A remarkable fraction of the exciting, forefront fields of astronomy originated from largely serendipitous discoveries. The passage of time and post facto rewrites of history sometimes obscure and confuse the result of special insight and hard work with the truly serendipitous outcome. But even tempered by this ambiguity, all of us can point to major, exciting research areas which started at least largely by accident: pulsars, extrasolar X-ray sources, the cosmic microwave background radiation, gamma-ray bursts, etc.
Of course, few or none of these wonderful serendipitous observations were derived from literally pointing an instrument at the sky at random; rather, they were enabled when a new observational capability became available. The SDSS, with precise photometric, astrometric, and morphological data for tens of millions of images, and the ability to acquire huge numbers of homogeneous, well-calibrated spectra, based on highly precise and well understood selection criteria, is another such new capability. Never before has it been possible, for example, to say "show me the spectrum of every object in pi ster of sky with colors and magnitudes in 5-color space, accurate to a few percent, between ranges x and y , for morphology z". Therefore it is worthwhile to ponder how to maximize the probability of serendipitous discoveries with SDSS.
It may seem tautological to discuss "planning for serendipity," and indeed there is a limited amount we can do in advance of data acquisition other than to keep an open mind, and to do our best to ensure that the data collection, reduction and archiving techniques do not preclude a variety of later lines of analysis. One area of SDSS not amenable to post facto manipulation, however, is the selection of spectroscopic targets from the imaging data base. Once the spectra are on hand and the data collection is over, the universe of SDSS spectra has been totally defined. Of course, the ~ 106 homogeneous spectra in the SDSS data bank can be automatically and quantitatively scrutinized for a huge variety of new phenomena at leisure after observations have ended, and these tasks will probably occupy a significant fraction of the astronomical community for many years after the close of the survey. However, we have also given considerable thought to optimizing serendipity during the data collection phase of the survey, via selection of spectroscopic targets.
To this end, the SDSS Scientific Advisory Committee has agreed to devote a modest fraction of the 640 spectroscopic fibers on each plate (a few percent) to "serendipity." A Serendipity Working Group of interested SDSS scientists is charged with specifying automated target selection criteria for these fibers, using data from the automated pipeline reductions in the imaging data bank, exactly analogous to how galaxies and QSOs are selected for spectroscopy. Unlike most of the survey target selection, however, there is no need for continuity and homogeneity of the selection criteria, as a homogeneous sample is not required. Indeed, most attempts at selecting in advance interesting spectroscopic targets for serendipitous science will doubtless fail. Therefore an additional task of this Working Group is to rapidly and continuously evaluate the 5-10 "serendipity spectra" from each plate, identify selection criteria which yield uninteresting spectra, and modify criteria for future plates. We anticipate that such modifications of selection algorithms, or more precisely of variable parameter values within algorithms, will occur at least many times per year, and possibly more frequently as experience and confidence are gained.
In addition, the target selection software allows positions to be put into a target list "by hand", i.e. if an astronomer scrutinizing recent photometric data notices an object of interest, its position is entered into a database, automatically updated, of serendipitous objects, which is input to the spectroscopic plate design algorithms and targeted if a fiber is available.
Of course, many serendipitous discoveries will be followed up using other telescopes, but the advantage of also incorporating this facility into the SDSS should not be underestimated; a few spectra per field adds up to a lot of spectra in a survey the size of the SDSS -- several tens of thousands over the whole sky. This provides the opportunity to find extremely rare objects (e.g. zero metallicity stars) and to measure spectra for decent-sized samples of the merely rare.
Even apart from the exciting science, attention to serendipity is of immense value to the SDSS because its obverse is quality control; the detailed scrutiny to which apparently new and unusual astronomical results must be subjected will also provide stringent checks on the quality, reliability and accuracy of the SDSS data.
Although it is not appropriate to quote specific detailed algorithms for selection of the serendipitous targets here, a few illustrative examples may be useful. The precise five-color photometry in the imaging survey permits us to target those objects in each 7 deg² field that are the most geometrically distant in five-color space from the clustered locus of all other stars and galaxies on the color-color plots. Objects already selected for spectroscopy by other SDSS Working Groups will be ignored in this process (although they will be tagged in the data base as of interest to the serendipity working group). The QSO Working Group, for example, also looks for "odd colors", but in specific regimes of the color-color plots meant to optimize detection of QSOs; the Stars Working Group similarly selects unusual objects. Thus this serendipity algorithm will potentially enable detection of odd objects, both galactic and extragalactic, which have colors and/or morphologies that are unprecedented. As an additional bonus, this technique serves as a quality-control monitor of the more conventional (e.g., QSO) selection algorithms: if the serendipity fibers consistently identify significant numbers of QSOs, this is an alert to the QSO group that their algorithm is incomplete.
Serendipity fibers will also be placed as available on optical counterparts of FIRST radio sources. The FIRST survey (Becker et al. 1995) is a comprehensive atlas of the sky at 1400 MHz, made by the VLA in the B configuration, to a limiting flux density of ~1 mJy ( 5 sigma ). The FIRST survey region has been chosen (not accidentally) to coincide with the pi ster of the northern SDSS survey area. The initial results show that FIRST detects ~ 90 sources per square degree, each with a positional accuracy of ~1" or better, thus normally yielding unambiguous optical identifications with SDSS images for all sources with optical magnitudes brighter than the SDSS detection thresholds. As each SDSS field will have almost 103 FIRST sources, it is not practical, given our other scientific goals, to obtain spectra of the majority of FIRST objects in the SDSS region as a dedicated program. However, only about 20% of these sources will have optical counterparts bright enough to be worth targetting with the SDSS. Furthermore, the vast majority of these sources are either quasars or galaxies. The majority of the quasars will have optical colors which cause them to be selected as quasar spectroscopic targets anyway, as discussed in Section 3.3.2, we will target stellar sources brighter than g' = 19.7 associated with FIRST radio sources anyway, to supplement our quasar color selection. Finally, all galaxies with r' < 18.2 will be targetted, and thus we expect only a handful (we estimate between 5 and 10) FIRST sources bright enough to get spectra of which we are not targetting anyway. Serendipity fibers will then be placed where possible on these FIRST sources that coincide with optical counterparts. This provides the potential to discover very odd objects, both galactic and extragalactic. For example, SS 433, were it located in the SDSS region, would be discovered by this technique.
Becker, R. H., White, R. L., and Helfand, D. J. 1995, ApJ 450, 559.