HST STIS Advanced Spectral Library Project

Thomas Ayres (for the ASTRAL Co-Investigators), University of Colorado (CASA)


Four metal-poor turn-off stars from GO-14161 (Peterson, PI) have been added as of June 2017.


ASTRAL is a Hubble Space Telescope (HST) Large Treasury Project, whose aim is to collect high-quality ultraviolet spectra of representative bright stars utilizing the high-performance Space Telescope Imaging Spectrograph. In Cycle 18 (2010–2011), ASTRAL focused on eight iconic late-type stars, devoting 146 HST orbits to the purpose. In Cycle 21 (2013–2015), the program shifted gears to the warm side of the H-R diagram, to capture 21 diverse early-type objects with an allocation of 230 orbits. Main objective is to record the targets — including well-known bright stars like Procyon, Betelgeuse, Sirius, and Vega — with broad uninterrupted UV coverage (1150–3100 Å) at the highest signal-to-noise and highest echelle spectral resolution achievable within the alloted spacecraft time, and given a variety of observing constraints. These UV "atlases" have enormous interpretive value in their own right, and will complement efforts from ground-based observatories, which now routinely achieve comparably high resolution and S/N in optical and near-infrared spectra of bright stars.

Broad ultraviolet coverage is achieved by splicing together echellegrams taken in multiple FUV (1150–1700 Å) and NUV (1600–3100 Å) grating settings of STIS. The observing strategy was designed to maximize S/N by spreading the total desired exposure depth in each setting over 2–5 separate "visits" of up to four orbits each, in a series of independent relatively short integrations (1–3 kiloseconds duration). Because of a slight randomness in the STIS grating positioning mechanism, and because of the changing projected velocity of the target (owing to telluric and spacecraft motions), each independent exposure will shift slightly on the detector and thus experience a different "fixed pattern noise." Combining the independent spectra after the fact mitigates these systematics, improving S/N. In addition, the observing sequences were designed to have at least one exposure of each distinct setting immediately after a target-centering "peak-up" so that the velocity zero point of the echellegram, which can be affected by thermal drifts, would be as accurate as possible. Other exposures of that type can be registered to the reference observation by cross-correlation. Furthermore — if practical, depending on target brightness — a few exposures in each sequence were taken through the photometric aperture (0.2"×0.2"), and again close to a peak-up, to ensure that the radiometric scale of the final spliced spectrum would be as close to the true absolute level as practical.

The post-processing of the ASTRAL spectra follows protocols developed for the earlier StarCAT project, an extensive catalog of STIS echelle spectra of objects classified as "stars." A full description can be found in Ayres (2010: ApJS 187, 149). The author strongly encourages consulting the latter as a general introduction before attempting to use the ASTRAL spectra for analysis purposes. At the same time, the ASTRAL experience has uncovered a number of observation-related issues, which have required modifications and additions to the original protocols.

Brief Description

Post-processing begins at the level of the calstis pipeline "x1d" file, a tabulation of extracted wavelengths, flux densities, photometric errors, and data quality flags for the up to several dozen orders of the particular grating setting (e.g., E140M-1425, where the first part is the mode and the second is central wavelength in Å). The x1d file contains at least one — sometimes several — sub-exposures, which were treated as separate observations. The initial processing includes a post facto correction for subtle wavelength distortions identified in a previous study of the STIS dispersion relations ("Deep Lamp Project"); a Bayesian reformulation of the photometric error; and more aggressive edge trimming of E230 settings to avoid unflagged "dropouts" that sometimes occur at the beginning of the low orders. Also, spurious tilts of the spectra, due to an unfortunate interaction between narrow slits and the telescope "breathing," are corrected, if necessary. The x1d orders then are merged, averaging the overlap regions weighted by the individual sensitivity functions s λ, but accounting for bad pixels and wavelength gaps. During this process, an "active blaze correction" determines an optimum blaze shift to balance the fluxes in the overlaps between adjacent orders.

Next, a series of different layers of coaddition and splicing are applied to the sets of order-merged 1D spectra of each object:

STAGE ZERO — sub-exposures, if any, of an observation are combined. The individual spectra are aligned in velocity by cross-correlating against the observation with the highest apparent throughput, determined by comparing the total net count rates integrated over the central zone of the echelle format. With the narrow slits used extensively in ASTRAL Hot-Stars, the effective throughput can vary significantly (up to ~30%) during a single orbit owing to telescope breathing. Next, the sub-exposures are interpolated onto the wavelength scale of the reference spectrum; scaled to the reference exposure in flux density according to the total net count rates; then coadded, weighting by the total net counts, but taking into account bad pixels and gaps. Resulting files are called "o-type" and have the same rootname as the original observation, e.g., obkk52040.

STAGE ONE/TWO — Same-setting exposures of an object taken in different orbits of a visit, or in different visits, are combined. This is a hybrid of the Stages ONE and TWO described in StarCAT, to take advantage of the specially designed observing strategy of ASTRAL. As in Stage ZERO, the cross-correlation alignments are relative to the exposure exhibiting the highest apparent throughput. Also, the individual observations are scaled to the reference exposure according to flux ratios determined by a global average over high S/N intervals. Again, the weighting was by the total net counts, which allows for possibly different integration times in an exposure set (or use of different apertures in different visits). Resulting files are called "E-type," and have an appended aperture code and MJD date range to reflect the diversity of the constituent exposures, e.g., "E140M-1425_020X020_55543-55554." If multiple apertures were used, the aperture code is changed to "MULTIaperture," e.g., "E230H-2513_MULTIaperture_52507-57585." (The latter applies to spectra processed after 1 January 2017.)

STAGE THREE — Different wavelength segments of an object are spliced together to produce, ideally, a seamless spectrum covering the full FUV + NUV range. As in the other steps, wavelengths of adjacent segments are aligned by cross-correlation. A "bootstrapping" calibration procedure takes advantage of the intentional broadly overlapping spectral coverage to refine the velocity zero point and the absolute flux scale (by pair-wise comparisons of the overlap regions in velocity and flux). Resulting file is called "U-type," e.g., "UVSUM_1X_55543-55554." The "UVSUM" part signals that a multi-wavelength splice was involved; the middle numeral indicates a particular grouping of spectra that were spliced; the adjacent letter tells whether the spectra all were medium resolution ("M"), all were high resolution ("H"), or mixed ("X"; latter is the case for all but one of the ASTRAL Cool Stars targets, but several of the Hot Stars are purely M and a few are purely H); and the trailing date range summarizes the minimum and maximum starting MJDs of the spliced group. In the splicing procedure, the general philosophy was to minimize the overlap regions between observations of different resolutions, to the extent possible given the desire to include enough overlap to determine an accurate flux ratio, as well as to contain at least one suitable cross-correlation feature. An exception was made for the frequently-used combination E140H-1291 + E140M-1425, in order to boost S/N in the key FUV interval below about 1350 Å. In this instance the entire overlap zone (1150–1350 Å) was coadded. In all cases of mixed resolution in an overlap zone, the higher resolution spectrum was filtered with the lower resolution line-spread-function, and vice versa for the lower resolution spectrum. Thus, the spectral resolution in a mixed resolution overlap zone is a convolution of the two lsf's, and is lower than the lower of the two. The photometric errors were adjusted for the filtering, and the coaddition weighting was according to a smoothed version of the respective inverse squared errors (in flux density units). The dual-filtering procedure avoids the awkward lsf that would result from simply adding the two mixed-resolution overlaps (as was done in StarCAT). NOTE, however, that if the highest resolution is desired, say for ISM studies, one should avoid the top-level spliced (U-type) spectrum below 1350 Å, and instead consult the E140H-1291 coadd one level down.

As of May 2015, a number of key improvements have been made to the x1d unpacking procedure: (1) Double-density resampling abandoned. Previously, the x1d parameters were resampled on a wavelength grid with half the spacing of the native x1d file. This was done to mitigate the smoothing that accompanies interpolations in the downstream processing (for velocity registrations), but recent tests indicated little real improvement. An undesirable side-effect of the resampling is to introduce correlated noise in the spectral traces, which apparently survives even in a multi-exposure coaddition. Thus, the resampling step was eliminated. (2) Spectral de-tilting implemented. An undesirable consequence of the telescope breathing, mentioned earlier, is that the slight focus changes can affect the slit throughputs, especially for the very narrow apertures used frequently in Hot Stars. Because the telescope psf is wavelength dependent, the blurring of the image on the slit can cause a wavelength-dependent change in the throughput, which manifests as a spurious tilt of the spectrum. Recall that the ASTRAL observing strategy yields usually several, in some cases many, independent observations in the same setting. If one assumes that the stellar source was constant over the span of those visits, and that differences in the net count rates of the exposures were due solely to breathing effects, then one can identify the exposure of a set that has the best apparent throughput (highest net count rate), and ratio the other observations to it in order to assess whether spurious tilts are present. This was done in practice by dividing the x1d echellegram into five or six zones (as a function of order number) and determining the net count rates averaged over those zones. If tilts were apparent in the ratio comparison, a low-order polynomial was fitted to the deviated ratio, and later applied to correct the flux density scales of the individual observations (at the sub-exposure level). In essence, the relative flux density scales of a set of observations were slaved to the one that exhibited the highest throughput. This approach is especially effective for the early-type stars, which usually have strong signal over the whole echellegram, and for the late-type stars at the longer, continuum-dominated wavelengths. The strategy does not work well for the shorter wavelengths of the late-type stars, which tend to be faint. It also is not an effective strategy, of course, if only a single exposure in a given setting is available. In cases where a tilt could not be measured definitively, no correction was applied. (3) Gaussian "tapers" to the weighting functions at opposite sides of overlapping echelle orders to ensure flux continuity across order edges. Previously, the untapered weights could cause small flux jumps at the order edges if intensities of the overlapping orders were different (as can happen if the blaze correction is not perfect). (4) Active compensation for blaze shifts. The post-SM4 calstis processing does not yet take into account time-evolution of the blaze correction (as had been noticed in the 1997–2004 STIS echelle material, and ultimately was compensated in the calstis of that epoch). Many of the new Hot Stars spectra showed conspicuous mismatches between the order overlap regions. These mismatches also were present in the earlier Cool Stars spectra, but were not so obvious because that program was closer in time to the post-SM4 blaze calibration, and the S/N usually was much lower than for the UV-bright Hot Stars. Although the blaze correction displays a systematic time dependence, measurements of Hot Stars echellograms indicated that short-term changes in the blaze shifts were common. To compensate, an "active" blaze correction was developed. It determines the appropriate blaze shift for a given spectrum to minimize flux imbalances over the middle orders with significant counts in the overlap zones. All of the ASTRAL echellegrams were reprocessed using the new approach, and this eliminated a number of annoying defects that previously had been cropping up in the order-merged spectra. Again, this strategy works well for the photon-rich early-type stars, but often obtains indeterminate results for the UV-starved settings of the late-type stars. If the blaze shift could not be determined actively, the default was to apply a time-dependent correction estimated from long-term behavior of post-SM4 observations of calibration WDs.

However, there still remain a number of isolated defects in the STIS spectra, which are noticeable in the Hot Stars tracings, especially for those targets with long stretches of smooth continuum. Most of these features appear as weak, narrow "emission lines" and mainly are confined to the NUV region. They apparently result from flat-field issues, and there is an on-going effort to identify and correct the cause. There also occasionally are small defects that occur in spectral overlap zones where one of the tracings has points flagged as "bad." If the other, "good" spectrum diverges from the average intensity at those locations, small "blips" will result. These can be recognized easily, however, because the "bad" flag is retained for those points. Finally, a new flag (EPSILON=450) marks places in the splicing zones between adjacent echelle settings where the difference between the coadded fluxes was larger than 3 times the average photometric noise at that location. This flag was introduced to help identify possible processing glitches.

The author, in collaboration with the STIS calibration group at STScI, has developed a new version of the echelle dispersion relations for the calstis pipeline. When it is fully implemented in the On-The-Fly system, the ASTRAL spectra will be reprocessed up to that standard. Other small revisions to the ASTRAL spectra likely will be made from time to time when upgrades to calstis warrant (take note of OTF processing date in the Tables below).

IMPORTANT NOTE: In the processed spectra below, the top-level dataset ("UVSUM") was constructed so that the maximum resolution was achieved in each spectral region, except below 1350 Å, as mentioned earlier, if an E140H was available to complement the normal E140M setting. In the overlaps, the cross-resolution filtering described earlier produces a hybrid, convolved line spread function, which is unimportant for the broad-line objects, but significant for the narrow-line stars and/or ISM features. In the latter cases, one should examine the highest-resolution coadded spectrum in the next layer down, e.g., E140H in the example mentioned above. In a few cases, such as HR-1840, there was small section of archival H-res spliced into the middle of the ASTRAL broad M-res setting, so beware of abrupt resolution changes. In fact, a new parameter "RESOLN" has been added to the U-type FITS files: it keeps track of the local resolution to help identify places where mixed-resolution spectra were spliced. Also, for the Cool Stars targets, if an E230M-2707 was taken for flux calibration purposes, it was spliced in only over a small interval at the longwavelength end of the merged H-res spectrum. Finally, because of the "breathing" throughput and spectral tilt issues, and the necessity to de-tilt and rescale many of the sub-exposures, the ASTRAL Hot Stars material is not well suited for time domain studies of continuum variations (the 0.2"×0.2" Cool Stars o-type exposures should be better in the time-domain regard). The fact that the exposures of each setting were scaled to the one with maximum throughput could introduce a bias, if the source had in fact been variable over the span of the STIS visits, or even if the post-SM4 photometric calibration of that setting had been affected by a slightly lower throughput due to breathing. Thus, the absolute fluxes of the ASTRAL spectra should be taken with a grain of salt, so to speak; although the relative flux distributions should be reliable, except for perhaps at the longest wavelengths for the objects where only a single spectrum was taken and a narrow slit (0.06", or less) was used.

Contents of the Catalog

Top-level object tables, below, summarize brief stellar characteristics, mostly taken from SIMBAD, and link to several layers of processed data. BE SURE TO RE-LOAD PAGE TO ACCESS MOST UP-TO-DATE DATA (Note OTF processing date in Tables).

ASTRAL Cool Stars

HD NumberProper Nameα2000δ2000VB-VTypeπOTF Date

ASTRAL Hot Stars

HD NumberProper Nameα2000δ2000VB-VTypeπOTF DateNotes
93129AHD-93129A160.989-59.5477.310.17O2If*0.0002015-FEB-13double: 0.05" sep
66811ZET-PUP120.896-40.0032.25-0.28O4I(n)fp0.0032015-FEB-13classic P-Cygni wind
46223HD-4622398.038+04.8237.320.13O4V((f))0.0002015-FEB-13MK standard
46202HD-4620298.043+04.9668.200.16O9.2V0.0002015-FEB-13MK standard
36959HR-188683.754-06.0095.67-0.23B1Vv0.0012015-FEB-13sharp-lined, `normal' B
37479SIG-ORI-E84.696-02.5946.66-0.18B2Vp0.0072015-FEB-13broad-lined, magnetic star
52089EPS-CMA104.656-28.9721.51-0.13B1.5II0.0082015-FEB-13low ISM
160762IOT-HER264.866+46.0063.80-0.17B3IV0.0072015-FEB-13sharp-lined, abundance standard
215573XI-OCT342.595-80.1235.31-0.12B6IV0.0072015-FEB-13sharp-lined, `normal' B
1757341-ARI42.495+27.2603.61-0.09B8Vn0.0202015-FEB-13broad-lined, LISM
87901ALP-LEO152.092+11.9671.40-0.16B8IVn0.0412015-FEB-13broad-lined, LISM
176437GAM-LYR284.735+32.6893.25-0.05B9III0.0052015-FEB-13broad-lined, MK standard
172167ALP-LYR279.234+38.7830.030.00A0V0.1302015-FEB-13photometric standard, LISM
48915ALP-CMA101.287-16.716-1.470.01A0V0.3792015-FEB-13Am, brightest star, LISM

Metal-Deficient Turn-Off Stars

HD NumberProper Nameα2000δ2000VB-VTypeπ[Fe/H]OTF Date

First layer down (linked through object HD number [Column 1]) contains the final spectrum (or spectra) for each target, typically a U-type covering all (or a significant part) of the 1150–3100 Å range.

This "final spectrum" layer then links down to the constituent spectra in the splice; and each of these in turn points to the grouping of exposures that constituted it. Thus, the lowest layers of a tree always are the o-type exposures; the next layer up has the E-types; and finally the top layer holds the U-type(s).

There are several kinds of processed data. First, each page displays a GIF preview of the coadded (o-type or E-type) or spliced (U-type) spectrum. Second, at the top level, a graphical timeline of all the observations is provided, color-coded by mode. Third, FITS files of the fundamental data are linked for downloading. Fourth, for the top-level datasets, an "ETC-ready" ASCII file is available. It is a highly streamlined version of the final spectrum, specifically intended to be used with an HST Exposure Time Calculator (or any that adheres to the HST ETC format standard). A description of the streamlining procedure and numerous warnings concerning the use of these ETC files can be found in ETC Summary. Finally, links to flat ASCII versions of all the coadded data files can be found at:

Fits File Data Formats. The o-type files have basic header information in the zeroth extension describing the target, exposure properties, and splice points from the merging process; and one or more trailing data extensions. If the observation consisted of a single exposure, there would be only one data extension — EXTEN=1 — and the spectral parameters would be found there. These are: WAVE — wavelength (Å); FLUX — flux density (erg/cm²/s/Å); ERROR — photomeric error (same units as flux density); and DQ — data quality (0 for no issues; higher values to flag various conditions such as bad pixels, camera blemishes, gaps, and so forth). On the other hand, if there were two or more sub-exposures, EXTEN=1 would hold the Stage ZERO coadded spectrum, while the subsequent extensions would have the parameters of the individual sub-exposures: sub#n in EXTEN=n+1. The EXTEN=0 header now would contain additional information, concerning the cross-correlation template and derived velocity shifts.

The E-type coadded and U-type spliced spectra both have similar structure to a single-exposure o-type, consisting of only two extensions. The zeroth extension again lists basic information concerning the target and exposure properties, and digests of cross-correlation templates, splice points (for the U-types), and flux scale factors. Extension 1 contains the spectral parameters for the coadded and/or spliced spectrum. In all cases, including o-types, the most refined dataset always is in EXTEN=1.

ASTRAL Cool Stars. A few words concerning the eight late-type targets. They cover a wide range of spectral type, F-M, mainly giants and supergiants. All have been observed by IUE, and all previously by HST, usually GHRS but two by STIS. However, especially for GHRS, the spectral coverage was incomplete, the resolution was not as high as the STIS H modes, or the S/N did not meet our goals. To round out the original ASTRAL Cool Stars group, the nearby, bright cool dwarfs Alpha Centauri A (G2 V) and B (K1 V) were added (from an ongoing joint Chandra/HST program to track the coronal activity cycles). Key characteristics of the targets:

Beta Cassiopeia (Caph: F2 IV)— Most extreme "X-ray deficient" case in a group of already anomalous fast-spinning Hertzsprung gap giants. These stars display powerful FUV emissions, but surprisingly underluminous X-ray coronae. Beta Cas, like Procyon (see next), falls at the edge of convection: an essential ingredient (together with rotation, which Beta Cas has in abundance, but Procyon does not) for the dynamo generation of magnetic fields, with their consequent effects on high-energy processes in the stellar outer atmosphere. Important link in magnetic evolution of Hertzsprung gap giants: soft 1 MK corona versus hard 10 MK for the later G0 IIIs.

Alpha Canis Minoris A (Procyon: F5 IV-V)— Nearby, bright, warmer analog of the low-activity Sun. Important cool-corona object (2 MK). Chandra transmission grating spectrum mid-way between solar-like Alpha Cen A and its more active K-type companion Alpha Cen B.

Alpha Centauri (Rigel Kentaurus) A (G2 V) and B (K1 V)— Nearest sunlike stars, only 1.3 pc away. Binary system with 80 year period and 17.5" semi-major axis, although only ~4" apparent separation in current epoch (ca. 2015). Resolved X-ray sources with Chandra, even at closest orbital approach (2016). Alpha Cen A is a near twin of Sun in its fundamental properties, including age and coronal activity; B is smaller, cooler, less luminous, but coronally more active. Important comparisons to more evolved stars of original Cool Stars sample.

Alpha Aquarii (Sadalmelik: G2 Ib)— "Hybrid chromosphere" supergiant in the class originally discovered by L. Hartmann and colleagues in early 1980s: harboring cool massive wind, imprinting blueshifted circumstellar absorptions on Mg II; but also displaying hot FUV lines like C IV, a combination usually avoided in the "noncoronal" giants like Arcturus (Alpha Boo: K2 III) and Aldebaran (Alpha Tau: K5 III). Weak coronal X-ray source detected by Chandra. Important comparison to its sibling, Beta Aqr (G0 Ib), another certified hybrid star, previously observed by STIS (and Chandra).

Beta Draconis (Rastaban: G2 Iab)— Yellow supergiants Beta Dra and Alpha Aqr, although superficially similar in spectral type and luminosity, are strikingly different at high energies: former is a strong X-ray source with bright FUV emissions; latter is cool-wind dominated star with suppressed FUV emissions and barely detected corona. A pivotal pair for understanding the dichotomy between coronally active and quiet supergiants.

Beta Geminorum (Pollux: K0 IIIb)— Early-K giant with solar-like coronal properties; key comparison to the noncoronal red giants mentioned above. Important contrast as well to the equally puzzling class of super-active helium-core-burning "clump giants" like Iota Cap (G8 III) and Beta Cet (K0 III), previously observed by STIS. One of few giant stars with suspected planetary companion as well as detected weak magnetic field.

Gamma Draconis (Etamin: K5 III)— Another hybrid chromosphere star, showing weaker fluoresced molecular lines (CO and H2) than archetype red giant Arcturus, latter extensively observed by STIS. Faint Chandra source, but stronger than the 3 events detected from Arcturus in a 19 kilosecond HRC-I pointing. Important link to more active hybrid stars like Alpha Aqr (above).

Gamma Crucis (Gacrux: M3.5 III)— Classic M giant crucial for understanding complex atmosphere, wind, and spectrum of more exotic red supergiant Betelgeuse (next): similar in surface temperature, equally extreme nocoronal object, but with a simpler, cleaner UV spectrum (e.g., narrower, less blended chromospheric and wind emission lines), and yet significant mass outflow. Bridge to the warmer noncoronal K giants like Gamma Dra.

Alpha Orionis (Betelgeuse: M2 Iab)— Iconic windy cool supergiant, with very clumpy surface convection and mysterious distant cold circumstellar shell, prominent in FUV absorptions of CO (as seen by GHRS at low resolution). An extreme object in terms of low surface temperature, high visual luminosity, and lack of coronal signatures. Again, an important player in the story of the hot-corona/cool-wind transition.

ASTRAL Hot Stars. The twenty-one targets of Hot Stars are too numerous to describe individually. Some brief characteristics are noted in the Hot Stars Table. The specific objects were chosen by the ASTRAL collaborators, after significant debate, to include: the full range of spectral types from early-O to early-A; Main sequence and evolved stars; normal plus chemically peculiar subtypes; fast and slow rotators; magnetic exotica; as well as nearby objects of relevance to ISM studies. A practical consideration was that early-type stars are UV-bright, and in many cases can trigger overlight conditions on STIS's highly sensitive MAMA cameras. While there is a pair of neutral density filters in the STIS slit wheel for bright targets, the initial ND step is 2 (100x attenuation relative to the normal clear slits), which means that many targets that are just over the bright limits will be very inefficient to observe with the supported ND2 aperture (or companion ND3). To work around this limitation, we carried out a calibration program in Cycle 19 ("Bridging STIS's Neutral Density Desert") to validate a set of three 31"×0.05" intermediate-ND slits (ND=0.6, 1.0, and 1.3) that were "available but not supported." This involved measurements of a hot WD to pin down the wavelength-dependent attenuation of each slit; and a spectral comparison between short ND2 (0.2"×0.05") and tall ND1.3 (31"×0.05") using the bright, relatively sharp-lined A0 star Vega (Alpha Lyrae) to determine whether the long slit degraded resolution or enhanced scattered light (neither turned out to be an issue). The addition of the intermediate-ND slits to the STIS toolkit greatly increased the number of potential objects that could be observed efficiently (albeit still requiring 6-12 orbits per target) to satisfy the objective of high S/N. This made it easier for the ASTRAL collaborators to select a list of candidates to meet the purely scientifically-motivated objectives. The case of Vega was deemed of special importance (zero point of photometric scale; ultra-fast rotator seen pole-on; possible Chandra X-ray source), even though it fell partly into the inefficient category. Vega accordingly was allocated additional orbits (20 total) to boost S/N for some of the underperforming echelle settings.

Full details of the dual phases of the ASTRAL program will be described in a series of forthcoming journal articles.

Final word. The author would like to express appreciation to all the ASTRAL Co-Investigators who have contributed to the project, especially to STIS colleagues at STScI who have helped materially in planning and executing the complex observing program. In a number of cases, the ASTRAL Co-Is are obtaining supporting ground-based observations, which ultimately will be linked here when available.

Acknowledgments. Based on observations made with NASA/ESA Hubble Space Telescope, obtained from the Mikulsky Archive at Space Telescope Science Institute, operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. Support for ASTRAL is provided by grants HST-GO-12278.01-A and HST-GO-13346.01-A from STScI. The project has made use of public databases hosted by SIMBAD, maintained by CDS, Strasbourg, France.