S. M. Pascarelle, R. A. Windhorst
Department of Physics & Astronomy, Arizona State University, Tempe, AZ
85287-1504, USA
W. C. Keel
Department of Physics & Astronomy, University of Alabama, Tuscaloosa,
AL 35487-0324, USA
in Deep Cycle 4--5 WFPC2
images. The
F410M filter (Lyman-alpha at 
) was used, in conjunction with F450W
for continuum subtraction, to find at least 18 potential 
objects
within a 0.5 Mpc 
0.5 Mpc region (H
=80, q
=0). Four
of these have been spectroscopically confirmed to be at 
. All
candidates are
much smaller than the median scale-length of WFPC2 field galaxies at the same
magnitude, with half-light radii 
0.'' 1--0.'' 2 or
0.8--1.5 kpc, and have luminosities in the range M
=-18 to M
=-23 (<0.1--1L
at 
). The very
small scale lengths of these subgalactic clumps may explain why
ground-based Lyman-
primeval galaxy searches have been largely
unsuccessful; in
typical ground-based seeing and sky brightness these would require much longer
exposure times than with WFPC2 and F410M, so that HST holds the secret to
success in searches for such faint, compact objects. We propose that these
sub-galactic clumps could have grown into the luminous giant galaxies (E/S0
and early-type spirals) seen today, through the process of repeated hierarchical
merging (and the subsequent development of disks).
Keywords: galaxies: clusters: general---galaxies: distances and redshifts---galaxies: evolution---galaxies: formation---quasars: general
To explain the formation of the luminous elliptical and grand-design spiral galaxies is one of the most challenging problems in observational cosmology. Two basic scenarios exist at present: one based on the so-called `top-down' models, in which the largest structures in the universe formed first (Sunyaev & Zel'dovich 1975), and then subsequently fragmented into smaller and smaller sub-structures. The other is based on the so-called `bottom-up' models, in which small structures formed first, to eventually merge and coalesce as they build up to become the giant luminous galaxies that we see today (Navarro & White 1994). Several reviews on these topics, and of galaxy formation in general, exist in the literature (Cowie 1988, White 1989, Larson 1990, 1992, Silk & Wyse 1993).
Direct observational evidence of galaxy formation is critical to resolve this issue. The difficulty lies in finding large numbers of faint and presumably distant galaxies at or near their time of formation. Unfortunately, there exist only a few known cases of galaxy clusters or groups with spectroscopic confirmation at high redshift (e.g., Lowenthal et al. 1991, Steidel, Dickinson, & Sargent 1991, Dressler et al. 1993, Giavalisco, Steidel, & Szalay 1994, Hutchings 1995, Francis et al. 1996).
A piece of the puzzle of galaxy formation may come from recent findings on the
subject of Faint Blue Galaxies (FBGs). The existence of this field galaxy
population has been known for some time (Kron 1982, Broadhurst, Ellis, &
Shanks 1988), but the true nature and
evolution of the FBGs has remained a mystery. Recent results from deep HST
images with the refurbished WFPC2 show that the FBG population is dominated by
late-type or irregular galaxies (Driver et al. 1995) that have undergone
substantial evolution since z
1.
Many of these late-type/irregulars
were found to be very compact galaxies with scale lengths, or half-light radii,
0.'' 3--0.'' 4 (Driver et al.
1995, Casertano et al. 1995, Odewahn et al. 1996). They have the
steepest object counts,
and clearly dominate the FBG counts at the faintest
flux levels. Odewahn et al. (1996) suggest that a good fraction of these
compact FBGs may play a role in the formation of giant galaxies at
z
1.
On the other hand, galaxy counts from deep HST images, along with the scale
length--redshift (or 
--z) relation, of luminous
early-type galaxies (E/S0's) and mid-type spiral galaxies (Sabc) indicate that
they have been assembled largely before z
1 and have
undergone little or
no evolution since z
1 (Mutz et al. 1994, Driver
et al. 1995, Odewahn et al. 1996). If the formation of luminous
early- to mid-type galaxies occurred sometime before z
1,
then it is
possible that they could have been assembled from the gradual merging of at
least a sub-fraction of these compact, late-type FBGs. There is increasing
evidence from both HST and ground-based work that the galaxy merger rate may
have been larger at earlier cosmic epochs, roughly increasing with redshift as
with m
2--3 (Burkey et al. 1994, Carlberg et
al. 1994, Yee & Ellingson 1995). It is thus
possible that the FBG population provided the reservoir of building blocks out
of which many of the luminous galaxies that we see today were formed through the
process of repeated hierarchical merging (Navarro & White 1994). This is
consistent with recent findings of a population of faint (B
24) blue galaxies at 1
z
2 with rather unusual morphologies suggestive of dynamical
formation processes or mergers (Cowie, Hu, & Songaila 1995).
Here we report our discovery of a cluster of 
18 subgalactic-sized
objects at 
. If these objects are typical of the building blocks
from which the giant luminous galaxies were made, they would have eluded most
ground-based searches in the past because of their small scale lengths, their
low flux levels, and possibly their clustering properties (or rather any
large-scale structure present at high redshifts).
The initial discovery of a possible cluster or group at 
was made
from ground-based photometry with a medium-band (150Å-wide) filter
centered at 4130Å (Lyman-
at 
) in the field around
the weak radio galaxy 53W002 at z=2.390 (Windhorst et al. 1991,
Windhorst & Keel 1996), resulting in two
other candidates at the same redshift (Pascarelle et al. 1996a). These
were later spectroscopically confirmed with the Multiple Mirror Telescope (MMT)
, as shown in Fig. 1. Fortunately, the existence of a nearly identical
medium-band filter on HST (F410M, centered at 4100Å) allowed the same
observations to be conducted with higher sensitivity at much higher spatial
resolution than could be achieved from the ground.
Figure: MMT spectra of four of the 
cluster candidates. The
spectra for 53W002 (object 6) and object 19 were taken with the Blue
Spectrograph and the 300 gpm grating, and have been smoothed to 
20Å
resolution (Windhorst et al. 1991). The
spectra for objects 18 and 12 were taken with the Red Spectrograph and the 150
gpm grating, and have been slightly smoothed to 
50Å (Pascarelle
et al. 1996a).
The new HST data are presented in the color-color diagram of Fig. 2,
which shows
photometry for 115 objects detected simultaneously in the broad-band WFPC2
filters F606W (V) and F450W (B), and in the medium-band filter F410M
(Lyman-
at 
). As with the ground-based data, the
Lyman-
passband is contained entirely inside the 
filter,
so that proper continuum subtraction is possible. At 
, these bands
sample well shortward of the redshifted 4000Å break,
where young galaxy spectral energy distributions are relatively
featureless (Windhorst et al. 1991).
Figure: The (F410M--
) vs. (
--
)
color-color diagram from 51 orbits in Cycles 4--5 on the 
cluster
in the 53W002 field. The plot contains 115 objects which were detected in all
three filters. The best signal-to-noise colors were determined from sufficiently
large object apertures selected to be the same in all three filters. The
sub-pixelated images were registered to well within 1 pixel. The sky was
interpolated underneath the object aperture by fitting a sloped plane to the
pixels unaffected by faint neighbors, hence correcting for any remaining small
gradients in the flat-fielded images. Compared to automated object finding in
the same field (Odewahn et al. 1996), the sky estimates are consistent
to within 0.07% and
the fluxes on average within 0.05 mag. Including the WFPC2 zeropoints of
Holtzman et al. (1995), photometry presented here is accurate to
0.05--0.10 mag.
Another 
250 objects were detected in 
and 
(down to 
26.0 mag) but not in the F410M
image, which was unavoidably underexposed. They have 2
upper limits of
(F410M--
)
-0.2 mag and are not plotted
here. For 26.0
27.5
mag the 
-band sample is 
90%
complete (Odewahn et al. 1996), but the underexposed F410M image does
not provide useful Lyman-
upper limits for such faint objects.
The solid line in Figure 2 shows the expected relation for field objects with a
featureless
power-law (F
) across the three adjacent filters,
labeled with values of 
, around which most of the general field objects
(open circles) are distributed. Approximate 2
error boundaries
for their WFPC2 photometry are indicated by the dotted lines. Error bars are
plotted for 18 objects that are at least 2
away from the power-law
line, which we believe are significant 
cluster candidates. The
four large triangles have a spectroscopic confirmation at 
, out
of five objects for which spectra were obtained, indicating that the
reliability of this method to find compact 
candidates is about
80%.
Apart from the three brighter objects, which were easily seen in ground-based
photometry (Pascarelle et al. 1996a) and
contain a weak active galactic nucleus, the rest of the sample has
Lyman-
emission with equivalent widths (estimated from the F410M
photometry) more typical of ionization arising from star formation.
Although some of these cluster candidates could be foreground galaxies with O
II] (at 3727Å) redshifted into the F410M filter (at 
), such
intrinsically faint and compact objects would be quite unusual at such a low
redshift. In addition, the differential volume element at 
is
18--86 times bigger than at 
for q
=0--0.5, and
no other strong emission lines exist for star-forming objects
between 1216Å and 3727Å, implying that most of the 18 significant
candidates are likely at 
.
The restframe ultraviolet reddening vector (Seaton 1979) expected at 
is indicated in Figure 2, and suggests that the few reddest cluster
candidates may have a visual absorption of 
2--3 mag. These reddest cluster
candidates should be viewed with some caution, however, because of the slight
dependence of the 
magnitudes on (
--
) color, which increases towards redder (
--
) (Holtzmann et al. 1995). We corrected for this to first order,
but higher order terms
may affect the reliability of the few 
candidates with (
--
)
1.5 mag.
The reliability of the 18 cluster candidates in Fig. 2 is further strengthened
by the fact that two of the spectroscopically confirmed members are actually at
the lower boundary in the (F410M--
) color of the general field
population. It is of course possible that there exist additional cluster members
that are not in Fig. 2 because they are significantly obscured by dust.
Figure: ( top) Histogram of WFPC2 continuum scale lengths for the
significant narrow-band redshifted Lyman-
emitting 
candidates in the 53W002 field from Figure 2. Scale lengths were determined
from the average
of their half-light radii in 
, 
, and 
. Most 
candidates have compact cores with 
0.'' 1--0.'' 2 or 
0.5--1.5
kpc, which is why they are relatively easy to detect with HST. ( bottom)
Histogram of WFPC2 
luminosities for
the significant 
Lyman-
emitting candidates. The upper
axis indicates their absolute magnitude distribution, assuming K-corrections
for young stellar populations at 
from Windhorst et al. (1991).
Figure 3 ( top) shows the size distribution for the cluster
candidates and indicates that most are very compact, with 
0.'' 1--0.'' 2, or 
0.5--1.5
kpc at 
(all sizes in the text are quoted using
H
=80
Mpc
and the range q
=0--0.5). Each object is typically seen at
least 3--4 scale lengths out in the WFPC2 images (see Fig. 4). Most candidates
are much smaller than the median scale length of WFPC2 field galaxies at the
same flux levels, which is about 0.'' 3--0.'' 4 (Casertano et al. 1995).
Average light profiles were generated for the cluster candidates, assuming that
they are all to first order similar in shape and size. These profiles are shown
in Fig. 4, and are in each filter better fit by an 
bulge-like profile
than by a disk-like exponential. To derive a mean intensity profile with the
greatest dynamic range, average intensity-weighted composite images of the 14
compact and isolated 
candidates were produced in each
filter, so that the total effective exposure times of the image stacks
become 14
5.7 hours.
To measure the scale length of the mean observed profile, model profiles were
convolved with an empirical point-spread function taken from a star in the
images (for the undersampled 
and 
images an
additional `pyramid' function was used to account for subpixel shifts among the
objects, since shifting by interpolation introduces artifacts in undersampled
images). The continuum scale lengths of the cluster
candidates are very similar in 
and 
, showing no dependency on restframe wavelength below the 4000Å
break (Odewahn et al. 1996). Their half-light radii were measured to be
0.'' 11--0.'' 14 in 
and 0.''
10--0.'' 12 in 
, consistent with a mean value of 0.'' 12
(
0.5--1.0 kpc at z
).
Figure: Light profiles of the intensity-weighted composite images of 14
compact and isolated 
cluster candidates in the 
and 
filters. The profiles are reliable for radii
between the two vertical dotted lines, but are affected by point spread
function and pixel-interpolation problems at smaller radii, and by errors in
the sky-determination at larger radii. In each filter, the profiles
are better-fit by an 
-law than by a disk-like exponential. The allowed
range of effective radii in WFC pixels (=0.'' 0996) is:
1.1--1.4 for 
(
=1.30) and
1.0--1.2 for 
(
=1.12), corresponding to 
0.5--1.0
kpc at 
(H
=80, q
=0--0.5).
The question arises as to whether or not we are seeing the full extent of these
objects. The K-correction for young spectral energy distributions at 
could have compensated for at least some of the cosmological 
surface brightness dimming (Windhorst et al. 1991). With the exception
of 53W002 itself, which already has at 
a well-developed
-law profile with a large scale length (
0.'' 72)
indicating a massive early-type galaxy (Windhorst,
Mathis, & Keel 1992, Windhorst et al. 1994, Windhorst & Keel 1996),
and was by selection included in this WFPC2 field, it appears that none of the
other cluster candidates are yet fully assembled massive ellipticals or
grand-design spirals. However, their scale lengths are quite comparable to
those of bulges in local spirals (Simien & de Vaucouleurs 1987), which range
from 0.2--4 kpc with a type-dependent median that is close to 1 kpc for types
S0--Sbc. Therefore, despite their small sizes, these objects are not unusually
small for the bulges of early- to mid-type disk galaxies. Given that the
ratio of bulge-to-disk scale lengths of nearby late-type galaxies is 
(Courteau, deJong, & Broeils 1996), these objects may be
subgalactic-sized (compact) and young (blue)
spheroids, possibly representing the bulges of young galaxies that have not
(yet) developed a significant disk around them at 
, or a disk
that is depressed in the HST images by the severe cosmological surface
brightness dimming.
The implied luminosities at 
for all cluster candidates are shown
in Fig. 3 ( bottom), and typically range from M
-23 to
-18 mag (based on the stellar population models, age estimates, and
K-corrections as described in Windhorst et al. 1991). With multicolor
BVI photometry, the K-corrections are straightforward, provided that their
redshifts are known to be at 
either from spectroscopy or their
(F410M-
) colors). Subtraction of
any contributions from active galactic nuclei were also performed following
the point-source subtraction method of Windhorst, Mathis, & Keel (1992).
Some of the faintest candidates have 
25--26 mag (the figure becomes incomplete for 
25.5 mag due to the lower sensitivity
in F410M than in F450W), so that it will require the concentrated efforts of the
world's largest telescopes to confirm the redshifts of all 18 
candidates spectroscopically. Given the apparent completeness limit, the initial
(luminous) mass spectrum of this cluster could thus be quite steep.
The (evolving) absolute magnitude of an L
galaxy
at 
(M
mag) was estimated by assuming that
there would have been 
2 mag of stellar evolution since 
for
a typical starburst 
0.3--0.5
10
years earlier (as the
unreddened colors suggest, see Fig. 2; Windhorst et al. 1991,1994,
Bruzual & Charlot 1993). Therefore, if indeed their stellar populations are
young, most of the cluster candidates have luminosities of 
0.1--1 L
, and so possibly had only 10
--10
M
processed into stars at 
. For these parameters, the
free-fall time expected for these clumps is 
20--40
10
years,
long enough that the short-lived O and B stars are gone, but much shorter than
the age of the dominant stellar population (A stars). Hence, there was indeed
enough time for their mass distributions to settle into regular 
-like
light profiles.
Most of the objects
that were seen in 
and 
were also detected in
, limiting the maximum redshift sampled in this 
2.'
4
2.' 4 field to typically z
3.5--4, or else
the expected Lyman limit would have significantly damped out the light in the
images (Guhathakurta, Tyson, & Majewski 1990). We suggest that
these subgalactic-sized objects exist throughout
the entire redshift range 
--4, and could have grown into the luminous
giant galaxies (ellipticals and early-type spirals, with disks subsequently
growing through accretion) seen today through the
process of repeated hierarchical merging (Navarro & White 1994). The
epoch-dependent merger rate,
mentioned earlier and roughly 
, would result in a time integral
of 
10--20 mergers of compact objects from 
to z=0, which
could yield a few L
galaxies today. Since the HST counts of early-type
galaxies show little evolution since z
1, this process of
repeated merging
would have to be largely complete by 
. Our 18 cluster candidates could
thus have merged to produce a few L
galaxies today. The total 
luminosity for the 18 objects is M
-24.7 to -25.8 at
, which agrees quite well with the combined luminosities of a few
typical L
galaxies today of M
-22.3 (including the
expected 
-2 mag from their K-corrections plus evolution; Windhorst
et al. 1991).
We note that the substantial number of luminous
galaxies found at z
1 by recent redshift surveys (Cowie,
Hu, & Songaila 1995, Koo et al. 1996) is not
inconsistent with our finding of small subgalactic-sized objects at
. Our detection of this substantial, previously unrecognized
population derives from their weak Lyman-
fluxes, near the limit
of ground-based narrow-band imaging surveys, and their characteristically
small sizes, so that the HST images realize nearly the full point-source
sensitivity gain over ground-based data. The medium-band
HST images are more efficient in selecting samples of faint compact
Lyman-
emitting candidates, but ground-based spectroscopic confirmation
is still necessary for unambiguous redshift determinations.
We find that only about 5% of the faint blue objects in our field can be
classified as the `chain galaxies' of Cowie, Hu, & Songaila (1995), which is
lower than the 20%--50% estimated in their sample. We believe that such
objects are likely the short-lived (
3
10
yr
out of a total of 
6
10
yr available for
q
=0 and z
1) merger events among the many faint
blue subgalactic clumps, in which the gas is drawn out of the merging objects
during the encounter (Navarro & White 1994, Mihos 1995).
In conclusion, we believe that a non-negligible fraction of the compact FBGs
observed in modern deep galaxy surveys may be such high-redshift
subgalactic-sized clumps. The merger rate would have to have been much higher in
the past in order to produce the little evolving early-type populations observed
out to 
. It is thus possible that these subgalactic clumps
may be hiding as
many of these compact FBGs, and have escaped proper recognition from the ground
until now because they are so small. Future work to test the universality of our
findings in the 53W002 field is necessary. Random WFPC2 observations through the
F450W and F410M filters will tell whether our 
cluster is unique
in the early universe, or is typical of the general redshift distribution at
2.4 (c.f., Francis et al. 1996), which may be arranged in
sheets on
scales up to 
125--156 Mpc as seen at lower redshifts (Broadhurst, Ellis,
& Shanks 1988, Landy et al. 1996). The results of the work presented
here are presented in detail in Pascarelle et al. (1996b).
All image data were obtained with the NASA/ESA Hubble Space Telescope through the Space Telescope Science Institute (STScI), which is operated by the Association of Universities for Research in Astronomy, Inc. (AURA), under contract to NASA. The spectroscopic observations were obtained at the Multiple Mirror Telescope Observatory, a joint facility of the University of Arizona and the Smithsonian Institution. We thank Simon Driver for providing his models for the late-type galaxies and for helpful discussions. We acknowledge support from HST grants GO.5308.0*.93A and GO.5985.0*.94A (to RAW & WCK).
Allington-Smith, J. R., Ellis, R. S., Zirbel, E. L., & Oemler, A. 1993, ApJ, 404, 521
Broadhurst, T. J., Ellis, R. S, & Shanks, T. 1988, MNRAS, 235, 827
Broadhurst, T. J., Ellis, R. S., Koo, D. C., & Szalay, A. S. 1990, Nature, 343, 726
Bruzual, A. G. & Charlot, S. 1993, ApJ, 405, 538
Burkey, J. M., Keel, W. C., Windhorst, R. A., & Franklin, B. E. 1994, ApJ, 429, L13
Calzetti, D. & Kinney, A. L. 1992, ApJ, 399, L39
Carlberg, R. G. et al. 1994, ApJ, 435, 540
Casertano, S., Ratnatunga, K. U., Griffiths, R. E., Im, M., Neuschaefer, L. W., Ostrander, E. J., & Windhorst, R. A. 1995, ApJ, 453, 599
Charlot, S. & Fall, S. M. 1993, ApJ, 415, 580
Colless, M., Schade, D., Broadhurst, T. J., & Ellis, R. S. 1994, MNRAS, 267, 1108
Courteau, S., deJong, R. S., & Broeils, A. 1996, ApJ, 457, L73
Cowie, L. L. 1988, in The Post Recombination Universe, ed. N. Kaiser & A. N. Lazenby, (Dordrecht, Kluwer), p. 1
Cowie, L. L., Hu, E. M., & Songaila, A. 1995, AJ, 110, 1576
Cowie, L. L., Hu, E. M., & Songaila, A. 1995, Nature, 377, 603
Crampton, D., Le 
, O., Lilly, S. J., & Hammer,
F. 1995, ApJ, 455, 96
Dressler, A., Oemler, A., Gunn, J. E., & Butcher, H. 1993, ApJ, 404, L45
Driver, S. P., Windhorst, R. A., Ostrander, E. J., Keel, W. C., Griffiths, R. E., & Ratnatunga, K. U. 1995, ApJ, 449, L23
Driver, S. P., Windhorst, R. A., & Griffiths, R. E. 1995, ApJ, 453 , 48
Francis, P. J. et al. 1996, ApJ, 457, 490
Giavalisco, M., Steidel, C. C., & Szalay, A. S. 1994, ApJ, 425, L5
Glazebrook, K. et al. 1995, MNRAS, 273, 157
Guhathakurta, P., Tyson, J. A., & Majewski, S. R. 1990, ApJ, 357, L9
Holtzman, J. A., Burrows, C. J., Casertano, J., Hester, J. J., Trauger, J. T., Watson, A. M., & Worthey, G. 1995, PASP, 107, 1065
Hutchings, J. B. 1995, AJ, 110, 994
Koo, D. C. & Kron, R. G. 1992, ARA&A, 30, 613
Koo, D. C. et al. personal communication
Kron, R. G. 1982, Vistas in Astronomy, 26, 37
Landy, D. L., Shectman, S. A., Lin, H., Kirshner, R. P., Oemler, A. A., & Tucker, D. 1996, ApJ, 456, L1
Larson, R. B. 1990, PASP, 102, 709
Larson, R. B. 1992, in Star Formation in Stellar Systems, ed. G. Tenorio-Tagle, M. Prieto, & F. Sanchez, (Cambridge, UK, Cambridge Univ. Press, p. 125
Lowenthal, J. D., Hogan, C. J., Green, R. F., Caulet, A., Woodgate, B. E., Brown, L., & Foltz, C. B. 1991, ApJ, 377, L73
Lowenthal, J. D., Hogan, C. J., Green, R. F., Woodgate, B. E., Caulet, A., Brown, L., & Bechtold, J. 1995, ApJ, 451, 484
Le 
, O., Crampton, D., Hammer, F., Lilly, S.
J., & Tresse, L. 1994, ApJ, 423, L89
Mihos, J. C. 1995, ApJ, 438, L75
Mutz, S. B. et al. 1994, ApJ, 434, L55
Navarro, J. F. & White, S. D. 1994, MNRAS, 267, 401
Neuschaefer, L. W. & Windhorst, R. A. 1995, ApJ, 439, 14
Odewahn, S. C. et al. 1996, ApJ, submitted
Pascarelle, S. M., Windhorst, R. A., Driver, S. P., Ostrander, E. J., & Keel, W. C. 1996, ApJ, 456, L21
Pascarelle, S. M., Windhorst, R. A., Keel, W. C., & Odewahn, S. C. Nature (in press)
Seaton, M. J. 1979, MNRAS, 187, 73P
Silk, J. & Wyse, R. F. 1993, Phys. Rep., 231, 293
Simien, F. & de Vaucouleurs, G. 1987, ApJ, 302, 564
Steidel, C. C., Dickinson, M., & Sargent, W. L. W. 1991, AJ, 101, 1187
Sunyaev, R. A. & Zel'dovich, Ya. B. 1975, MNRAS, 171, 375
Thompson, D., Djorgovski, S. G., & Beckwith, S. 1994, AJ, 107, 1
Thompson, D., Djorgovski, S. G., & Trauger, J. 1995, AJ, 110, 963
Thompson, D. & Djorgovski, S. G. 1995, AJ, 110, 982
White, S. D. M. 1989, in The Epoch of Galaxy Formation, ed. C. S. Frenk et al., (Dordrecht, Kluwer), p. 15
Windhorst, R. A. et al. 1991, ApJ, 380, 362
Windhorst, R. A., Mathis, D. F., & Keel, W. C. 1992, ApJ, 400, L1
Windhorst, R. A., Fomalont, E. B., Partridge, R. B., & Lowenthal, J. D. 1993, ApJ, 405, 498
Windhorst, R. A., Gordon, J. M., Pascarelle, S. M., Schmidtke, P. C., Keel, W. C., Burkey, J. M., & Dunlop, J. S. 1994, ApJ, 435, 577
Windhorst, R. A. & Keel, W. C. 1996, ApJ, submitted
Yee, H. K. C. & Ellingson, E. 1995, ApJ, 445, 37