Special plates

Overview table of special programs

The Programname is stored in the platex table in the CAS. We have begun to add SQL example queries showing how to use the extra and special spectra. We will soon provide SQL example queries for accessing the special spectra in the CAS. The links in the Program column point to the tables below describing individual target flags and targeting criteria.

Descriptions of individual programs and primtarget categories

Note The descriptions on this page pertain to targeting cateogries, i.e., the way in which objects where selected from the photometry catalog. Before using these spectra, you need to check both the spectral classification (specClass in the CAS, spec_cln in the spSpec*.fits primary header) and the redshift status (as for all SDSS spectra).

Galactic kinematic programs

F stars

F stars are numerous in the Galaxy, have sharp spectral features allowing accurate radial velocities to be measured, are approximate standard candles if the stars are on the main sequence, and are of high enough luminosity that they can be seen to great distances. This program aims to use F star radial velocities to understand the kinematics of the outer parts of the Milky Way. Most of these plates were in the Southern Galactic Cap Equatorial Stripe, but there are also three plates in the vicinity of M31, and another two plates centered roughly on the Perseus cluster. The M31 (e.g., Zucker et al. 2004a and Zucker et al. 2004b) and Perseus imaging data are not included in DR4, but will be included in a future data release.

Main sequence turnoff

This program was designed to study the kinematics and metallicities of high-latitude thick disk and halo stars. One of the plates overlaps with the area of the Thick/Thin disk program described below and uses the extra color cut i-z>0.2 to avoid overlap with that program. The DR4 spectra from this program are a random subset of all the objects identified by the selection algorithm.

Thick/thin disk stars

A third program was focused on the kinematics on the thin and thick disks, targeting a complete sample of 8880 bright, red stars in three adjacent spectroscopic tiles centered at high Galactic latitude. It provides an in situ sample of thin and thick disk stars with radial velocities, proper motions, and spectroscopic metallicity determinations, densely sampling a single line of sight out to 2 kpc above the Galactic plane.

SEGUE test plates

SEGUE ("Sloan Extension for Galactic Understanding and Exploration") is part of a follow-on project to the SDSS, emphasizing spectra of stars to study stellar populations and Galactic structure. The targets on the SEGUE test plates in DR4 are Galactic stars of a variety of colors and magnitudes, meant to sample the range of available spectral types and to test the reproducibility of radial velocity measurements at faint magnitudes. SEGUE aims to probe the structure of the Milky Way by sampling Horizontal Branch, F turnoff, G dwarf and K dwarf and giant stars, representing a variety of distances in the disk and halo. These plates are being used to refine target selection for the SEGUE program itself.

The SEGUE test plates use dereddened PSF magnitudes in their selection. Most selection categories use a color and/or magnitude weighting to get a more even sampling of the color/magnitude distribution where the number counts rise steeply with color or magnitude.

NOTE Among the SEGUE plates, plate 1664 is a special radial-velocity standards plate which targets F, G and K stars (mostly dwarfs). It includes very faint stars and has been observed twice to explore the accuracy of radial velocity determinations at faint magnitudes. Thus, the table below does not include counts for objects from plate 1664 with its fainter magnitude limits.

In addition, not all plates from the SEGUE test chunks have been observed, so that some targeting categories do not have any spectra in DR4.

Other stellar/point-source target selection algorithms

Spectra of Everything

As part of an exploration of the full stellar locus, as well as a search for unusual objects of all sorts, we carried out a survey of all stellar objects ("Spectra of Everything"), which was used as a filler for a series of so-called merged program plates, which included a mixture of mostly extragalactic targets. In 2002 (chunk merged48), this included a random sampling of all point sources with clean photometry (see the discussion of fatal and non-fatal flags in Richards et al. 2002) with reddening-corrected i-band PSF magnitudes brighter than 19.1. Not surprisingly, the vast majority of the targets were chosen from the densest core of the stellar locus in color-color space.

In an attempt to put greater weight on the wings of the stellar locus, we revised the selection algorithm slightly for chunk merged73. We defined a distance from the ridge of the stellar locus, by asking for the median and standard deviation u-g, g-r, and i-z of stars in narrow bins of r-i. Having tabulated these, we calculated a crude χ²-like quantity:

L = 1/3 Sum_{u-g,g-r,i-z} ((color-median color)/(standard deviation))2

75% of all stars had L < 1. We gave all stars with L > 1 highest priority, and for L < 1, the priority decreased smoothly with L.

These spectra were used for a determination of the completeness of the quasar target selection algorithm (Vanden Berk et al. 2005). The overwhelming majority of these objects are confirmed to be stars; only 10 of the 19,543 of the Spectra of Everything targets analyzed in that paper are quasars not previously targeted as such.

High proper motion

A sample of high proper motion stars was defined using proper motions determined using the methods of Munn et al. (2004), where a consistent astrometric solution was found matching USNO-B data (Monet et al. 2003) with SDSS. There were two cuts:

1. A simple cut of proper motion μ > 100 mas/y
2. Cuts using reduced proper motion H_r defined by:

         Hr = r + 5 log10 μ + 5

   The sample was defined by:

           g-i < 2      Hr > 16
       2 < g-i < 2.375  Hr > 8 + 4 (g-i)
   2.375 < g-i          Hr > 17.5

Stellar locus

For plates 323 and 324, stellar targets were chosen from SDSS imaging data, randomly sampling the stellar locus in color space, in order to explore the full range of stellar spectra. In particular, grids of width 0.04 magnitude in the (u-g),(r-i) and (r-i),(i-z) color planes were set down, and where they existed, a single star was selected in each grid. The process was iterated until about 600 targets on each plate had been assigned.

Galaxies at low redshifts

Main extension: galaxies

In Chunk 22, observed in Fall 2001, we used direct extensions of the main survey target selection algorithms. In particular, the main galaxy sample (Strauss et al. 2002) was modified only slightly, by removing the cut on objects with half-light Petrosian r band surface brightness μ_50,r below 24.5 mag/arcsec². This adds less than one object per square degree.

For LRGs, the sample was changed to probe about 0.3 mag fainter than for the main survey selection described by Eisenstein et al (2001). In particular, the cuts that were used garner about 40 objects/deg²:

        r < 19.5
        r < 13.4 + c||/0.3
    μ50,r < 25
     |c⊥| < 0.4

        r < 19.5
      g-r > 1.65 - c⊥
    μ50,r < 25
    rPSF-rmodel > 0.3

        c|| = 0.7(g-r) + 1.2(r-i-0.18)
        c⊥ = (r-i) - (g-r)/8

Complete main galaxies

The main galaxy sample, as described by (Strauss et al. 2002) is as complete as we can make it. However, the subset for which we actually obtain a spectrum is only about 90%. This incompleteness has several causes, including the fact that two spectroscopic fibers cannot be placed closer than 55′′ on a given plate, possible gaps between the plates, fibers that fall out of their holes, and so on. This program aims to observe the remaining 10\% of galaxy targets in order to have a region of sky with truly complete galaxy spectroscopic coverage. This is particularly important for studies of galaxy pairs, which are by definition strongly affected by the 55′′ rule. The target selection algorithm is simple: all galaxies selected by the algorithm described in (Strauss et al. 2002) in the Southern Equatorial Stripe, minus those galaxies that actually have a successfully measured redshift in routine targeting.

u-band galaxies

The main galaxy target selection is carried out in the r band, and is flux-limited at r = 17.77. The bluest galaxies have $u-r \approx 0.6$, thus a magnitude-limited in the $u$ band corresponds to $u \approx 18.4$. In order to explore the u-band luminosity density of the universe, and explore the recent history of star formation in the universe, we carried out a u-band survey of galaxies in the SDSS. Baldry et al. (2005) describe the sample in detail and present the resulting u-band luminosity function.

In the selection criteria below, we use a (dereddened) "Pseudo-Petrosian" u-band magnitude

        uselect = umodel - rmodel + rPetro.

Note that the objects targeted in this program do not include objects with spectroscopic observations already in hand. Thus one needs to combine objects from various programs to define a complete sample. See the discussion in Baldry et al. (2005) for determination of the completeness of the resulting sample.

Low-redshift galaxies

We have carried out a survey of low-redshift galaxies to 2 magnitudes fainter than the SDSS main sample limit in order to add more low-luminosity galaxies to the sample. Our redshift selection used photometric redshifts derived from second-order polynomial fits to observed Petrosian r magnitudes and model colors, with separate fits done in bins of model g-r color. For Chunks 45, 52, and 62, we used the SDSS EDR photometry and spectroscopy then available to derive photometric redshifts, while for Chunks 74 and 97, we were able to derive improved photometric redshifts using catalog-coadded Stripe 82 SDSS photometry, combined with all available SDSS redshift data on the Southern Equatorial Stripe as of 11 July 2003. This included much of the data taken for the express purpose of calibrating the photometric redshift relation in the SDSS photometric system.

Galaxies were then chosen for observations based on their photometric redshift z_p and Petrosian magnitude r_Petro. In particular, the aim was to target as complete a sample as possible for 17.77 ≤ r < 19.0 and true redshift below 0.15, and sparse samples to higher redshifts, as well as at fainter magnitudes 19.0 ≤ r < 19.5. The specific target categories, in order of highest to lowest priority for fiber assignment, are listed in the table below. As there are many more such targets than available fibers, the available targets were sampled sparsely. The sparse sampling fraction values were chosen to get reasonable distributions of objects over the target categories and to keep approximately similar target distributions from chunk to chunk, which resulted in somewhat different sampling fractions for each of the five chunks in which this algorithm was used.

A caveat to note is that Chunk 45 used the star/galaxy separation criteria of the SDSS photometric pipeline to select galaxies, and this resulted in noticeable contamination of stars in several of the Chunk 45 plates located at lower galactic latitudes. The other chunks used the star/galaxy cut employed by the SDSS main sample target selection algorithm (r_PSF - r_model ≥ 0.3 or 0.24, depending on the version of the photometric pipeline), which is more conservative for selecting galaxies.

Galaxy target selection for calibration of photometric redshifts

The SDSS five-band photometry goes substantially fainter than does the spectroscopy, suggesting the opportunity to derive photometric redshifts for vastly more objects than have spectroscopy (see, for example, Csabai etal 2003). Calibrating the photometric redshift relation requires a training sample exploring the same range of apparent magnitudes and colors as the objects for which photometric redshifts will eventually be derived. The SDSS LRG sample (Eisenstein et al. 2001) obtains spectra for red faint (r < 19.5) galaxies; photometric redshifts of this relatively uniform population (e.g., Eisenstein et al. 2003) are fairly robust (e.g., Padmanabhan et al. 2005). However, we do not have a corresponding sample of faint blue galaxies for the calibration of photometric redshifts. Therefore, a series of spectroscopic plates was designed to obtain redshifts for the blue end of the galaxy color distribution at the faint end.

 0.40 + 0.6(u-g) < g-r < 1.7 - 0.1(u-g)
            -0.5 < u-g < 3.0 
               0 < g-r < 1.8 
            -0.5 < r-i < 1.5 
              -1 < i-z < 1.5 
            18.0 < u   < 24.0 
            18.0 < g   < 21.5 
            17.8 < r   < 19.5 
            16.5 < i   < 20.5 
            16.0 < z   < 20.0 
              σu < 0.6
        σg,r,i,z < 0.25

For objects that satisfied the above cuts, the quantity exp[c((g-r) - (0.40 +0.6(u-g)))] was calculated; if it was larger than a random number chosen between zero and one, the object was targeted for spectroscopy. The coefficient c=0.1411 was chosen to obtain an appropriate density of targets. Note that plates 672 and 809 have the same center, and some of the same objects were inadvertently observed twice.

Perseus-Pisces galaxies

Imaging scans were taken centered roughly on the Perseus cluster at z ≅ 0.018, and were used to target galaxies. The double exposure plates 1665, 1666 targeted, and obtained redshifts for, approximately 400 galaxies as faint as r_fiber = 18.8 in a region centered on the cluster at (α,δ) = (49.96°r;, 41.53°r;). The majority of the galaxies are associated with the cluster, although there are 50 objects in a background overdensity at z≅0.05 (Brunzendorf & Meusinger 1999). These plates also included approximately 300 foreground F-stars.

Higher-redshift galaxies

Deep LRG exposures

The SDSS LRG sample (Eisenstein et al. 2001) targets high-redshift (0.2 < z < 0.55), luminous galaxies by their colors and magnitudes. As part of Southern targeting, we have explored this algorithm in several ways. The first, the Deep LRG sample, uses double-length spectroscopic exposures to get higher S/N spectra of LRGs with z > 0.25 which satisfied the Cut I criteria from Eisenstein et al. 2001. These serve two purposes: first, to obtain measurements of velocity dispersion for galaxies where the current, single-pass spectroscopy is only "good enough for a redshift". Second, given the discontinuity in targeting algorithm at z ≅ 0.4 (the distinction between Cut I and Cut II; see Eisenstein et al. 2001) higher S/N spectra allow the exploration of possible spectroscopic differences at this transition, implying differences in stellar populations. XXX This sounds a bit rough.

Faint LRGs

The LRG Cut II sample aims for a flux-limited sample of LRG with redshifts roughly between 0.40 and 0.55. We also experimented with an extension of this cut, going substantially fainter.

        17.5 < ideV < 20
      iPetro < 19.1 
         0.5 < g-r  < 3.0      
         0.0 < r-i  < 2.0 
         c|| = 0.7(g-r) + 1.2(r-i-0.18)
             > 1.6
         c⊥ = (r-i) - (g-r)/8
             > 0.5 

Here, deV refers to deVaucouleurs model magnitude, Petro to Petrosian magnitude, and all colors are based on model magnitudes (see Which magnitudes should I use?)

Brightest Cluster Galaxies (BCGs)

A separate program explicitly targeted the brightest galaxies in clusters. While LRGs are often the brightest galaxies in their clusters, they need not be. The so-called MaxBCG method described by Bahcall et al. (2003) searches for galaxies with the apparent magnitudes and colors of LRGs, together with a red sequence of fainter ellipticals in the vicinity (cf., Gladders & Yee 2000). The BCG program targeted BCG candidates found with this method, with estimated redshifts in the range 0.4 < z < 0.7. The MaxBCG algorithm was run on photometry derived from co-adding the detections (at the catalog level) of multiple scans of the Southern Equatorial Stripe.

Quasar and variability target selection

Main extension: quasars

In Chunk 22, the quasar target selection algorithm (Richards et al. 2002) was extended as follows. For objects selected from the ugri color cube, the magnitude limit was changed from i = 19.1 to 19.9, while for the griz color cube (where high-redshift quasars are selected), we changed the limit from i = 20.2 to 20.4.

As Richards et al. (2002) describe, there are regions outside the stellar locus that are heavily contaminated by hot white dwarfs, M dwarf-white dwarf pairs (Smolcic et al. 2004), and other non-quasars. These regions are explicitly excluded from quasar target selection in the main survey; however, in the extension, they are allowed back in. Similarly, there is a region of color space where z ≅ 2.7 quasars intersect the stellar locus. In the main survey, objects falling in this region are sparse-sampled to 10% (to reduce the number of stars); in the extension, all objects falling in the mid-z box defined in Richards et al. (2002) are targeted.

Faint quasars

The quasar target selection was further modified for the faint quasar targets on the merged plates on Chunk 48. In particular, the following are the changes over the standard quasar target selection algorithm described in Richards et al. 2002:

• The magnitude limit is set to i= 20.1, rather than i = 19.1 for the ugri color cube;
• The magnitude limit for optical counterparts to FIRST sources is set to i = 20.65, rather than i = 19.1;
• The standard quasar target selection algorithm requires that the estimated PSF magnitude errors in u and g both be less than 0.1 for UV excess sources (u-g<0.6). This limit is now set to 0.2.
• For UV excess objects with 19.1 ≤ i < 20.1, we add a requirement that g - r < 0.7 to minimize stellar contamination.
• Objects in the "mid-z box" are excluded altogether.
• In addition to the usual "distance from the stellar locus" algorithm used to target quasars, in the main sample there are hard color cuts used to select high-redshift quasars in the griz color cube. These color cuts are not used in the Southern targeting.
• Finally, there were additional color cuts to reject unphysical objects affected by bad CCD columns; we required g-r>-0.5, r-i>-0.5, and i-z>-0.6.

M31 Quasars

Low-redshift quasars were targeted on the M31 imaging data (Chunk 72) using the standard quasar selection algorithm (Richards et al. 2002), but excluding the high-redshift candidates selected from the griz cube. The confirmed quasars can be used to probe gas in the halo of M31. The plates used for this program also included F star targets.

Double-lobed radio sources

Quasar target selection targets unresolved optical counterparts to FIRST radio sources, while "serendipity" target selection (see the EDR paper, Stoughton et al. 2002; we also provide this html version of the EDR paper) selects FIRST sources with extended optical counterparts. This works fine for compact or core-dominated radio sources, but is less effective for double-lobed radio sources, in which the optical counterpart is associated with a point between the two. An algorithm was developed to find these double-lobed sources, including allowing for the more challenging case of bent double sources. In particular, pairs of FIRST sources separated by 90′′ or less without SDSS optical counterparts were identified. Given the distance d between the centers of these two sources, a rectangle is drawn centered at the midpoint between the sources, with dimensions 0.57d, 1.33d, with the short axis parallel to the line connecting the two sources. This box size was chosen empirically to include the core of a sample of bent double sources compiled by E. Blanton (see Blanton et al. 2001). Optical counterparts with r < 19.8 that fell into the box were selected as bent double counterparts. A subset of those objects, which fell into a 12 arcsec square centered on the midpoint between the sources, were selected as straight double counterparts.

Variability

As discussed above, the Southern Equatorial Stripe has been imaged multiple times in the course of the SDSS, allowing photometric variability to be studied. Variable sources were selected for follow-up spectroscopy using pairs of observations of unique unsaturated point sources with i < 21, and requiring that the changes in the g and r bands exceed 0.1 mag, and are at least 3 sigma significant (using error estimates computed by the photometric pipeline). The time difference between the two observations varied from 56 days to 1212 days.

This target selection produced targets over the Southern Equatorial stripe with a surface density of about 10 objects deg^-2. The majority of these targets have colors consistent with z < 2 (UV excess) quasars and RR Lyrae stars, classifications confirmed by the spectroscopy. Out of these, 978 have been observed on DR4 special plates (as for other special plates, objects which had already been observed on regular survey plates were not reobserved).