A number of surveys were carried out to overcome all the shortcomings of previous optical surveys for QSOs, such as difficult selection functions, narrow luminosity ranges and small samples. I list six important and complementary surveys here, which have produced results in recent years:
the brightest (~ 400) objects were observed in the wide-area objective-prism Hamburg-ESO-Survey (HES, Wisotzki 2000) at B < 17, z = [0.0, 3.2]
the 2dF Quasar Redshift Survey (2QZ, Boyle et al. 2000, Croom et al. 2004) found large numbers (~ 23, 000) of QSOs at z = [0.3, 2.2] and B < 21.
the Sloan Digital Sky Survey (SDSS, York et al. 2000) collects the largest QSO sample to date with important complete subsamples at z > 3.6, i < 20 (Fan et al. 2001)
the COMBO-17 survey (Wolf et al., 2003) selected QSOs reliably at z = [1, 5] with detailed SEDs from 17 filters, providing the deepest large sample
the BTC-40 survey (Monier et al. 2002) targetted specifically the very high-z end at z > 4.8, confirming just two of these rare QSOs to date
a Lyman-break-selected galaxy sample (Steidel et al. 2002) providing a small but the deepest probe into nuclear activity at z ~ 3 (Hunt et al. 2004)
All these surveys selected their object samples by colour, and all except for HES and COMBO-17 assembled a QSO sample by spectroscopic follow-up of a subset. In contrast, HES and COMBO-17 had their spectroscopic information already obtained in the first observational step: HES is based on object prism spectroscopy and collapsed spectra into colours for selection; the COMBO-17 filters have sufficient number and spectral resolution to identify QSOs and measure accurate redshifts from photometry alone.
Indeed, the enlarged statistics on colours and spectra of QSOs allowed to develop the technique of photometric redshifts for QSOs to maturity: broad-band colours such as the SDSS ugriz bands lead to photo-z errors of z ~ 0.2 or ~ 0.1 in photo-z codes based on QSO templates (Richards et al. 2001) or using neural nets (Budavari et al. 2001), respectively. In the COMBO-17 filter set the larger number of filters leads to a much more complete identification of QSOs, especially in the range of the suspected turnover at z = [2, 4]. The higher spectral resolution of medium-band filters leads to much smaller photo-z errors of z / (1 + z) 0.015 for QSOs (Wolf et al. 2004). Similar work in the CADIS survey did not take variability into account, and the resulting variation in observed colours led to a number of redshift outliers there (Wolf et al. 1999). The latter comparison demonstrates the importance of simultaneous colours for photometric redshifts of QSOs.
The analysis of these surveys has resulted in LFs covering different parts of the redshift-luminosity plane, and we have found that space densities mostly agree well between surveys where redshift and luminosity ranges overlap. Especially, it was shown that the COMBO-17 LF is consistent with 2QZ at low redshift and smoothly turns over into the SDSS LF at high redshift (see Fig. 1 and Wolf et al. 2003), providing the previously missing link.
Figure 1. QSO space densities from 2QZ, COMBO-17 and SDSS (Fan et al. 2001) agree very well. The different curves show densities according to the best-fitting models from each survey, integrated to different limiting luminosities of M145, Vega < [- 28 ... - 24]. Dashed lines are extrapolations to fainter magnitudes.
3.1. Shape of the luminosity function
Where luminosity functions from different surveys overlap, they tend to agree. However, the parts of the redshift-luminosity plane measured by more than one survey are not large, and there are still some remaining issues about survey incompleteness and its correction.
The largest newly observed sample is from 2QZ and covers redshifts of z = [0.3, 2.2]. The resulting LFs can be fit to broken power laws and give satisfactory fits for a PLE-type model. However, 2QZ selects candidates from broad-band colours and thus has a bias against low-luminosity QSOs with bright host galaxies, where the latter dominate the colours. Some of the flattening of the towards the faint end may result from this incompleteness, and it is not clear that a broken power law is the required function.
At the brightest end, the HES has found some mild curvature in the LF. At very low redshifts of z < 0.3, it should clearly allow to see the L* knee in the LF suggested by the early broken power-laws and seen again by the 2QZ survey. However, while there is curvature in the LF and flattening towards lower luminosities, a broken power-law does not match the HES results, which instead suggest a steeper faint LF than 2QZ.
At high redshifts of z > 3.6 the SDSS sample has improved over the original SSG95 result. The results do not allow to constrain LF curvature due to small sample size and a narrow luminosity range, yet. However, the slope of the SDSS high-z LF is flatter than the 2QZ slope at lower redshift, suggesting evolution in the LF shape.
The COMBO-17 sample bridged the gap between 2QZ and SDSS with complete selection, and furthermore extended the LFs to much fainter limits. The COMBO-17 LF shows again mild curvature but no L* break in its low-luminosity range. Combined with the SDSS LF, COMBO-17 shows a flattening of the high-z LF, as a faint extrapolation of the single power-law SDSS LF exceeds the COMBO-17 counts.
The Steidel et al. Lyman-break sample provides the deepest probe, although the object number does not help to constrain the faint-end slope or LF curvature reliably. It was argued, that this sample combined with the WHO94 results suggested an L* break with a strong slope change. However, combining the larger samples from COMBO-17 and SDSS gives consistent results in terms of a smoothly curved LF without a distinct break.
Altogther, these results suggest that the LF of QSOs is smoothly curved, but we have not seen a compelling case for a strong L* break, which could be traced as a clue to the physical evolution of QSOs underlying our observations.
3.2. Peak of AGN activity
The peak of the cosmic AGN activity would be measured best by the integrated AGN luminosity density, which could potentially be used as a crude proxy for the total BH accretion, once the conversion becomes clearer. The faint COMBO-17 observations have demonstrated that the L-range of existing samples probably covers > 95% of the total blue AGN luminosity density. Thus, variations in the further faint extrapolation would change the total luminosity density only little. COMBO-17 found an activity peak at z 2.0 based on its LF at MB < - 22. In contrast, the much shallower HES found the luminosity density to rise further out to its survey limit at z = 3.2 from an LF integrated at MB < - 28. This comparison clarifies that high-luminosity QSOs reached their activity peak earlier than low-luminosity objects and implies that the shape of the QSO LF changes with time. The same behaviour was also found in X-ray selected samples, where low-Lx AGN peak at z ~ 0.75 (Cowie et al. 2003, Steffen et al. 2003). This phenomenon is called cosmic downsizing and describes a scenario, in which comparatively large members of a population undergo the critical evolutionary epoch first, with members of a progressively smaller kind showing up only later. This downsizing trend provides an important constraint for theories of co-evolution between AGN and their host galaxies.
3.3. Reddened QSOs
Aside from the subject of optically-obscured QSOs, the optical community occasionally wondered about missing red or reddened QSOs which may escape colour selection or show unexpected spectral shapes. Recently, a careful study of a large SDSS QSO sample in combination with 2MASS photometry investigated the detailed continuum properties of QSOs at z < 2.2 (Richards et al. 2003, Hopkins et al. 2004). This work showed that the QSO population could best be explained by a distribution of intrinsic spectral slopes = [- 0.75, - 0.25], combined with dust reddening at the redshift of the QSO, possibly with an SMC-type dust law. It was shown that the AGN continuum and the broad-line region (BLR) were reddened to the same degree but the narrow-line region (NLR) remained unaffected. The mean reddening was measured to be <EB-V> = 0.03, while the SDSS selection procedure was shown to be sensitive to EB-V < 0.5. Only 2% of the parent population of type-1 QSOs was found to be reddened with EB-V > 0.1. A total of ~ 10% of QSOs are lost from the SDSS sample by extinction pushing objects below the flux limit, not due to reddening-induced colour changes!
3.4. Optically obscured QSOs
The canonical model of AGN involves a dusty torus around the accreting black hole, which absorbs optical signatures of the central engine on certain lines-of-sight, while keeping a type-1 appearance at other viewing angles. Among low-luminosity AGN, this picture had been confirmed through the unification of Seyfert-1 galaxies with Seyfert-2 galaxies. The key observation in the latter was a detection of unobscured AGN light, which was scattered (and polarized) by low-density dust residing far above the absorbing torus plane and being freely illuminated by the AGN. Seyfert-2 galaxies had been found abundantly as counterparts to Seyfert-1 galaxies, but only with low luminosity. Five years ago, optically obscured QSOs (with high luminosity) were still at large, although a radio galaxy at z = 0.44, allegedly the most powerful IRAS source, was a good candidate for a type-2 QSO (Kleinmann et al. 1988) and further X-ray-based evidence was obtained by Franceschini et al. (2000).
More recently, very significant progress was made with deep X-ray surveys and optical identification of faint X-ray sources by spectroscopic follow-up. Several type-2 QSOs were found with Chandra and XMM, reaching up to z = 3.7 (Norman et al. 2002). It has been confirmed that type-2 objects dominate the population at low luminosity, but their fraction drops when going to high luminosities. At this point, it seems unclear which population dominates the nuclear accretion budget, because of the debate on how to correct for completeness in follow-up identification. E.g., Treister et al. (this volume) suggest, that with proper correction type-2 objects account for 3/4 of the integrated X-ray luminosity even at z > 2, where we only see higher-Lx objects.
Maybe somewhat unexpectedly, the subject of optically obscured AGN has been advanced through extensive work based on optical selection in the SDSS spectroscopic database. Kauffmann et al. (2003) have selected type-2 AGN at z < 0.3 by subtracting the stellar continuum and star formation-related components from emission lines to isolate the AGN emission from the NLR. Zakamska et al. (2003, 2004) have pushed further in redshift by applying such a technique to the whole SDSS spectral database. They isolated a serendipitous sample at z = [0.3, 0.8] with ~ 150 type-2 QSOs, defined by LOIII > 3 × 108Lsol, and many lower-luminosity objects. Their AGN nature was further confirmed by showing that this class of objects is MIR/FIR-luminous and has hard X-ray colours. However, in contrast to the Kauffmann et al. sample at z < 0.3, the statistical completeness of the Zakamska et al. sample is presently unclear.
While X-ray surveys played a fundamental role in mapping out the obscured population, they still need to improve in area, and optical/NIR follow-up needs to reach deeper, before the obscured samples at high-z will catch up in usefulness with type-1 samples. We will need a good z, L-map of the type-1/2 fraction, not only to observe the entire nuclear accretion budget, but also in order to constrain models of torus evolution.
3.5. Evolution of spectral properties
Any evolution in spectral properties of QSOs would hold important clues about changes in the physical conditions in AGN, which would help very much to interpret the AGN phenomenon beyond the above discussed 'bean counting' exercises of the many surveys. However, no significant redshift evolution has been observed in the properties of either the continuum or the emission lines. Pentericci et al. (2003) complemented the optical data of 45 SDSS QSOs at z = [3.6, 5.0] with JHK photometry and found a mean spectral slope of = - 0.57 ± 0.33, which is fully consistent with the QSOs at low redshift. Emission lines were found to display various (anti-) correlations between their equivalent widths (EWs) and the continuum luminosity of the source, also known as Baldwin effects. But no evolution was found for EWs with redshift, leaving us with no evidence for chemical evolution (Croom et al. 2002, Dietrich et al. 2002). Also, no evolution with redshift has been found in a relation between black hole mass and QSO luminosity, MBH LQSO (see Corbett et al. 2003 and references therein).