THE ASTROPHYSICAL JOURNAL, 512 :30È47, 1999 February 10
1999. The American Astronomical Society. All rights reserved. Printed in U.S.A.(
CALTECH FAINT GALAXY REDSHIFT SURVEY. VIII. ANALYSIS OF THE FIELD J0053]12341
JUDITH G. COHEN,2 ROGER BLANDFORD,3 DAVID W. HOGG,3,4,5 MICHAEL A. PAHRE,2,5,6 AND PATRICK L. SHOPBELL2
Received 1998 June 11 ; accepted 1998 September 17
ABSTRACT
The results of a spectroscopic investigation of a complete sample of objects with mag in a 2@K
s
\ 20
by Ðeld at J005325]1234 are reported. Redshifts were successfully obtained for 163 of the 1957@.3
objects in the sample; these redshifts lie in the range [0.173, 1.44] and have a median of 0.58 (excluding
24 Galactic stars). The redshift identiÐcations are believed to be almost complete for z \ 0.8. Approx-
imately one-half of the galaxies lie in Ðve narrow redshift features with local velocity dispersions of D300
km s~1. These narrow redshift ““ peaks ÏÏ are primarily populated both by absorption-line galaxies and the
most luminous galaxies in the sample, although the incidence of emission lines in the luminous galaxies
increases with redshift. The estimated dynamical masses of these redshift peaks, and the sky distribution
of the galaxies within them, appear similar to groups or poor clusters of galaxies in the local universe at
various stages of virialization. Some groups of galaxies therefore form at epochs z [ 1.5, and the galaxies
in such groups appear to be coeval and to show little sign of ongoing star formation. The galaxies
outside the redshift peaks are also clustered, albeit more weakly, are less luminous and more frequently
exhibit strong emission lines. These ““ isolated ÏÏ galaxies therefore appear, on average, to form stars at
later epochs than the strongly clustered galaxies. The galaxy spectral energy distributions (SEDs) derived
from our UBV RIK photometry are also very closely correlated with the galaxy spectral types and lumi-
nosities. These results have strong implications for the analysis of redshift surveys at intermediate red-
shift. The sample is used to investigate the evolution of the combined galaxy luminosity function back to
z \ 0.8. No signiÐcant change is found in the characteristic luminosity L*, and only weak color changes
are detected, consistent with passive evolution. The blue galaxy-luminosity function is more dwarf rich
than the red galaxy-luminosity function. No signiÐcant change in the comoving density is found in this
sample out to z D 1.4, assuming that the objects without redshifts (16% of the sample) are galaxies,
essentially all of which have z [ 0.8. This suggests that mergers are not important among the objects in
this sample. A population of extremely red objects with (R[K) [ 5 mag exists in the infrared-selected
sample; all four such objects with redshifts are found to be absorption-line galaxies with z D 1. Most of
the very red objects therefore appear to be galaxies with that are not heavily reddened by dust. Az Z 1
measure of the UV extinction at 2400 for the emission-line galaxies of a factor of 2 is obtained, imply-A
ing only modest UV extinction in high-redshift star-forming galaxies.
Subject headings: cosmology: observations È galaxies: distances and redshifts È galaxies: evolution È
galaxies: fundamental parameters È galaxies: luminosity function, mass function È
surveys
1. INTRODUCTION
We observe that galaxies evolve. The number of very
faint (and, presumably, distant) galaxies on the sky is at
least 8 ] 1010 (Williams et al. 1996), roughly 30 times more
than would be naively predicted on the basis of the co-
moving volume of the universe and the local bright-galaxy
luminosity function (Loveday et al. 1992; Marzke, Huchra,
& Geller 1994; Lin et al. 1996a; Gardner et al. 1997 ; Hogg
et al. 1997; Ratcli†e et al. 1998). In addition, faint galaxies
are bluer (Koo & Kron 1992 ; Smail et al. 1995), smaller
(Smail et al. 1995; Griffiths et al. 1994b), and more irregular
1 Based in large part on observations obtained at the W. M. Keck
Observatory, which is operated jointly by the California Institute of Tech-
nology and the University of California
2 Palomar Observatory, Mail Stop 105-24, California Institute of Tech-
nology, Pasadena, CA, 91125
3 Theoretical Astrophysics, California Institute of Technology, Mail
Stop 130-33, Pasadena, CA, 91125
4 Current Address : Institute for Advanced Study, Olden Lane, Prin-
ceton, NJ, 08540
5 Hubble Fellow
6 Current Address : Harvard-Smithsonian Center for Astrophysics, 60
Garden St., Mail Stop 20, Cambridge, MA, 02138
(Griffiths et al. 1994a, 1994b; Glazebrook et al. 1995;
Driver, Windhorst, & Griffiths 1995; Abraham et al. 1996;
Odewahn et al. 1996) than bright galaxies at the present
day.
In order to translate these observations into a description
of the physical evolution of the whole galaxy population
with cosmic time, it is necessary to understand the galaxy
redshift distribution. For this reason much telescope time
has been devoted to large redshift surveys, which allow us
to measure the luminosities and ages of statistical samples
of galaxies. In addition, they also explore, directly, the
variation with cosmic time of the star formation rate, the
chemical abundance, and nonthermal activity, which are
strong markers of the evolutionary history of galaxies. Fur-
thermore, these surveys can be used, globally, to study the
clustering and spatial distribution of faint galaxies.
The Caltech Faint Galaxy Redshift Survey (CFGRS) is
designed to measure the properties of Ðeld galaxies in the
redshift interval The survey is described in0.3 [ z [ 1.3.
Cohen et al. (1999a), where it is contrasted with the many
other recent redshift surveys of Ðeld galaxies. The CFGRS
uses complete samples to a Ðxed limiting magnitude in a
particular bandpass within a small solid angle on the sky.
30
REDSHIFT SURVEY 31
Spectra are obtained for every object in the sample with the
Low-Resolution Imaging Spectrograph (LRIS ; Oke et al.
1995) on the 10 m Keck Telescope.
This paper presents spectroscopic results from one Ðeld
located at J005325]1234, the central region of which is
part of the Medium Deep Survey (Griffiths et al. 1994a) and
among the deepest Ðelds imaged with the Hubble Space
T elescope (HST ) prior to the Hubble Deep Field. The Ðeld
measures 2@ by with a statistical sample containing 1957@.3
infrared-selected objects complete to K \ 20 mag, of which
24 are spectroscopically conÐrmed Galactic stars and 32
cannot be assigned spectroscopic redshifts. (21 of these have
spectra). A preliminary report on the Ðeld J0053]1234 was
given by Cohen et al. (1996b), who showed that roughly
one-half of the galaxies are located in dense groups. The
sample deÐnition and UBV RIK photometric catalog are
presented in a companion paper (Pahre et al. 1999), while
the redshifts are given in a second companion paper, Cohen
et al. (1999a). The properties of the Galactic stars are dis-
cussed in Reid et al. (1997). Large-scale structure will be
discussed, in the context of this survey, in a forthcoming
paper (Cohen et al. 1999c). A redshift survey of another
region, the Hubble Deep Field (HDF; Williams et al. 1996),
will be the subject of the next paper in this series (Cohen et
al. 1999b).
This paper is structured as follows. The galaxy lumi-
nosities and colors in the observed frame are discussed the
° 2. The extremely red galaxies in our sample are discussed
in ° 2.3. Then the galaxy spectral energy distributions
(SEDs) are constructed in the rest frame, and it is shown
that these are strongly correlated with the galaxy spectral
classes in ° 3. On this basis, redshifts are provisionally
assigned for the 32 galaxies whose spectra lack adequate
line features in our complete sample. After using our SEDs
to deduce the ultraviolet extinction for strong emission-line
galaxies with z D 1.2 in ° 3.4, we then discuss the overall
distribution of galaxies in redshift and comoving volume.
Next, we distinguish the clustering of the di†erent spectral
classes in ° 4.2 and demonstrate that the prominent groups
contain apparently older galaxies. The limited morphologi-
cal information that we have on our sample is presented
along with some predictions in ° 6. Finally, our conclusions
are collected and discussed in ° 7, along with some sugges-
tions for future research.
We adopt the values km s~1 Mpc~1 andH
0
\ 60 )
M
\
0.3 with " \ 0 throughout. The present age of the universe
in this cosmology is 13 Gyr and the current cosmological
density is 2 ] 10~30 gcm~3.
2. GALAXY PROPERTIES IN THE OBSERVED FRAME
2.1. Hubble Diagrams
The conventional way to represent the results of redshift
surveys is with a Hubble diagram. Three Hubble diagrams,
B(z), R(z), and K(z) are presented as Figures 1a,1b, and 1c,
respectively, for the redshift sample. The four di†erent
galaxy spectral classes deÐned in Cohen et al. (1999a) (““ E ÏÏ
for emission-lineÈdominated galaxies, ““ A ÏÏ for galaxies
showing only absorption features, ““ C ÏÏ for galaxies with
composite spectra, and ““ Q ÏÏ for active galactic nuclei) are
distinguished on these diagrams. (A few galaxies were not
detected at B, and these are ignored in Fig. 1a.) We also
exhibit the magnitudes of the 32 objects from the galaxy
sample that do not have redshifts when these are detected.
As is usual, we ignore reddening internal to the galaxies
themselves and from the intergalactic medium (IGM).
As the rest wavelengths are blueshifted with respect to
our six bandpasses, it is also conventional to correct the
magnitudes for this e†ect using k-corrections. These have
been tabulated by Poggianti (1997), who conveniently
separates this purely spectral e†ect from the change due to
the evolution of the stars that we observe locally. Together,
these two e†ects are called ““ passive ÏÏ evolution. (We have
corrected for our di†erent Ðlters and cosmology.) The spec-
tral evolution predicted by Poggianti for a galaxy with
L \ 2L* with no evolution is also exhibited on the Hubble
diagrams for local galaxies with SEDs of elliptical, Sa, and
Sc types.7 The passive-evolution predictions from Poggianti
(1997) are also shown in Figure 1, again for a 2L* galaxy.8
An enormous range in galaxy luminosities is apparent in
Figure 1. For intermediate-redshift galaxies observed in the
R-band, there is a 5 mag spread in luminosities. The second
notable feature of this diagram is the concentration of
roughly one-half of the galaxies, predominantly A galaxies,
in redshift clumps. A third peculiarity is the dearth of gal-
axies, especially A galaxies, for andz [ 0.3 0.8 [ z [ 1.1.
Fourth, we note that the Poggianti (1997) passive-evolution
tracks do not appear to represent the observed run of gal-
axies, particularly at B and R. Including the evolutionary
correction term for passive evolution leads to luminosities
that decrease as z increases, while use of just the k-
correction term (i.e., no evolution) is in fact better.
It has long been argued that elliptical galaxies in clusters
of galaxies show only passive evolution, at least out to
z D 0.5 et al. 1993; Kelson et al. 1997;(Arago
n-Salamanca
Pahre 1998; Postman, Lubin, & Oke 1998). Evolutionary
brightening of disk galaxies in the Ðeld is also quite modest
as measured using surface photometry (Schade et al. 1996;
Barger et al. 1998a) or the Tully-Fisher relation (Vogt et al.
1996, 1997), reaching only 0.5 mag in the B-band at z D 0.5.
More recently, Hogg (1998) in an analysis of the HDF has
shown that the evolution of is at most modest out toL
B
*
z D 1. Hamilton (1985) has demonstrated that the ampli-
tude of the 4000 jump is approximately constant forA
bright Ðeld elliptical galaxies out to z D 0.8.
2.2. Galaxy Colors
A conventional view of galaxy colors is given in Figures
2a,2b, and 2c, which show the dereddened galaxy colors
U[R, R[K, and U[K, respectively, for the full sample as
a function of redshift. Galaxies without a detection at U are
not plotted. The same symbols as in Figure 1 are used to
indicate the galaxy spectral types. The thin and thick lines
represent the predictions of the Poggianti (1997) models for
no evolution and for passive evolution, respectively. For
galaxy colors, only di†erences of the values of the evolution-
ary corrections computed in each of two colors for the
passive-evolution models of Poggianti (1997) are relevant,
7 The SED for the local elliptical is from Bruzual & Charlot (1993).
8 The k-correction assigned to an elliptical galaxy as computed from the
Worthey (1994) and Bruzual & Charlot (1999) models (as provided in
Leitherer et al. 1996) show good agreement with each other, and with the
Bruzual & Charlot (1993) model described above. They also show good
agreement with the ““ empirical ÏÏ approach to estimating k-corrections by
Cowie et al. (1994), which was based on broadband photometry of nearby
elliptical galaxies. These k-corrections di†er substantially by D0.4 mag
from the Poggianti (1997) models at z D 0.6 for the K Ðlter, suggesting that
the latter models are problematical at near-infrared wavelengths.
32 COHEN ET AL. Vol. 512
FIG.1a FIG.1b
FIG.1c
FIG. 1.ÈHubble diagrams for the main sample for three observed magnitudes, (a) B,(b) R, and (c) K, using the photometry of Pahre et al. (1999). The
abscissa is the redshift z rather than the conventional log z. Also shown on the upper scale is the lookback time for our adopted cosmography. The ordinate
is the magnitude expressed as an equivalent Ñux in measured in watts per square meter on the right-hand axis. The open circles are E emission-linelF
l
,
galaxies, while the Ðlled circles are A absorption-line galaxies. C composite galaxies are designated by half-Ðlled circles and the three AGNs by Ðve-pointed
stars. The measured magnitudes of the 32 remaining members of the galaxy sample for which spectroscopic redshifts are not available are shown as
horizontal bars to the left of the diagrams. Note the Ðrm upper limit on K and the large variation in color apparent in the broader distributions in R and B.
Each of these diagrams includes theoretical evolutionary tracks computed from the simulations of Poggianti (1997) (see text) under the assumption that an Sc
Hubble type is equivalent to an emission-line galaxy (dot-dashed line), type Sa to a composite galaxy (dashed line), and type E to an absorption line galaxy
(solid line). The k-corrections alone produce tracks for 2L* indicated by the lighter lines, while the thick lines show the tracks for passive evolution.
while, in comparing galaxy luminosities with predictions
from models, the values themselves are at issue. Thus the
e†ects of any errors in the evolutionary corrections in
Figure 2 are more subtle than those seen in the Hubble
diagrams (Fig. 1).
Figure 2 shows overall agreement with the color predic-
tions from no-evolution models, as has been noted earlier
by, for example, Oke, Gunn, & Hoessel (1996) for galaxies
in clusters with z D 0.5 and in the Canada-France Redshift
Survey (Crampton et al. 1995), but still many minor con-
cerns persist. In each panel of this Ðgure, the range of
observed galaxy colors (equivalent to the range of galaxy
SEDs, i.e., the range of star formation histories considered
valid for galaxies) is somewhat larger at all redshifts than
the range predicted by the models.
The evolutionary corrections are better studied by a
more carefully deÐned sample of galaxies in clusters where
the galaxy spectral type can be more tightly constrained.
Such an approach has been taken by Pahre (1998).
2.3. T he V ery Red Objects
At high Galactic latitude, there is a population of faint,
very red objects. Among the main sample (Pahre et al. 1999)
are 19 objects with (R[K) º 5 magÈthree of which have
(R[K) [ 6 magÈproducing a surface density on the sky of
D1.3 arcmin~2. Ignoring Galactic stars, this class of objects
comprises 11% of the number count for K \ 20 mag.
The spectroscopically conÐrmed stars in the total sample
all have (R[K) \ 4.6 mag and are mostly M dwarfs. While
the reddest Galactic M dwarfs reach (V [K) B 6 mag
(Leggett 1992), such extremely red stars are not common in
magnitude-limited samples.
Lawrence et al. (1995), among others, advocate that a
signiÐcant fraction of galaxies at high redshift are dusty.
However we do not Ðnd signiÐcant UV extinction among
the emission-line galaxies in our sample (° 3.4), and hence
we do not believe that the many of the very red objects in
our sample without measured redshifts are heavily
reddened blue galaxies.
Four of the 19 very red objects have measured redshifts.
All four of them are classiÐed as A, with 0.78 \ z \ 1.23 ;
three of the four have z [ 1. Three of these very red objects
have good images in the HST /MDS database and they are
classiÐed as galaxies, not stars. Persson et al. (1993) have
speculated that such objects are passively evolved elliptical
galaxies with z [ 1, while Graham & Dey (1996) suggest
that they are reddened star-forming galaxies. The very red
objects for which we were successful in measuring redshifts
No. 1, 1999 REDSHIFT SURVEY 33
FIG.2a FIG.2b
FIG.2c
FIG. 2.ÈDereddened observed colors for the extragalactic objects are shown as a function of redshift. The symbols indicating the various galaxy spectral
types are the same as in Fig. 1. The lines denote the predictions from the Poggianti models for no evolution and for passive evolution for elliptical, Sa, and Sc
galaxies as in Fig. 1. The three panels show (a)(U[R), (b)(R[K), and (c)(U[K). Galaxies with upper limits for U are not plotted in (a) and (c).
support the hypothesis that such objects are high-redshift
elliptical galaxies, speciÐcally that they are galaxies with
that are not heavily reddened by dust.z Z 1
3. SPECTRAL ENERGY DISTRIBUTIONS
3.1. Computation of the Rest Frame Spectral
Energy Distributions
For each galaxy in our redshift sample, we can construct
a rest frame spectral energy distribution. We do this in a
slightly nonstandard way. For each galaxy, we compute the
luminosity per ln l, as a function of rest frequencyL 4 lL
l
,
using the six photometric magnitudes of Pahre et al. (1999).
The calibrations for absolute Ñux were adopted from Bessell
(1979) for U through R and were calculated from the
material in Pahre et al. (1999) for I and for A selection ofK
s
.
raw SEDs is displayed in Figure 3. In constructing raw
SEDs, we interpolate linearly between measurements. The
nominal error on each measured point is dominated by
systematic e†ects. We estimate this to be 0.2 in magnitude
or 0.08 in log L . For a variety of reasons discussed in Pahre
et al. (1999), many photometric measurements have much
larger errors, and these are shown as vertical bars. In addi-
tion, several photometric measurements are only 2 p upper
limits and these are also designated in Figure 3.
We next transform these raw SEDs into corrected SEDs
by interpolation and extrapolation through the inaccurate
measurements and upper limits and extension redward to
2.2 km using the observed I[K colors. (We believe this
procedure to be quite robust because the near-infrared
spectra of most galaxies are well represented as power laws.)
We also introduce two new spectral bands, P and Q, at rest
frequencies log l \ 15, 15.1 in the vacuum ultraviolet,
which are directly observed in most cases. (We extrapolate
to these frequencies in the lowest redshift galaxies.) This
operation leaves us with continuum spectra and a set of
eight rest spectral luminosities where a \ K, I, R, V , B,L
a
,
U, P, and Q. (The biggest concern about this procedure is
the large gap in log l between K and I. This can introduce a
systematic steepening of the derived infrared rest spectra for
high-redshift galaxies as the observed I band corresponds
to rest B where the rest SED can be intrinsically curved.
However, none of our results is strongly dependent upon
this extrapolation.)
34 COHEN ET AL.
FIG. 3.ÈRaw SEDs for a few selected galaxies denoted by their iden-
tifying D0K numbers. The abscissa is the rest frequency in hertz. The
ordinate is the spectral power in units of the Ðducial power (measuredL
B
*
in watts), where The errors in the measured magnitudeslog (L
B
*) \ 36.86.
are taken to be 0.2 except where indicated by vertical lines extending above
and below the measurement. Upper limits (2 p) are indicated by lines that
extend downward from the limit. In order to generate corrected SEDs,
low-accuracy measurements, such as the I-band measurement in D0K 149,
are replaced by interpolations. Upper limits, such as the B and U measure-
ments in D0K 116, are replaced by extrapolations through the Ðrst upper
limit adopting it as a measured value. Galaxy D0K 12 is an AGN and
appears to vary.
These SEDs are listed in Table 1 and are exhibited for a
selection of galaxies in the redshift sample in Figure 4.
There are some striking (though not unexpected) regu-
larities. The absorption-line galaxy SEDs exhibit quite red
infrared and ultraviolet spectra. For convenience we deÐne
two spectral indices in the rest frame. is the spectrala
IR
index, measured between the rest B- and[d log L
l
/d log l,
K-bands, and is the corresponding quantity betweena
UV
the Q- and B-bands. (Again our results are quite robust to
this choice, which maximizes the tightness of the corre-
lations that follow.) If we restrict our attention to A gal-
axies with high-quality redshifts as deÐned in Cohen et al.
1999a and with z \ 0.8, then nearly all the galaxies are
located in the region deÐned by anda
IR
[ 1, a
UV
[ 3,
almost all of E galaxies with high-quality redshifts lie
outside this region (Fig. 5). (The association of a hard-
ultraviolet continuum with emission lines is not, of course, a
surprise. However, this correlation does demonstrate that
internal reddening to be discussed in ° 3.4 is not a big factor
in these galaxies.) We can then use this correlation to deÐne
two SED classes for the whole redshift sample, including the
C spectral class, which covers the whole two-color plane,
and those of low quality (mostly A class). We call these
classes ““ old ÏÏ and ““ young, ÏÏ although the(a
IR
[ 1, a
UV
[ 3)
latter may also be rejuvenated, consisting of an older popu-
lation plus a recent starburst.
The correlation of with is shown in Figure 6 for thea
IR
L
K
same set of galaxies with high-quality redshifts that are dis-
played in Figure 5. A strong correlation between andL
K
a
IR
is apparent ; the most luminous galaxies have redder spec-
tral indices. There is also a clear separation with galaxy
spectral type. More luminous galaxies tend to be of A spec-
tral class, while the least luminous ones are those that show
strong signs of recent star formation (the E galaxies).
The tight correlation between galaxy SED shapes
(deÐned from 2400 to 2.2 km in the rest frame) and galaxyA
spectral classes assigned on the basis of the presence or
absence of key diagnostic features ([O II] j3727, H]K,[O
III] j5007, etc.) is one reason that photometric redshift tech-
niques such as that of Connolly et al. (1995) work reason-
ably well at least out to z D 1 for high-precision
photometric data sets.
3.2. L uminosity-V olume Diagram
In order to compare galaxies at di†erent redshifts (and
also with galaxies studied in other surveys), it is necessary to
deÐne a Ðducial luminosity (or absolute magnitude). We
again follow convention and label the galaxies by their rest
B luminosities. We do this in a slightly nonstandard way,
eschewing tabulated k-corrections and models of galaxy
evolution. Instead, for each galaxy, we compute the spectral
energy distribution (SED) of the luminosity per ln l, lL
l
,
as a function of rest frequency and extrapolate to the rest B
frequency (nominally from the red so as tolog l
B
\ 14.83)
deÐne a rest B luminosity We do this becauseL
B
4 lL
l
(l
B
).
of the presence of variable 4000 breaks just to the blue ofA
the B-band.
We normalize the galaxy luminosity to the local, Ðducial
luminosity W, or mag, orlog (L
B
*) \ 36.86 M
B
* \[20.8
(Bingelli, Sandage, & Tammannlog (L
B
/1 L
B_
) \ 10.52
1988) in our cosmography. [Note that log (L
K
* \ 36.91W 4
mag (Mobasher, Sharples, & Ellis 1993 ;M
K
* \[24.6
Cowie et al. 1996) and so a median L* galaxy should have
somewhat bluer than our median galaxy, just asa
IR
D 1,
might be expected from a K-selected sample.] We also esti-
mate the total luminosity in the wavelength interval 0.4
km \j\2 km, which is probably a fair measure of the
bolometric luminosity, by assuming a power-law Ðt to the
SED so that where T (0),T (1),T (2) \ 0.9,1.6,L
tot
\ T (a
IR
)L
B
,
3.6, respectively. For most of the galaxies in our sample,
0.5 \ T (a
IR
) \ 2.5.
Figure 7 shows the luminosity for each galaxy in our
redshift sample as a function of both comoving volume and
redshift separated by spectral class. It is immediately appar-
ent that the luminosity function does not evolve strongly
out to z D 0.8 and that there is a serious deÐcit of galaxies
with an issue to which we turn next.0.8 [ z [ 1.3,
3.3. Galaxies without Redshifts
The sample contains 32 objects believed to be galaxies for
which we cannot assign redshifts. However, we are fairly
conÐdent that these are not normal galaxies with z [ 0.8
because normal galaxies with strong emission lines in that
redshift range are detected with an identiÐable line or lines
all the way up to the survey limit. While faint absorption-
line galaxies in this regime are more problematic, the char-
acteristic shape of the 4000 break is still visible for suchA
galaxies with z \ 0.8, at least to R D 24 mag, corresponding
to (R[K) [ 4 mag at the survey limit. Although it is pos-
sible that we could be dealing with a new population of
galaxies that exhibit neither emission nor absorption lines,
similar to BL Lac objects, we regard this as quite unlikely
because this population would have to be spectrally heter-
ogeneous and to have, at a given redshift, maximum lumi-
nosity that evolved so as to track our survey limit.
Conversely, if these galaxies are at high redshift (z [ 2), they
would be anomalously luminous. Their median B lumi-
nosity at would be requiring a discontinuousz Z 1.3 Z3L
B
*,