A solution to this discrepancy has been recently proposed by
Franceschini, Braito & Fadda (2002),
who suggested that obscured AGN
undergo a very fast evolution up to z = 0.8. The physical scenario
supporting this idea is that obscured AGN are related to the fast
evolving starburst population necessary to reproduce the ISO
mid-infrared counts
(Franceschini et al. 2001).
By shifting most of the obscured AGN at z < 1, that model
nicely reproduces the redshift
distribution observed in the deep surveys; unobscured sources are
however still needed to explain the high redshift tail of the
distribution (see Fig. 5 in
Franceschini et al. 2002).
A more refined model has been recently worked out by
Gandhi & Fabian (2002),
who also make a connection between obscured AGN with the infrared
starburst population. Even this model is able to reproduce the low-z
peak in the redshift distribution, with the main contribution at
z < 1
provided by obscured AGN. Both these new models are bound to predict a
decrease with redshift in the ratio between obscured and unobscured
AGN, which can be checked on the CDFN and CDFS data. In
Fig. 4 (bottom)
it is shown the ratio between the number of sources with
logNH > 22
and logNH < 22 in the CDFS and CDFN as a function
of redshift. Only sources with
f2-10 > 5 × 10-16 erg
cm-2 s-1 and in the inner regions of
the two fields have been considered
1, to get a spectroscopic
completeness of ~ 60%. The combined sample contains 194 sources with
measured redshift, 85 from the CDFS and 109 from the CDFN. As in
Fig. 2, the
absorption column density for the CDFS sources has been calculated by
fitting the X-ray spectra with a simple absorbed power-law, fixing the
slope to 
= 1.8 when
the photon statistics is low. About 64% of
the considered CDFS sources have absorption in excess of
1022 cm-2 . The absorption column density for the
CDFN sources has been taken from Fig. 18 of
Barger et al. (2002),
who derived the NH
values from the source hardness ratios (fixing the photon index to
= 1.8). While the
redshifts considered in the CDFS subsample
are all spectroscopic, one third of the redshifts in the CDFN
subsample are photometric. Similarly to what found in the CDFS, 72%
of the considered CDFN sources have
NH > 1022 cm-2 . The shaded area
in Fig. 4 (bottom) shows the
possible range covered by the ratio
between the number of sources with column density above and below
1022 cm-2 under the two extreme assumptions that
the unidentified
sources are either all obscured or all unobscured. Although the
incompleteness is likely to increase with redshift, it is assumed to
be 40% in each redshift bin. The ratio predicted by the
Franceschini et al. (2002)
model, calculated at a comparable limiting
flux is also shown as a solid line. At low redshifts the predicted
ratio highly overestimates the data, while the opposite is true at
high redshifts. It is noted that the Franceschini et al. model is a
simple approximation, since only one class of obscured sources is
considered (with NH ~ 2 × 1023
cm-2 ). In the more refined model
by Gandhi & Fabian
(2002),
where several classes of sources with
different obscuration are assumed, the discrepancy is less
significant, but still the ratio between AGN with
logNH > 22 and
logNH < 22 is overestimated at z < 1.
Fig. 4 (bottom) indicates that at z < 1 the ratio between AGN with logNH > 22 and logNH < 22 is lower than ~ 3, suggesting that the low-z peak in the redshift distribution is not due exclusively to obscured sources. Since the XLF of unobscured AGN is not properly sampled by ROSAT at low luminosities and moderate redshifts (1042erg s-1 at z ~ 1), a regime now accessible to Chandra, the assumed extrapolations might not be correct (preliminary results suggest this is indeed the case; see Cowie et al. 2003 and Hasinger et al. 2003) and a new determination of the AGN XLF is therefore needed.
1 About 1/4 of the CDFN sample is actually selected at fluxes above 5 × 10-15 erg cm-2 s-1 (see Barger et al. 2002). Back.