The fundamentals behind quasar discovery have not changed since the earliest surveys. Historically the approaches taken in the optical have been rather different from those applied in the radio and X-ray regimes, although recent developments are producing something of a convergence in approach.
3.1 The Optical Regime
In the optical ( ~ 3300-9000 Å), where thermal sources, both Galactic and extragalactic, dominate the source counts, the detection is based on the identification of some recognizable non-stellar feature(s) of the quasar spectra. The two classic examples are the so-called ultraviolet excess and the presence of strong, broad emission lines (see Smith 1975). The detection problem is essentially one of eliminating the large number of contaminating objects, and a successful strategy to eliminate ``false positives'' is a prerequisite for a viable survey. Far too little attention has been paid to this problem, with many workers concentrating on the development of particularly clever or sophisticated detection algorithms. There is usually little point in developing a ``superior'' emission line detection algorithm, thereby extending detection to lower equivalent widths, if large numbers of non-quasars are also then flagged as candidates; the overall efficiency is reduced, sometimes by a large factor. However, in a project to determine the relative proportions of quasars with emission lines of different equivalent width, a decrease in efficiency for objects with lines of low equivalent width may be unavoidable. The viability of a survey can be ascertained only if the detection limit and the corresponding success rate are established. The demonstration that a technique can identify a quasar with, say, a C IV emission line of 10 Å equivalent width may be an outstanding achievement if the success rate is one in three, but only of academic interest if the success rate is one in a hundred.
The need to reduce the number of contaminants can be appreciated through consideration of the surface density of objects on the sky. At high Galactic latitudes the integral surface density of sources, predominantly stars, at mB = 18 is ~ 500 deg-2, while the integral surface density of quasars is ~ 1 deg2. Compilation of a sample of modest size, say 200 quasars, thus requires that a minimum of 100,000 objects be examined. If the selection technique has a probability of only 1% of misclassifying a star as a quasar candidate, a candidate list of ~ 1200 objects will result, with a resulting overall efficiency of 17%. All of these must be followed-up with spectroscopy on large telescopes. Even these daunting statistics assume that the selection technique has unit probability of correctly identifying a quasar. For magnitudes mB 19.5, the quasar number-magnitude counts show a rapid rise with increasing magnitude, log (N) 0.8 m, and this increase outpaces that of stars and galaxies. At mB = 20, the surface density of all objects is ~ 2000 deg2, and there are ~ 20 quasars per square degree; fully 1% of sources are quasars. Fainter than mB = 20 the quasar number-magnitude counts become much shallower, log (N) 0.3 m (Koo and Kron 1982), and the statistics again become more unfavorable. However, the number of quasars per square degree is numerically large, ~ 100 deg2 at mB = 22, and a substantial multiplex advantage becomes possible through the use of wide-field, 0.5-1.0 deg2, multifiber systems. The faint ultraviolet-excess survey of Boyle, Shanks and collaborators (420 quasars, mB 21; Boyle et al. 1990) is an outstanding example. Their work demonstrates how a major survey can be achieved with relatively modest amounts of telescope time by matching the region of the apparent magnitude-surface density parameter space to an available telescope/instrumentation combination. Unfortunately such an approach is not possible for the majority of the absolute magnitude-redshift plane and few additional opportunities exist to undertake such efficient surveys. The development of the two-degree diameter multifiber field for the Anglo Australian Telescope (Gray et al. 1993) will enable several major investigations, such as quantifying the spatial clustering behavior of the quasar population, to be undertaken. However, in general, surveys will continue to be based on spectroscopy of single (or a few) objects at a time.
3.2 The Wavelength Extremes
At the wavelength extremes, notably in the radio and X-ray regimes, the surface density of sources at bright flux levels is low, and quasar surveys have consisted essentially of optical spectroscopy of entire flux-limited catalogs. The recently-completed EMSS survey (Gioia et al. 1990) is an important modern example, while the original optical identification program of the 3rd Cambridge Radio Survey led to the discovery of quasars (Schmidt 1963) and resulted in the first sample suitable for statistical analysis (Schmidt 1968).
While spectroscopy of entire flux-limited samples offers the prospect of producing a well-defined sample that contains all objects in an area of sky with fluxes within specified limits, practical problems remain. The positional accuracy of objects identified at far-infrared and X-ray wavelengths has been poor by optical standards, with error ellipses extending to substantial fractions of an arcminute common, frequently resulting in an ambiguity in identifying the optical counterparts. Radio catalogs derived from single-dish observations suffer similarly, although re-observation of many sources with radio interferometers has resulted in positional accuracies smaller than an optical seeing disc. Once positional accuracies are ~ 1 arcsecond, the ambiguity in the optical identification is small, except in extreme circumstances, and spectroscopic identification is therefore unbiased. Hazard (1989) has stressed how poor positional accuracy coupled with the imposition of preconceived ideas concerning the target ``identifications'' can lead to important classes of object being overlooked. In recent surveys, such as the EMSS, care has been taken to eliminate any bias at the expense of substantial amounts of large telescope time, with many sources requiring spectra of several optical candidates. The EMSS program has been highly successful in this regard although the possibility of some minor bias has been suggested (Scott 1991).
Unless the fraction of quasars in the candidate list is high, optical spectroscopy of radio and X-ray selected catalogs is expensive in terms of telescope time. Improvements in the positional accuracy of source detection at all wavelengths has reduced the overhead in performing optical identifications, but as sensitivity has improved, new populations of objects (other than quasars) constitute a significant fraction, or even the bulk, of the source counts (Windhorst et al. 1985, Benn et al. 1993). At the faintest flux levels complete spectroscopic identification is no longer viable and additional information is required to target the quasars. Thus, the strategy required to perform a successful survey is becoming similar to that employed in the optical; the reduction of false-positives is essential to make the survey viable. Information such as radio spectral index, or X-ray hardness, allows the elimination of certain classes of objects, such as nearby stars, and data from other wavelength regimes can also provide valuable discrimination. In particular, the ``all-sky'' optical catalogs from the digital scanning of the UK Schmidt Telescope and Palomar Schmidt Telescope plates are revolutionizing follow-up spectroscopy of objects identified at other wavelengths. As optical color information becomes available for large areas of sky it will be possible to dramatically restrict the target lists for follow-up spectroscopy. The intrinsic spread in the SEDs suggests a cautious approach when ruling out objects, but even if very liberal criteria for the definition of candidates are employed, the demands on follow-up spectroscopy can nonetheless be reduced by a large factor in some cases. McMahon's (1991) use of optical colors to identify the small fraction of sources in a faint radio survey that are possible high-redshift quasars is a case in point.