe K) 1keV with a Compton thin, low absorption model would require an iron gas abundance several times higher than the cosmic value (e.g. Matt et al. 1997a).

The K lines from He- and H-like iron (6.7 and 6.96 keV respectively) are emitted from highly ionized gas. Their EW with respect to the scattered component can be as high as a few keV (Matt et al. 1996), while, if the primary radiation is visible, they are likely to be diluted to invisibility. Therefore, detection of lines with 1 keV is a clue that the primary radiation is heavily absorbed, i.e. N_H 10²⁴cm^-2. However, determining the properties of the obscuring material and of the scattering medium from the EW of these high ionization lines presents more uncertainties than the 6.4 keV line. Indeed, the intensity of the He- and H- lines depends non linearly from the optical thickness of the scattering gas. Also, starburst activity can increase the intensity of these lines. On the other hand cases where the 6.7-6.96 keV lines dominate are rare: in most Sy2s the 6-7 keV region is dominated by the line at 6.4 keV (Turner et al. 1997a, also this paper Sect. 4.4).

Finally, we should mention that, both in terms of EW (Fe K) and continuum shape, objects obscured by a column density N_H 10²⁵cm^-2 look like purely reflected sources in the 2-10 keV range (ASCA); however, their Compton thickness, even if larger than unity, is still small enough to permit transmission of a significant fraction of photons in the 10-100 keV range. So far, only NGC4945 and Mkn3 are known to belong to this class (Done et al. 1996, Turner et al. 1997b). Such class of objects would be readily recognized in the 10-100 keV range of BeppoSAX data (PDS).

4.3. Other nuclear absorption indicators

Various authors have also used other methods to determine the nuclear absorption along our line of sight and, specifically, to distinguish between Compton thin and Compton thick sources. By assuming the unified model correct, the ratio between the hard X-ray luminosity and an isotropic indicator of the intrinsic luminosity should provide indications on the amount of absorption affecting the nuclear X-ray source. The luminosity of the [OIII] line can be considered an isotropic indicator of the nuclear intrinsic luminosity, though caveats discussed in Sect. 2must be taken into account (in particular the [OIII] luminosity must be corrected for the NLR extinction as deduced from the Balmer decrement). The effect of a high absorbing column density is to lower the ratio with respec to Sy1s. The reduction is at most by a factor of when N_H a few times cm^-2, and by about two orders of magnitude when N_H 10²⁴cm^-2.

Mulchaey et al. (1994) and Alonso-Herrero et al. 1997 used the ratio to identify absorption effects in Sy2s. They do not find significant differences between Sy2s and Sy1s (except for NGC1068). However, their Sy2 sample is seriously biased toward X-ray bright sources (see Sect. 1and Sect. 2), hence little absorbed; also, they do not correct for extinction. Turner et al. (1997b) adopt this method on a better selected sample (by including weaker sources with respect to the Mulchaey et al. sample). As a result, they identify some Sy2s that are suspected to be reflection dominated, based on a much lower than observed in Sy1s. We will use the ratio (where is corrected for extinction in the narrow line region) as an aid to identify heavily absorbed sources. The distribution of for the sources in our sample, compared to a sample of Sy1s, is shown in Fig. 3 ([OIII] data are from Dadina et al. 1998). A more thorough analysis of the ratio for a large sample of Sy2s, including the ones presented in this paper, is discussed in Bassani et al. (in prep.).

Mulchaey 1994, Mas-Hesse et al. 1995 and Awaki (1991) use also the Far-IR (FIR) emission (as deduced from the 60 and 100 IRAS data) as isotropic indicator of the intrinsic nuclear luminosity, hence the as indicator of nuclear absorption. However, in these low luminosity AGNs the FIR luminosity is often dominated by star formation in the host galaxy (Maiolino et al. 1995). Therefore, we do not consider the FIR emission a reliable indicator of the AGN luminosity.

Rapid variability (on scales of a few 10 ksec) of the X-ray continuum is indicative that the observed radiation is primary emission seen directly and not reprocessed by pc-scale reflecting media, i.e. that the source is Compton thin. Unfortunately, as discussed in Sect. 3, our data do not provide good constraints on the short term variability. Yet, the lack of significant long term variability (i.e. on a time scale of a few years) is in favor of a reprocessed origin of the observed emission.

4.4. Results on single objects

In this section we describe the results of spectral fits for the objects in our sample. Also, we apply the considerations discussed in the former sections to each single object.

Table 2 shows the parameters of the three fitting models for each of the sources in our sample. Bold face entries (also marked with an asterix) indicate sources for which the corresponding model provides the best fit, according to the considerations discussed below. Instead, spectral fits that are inconsistent with the data at a high significance level are not reported. The second column (kT) indicates the temperature of the black body used to fit the soft excess. As for the Fe line we report line energy, normalization and equivalent width. In the transmission models we also report the EW(Fe) once the underlying continuum is corrected for the absorbing column density N_H, listed in column 3 (EW_corr): this value provides a lower limit to the absorption corrected EW(Fe), since it does not take into account the absorption affecting the Fe line itself; if this lower limit turns out to be higher than about 500 eV (i.e. significantly higher than the eV observed in Sy1s) this would indicate that the Compton thin model is little plausible on physical grounds. All errors are at the 90% confidence level for one interesting parameter. The last column indicates the /degrees of freedom for the whole fit, i.e. including the low energy data and the black body component. However, when discussing various spectral models in the following sections, we will often refer to differences relative to the high energy (3keV) data alone.

Table 2. Spectral fits
Notes:
In these models the photon index of the primary radiation is frozen to 1.7. Bold face entries (also marked with an asterix) indicate sources for which the model provides the best fit, both in terms of and in terms of EW(Fe) and properties (see text). Spectral models that are inconsistent with the data at a high significance level are not reported. MCG-05-18-002 was not fitted with any of these models because of the low signal-to-noise; for this source we only report a simple power law fit in Table A1. The cold reflection spectrum was modeled with the XSPEC routine PEXRAV.
a) In the case of NGC1386 the Compton thick model constrains only N_H 10²⁴cm^-2 (see text);
b) rest frame;
c) frozen parameter.

Figs. 1 and 2 show the data along with the folded best fit models, the residuals from the model and the unfolded models. Hereafter we discuss each object individually.

Fig. 1. Each box shows the data and the folded best fit model (top), the residuals (middle) and the unfolded model (bottom) for four of the objects in our sample. The shaded regions in the top panels indicate the energy bands selected for each instrument.

Fig. 2a-d. Each box shows the data and the folded best fit model (top), the residuals (middle) and the unfolded model (bottom) for four of the objects in our sample. The shaded regions in the top panels indicate the energy bands selected for each instrument.

4.4.1. NGC 1386

This source was detected also in the 20-100 keV spectral region by the PDS. However, further analysis indicated that most of the flux is due to a nearby Seyfert galaxy (NGC1365) that happens to be located at the edge of the PDS beam. As a consequence, we could not use the PDS data to constrain the X-ray properties of the source.

The MECS spectrum is characterized by a prominent iron line and a very low continuum level. The low continuum makes estimates of the EW(Fe K) very uncertain, since the latter becomes very sensitive to the continuum fit. However, we could set a 90% confidence lower limit of 2 keV to the EW(Fe K), that strongly supports the Compton thick nature of this source.

The Compton, cold reflection model provides the best fit to the high energy ( 3 keV) data. However, from a statistical point of view this model is only marginally better than the warm scattering model: with the same number of degrees of freedom. The main problem is that the low continuum level makes difficult discriminating different continuum shapes.

On the other hand, the Fe K line center is consistent with 6.4 keV, and it is not consistent with 6.7 or 6.96 keV at a high confidence level. There is no evidence for additional components at 6.7 or 6.96 keV. This result provide further support to the cold reflection model with respect to the warm scattering model.

As discussed above, the PDS data cannot be used to constrain the emission in the 20-100 keV range of NGC 1386. As a consequence, in Compton thick models we cannot rule out an absorbing colummn density in the range cm^-2 that would provide an excess of emission in the PDS band with respect to the MECS flux. Therefore, in Compton thick models we can only determine a lower limit of cm^-2 for the absorbing column density.

In the Compton thin, transmission model the absorbing column density N_H is indetermined. A value of N_H cm^-2 provides a fit that statistically is not significantly worse than the cold reflection model. The Compton thin model is ruled out on physical grounds because of the large EW(Fe K). Even the EW computed by correcting the continuum for absorption (EW_corr) has still a 90% lower limit of 600 eV.

The Compton thick model is also supported by the ratio which, as shown in Fig. 3, is about two orders of magnitude lower than in Sy1s.

Fig. 3. Distribution of the ratio between the observed 2-10 keV luminosity and the (reddening corrected) [OIII] line luminosity for the sources in our sample, compared to the same distribution for the Sy1s in the sample of Mulchaey et al. (1994) (we excluded the Sy1s for which the narrow-lines Balmer decrement was not available to correct the [OIII] flux).

NGC 1386 has been observed also by ASCA on 1995 January 26, with a similar integration time as our SAX observation. We reduced and analyzed the ASCA data and found that both flux and spectral shape of the ASCA spectrum are consistent with our data within the uncertainties. This indicates that the source does not show evidence for long term variability in excess of about 25% , that is our uncertainty on the flux. This is consistent with the idea that the observed flux is not seen directly, but is reprocessed by a large scale ( 1 pc) medium.

The ASCA data of NGC 1386 were also analyzed by Iyomoto et al. (1997) who interpret the observed spectrum with a Compton thin transmission model (N_H cm^-2). As discussed above, statistically the transmission model would fit also our data, but is inconsistent with the large EW(Fe K).

Finally, we estimated the contribution to the soft X rays from the Fornax cluster thermal emission by extracting the spectrum in two regions of the sky located at the same distance from the cluster center, and by using the same aperture size. The contribution from the cluster to the observed soft X-ray flux turns out to be about one third of the total observed in NGC 1386.

4.4.2. NGC 2273

The Compton thick, cold reflection model fits the high energy data better than the warm scattering model at a high confidence level: above 3 keV, with the same number of degrees of freedom. The Compton thin, transmission model indicates N_H = cm^-2 and is worse than the cold reflection model only at the 70% confidence level (same with only one degree of freedom less). However, the Compton thick nature of this spectrum is supported by the large EW of the iron line(s). Also, if in the transmission model the photon index is thawed (see Appendix) the best fitting value is , much flatter than expected for the intrinsic spectrum, thus further supporting the Compton thick, cold reflection model.

In Compton thick models the upper limit provided by the PDS in the 20-100 keV range rules out an absorbing column density in the range cm^-2, otherwise the observed 20-100 keV flux would be significantly higher than our upper limits. Therefore, if the Compton thick model is valid the column density along our line of sight must be larger than cm^-2.

The fit of the iron line improves significantly (99% confidence level) by splitting the line in two components at 6.4 and 6.96 keV. The bump at 3.1 keV observed in the residuals could be interpreted as a line of ArXVII, but by adding a gaussian at this location the fit does not improve significantly.

4.4.3. NGC 3081

The Compton thin, transmission model (N_H = cm^-2) fits the high energy data better than the Compton thick models at a high confidence level (the latter ones have reduced ). The absorbing column density is very large, and is responsible for making the observed EW(Fe K) intermediate between typical Compton thin sources and Compton thick ones.

4.4.4. NGC 3393

The MECS data of this source were already presented in Salvati et al. (1997), in this paper we include the PDS detection in the 20-100 keV range.

As discussed in Salvati et al. (1997), statistically the Compton thick, cold reflection model is only marginally better than the transmission model: the latter (N_H = cm^-2) fits the data with a with respect to the former, at the expense of only one extra degree of freedom. However, the data above 10 keV are reasonably consistent with cold reflection, while the transmission model falls short of accounting for the high energy data. If in the transmission model the photon index is thawed (see Appendix) the best fitting value of is negative, that is certainly not acceptable for the intrinsic spectrum. The EW(Fe K) is large even when the continuum is corrected for the absorbing column deduced by the transmission model. Both findings are in favor of the Compton thick interpretation (with N_H 10²⁵ cm^-2). The warm scattering model is ruled out at a confidence level larger than 90% and, in particular, falls short of accouting for the PDS flux.

4.4.5. NGC 4939

The Compton thick, cold reflection model (with N_H 10²⁵cm^-2, i.e. reflection throughout the PDS range) provides the best fit to the data in the 3-100 keV spectral range. The warm scattering model is ruled out at a high significance level (reduced ); the Compton thin, transmission model (N_H = cm^-2) is worse than the cold reflection model at a confidence level larger than 90% (above 3 keV, with only one degree of freedom less); both alternatives are especially bad in the 10-100 keV band, where they fall short of accounting for the observed flux. As in the case of NGC2273, if in the transmission model the photon index is thawed (see Appendix) the best fitting value is 0.7, adding further support to the Compton thick, cold reflection interpretation.

The iron line appears to contain a H-like component at 6.96 keV that is as strong as the neutral-fluorescence line at 6.4 keV. Perhaps a warm scattered component is present at lower energies (i.e. NGC 4939 might be similar to NGC 1068, Matt et al. 1997b), but the signal-to-noise in our data is not high enough to detect any additional component of the continuum.

We should note that the EW of the 6.4 keV line is only 400 eV and that the only about one order of magnitude lower than the average in Sy1s (Fig. 3). As a consequence, the preference for the cold reflection, Compton thick model over a transmission model with N_H = cm^-2 is still questionable.

4.4.6. NGC 4941

The MECS data of NGC 4941 were also presented in Salvati et al. (1997). In this paper we include the LECS data and a marginal PDS detection in the 30-60 keV range. However, the latter datum was not used during the fitting procedure because of its low statistical significance (though it is plotted in Fig. 2). The transmission model (N_H = cm^-2) provides the best fit to the data above 3 keV. The Compton thick warm scattering model is ruled out at a high confidence level, and the Compton thick, cold reflection model is also worse than the transmission model (though at a lower confidence level: above 3 keV with only one degree of freedom less). The is high, but consistent with 300 eV at the 90% level once the effects of absorption are taken into account. Summarizing, although the transmission scenario is preferred, we consider the nature of this spectrum still uncertain.

4.4.7. NGC 5643

This source is close to a cluster of galaxies. To avoid contamination we reduced the extraction radius to 1´. We also checked the level of contamination by extracting the spectrum of two regions located at the same distance from the cluster as NGC5643 and found that the flux from the cluster contributes a negligible fraction of the total flux observed in NGC5643.

A simple power law with , and no excess absorption with respect to the Galactic value (N_H = cm^-2) fits the continuum significantly better than the cold reflection model. Both the absence of an absorption cutoff, in addition to the Galactic one, and the large EW of the iron line strongly support the idea that the source is Compton thick, and warm scattering dominated.

If this were the case, ionized iron lines would be expected. The best fit energy of the gaussian in the rest frame of the galaxy is 6.46, i.e. the iron is less ionized than Fe XXII . These ions can still emit a significant line by fluorescence and resonant scattering; lighter elements are highly ionized so that the scattered spectrum is not dramatically distorted in shape with respect to the incident one.

NGC 5643 has been observed also by ASCA for ksec on 1996 February 21. The 2-10 keV flux observed by ASCA is about 30% lower than that derived from our data, while there is agreement on the EW and the energy of the Fe line. Above 3 keV the slope of the power law in the ASCA spectrum is flatter than that observed in the BeppoSAX spectrum. The difference in flux is puzzling, calibration problems might have occurred. The difference in slope could be ascribed to the low statistics in both spectra (actually an intermediate model could fit them both with a reduced .), but it calls into question our interpretation of the spectrum as warm scattering dominated. However, be it warm scattering or cold reflection dominated, the Compton thick nature of the source is supported by both ASCA and BeppoSAX data.

As discussed previously, the upper limit provided by the PDS requires an absorbing column density higher than cm^-2.

4.4.8. MGC-05-18-002

This source was observed after the failure of one of the three MECS units. Both the reduced effective area and the short integration time are responsible for the low signal-to-noise ratio of its spectrum.

In this case we only used a simple power law fit (Table A1). Although a detailed analysis cannot be performed, both the very flat spectrum (Table A1) and the very low ratio (two orders of magnitude lower than the average Sy1, as shown in Fig. 3) strongly suggest that this is another Compton thick source, probably cold reflection dominated, and with N_H 10²⁵cm^-2 because of the lack of detection in the PDS range. Higher signal-to-noise data are required to confirm this interpretation.

© European Southern Observatory (ESO) 1998

Online publication: September 17, 1998
helpdesk.link@springer.de