Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.stsci.edu/stsci/meetings/irw/proceedings/keelw.ps.gz Äàòà èçìåíåíèÿ: Mon May 23 17:23:52 1994 Äàòà èíäåêñèðîâàíèÿ: Sun Dec 23 02:57:09 2007 Êîäèðîâêà: IBM-866 Ïîèñêîâûå ñëîâà: arp 220

The Restoration of HST Images and Spectra II
Space Telescope Science Institute, 1994
R. J. Hanisch and R. L. White, eds.
Scientiîc Results from Deconvolved Images
William C. Keel
Department of Physics and Astronomy, University of Alabama, Box 870324, Tuscaloosa,
AL 35487
Abstract. I review various regimes in which deconvolution is and is not the technique of
choice for analysis of HST images. The difîculty of getting adequately deconvolved images
depends largely on required angular îeld, attainable signalítoínoise, and the dynamic range
of the target; this last factor limits many interesting investigations in the presence of spherical
aberration, exacerbating noise ampliîcation and uncertain knowledge of the pointíspread
function. Photometric validation issues are also important; real data are used to show how
well completely different approaches agree on the intensity proîles of faint galaxies.
Some examples of scientiîc results in the realm of galaxy structure and evolution
are given, which have required deconvolved data. Some highlight are disk structures near
galactic nuclei, îne structure in synchrotron jets, morphological evolution of mediumí and
highíredshift galaxies, and signiîcant galaxy merging at moderate redshifts. An additional
set of results has been greatly aided by deconvolution, including study of concentrated cores
in galaxies and structural parameters of faint galaxies.
1. Introduction
Many of the presentations at this workshop have, quite properly, centered on problems in image
restoration # mathematical underpinnings of algorithms, proper image bases for reconstruction,
biases in measurement from deconvolved images. These might leave a depressing view of the
prospects for getting scientiîc results from deconvolution, but the opposite is more nearly the case.
I will review here some experience as to when image restoration is the right tool for the job, and
describe some of the signiîcant science that has come from analysis of deconvolved HST imagery.
Deconvolution is clearly a powerful analysis tool. The astronomical community's experience
with HST and IRAS data has started to spread expertise in use of deconvolution techniques beyond
its traditional base in aperture synthesis, and is leading to a more sophisticated view of data and
measurement than was needed to interpret images blurred by the nearly Gaussian PSF of longí
exposure images from the ground. In retrospect, we made it so long without them because eye,
brain, and mathematics can deal with this Gaussianílike PSF very naturally; in many cases, we
didn't even realize what visual processing was going on! Deconvolution algorithms are especially
potent for astronomers as part of a software toolkit including image modelling, convolution, noise
tracking, and measurement routines.
2. Practical Validation of Results
A major question in the minds of most astronomical users was discussed by Ivan King at this
workshop # the issue of validation of the output from deconvolution. Astronomers are deeply
concerned not only about morphology but with photometric integrity. It is of limited use to
accurately distinguish thousands of stars in a restored cluster image if the positions and intensities
are subject to unknown biases produced by the reconstruction procedures. Several presentation
here have dealt with photometric issues for stellar images. For regimes useful in analyzing galaxy
structure, existing techniques already yield sufîcient accuracy to learn interesting and otherwise
359

360 Keel
inaccessible parameters. Fig. 1 shows an example of crossívalidation, using data taken from Keel
& Windhorst (1993). The image is of the faint radio galaxy 53W044 at redshift z = 0:311. The
surfaceíbrightness proîle along the local major axis has been evaluated in three quite different
ways. Direct measurements were made on deconvolved images, one made using the STSDAS
implementation of the LucyíRichardson algorithm run to ¼ 2 = 1, and the other using a hybrid
CLEAN with noise model. The CLEAN result was constrained to have a resolution equal to the
Nyquist limit for sampling in the WFC, and the LucyíRichardson result is demonstrably very close
to the same effective resolution. For galaxies with smooth, symmetric structure, modelling allows
an independent comparison. In this case, a family of models with various bulge:disk intensity ratio,
and scale lengths for bulge and disk components, was generated and convolved with the empirical
PSF (from a foreground star projected 5 arcseconds away). The model from this family that ît
best in a ¼ 2 sense was numerically realized and measured in the same way as the deconvolved
images. All three surfaceíbrightness proîles track to within a few per cent at all radii over a
dynamic range exceeding 100:1 (5 magnitudes). The three techniques could scarcely have more
different basic principles and sensitivity to details of application, so this treatment is an empirical
demonstration that HST, in its aberrated state, can deliver reliable surfaceíbrightness proîles for
galaxies at substantial redshifts; this means that we can do reliable quantitative studies of galaxy
morphology.
The twoídimensional galaxy modelling used in this validation exercise introduces a crucial
issue. When is deconvolution the analysis tool of choice, and just as important, when is it the wrong
tool? Frequently, the science we seek is not directly in the image, however crisp. In imaging a star
cluster, the desired result may be a HertzsprungíRussell diagram; in observing the inner regions
of an elliptical galaxy, the goal may be a highíresolution azimuthally averaged intensity proîle.
Problems of this kind, where the astrophysical background gives us strong a priori knowledge of the
size or shape of the target objects, lend themselves to modelling and direct comparison with data in
the observed domain, thus avoiding any possible biases from deconvolution. Speciîc tools for these
applications exist, both for stars (Stetson 1994) and galaxies (Keel & Windhorst 1993, Windhorst
et al. 1993b, Ratnatunga 1994). However, there are many problems for which the universe is not so
cooperative: the targets may have unknown or complex structure. In these cases, deconvolution is
the only way to retrieve a faithful representation of the object's properties. It may be the only way
to match measurements made with various instruments. As a concrete example, if we observe some
galaxy with a small aperture and the FOS, the only way to înd what fraction of the galaxy's total light
was included if an HST image is available involved deconvolving the image and reconvolving with
a PSF appropriate to the wavelength range observed spectroscopically. Finally, producing faithful
images for public release is not a trivial need, for use in education and in letting the taxpayers know
they're still getting something for their money.
In seeing how different investigators use or avoid deconvolution algorithms, and in the choice
for various problems, one can see some philosophical differences in how the results are approached,
and what deconvolution is supposed to do. Some users see it as an operation on the data, approxií
mating what would have been seen with a more favorable PSF, including noise and artifacts so that
the quality of the processed data is immediately apparent. Other users want to go straight to #truth#,
and want to see a bestíestimate model for the object with noise suppressed as irrelevant. Algorithms
are certainly available to do both, and cases in between as well.
Before discussing speciîc science results, it is useful consider various regimes of deconvoluí
tion, where its practice may be affected by angular extent (in choice of instrument, mode, sampling,
and extent of PSF changes) and signalítoínoise ratio (controlling the available dynamic range). For
example, a planetary image with a good PSF needs only Fourier reconstruction (Cunningham &
Anthony 1993; note that their reconstruction tests used only a spaceíinvariant PSF and are thus
applicable only to small objects), while this is completely unacceptable in noise properties for faint
galaxies. As size increases and S/N drops, the degree of difîculty (measured in computational
expense, investigator's time, trouble with subtle instrumental effects, and perhaps number of false
starts) grows. Contrast issues are also important in determining which classes of problems are
amenable to deconvolution. Fig. 2 shows where some wellíknown observations fall in these terms.

Scientiîc Results from Deconvolved Images 361
1 2 3 4
Semimajor axis, arcseconds
18
20
22
24
Raw data Hybrid CLEAN
s-CLEAN
Lucy-Richardson
2D model fit
I
surface
brightness
(mag/arcsec
2
)
53W044 (z=0.311)
WFC/F785LP
Figure 1. Comparison of surfaceíbrightness proîles obtained through modelling and dií
rect comparison with the aberrated data, and using both LucyíRuchardson and oeíCLEAN
deconvolution.

362 Keel
The faintígalaxy results I will stress fall across the bottom in this diagram, which makes them
relatively forgiving subjects for restoration; the dynamic range is limited by signalítoínoise ratio
rather than PSF errors or sampling.
3. HST Science: Structure and Evolution of Galaxies
This could easily be a shopping list of HST's greatest hits. I will exercise the reviewer's arbitrary
prerogative to limit the discussion to a îeld of particular interest where deconvolved images have
had a profound impact, the structure and evolution of galaxies. This has been effective ground for
deconvolution, because many targets have less contrast than, say, QSO host galaxies, and we are
limited by S/N before PSF uncertainties or algorithmic problems. Many of these results require
restoration only to the Nyquist limit of WFC sampling, sidestepping issues of undersampling.
There are many results that have required deconvolution, things we simply wouldn't know
without this capability. The discussion moves roughly in order of typical distance.
3.1. Galactic Nuclei
These regions have proven fertile ground for imaging and deconvolution, being strongly condensed
and bright. Numerous surprises have appeared in such images, especially of active nuclei.
It has long been suspected that active galactic nuclei are driven by central massive objects
(massive black holes) surrounded by accretion disks, sometimes with jets arising along the rotation
axis of the accretion disk. In the radio galaxies NGC 4261 and 3C 449 (Jaffe & Ford 1993), HST
images have revealed dark disklike structures around the nuclei, oriented perpendicular to twin jets
seen at centimeter wavelengths. These structures have characteristic dimensions of parsecs, much
too large to be the accretion disks responsible for the emission properties of these nuclei and mass
loss into the putative black holes, but may represent material which has already settled into orbits
aligned with the yetíunseen inner disk.
Most active galactic nuclei fall into two types, classiîed on the basis of their emissioníline
spectra. Broadíline objects have emission from gas at a wide range of velocities, covering in the
most extreme cases a range 0:1c, while narrowíline objects show similar ionization and energyíinput
requirements but show only emission from lowerídensity gas with characteristic velocity of a few
hundred km s \Gamma1 (which is also often present in broadíline objects). An important breakthrough was
the recognition, driven by polarimetry, that some narrowíline active nuclei possess a dense inner
broadíline region that is blocked from our line of sight, but seen by (polarized) scattered light (as
reviewed by Antonucci 1993). This can ît nicely with an accretionídisk picture if the scattering
region is a thick disk aligned with (perhaps outside) the accretion disk proper. Early PC observations
of the nearby Seyfert galaxy NGC 1068 (Lynds et al. 1991) in fact resolved the continuum emission
of the nucleus into a region of typical size 10 parsecs, perhaps a direct detection of this scattering
region.
Some active nuclei exhibit jets that are detectable in the optical and ultraviolet. Most are
dominated by synchrotron radiation, so that observations in this range are sensitive to highly
relativistic electrons with relativistic Lorentz factors fl ¦ 10 6 as well as to the details of the magnetic
îeld structure in which they radiate. The FOC has proven very effective for these objects, since
they have high surface brightness, small angular size, and greatest contrast against the background
starlight in the blue wavelengths where the FOC is most sensitive. The wellíknown jet in M87 was
a prime HST target, reported by Boksenberg et al. (1992) with the FOC and also observed with
the PC in the deep red (e.g. Lauer et al. 1992a). Restoration shows many of the same features
observed at 2 cm by the VLA (Biretta et al. 1983, not without its own level of deconvolution). Subtle
differences between the radio and UV structures are particularly intriguing, offering the possibility
of tracing the îne structures in which electrons are accelerated and respond to the magnetic îeld
geometry. The subtlety of any such differences helps focus attention on the detailed behavior of
different deconvolution algorithms, taxing îdelity of both intensity and geometry at low count
rates. Observations of additional jets have revealed contrasts between rather smooth structure (PKS

Scientiîc Results from Deconvolved Images 363
Difficulty
JETS
BRIGHT PLANETS
DISTANT CLUSTERS
HIGH-REDSHIFT GALAXIES
ISM FINE STRUCTURE
GALACTIC NUCLEI
Difficulty
ISM FINE STRUCTURE
ANGULAR SIZE
SIGNAL-TO-NOISE
RATIO
Figure 2. Schematic display of factors affecting the difîculty of image restoration of
HST data. Angular size enters primarily through changes in the form of the PSF. Some
wellíknown images are included at the appropriate locations to graphically illustrate the
various regimes relevant to different scientiîc aims.

364 Keel
0521í36, Macchetto et al. 1991a) and hints of the kinds of îlamentary features seen in parts of the
M87 jet (3C 66B, Macchetto et al. 1991b). A further synchrotron jet was discovered serendipitously
in NGC 3862 (Crane et al. 1993).
3.2. Galaxy Mergers and Star Formation
Groundíbased data obtained at a variety of wavelengths have shown that bursts of star formation can
be induced by strong tidal interactions and mergers. The high resolution allowed by deconvolved
HST images has added a new richness to this picture, by showing very luminous blue star clusters.
These have been seen now in the cases of NGC 1275 (Holtzmann et al. 1992), Arp 220 (Dowling
& Shaya 1992), NGC 7252 (Whitmore et al. 1993), and a #quiet# merger in the compact group
NGC 6027 (Seyfert's Sextet; Sulentic et al. 1994). The properties of these clusters open new
windows on the history of mergers, with timescales and cluster ages deduced from their colors and
in some cases spectra. These may be the precursors of globular clusters, an important issue in
understanding the merger history of galaxies.
3.3. Morphology of Distant Galaxies
High angular resolution opens the possibility of seeing transformations in the forms of galaxies
over cosmic time. For any but elliptical and S0 galaxies, morphological classiîcation requires
deconvolution rather than modelling with a manageable number of components. The galaxy content
of rich clusters is a particularly rich îeld for inquiry; the color distributions of cluster members have
long been known to show excess blue galaxies at redshifts z ? 0:3 (the ButcheríOemler effect; see
Lavery, Pierce, & McClure 1992 and references therein). One mechanism for the color change is
shown in a spectacular way by WFC imaging of the cluster 0939+4713 at z = 0:4 (lookback time
5:2 \Theta 10 9 years for H 0 = 50 km s \Gamma1 Mpc \Gamma1 and q 0 = 1=2) presented by Dressler et al. (1994).
While nearby rich clusters are uniformly dominated by E and S0 galaxies, all Hubble types are
present in 0939+4713; many of the blue galaxies are structurally lateítype spirals, along with a few
apparent mergers. WFPC2 should allow similar observations to redshifts z ¦ 0:8. The surprisingly
rapid evolution of the galaxy population in clusters with cosmic time may be due to a combination
of gas stripping from the hot intracluster medium and merging with subsequent starburstídriven
galactic winds.
The ability to discriminate arbitrary structures of small angular scale has allowed detection
of gravitationally lensed galaxies, as in the cluster AC 114 (Couch et al. 1992). This presents the
possibility of measuring structure in extremely distant galaxies, using the #gravitational telescope#
of deep cluster potentials. Distinguishing between even and oddíparity structure in such a lensed
image will in principle allow distinction between features in the background galaxy and foreground
lens, and may eventually allow us to probe structures on angular scales beyond the direct resolution
of even corrected HST optics.
The observation of merging galaxies locally has led to the suggestion that mergers might be an
important part of galaxy evolution, perhaps driving important bursts of star formation and nuclear
activity, changing disk galaxies into ellipticals, and certainly changing the comoving space density
of galaxies. The ability of deconvolved HST images to resolve îne structure at large redshifts
allows a new test for this. Since mergers seen today are dominantly between galaxies that are
gravitationally bound to one another, rather than from random encounters of unrelated galaxies, one
signature of a strong role for mergers will be increasing numbers of galaxy pairs and tight groups
wih increasing redshift and lookback time. Such an effect has indeed been reported, by Burkey et
al. (1994) for pairs in a set of parallel WFC images, and by Casertano et al. (1994) for pairs and
groups from the MediumíDeep Survey. Statistically wellídeîned criteria for such associations give
membership rates of about 7% for nearby galaxies and 34‘40% for faint galaxies (I magnitudes
21‘24, redshifts z = 0.3‘0.7), suggesting that the merger rate changes with redshift approximately
as (1 + z) 3 . This change is important in interpreting counts of faint galaxies, and is interestingly
close to the rate at which QSOs and radio galaxies evolve with cosmic time.

Scientiîc Results from Deconvolved Images 365
Working to yet higher redshift has been more difîcult, but feasible, with spherical aberration.
Galaxies have been found with #normal# morphology and size out to z = 2:4, as in the case of the
radio galaxy 53W002 (Windhorst et al. 1992). This galaxy has an ellipticalílike intensity proîle
and scale length comparable to those of nearby ellipticals with radio sources. Seen at a lookback
time only 1‘2\Theta10 9 years later than the peculiar galaxies at the highest redshift, such objects provide
clues to the timescales of galaxy formation. So far, galaxies with strong central condensation have
been reported out to z = 2:4. Further imaging studies will be crucial in understanding whether
this epoch is when most galaxies approach their present dynamically relaxed structures. We may
already be seeing galaxy formation, if we but had a clue what the process looks like.
The highestíredshift galaxies known are the very powerful radio sources found from the 3C
and 4C surveys. At redshifts as high as z = 3:8, they show spectacular, if poorlyíunderstood,
morphologies. Their optical (that is, emitted UV) images are frequently elongated and clumpy. The
elongation is correlated with the direction of the radio structure in the soícalled alignment effect,
which is in evidence only for z ? 0:5. HST images show striking correlations in detail between
emitted ultraviolet and radio structures, in the cases of 4C 41.17 (z = 3:8, Miley et al. 1992) and
4C 28.58 (z = 2:9, Miley 1993). Connecting these early views of galaxy evolution with what
we see at lower redshifts may be the outstanding problem here. Are these systems rare examples
of galaxian pathology, irrelevant to the evolution of galaxies in general, or are they symptomatic
of many galaxies during a turbulent initial epoch? Once again, only the ability to image larger
samples to fainter levels will clarify these matters. It is intriguing that we see symmetric, centrally
condensed galaxies at redshifts less than 2.5, and that the peculiar elongated and clumpy systems
occur mostly at yet higher redshifts; one might be tempted to call this an evolutionary pathway,
invoking astronomers' wellíknown ability to generalize from a sample of two or more.
4. Science Aided by Deconvolution
There is another category of scientiîc results where deconvolution plays a signiîcant role, even
when the înal measurements use another technique. This operates on the principle that if you don't
know it's there, you can't measure it. In many cases, deconvolved images are important in knowing
what analysis technique to use for the problem at hand, and whether in fact another technique
might be more appropriate. Several examples come to mind from the realm of galaxy structure and
evolution.
4.1. Central Peaks in Galaxy Proîles
The presence of strong central concentrations in the starlight of galaxies is important in understandí
ing their dynamics, and in searching for the massive compact objects thought to exist in (at least)
active galactic nuclei. They have been probed using combinations of modelling and deconvolution
in several galaxies. From an early science veriîcation image, Lauer et al. 1991 found a strong cení
tral peak in the previously undistinguished galaxy NGC 7457. Further PC observations conîrmed
groundíbased measurements of a central #spike# in M87 (Lauer et al. 1992a), and showed that the
central concentration is more pronounced that could have been demonstrated from groundíbased
data alone. A similar concentration also appears in the Local Group elliptical M32 (Lauer et al.
1992b). Pending spectroscopic observations postíCOSTAR, galaxies with these features are the
best candidates for hosting massive central black holes. It is noteworthy that this signature is not
uniquely linked to nuclear activity, suggesting that the prerequisites for such activity are more
common than its actual occurrence.
The core of the Andromeda galaxy, M31, has also been suspected to harbor a massive central
object from groundíbased spectroscopic data (Dressler & Richstone, 1988, Kormendy 1988). The
HST images analyzed by Lauer et al. (1993) show an unexpectedly complex situation # two cores
of different scale size and brightness, separated by 0.49 arcseconds. This discovery accounts for
a longístanding puzzle, the offset of the brightness peak of M31 from the isophotal center (Nieto
et al. 1986). This nucleus was observed using the Stratoscope II ballooníborne telescope (Light

366 Keel
et al. 1974), which showed this central asymmetry; had their image been somewhat deeper the
character of the nucleus might have become apparent. The double core of M31 poses outstanding
theoretical questions. The most likely explanation is that the more compact object is the remnant
core of a galaxy that has merged with M31; to have such a conîguration last long enough to have a
reasonable chance of our observing it probably requires that both nuclei harbor massive black holes.
These conclusions could have been reached from modeling and îtting alone, but deconvolution was
a strong aid in knowing what to ît.
4.2. Modelling Galaxy Images
For more distant galaxies, structural parameters are often better estimated by modelíîtting (Schade &
Elson 1993) # once deconvolution has given assurance that there is not too much nonaxisymmetric
structure. Even in îtting multicomponent models to the aberrated data in a ¼ 2 sense, a deconvolved
image can give initial values for the parameters that vastly speeds convergence to the bestíît set
(Windhorst et al. 1993a, 1993b). Results from such procedures include an angularísize ‘ redshift
relation for îeld galaxies (Grifîths et al. 1994) and identiîcation of weak disks in earlyítype galaxies
(Keel & Windhorst 1993).
5. Current Challenges
Many of the successes listed above are relatively forgiving targets for image reconstruction. In
faintígalaxy work, we are limited more often by signalítoínoise ratio than by knowledge of the
PSF or residual effects from the restoration. For galactic nuclei, the data quality is usually high
enough that some noise ampliîcation can be tolerated. Some of the most challenging restoration
tasks are in the regime of high dynamic range, exempliîed by QSO host galaxies, gravitational
lenses, stellar ejecta, and even comets. In all these cases, reconstruction is vulnerable to small
errors in the PSF, and the inevitable photon noise in lowísignal regions is dominated by the Poisson
noise from aberrated photons properly belonging to bright parts of the image. This problem is
particularly severe for the kinds of problems just listed # where the primary interest is in faint
extended structure close to very bright (usually unresolved) sources. Here, noise ampliîcation can
destroy the interesting signal, since the noise is set by statistics unrelated to the true local brightness.
Nisenson (1994) has described in these proceedings some indirect analysis methods for gravitational
lenses; early imageírestoration work on the multiply imaged QSO PG 1115+080 (Groth et al. 1991;
see also Kristian et al. 1993) amply illustrated how sensitive such results can be to both noise and
PSF accuracy.
The issue of PSF accuracy has been addressed in several contributions. Analysis of existing
imagery has already run into the limitsof present calculations (with TinyTIM) and empirical libraries
in several instances. Crowdedíîeld photometry has reached a precision limited by the our ability
to track changes in the PSF across the WF/PC îeld. Empirical libraries, though guaranteed to have
the right optical mapping, are usually too sparsely sampled near the edges of the WF/PC chips, and
break down for objects with extreme spectral shapes (including strong emission lines). This makes
both wide îeld and high dynamic range restorations some of the most stringent tests for image
restoration techniques, and just as important of our ability to properly compute the appropriate PSF.
6. Where Are We Going Now?
What will optical astronomers be doing with deconvolution in the coming years, especially with
HST? Algorithms and PSF characterization will only improve with experience, so that we will be
able to do a better job on data already in the archive. The amount of data already taken insures a
rich harvest for the taking as our understanding of the instruments and optics improves. Further,
as we have heard here, the postíCOSTAR images will still display residual aberrations that can
be corrected by the techniques that were utterly necessary for handling the aberrations in the îrst

Scientiîc Results from Deconvolved Images 367
place. Crudely put, there will be little excuse for seeing even diffraction rings in fully processed
postíCOSTAR images!
What have optical astronomers learned from handling HST imagery? Any experience that
forces more of the community to understand the dataítaking and data analysis processes more
completely is to that extent a positive step, albeit in this case not one that any of us would have
chosen. Many optical astronomers have now followed their colleagues in radio interferometry into
regular and informed use of deconvolution algorithms. The HST optics have driven a signiîcant
amount of work in algorithm development and in understanding their theoretical underpinnings;
comparison of the presentations here with the îrst imageírestoration workshop shows how much
these applications have matured. More astronomical results than ever before are beneîting from
restoration, and we can look forward to seeing more complete analyses of hardíwon data as a result.
Finally, prospects for adaptive optics on the ground open new avenues for science through image
restoration. Systems in the optical are likely to produce PSFs with a sharp core and signiîcant diffuse
halo; if the PSF can be determined locally, this reduces to something that is, in the mathematicians'
terms, #a problem previously solved.#
References
Antonucci, R. R. J. 1993, ARAA 31, 473
Biretta, J. A., Owen, F. N., & Hardee, P. E. 1983, ApJ, 274, L27
Boksenberg, A. et al. 1992, A&A 261, 393
Burkey, J. M., Keel, W. C., Windhorst, R. A., & Franklin, B. E. 1994, ApJ, submitted
Casertano, S. et al. 1994, BAAS, 25(4), 1399
Couch, W., Ellis, R., & Sharples, R. 1992, STíECF Newsletter 17, 13
Crane, P., et al. 1993, ApJ, 402, L37
Cunningham, C. C. & Anthony, D. 1993, Icarus 102, 307
Dowling, D. & Shaya, E. 1992, BAAS, 24, 728
Dressler, A. & Richstone, D. O. 1988, ApJ, 324, 701
Dressler, A. et al. 1994, ApJ, submitted
Grifîths, R. E. et al. 1994, ApJ, submitted
Groth, E. J. et al. 1991, in The First Year of HST Observations, A. L. Kinney & J. C. Blades, eds.,
Space Telescope Science Institute, Baltimore, 192
Holtzmann, J. et al. 1992, AJ, 103, 691
Jaffe, W. & Ford, H. 1993, reported in Maran, S. P., & Kinney, A. L. 1993, PASP, 105, 447
Keel, W. C. & Windhorst, R. A. 1993, AJ, 106, 455
Kormendy, J. 1988, ApJ, 325, 128
Kristian, J. et al. 1993, AJ, 106, 1330
Lauer, T. R. et al. 1991, ApJ, 369, L41
Lauer, T. R. et al. 1992a, AJ, 103, 703
Lauer, T. R. et al. 1992b, AJ, 104, 552
Lauer, T. R. et al. 1993, AJ, 106, 1436
Lavery, R. J., Pierce, M. J., & McClure, R. D. 1992, AJ, 104, 2067
Light, E. S., Danielson, R. E., & Schwarzschild, M. 1974, ApJ, 194, 257
Lynds, C. R. et al. 1991, ApJ, 369, L31
Macchetto, F. et al. 1991a, ApJ, 369, L55
Macchetto, F. et al. 1991b, ApJ, 373, L55
Miley, G. K., Chambers, K. C., van Breugel, W. J. M., & Macchetto, F. 1992, ApJ, 401, L69

368 Keel
Miley, G. K. 1993, in Science with the Hubble Space Telescope, ESO Conference and Workshop
Proceedings, 44, P. Benvenuti & E. Schreier, eds., European Southern Observatory, Munich,
1
Nisenson, P., et al. 1994, this volume
Ratnatunga, K., et al. 1994, this volume
Schade, D. J. & Elson, R. A. W. 1993, AJ, 105, 1581
Stetson, P. 1994, this volume
Sulentic, J. W., Arp. H., & Raba#ca, C. 1994, ApJ, submitted
Whitmore, B. C., Schweizer, F., Leitherer, C., Borne, K., & Robert, C. 1993, AJ, 106, 1354
Windhorst, R. A., Mathis, D. F., & Keel, W. C. 1992, ApJ, 400, L1
Windhorst, R. A., Gordon, J. M., Mathis, D. F., Neuschaefer, L. W., Pasquerelle, S. M., Schmidtke,
P. C., Ferro, A. J., Keel, W. C., & Burkey, J. M. 1993a, ApJ, submitted
Windhorst, R. A. et al. 1993b, ApJ, submitted