2.1. Optical Color Selection and the Presence of a Semi-stellar Nucleus
It was originally realized by Sandage (1971) that the optical colors of an AGN (modeled as a power law plus broad emission lines) and the starlight from a host galaxy deviated in a systematic way from pure stellar colors as a function of power-law index, redshift, and the fraction of total light in the non-thermal continuum. Using this technique and multicolor data, it is possible to construct color ratios that efficiently select objects with unusual colors (e.g., Koo & Kron 1988). These are then candidates for follow-up optical spectroscopy and radio and X-ray observations. This technique has reached its present apogee with the SDSS, which finds objects by their deviations in five-color space (Fan 1999; Richards et al. 2001). The technique is essentially "trained" on known objects, and the colors of the quasars so obtained are sufficiently narrowly distributed that photometric redshifts can be estimated. While there are claims of "completely unbiased" optical samples (Meyer et al. 2001), this reviewer believes that they are rather overstated. In particular, X-ray images with the XMM-Newton and Chandra X-ray Observatories of the 2dF fields find many AGN not detected by this optical survey (see Fig. 1).
Figure 1. XMM-Newton image of a 2dF quasar field with the 2dF quasars marked by crosses. Notice the large numbers of X-ray point sources, virtually all AGN, that are not identified in the 2dF.
Over the last 30 years, there have been a large number of programs (see Hartwick & Schade 1990) that selected AGN using Schmidt's (1969) criteria for selecting quasars:
Sources that do not have the colors of normal stars in their nuclei at any redshift. These AGN are found by "exclusion".
Sources that show strong, relatively broad UV/optical emission lines.
Stellar sources that lack proper motion. This criterion has recently been revived by analyses of deep Hubble Space Telescope (HST) images.
Luminous high-redshift sources that are selected by their optical colors via the Lyman break.
These criteria are optimized for quasar-like sources and will not select Seyfert 2-like sources very well. (Seyfert 2 galaxies have lines that are not as strong or as broad, nuclei that are not as bright, and they show little or no time variability in the optical bands.) The realization that AGN bolometric luminosity may not be well correlated with galaxy luminosity (Urry 2003) indicates that low-luminosity AGN are very difficult to find in massive galaxies via optical selection techniques. Essentially, the starlight dilutes the signatures of the AGN, reducing the equivalent widths of all of the lines; at moderate redshifts (z > 0.2), the starlight reduces or eliminates the color signatures of the nucleus.
Hartwick & Schade (1990) review the results of many of these surveys and provide an excellent summary of optical AGN selection techniques. However, they do not compare the results to techniques obtained via other methods. It is somewhat amusing to find stated in the Hartwick & Schade (1990) paper that the samples found by these optical techniques are a complete sample of quasars. As we shall see below, optical techniques miss most radio-selected AGN and a large fraction of IR and X-ray-selected AGN.
The presence of a semi-stellar nucleus was one of the original defining criteria for quasars. However, this criterion is highly sensitive to the angular resolution of the telescope and to the relative contrast of the nucleus and the host galaxy. It also finds bright super star clusters and other compact objects. Sarajedini et al. (1999) searched for stellar nuclei in a relatively large sample of galaxies serendipitously observed by HST and found that ~ 10% show unresolved semi-stellar nuclei that contribute more than 5% of the total optical light. So far, this is the best optical survey for low-luminosity AGN-like activity at moderate redshifts.
While optical techniques are very powerful, they have several problems. Their use requires that the nuclei either be bright enough to outshine the stars in the photometric extraction region or have sufficiently different colors to be recognizable. The effects of copious star formation on the optical colors are considerable. For example, in the Byurakan surveys, which selected objects by searching for blue continua, 90% of the objects found are starbursts rather than Seyferts. Thus, this selection technique is only five times more efficient than a blind survey. The effect of stellar dilution has been quantified for the SDSS by Richards et al. (2001), who noticed that more luminous objects are bluer at a fixed redshift due presumably to the increased importance of starlight in lower luminosity objects. At an absolute magnitude in the Sloan g band of -23 (or an optical luminosity of 6 × 1044 ergs s-1), this effect seems to go away, indicating that the effects of starlight are minimal for these extremely luminous objects. Thus, there are strong selection effects with the luminosities of the nuclei and of the host galaxies, and with redshift. In principle, these selection effects are quantifiable, but in practice, this only works at low redshifts, where one can properly model the host galaxy and remove the starlight (Moran, Filippenko, & Chornock 2002; see also Moran, this volume). Because there is no simple relation between the host galaxy and the nucleus properties, the problem is rather intractable. There are also large color selection terms because of the varying contrast between a "typical" stellar spectrum and the SED of an AGN with redshift.
Color selection is a very efficient technique, and with a large database like the SDSS, produces copious samples of AGN over a wide redshift range. However, such methods must be evaluated carefully because of the large ratio of stars to quasars. At m ~ 18 there are ~ 500 stars to every quasar, and thus the classification accuracy must be better than 0.2% to avoid severe contamination! Therefore, techniques have focused on stellar "rejection" rather than on "finding all the quasars", unless extensive spectroscopic follow-up is also available for the survey.
In addition, since it is now well-known (see discussion in Section 3) that many (most?) AGN have large amounts of dust and gas in the line of sight, the effects of extinction can be very large and have to be very well modeled to produce reliable samples. The effects of extinction can be partially ameliorated via the use of near-IR data, but only up to some limit. It has been known for very many years that the reddening curve in AGN is different from that in the interstellar medium of the Milky Way (Maiolino et al. 2001), and thus it is not at all clear how to "de-redden" the AGN spectrum to derive reliable optical fluxes.