Astron. Astrophys. 346, 407-414 (1999)

4. Spectral analysis

The spectra of the imaging instruments have been rebinned in order to sample the intrinsic energy resolution of the detectors with 3 (LECS) or 4 (MECS) energy channels. Each channel has at least 30 counts, which ensures the applicability of test. The PDS spectrum has been quasi-logarithmically rebinned, in order to have 16 energy channels in the 14-200 keV band. The spectra of the three detectors have been fitted simultaneously. Numerical relative normalization factors among the BeppoSAX instruments have been added to all the following spectral fits. The reasons is two-fold: a) the BeppoSAX instrument response matrices employed in this paper (September 1997 release) exhibit slight mismatches in the absolute flux calibration; b) the sampling of the instruments is not strictly simultaneous, due to the need for operating the LECS only during satellite nights or the different data selection criteria between imaging instruments and the PDS. This can affect the flux measured in variable sources as NGC 4593. The LECS to MECS factor has been left free in the fitting procedure, and turns out to be comprised in the range 0.74-0.76, which is consistent with typical values observed so far (0.7-1.1: Grandi et al. 1997; Haardt et al. 1998; Cusumano et al. 1998). The PDS to MECS factor has been instead held fixed to 0.8, as the available statistics was not good enough to provide independent constraints on it. This value corresponds to the multiplication of the best fiducial value estimated by Cusumano et al. (1998) by 0.82, to account for the effect of the PDS RT selection algorithm employed (see Sect. 1). The systematic uncertainty on this parameter can be estimated . Spectral fits have been performed in the 0.1-4 keV (LECS), 1.8-10.5 (MECS), 14-200 keV (PDS) energy bands.

4.1. Continuum shape

In Fig. 4 the result is shown, when a simple power-law model with photoelectric absorption is applied. The quality of the fit is rather poor ( dof). The main deviations are due to: (a) an absorption feature starting at  keV; (b) a prominent emission line with centroid energy  keV; (c) a "bump" in the PDS band, peaking at energy  keV. The residuals between 0.3 and 0.6 keV (immediately red wards the absorption feature) are systematically positive. In principle, this may be due to the emergence of a soft excess in this energy range. However, the recovery of the residuals to values consistent with zero below 0.3 keV leads to the suggestion the the above feature is simply the typical wavy residual produced by a mis-fit absorption edge (see e.g. Nandra & Pounds 1992). The BeppoSAX broadband allows to study simultaneously the whole spectral complexities expected on the basis of previous band-limited measures of this object and of the Seyfert 1s as a class (cf. Sect. 1). We have then defined a "baseline" model, where a Compton reflection component from a neutral slab (model dollar;E blank; bsol;simeq blank;25 dollar; in XSPEC , Magdziarz & Zdziarski 1995), an emission line and a photoionization absorption edge are superposed to the photoelectric absorbed power-law. The possibility that the reflecting matter is substantially ionized is not required by the data. The model depends on the heavy element abundance (which has been held fixed to the solar one), and on the angle between the normal to the slab and the line of sight ("inclination angle" ). has been held fixed to hereinafter (the best-fit value arising from a fit of the iron line profile with a relativistic model, see Sect. 4.2). The only free parameter in addition to the power-law model of the continuum is the relative normalization between the reflected and the primary components (equal to 1 for an isotropic source, illuminating a plane-parallel infinite slab). The best-fit and are basically unaffected, if one assumes that the reflector is seen face-on (i.e. : ). The "baseline" model yields a very good  dof. The Table 1 reports the best-fit parameters and results.

Fig. 4. Spectra and best-fit model (upper panel ) and residuals in units of standard deviations (lower panel ) when a simple power-law model with photoelectric absorption is applied

Table 1. BeppoSAX best-fit results. wa = photoelectric absorption from neutral matter; po = power-law; px = Compton reflection; ga = Gaussian line; ed = photoionization absorption edge; bk = broken power-law.
Notes:
a) ;
b) , "baseline model" in text.
c) fit on the ASCA data (see Sect. 5)

It is worth noticing that no soft excess above the extrapolation of the high-energy power-law is required. A model, where one replaces the absorption edge with a continuous break of the primary power-law provides a much worse fit ( dof) and leaves significant residuals in the 0.2-1.3 keV energy range. Moreover, the best-fit spectrum is convex (i.e. : ), further demonstrating that no soft excess is present in the data. The inferred absorbing column density is consistent with the Galactic contribution along the line of sight as measured by Elvis et al. (1989,  cm^-2). The best-fit model and deconvolved spectra are shown in Fig. 5. The 0.1-2 keV, 2-10 keV and 20-100 keV fluxes are , and  erg cm^-2 s^-1, respectively. They correspond to rest frame luminosities of , and  erg s^-1, respectively.

Fig. 5. Left: spectra and best-fit model (upper panel ) and residuals in units of standard deviations (lower panel ) when the "baseline" model is applied. Right: unfolded energy spectrum and best-fit model (solid lines). The direct and reflected continuum and the emission line are separately indicated with dotted lines. The location of the O VII photoionization absorption edge in the observers frame is labeled

The unprecedented BeppoSAX energy coverage allows the simultaneous determination of the primary radiation steepness and of the Compton reflection intensity with the best accuracy ever achieved. In Fig. 6 the contour plot for the photon index versus the relative normalization between the reprocessed and primary components is shown. The latter parameter is 1 if the reflection occurs in an infinite, plane-parallel slab and the primary source emits isotropically. Higher values are formally possible, and may be due to the geometry of the reflecting matter, covering more than (e.g. in a concave or warped accretion disk), to an intrinsic anisotropy of the primary source, or to a delayed response of the reflecting matter to changes of the primary flux. At 90% level of confidence for two interesting parameters, the photon index is constrained between 1.81 and 1.91. R is consistent with a plane-parallel geometry of the reflecting matter (90% confidence interval between 0.6 and 1.6). Any cut-off of the intrinsic power-law is constrained to lay at energies higher then 150 keV (see Fig. 7). It is worth noticing that the statistical relative uncertainties on and are rather small, despite the strong correlation between the two parameters (note the strongly inclined contour plot in Fig. 7). For comparison, the single parameter 68% statistical errors ( and ) are 4.5 and 3.5 times smaller than the ones obtained from the Ginga observations of the same Seyfert 1 (Nandra & Pounds 1994). It should however be noticed that the residual systematic uncertainties on the relative PDS to MECS normalization factor affect the accuracy of both these parameters, with additional uncertainties of about 1% and 30%, respectively.

Fig. 6. Contour plot of the relative normalization between the primary and Compton reflected spectral components versus the intrinsic photon index, when the "baseline" model is employed. Iso- curves are at 68%, 90% and 99% confidence level for two interesting parameters (, 4.6 and 9.2)

Fig. 7. Contour plot of the cutoff energy of the primary power-law component versus the photon index when the "baseline" model is employed

4.2. On the iron line

The centroid energy of the iron line is well consistent with K fluorescence from neutral iron. The line is broad if a simple broad Gaussian profile is used to describe it (see Fig. 8); the intrinsic width is comprised in the range 60-600 eV at 90% level of confidence for two interesting parameters (best-fit value  eV). We have tried also alternative parameterization of the line. Adding a further "narrow" line (i.e. : intrinsic width held fixed at zero) to the "baseline" model results in no improvement of the quality of the fit (). Given the intrinsic width of the line and the agreement between the derived best-fit line EW and the theoretical expectations if the line is produced in an X-ray illuminated relativistic accretion disk (Matt et al. 1992), we have tried to use a self-consistent model of line emission from a relativistic accretion disk (model dollar; bsol;Delta blank; bsol;chi circ;2 blank; equals; blank;0 dollar; in XSPEC , Fabian et al. 1989). If all its parameters are allowed to be free, most of them are totally unconstrained. No further constraint comes from the ASCA data as well (Nandra et al. 1997). If it is assumed that the inner radius of the emitting region is 6 gravitational radii (); the emissivity law index is equal to -2.5 (Nandra et al. 1997); the inclination of the line and Compton reflection continuum emitting region is the same; and the line is neutral (i.e. :  keV); then , and  eV. The quality of the iron line modeling ( dof) is comparably good as for the broad Gaussian.

Fig. 8. Contour plot of the intrinsic width versus the centroid energy of the iron line, when the "baseline" model is employed

4.3. The warm absorber

In 12 out of 24 Seyfert galaxies observed by ASCA, absorption edges from ionized species of oxygen have been detected (Reynolds 1997), which have been interpreted as the imprinting of warm gas along the path from the nucleus and us. NGC 4593 is one of these objects, and the BeppoSAX observation confirms this outcome, thanks to the detection of an absorption edge with threshold energy  keV. The edge energy is consistent with the K-photoionization threshold energy of OVII . We have tried to give a qualitative characterization of the ionization and chemical structure of the absorbing matter, by tentatively including in the fit four absorption edges, with threshold energies held fixed to the values expected from OVII (0.739 keV), OVIII (0.871 keV), NeIX (1.196 keV) and NeX (1.362 keV). However, the available statistics is not good enough to give us significant constraints. Only the OVII edge yields a significant detection (), while only upper limits can be obtained for the other three edges (, , ). Using CLOUDY (Ferland 1996), we have constructed a grid of self-consistent models of the spectra transmitted through a ionized gas in thermal and ionization equilibrium, when the SED is the one observed in NGC 4593 (Santos-Lleó et al. 1995). They depend on the ionization parameter (defined as the dimensionless ratio between the number of Hydrogen ionizing photons and the electron density of the gas), and the warm column density . The fit is comparably good as the phenomenological description of the "baseline" model (). The best-fit parameters of the warm absorbing matter are  cm^-2 and . For a source with , the best-fit corresponds to a value of the more common a few. The continuum parameters are slightly affected (; ), but remain consistent with the one of the "baseline" model within the statistical uncertainties.

© European Southern Observatory (ESO) 1999

Online publication: May 21, 1999
helpdesk.link@springer.de