Deep Infrared Studies of Massive High Redshift Galaxies Proefschrift

Deep Infrared Studies of Massive High Redshift Galaxies Proefschrift
Deep Infrared Studies of
Massive High Redshift
Galaxies
Proefschrift
ter verkrijging van
de graad van Doctor aan de Universiteit Leiden,
op gezag van de Rector Magnificus Dr. D.D. Breimer,
hoogleraar in de faculteit der Wiskunde en
Natuurwetenschappen en die der Geneeskunde,
volgens besluit van het College voor Promoties
te verdedigen op woensdag 13 oktober 2004
klokke 14.15 uur
door
Ivo Ferdinand Louis Labbé
geboren te Alphen a/d Rijn
in 1972
Promotiecommissie
Promotor:
Prof. dr. M. Franx
Referent:
Prof. dr. P.D. Barthel (Universiteit Groningen)
Overige leden:
Prof.
Prof.
Prof.
Prof.
dr.
dr.
dr.
dr.
P.G. van Dokkum (Yale, New Haven, USA)
K.H. Kuijken
H.-W. Rix (MPIA, Heidelberg, Germany)
P.T. de Zeeuw
iv
Table of contents
Table of contents
Chapter 1. Introduction and Summary
1 Introduction . . . . . . . . . . . . . . .
1.1
Observational Cosmology . . . .
1.2
Galaxy Formation . . . . . . . .
1.3
Massive High Redshift Galaxies .
1.4
The Faint InfraRed Extragalactic
2 Outline and Summary . . . . . . . . . .
3 Conclusions and Outlook . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
Survey
. . . . .
. . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
1
.
1
.
1
.
2
.
4
.
6
.
8
. 11
Chapter 2. Ultradeep NIR ISAAC observations of the HDF-South:
observations, reduction, multicolor catalog, and photometric
redshifts
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1
Field Selection and Observing Strategy . . . . . . . . . . . .
2.2
Observations . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 Data Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1
Flatfields and Photometric Calibration . . . . . . . . . . . . .
3.2
Sky Subtraction and Cosmic Ray Removal . . . . . . . . . .
3.3
First Version and Quality Verification . . . . . . . . . . . . .
3.4
Additional Processing and Improvements . . . . . . . . . . .
3.5
Final Version and Post Processing . . . . . . . . . . . . . . .
4 Final Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1
Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2
Image Quality . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3
Astrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4
Backgrounds and Limiting Depths . . . . . . . . . . . . . . .
5 Source Detection and Photometry . . . . . . . . . . . . . . . . . . .
5.1
Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2
Optical and NIR Photometry . . . . . . . . . . . . . . . . . .
6 Photometric Redshifts . . . . . . . . . . . . . . . . . . . . . . . . . .
6.1
Photometric Templates . . . . . . . . . . . . . . . . . . . . .
6.2
Zphot Uncertainties . . . . . . . . . . . . . . . . . . . . . . . .
6.3
Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7 Catalog Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . .
8 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.1
Completeness and Number Counts . . . . . . . . . . . . . . .
8.2
Color-Magnitude Distributions . . . . . . . . . . . . . . . . .
15
16
17
17
19
20
21
22
23
23
24
25
25
27
28
28
30
31
32
35
36
36
37
38
39
39
44
v
Table of contents
9
Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . .
Chapter 3. Large disk-like galaxies at high redshift
1 Introduction . . . . . . . . . . . . . . . . . . . . . .
2 Observations . . . . . . . . . . . . . . . . . . . . . .
3 Rest-frame Optical versus UV Morphology . . . . .
4 Profile fits and sizes . . . . . . . . . . . . . . . . . .
5 Spectral Energy Distribution . . . . . . . . . . . . .
6 Discussion . . . . . . . . . . . . . . . . . . . . . . .
48
.
.
.
.
.
.
53
54
55
55
57
58
59
Chapter 4. The rest-frame optical luminosity density, color, and
stellar mass density of the universe from z=0 to z=3
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 Measuring Photometric Redshifts and Rest-Frame Luminosities . .
3.1
Photometric Redshift Technique . . . . . . . . . . . . . . . .
3.1.1
Star Identification . . . . . . . . . . . . . . . . . . .
3.2
Rest-Frame Luminosities . . . . . . . . . . . . . . . . . . . .
3.2.1
Emission Lines . . . . . . . . . . . . . . . . . . . . .
4 The Properties of the Massive Galaxy Population . . . . . . . . . .
4.1
The Luminosity Density . . . . . . . . . . . . . . . . . . . . .
4.1.1
The Evolution of jλrest . . . . . . . . . . . . . . . . .
4.2
The Cosmic Color . . . . . . . . . . . . . . . . . . . . . . . .
4.3
Estimating M/L∗V and The Stellar Mass Density . . . . . . .
5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1
Comparison with other Work . . . . . . . . . . . . . . . . . .
5.2
Comparison with SFR(z) . . . . . . . . . . . . . . . . . . . .
5.3
The Build-up of the Stellar Mass . . . . . . . . . . . . . . . .
6 Summary & Conclusions . . . . . . . . . . . . . . . . . . . . . . . .
A Derivation of zphot Uncertainty . . . . . . . . . . . . . . . . . . . . .
B Rest-Frame Photometric System . . . . . . . . . . . . . . . . . . . .
C Estimating Rest-Frame Luminosities . . . . . . . . . . . . . . . . . .
63
64
66
67
67
68
68
70
73
73
76
78
81
88
88
90
91
92
97
99
99
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Chapter 5. The color magnitude distribution of field galaxies at
1 < z < 3: the evolution and modeling of the blue sequence
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1
The Observations and Sample Selection . . . . . . . . . . . .
2.2
Photometric Redshifts and Rest-Frame Colors . . . . . . . .
3 The rest-frame Color-Magnitude Distribution of Galaxies from z ∼ 1
to z ∼ 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 The Color-Magnitude Relation of Blue Field Galaxies . . . . . . . .
4.1
The Slope and its Evolution . . . . . . . . . . . . . . . . . . .
4.2
The Zeropoint and its Evolution . . . . . . . . . . . . . . . .
4.3
The Origin of the Blue Sequence in the Local Universe . . . .
103
104
106
106
107
108
109
109
112
114
vi
Table of contents
4.4
5
6
7
Comparison to z ∼ 3 Galaxies . . . . . . . . . . . . . .
4.4.1
The Models . . . . . . . . . . . . . . . . . . .
4.4.2
Results . . . . . . . . . . . . . . . . . . . . . .
4.4.3
The Blue CMR in Various Rest-Frame Colors
Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3 . . . . .
5.1
A library of Star Formation Histories . . . . . . . . .
5.2
Fitting Method . . . . . . . . . . . . . . . . . . . . . .
5.2.1
Creating Mock Observations . . . . . . . . . .
5.2.2
The Fitting . . . . . . . . . . . . . . . . . . .
5.3
Results . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1
Constant Star Formation . . . . . . . . . . . .
5.3.2
Exponentially Declining Star Formation . . .
5.3.3
Repeated Bursts . . . . . . . . . . . . . . . .
5.3.4
Episodic Star Formation: The Duty Cycle . .
5.4
Discussion . . . . . . . . . . . . . . . . . . . . . . . . .
The Onset of the Red galaxies . . . . . . . . . . . . . . . . .
Summary and Conclusions . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
117
117
118
119
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
120
120
121
121
122
123
123
125
126
128
130
132
134
Chapter 6. IRAC Mid-infrared imaging of red galaxies at z >
new constraints on age, dust, and mass
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 The Observations, Photometry, and Sample Selection . . . . . .
3 Mid-Infrared Properties of Red Galaxies at z > 2 . . . . . . . .
4 Comparison to Lyman Break Galaxies . . . . . . . . . . . . . . .
5 The rest-frame K-band mass-to-light ratio . . . . . . . . . . . .
6 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . .
2:
.
.
.
.
.
.
.
.
.
.
.
.
139
140
141
143
147
149
150
Nederlandse samenvatting (Dutch summary)
155
Curriculum Vitae
161
Nawoord / Acknowledgments
163
CHAPTER
ONE
Introduction and Summary
1
1.1
Introduction
Observational Cosmology
C
osmological studies have resulted in a “standard” model: a flat, homogeneous, isotropic universe on large scales, composed of ordinary matter, nonbaryonic cold dark matter, and dark energy (Spergel et al. 2003). In this model,
the dozen or so of parameters that characterize the universe, most importantly
the density of baryons, dark matter, and the expansion of the universe, successfully describe astronomical observations on scales from a few Mpc to several 1000
Mpc (Freedman et al. 2001; Efstathiou et al. 2002; Spergel et al. 2003). Large
scale structure and galaxies grew gravitationally, from tiny, nearly scale-invariant
adiabatic Gaussian fluctuations.
I remember vividly the first cosmology conference I attended at the start of
this thesis work. I confess the discussions about cosmic microwave background
fluctuations seemed endless, as were the discussions on the large scale structure
seen in galaxy surveys. I realize now that I did not fully appreciate how much
easier my life was going to be knowing that the universe was spatially flat and
13.7 Gyr old. I could focus for many years on questions that would have been
much harder to address until these major issues were solved. Given the enormity
of the subject in both distance and time, what cosmologists have learned in the
last century, leading up to the arrival of this “standard model”, is nothing short
of an extraordinary success.
Leaving the mother of all questions “What is the the origin of the universe?”
to the realm of literature and natural philosophy, there exist two major, widely
recognized, unresolved issues in contemporary cosmology.
The first one relates to the seminal revelation by Fritz Zwicky who noted the
existence of dark matter in the Coma cluster (Zwicky 1937), or the problem of
missing mass, as it was then known. Zwicky measured the velocity dispersion of
2
1 Introduction and Summary
galaxies in the Coma cluster and used the virial theorem (Eddington 1916), which
relates the total internal kinetic energy of the cluster to its gravitational potential,
to argue that there was much more mass in the cluster than could be attributed
to the stars in galaxies. In solar units, the ratio of mass to optical luminosity of
a galaxy such as the Milky Way is ∼ 3M¯ /L¯ , whereas for the Coma cluster the
ratio was about 400. This implies that there must be about 100 times more hidden
or “dark” matter as compared with matter in stars. The realization grew that all
visible matter is only a minor constituent of the universe.
Dark matter has proved remarkably elusive and despite more than 70 years
of observational astronomy and experimental physics – and a host of respectable
and less-respectable candidates – there has been no confirmation of its nature,
except that it is gravitating, does not emit or absorb light, is non-baryonic, and
non-leptonic. Worse, since the rise of the standard cosmological model it has been
joined by its even more elusive cousin, Vacuum or Dark Energy: the mysterious
substance that apparently dominates the energy budget of the Universe, and is
believed to fuel its accelerated expansion. Yet in contrast to the fundamentally
impenetrable mist of the Big Bang, Dark Matter and Dark Energy are not so
far removed that their effects cannot be measured with the present-day methods
and technology. If anything, its presence challenges us towards new physics and
promises to keep cosmology vibrant for quite some time.
The second great question in cosmology is the main inspiration of this thesis.
How did galaxies, harboring most of the visible matter in the universe, form out
of an almost perfectly smooth distribution of matter, 379.000 years after the Big
Bang? In the standard model, the growth of structure with time on a range of
scales, from the largest scale structure, to galaxies, and to perhaps even the first
stars, is driven by the gradual hierarchical merging of Cold Dark Matter (CDM)
Halos (White & Rees 1978). Cold intergalactic gas cooled onto these halos and
was subsequently converted into the stars which shine as galaxies. Observation of
galaxies can thus be used to trace the evolution of both normal and dark matter.
Given the theoretical foundation of structure formation and the general principles thought to govern the formation of galaxies, the formidable task is to understand the physics of star formation on galactic scales, the resultant feedback of
energy and material into the interstellar and intergalactic medium, and the role of
feedback from supermassive black holes that lurk in the center of galaxies. However formidable, the problem is likely one of complexity, but not in any sense a
fundamental problem requiring new physics to be discovered.
1.2
Galaxy Formation
One of the most prominent and important clues to understanding galaxy evolution
is the range and distribution of galaxy morphologies present in the local universe.
These morphologies were first classified by Hubble (1930) and consist of four major
classes: (i) elliptical galaxies (E), (ii) lenticular galaxies (S0), (iii) ’spiral’ galaxies,
1.1 Introduction
3
(iv) all remaining galaxies, considered ‘irregular’. Elucidating the origin of the
“Hubble Sequence” is a crucial and necessary constraint on all models of galaxy
formation.”
One of the strongest constraints comes from E+S0s galaxies, collectively called
“early-type” galaxies, which are the most massive galaxies in the local universe.
The standard CDM scenario prescribes that early-type galaxies formed by mergers
of spiral galaxies at relatively recent times (Toomre & Toomre 1972; Toomre 1977;
White & Rees 1978). In this scenario, the appearance of galaxies should reflect
the growing and merging pattern of their dark matter halos, and as such galaxies
should be growing through merging and/or star formation to the present day (e.g.,
White & Frenk 1991; Kauffmann & White 1993).
Strangely enough, extensive study of nearby early-type galaxies in clusters
contradicts this picture. The massive early-type galaxies form an extremely homogeneous class, obeying tight relations in their properties, such as colors and
luminosities (Sandage & Visvanathan 1978). Since the colors of stellar populations change with age, the small scatter in the color-magnitude relation constrains
the scatter in age at a given magnitude. For ellipticals in the Coma Cluster, the
estimated intrinsic scatter of 0.04 in U − V colors implied a spread in age of less
than 15% (Bower, Lucey, & Ellis 1992). The usual interpretation is that the stars
in these galaxies formed at high redshift z > 2. Detailed modelling of the spectra
showed a similar picture (Trager et al. 2000).
The formation scenario of nearby early-type galaxies in clusters is closer to
the classical picture of the monolithic collapse, the antipole of the hierarchichal
galaxy formation. In the monolithic scenario, E+S0s assembled their mass and
formed their stars in a rapid event, of much shorter duration than their average
age (Eggen, Lynden-Bell, & Sandage 1962; Larson 1975; van Albada 1982). The
formation process happened at high redshifts predicting that and the progenitors
of todays massive early-type galaxies would be vigorously star-forming with star
formation rates of much more than 100M¯ yr−1
For decades, the chief way to make the distinction between these opposite formation scenarios was through minute observations of the colors and spectras of
galaxies in the local universe. However, direct inference from the fossil evidence
to explain its origin is impossible given the complexity of galaxy formation. Even
with powerful computational solutions progressing at high pace, neither hydrodynamical simulations (e.g., Katz & Gunn 1991; Springel et.al. 2001; Steinmetz
& Navarro 2002) nor state-of-the-art “semi-analytic” models (e.g., Kauffmann,
White, & Guiderdoni 1993; Somerville & Primack 1999; Cole, Lacey, Baugh, &
Frenk 2000) provide accurate predictions for the formation history of galaxies. The
major problem still to hurdle is the correct description of galactic scale baryonic
processes involved in stellar birth, evolution, and death. In contrast to the cosmic
microwave background, where we were confident of the physics, were able to constructed elaborate deductive models, and test them observationally, the physics
of galaxy formation is perhaps too complicated to consider from first principles.
4
1 Introduction and Summary
Here, it is more robust to restrict our science to simple inferences from direct
observations.
Clearly, observing galaxies while they are forming provides a powerful and
complementary approach. Nature’s gift to cosmologists, the finite speed of light,
allows us to look back in time as we observe galaxies at increasing distance, as
measured by their redshifts. This has been the corner-stone of the “look-back”
approach to studying galaxy evolution. Observations of extremely distant and
extremely faint galaxies at z = 2 and higher allow us a direct look into the process
of their formation.
1.3
Massive High Redshift Galaxies
Answers to many of the questions are now becoming available from large statistical
surveys of galaxies, using new instruments on the largest ground-based and spacebased telescopes. Multicolor imaging is our most powerful tool for understanding
galaxy evolution, as broadband filter measurements can go very deep and cover
large areas. However, even the most ambitious apparent magnitude-selected surveys showed that simply observing fainter galaxies is a relatively inefficient means
of assembling significant samples of galaxies at high redshift (e.g., Lilly et al. 1996;
Ellis et al. 1996 Songaila et al. 1994; Cohen et al. 1996; Cowie et al. 1996). Most
faint galaxies on the sky are in fact nearby.
The study of distant galaxies was revolutionized in the past decade, when the
increased surveying power was coupled with advances in photometric pre-selection
techniques. Of particularly importance was the identification of a large sample of
high redshift galaxies from a specific color signature – red in U − B colors, blue
in B − V colors. This color signature is produced when neutral hydrogen, from
the intergalactic medium and within the galaxies, absorbs the flux of young hot
stars (t . 107 yr) shortward of Lyman 1216 Å and the 912 Å Lyman limit, which
corresponds to the U -band at z ∼ 3 (e.g., Guhathakurta, Tyson, & Majewski
1990; Songalia, Cowie, & Lilly 1990, Steidel & Hamilton (1992, Steidel, Pettini, &
Hamilton 1995).
At the time this thesis work commenced, Lyman Break Galaxies (LBGs), as
they are commonly called, were among the best studied classes of distant galaxies.
Close to a 1000 had been spectroscopically confirmed (Steidel et al. 1996a,b, 2003)
at z & 2, and they were found to be a major constituent of the early universe, with
space densities similar to local luminous galaxies. Could LBGs be the forerunners
of todays massive ellipticals?
The strong clustering of LBGs indicated that they were luminous tracers of
the underlying massive dark matter halos (Adelberger et al. 1998; Giavalisco et al.
1998; Giavalisco & Dickinson 2001) suggesting that they evolve into massive galaxies. Initially it was thought that they were very massive (Steidel et al. 1996b),
but direct internal kinematic measurements revealed velocity dispersions of 60100 kms−1 (Pettini et al. 2001; Shapley et al. 2001), which combined with their
1.1 Introduction
5
compact sizes of a few kiloparsecs (Giavalisco, Steidel, & Macchetto 1996) suggested virial masses of only 1010 M¯ , less than 1/10th of todays most massive
galaxies. Their spectra were consistent with on-going star forming activity, resembling that of local star bursting galaxies, with only moderate extinction by
dust (Adelberger & Steidel 2000). The estimated star formation rates were only
10 − 15 M¯ yr−1 , much lower than expected for the progenitors of local earlytypes in the monolithic scenario. Studies of the stellar composition, using optical
and near-infrared broadband photometry, reinforced this picture of relatively lowmass, moderately star forming objects (Sawicki & Yee 1998; Shapley et al. 2001;
Papovich, Dickinson, & Ferguson 2001). These properties are expected in the
standard hierarchical scenario of galaxy formation, where LBGs are the low-mass
building blocks that merge to become present-day massive galaxies in groups and
clusters (e.g., Mo & Fukugita 1996; Baugh et al. 1998).
It was quickly realized that LBGs were selected in a very peculiar way. Galaxies
that were dust-reddened by more than E(B−V ) & 0.4 were not selected(Adelberger
& Steidel 2000). Could there still be a population of vigorously star forming
galaxies that escaped detection because it was enshrouded in dust? We know that
complementary samples of heavily obscured high-redshift galaxies are found in
sub-mm, radio (Smail et al. 2000), and X-ray surveys (Cowie et al. 2001; Barger
et al. 2001). Their number densities are much lower than LBGs, although their
contribution to the total star formation rate can be significant (cf., Adelberger &
Steidel 2000; Barger, Cowie, & Sanders 1999).
Even more important, have we overlooked a major population at z > 2 of
evolved galaxies that resemble the present-day elliptical and spiral galaxies? None
of the techniques described above would select those normal galaxies, whose light
is dominated by evolved stars. Specifically, even the massive Milky Way would
have never been selected with the Lyman Break technique if placed at z ∼ 3, as it
does not have the required high far-UV surface brightness.
It is much easier to detect evolved galaxies at z > 2 in the near-infrared (NIR),
where one can access their rest-frame optical light, and where evolved stars emit
the bulk of their light. The rest-frame optical light is also much less sensitive
to the effects of dust obscuration and on-going star formation than the UV, and
expected to be a better (yet imperfect) tracer of stellar mass.
Researchers had already begun using NIR data to look for evolved galaxies at
z > 1. The techique relied on photometric pre-selection on the stellar Balmer and
4000 Å break. The Balmer discontinuity at 3650 Å is generally strong in stellar
populations of age 108 to 109 yr, and is most pronounced in the photosphere of
A-stars, while even older stellar populations show a charactaristic 4000 Å break
due to the sudden onset below 4000 Å of metallic and molecular absorption in cool
stars.
Evolved galaxies were targeted by selecting on red optical-NIR colors R−K > 5
or I − K > 4, which can be produced as the Balmer/4000 Å break redshifts
6
1 Introduction and Summary
out of the optical filters. Recent surveys yielded large samples of these so-called
Extremely Red Objects at redshifts 0.8 < z < 1.8 (EROs; e.g., Elston, Rieke,
& Rieke 1988; Hu & Ridgway 1994; Thompson et al. 1999; Yan et al. 2000;
Scodeggio & Silva 2000; Daddi et al. 2000; McCarthy et al. 2001; Smith et
al. 2002, Moustakas et al. 2004), but their relation to the well-formed massive
early-type galaxies at z < 1 has not been firmly established. Apparently, a minor
fraction of 30% are strongly clustered and evolved galaxies (e.g., Cimatti et al 2002,
Daddi et al. 2002), while the rest are believed to be dust reddened star-forming
objects (e.g., Yan & Thompson 2003; Moustakas et al. 2004).
To observe evolved galaxies at even higher redshifts z > 2 one has to develop
criteria to select the galaxies and at the same time go deep enough to overcome
the cosmological effect of the (1 + z)4 surface brightness dimming. Normal evolved
galaxies at z ∼ 3 would be incredibly faint. In addition as the Balmer/4000 Å
break is less strong than the Lyman break, it requires a combination of extremely
deep optical and NIR imaging to select them.
There were very few datasets that reached the required depths, most of them
taken with the WFPC2 and NICMOS instruments on the Hubble Space Telescope (Thompson et al 2000; Dickinson et al 2000; Williams et al 2000). The
largest survey to date is that of Dickinson et al. (2000), who imaged the Hubble
Deep Field North (HDFN) WFPC2 field in J110 and H160 with NICMOS, finding that most high redshift galaxies would have been picked up by the Ly-break
technique. Nevertheless, the total area studied is still very small, and the depth of
the K−band data, which is important in constraining the constribution of evolved
stars at z ∼ 3, is not well matched to that of the rest of the NICMOS or WFPC2
imaging data.
To remedy this situation, we started the Faint InfraRed Extragalactic Survey:
an ultradeep optical-to-infrared multicolor survey of high redshift galaxies.
1.4
The Faint InfraRed Extragalactic Survey
This dissertation is based on the Faint InfraRed Extragalactic Survey (FIRES;
Franx et al. 2000) and its aims are intimately connected with the goals of this
survey. FIRES is a large public program at the Very Large Telescope (VLT)
consisting of very deep NIR imaging of two fields. The fields are the WPFC2-field
of Hubble Deep Field South (HDFS), and the field around the z = 0.83 cluster
MS1054-03, both selected for their exquisite, deep optical WFPC2 imaging with
the HST.
The central question in this thesis is:
How did massive galaxies assemble over time?
This general question is addressed in the following chapters by analyzing the following specific issues:
1.1 Introduction
7
1. Have we overlooked a major population at high redshift perhaps
resembling the present-day normal elliptical and spiral galaxies?
2. When did the Hubble Sequence of galaxies start manifesting?
3. How and when did galaxies assemble the bulk of their stellar mass?
4. Can we use the distribution of galaxy properties to constrain formation scenarios?
5. What is the detailed stellar composition of z > 2 galaxies: are there
passively evolving galaxies at z ∼ 3, and what is the role of dust?
The body of this work deals with the properties of faint distant galaxies as
observed in the HDFS, a small, otherwise undistinguished high-galactic latitude
patch of sky. The HDFS and its counterpart in the north, the HDFN, constitute
milestones in optical imaging, as the WFPC2 camera on the HST was pushed to
its limits. The ultradeep imaging in four optical bands (U300 , B450 , V606 , I814 ) of
these “empty” fields, named after the absence of any large foreground galaxies,
allowed an unprecedented deep look of the distant universe, and opened the door
to the study of normal galaxies at z > 2.
It is often wondered whether such a small field presents us a fair view of the
universe. Luckily, the universe seems not to be fractal or hierarchically structured
beyond a few hundred Mpc, and voids and superclusters like the ones near us
simply repeat. Therefore at large distances, even a pinhole survey such as the
HDFS may sample enough volume as to obtain a representative picture. We
should always remember, however, that the field size and volume of the HDFS is
small.
Capitalizing on the advances in NIR detector capabilities, we took to the Infrared Spectrometer and Array Camera (ISAAC, Moorwood 1997) on the VLT,
and observed this tiny field in the NIR Js , H, and Ks filters for more than 100
hours total, and only under the best seeing conditions. The second field, centered
on the z = 0.83 cluster MS1054-03, was observed for 80 hours (Förster Schreiber
et al. 2004a). While not as spectacularly deep, the area surveyed in MS1054-03 is
nearly five times larger, and turned out to be a crucial element for its ability to
reinforce our findings in the HDFS.
A special asset in the deep imaging set was the Ks −band, the reddest band
from the ground where achievable sensitivity and resolution were still somewhat
comparable to the space-based optical data. At z ∼ 3 it probes rest-frame wavelengths λ ∼ 5400Å, comfortably redward of the Balmer/4000 Å break and therefore
crucial to assess the build-up of evolved stars.
Aiming for evolved galaxies at z > 2, we experimented with simple color-
8
1 Introduction and Summary
selection criteria, analogous to those applied to EROs at lower redshift 1 < z < 2.
We selected high-redshift galaxies with the simple color criterion Js − Ks > 2.3,
specifically designed to target the Balmer/4000 Å breaks at redshifts between
2 < z < 4. While candidate high-redshift galaxies with even redder J − K colours
have been reported by other authors as well (e.g. Scodeggio & Silva 2000; Hall et
al. 2001; Totani et al. 2001; Saracco et al. 2001, 2003), the focus was usually on
the objects with the most extreme colors. The Js − Ks > 2.3 criterion, in fact, is
not extreme at all as it corresponds to a U − V color of 0.1 at z = 2.7. Such a
selection would include all but the bluest present-day Hubble Sequence galaxies.
2
Outline and Summary
We present in Chapter 2 the FIRES observations of the HDFS, the data reduction, the assembly of the photometric source catalogs, and the photometric
redshifts. These data constitute the deepest groundbased NIR images to date,
and the deepest Ks −band in any field, even from space (Labbé et al. 2003). We
released the reduced data, catalog of sources, and photometric redshifts to the
community, and they are now in use by many researchers worldwide.
One immediate scientific breakthrough was the identification of a significant
population of galaxies with red Js −Ks > 2.3 colors at z > 2. We find these Distant
Red Galaxies, or DRGs as we shall now call them, in substantial numbers, with
space densities about half of that of LBGs selected from ground-based imaging
(Franx et al. 2003). Our follow-up studies suggested that DRGs, at a given
rest-frame optical luminosity, have higher ages, contain more dust, and are more
massive than LBGs (Franx et al. 2003; van Dokkum et al. 2004; Förster Schreiber
et al. 2004),
These galaxies had been previously missed because they are extremely faint
in the optical (rest-frame UV) and emit most of their light in the NIR (restframe optical). Surprisingly, galaxies with comparable colors at 2 < z < 3.5 were
almost absent in HDFN (cf., Labbé et al. 2003, Papovich, Dickinson, & Ferguson
2001) even though it was surveyed over a similar area and depth (Dickinson 2000).
Clearly, cosmic variance due to large scale structure in the universe plays a role
here, and the possibility existed that neither of the two fields was representative.
Later, our findings in the HDFS were confirmed by the discovery of DRGs at
similar densities in the much larger MS1054-03 field (see, e.g., van Dokkum et al.
2003; Förster Schreiber et al. 2004).
Another galaxy population that believed to be absent at high redshift, were
large disk galaxies. Disk galaxies are thought to undergo a relatively simple formation process in which gas cools and contracts in dark matter halos to form rotationally supported disks with exponential light profiles (Fall & Efstathiou 1980;
Mo, Mao, & White 1998). In the standard hierarchical model, large disks form
relatively late z < 1, and very few of them are expected as early as z ∼ 2.
1.2 Outline and Summary
9
In Chapter 3 we report the discovery of 6 galaxies in the HDF-S at 1.5 . z . 3
with colors, morphologies and sizes comparable to local spiral galaxies (Labbé et al.
2003b). The irregular optical WFPC2 (rest-frame far-UV) morphologies galaxies
had previously been misinterpreted, because they traced the sites of unobscured
star-formation rather than the underlying evolved population. Here, the combination of bandpass shifting and surface brightness dimming had given an exaggerated
impression of evolution towards high redshift. In the NIR, however (rest-frame optical), the morphologies were much more regular than in the rest-frame far-UV,
with well resolved exponential profiles as expected for rotating stellar disks. Models of disk formation in the standard CDM scenario (Mo, Mao, & White 1999)
currently underpredict the number of large disks at high redshift by a factor of
two. Only larger samples and kinematical studies, to establish the presence of
rotating disks, can tell how serious this discrepancy is, and when classical Hubble
sequence spiral galaxies came into existence.
In Chapter 4 we analyze the cosmological growth of the stellar mass density
from redshift z ∼ 3 to z = 0, as traced by optically luminous galaxies in the HDFS.
Measuring accurate stellar masses from broadband photometry is quite hard. We
resorted to interpreting the mean cosmic colors (U − B)rest and (B − V )rest using
stellar population synthesis models. We assumed an appropriate star formation
history for the universe as a whole, and used the models to derive the global
mass-to-light ratio M/LV in the V −band.
We found that the universe at z ∼ 3 had a ∼ 10 times lower stellar mass
density than it does today, and half of the stellar mass of the universe was formed
by z = 1 − 1.5, broadly consistent with independent results obtained in the HDFN
(Dickinson et al. 2003). Interestingly, the distant red galaxies discovered earlier
in the survey may have contributed as much as ∼ 50% to the cosmic stellar mass
density at z ∼ 3.
In Chapter 5 we studied the rest-frame optical color-magnitude distribution
of galaxies in the FIRES fields. We focused in particular on the redshift range
z ∼ 3, where observations in the HDFN showed that blue star-forming galaxies
followed a clear trend, such that galaxies more luminous in the rest-frame V −band
had slighly redder U − V colors (Papovich, Dickinson, & Ferguson 2001; Papovich
et al. 2004). The origin of this color-magnitude relation (CMR) for blue galaxies,
or blue sequence, was not clearly understood, even at low redshift (cf., Peletier &
de Grijs 1998; Tully et al. 1998; Zaritsky, Kennicutt, & Huchra 1994; Bell & De
Jong 2001).
We analyzed spectra of nearby galaxies in the Nearby Field Galaxy Survey
(Jansen, Fabricant, Franx, & Caldwell 2000; Jansen, Franx, Fabricant, & Caldwell
2000; Jansen, Franx, & Fabricant 2001), and found that the relation is mainly one
of increasing dust-opacity with increasing luminosity. We also showed that the
slope of this relation does not change with redshift, and may have a similar origin
already at z ∼ 3.
10
1 Introduction and Summary
Similarly to studies of the red sequence cluster ellipticals (Bower, Lucey, & Ellis
1992; van Dokkum et al. 1998), we interpreted the scatter around the relation as
the result of a spread in ages of the stellar populations. The blue-sequence scatter
is fairly narrow, has a conspicuous blue envelope, and skew to red colors. After
exploring a range of formation models for the galaxies, we concluded that models
with episodic star formation explain most aspects of the z = 3 color-magnitude
distribution. The episodic models cycle through periods of star formation and
quiescence, rejuvenating the stellar population during each active episode. The
result is that the luminosity weighted ages of the stars are smaller than the age
of the galaxy, i.e., the time since the galaxy first started forming stars. They may
provide a solution of the enigmatic observation that z = 3 galaxies are much bluer
than expected if they were as old as the universe (e.g., Papovich, Dickinson, &
Ferguson 2001).
Chapter 6 presents a study of the stellar composition of distant red galaxies
and Lyman Break galaxies, using mid-IR imaging from IRAC on the Spitzer Space
Telescope. Our previous studies indicated that DRGs have higher ages, contain
more dust, and are more massive than LBGs at a given rest-frame optical luminosity (Franx et al. 2003; van Dokkum et al. 2004; Förster Schreiber et al. 2004), and
may contribute comparably to the cosmic stellar mass density (Franx et al. 2003;
Rudnick et al. 2003). Nevertheless, the nature of their red colors is still poorly
understood, and the masses are somewhat uncertain. Are they all truly old, or are
some also very young and very dusty? What is the fraction of passively, evolving
“dead” systems? How much do the DRGs contribute to the stellar mass density in
a mass-selected sample? And how do they relate to the blue Lyman break galaxies.
Finally, what is their role in the formation and evolution of massive galaxies?
In this chapter, we present deep IRAC 3.6 − 8 micron imaging of the HDFS
field. The new IRAC data reached rest-frame NIR wavelengths, which were crucial
in determining the nature of DRGs in comparison to LBGs. We uniquely identified
3 out of 11 DRGs as old passively evolving systems at z ∼ 2.5. Others were heavily
reddened star-forming galaxies, for which we are now better able to distinguish
between the effects of age and dust. Furthermore, the rest-frame NIR data allowed
more robust estimates of the stellar mass and stellar mass-to-light ratios (M/L K ).
We found that in a mass-selected sample DRGs contribute 1.5 − 2 times as much
as the LBGs to the cosmic stellar mass density at 2 < z < 3.5. Also, at a given
rest-frame K luminosity the red galaxies are twice as massive with average stellar
masses ∼ 1011 M¯ , and their M/LK mass-to-light ratios exhibit only a third of
the scatter compared to the LBGs. This is consistent with a picture where DRGs
are more massive, more evolved, and have started forming at higher redshift than
most LBGs.
1.3 Conclusions and Outlook
3
11
Conclusions and Outlook
We have presented evidence in this thesis that previous imaging surveys gave a
biased view of the early universe. The immediate conclusions of the Faint InfraRed
Extragalactic Survey are that large numbers of evolved and dust-obscured galaxies
at z = 2 − 3 have been overlooked, that up to half of the stellar mass in the
universe at z = 2 − 3 was unaccounted for, and that the morphologies of galaxies
were misinterpreted.
With the newest optical-to-MIR multiwavelength surveys, we are for the first
time obtaining a better census of the massive galaxies in the early universe. We are
one step closer to tracing the assembly of massive galaxies directly, and it seems
warranted now to interpolate between the properties of galaxies populations at
different epochs. Ultimately, that approach will help us to understand how galaxies
evolved from the cradle to the present-day.
Even so, for a full comparison with local samples, our data sets are still much
too small. If we would dissect the galaxy population by redshift, luminosity, color,
morphology, or environment, to analyze galaxy formation in all its complexity, we
would be left with few galaxies in every subsample. A straightforward extension
of the current work is thus to obtain much larger samples.
Apart from the obvious enlargement of the samples, it is now crucial to followup the current samples with high-resolution NIR imaging, and optical and NIR
spectroscopy. High resolution NIR imaging allows to unambiguously determine the
rest-frame optical morphologies of galaxies to high redshift. Unfortunately, given
the limited availability of the Hubble Space Telecope, and the slow survey speed of
the NICMOS camera in particular, we must await the arrival of next-generations of
space telescopes (e.g., JWST or JDEM) or the maturing of ground-based solutions,
such as active optics with laser guide stars. A dearth of high-resolution imaging
of distant galaxies will continue to exist for some time to come.
On the other hand, new NIR spectroscopic instruments promise spectacular
advances. In the coming years a number of multi-object NIR spectrographs will
arrive at 8 − 10 meter class telescopes. With these instruments we can obtain deep
spectra for hundreds of distant galaxies at the same time. Spectroscopic redshifts
constrain the space densities of the sources and allow to get accurate rest-frame
colors. Kinematic measurements of nebular lines provide direct estimates of the
dynamical masses involved, and reveal any ordered rotation. Finally, their emission
line strengths provide independent constraints on star formation rates (SFRs) and
dust extinction, helping to break the degeneracies inherent to modeling of the
broadband SEDs (e.g., Papovich et al. 2001, Shapley et al 2001). Every topic
discussed in this thesis, from the study of high-redshift disk galaxies, via the
stellar populations of DRGs, to the origin of the high-redshift color-magnitude
relation (and its scatter), benefits directly from deep spectra.
12
1 Introduction and Summary
References
Adelberger, K. L., Steidel, C. C., Giavalisco, M., Dickinson, M., Pettini, M., & Kellogg,
M. 1998, ApJ, 505, 18
Baugh,C. M., Cole, S., Frenk, C. S., & Lacey, C. G. 1998, ApJ, 498, 504
Bower, R. G., Lucey, J. R., & Ellis, R. S. 1992, MNRAS, 254, 601
Casertano, S. et al., AJ, 120, pp. 2747–2824, 2000
Cimatti, A., et al. 2002, A&A, 391, L1
Cohen, J. G., Cowie,L. L., Hogg, D. W., Songaila, A., Blandford, R., Hu, E. M., &
Shopbell,P. 1996, ApJ, 471, L5
Cowie,L. L., Songaila, A., Hu, E. M., & Cohen, J. G. 1996, AJ, 112, 839
Daddi, E., Cimatti, A.,Pozzetti, L., Hoekstra, H., Röttgering, H. J. A., Renzini, A.,Zamorani,
G., & Mannucci, F. 2000, A&A, 361, 535
Daddi, E., et al. 2003, ApJ, 588, 50
Dickinson, M., Philos. Trans. R. Soc. London A 358, p. 2001, 2000
Dickinson, M., Papovich, C., Ferguson,H. C., & Budavári, T. 2003, ApJ, 587, 25
Eddington, A. S. 1916, MNRAS, 76, 525
Efstathiou, G., et al. 2002, MNRAS, 330, L29
Eggen, Lynden-Bell, & Sandage 1962
Ellis, R. S., Colless,M., Broadhurst, T., Heyl, J., & Glazebrook, K. 1996, MNRAS, 280,
235
Elston, R.,Rieke, G. H., & Rieke, M. J. 1988, ApJ, 331, L77
Erb, D. K., Shapley, A. E., Steidel, C. C., Pettini, M., Adelberger, K. L., Hunt, M. P.,
Moorwood, A. F. M., & Cuby, J. 2003, ApJ, 591, 101
Fontana, A., Poli, F.,Menci, N., Nonino, M., Giallongo, E., Cristiani, S., & D’Odorico,
S. 2003, ApJ, 587, 544
Fontana, A., et al. 2003, ApJ, 594, L9
Forster Schreiber et al. 2004,ApJ, accepted
Forster Schreiber et al. 2004, in preparation
Franx, M. et al., The Messenger 99, pp. 20–22, 2000
Franx, M. et al. 2003, ApJ, 587, L79
Freedman, W. L., et al. 2001, ApJ, 553, 47
Giavalisco, M. & Dickinson, M. 2001, ApJ, 550, 177
Giavalisco, M., Steidel, C. C., Adelberger, K. L., Dickinson, M. E., Pettini, M., & Kellogg,
M. 1998, ApJ, 503, 543
Guhathakurta, P., Tyson, J. A., & Majewski, S. R. 1990, ApJ, 357, L9
Hall, P. B., et al. 2001,AJ, 121, 1840
Hu, E. M. & Ridgway,S. E. 1994, AJ, 107, 1303
Hubble, E. P. 1930, ApJ, 71, 231
Ivison, R. J., Smail, I., Barger, A. J., Kneib, J.-P., Blain, A. W., Owen, F. N., Kerr,
T. H., & Cowie, L. L. 2000, MNRAS, 315, 209
Katz, N. & Gunn, J. E. 1991, ApJ, 377, 365
Kauffmann, G., Charlot, S., & White, S. D. M. 1996, MNRAS, 283,
Kauffmann, G. & White, S. D. M. 1993, MNRAS, 261, 921
Kauffmann, G., White, S. D. M., & Guiderdoni, B. 1993, MNRAS, 264, 201
Labbé, I., et al. 2003, AJ, 125, 1107
Labbé et al 2004, in preparation
Larson, R. B. 1975, MNRAS, 173, 671
Lilly, S. J., Le Fevre, O., Hammer, F., & Crampton, D. 1996, ApJ, 460,L1
1.3 Conclusions and Outlook
13
Lowenthal, J. D. et al. 1997, ApJ, 481, 673
Madau, P., Ferguson, H. C., Dickinson, M. E., Giavalisco, M., Steidel, C. C., & Fruchter,
A., MNRAS283, pp. 1388–1404, 1996
McCarthy, P. J., etal. 2001, ApJ, 560, L131
Mo, H. J. & Fukugita,M. 1996, ApJ, 467, L9
Moorwood, A. F. 1997, Proc. SPIE, 2871, 1146
Moorwood, A. F. M., van der Werf, P. P., Cuby, J. G., & Oliva, E. 2000,A&A, 362, 9
Papovich, C., Dickinson, M., & Ferguson, H. C. 2001, ApJ, 559, 620
Pettini, M., Kellogg, M., Steidel, C. C., Dickinson, M., Adelberger, K. L., & Giavalisco,
M. 1998, ApJ, 508, 539
Pettini, M., Shapley, A. E., Steidel, C. C., Cuby, J., Dickinson, M., Moorwood, A. F. M.,
Adelberger, K. L., & Giavalisco, M. 2001, ApJ, 554, 981
Pettini, M., Shapley, A. E., Steidel, C. C., Cuby, J., Dickinson, M., Moorwood, A. F. M.,
Adelberger, K. L., & Giavalisco, M. 2001, ApJ, 554, 981
Rudnick, G. et al. 2001, AJ, 122, 2205
Rudnick, G., et al. 2003, ApJ, 599, 847
Sandage, A. & Visvanathan, N. 1978, ApJ, 223, 707
Saracco, P., D’Odorico, S., Moorwood, A., & Cuby, J. G. 2001, Ap&SS, 276,991
Saracco, P., et al. 2003, A&A, 398, 127
Sawicki, M. & Yee, H. K. C. 1998, AJ, 115, 1329
Scodeggio, M. &Silva, D. R. 2000, A&A, 359, 953
Shapley, A. E., Steidel, C. C., Adelberger, K. L., Dickinson, M., Giavalisco, M., & Pettini,
M. 2001, ApJ, 562, 95
Smail, I., Ivison, R. J., Owen, F. N., Blain, A. W., & Kneib, J.-P. 2000, ApJ, 528, 612
Smith, N. 2002, MNRAS, 337,1252
Smith, V. V., et al. 2002, AJ, 124, 3241
Somerville, R. S. & Primack, J. R. 1999, MNRAS, 310, 1087
Songaila, A., Cowie, L. L., Hu, E. M., & Gardner, J. P. 1994, ApJS, 94,461
Songaila,A., Cowie, L. L., & Lilly, S. J. 1990, ApJ, 348, 371
Spergel, D. N., et al. 2003, ApJS, 148, 175
Steidel, C. C., Giavalisco, M., Dickinson, M., & Adelberger, K. L. 1996, AJ, 112, 352
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K. L. 1996, ApJ,
462, L17
Steidel, C. C., Adelberger, K. L., Giavalisco, M., Dickinson, M., & Pettini, M. 1999, ApJ,
519, 1
Steidel, C. C., Adelberger, K. L., Shapley, A. E., Pettini, M., Dickinson, M., & Giavalisco,
M. 2003, ApJ, 592, 72
Steidel, C. C. & Hamilton, D. 1992, AJ, 104, 941
Steidel, C. C., Pettini, M., & Hamilton, D. 1995, AJ, 110, 2519
Steinmetz, M. & Navarro, J. F. 2002, New Astronomy, 7, 155
Thompson, R. I.,Storrie-Lombardi, L. J., Weymann, R. J., Rieke, M. J., Schneider, G.,
Stobie, E., & Lytle, D. 1999, AJ, 117, 17
Totani, T., Yoshii, Y.,Maihara, T., Iwamuro, F., & Motohara, K. 2001, ApJ, 559, 592
Trager, S. C., Faber, S. M., Worthey, G., & González, J. J. 2000, AJ, 120, 165
van Albada, T. S. 1982, MNRAS, 201, 939
van Dokkum, P. G., Franx, M., Kelson, D. D., Illingworth, G. D., Fisher, D., & Fabricant,
D. 1998, ApJ, 500, 714
van Dokkum, P. G., Franx, M., Fabricant, D., Illingworth, G. D., & Kelson, D. D. 2000,
ApJ, 541, 95
14
1 Introduction and Summary
van Dokkum, P. G. et al. 2003, ApJ, 587, L83
van Dokkum, P. G., et al. 2004, ApJ, 611, 703
White, S. D. M. & Frenk, C. S. 1991, ApJ, 379, 52
White, S. D. M. & Rees, M. J. 1978, MNRAS, 183, 341
Williams, R. E. et al. 1996, AJ, 112, 1335
Williams, R. E. et al. 2000, AJ, 120, 2735
Yan, L., McCarthy, P. J.,Weymann, R. J., Malkan, M. A., Teplitz, H. I., StorrieLombardi, L. J.,Smith, M., & Dressler, A. 2000, AJ, 120, 575
Zwicky, F. 1937, ApJ, 86, 217
CHAPTER
TWO
Ultradeep Near-Infrared ISAAC Observations of
the Hubble Deep Field South
observations, reduction, multicolor catalog
and photometric redshifts
ABSTRACT
We present deep near-infrared (NIR) Js , H, and Ks -band ISAAC imaging
of the WFPC2 field of the Hubble Deep Field South (HDF-S). The 2.50 ×
2.50 high Galactic latitude field was observed with the VLT under the best
seeing conditions with integration times amounting to 33.6 hours in Js , 32.3
hours in H, and 35.6 hours in Ks . We reach total AB magnitudes for point
sources of 26.8, 26.2, and 26.2 respectively (3σ), which make it the deepest
ground-based NIR observations to date, and the deepest Ks -band data in
any field. The effective seeing of the coadded images is ≈ 0.00 45 in Js , ≈
0.00 48 in H, and ≈ 0.00 46 in Ks . Using published WFPC2 optical data, we
constructed a Ks -limited multicolor catalog containing 833 sources down
tot
. 26, of which 624 have seven-band optical-to-NIR photometry.
to Ks,AB
These data allow us to select normal galaxies from their rest-frame optical
properties to high redshift (z . 4). The observations, data reduction and
properties of the final images are discussed, and we address the detection and
photometry procedures that were used in making the catalog. In addition,
we present deep number counts, color distributions and photometric redshifts
of the HDF-S galaxies. We find that our faint Ks -band number counts are
flatter than published counts in other deep fields, which might reflect cosmic
variations or different analysis techniques. Compared to the HDF-N, we
find many galaxies with very red V − H colors at photometric redshifts
1.95 < zphot < 3.5. These galaxies are bright in Ks with infrared colors
redder than Js − Ks > 2.3 (in Johnson magnitudes). Because they are
extremely faint in the observed optical, they would be missed by ultravioletoptical selection techniques, such as the U-dropout method.
Ivo Labbé, Marijn Franx, Gregory Rudnick, Natascha M. Förster Schreiber,
Hans-Walter Rix, Alan Moorwood, Pieter G. van Dokkum, Paul van der Werf,
Huub Röttgering, Lottie van Starkenburg, Arjen van de Wel, Konrad Kuijken, &
Emanuele Daddi
Astronomical Journal, 125, 3, 1107 (2003)
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
16
multicolor catalog, and photometric redshifts
1
Introduction
I
n the past decade, our ability to routinely identify and systematically study
distant galaxies has dramatically advanced our knowledge of the high-redshift
universe. In particular, the efficient U-dropout technique (Steidel et al. 1996a,b)
has enabled the selection of distant galaxies from optical imaging surveys using
simple photometric criteria. Now more than 1000 of these Lyman break galaxies
(LBGs) are spectroscopically confirmed at z & 2, and have been subject to targeted studies on spatial clustering (Giavalisco & Dickinson 2001), internal kinematics (Pettini et al. 1998, 2001), dust properties (Adelberger & Steidel 2000), and
stellar composition (Shapley et al. 2001; Papovich, Dickinson, & Ferguson 2001).
Although LBGs are among the best studied classes of distant galaxies to date,
many of their properties like their prior star formation history, stellar population
ages, and masses are not well known.
More importantly, it is unclear if the ultraviolet-optical selection technique
alone will give us a fair census of the galaxy population at z ∼ 3 as it requires
galaxies to have high far-ultraviolet surface brightnesses due to on-going spatially
compact and relatively unobscured massive star formation. We know that there
exist highly obscured galaxies, detected in sub-mm and radio surveys (Smail et al.
2000), and optically faint hard X-ray sources (Cowie et al. 2001; Barger et al. 2001)
at high redshift that would not be selected as LBGs, but their number densities
are low compared to LBGs and they might represent rare populations or transient
evolutionary phases. In addition, the majority of present-day elliptical and spiral
galaxies, when placed at z ∼ 3, would not satisfy any of the current selection
techniques for high-redshift galaxies. Specifically, they would not be selected as
U-dropout galaxies because they are too faint in the rest-frame UV. It is much
easier to detect such galaxies in the near-infrared (NIR), where one can access
their rest-frame optical light.
Furthermore, observations in the near-infrared allow the comparison of galaxies
of different epochs at fixed rest-frame wavelengths where long-lived stars may
dominate the integrated light. Compared to the rest-frame far-UV, the rest-frame
optical light is less sensitive to the effects of dust extinction and on-going star
formation, and provides a better tracer of stellar mass. By selecting galaxies in
the near-infrared Ks -band, we expect to obtain a more complete census of the
galaxies that dominate the stellar mass density in the high-redshift universe, thus
tracing the build-up of stellar mass directly.
In this context we initiated the Faint InfraRed Extragalactic Survey (FIRES;
Franx et al. 2000), a large public program carried out at the Very Large Telescope
(VLT) consisting of very deep NIR imaging of two selected fields. We observed
fields with existing deep optical WFPC2 imaging from the Hubble Space Telescope (HST): the WPFC2-field of Hubble Deep Field South (HDF-S), and the
field around the z ≈ 0.83 cluster MS1054-03. The addition of NIR data to the
optical photometry is required not only to access the rest-frame optical, but also to
2.2 Observations
17
determine the redshifts of faint galaxies from their broadband photometry alone.
While it may be possible to go to even redder wavelengths from the ground, the
gain in terms of effective wavelength leverage is less dramatic compared to the
threefold increase going from the I to K-band. This is because the Ks -band is
currently the reddest band where achievable sensitivity and resolution are reasonably comparable to deep space-based optical data. Preliminary results from this
program were presented by Rudnick et al. (2001, hereafter R01).
Here we present the full NIR data set of the HDF-S, together with a Ks -selected
multicolor catalog of sources in the HDF-S with seven-band optical-to-infrared
photometry (covering 0.3 − 2.2µm), unique in its image quality and depth. This
paper focusses on the observations, data reduction and characteristic properties
of the final images. We also describe the source detection and photometric measurement procedures and lay out the contents of the catalog, concluding with the
NIR number counts, color distributions of sources, and their photometric redshifts.
The results of the MS1054-03 field will be presented by Förster Schreiber et al.
(2002) and a more detailed explanation of the photometric redshift technique can
be found in Rudnick et al. (2001, 2002b). Throughout this paper, all magnitudes
are expressed in the AB photometric system (Oke 1971) unless explicitly stated
otherwise.
2
2.1
Observations
Field Selection and Observing Strategy
The high Galactic latitude field of the HDF-S is a natural choice for follow-up
in the near-infrared given the existing ultradeep WFPC2 data in four optical
filters (Williams et al. 1996, 2000; Casertano et al. 2000). The Hubble Deep
Fields (North and South) are specifically aimed at constraining cosmology and
galaxy evolution models, and in these studies it is crucial to access rest-frame
optical wavelengths at high redshift through deep infrared observations. Available
ground-based NIR data from SOFI on the NTT (da Costa et al. 1998) are not
deep enough to match the space-based data. To fully take advantage of the deep
optical data requires extremely deep wide-field imaging in the infrared at the best
possible image quality; a combination that in the southern hemisphere can only be
delivered by the Infrared Spectrometer And Array Camera (ISAAC; Moorwood
1997), mounted on the Nasmyth-B focus of the 8.2 meter VLT Antu telescope.
The infrared camera has a 2.50 × 2.50 field of view similar to that of the WFPC2
(2.70 × 2.70 ). ISAAC is equipped with a Rockwell Hawaii 1024 × 1024 HgCdTe
array, offering imaging with a pixel scale of 0.00 147 pix−1 in various broad and
narrow band filters.
Our NIR imaging consists of a single ISAAC pointing centered on the WFPC2
main-field of the HDF-S (α = 22h 32m 55.464, δ = −60◦ 330 05.0100 , J2000) in the
Js , H and Ks filters, which gives good sampling of rest-frame optical wavelengths
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
18
multicolor catalog, and photometric redshifts
Figure 1 — Shown are the raw data in the filters Js (dotted line or circles), H (dashed line
or squares), and Ks (solid line or triangles). (a) Histogram of the median seeing in the raw
ISAAC images weighted by the weight function of Eq. 2 used to combine the images. (b) Relative
instrumental counts in a ≈ 300 radius aperture of four bright non-saturated stars in individual
sky-subtracted exposures, plotted against Julian Date. The relative increase in counts, slightly
dependent on wavelength, after cleaning and re-aluminization of the mirror directly reflects the
increase in efficiency of the telescope, because the sky background levels (c) remained the same.
Presumably, the photons were scattered by dirt rather than absorbed before cleaning. (d) Nightly
sky variations are largest and most rapid in the H-band and mean sky levels are highest at the
beginning and ending of the night. Js -band varies less and peaks at the start of the night,
whereas K-band levels are most stable.
over the redshift range 1 < z < 4. The Js filter is being established as the
new standard broadband filter at ≈ 1.24µm by most major observatories (Keck,
Gemini, Subaru, ESO), and is photometrically more accurate than the classical
J because it is not cut off by atmospheric absorption. It is a top-hat filter with
2.2 Observations
19
sharp edges, practically the same effective wavelength as the normal J filter, and
half-transmittance points at 1.17µm and 1.33µm. We used the Ks filter which is
bluer and narrower than standard K, but gives a better signal-to-noise ratio (SNR)
for faint sources because it is less affected by the high thermal background of the
atmosphere and the telescope. The ISAAC H and Ks filters are close to those
used to establish the faint IR standard star system (Persson et al. 1998), while the
Js filter requires a small color correction. The WFPC2 filters that are used are
F 300W, F 450W, F 606W and F 814W which we will call U300 , B450 , V606 and I814 ,
respectively, where the subscript indicates the central wavelength in nanometers.
The observing strategy for the HDF-S follows established procedures for groundbased NIR imaging. The dominance of the sky background and its rapid variability
in the infrared requires dithering of many short exposures. We used a 2000 jitter
box in which the telescope is moved in a random pattern of Poissonian offsets between successive exposures. This jitter size is a trade-off between keeping a large
area at maximum depth and ensuring that each pixel has sufficient exposures on
sky. Individual exposures have integration times of 6 × 30 s in Js , 6 × 20 s in H,
and 6 × 10 s in Ks (subintegrations × detector integration times). We requested
service mode observations amounting to 32 hours in each band with a seeing requirement of . 0.00 5, seeing conditions that are only available 25% of the time at
Paranal. The observations were grouped in 112 observation blocks (OBs), each of
which uniquely defines a single observation of a target, including pointing, number
of exposures in a sequence, and filter. The calibration plan for ISAAC provides the
necessary calibration measurements for such blocks, including twilight flats, detector darks, and nightly zero points by observing LCO/Palomar NICMOS standard
stars (Persson et al. 1998).
2.2
Observations
The HDF-S was observed from October to December 1999 and from April to October 2000 under ESO program identification 164.O-0612(A). A summary of the
observations is shown in 1. We obtained a total of 33.6, 32.3 and 35.6 hours in
Js , H and Ks , distributed over 33, 34 and 55 OBs, or 1007, 968 and 2136 frames,
respectively. This represents all usable data, including aborted and re-executed
OBs that were outside weather specifications or seeing constraints. In the reduction process these data are included with appropriate weighting (see section
3.4). Sixty-eight percent of the data was obtained under photometric conditions
and the average airmass of all data was 1.25. A detailed summary of observational parameters with pointing, observation date, image quality and photometric conditions can be found on the FIRES homepage on the World Wide Web
(http://www.strw.leidenuniv.nl/~fires).
An analysis of various observational parameters reveals some surprising trends
in the data, whereas other expected relations are less apparent. An overview is
given in Figure 1. The median seeing on the raw images is better than 0.00 5 in all
bands, with the seeing of 90% of the images in the range 0.00 4 − 0.00 65, as can be
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
20
multicolor catalog, and photometric redshifts
seen in Figure 1.
Seeing may vary strongly on short timescales but it is not related to any other
parameter. The most drastic trend in the raw data is the change of sensitivity
with date. Since the cleaning and re-alumization of the primary mirror in March
2000 the count rates of bright stars within a ≈ 300 aperture increased by +29% in
Js , +45% in H, and +45% Ks , which is reflected by a change in zero points before
and after this date. Because the average NIR sky levels remained the same, this
increase proportionally improved the achievable signal-to-noise for backgroundlimited sources. The change in throughput was caused by light scattering, which
explains why the sky level remained constant. Sky levels in Js and H, dominated
by airglow from OH-emission lines in the upper atmosphere (typically 90 km altitude), vary unpredictably on the timescale of minutes, but also systematically
with observed hour. The average sky level is highest at the beginning and end
of each night with peak-to-peak amplitudes of the variation being 50% relative to
the average sky brightnesses over the night. The background in Ks is dominated
by thermal emission of the telescope, instrument, and atmosphere and is mainly a
function of temperature. The Ks background is the most stable of all NIR bands
and only weakly correlated with airmass; our data do not show a strong thermal atmospheric contribution, which should be proportional to atmospheric path
length. We take into account the variations of the background and seeing through
weighting in the data coadding process.
3
Data Reduction
The reduction process included the following steps: quality verification, flatfielding, bad pixel correction, sky subtraction, distortion correction, registration,
photometric calibration and weighting of individual frames, and combination into
a single frame. We used a modified version of the DIMSUM1 package and standard
routines in IRAF2 for sky subtraction and coadding, and the ECLIPSE3 package
for creating the flatfields and the initial bad pixel masks. We reduced the ISAAC
observations several times with an increasing level of sophistication, applying corrections to remove instrumental features, scattered light, or clear artifacts when
required. Here we describe the first version of the reduction (v1.0) and the last
version (v3.0), leaving out the intermediate trial versions. The last version produced the final Js , H and Ks images, on which the photometry (see section 5) and
analysis (see section 8) is based.
1 DIMSUM is the Deep Infrared Mosaicing Software package developed by Peter Eisenhardt,
Mark Dickinson, Adam Stanford, and John Ward, and is available via ftp to ftp://iraf.noao.
edu/contrib/dimsumV2/
2 IRAF is distributed by the National Optical Astronomy Observatories, which are operated
by the AURA, Inc., under cooperative agreement with the NSF.
3 ECLIPSE is an image processing package written by N. Devillard, and is available at ftp:
//ftp.hq.eso.org/pub/eclipse/
2.3 Data Reduction
3.1
21
Flatfields and Photometric Calibration
We constructed flatfields from images of the sky taken at dusk or dawn, grouped
per night and in the relevant filters, using the flat routine in ECLIPSE, which
also provided the bad pixel maps. We excluded a few flats of poor quality and
flats that exhibited a large jump between the top row of the lower and bottom
row of the upper half of the array, possibly caused by the varying bias levels of
the Hawaii detector. We averaged the remaining nightly flats per month, and
applied these to the individual frames of the OBs taken in the same month. If
no flatfield was available for a given month we used an average flat of all months.
The stability of these monthly flats is very good and the structure changes little
and in a gradual way. We estimate the relative accuracy to be 0.2 − 0.4% per pixel
from the pixel-to-pixel rms variation between different monthly flats. Large scale
gradients in the monthly flats do not exceed 2%. We checked that standard stars,
which were observed at various locations on the detector, were consistent within
the error after flatfielding.
Standard stars in the LCO/Palomar NICMOS list (Persson et al. 1998) were
observed each night, in a wide five-point jitter pattern. For each star, on each
night, and in each filter, we measured the instrumental counts in a circular aperture of radius 20 pixel (2.00 94) and derived zero points per night from the magnitude of that star in the NICMOS list. We identify non-photometric nights after
comparison with the median of the zero points over all nights before and after
re-aluminization in March 2000 (see section 2.2). The photometric zero points
exhibit a large increase after March 2000 but, apart from this, the night-to-night
scatter is approximately 2%. We adopted the mean of the zero points after March
2000 as our reference value. See Table 2 for the list of the adopted zero points.
By applying the nightly zero points to 4 bright unsaturated stars in the HDF-S,
observed on the same night under photometric conditions, we obtain calibrated
stellar magnitudes with a night-to-night rms variation of only ≈ 1 − 1.5%. No
corrections for atmospheric absorption were required because the majority of the
science data were obtained at similar airmass as the standard star observations.
In addition, instrumental count rates of HDF-S stars in individual observation
blocks reveal no correlation with airmass. We used the calibrated magnitudes
of the 4 reference stars, averaged over all photometric nights, to calibrate every
individual exposure of the photometric and non-photometric OBs. The detector
non-linearity, as described by Amico et al. (2001), affects the photometric calibration by . 1% in the H-band, where the exposure levels are highest. Because the
effect is so small, we do not correct for this. We did not account for color terms due
to differences between the ISAAC and standard filter systems. Amico et al. (2001)
report that the ISAAC H and Ks filters match very well those used to establish
the faint IR standard star system of Persson et al. (1998). Only the ISAAC J s
filter is slightly redder than Persson’s J and this may introduce a small color term,
≈ −0.04 · (J − K)LCO . However, the theoretical transformation between ISAAC
magnitudes and those of LCO/Palomar have never been experimentally verified.
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
22
multicolor catalog, and photometric redshifts
Furthermore, the predicted color correction is small and could not be reproduced
with our data. In the absence of a better calibration we chose not to apply any
color correction. We did apply Galactic extinction correction when deriving the
photometric redshifts, see section 6, but it is not applied to the catalog.
As a photometric sanity check, we compared 200 circular diameter aperture
magnitudes of the brightest stars in the final (version 3.0, described below) images
to magnitudes based on a small fraction of the data presented by R01. Each
data set was independently reduced, the calibration based on different standard
stars, and the shallower data were obtained before re-aluminization of the primary
mirror. The magnitudes of the brightest sources in all bands agree within 1%
between the versions, indicating that the internal photometric systematics are well
under control. For the NIR data, the adopted transformations from the Johnson
(1966) system to the AB system are taken from Bessell & Brett (1988) and we
apply Js,AB = Js,V ega + 0.90, HAB = HV ega + 1.38 and Ks,AB = Ks,V ega + 1.86.
3.2
Sky Subtraction and Cosmic Ray Removal
The rapidly varying sky, typically 25 thousand times brighter than the sources we
aim to detect, is the primary limiting factor in deep NIR imaging. In the longest
integrations, small errors in sky subtraction can severely diminish the achievable
depth and affect faint source photometry. The IRAF package DIMSUM provides a
two-pass routine to optimally separate sky and astronomical signal in the dithered
images. We modified it to enable handling of large amounts of data and replaced
its co-adding subroutine, which assumes that the images are undersampled, by the
standard IRAF task IMAGES.IMMATCH.IMCOMBINE. The following is a brief
summary of the steps performed by the REDUCE task in DIMSUM.
For every science image in a given OB a sky image is constructed. After scaling
the exposures to a common median, the sky is determined at each pixel position
from a maximum of 8 and a minimum of 3 adjacent frames in time. The lowest
and highest values are rejected and the average of the remainder is taken as the
sky value. These values are subtracted from the scaled image to create a sky
subtracted image. A set of stars is then used to compute relative shifts, and the
images are integer registered and averaged to produce an intermediate image. All
astronomical sources are identified and a corresponding object mask is created.
This mask is used in a second pass of sky subtraction where pixels covered by
objects are excluded from the estimate of the sky.
The images show low-level pattern due to bias variations. Because they generally reproduce they are automatically removed in the skysubtraction step. We
find cosmic rays with DIMSUM, using a simple threshold algorithm and replacing
them by the local median, unless a pixel is found to have cosmic rays in more than
frames per OB. In this case the pixel is added to the bad pixel map for that OB.
2.3 Data Reduction
3.3
23
First Version and Quality Verification
The goal of the first reduction of the data set is to provide a non-optimized image,
which we use to validate and to assess the improvements from more sophisticated
image processing. The first version consists of registration on integer pixels and
combination of the sky subtracted exposures per OB. For each of the 122 OBs, we
created an average and a median combined image to verify that cosmic rays and
other outliers were removed correctly, and we visually inspected all 4149 individual
sky subtracted frames as well, finding that many required further processing as
described in the following section. Finally, we generated the version 1.0 images
(the first reduction of the full data set) by integer pixel shifting all OBs to a
common reference frame, and coaveraging them into the Js , H and Ks images.
While this first reduction is not optimal in terms of depth and image quality, it is
robust owing to its straightforward reduction procedure.
3.4
Additional Processing and Improvements
The individual sky subtracted frames are affected by a number of problems or
instrumental features, which we briefly describe below, together with the applied
solutions and additional improvements that lead to the version 3.0 images. The
most important problems are:
• Detector bias residuals, most pronounced at the rows where the read-out of
the detector starts at the bottom (rows 1, 2, ...) and halfway (rows 513, 514, ...),
caused by the complex bias behaviour of the Rockwell Hawaii array. These
variations are uniform along rows, and we removed the residual bias by subtracting the median along rows in individual sky subtracted exposures, after
masking all sources.
• Imperfect sky subtraction, caused by stray light or rapid background variations. Strong variations in the backgrounds, reflection from high cirrus,
reflected moonlight in the ISAAC optics or patterns of less obvious origin
can lead to large scale residuals in the sky subtraction, particularly in J s
and H. For some OBs, we succesfully removed the residual patterns by
splitting the sequence in two (in case of a sudden appearance of stray light),
or subtracting a two-piece cubic spline fit along rows and columns to the
background in individual frames, after masking all sources. We rejected a
few frames, or masked the affected areas, if this simple solution did not work.
• Unidentified cosmic rays or bad pixels. A small amount of bad pixels were not
detected by ECLIPSE or DIMSUM routines but need to be identified because
we average the final images without additional clipping or rejection. By
combining the sky subtracted frames in a given OB without shifts and with
the sources masked, we identified remaining cosmic rays or outliers through
sigma clipping. We added ∼ 60 − 100 pixels per OB to the corresponding
bad pixel map.
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
24
multicolor catalog, and photometric redshifts
Several steps were taken to improve the quality and limiting depths of the
version 1.0 images, the most important of which are:
• Distortion correction of the individual frames and direct registration to the
3×3 blocked I814 image (0.11900 pixel−1 ), our preferred frame of reference. We
obtained the geometrical distortion coefficients for the 3rd order polynomial
solution from the ISAAC WWW-page4 . The transformation procedure involves distortion correcting the ISAAC images, adjusting the frame-to-frame
shifts, and finding the linear transformation to the WFPC2 I814 frame of
reference. This linear transformation is the best fit mapping of source positions in the blocked WFPC2 I814 image to the corresponding positions in
the corrected Js -band image5 . Compared to version 1.0 described in the
previous section, this procedure increases registration accuracy and image
quality, decreases image smearing at the edges introduced by the jittering
and differential distortion. Given the small amplitude of the ISAAC field
distortions, the effect on photometry is negligible. In the linear transformation and distortion correction step the image is resampled once using a
third-order polynomial interpolation, with a minimal effect of the interpolant
on the noise properties.
• Weighting of the images. We substantially improved the final image depth
and quality by assigning weights to individual frames that take into account
changes in seeing, sky transparency, and background noise. Two schemes
were applied: one that optimizes the signal-to-noise ratio (SNR) within an
aperture of the size of the seeing disk, and one that optimizes the SNR per
pixel. The first improves the detection efficiency of point sources, the other
optimizes the surface brightness photometry. The weights wi of the frames
are proportional to either the inverse scaled variance zpscalei × vari within
a seeing disk of size si , or to the inverse scaled variance per pixel, where
the scaling zpscalei is the flux calibration applied to bring the instrumental
counts of our four reference stars in the HDF-S to the calibrated magnitude.
wi,point ∝ (zpscalei × vari × s2i )−1
wi,extended ∝ (zpscalei × vari )−1
3.5
(1)
(2)
Final Version and Post Processing
The final combined Js , H, and Ks images (version 3.0) were constructed from
the individually registered, distortion corrected, weighted and unclipped average
of the 1007, 968, and 2136 NIR frames respectively. Ultimately, less than 3% of
individual frames were excluded in the final images because of poor quality. In
this step we also generated the weight maps, which contain the weighted exposure
time per pixel. We produced three versions of the images, one with optimized
4 ISAAC
home page: http://www.eso.org/instruments/isaac
have noticed that the mapping solution changed slightly after the remount of ISAAC in
March 2000, implying a 0.1% scale difference.
5 We
2.4 Final Images
25
weights for point sources, one with optimized weights for surface brightness, and
one consisting of the best quartile seeing fraction of all exposures, also optimized
for point sources. The weighting has improved the image quality by 10–15% and
the background noise by 5–10%, and distortion correction resulted in subpixel
registration accuracy between the NIR images and I814 -band image over the entire
field of view.
The sky subtraction routine in DIMSUM and our additional fitting of rows
and columns (see section 3.4) have introduced small negative biases in combined
images, caused by systematic oversubtraction of the sky which was skewed by
light of the faint extended PSF wings or very faint sources, undetectable in a
single OB. Because of this, the flatness of the sky on large scales was limited to
about 10−5 . The negative bias was visible as clearly defined orthogonal stripes at
P.A.≈ 6◦ , as well as dark areas around the brightest stars or in the crowded parts
of the images. To solve this, we rotated a copy of the final images back to the
orientation in which we performed sky substraction, fitted a 3-piece cubic spline
to the background along rows and columns (masking all sources), re-rotated the
fit, and subtracted it. The sky in the final images is flat to a few ×10−6 on large
(> 2000 ) scales.
4
Final Images
The reduced NIR Js , H, and Ks images and weight maps can be obtained from the
FIRES-WWW homepage (http://www.strw.leidenuniv.nl/~fires). Throughout the rest of the paper we will only consider the images optimized for point source
detection which we will use to assemble the catalog of sources.
4.1
Properties
The pixel size in the NIR images equals that of the 3 × 3 blocked WFPC2 I814 band image at 0.11900 pixel−1 . The combined ISAAC images are aligned with the
HST version 2 images (Casertano et al. 2000) with North up, and are normalized
to instrumental counts per second. The images are shallower near the edges of
the covered area because they received less exposure time in the dithering process,
which is reflected in the weight map containing the fraction of total exposure
time per pixel. The area of the ISAAC Ks -band image with weight per pixel
wK ≥ 0.95, 0.2, and 0.01 covers 4.5, 7.2 and 8.3 arcmin2 , while the area used
for our preferred quality cut for photometry (w ≥ 0.2 in all seven bands) is 4.7
arcmin2 . The NIR images have been trimmed where the relative exposure time
per pixel is less than 1%.
Figure 2 shows the noise-equalized Ks -detection image obtained by division
with the square root of the exposure-time map. The richness in faint compact
sources and the flatness of the background are readily visible. Figure 3 shows a
RGB color composite image of the I814 , Js , and Ks images. The PSF of the space-
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
26
multicolor catalog, and photometric redshifts
Figure 2 — The HDF-S field in the ISAAC Ks -band divided by the square root of the weight
map (based on the fractional exposure time per pixel) and displayed at linear scaling. The total
integration time is 35.6 hours, the stellar FWHM≈ 0.00 46 and the total field size is 2.850 × 2.850 .
based I814 image has been matched to that of the NIR images at FWHM≈ 0.00 46
(see section 4.2) and three adjacent WFPC2 I814 flanking fields have been included
for visual purposes. We set the linear stretch of both images to favor faint objects.
Immediately striking is the rich variety in optical-NIR colors, even for the faint
objects, indicating that the NIR observations are very deep and that there is a
wide range of observed spectral shapes, which can result from different types of
galaxies over a broad redshift range.
2.4 Final Images
27
Figure 3 — Three-color composite image of the ISAAC field on top of the WFPC2 main-field
and parts of three WFPC2 flanking fields. The main-field is outlined in white and North is
up. The images are registered and smoothed to a common seeing of FWHM≈ 0.00 46, coding
WFPC2 I814 in blue, ISAAC Js in green and ISAAC Ks in red. There is a striking variety
in optical-to-infrared colors, especially for fainter objects. A number of sources with red colors
have photometric redshifts z > 2 and they are candidates for relatively massive, evolved galaxies.
These galaxies would not be selected by the U-dropout technique because they are too faint in
the observer’s optical.
4.2
Image Quality
The NIR PSF is stable and symmetric over the field with a gaussian core profile
and an average ellipticity < 0.05 over the Js , H, and Ks images. The median
FWHM of the profiles of ten selected isolated bright stars is 0.00 45 in Js , 0.00 48 H,
and 0.00 46 Ks with 0.00 04 amplitude variation over the images.
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
28
multicolor catalog, and photometric redshifts
For consistent photometry in all bands we convolved the measurement images
to a common PSF, corresponding to that of the H-band which had worst effective
seeing (FWHM = 0.00 48). The similarity of PSF structure across the NIR images
allowed simple gaussian smoothing for a near perfect match. The complex PSF
structure of the WFPC2 requires convolving with a special kernel, which we constructed by deconvolving an average image of bright isolated non-saturated stars
in the H-band with the average I814 -band image of the same stars. Division of the
stellar growth curves of the convolved images by the H-band growth curve shows
that the fractional enclosed flux agrees to within 3% at radii r ≥ 0.00 35.
4.3
Astrometry
The relative registration between ISAAC and WFPC2 images needs to be very
precise, preferably a fraction of an original ISAAC pixel over the whole field of
view, to allow correct cross-identification of sources, accurate color information
and morphological comparison between different bands. To verify our mapping
of ISAAC to WFPC2 coordinates, we measured the positions of the 20 brightest
stars and compact sources in all registered ISAAC exposures, and we compared
their positions with those in the I814 image. The rms variation in position of
individual sources is about 0.2 − 0.3 pixel at 0.11900 pixel−1 (25 − 35 mas), but for
some sources systematic offsets between the NIR and the optical up to 0.85 pixel
(100 mas) remain. The origin of the residuals is unclear and we cannot fit them
with low order polynomials. They could be real, intrinsic to the sources, or due to
systematic errors in the field distortion correction of ISAAC or WFPC26 . However,
for all our purposes, the effect of positional errors of this amplitude is unimportant.
The error in absolute astrometry of the HST HDF-S coordinate system, estimated
to be less than 40 mas, is dominated by the systematic uncertainty in the positions
of four reference stars (Casertano et al. 2000; Williams et al. 2000).
4.4
Backgrounds and Limiting Depths
The noise properties of the raw individual ISAAC images are well described by
the variance of the signal collected in each pixel since both Poisson and read
noise are uncorrelated. However, image processing, registration and combination
have introduced correlations between neighbouring pixels and small errors in the
background subtraction may also contribute to the noise. Understanding the noise
properties well is crucial because limiting depths and photometric uncertainties
rely on them.
Instead of a formal description based on the analysis of the covariance of correlated pixel pairs, we followed an empirical approach where we fit the dependence
6 The
ISAAC field distortion might have changed over the years, but this cannot be checked
because recent distortion measurements are unavailable. The worst case errors of relative positions across the four WFPC2 chips can be 0.00 1 (Vogt et al. 1997), but is expected to be smaller
for the HDF-S images.
2.4 Final Images
29
Figure 4 √
— Scaling relation of the measured background rms noise as a function of linear
size N = A of apertures with area A. (a) Gaussians are fitted to histograms of K s counts
in randomly placed apertures of increasing size, excluding pixels belonging to sources. This
correctly accounts for pixel-to-pixel correlations and other effects, allowing us to measure the
true rms variation as a function of linear size of aperture. (b) The Ks -band results (solid points),
together with the best-fit scaling relation of Eq. 3 (solid line), show that the measured variation
in large apertures exceeds the variation expected from linear (Gaussian) scaling of the pixelto-pixel noise (dashed line), likely due to large scale correlated fluctuations of the background.
√
of the rms background variation in the image as a function of linear size N = A
of apertures with area A. Directly measuring the effective flux variations in apertures of different sizes provides a more realistic estimate of signal variations than
formal Gaussian scaling σ(N ) = N σ̄ of the pixel-to-pixel noise σ̄, as is often done.
We measured fluxes in 1200 non-overlapping circular apertures randomly placed
on the registered convolved images, which were also used for photometry. We ex-
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
30
multicolor catalog, and photometric redshifts
cluded all pixels belonging to sources detectable in Ks at the 5σ level (see section
5.1 for detection criteria). We used identical aperture positions for each band i and
measured fluxes for circular aperture diameters ranging from 0.00 5 to 300 . Then we
obtained the flux dispersions by fitting a Gaussian distribution to the histogram of
fluxes at each aperture size. Finally, we fitted a parameterized function of linear
size to the different dispersions:
√
σi (N ) = N σ̄i (ai + bi N )/ wi
(3)
This equation describes the signal variation versus aperture size N over the
entire image, taking into account spatial variations as a result of relative weight
wi for each passband i. As can be seen in Figure 4, it provides a good fit to the noise
characteristics. The noise is significantly higher than expected from uncorrelated
(Gaussian) noise, indicated by a dashed line in Figure 4b. Table 3 shows the best
fit values in all bands and the corresponding limiting depths. The parameter a
reflects the correlations of neighbouring pixels (a > 1), which is important in the
WFPC2 images because of heavy smoothing, but also in the ISAAC images given
the resampling from 0.00 147 to 0.00 119 pixel−1 . The parameter b accounts for large
scale correlated variations in the background (b > 0). This may be caused by
the presence of sources at very faint flux levels (confusion noise) or instrumental
features. Typically, the large scale correlated contribution per pixel is only 315% relative to the gaussian rms variation, but due to the N 2 proportionality
the contribution to the variation in large apertures increases to significant levels.
While the signal variations grow faster with area than expected from a Gaussian,
at any specific scale the variation is consistent with a pure Gaussian.
From the analysis of the scaling relation of simulated colors we find that part
of the large scale irregularities in the background are spatially correlated between
bands. In particular, we measured the rms variation of the I814 − V606 colors
directly by subtracting in registered apertures the I814 -band fluxes from the V606
fluxes and fitting the dispersion of the difference at each linear size. On large
scales rms variations are 30% smaller than predicted from Eq. 3 if the noise were
uncorrelated. Yet, if we subtract the two fluxes in random apertures, the scaling of
the background variation is consistent with the prediction. A similar effect is seen
for the I814 − Js color, but at a smaller amplitude. The spatial coherence of the
background variations between filters and across cameras suggests that part of the
background fluctuations may be associated with sources at very faint flux levels.
Other contributions are likely similar flatfielding or skysubtraction residuals from
one band to another.
5
Source Detection and Photometry
The detection of sources at very faint magnitudes against a noisy background forces
us to trade off completeness and reliability. A very low detection threshold may
2.5 Source Detection and Photometry
31
generate the most complete catalog, but we must then apply additional criteria to
assess the reliability of each detection given that such a catalog will contain many
spurious sources. More conservatively, we choose the lowest possible threshold for
which contamination by noise is unimportant. We aim to produce a catalog with
reliable colors suitable for robustly modeling of the intrinsic spectral energy distribution. Using SExtractor version 2.2.2 (Bertin & Arnouts 1996) with a detection
procedure that optimizes sensitivity for point-like sources, we construct a K s -band
selected catalog with seven band optical-to-infrared photometry.
5.1
Detection
To detect objects with SExtractor using a constant signal-to-noise criterion over
the entire image, including the shallower outer parts, we divide the point source
optimized Ks -image by the square root of the weight (exposure time) map to create
a noise-equalized detection image. A source enters the catalog if, after low-pass
filtering of the detection image, at least one pixel is above ≈ 5 times the standard
deviation of the filtered background, corresponding to a total Ks -band magnitude
limit for point sources of Ks ≈ 26.0. This depth is reached for the central 4.5
arcmin2 . In total we have 833 detections in the entire survey area of 8.3 arcmin2 .
Initially 820 sources are found but the detection software fails to detect sources
lying in the extended wings of the brightest objects. To include these, we fit the
surface brightness profiles of the brightest sources with the GALPHOT package
(Franx et al. 1989) in IRAF, subtract the fit, and carry out a second detection pass
with identical parameters. Thirteen new objects enter the catalog, and 9 sources
detected in the first pass are replaced with improved photometry. The catalog
identification numbers of all second-pass objects start at 10001, and the original
entries of the updated sources are removed.
Filtering affects only the detection process and the isophotal parameters; other
output parameters are affected only indirectly through barycenter and object extent. We chose a simple two-dimensional gaussian detection filter (FWHM= 0.00 46),
approximating the core of the effective Ks -band PSF well. Hence, we optimize detectability for point-like sources, introducing a small bias against faint extended
objects. In principle it is possible to combine multiple catalogs created with different filter sizes but merging these catalogs consistently is a complicated and
subjective process yielding a modest gain only in sensitivity for larger objects.
We prefer the small filter size equal to the PSF in the detection map because the
majority of faint sources that we detect are compact or unresolved in the NIR and
because we wish to minimize the blending effect of filtering on the isophotal parameters and on the confusion of sources. SExtractor applies a multi-thresholding
technique to separate overlapping sources based on the distribution of the filtered
Ks -band light. About 20% of the sources are blended because of the low value
of the isophotal threshold; in SExtractor this must always equal the detection
threshold. With the deblending parameters we used, the algorithm succeeds in
splitting close groups of separate galaxies, without “oversplitting” galaxies with
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
32
multicolor catalog, and photometric redshifts
rich internal structure.
We tested sensitivity to false detections by running SExtractor on a specially
constructed Ks -band noise map created by subtracting, in pairs, individual Ks band images of comparable seeing, after zero point scaling, and coaveraging the
weighted difference images. This noise image has properties very similar to the
noise in the original reduced image, including contributions from the detector
and reduction process, but with no trace of astronomical sources. Our detection
algorithm resulted in only 11 spurious sources over the full area.
5.2
Optical and NIR Photometry
We use SExtractor’s dual-image mode for spatially accurate and consistent photometry, where objects are detected and isophotal parameters are determined from
the Ks -band detection image while the fluxes in all seven bands are measured in
the registered and PSF matched images. We used fluxes measured in circular
apertures AP ER(D) with fixed-diameters D, isophotal apertures AP ER(ISO)
determined by the Ks -band detection isophote at the 5σ detection threshold, and
AP ER(AU T O) (autoscaling) apertures inspired by Kron (1980), which scales an
elliptical aperture based on the first moments of the Ks -band light distribution.
We select for each object the best aperture based on simple criteria to enable
detailed control of photometry. We define two types of measurements:
• “color” flux, to obtain consistent and accurate colors. The optimal aperture
is chosen based on the Ks flux distribution, and this aperture is used to
measure the flux in all other bands.
• “total” flux, only in the Ks -band, which gives the best estimate of the total
Ks flux.
For both measurements we treat blended sources differently from unblended ones,
and consider a source blended when its BLENDED or BIAS flag is set by SExtractor, as described in Bertin & Arnouts (1996).
Our color aperture is chosen
p as follows, introducing the equivalent of a circular
isophotal diameter Diso = 2 Aiso /π based on Aiso , the measured non-circular
isophotal area within the detection-isophote:
33
2.5 Source Detection and Photometry
if unblended
if blended

00
00

AP ER(ISO) (0. 7 < Diso < 2. 0)
AP ER(COLOR) = AP ER (0.00 7) (Diso ≤ 0.00 7)


AP ER (2.00 0) (Diso ≥ 2.00 0)


AP ER (Diso /s)
AP ER(COLOR) = AP ER (0.00 7)


AP ER (2.00 0)
(0.00 7 < Diso /s < 2.00 0)
(Diso /s ≤ 0.00 7)
(Diso /s ≥ 2.00 0)
(4)
The parameter s is the factor with which we shrink the circular apertures centered on blended sources, increasing the separation to the blended neighbour such
that mutual flux contamination is minimal. This factor depends on the data set,
and for our ISAAC Ks image we find that s = 1.4 is most successful. The smallest
aperture considered, APER(0.00 7), ≈ 1.5 FWHM of the effective PSF, optimizes
the S/N for photometry of point sources in unweighted apertures and prevents
smaller more error-prone apertures. The largest allowed aperture, APER(2.00 0),
prevents large and inaccurate isophotal apertures driven by the filtered Ks -light
distribution. We continuously assessed the robustness and quality of color flux
measurements by inspecting the fits of redshifted galaxy templates to the flux
points, as described in detail in §6.
We calculate the total flux in the Ks -band from the flux measured
p in the
AUTO aperture. We define a circularized AUTO diameter Dauto = 2 Aauto /π
with Aauto the area of the AUTO aperture, and define the total magnitude as:
if unblended
AP ER(T OT AL) = AP ER(AU T O)
if blended
(5)
AP ER(T OT AL) = AP ER(COLOR)
Finally, we apply an aperture correction using the growth curve of brighter
stars to correct for the flux lost because it fell outside the “total” aperture. This
aperture correction is necessary because it is substantial for our faintest sources,
as shown in Figure 5 where we compare different methods to estimate magnitude.
The aperture correction reaches 0.7 mag at the faint end, therefore magnitudes
are seriously underestimated if the aperture correction is ignored.
We derive the 1σ photometric error for all measurements from Eq. 3 with the
best-fit values shown in Table 3. These errors may overestimate the uncertainty in
colors of adjacent bands (see section 4.4) but it should represent well the photo-
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
34
multicolor catalog, and photometric redshifts
Figure 5 — Comparison of methods to estimate total Ks -band magnitude. Shown are isophotal
(top), SExtractor’s auto-scaling AUTO (middle), and our “total” magnitudes (bottom) as defined
in Eq. 5 and which are aperture corrected using stellar growth curve analysis. We subtracted
the aperture corrected magnitude measured in an aperture of 0.00 7 (M AGAP ER(0.00 7) − 0.7),
which produces the correct total magnitudes for stars and pointlike sources. Stars are marked by
star symbols and fluxes are plotted with ±1σ error bars. The turn-up at K s ≈ 24 of isophotal
and at Ks ≈ 25 of the AUTO magnitudes shows that these photometric schemes systematically
underestimate the total flux at faint levels, due to the decreasing size of the used aperture with
magnitude. This effect is nearly absent in the bottom panel, which shows the total magnitudes
measured in this paper.
metric error over the entire 0.3µ − 2.2µm wavelength range. The magnitudes may
suffer from additional uncertainties because of surface brightness biases or possible
biases in the sky subtraction procedure which could depend on object magnitude
and size.
35
2.6 Photometric Redshifts
Figure 6 — Direct comparison of photometric redshifts to the 39 spectroscopic redshifts of
objects in the HDF-S with good photometry in all bands. The 68% error bars are derived from
our Monte Carlo simulations and the diagonal line corresponds to a one-to-one relation to guide
the eye. While the agreement is excellent with no failures for this small sample and with mean
∆z/(1+z) = 0.08, large asymmetric uncertainties remain for some objects indicating the presence
of a second photometric redshift solution of comparable likelihood at a different redshift.
6
Photometric Redshifts
To physically interpret the seven-band photometry for our Ks -band selected sample, we use a photometric redshift (zphot ) technique explained in detail by R01. In
summary, we correct the observed flux points for Galactic extinction (see Schlegel,
Finkbeiner, & Davis 1998) and we model the rest-frame colors of the galaxies by
fitting a linear combination of redshifted empirical galaxy templates. The redshift
with the lowest χ2 statistic, where
Nf ilter
X · F data − F model ¸2
i
i
χ (z) =
data
σ
i
i=1
2
(6)
is then chosen as the most likely zphot . Using a linear combination of SEDs as
F model minimizes the a priori assumptions about the nature and stellar composition of the detected sources.
Our data set with three deep NIR bands samples the position of Balmer/4000 Å
break over 1 . z . 4, allowing us to probe the redshift distribution of more evolved
galaxy types that may have little rest-frame UV flux and hence a weak or virtually
absent Lyman break.
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
36
multicolor catalog, and photometric redshifts
6.1
Photometric Templates
We used the local Hubble type templates E, Sbc, Scd, and Im from Coleman,
Wu, & Weedman (1980), the two starburst templates, SB1 and SB2, with low
derived reddening from Kinney et al. (1996), and a 10 Myr old single age template
model from Bruzual & Charlot (2002). The starburst templates are needed because many galaxies even in the nearby Universe have bluer colors than the bluest
CWW templates. The observed templates are extended beyond their published
wavelengths into the far-ultraviolet by power law extrapolation and into the NIR
using stellar population synthesis models from Bruzual & Charlot (2002), with
the initial mass functions and star-formation timescales for each template Hubble
type from Pozzetti, Bruzual, & Zamorani (1996). We accounted for internal hydrogen absorption of each galaxy by setting the flux blueward of the 912 Å Lyman
limit to zero, and for the redshift-dependent cosmic mean opacity due to neutral
intergalactic hydrogen by following the prescriptions of Madau (1995).
6.2
Zphot Uncertainties
The best test of photometric redshifts is direct comparison to spectroscopic redshifts, but spectroscopic redshifts in the HDF-S are still scarce. We calculate the
uncertainty in the photometric redshift due to the flux measurement errors using
a Monte-Carlo (MC) technique derived from that used in R01 and fully explained
in Rudnick et al. (2002b). At bright magnitudes template mismatch dominates
the errors, something that is not modeled by the MC simulation. Hence, the MC
error bars for bright galaxies are severe underestimates. At fainter magnitudes,
the uncertainty is driven by errors in photometry (Fernández-Soto, Lanzetta, &
Yahil 1999) and the MC technique should provide accurate zphot uncertainties.
Experience from R01 showed that two ways to correct for the template mismatch,
setting a minimum fractional flux error or setting a minimum zphot error based on
the mean disagreement with zspec , either degrade the accuracy of the zphot measurement or reflect the systematic error only in the mean, while template mismatch
can be a strong function of SED shape and redshift. A method based completely
on Monte-Carlo techniques is preferable because it has a straightforwardly computable redshift probability function. This approach is desirable for estimating
the rest-frame luminosities and colors (Rudnick et al. 2002b).
Therefore, we modify the MC errors directly using the FIRES photometry.
In summary, we estimate the systematic component of the zphot uncertainty by
scaling up all the photometric errors for a given galaxy with a constant to bring
residuals of the fit in agreement with the errors. This will not change the best fit
redshift and SED and will not modify the MC errorbars of faint objects, but it
will enlarge the redshift interval over which the templates can satisfactorily fit the
bright objects. Only in case of widely different photometric errors between the
visible and infrared might the modified MC uncertainties still underestimate the
true zphot uncertainty.
2.6 Photometric Redshifts
37
Figure 7 — Js − Ks versus I814 − Ks color-color diagram (on the AB system) for the sources
with Ks < 24 in the HDF-S with a minimum of 20% of the total exposure time in all bands.
Identified stars are marked by a star symbol. The colors are plotted with ±1σ error bars. There
is a large variation in both I − Ks and Js − Ks colors. Redshifted galaxies are well separated
from the stellar locus in color-color space.
In Figure 6 we show a direct test of photometric redshifts of the 39 objects in the
HDF-S with available spectroscopy and good photometry in all bands. The current
set of spectroscopic redshifts in the HDF-S will appear in Rudnick et al. (2002a).
For the small sample that we can directly compare, we find excellent agreement
with no failures and with a mean ∆z/(1+zspec ) ≈ 0.08 with ∆z = |zspec −zphot |. It
is encouraging to see that the modified 68% error bars that were derived from the
Monte Carlo simulations are consistent with the measured disagreement between
zphot and the zspec in the HDF-S. However, large asymmetric uncertainties remain
for some objects, clearly showing the presence of a second photometric redshift
solution of comparable likelihood at a vastly different redshift, revealing limits on
the applicability of the photometric redshift technique.
6.3
Stars
In a pencil beam survey at high Galactic latitude such as the HDF-S, a limited
number of foreground stars are expected. We identify stars as those objects which
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
38
multicolor catalog, and photometric redshifts
have a better raw χ2 for a single stellar template fit than the χ2 for the galaxy template combination. The stellar templates are the NEXTGEN model atmospheres
from Hauschildt et al. (1999) for main sequence stars with temperatures of 3000
to 10000K, assuming local thermodynamical equilibrium (LTE). Models of cooler
and hotter stars cannot be included because non-LTE effects are important. We
checked the resulting list of stars using the FWHM in the original B450 -band image
and the Js − Ks color, excluding two objects (catalog IDs 207 and 296) that were
obviously extended in B450 , and we find a total of 57 stars. As shown in Figure 7,
most galaxies are clearly separated from the stellar locus in I814 −Ks versus Js −Ks
color-color space. Other cooler stars might still resemble SEDs of redshifted compact galaxies but the latter are generally redder in the infrared Js − Ks than most
known M or methane dwarfs. Known cool L-dwarfs fall along a redder extension
of the track traced by M-dwarfs in color-color space and have progressively redder
Js − Ks colors for later spectral types. However measurements by Kirkpatrick et
al. (2000) show the L-dwarf sequence abruptly stopping at (Js − Ks )J ≈ 2.1 (the
subscript noting Johnson magnitudes, see section 3.1 for the transformations to the
AB system) whereas even cooler T-dwarfs have much bluer (Js − Ks )J ≈ 0 colors
than expected from their temperatures due to strong molecular absorption. This
is important because if we would apply a (Js − Ks )J > 2.3 photometric criterion
to select z > 2 galaxies (as discussed in section 8.2), then we should ensure that
cool Galactic stars are not expected in such a sample. The published data on the
lowest-mass stars suggest that they are too blue in infrared colors to be selected
this way. Only heavily reddened stars with thick circumstellar dust shells, such
as extreme carbon stars or Mira variables, or extremely metal-free stars having a
hypothetical . 1500K blackbody spectrum could also have red (Js − Ks )J > 2.3
colors but it seems unlikely that the tiny field of the HDF-S would contain such
unusual sources.
7
Catalog Parameters
The Ks -selected catalog of sources is published electronically. We describe here a
subset of the photometry containing the most important parameters. The catalog
with full photometry and explanation can be obtained from the FIRES homepage
7
.
• ID. — A running identification number in catalog order as reported by
SExtractor. Sources added in the second detection pass have numbers higher
than 10000.
• x, y. — The pixel positions of the objects corresponding to the coordinate
system of the original (unblocked) WFPC2 version 2 images.
• RA, DEC. – The right ascension and declination in equinox J2000.0 coordinates of which only the minutes and seconds of right ascension, and negative
arcminutes and arcseconds of declination are given. To these must be added
7 http://www.strw.leidenuniv.nl/~fires
2.8 Analysis
•
•
•
•
•
•
•
•
•
8
8.1
39
22h (R.A.) and −60◦ (DEC).
fcol,i ± σi . — The sum of counts in the “color” aperture fcol,i in band
i = {U300 , B450 , V606 , I814 , Js , H, Ks } and its simulated uncertainty σi , as
described in § 5.2. The fluxes are given in units of 10−31 ergs s−1 Hz−1
cm−2 .
Ktot ± σ(Ktot ). — Estimate of the total Ks -band flux and its uncertainty.
The sum of counts in the “total” aperture is corrected for missing flux assuming a PSF profile outside the aperture, as described in § 5.2.
ap col. — An integer encoding the aperture type that was used to measure
fcol,i . This is either a (1) 0.00 7 diameter circular aperture, (2) 2.00 0 diameter
circular aperture, (3) isophotal aperture determined by the detection-image
isophote,
or a (4) circular aperture with a reduced isophotal diameter D =
p
(Aiso /π)/1.4.
ap tot. — An integer encoding the aperture type that was used to measure
Ktot . This is either a (1) automatic Kron-like aperture, or a (2) circular
aperture within a reduced isophotal p
diameter.
rcol , rtot . — Circularized radii r = A/π, corresponding to the area A of
the specified “color” or “total” aperture.
Aiso , Aauto . — Area of the detection isophote Aiso and area of the autoscaling elliptical aperture Aauto = π∗a∗b with semi-major axis a and semi-minor
axis b.
F W HMK , F W HMI . — Full width at half maximum of a source in the Ks
detection image F W HMK , and that of the brightest I814 -band source that
lies in its detection isophote F W HMI . We obtained the latter by running
SExtractor separately on the original I814 -image and cross correlating the
I814 -selected catalog with the Ks -limited catalog.
wi . — The weight wi represents, for each band i, the fraction of the total
exposure time at the location of a source.
f lags.— Three binary flags are given. The bias flag indicates either that the
AUTO aperture measument is affected by nearby sources, or marks apertures
containing more than 10% bad pixels. The blended flag indicates overlapping
sources, while the star flag shows that the source SED is best fit with a stellar
template (see section 6.3).
Analysis
Completeness and Number Counts
The completeness curves for point-sources in the Js and Ks -band as a function of
input magnitude are shown in Figure 8. Our 90% and 50% completeness levels on
the AB magnitude system are 25.65 and 26.25, respectively, in Ks , and 26.30 and
26.90 in Js .
We derived the limits from simulations where we extracted a bright nonsaturated star from the survey image and add it back 30000 times at random
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
40
multicolor catalog, and photometric redshifts
Figure 8 — Completeness curves (on the AB system) for the detectability of point sources in
Js (triangles) and Ks (points), based on simulations where we calculated the recovered fraction
of stars that were dimmed to magnitudes between 22 and 28 and embedded in the survey images. The detection threshold of the source extraction software was set to 3.5σ of the filtered
background rms. The dotted lines indicate the 50% and 90% completeness levels.
locations, applying a random flux scaling drawn from a rising count slope (or an
increasing surface density of galaxies with magnitude) to bring it to magnitudes
between 22 ≤ Ks,AB ≤ 28. We added the dimmed stars back in series of 30 realizations so that they do not overlap each other. The rising count slope needs to
be considered because the slope influences the number of recovered galaxies per
apparent magnitude, as described below. The input count slope is based on the
observed surface densities in the faint magnitude range where the signal-to-noise
is 60 . SN R . 10 (or 23 . Ks,tot . 25) and where incompleteness does not yet
play a role. We used only the deepest central 4.5 arcmin2 of the Js and Ks images
(w > 0.95) with near uniform image quality and exclude four small regions around
the brightest stars. In the simulation images we extract sources following the
same procedures as described in §5.1, but applying a reduced (≈ 3.5σ) detection
threshold. We measure the recovered fraction of input sources against apparent
magnitude, and from this we estimate the detection efficiency of point-like sources
which we use to correct the observed number counts. We executed this procedure
in the Js and Ks -band.
The resulting completeness curves assume that the true profile of the source
is point-like and therefore they should be considered upper limits. An extended
source would have brighter completeness limits depending on the true source size,
2.8 Analysis
41
Figure 9 — Differential Ks -band counts (on the AB system) of galaxies in the HDF-S. The
counts are based on auto-scaling apertures (Kron 1980) for isolated sources and adapted isophotal
apertures for blended sources, both corrected to total magnitudes using stellar growth curve
measurements. Raw counts (open circles) and counts corrected for incompleteness and false
positive detections using point source simulations (filled circles) are shown. The small corrections
at magnitudes & 23 reflect missed sources due to confusion. Effective corrections at the faintest
magnitudes Ks ∼ 25 − 25.5 are very small because the loss of sources on negative noise regions
(incompletenesss) is compensated by the number of sources pushed above the detection limit
by positive noise fluctuations. Only the faintest 0.5 mag bin centered on K s = 26.0, bordering
the 3.5σ detection limit (Ks ≈ 26.3), is significantly corrected because of a contribution of false
positive detections.
its flux profile, and the filter that is used in the detection process. However, the
detailed treatment of detection efficiency as a function of source morphology and
detection criteria is beyond the scope of this paper.
When using the completeness simulations to correct the number counts, we
choose a simple approach and apply a single correction down to ≈ 50% completeness, based on the ratio of the simulated counts per input magnitude bin
to the total recovered counts per observed total magnitude bin. More sophisticated modeling is possible but requires detailed knowledge of the intrinsic size
and shape distribution of faint NIR galaxies. The simple approach corrects for
all effects resulting from detection criteria, photometric scheme, incompleteness,
and noise peaks. We find it works well if the total magnitude of sources is measured correctly, with little systematic difference between the input and recovered
magnitudes, which is the case for our photometric scheme (see section 5.2). It is
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
42
multicolor catalog, and photometric redshifts
Figure 10 — FIRES Ks −band galaxy counts (on the conventional Johnson system) compared
to published counts in deep K-band fields. The corrected counts (filled circles) are shown for
FIRES data. The Maihara et al. (2001) counts have been plotted to their S/N∼ 3 limit. The
slope at magnitudes KJ > 21 is flatter than reported in other surveys although straightforward
comparisons are difficult, due to model-dependent correction factors of ∼ 2 − 3 applied to the
faintest data points in these surveys. The nature of the scatter in count slopes is unclear but
field-to-field variations as well as different photometry and corrections procedures likely play a
role. The FIRES counts need little correction for completeness effects or false positive detections,
except for the Ks,J = 24.25 bin.
worth noticing that at Ks,AB > 25 we actually recovered slightly more counts in
the observed magnitude bins than we put in. This is caused by the fact that, in
the case of a rising count slope, there are more faint galaxies boosted by positive
noise peaks than bright galaxies lost on negative noise peaks. This effect is strong
at low signal-to-noise fluxes and results in a slight excess of recovered counts. This
is the main reason that we required little correction up to the detection threshold,
except for the faintest 0.5 mag bin centered on Ks = 26.0, which contained false
positive detections due to noise. After removing stars (see section 6.3), we plot in
Figure 9 the raw and corrected source counts against total magnitude.
Figure 10 presents a compilation of other deep K-band number counts from
a number of published studies. The FIRES counts follow a dlog(N)/dm relation
with a logarithmic slope α ≈ 0.25 at 20 . Ks,J . 22 (Johnson magnitudes) and
decline at fainter magnitudes to α ≈ 0.15 at 22.0 . Ks,J . 24. This flattening
of the slope has not been seen in other deep NIR surveys, where we emphasize
2.8 Analysis
43
Figure 11 — I814 − Ks versus Ks color-magnitude relation (on the AB system) for Ks -selected
objects in the HDF-S. Only sources with a minimum of 20% of the total exposure time in all
bands are included and identified stars are marked by a star symbol. Colors are plotted with ±1σ
error bars, and I814 measurements with S/N < 2 (triangles) are plotted at their 2σ confidence
interval, indicating lower limits for the colors. There are more red sources with I 814 − Ks > 2.6
at K ∼ 23 than at at K ∼ 24 where the I814 is still sufficiently deep to select them. The
transformation of the I814 − Ks color from the AB system to the Johnson magnitude system is
(I814 − Ks )J = (I814 − Ks )AB + 1.43.
that the FIRES HDFS field is the largest and the deepest amongst these surveys,
and that only the counts in the last FIRES bin at Ks,J = 24.25 were substantially
corrected. It is remarkable that the SUBARU Deep Field count slope α ≈ 0.23 of
Maihara et al. (2001) looks smooth compared to the HDF-S although their survey
area and the raw count statistics are slightly smaller.
Other authors (Djorgovski et al. 1995; Moustakas et al. 1997; Bershady, Lowenthal, & Koo 1998) find logarithmic counts slopes in K ranging from 0.23 to 0.36
over 20 . KJ . 23 − 24, however the counts in the faintest bins in these surveys
were boosted by factors of ∼ 2 − 3, based on completeness simulations. The origin
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
44
multicolor catalog, and photometric redshifts
Figure 12 — Same as Figure 11 for the Js − Ks color. Striking is the number the galaxies with
very red NIR colors Js − Ks & 1.34 (on the AB system) or Js − Ks & 2.3 (Johnson). These
systems have photometric redshifts z > 2 and are extremely faint in the observer’s optical; as such
they would not be selected with the U-dropout technique. Identified stars are well separated from
redshifted galaxies and almost all have Js − Ks ≤ 0 colors. The transformation of the Js − Ks
color from the AB system to the Johnson magnitude system is (Js − Ks )J = (Js − Ks )AB + 0.96.
of the faint-end discrepancies of the K counts is unclear. Cosmic variance can play
a role, because the survey areas never exceed a few armin2 , but also differences in
the used filters (Ks , K0, K) and differences in the techniques and assumptions used
to estimate the total magnitude (see 5) or to correct the counts for incompleteness
may be important. Further analysis is needed to ascertain whether size-dependent
biases in the completeness correction play a role in the faint-end count slope.
8.2
Color-Magnitude Distributions
Figures 11 – 14 show color-magnitude diagrams of Ks -selected galaxies in the
HDF-S. The I814 − Ks versus Ks color-magnitude diagram in Figure 11 shows a
2.8 Analysis
45
Figure 13 — Same as Figure 11 for the H − Ks color. One of the galaxies is extremely red
with H − Ks ≈ 2.2 and is barely visible in Js and H. The transformation of the H − Ks color
from the AB system to the Johnson magnitude system is (H − Ks )J = (H − Ks )AB + 0.48.
large number of extremely red objects (EROs) with I814 − Ks & 2.6 (on the AB
system) or (I814 − Ks )J & 4 (Johnson). There appears to be an excess of EROs
at total magnitudes Ks,AB ∼ 23 compared to magnitudes Ks,AB ∼ 24. This is
not caused by the insufficient signal-to-noise ratio in the I814 measurements. In
a similar diagram for the Js − Ks color shown in Figure 12, there is a striking
presence at the same Ks magnitudes of sources with very red (Js − Ks )AB & 1.34
or (Js − Ks )J & 2.3 colors. Such sources were also found by Saracco et al. (2001),
using shallow NIR data, who suggested they might be dusty starbursts or ellipticals
at z > 2. Interestingly, any evolved galaxy with a prominent Balmer/4000 Å
discontinuity in their spectrum, like most present-day Hubble Type galaxies, would
have such very red observed NIR colors if placed at redshifts z > 2. While the
(Js − Ks )J & 2.3 sources we find are generally morphologically compact, with
exceptions, we do not expect the sources with good photometry to be faint cool
L-dwarf stars because known colors of such stars are (Js − Ks )J . 2.1 (see section
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
46
multicolor catalog, and photometric redshifts
Figure 14 — V606 − H versus H color-magnitude diagram (on the AB system) for galaxies
in the HDF-S Ks -selected catalog with 1.95 < zphot < 3.5. Filled symbols indicate galaxies
with spectroscopy. The number of candidates for red, evolved galaxies is much higher than
in the HDF-N for a similar survey area, as shown in a identical plot in Fig. 1 of Papovich,
Dickinson, & Ferguson (2001): we find 7 galaxies redder than V606,AB − HAB & 3 and brighter
than HAB . 25.5, compared to only one in the HDF-N. Galaxies with S/N < 2 for the V606
measurement (triangles) are plotted at the 2σ confidence limit in V606 , indicating a lower limit
on the V606 − H color. The subsample of galaxies having red (Js − Ks )J > 2.3 colors (open
squares) is also shown. The transformation of the V606 − Hs color from the AB system to the
Johnson magnitude system is (V606 − H)J = (V606 − H)AB + 1.26.
6.3). The photometric redshifts of all red NIR galaxies are zphot & 2, but they
would be missed by ultraviolet-optical color selection techniques such as the Udropout method, because most of them are barely detectable even in the deepest
optical images. One bright NIR galaxy is completely undetected in the original
WFPC2 images. The (Js − Ks )J & 2.3 sources are studied in more detail by Franx
et al. (2002) and the relative contributions of these galaxies and U-dropouts to the
rest-frame optical luminosity density will be presented in Rudnick et al. (2002b).
If we select sources with 1.95 < zphot < 3.5, we find clear differences in the
V606 − H versus H color-magnitude diagram between our NIR-selected galaxies in
2.8 Analysis
47
Figure 15 — The surface densities of galaxies selected by color in the 5σ catalog of sources
of HDF-S. Presented are galaxies with (Js − Ks )J > 2.3 (in Johnson magnitudes) (diamonds),
extremely red objects with (I814 −Ks )J > 4 (squares), and all Ks -selected galaxies (filled; points)
as a function total Ks -band AB magnitude. Only sources with a minimum exposure time of 40%,
40% and 90% of the total in I814 , Js and Ks are plotted, so that the selection in Ks is uniform
over the area, and the I814 and Js observations are sufficiently deep to prevent a bias against
objects with very red I814 − Ks and Js − Ks colors. No corrections to the counts have been
applied. The errorbars are poissonian and might underestimate the true uncertainty which would
also contain contributions from large scale structure.
the HDF-S and those of the HDF-N (compare Figure 14 to Figure 1 of Papovich,
Dickinson, & Ferguson 2001). Over a similar survey area and to similar limiting
depths, we find 7 galaxies redder than (V606 − H)AB & 3 and brighter than total
magnitude HAB . 25.5, compared to only one in the HDF-N. While the surface
density of such galaxies is not well known, it is clear that the HDF-N contains
far fewer of them than the HDF-S. It remains to be seen if this is just field-tofield variation, or that one of the two fields is atypical. The results of the second
much larger FIRES field centered on MS1054-03 (Förster Schreiber et al. 2002)
should provide more insight into this issue. We note that all 7 (V606 − H)AB & 3
galaxies in the HDF-S are also amongst the brightest 16 (Js − Ks )J > 2.3 sources.
Figure 15 shows the surface densities of EROs and galaxies with (Js − Ks )J > 2.3
colors as function of Ks -band total magnitude. The surface density of EROs peaks
around Ks,AB ≈ 23 and then drops or flattens at fainter magnitudes, contrary to
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
48
multicolor catalog, and photometric redshifts
the number of (Js − Ks )J > 2.3 galaxies which keeps rising to the faintest Ks total
magnitudes in our catalog.
9
Summary and Conclusions
We have presented the results of the FIRES deep NIR imaging of the WFPC2field of the HDF-S obtained with ISAAC at the VLT: the deepest ground-based
NIR data available, and the deepest Ks -band of any field. We constructed a Ks selected multicolor catalog of galaxies, consisting of 833 objects with K s,AB . 26
and photometry in seven-bands from 0.3 to 2.2µ for 624 of them. These data are
available electronically together with photometric redshifts for 567 galaxies. Our
unique combination of deep optical space-based data from the HST together with
deep ground-based NIR data from the VLT allows us to sample light redder than
the rest-frame V-band in galaxies with z . 3 and to select galaxies from their
rest-frame optical properties, obtaining a more complete census of the stellar mass
in the high-redshift universe. We summarize our main findings below:
• The Ks -band galaxy counts in HDF-S turn over at the faintest magnitudes
and flatten from α ≈ 0.25 at total AB magnitudes 22 . Ks . 24 to α ≈ 0.15
at 24 . Ks . 26; this is flatter than counts in previously published deep
NIR surveys, where the FIRES HDF-S field is largest and deepest amongst
these surveys. The nature of the scatter in the faint-end counts is yet unclear
but field-to-field varations as well as different analysis techniques likely play
a role.
• The HDF-S contains 7 sources redder than (V606 − H)AB & 3 and brighter
than total magnitude HAB . 25.5 at photometric redshifts 1.95 < zphot <
3.5, while such galaxies were virtually absent in the HDF-N. They are much
redder than regular U-dropout galaxies in the same field and are candidates
for relatively massive, evolved systems at high redshift. The difference with
the HDF-N might just reflect field-to-field variance, calling for more observations to similar limits with full optical-to-infrared coverage. Results from
the second and larger FIRES field centered on MS1054-03 (Förster Schreiber
et al. 2002) should provide more insight into this issue.
• We find substantial numbers of red galaxies with (Js − Ks )J > 2.3 that
have photometric redshifts zphot > 2. These galaxies would be missed by
ultraviolet-optical color selection techniques such as the U-dropout method
because most of them are barely detectable even in the deepest optical images. The surface densities of these sources in our field keeps rising down to
the detection limit in Ks , in contrast to the number counts of EROs which
peak at Ks,AB ∼ 23 and then drop or flatten at fainter magnitudes.
The results of the HDF-S presented in this paper demonstrate the necessity of
extending optical observations to near-IR wavelengths for a more complete census
of the early universe. Our deep Ks -band data prove invaluable for they probe well
into the rest-frame optical at 2 < z < 4, where long-lived stars may dominate the
2.9 Summary and Conclusions
49
light of galaxies. We are pursuing follow-up programs to obtain more spectroscopic
redshifts needed to confirm the above results. Updates on the FIRES programme
and access to the reduced images and catalogues can be found at our website
http://www.strw.leidenuniv.nl/~fires.
Acknowledgments
The data here presented have been obtained as part of an ESO Service Mode
program. We would very much like to thank the ESO staff for their kind assistance and enormous efforts in taking these data and making them available to us.
This project would not be possible without their dedicated work. This work was
supported by a grant from the Netherlands Organization for Scientific Research.
We would like to thank the Lorentz Center of Leiden University for its hospitality
during several workshops. The comments of the referee helped to improve the
paper.
References
Adelberger, K. L. & Steidel, C. C. 2000, ApJ, 544, 218
P. Amico et al. eds, 2001, ISAAC-SW Data Reduction Guide (version 1.5)
Barger, A. J., Cowie, L. L., Bautz, M. W., Brandt, W. N., Garmire, G. P., Hornschemeier,
A. E., Ivison, R. J., & Owen, F. N. 2001, AJ, 122, 2177
Bershady, M. A., Lowenthal, J. D., & Koo, D. C. 1998, ApJ, 505, 50
Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393
Bessell, M. S. & Brett, J. M. 1988, PASP, 100, 1134 (JHKLMN photometric system)
Bruzual A., G., & Charlot, S. 2001, in preparation
Casertano, S. et al. 2000, AJ, 120, 2747
Coleman, G. D., Wu, C.-C., & Weedman, D. W. 1980, ApJS, 43, 393 (CWW80)
Cowie, L. L. et al. 2001, ApJ, 551, L9
da Costa, L et al. 1998 (astro-ph/9812105)
Dickinson, M., 2001a, in preparation
Djorgovski, S. et al. 1995, ApJ, 438, L13
Fernández-Soto, A., Lanzetta, K. M., & Yahil, A. 1999, ApJ, 513, 34 (FLY99)
Forster Schreiber, N.M. et al. 2002, in preparation
Franx, M., Illingworth, G. D., Heckman, T. M., 1989, A.J., 98 538-576
Franx, M. et al. 2000, The Messenger, 99, 20
Franx, M. et al. 2002, in preparation
Giavalisco, M. & Dickinson, M. 2001, ApJ, 550, 177
Hauschildt, P. H., Allard, F., Ferguson, J., Baron, E., & Alexander, D. R. 1999, ApJ,
525, 871
Johnson, H. L. 1966, ARA&A, 4, 193
Kinney, A. L., Calzetti, D., Bohlin, R. C., McQuade, K., Storchi-Bergmann, T., &
Schmitt, H. R. 1996, ApJ, 467, 38
Kirkpatrick, J. D. et al. 2000, AJ, 120, 447
Kron, R. G. 1980, ApJS, 43, 305
Madau, P. 1995, ApJ, 441, 18
Maihara, T. et al. 2001, PASJ, 53, 25
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
50
multicolor catalog, and photometric redshifts
Moorwood, A. F. 1997, Proc. SPIE, 2871, 1146
Moustakas, L. A., Davis, M., Graham, J. R., Silk, J., Peterson, B. A., & Yoshii, Y. 1997,
ApJ, 475, 445
Oke, J. B. 1971, ApJ, 170, 193
Papovich, C., Dickinson, M., & Ferguson, H. C. 2001, ApJ, 559, 620 (P01)
Persson et al. 1998 AJ, 116, 2475
Pettini, M., Kellogg, M., Steidel, C. C., Dickinson, M., Adelberger, K. L., & Giavalisco,
M. 1998, ApJ, 508, 539
Pettini, M., Shapley, A. E., Steidel, C. C., Cuby, J., Dickinson, M., Moorwood, A. F. M.,
Adelberger, K. L., & Giavalisco, M. 2001, ApJ, 554, 981
Pozzetti, L., Bruzual A., G., & Zamorani, G. 1996, MNRAS, 281, 953
Rudnick, G. et al. 2001, AJ, 122, 2205 (R01)
Rudnick, G. et al. 2002, in preparation (R02a)
Rudnick, G. et al. 2002, in preparation (R02b)
Saracco, P., Giallongo, E., Cristiani, S., D’Odorico, S., Fontana, A., Iovino, A., Poli, F.,
& Vanzella, E. 2001, A&A, 375, 1
Shapley, A. E., Steidel, C. C., Adelberger, K. L., Dickinson, M., Giavalisco, M., & Pettini,
M. 2001, ApJ, 562, 95
Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, ApJ, 500, 525
Steidel, C. C., Giavalisco, M., Dickinson, M., & Adelberger, K. L. 1996, AJ, 112, 352
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K. L. 1996, ApJ,
462, L17
Smail, I., Ivison, R. J., Owen, F. N., Blain, A. W., & Kneib, J.-P. 2000, ApJ, 528, 612
Voit, M., ed. 1997, HST Data Handbook, Vol. 1 (version 3.1; Baltimore: STScI)
Williams, R. E. et al. 1996, AJ, 112, 1335
Williams, R. E. et al. 2000, AJ, 120, 2735
2.9 Summary and Conclusions
Summary of the HDF-S Observations
Filter
Number of Integration Time FWHM1
Exposures
(h)
(arcsec)
F 300W
102
36.8
0.00 16
F 450W
51
28.3
0.00 14
F 606W
49
27.0
0.00 13
F 814W
56
31.2
0.00 14
Js
1007
33.6
0.00 45
H
968
32.3
0.00 48
Ks
2136
35.6
0.00 46
Table 1 —
Camera
WFPC2
WFPC2
WFPC2
WFPC2
ISAAC
ISAAC
ISAAC
1
The full width at half-maximum of the best-fitting Gaussian.
Zero Points for the HDF-S
Zero Point
Data Set Johnson mag AB mag
U300
19.43
20.77
B450
22.01
21.93
V606
22.90
23.02
I814
21.66
22.09
Js
24.70
25.60
H
24.60
25.98
Ks
24.12
25.98
Table 2 —
51
2 Ultradeep NIR ISAAC observations of the HDF-South: observations, reduction,
52
multicolor catalog, and photometric redshifts
Background
rms background2
1.34e-05
1.88e-05
3.80e-05
2.79e-05
0.0069
0.0165
0.0163
Table 3 —
Data Set
U300
B450
V606
I814
Js
H
Ks
1
2
3
4
1
noise in the HDF-S images
a3
b3
1σ sky noise limit4
2.51 0.38
29.5
2.65 0.40
30.3
2.49 0.39
30.6
2.59 0.39
30.0
1.46 0.047
28.6
1.43 0.044
28.1
1.49 0.038
28.1
The images are all at the 0.00 119 pixel−1 scale. The WFPC2 are 3x3 block summed,
and all images are smoothed to match the image quality of the H-band.
Pixel-to-pixel rms variations (in instrumental counts per second) as measured
directly in empty parts of the registered convolved images which were used for photometry.
Best-fit parameters of Eq. 3 which gives the effective rms variation of the background
as a function of linear size of the aperture.
The 1σ sky noise limit in a 0.00 7 circular diameter aperture (≈ 0.4 arcsec2 ) in
AB magnitudes using Eq. 3.
CHAPTER
THREE
Large Disk-Like Galaxies at High Redshift
ABSTRACT
Using deep near-infrared imaging of the Hubble Deep Field South with
ISAAC on the Very Large Telescope we find 6 large disk-like galaxies at
redshifts z = 1.4 − 3.0. The galaxies, selected in Ks (2.2µm), are regular
and surprisingly large in the near-infrared (rest-frame optical), with face-on
effective radii re = 0.00 65 − 0.00 9 or 5.0 − 7.5 h−1
70 kpc in a ΛCDM cosmology,
comparable to the Milky Way. The surface brightness profiles are consistent
with an exponential law over 2 − 3 effective radii. The WFPC2 morphologies
in Hubble Space Telescope imaging (rest-frame UV) are irregular and show
complex aggregates of star-forming regions ∼ 200 (∼ 15 h−1
70 kpc) across,
symmetrically distributed around the Ks -band centers. The spectral energy
distributions show clear breaks in the rest-frame optical. The breaks are
strongest in the central regions of the galaxies, and can be identified as the
age-sensitive Balmer/4000 Å break. The most straightforward interpretation
is that these galaxies are large disk galaxies; deep NIR data are indispensable for this classification. The candidate disks constitute 50% of galaxies
with LV & 6 × 1010 h−2
70 L¯ at z = 1.4 − 3.0. This discovery was not expected
on the basis of previously studied samples. In particular, the Hubble Deep
Field North is deficient in large galaxies with the morphologies and profiles
we report here.
Ivo Labbé, Gregory Rudnick, Marijn Franx, Emanuele Daddi, Pieter G. van
Dokkum, Natascha M. Förster Schreiber, Konrad Kuijken, Alan Moorwood,
Hans-Walter Rix, Huub Röttgering, Ignacio Trujillo, Arjen van de Wel, Paul van
der Werf, & Lottie van Starkenburg,
Astrophysical Journal Letters, 591, L95
54
1
3 Large disk-like galaxies at high redshift
Introduction
D
isk galaxies are believed to undergo a relatively simple formation process
in which gas cools and contracts in dark matter halos to form rotationally
supported disks with exponential light profiles (Fall & Efstathiou 1980; Mo, Mao,
& White 1998). A critical test of any theory of galaxy formation is to reproduce
the observed properties and evolution of galaxy disks.
Previous optical spectroscopy and HST imaging have yielded a wealth of data
on disk galaxies at z . 1. (e.g., Vogt et al. 1996, 1997; Lilly et al. 1998; Barden et al.
2003), although contradictory claims have been made regarding the implications
for the size and luminosity evolution with redshift (see Lilly et al. 1998; Mao,
Mo, & White 1998; Barden et al. 2003), and the importance of surface brightness
selection effects (see Simard et al. 1999; Bouwens & Silk 2002).
It is still unknown what the space density and properties are of disk galaxies
at substantially higher redshift. Many galaxies at z ∼ 3 have been identified
using the efficient U-dropout technique (Steidel et al. 1996a,b). Most of these
objects are compact with radii ∼ 1 − 2 h−1
70 kpc, while some are large and irregular
(Giavalisco, Steidel, & Macchetto 1996; Lowenthal et al. 1997). However, the
U-drop selection requires high far-UV surface brightness due to active, spatially
compact, and unobscured star formation. As a result, large and UV-faint disk
galaxies may have been overlooked and, additionally, the morphologies of LBGs
could just reveal the unobscured star-forming regions rather than the more evolved
underlying population which forms the disk.
The most direct evidence for the existence of large disks at high redshift has
come from observations in the NIR, which provide access to the rest-frame optical.
Here the continuum light is more indicative of the distribution of stellar mass
than in the UV and nebular lines are accessible for kinematic measurements. van
Dokkum & Stanford (2001) discuss a K-selected galaxy at z = 1.34 with a rotation
velocity of ∼ 290 km s−1 . Erb et al. (2003) detect ∼ 150 km s−1 rotation at ∼ 6
h−1
70 kpc radii in the Hα emission line of galaxies at z ∼ 2.3 and Moorwood et al.
(2003) find & 100 km s−1 rotation at ∼ 6 h−1
70 kpc from the center of a galaxy at
z = 3.2, seen in the NIR spectrum of the [O III]λ5007 Å emission line.
The imaging data in these studies, however, are of limited depth and resolution, making it difficult to determine morphological properties. In this Letter, we
present an analysis of the rest-frame ultraviolet-to-optical morphologies and spectral energy distributions (SEDs) of 6 large candidate disk galaxies at z ∼ 1.4 − 3
using the deepest groundbased NIR dataset currently available (Labbé et al. 2003).
Throughout, we adopt a flat Λ-dominated cosmology (ΩM = 0.3, Λ = 0.7, H0 =
70 h70 kms−1 Mpc−1 ). All magnitudes are expressed in the Johnson photometric
system.
3.2 Observations
2
55
Observations
We obtained 102 hours of NIR Js , H, and Ks imaging in the HDF-S (2.0 5×2.0 5) under excellent seeing (FWHM≈0.00 46), using ISAAC (Moorwood 1997) on the VLT.
The observations were taken as part of the Faint InfraRed Extragalactic Survey
(FIRES; Franx et al. 2000). We combined our data with existing deep optical
HST/WFPC2 imaging (version 2; Casertano et al. 2000), in the U300 , B450 , V606
and I814 bands, and we assembled a Ks -selected catalog of sources with SExtractor
(Bertin & Arnouts 1996). Photometric redshifts and rest-frame luminosities were
derived by fitting a linear combination of empirical galaxy spectra and stellar population models to the observed flux points (Rudnick et al. 2001, 2003a). The reduced
images, photometric catalog, and redshifts are presented in Labbé et al. (2003) and
are all available on-line at the FIRES homepage1 . Furthermore, we obtained optical spectroscopy with FORS1 on the VLT for some of the sources (Rudnick et
al. 2003b). Additional redshifts were obtained from Vanzella et al. (2002). As discussed in Rudnick et al. (2001, 2003a), our photometric redshifts yield good agreement with the spectroscopic redshifts, with rms |zspec − zphot |/(1 + zspec ) ≈ 0.05
for zspec > 1.4.
Large disk galaxies in the HDF-South were identified by fitting exponential
profiles convolved with the Point Spread Function (PSF) to the Ks -band images.
Six objects at z > 1.4 have effective radii re > 3.6 h−1
70 kpc, three of which have
spectroscopic redshifts. The mean redshift of the sample is 2.4. We will focus on
these large galaxies in the remainder of the Letter. The structural properties of
the full Ks -selected sample will be discussed in Trujillo et al. (2003).
3
Rest-frame Optical versus UV Morphology
The large galaxies are shown in Figure 1. They have a regular morphology in the
ISAAC Ks -band (2.2µm), which probes rest-frame optical wavelengths between
5400 and 9000 Å. In contrast, the WFPC2 V606 and I814 -band morphologies, which
map the unobscured star-forming regions at rest-frame UV wavelengths between
1500 and 3300 Å, are irregular with several knots up to ∼ 200 (∼ 15 h−1
70 kpc) apart,
symmetrically distributed around the Ks -band centers. In a few cases the observed
optical light is spatially almost distinct from the NIR.
As a result of the structure in the WFPC2 imaging, 4 of the large objects
have been split up into two sources by Casertano et al (2003). Fig. 1 shows
the corresponding “segmentation” map by “SExtractor” which illustrates how the
pixels in each image are allocated to different sources. The galaxies were not split
up when the Ks band image was used to detect objects. However, the broader
PSF in the Ks -band image can play a role: if we smooth the I814 -band data to
the same resolution, we find that SExtractor only splits up 1 galaxy.
1 http://www.strw.leidenuniv.nl/˜fires
56
3 Large disk-like galaxies at high redshift
Figure 1 — Left panels: The WFPC2 U300 , averaged V606 + I814 (rest-frame UV), and
our averaged ISAAC H + Ks images (rest-frame optical), scaled proportional to Fλ with
arbitrary normalization per galaxy. The rest-frame UV morphologies are complex and
symmetric with respect to the center of the smoother optical distribution. Middle panels:
SExtractor’s I814 -band segmentation map, the smoothed I814 images, and the Ks -band
images after subtracting the scaled, smoothed I814 -band. From the segmentation map
it follows that detection in I814 would likely split up most of the sources. The central
residuals in Ks −Ismooth demonstrate that optical and NIR light are distributed differently
and that all galaxies have a “red” nucleus. Right panels: The radial profiles in the K s band. The abscissa and the ordinate are respectively the mean geometric radial distance
and the surface brightness along elliptical isophotes. The arrows mark 1σ confidence
intervals for measurements with signal-to-noise less than 1. Overplotted are the best-fit
exponential law (dashed), r 1/4 law (dotted), and point+exponential (dash-dot) for galaxy
494 and 611.
Hence, the question remains whether these 4 objects split up in I814 are superpositions or whether they are part of larger systems. We tested this by subtracting
the PSF-matched I814 -band images from the Ks -band images. The I814 -band images were scaled to the Ks -band images to minimize the residuals. The residuals
are shown in Fig. 1. In all cases, we find strong positive residuals close to the
centers of the objects as defined in the Ks -band, whereas residuals at any of the
57
3.4 Profile fits and sizes
Table 1 — Properties of high redshift disk galaxies in the HDF-S
Galaxya
302
267
257
657
611
494
b
Ks,tot
19.70
19.98
20.25
20.68
20.53
21.14
z
1.439i
1.82
2.027i
2.793i
2.94
3.00
c
MB,rest
-22.70
-22.88
-23.08
-23.56
-23.59
-23.31
a
µd0,B,rest
19.70
19.92
19.53
19.33
18.51
18.84
Catalog identification (Labbé et al. 2003)
Ks -band total magnitudes
c
Rest-frame absolute B-band magnitudes
d
Face-on rest-frame B-band surface brightnesses
e
Face-on best-fit effective radii (arcsec)
b
e
re,K
0.89
0.75
0.74
0.76
0.65j
0.75j
f
r1/2,K
0.70
0.74
0.74
0.70
0.52
0.56
g
r1/2,I
0.86
0.88
0.84
0.74
0.97
0.86
²h
0.46
0.37
0.36
0.18
0.27
0.47
f
Ks half-light radii (arcsec)
I814 half-ligt radii, matched to Ks
h
Ellipticity
i
Spectroscopic redshifts
j
Two-component models
(point + exponential)
g
I814 -band peaks might be expected in case of a chance superposition. Furthermore, we performed photometric redshift analyses for subsections of the images
and found no evidence for components at different redshifts.
4
Profile fits and sizes
Next, we fitted simple models convolved with the PSF (FWHM≈0.00 46) to the
two-dimensional surface brightness distributions in the Ks -band. The images are
well-described by a simple exponential law over 2 − 3 effective radii (galaxy 302,
267, 257 and 657) or by a point source plus exponential (galaxy 611 and 494),
where the point source presumably represents the light emitted by a compact
bulge contributing about 40% of the light. We also derived intensity profiles by
ellipse fitting. As can be seen in Figure 1, most galaxies are well described by
an exponential. The central surface brightnesses and effective radii, enclosing half
of the flux of the model profile, are corrected to √face-on and shown in Table 1.
The central surface brightness is multiplied with 1 − ², as an intermediate case
between optically thin and optically thick, and corrected for cosmological dimming.
The effective radii (semi-major axes) are surprisingly large, re = 0.00 65 − 0.00 9
(5.0 − 7.5 h−1
70 kpc in a ΛCDM cosmology), comparable to the Milky Way and
much larger than typical sizes of “normal” Ly-break galaxies (Giavalisco, Steidel,
& Macchetto 1996; Lowenthal et al. 1997). As might be expected from the previous
section, the I-band images have even larger effective radii. All galaxies have a “red”
nucleus, and the colors become bluer in the outer parts.
Overall, the optical-to-infrared morphologies and sizes are strikingly similar
to L∗ disk galaxies in the local universe, with red bulges, more diffuse bluer ex-
58
3 Large disk-like galaxies at high redshift
Figure 2 — The spectral energy distribution of source 611 in a 0.00 7 circular diameter
aperture (left) and in a concentric 100 − 200 diameter ring (right), normalized to the Ks band flux. Overplotted are independent model fits from Rudnick et al. (2003a).
ponential disks and scattered, UV-bright star forming regions. Some even show
evidence of well-developed grand-design spiral structure. However, the mean central surface brightness of the disks is 1 − 2 mag higher than that of nearby disk
galaxies and the mean rest-frame color (U − V ) ≈ 0 is ∼ 1 mag bluer (c.f. Lilly et
al. 1998). Passive evolution can lead to disks with normal surface brightnesses at
low redshift. Alternatively, the disks are disrupted later by interactions or evolve
into S0’s, which have higher surface brightnesses (e.g., Burstein 1979).
5
Spectral Energy Distribution
The overall SEDs of the galaxies show a large variety. Four of the galaxies (257,
267, 494 and 657) satisfy conventional U-dropout criteria (Madau et al. 1996;
Giavalisco & Dickinson 2001). One galaxy is at too low redshift (302) to be
classified as a U-dropout, and one other galaxy is too faint in the rest-frame UV
(611). It has Js − Ks > 2.3, and is part of the population of evolved galaxies
identified by Franx et al (2003) and van Dokkum et al (2003).
The red colors of the central components can be due to either dust, higher age,
emission line contamination, or a combination of effects. The SEDs show stronger
Balmer/4000 Å breaks in the inner parts than in the outer parts (see Fig. 2 for an
example). We derived colors inside and outside of a 0.7 arcsec radius centered on
the Ks -band center. The mean differences are ∼ 0.5 mag in observed I814 − Ks
and ∼ 0.2 mag in rest-frame U − V . Higher resolution NICMOS data are required
to address the population differences in more detail.
3.6 Discussion
6
59
Discussion
We have found 6 large galaxies with characteristics similar to those of nearby disk
galaxies: exponential profiles with large scale lengths, more regular and centrally
concentrated morphologies in the restframe optical than in the rest-frame UV,
and, as a result, red nuclei. It is very tempting to classify these galaxies as disk
galaxies, given the similarities with low redshift disk galaxies. However, kinematic
studies are necessary to confirm that the material is in a rotating disk. Photometric
studies of larger samples are needed to constrain the thickness of the disks. We
note that simulations of Steinmetz & Navarro (2002) can show extended structures
during a merging or accretion event. The expect duration of this phase is short,
however, while the disk galaxies comprise a high fraction of the bright objects. It
is therefore unlikely that a significant fraction of the galaxies presented here are
undergoing such an event.
The density of these large disk galaxies is fairly low: over a survey area of 4.7
arcmin2 and to a magnitude limit of Ks,tot = 22 they make up 6 out of 52 galaxies
at 1.4 . z . 3.0. However, they do constitute 6 out of the 12 most rest-frame
luminous galaxies LV & 6 × 1010 h−2
70 L¯ in the same redshift range. The comoving
volume density is ∼ 3 × 10−4 h370 Mpc−3 at a mean redshift z ≈ 2.3. Obviously,
larger area surveys are needed to establish the true density. We note that three
of the galaxies only have photometric redshifts, one of which (a U-drop galaxy at
zp hot = 1.82) is in the poorly tested range 1.4 < z < 2.0. The volume density
of disk galaxies with re > 3.6 h−1
70 kpc in the local universe is much higher at
∼ 3 × 10−3 h370 Mpc−3 (de Jong 1996), although many nearby disks would not be
present in our high redshift sample because their surface brightness would be too
low.
We note that similar galaxies are absent in the very deep Near-IR imaging
data on the HDF-N ( Williams et al. 1996; Dickinson 2000) Although notable
differences between optical and NIR morphologies were reported for two of the
largest LBGs in the HDF-N, no large galaxies were reported to have red nuclei
and exponential profiles as in the HDF-S. The two fields are different in other
aspects as well. We found earlier that the HDF-N is deficient in red sources (e.g.,
Labbé et al. 2003 versus Papovich, Dickinson, & Ferguson 2001) and the disk
galaxies are more luminous in the H-band than most of the high-redshift galaxies
found in the HDF-N (Papovich, Dickinson, & Ferguson 2001). The larger number
of red galaxies in the HDF-S (e.g., Labbé et al. 2003, Franx et al. 2003) and their
strong clustering (Daddi et al. 2003) may indicate that the red galaxies and large
disks are both part of the same structures with high overdensities, and evolve into
the highest overdensities at low redshift, i.e. clusters. If this is the case, the large
disk galaxies may be the progenitor of large S0 galaxies in the nearby clusters,
which have very similar colors as elliptical galaxies (e.g., Bower, Lucey, & Ellis
1992).
Finally, we can compare the observed disk sizes to theoretical predictions. It
60
3 Large disk-like galaxies at high redshift
is often assumed (Fall & Efstathiou 1980; Mo, Mao, & White 1998) that the disk
scale length is determined by the spin parameter λ and the circular velocity of the
virialized dark matter halo (Fall & Efstathiou 1980; Mo, Mao, & White 1998). For
a ΛCDM cosmology, Mo, Mao, & White (1999) predict that the z ∼ 3 space density
−4 3
h70 Mpc−3 , whereas
of large (re & 3.6 h−1
70 kpc) bright U-dropouts is 1.1×10
−4 3
−3
we find ∼ 2 × 10 h70 Mpc for our 4 U-drops at a mean < z >∼ 2.4. This
difference of a factor of two is not very significant, given our low number statistics
and small survey volume. The combination of sizes and rotation velocities will
give much stronger constraints on these models; it may be possible to measure
the kinematics of some of these large galaxies using NIR spectrographs on large
telescopes.
Acknowledgments
We thank the staff at ESO for their hard work in taking these data and making
them available. This research was supported by grants from the Netherlands
Foundation for Research (NWO), the Leids Kerkhoven-Bosscha Fonds, and the
Lorentz Center. GR thanks Frank van den Bosch and Jarle Brinchmann for useful
discussions.
References
Barden et al., 2003, ApJ, submitted (astro-ph/0302392)
Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393
Bouwens, R. & Silk, J. 2002, ApJ, 568, 522
Bower, R. G., Lucey, J. R., & Ellis, R. S. 1992, MNRAS, 254, 601
Burstein, D. 1979, ApJ, 234, 435
Casertano, S. et al., AJ, 120, pp. 2747–2824, 2000
Daddi, E., et al., ApJ, in press
de Jong, R. S. 1996, A&A, 313, 45
Dickinson, M., Philos. Trans. R. Soc. London A 358, p. 2001, 2000
Erb et al., 2003, ApJ, in press
Fall, S. M. & Efstathiou, G. 1980, MNRAS, 193, 189
Förster Schreiber, N.M. et al. 2003, in preparation
Franx, M. et al., The Messenger 99, pp. 20–22, 2000
Franx, M. et al. 2003, ApJ, 587, L79
Giavalisco, M., Steidel, C. C., & Macchetto, F. D. 1996, ApJ, 470, 189
Giavalisco, M. & Dickinson, M. 2001, ApJ, 550, 177
Labbé, I. et al. 2003, AJ, 125, 1107
Lilly, S. et al. 1998, ApJ, 500, 75
Lowenthal, J. D. et al. 1997, ApJ, 481, 673
Madau, P., Ferguson, H. C., Dickinson, M. E., Giavalisco, M., Steidel, C. C., & Fruchter,
A., MNRAS283, pp. 1388–1404, 1996
Mao, S., Mo, H. J., & White, S. D. M. 1998, MNRAS, 297, L71
Mo, H. J., Mao, S., & White, S. D. M. 1998, MNRAS, 295, 319
Mo, H. J., Mao, S.,& White, S. D. M. 1999, MNRAS, 304, 175
3.6 Discussion
61
Moorwood, A. F. 1997, Proc. SPIE, 2871, 1146
Moorwood, A., van Der Werf, P., Cuby, J.-G., Oliva, E. 2003, Proceedings of the Workshop on The Mass of Galaxies at Low and High Redshift, eds. R. Bender and A.
Renzini, Springer-Verlag, p302.
Steinmetz, M. & Navarro, J. F. 2002, New Astronomy, 7, 155
Papovich, C., Dickinson, M., & Ferguson, H. C. 2001, ApJ, 559, 620
Rudnick, G. et al. 2001, AJ, 122, 2205
Rudnick, G. et al. 2003a, in preparation
Rudnick, G. et al. 2003b, in preparation
Schade, D., Lilly, S. J., Le Fevre, O., Hammer, F., & Crampton, D. 1996, ApJ, 464, 79
Simard, L. et al. 1999, ApJ, 519, 563
Steidel, C. C., Giavalisco, M., Dickinson, M., & Adelberger, K. L. 1996, AJ, 112, 352
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K. L. 1996, ApJ,
462, L17
Trujillo et al. 2003, in preparation
van Dokkum, P. G. & Stanford, S. A. 2001, ApJ, 562, L35
van Dokkum, P. G. et al. 2003, ApJ, 587, L83
Vanzella, E. et al. 2002, A&A, 396, 847
Vogt, N. P., Forbes, D. A., Phillips, A. C., Gronwall, C., Faber, S. M., Illingworth, G. D.,
& Koo, D. C. 1996, ApJ, 465, L15
Vogt, N. P. et al. 1997, ApJ, 479, L121
Williams, R. E. et al. 1996, AJ, 112, 1335
CHAPTER
FOUR
The Rest-Frame Optical Luminosity Density
Color, and Stellar Mass Density of
the Universe from z=0 to z=3
ABSTRACT
We present the evolution of the rest-frame optical luminosity density, jλrest ,
of the integrated rest-frame optical color, and of the stellar mass density,
ρ∗ , for a sample of Ks -band selected galaxies in the HDF-S. We derived
jλrest in the rest-frame U , B , and V -bands and found that jλrest increases
by a factor of 1.9 ± 0.4, 2.9 ± 0.6, and 4.9 ± 1.0 in the V , B , and U restframe bands respectively between a redshift of 0.1 and 3.2. We derived the
luminosity weighted mean cosmic (U − B)rest and (B − V )rest colors
as a function of redshift. The colors bluen almost monotonically with increasing redshift; at z = 0.1, the (U − B)rest and (B − V )rest colors
are 0.16 and 0.75 respectively, while at z = 2.8 they are -0.39 and 0.29
respectively. We derived the luminosity weighted mean M/L∗V using the
correlation between (U − V )rest and log10 M/L∗V which exists for a range
in smooth SFHs and moderate extinctions. We have shown that the mean of
individual M/L∗V estimates can overpredict the true value by ∼ 70% while
our method overpredicts the true values by only ∼ 35%. We find that the
universe at z ∼ 3 had ∼ 10 times lower stellar mass density than it does
today in galaxies with Lrest
> 1.4 × 1010 h−2
V
70 L¯ . 50% of the stellar mass
of the universe was formed by z ∼ 1 − 1.5. The rate of increase in ρ∗ with
decreasing redshift is similar to but above that for independent estimates
from the HDF-N, but is slightly less than that predicted by the integral of
the SFR(z) curve.
Gregory Rudnick, Hans-Walter Rix, Marijn Franx, Ivo Labbé, Michael Blanton,
Emanuele Daddi, Natascha M. Förster Schreiber, Alan Moorwood, Huub
Röttgering, Ignacio Trujillo, Arjen van de Wel, Paul van der Werf, Pieter G. van
Dokkum, & Lottie van Starkenburg,
Astrophysical Journal, 599, 847 (2003)
4 The rest-frame optical luminosity density, color, and stellar mass density of the
64
universe from z=0 to z=3
1
Introduction
A
primary goal of galaxy evolution studies is to elucidate how the stellar content
of the present universe was assembled over time. Enormous progress has been
made in this field over the past decade, driven by advances over three different
redshift ranges. Large scale redshift surveys with median redshifts of z ∼ 0.1 such
as the Sloan Digital Sky Survey (SDSS; York et al. 2000) and the 2dF Galaxy
Redshift Survey (2dFGRS; Colless et al. 2001), coupled with the near infrared
(NIR) photometry from the 2 Micron All Sky Survey (2MASS; Skrutskie et al.
1997), have recently been able to assemble the complete samples, with significant
co-moving volumes, necessary to establish crucial local reference points for the
local luminosity function (e.g. Folkes et al. 1999; Blanton et al. 2001; Norberg
et al. 2002; Blanton et al. 2003c) and the local stellar mass function of galaxies
(Cole et al. 2001; Bell et al. 2003a).
At z . 1, the pioneering study of galaxy evolution was the Canada France
Redshift Survey (CFRS; Lilly et al. 1996). The strength of this survey lay not
only in the large numbers of galaxies with confirmed spectroscopic redshifts, but
also in the I -band selection, which enabled galaxies at z . 1 to be selected in
the rest-frame optical, the same way in which galaxies are selected in the local
universe.
At high redshifts the field was revolutionized by the identification, and subsequent detailed follow-up, of a large population of star-forming galaxies at z > 2
(Steidel et al. 1996). These Lyman Break Galaxies (LBGs) are identified by the
signature of the redshifted break in the far UV continuum caused by intervening
and intrinsic neutral hydrogen absorption. There are over 1000 spectroscopically
confirmed LBGs at z > 2, together with the analogous U-dropout galaxies identified using Hubble Space Telescope (HST) filters. The individual properties of
LBGs have been studied in great detail. Estimates for their star formation rates
(SFRs), extinctions, ages, and stellar masses have been estimated by modeling the
broad band fluxes (Sawicki & Yee 1998; hereafter SY98; Papovich et al. 2001;
hereafter P01; Shapley et al. 2001). Independent measures of their kinematic
masses, metallicities, SFRs, and initial mass functions (IMFs) have been determined using rest-frame UV and optical spectroscopy (Pettini et al. 2000; Shapley
et al. 2001; Pettini et al. 2001, 2002; Erb et al. 2003).
Despite these advances, it has proven difficult to reconcile the ages, SFRs, and
stellar masses of individual galaxies at different redshifts within a single galaxy
formation scenario. Low redshift studies of the fundamental plane indicate that
the stars in elliptical galaxies must have been formed by z > 2 (e.g., van Dokkum et
al. 2001) and observations of evolved galaxies at 1 < z < 2 indicate that the present
population of elliptical galaxies was already in place at z & 2.5 (e.g., Benı́tez et
al. 1999; Cimatti et al. 2002; but see Zepf 1997). In contrast, studies of star
forming Lyman Break Galaxies spectroscopically confirmed to lie at z > 2 (LBGs;
Steidel et al. 1996, 1999) claim that LBGs are uniformly very young and a factor of
4.1 Introduction
65
10 less massive than present day L∗ galaxies (e.g., SY98; P01; Shapley et al. 2001).
An alternative method of tracking the build-up of the cosmic stellar mass is
to measure the total emissivity of all relatively unobscured stars in the universe,
thus effectively making a luminosity weighted mean of the galaxy population. This
can be partly accomplished by measuring the evolution in the global luminosity
density j(z) from galaxy redshift surveys. Early studies at intermediate redshift
have shown that the rest-frame UV and B -band j(z) are steeply increasing out
to z ∼ 1 (e.g., Lilly et al. 1996; Fried et al. 2001). Wolf et al. (2003) has recently
measured j(z) at 0 < z < 1.2 from the COMBO-17 survey using ∼25,000 galaxies
with redshifts accurate to ∼ 0.03 and a total area of 0.78 degrees. At rest-frame
2800Å these measurements confirm those of Lilly et al. (1996) but do not support
claims for a shallower increase with redshift which goes like (1 + z)1.5 as claimed
by Cowie, Songaila, & Barger (1999) and Wilson et al. (2002). On the other hand,
the B -band evolution from Wolf et al. (2003) is only a factor of ∼ 1.6 between
0 < z < 1, considerably shallower than the factor of ∼ 3.75 increase seen by Lilly et
al. (1996). At z > 2 measurements of the rest-frame UV j(z) have been made using
the optically selected LBG samples (e.g., Madau et al. 1996; Sawicki, Lin, & Yee
1997; Steidel et al. 1999; Poli et al. 2001) and NIR selected samples (Kashikawa et
al. 2003; Poli et al. 2003; Thompson 2003) and, with modest extinction corrections,
the most recent estimates generically yield rest-frame UV j(z) curves which, at
z > 2, are approximately flat out to z ∼ 6 (cf. Lanzetta et al. 2002). Dickinson et
al. (2003; hereafter D03) have used deep NIR data from NICMOS in the HDF-N
to measure the rest-frame B -band luminosity density out to z ∼ 3, finding that
it remained constant to within a factor of ∼ 3. By combining j(z) measurements
at different rest-frame wavelengths and redshifts, Madau, Pozzetti, & Dickinson
(1998) and Pei, Fall, & Hauser (1999) modeled the emission in all bands using an
assumed global SFH and used it to constrain the mean extinction, metallicity, and
IMF. Bolzonella, Pelló, & Maccagni (2002) measured NIR luminosity functions in
the HDF-N and HDF-S and find little evolution in the bright end of the galaxy
population and no decline in the rest-frame NIR luminosity density out to z ∼ 2.
In addition, Baldry et al. (2002) and Glazebrook et al. (2003a) have used the mean
optical cosmic spectrum at z ∼ 0 from the 2dFGRS and the SDSS respectively to
constrain the cosmic star formation history.
Despite the wealth of information obtained from studies of the integrated
galaxy population, there are major difficulties in using these many disparate measurements to re-construct the evolution in the stellar mass density. First, and
perhaps most important, the selection criteria for the low and high redshift surveys are usually vastly different. At z < 1 galaxies are selected by their rest-frame
optical light. At z > 2, however, the dearth of deep, wide-field NIR imaging has
forced galaxy selection by the rest-frame UV light. Observations in the rest-frame
UV are much more sensitive to the presence of young stars and extinction than
observations in the rest-frame optical. Second, state-of-the-art deep surveys have
only been performed in small fields and the effects of field-to-field variance at faint
magnitudes, and in the rest-frame optical, are not well understood.
4 The rest-frame optical luminosity density, color, and stellar mass density of the
66
universe from z=0 to z=3
In the face of field-to-field variance, the globally averaged rest-frame color
may be a more robust characterization of the galaxy population than either the
luminosity density or the mass density because it is, to the first order, insensitive
to the exact density normalization. At the same time, it encodes information
about the dust obscuration, metallicity, and SFH of the cosmic stellar population.
It therefore provides an important constraint on galaxy formation models which
may be reliably determined from relatively small fields.
To track consistently the globally averaged evolution of the galaxies which
dominate the stellar mass budget of the universe – as opposed to the UV luminosity
budget – over a large redshift range a different strategy than UV selection must be
adopted. It is not only desirable to measure j(z) in a constant rest-frame optical
bandpass, but it is also necessary that galaxies be selected by light redward of the
Balmer/4000Å break, where the light from older stars contributes significantly to
the SED. To accomplish this, we obtained ultra-deep NIR imaging of the WFPC2
field of the HDF-S (Casertano et al. 2000) with the Infrared Spectrograph And
Array Camera (ISAAC; Moorwood et al. 1997) at the Very Large Telescope (VLT)
as part of the Faint Infrared Extragalactic Survey (FIRES; Franx et al. 2000). The
FIRES data on the HDF-S, detailed in Labbé et al. (2003; hereafter L03), provide
us with the deepest ground-based Js and H data and the overall deepest Ks -band
data in any field allowing us to reach rest-frame optical luminosities in the V -band
at z ∼ 3. First results using a smaller set of the data were presented
of ∼ 0.6 Llocal
∗
in Rudnick et al. (2001; hereafter R01). The second FIRES field, centered on the
z = 0.83 cluster MS1054-03, has ∼ 1 magnitude less depth but ∼ 5 times greater
area (Förster Schreiber et al. 2003).
In the present work we will draw on photometric redshift estimates, zphot for
the Ks -band selected sample in the HDF-S (R01; L03), and on the observed SEDs,
to derived rest-frame optical luminosities Lrest
for a sample of galaxies selected by
λ
light redder than the rest-frame optical out to z ∼ 3. In § 2 we describe the
observations, data reduction, and the construction of a Ks -band selected catalog
with 0.3 − 2.2µm photometry, which selects galaxies at z < 4 by light redward
of the 4000Å break. In § 3 we describe our photometric redshift technique, how
for
we estimate the associated uncertainties in zphot , and how we measure Lrest
λ
for the individual galaxies to
our galaxies. In § 4 we use our measures of Lrest
λ
derive the mean cosmic luminosity density, jλrest and the cosmic color and then
use these to measure the stellar mass density ρ∗ as a function of cosmic time. We
discuss our results in § 5 and summarize in § 6. Throughout this paper we assume
ΩM = 0.3, ΩΛ = 0.7, and Ho = 70 h70 km s−1 Mpc−1 unless explicitly stated
otherwise.
2
Data
A complete description of the FIRES observations, reduction procedures, and the
construction of photometric catalogs is presented in detail in L03; we outline the
4.3 Measuring Photometric Redshifts and Rest-Frame Luminosities
67
important steps below.
Objects were detected in the Ks -band image with version 2.2.2 of the SExtractor software (Bertin & Arnouts 1996). For consistent photometry between the
space and ground-based data, all images were then convolved to 0.00 48, the seeing
in our worst NIR band. Photometry was then performed in the U300 , B450 , V606 ,
I814 , Js , H , and Ks -band images using specially tailored isophotal apertures
defined from the detection image. In addition, a measurement of the total flux in
tot
the Ks -band, Ks,AB
, was obtained using an aperture based on the SExtractor
1
AUTO aperture . Our effective area is 4.74 square arcminutes, including only
areas of the chip which were well exposed. All magnitudes are quoted in the Vega
system unless specifically noted otherwise. Our adopted conversions from Vega
system to the AB system are Js,vega = Js,AB - 0.90, Hvega = HAB - 1.38, and
Ks,vega = Ks,AB - 1.86 (Bessell & Brett 1988).
3
Measuring Photometric Redshifts and Rest-Frame Luminosities
3.1
Photometric Redshift Technique
We estimated zphot from the broad-band SED using the method described in R01,
which attempts to fit the observed SED with a linear combination of redshifted
galaxy templates. We made two modifications to the R01 method. First, we added
an additional template constructed from a 10 Myr old, single age, solar metallicity
population with a Salpeter (1955) initial mass function (IMF) based on empirical
stellar spectra from the 1999 version of the Bruzual A. & Charlot (1993) stellar
population synthesis code. Second, a 5% minimum flux error was adopted for
all bands to account for the night-to-night uncertainty in the derived zeropoints
and for template mismatch effects, although in reality both of these errors are
non-gaussian.
Using 39 galaxies with reliable FIRES photometry and spectroscopy available from Cristiani et al. (2000), Rigopoulou et al. (2000), Glazebrook (2003b) 2 ,
Vanzella et al. (2002), and Rudnick et al. (2003) we measured the redshift accuracy
of our technique to be h |zspec − zphot | / (1 + zspec ) i = 0.09 for z < 3 . There is
one galaxy at zspec =2.025 with zphot = 0.12 but with a very large internal zphot uncertainty. When this object is removed, h |zspec − zphot | / (1 + zspec ) i = 0.05 at
zspec > 1.3.
For a given galaxy, the photometric redshift probability distribution can be
highly non-Gaussian and contain multiple χ2 minima at vastly different redshifts.
An accurate estimate of the error in zphot must therefore not only contain the two1 The
reduced images,
photometric catalogs,
photometric redshift estimates,
and rest-frame luminosities are available online through the FIRES homepage at
http://www.strw.leidenuniv.nl/∼fires.
2 available at http://www.aao.gov.au/hdfs/
4 The rest-frame optical luminosity density, color, and stellar mass density of the
68
universe from z=0 to z=3
sided confidence interval in the local χ2 minimum, but also reflect the presence
of alternate redshift solutions. The difficulties of measuring the uncertainty in
zphot were discussed in R01 and will not be repeated in detail here. To improve
on R01, however, we have developed a Monte Carlo method which takes into
account, on a galaxy-by-galaxy basis, flux errors and template mismatch. These
uncertainty estimates are called δzphot . For a full discussion of this method see
Appendix Appendix A..
tot
Galaxies with Ks,AB
≥ 25 have such high photometric errors that the zphot estot
timates can be very uncertain. At Ks,AB
< 25, however, objects are detected at
better than the 10-sigma level and have well measured NIR SEDs, important for
locating redshifted optical breaks. For this reason, we limited our catalog to the
tot
< 25, lie on well exposed sections of the chip, and
329 objects that have Ks,AB
are not identified as stars (see §3.1.1).
3.1.1
Star Identification
To identify probable stars in our catalog we did not use the profiles measured from
the WFPC2-imaging because it is difficult to determine the size at faint levels. At
the same time, we verified that the stellar template fitting technique identified all
bright unsaturated stars in the image. Instead, we compared the observed SEDs
with those from 135 NextGen version 5.0 stellar atmosphere models described in
Hauschildt, Allard, & Baron (1999)3 . We used models with log(g) of 5.5 and 6,
effective temperatures ranging from 1600 K to 10,000 K, and metallicities of solar
and 1/10th solar. We identified an object as a stellar candidate if the raw χ2 of
the stellar fit was lower than that of the best-fit galaxy template combination.
Four of the stellar candidates from this technique (objects 155, 230, 296, and 323)
are obviously extended and were excluded from the list of stellar candidates. Two
bright stars (objects 39 and 51) were not not identified by this technique because
they are saturated in the HST images and were added to the list by hand. We
tot
< 25 and lie on well exposed
ended up with a list of 29 stars that had Ks,AB
sections of the chip. These were excluded from all further analysis.
3.2
Rest-Frame Luminosities
To measure the Lrest
of a galaxy one must combine its redshift with the observed
λ
SED to estimate the intrinsic SED. In practice, this requires some assumptions
about the intrinsic SED.
In R01 we derived rest-frame luminosities from the best-fit combination of
spectral templates at zphot , which assumes that the intrinsic SED is well modeled
by our template set. We know that for many galaxies the best-fit template matches
the position and strength of the spectral breaks and the general shape of the SED.
There are, however, galaxies in our sample which show clear residuals from the
3 available
at http://dilbert.physast.uga.edu/∼yeti/mdwarfs.html
4.3 Measuring Photometric Redshifts and Rest-Frame Luminosities
69
best fit template combination. Even for the qualitatively good fits, the model and
observed flux points can differ by ∼ 10%, corresponding to a ∼ 15% error in the
derived rest-frame color. As we will see in §4.2, such color errors can cause errors
of up to a factor of 1.5 in the V -band stellar mass-to-light ratio, M/L∗V .
Here we used a method of estimating Lrest
which does not depend directly on
λ
template fits to the data but, rather, interpolates directly between the observed
bands using the templates as a guide. We define our rest-frame photometric sysin
tem in Appendix Appendix B. and explain our method for estimating Lrest
λ
Appendix Appendix C..
We plot in Figure 1 the rest-frame luminosities vs. redshift and enclosed volume
tot
for the Ks,AB
< 25 galaxies in the FIRES sample. The different symbols represent
different δzphot values and since the derived luminosity is tightly coupled to the
redshift, we do not independently plot Lrest
errorbars. The tracks indicate the
λ
tot
for
different
SED
types
normalized
to
K
Lrest
s,AB = 25, while the intersection
λ
of the tracks in each panel indicates the redshift at which the rest-frame filter
at
passes through our Ks -band detection filter. There is a wide range in Lrest
λ
much
in
excess
of
the
all redshifts and there are galaxies at z > 2 with Lrest
λ
local L∗ values. Using the full FIRES dataset, we are much more sensitive than
tot
= 25 have Lrest
≈ 0.6∗Llocal
in R01; objects at z ≈ 3 with Ks,AB
V
∗,V , as defined
from the z=0.1 sample of Blanton et al. (2003c; hereafter B03). As seen in R01
there are many galaxies at z > 2, in all bands, with Lrest
≥Llocal
. R01 found 10
∗
λ
11 −2
rest
galaxies at 2 ≤ z ≤ 3.5 with LB > 10 h70 L¯ and inferred a brightening in the
luminosity function of ∼ 1 − 1.3 magnitudes. We confirm their result when using
the same local luminosity function (Blanton et al. 2001). Although this brightening
is biased upwards by photometric redshift errors, we find a similar brightening of
approximately ∼ 1 magnitude after correction for this effect. As also noticed in
R01, we found a deficit of luminous galaxies at 1.5 . z . 2 although this deficit
is not as pronounced at lower values of Lrest
. The photometric redshifts in the
λ
HDF-S, however, are not well tested in this regime. To help judge the reality of
this deficit we compared our photometric redshifts on an object-by-object basis to
those of the Rome group (Fontana et al. 2000)4 who derived zphot estimates for
galaxies in the HDF-S using much shallower NIR data. We find generally good
agreement in the zphot estimates, although there is a large scatter at 1.5 < z < 2.0.
Both sets of photometric redshifts show a deficit in the zphot distribution, although
the Rome group’s gap is less pronounced than ours and is at a slightly lower
redshift. In addition, we examined the photometric redshift distribution of the
NIR selected galaxies of D03 in the HDF-N, which have very deep NIR data.
These galaxies also showed a gap in the zphot distribution at z ∼ 1.6. Together
these results indicate that systematic effects in the zphot determinations may be
significant at 1.5 < z < 2.0. On the other hand, we also derived photometric
redshifts for a preliminary set of data in the MS1054-03 field of the FIRES survey,
whose filter set is similar, but which has a U instead of U300 filter. In this field,
4 available
at http://www.mporzio.astro.it/HIGHZ/HDF.html
4 The rest-frame optical luminosity density, color, and stellar mass density of the
70
universe from z=0 to z=3
no systematic depletion of 1.5 < z < 2 galaxies was found. It is therefore not
clear what role systematic effects play in comparison to field-to-field variations
in the true redshift distribution over this redshift range. Obtaining spectroscopic
redshifts at 1.5 < z < 2 is the only way to judge the accuracy of the zphot estimates
in this regime.
We have also split the points up according to whether or not they satisfied the
U-dropout criteria of Giavalisco & Dickinson (2001) which were designed to pick
unobscured star-forming galaxies at z & 2. As expected from the high efficiency
of the U-dropout technique, we find that only 15% of the 57 classified U-dropouts
have zphot < 2. As we will discuss in §4.1 we measured the luminosity density
for objects with Lrest
> 1.4 × 1010 h−2
70 L¯ . Above this threshold, there are 62
V
galaxies with 2 < z < 3.2, of which 26 are not classified as U-dropouts. These non
U-dropouts number among the most rest-frame optically luminous galaxies in our
sample. In fact, the most rest-frame optically luminous object at z < 3.2 (object
611) is a galaxy which fails the U-dropout criteria. 10 of these 26 objects, including
object 611, also have J − K > 2.3, a color threshold which has been shown by
Franx et al. (2003) and van Dokkum et al. (2003) to efficiently select galaxies at
z > 2. These galaxies are not only luminous but also have red rest-frame optical
colors, implying high M/L∗ values. Franx et al. (2003) showed that they likely
contribute significantly (∼ 43%) to the stellar mass budget at high redshifts.
3.2.1
Emission Lines
There will be emission line contamination of the rest-frame broad-band luminosities when rest-frame optical emission lines contribute significantly to the flux in
our observed filters. P01 estimated the effect of emission lines in the NICMOS
F160W filter and the Ks filter and found that redshifted, rest-frame optical emission lines, whose equivalent widths are at the maximum end of those observed for
starburst galaxies (rest-frame equivalent width ∼ 200Å ), can contribute up to 0.2
magnitudes in the NIR filters. In addition, models of emission lines from Charlot &
Longhetti (2001) show that emission lines will tend to drive the (U − B) rest color
to the blue more easily than the (B − V )rest color for a large range of models.
Using the U BV photometry and spectra of nearby galaxies from the Nearby Field
Galaxy Survey (NFGS; Jansen et al. 2000a; Jansen et al. 2000b) we computed
the actual correction to the (U − B)rest and (B − V )rest colors as a function
of (B − V )rest . For the bluest galaxies in (B − V )rest , emission lines bluen
the (U − B)rest colors by ∼ 0.05 and the (B − V )rest colors only by < 0.01.
Without knowing beforehand the strength of emission lines in any of our galaxies,
we corrected our rest-frame colors based on the results from Jansen et al. We
ignored the very small correction to the (B − V )rest colors and corrected the
(U − B)rest colors using the equation:
(U − B)corrected = (U − B) − 0.0658 × (B − V ) + 0.0656
(1)
4.3 Measuring Photometric Redshifts and Rest-Frame Luminosities
71
Figure 1 — The distribution of rest-frame V , B , and U -band luminosities as a function of
enclosed co-moving volume and zphot is shown in figures (a), (b), and (c) respectively for galaxies
tot
with Ks,AB
< 25. Galaxies which have spectroscopic redshifts are represented by solid points
and for these objects Lrest
is measured at zspec . Large symbols have δzphot /(1 + zphot ) < 0.16
λ
and small symbols have δzphot /(1 + zphot ) ≥ 0.16. Triangle points would be classified as Udropouts according to the selection of Giavalisco & Dickinson (2001). As is expected, most of
the galaxies selected as U-dropouts have zphot & 2. Note, however, the large numbers of restframe optically luminous galaxies at z > 2 which would not be selected as U-dropouts. The
large stars in each panel indicate the value of Llocal
from Blanton et al. (2003c). In the V -band
∗
we are sensitive to galaxies at 60% of Llocal
even at z ∼ 3 and there are galaxies at zphot ≥ 2
∗
rest for our seven template
with Lrest
≥ 1011 h−2
70 L¯ . The tracks represent the values of Lλ
λ
tot
= 25. The specific tracks correspond to the E
spectra normalized at each redshift to Ks,AB
(solid), Sbc (dot), Scd (short dash), Im (long dash), SB1 (dot–short dash), SB2 (dot–long dash),
and 10my (dot) templates. The horizontal dotted line in (a) indicates the luminosity threshold
Lthresh
above which we measure the rest-frame luminosity density jλrest and the vertical dotted
V
lines in each panel mark the redshift boundaries of the regions for which we measure j λrest .
which corresponds to a linear fit to the NFGS data. These effects might be greater
for objects with strong AGN contribution to their fluxes.
4 The rest-frame optical luminosity density, color, and stellar mass density of the
72
universe from z=0 to z=3
4.4 The Properties of the Massive Galaxy Population
4
73
The Properties of the Massive Galaxy Population
In this section we discuss the use of the zphot and Lrest
estimates to derive the
λ
integrated properties of the population, namely the luminosity density, the mean
cosmic rest-frame color, the stellar mass-to-light ratio M/L∗ , and the stellar mass
density ρ∗ . As will be described below, addressing the integrated properties
of the population reduces many of the uncertainties associated with modeling
individual galaxies and, in the case of the cosmic color, is less sensitive to field-tofield variations.
4.1
The Luminosity Density
tot
< 25 galaxies (see §3.2), we traced
Using our Lrest
estimates from the Ks,AB
λ
the redshift evolution of the rest-frame optically most luminous, and therefore
presumably most massive, galaxies by measuring the rest-frame luminosity density
jλrest of the visible stars associated with them. The results are presented in Table 1
and plotted against redshift and elapsed cosmic time in Figure 2. As our best
alternative to a selection by galaxy mass, we selected our galaxies in our reddest
rest-frame band available at z ∼ 3, i.e. the V -band. In choosing the z and
Lrest
regime over which we measured jλrest we wanted to push to as high of a redshift
λ
as possible with the double constraint that the redshifted rest-frame filter still
overlapped with the Ks filter and that we were equally complete at all considered
= 1.4 × 1010 h−2
threshold, Lthresh
redshifts. By choosing an Lrest
70 L¯ , and a
V
V
with
maximum redshift of z = 3.2, we could select galaxies down to 0.6 Llocal
∗,V
constant efficiency regardless of SED type. We then divided the range out to
z = 3.2 into three bins of equal co-moving volume which correspond to the redshift
intervals 0–1.6, 1.6–2.41, and 2.41–3.2.
In a given redshift interval, we estimated jλrest directly from the data in two
steps. We first added up all the luminosities of galaxies which satisfied our
criteria defined above and which had δzphot /(1 + zphot ) ≤ 0.16, roughly
Lthresh
V
twice the mean disagreement between zphot and zspec (see §3.1). Galaxies rejected
by our δzphot cut but with Lrest
>Lthresh
, however, contribute to the total lumiV
V
nosity although they are not included in this first estimate. Under the assumption
that these galaxies are drawn from the same luminosity function as those which
passed the δzphot cut, we computed the total luminosity, including the light from
the nacc accepted galaxies and the light lost from the nrej rejected galaxies as
Ltot = Lmeas × (1 +
nrej
).
nacc
(2)
As a test of the underlying assumption of this correction we performed a K-S test
on the distributions of Ks magnitudes for the rejected and accepted galaxies in
each of our three volume bins. In all three redshift bins, the rejected galaxies have
tot
Ks,AB
distributions which are consistent at the > 90% level with being drawn from
the same magnitude distribution as the accepted galaxies. The total correction
4 The rest-frame optical luminosity density, color, and stellar mass density of the
74
universe from z=0 to z=3
per volume bin ranges from 5 − 10% in every bin. Our results do not change if we
omit those galaxies whose photometric redshift probability distribution indicates
a secondary minima in chisquared.
Uncertainties in the luminosity density were computed by bootstrapping from
tot
the Ks,AB
< 25 subsample. This method does not take cosmic variance into
account and the errors may therefore underestimate the true error, which includes
field-to-field variance.
Redshift errors might effect the luminosity density in a systematic way, as
they produce a large error in the measured luminosity. This, combined with a
steep luminosity function can bias the observed luminosities upwards, especially
at the bright end. This effect can be corrected for in a full determination of the
luminosity function (e.g., Chen et al. 2003), but for our application we estimated
the strength of this effect by Monte-Carlo simulations. When we used the formal
redshift errors we obtained a very small bias (6%), if we increase the photometric
redshift errors in the simulation to be minimally as large as 0.08 ∗ (1 + z), we still
obtain a bias in the luminosity density on the order of 10% or less.
Because we exclude galaxies with faint rest-frame luminosities or low apparent magnitudes, and do not correct for this incompleteness, our estimates should
be regarded as lower limits on the total luminosity density. One possibility for
estimating the total luminosity density would be to fit a luminosity function as
a function of redshift and then integrate it over the whole luminosity range. We
don’t go faint enough at high redshift, however, to tightly constrain the faint-end
slope α. Because extrapolation of jλrest to arbitrarily low luminosities is very dependent on the value of α, we choose to use this simple and direct method instead.
tot
< 25 raises the jλrest values in the z = 0 − 1.6
Including all galaxies with Ks,AB
redshift bin by 86%, 74%, and 66% in the U , B , and V -bands respectively. Likewise, the jλrest values would increase by 38%, 35%, and 44% for the z = 1.6 − 2.41
bin and would increase by 5%, 5%, and 2% for the z = 2.41 − 3.2 bin, again in the
U , B , and V -bands respectively.
The dip in the luminosity density in the second lowest redshift bin of the (a)
and (b) panels of Figure 2 can be traced to the lack of intrinsically luminous
galaxies at z ∼ 1.5 − 2 (§3.2; R01). The dip is not noticeable in the U -band
because the galaxies at z ∼ 2 are brighter with respect to the z < 1.6 galaxies in
the U -band than in the V or B -band, i.e. they have bluer (U − B)rest colors
and (U − V )rest colors than galaxies at z < 1.6. This lack of rest-frame optically
bright galaxies at z ∼ 1.5 − 2 may result from systematics in the zphot estimates,
which are poorly tested in this regime and where the Lyman break has not yet
entered the U300 filter, or may reflect a true deficit in the redshift distribution of
Ks -band luminous galaxies (see §3.2).
At z . 1 our survey is limited by its small volume. For this reason, we supplement our data with other estimates of jλrest at z . 1.
We compared our results with those of the SDSS as follows. First, we selected
4.4 The Properties of the Massive Galaxy Population
75
SDSS Main sample galaxies (Strauss et al. 2002) with redshifts in the SDSS Early
Data Release (Stoughton et al. 2002) in the EDR sample provided by and described
by Blanton et al. (2003a). Using the product kcorrect v1 16 (Blanton et al.
2003b), for each galaxy we fit an optical SED to the 0.1u0.1g 0.1r0.1i0.1z magnitudes,
after correcting the magnitudes to z = 0.1 for evolution using the results of Blanton
et al. (2003c). We projected this SED onto the U BV filters as described by
Bessell (1990) to obtain absolute magnitudes in the U BV Vega-relative system.
Using the method described in Blanton et al. (2003a) we calculated the maximum
volume Vmax within the EDR over which each galaxy could have been observed,
accounting for the survey completeness map and the flux limit as a function of
position. 1/Vmax then represents the number density contribution of each galaxy.
From these results we constructed the number density distribution of galaxies as a
function of color and absolute magnitude and the contribution to the uncertainties
in those densities from Poisson statistics. While the Poisson errors in the SDSS
are negligible, cosmic variance does contribute to the uncertainties. For a more
realistic error estimate, we use the fractional errors on the luminosity density from
Blanton et al. (2003c). For the SDSS luminosity function, our Lthresh
encompasses
V
54% of the total light.
In Figure 2 we also show the jλrest measurements from the COMBO-17 survey
(Wolf et al. 2003). We used a catalog with updated redshifts and 29471 galaxies
(the J2003 catalog; Wolf, C. private
>Lthresh
at z < 0.9, of which 7441 had Lrest
V
V
communication). Using this catalog we calculated jλrest in an identical way to how
it was calculated for the FIRES data. We divided the data into redshift bins
of ∆z = 0.2 and counted the light from all galaxies contained within each bin
. The formal 68% confidence limits were calculated via
>Lthresh
which had Lrest
V
V
bootstrapping. In addition, in Figure 2 we indicate the rms field-to-field variations
between the three spatially distinct COMBO-17 fields. As also pointed out in Wolf
et al. (2003), the field-to-field variations dominate the error in the COMBO-17
jλrest determinations.
Bell et al. (2003b) point out that uncertainties in the absolute calibration and
relative calibration of the SDSS and Johnson zeropoints can lead to . 10% errors
in the derived rest-frame magnitudes and colors of galaxies. To account for this,
we add a 10% error in quadrature with the formal errors for both the COMBO-17
and SDSS luminosity densities. These are the errors presented in Table 1 and
Figure 2.
rest
value of luminous LBGs determined by
In Figure 2a we also plot the jV
integrating the luminosity function of Shapley et al. (2001) to Lthresh
. A direct
V
comparison between our sample and theirs is not entirely straightforward because
the LBGs represent a specific class of non-obscured, star forming galaxies at high
redshift, selected by their rest-frame far UV light. Nonetheless, our jλrest determination at z = 2.8 is slightly higher than their determination at z = 3, indicating
either that the HDF-S is overdense with respect to the area surveyed by Shapley
et al. or that we may have galaxies in our sample which are not present in the
4 The rest-frame optical luminosity density, color, and stellar mass density of the
76
universe from z=0 to z=3
ground-based LBG sample.
D03 have also measured the luminosity density in the rest-frame B -band but,
because they do not give their luminosity function parameters except for their
lowest redshift bin, it is not possible to overplot their luminosity density integrated
down to our Lrest
limit.
V
4.1.1
The Evolution of jλrest
We find progressively stronger luminosity evolution from the V to the U -band:
whereas the evolution is quite weak in V , it is very strong in U . The jλrest in
our highest redshift bin is a factor of 1.9 ± 0.4, 2.9 ± 0.6, and 4.9 ± 1.0 higher
than the z = 0.1 value in V , B , and U respectively. To address the effect
of cosmic variance on the measured evolution in jλrest we rely on the clustering
analysis developed for our sample in Daddi et al. (2003). Using the correlation
length estimated at 2 < z < 4, ro = 5.5 h−1
100 Mpc, we calculated the expected
1-sigma fluctuations in the number density of objects in our two highest redshift
bins. Because for our high-z samples the poissonian errors are almost identical to
the bootstrap errors, we can use the errors in the number density as a good proxy
to the errors in jλrest . The inclusion of the effects of clustering would increase the
bootstrap errors on the luminosity density by a factor of, at most, 1.75 downwards
and 2.8 upwards. This implies that the inferred evolution is still robust even in the
face of the measured clustering. The COMBO-17 data appear to have a slightly
steeper rise towards higher redshift than our data, however there are two effects to
remember at this point. First, our lowest redshift point averages over all redshifts
z < 1.6, in which case we are in reasonably good agreement with what one would
predict from the average of the SDSS and COMBO-17 data. Second, our data may
simply have an offset in density with respect to the local measurements. Such an
offset affects the values of jλrest , but as we will show in §4.2, it does not strongly
affect the global color estimates. Nonetheless, given the general increase with
jλrest towards higher redshifts, we fit the changing jλrest with a power law of the
form jλrest (z) = jλrest (0) ∗ (1 + z)β . These curves are overplotted in Figure 2 and the
best fit parameters in sets of (jλrest (0), β) are (5.96 × 107 , 1.41), (6.84 × 107 , 0.93),
and (8.42 × 107 , 0.52) in the U , B , and V bands respectively, where jλrest (0) has
units of h70 L¯ Mpc−3 . At the same time, it is important to remember that our
power law fit is likely an oversimplification of the true evolution in jλrest .
The increase in jλrest with decreasing cosmic time can be modeled as a simple
brightening of L∗ . Performing a test similar to that performed in R01, we deterneeded to match the observed
mine the increase in L∗,V with respect to Llocal
∗,V
increase in jλrest from z = 0.1 to 2.41 < z < 3.2, assuming the SDSS Schechter
function parameters. To convert between the Schechter function parameters in the
SDSS bands and those in the Bessell (1990) filters we transformed the LSDSS
∗,0.1 r values
to the Bessell V filter using the (V −0.1 r) color, where the color was derived from
the total luminosity densities in the indicated bands (as given in B03). We then ap-
0.10 ± 0.10 a
0.30 ± 0.10 b
0.50 ± 0.10 b
0.70 ± 0.10 b
0.90 ± 0.10 b
+0.48
1.12−1.12
c
+0.40 c
2.01−0.41
+0.40 c
2.80−0.39
rest
log jU
[h70 L¯,U Mpc−3 ]
+0.04
7.89−0.05
+0.05
7.84−0.05
+0.04
8.01−0.05
+0.04
8.18−0.05
+0.04
8.22−0.05
+0.08
8.11−0.09
+0.08
8.21−0.10
+0.07
8.58−0.08
rest
log jB
[h70 L¯,B Mpc−3 ]
+0.04
7.87−0.05
+0.05
7.85−0.05
+0.04
7.99−0.05
+0.04
8.13−0.05
+0.04
8.13−0.05
+0.08
8.02−0.08
+0.08
8.00−0.10
+0.07
8.32−0.08
rest
log jV
[h70 L¯,V Mpc−3 ]
+0.04
7.91−0.05
+0.05
7.93−0.05
+0.04
8.01−0.05
+0.04
8.12−0.05
+0.04
8.09−0.05
+0.08
8.00−0.09
+0.08
7.89−0.09
+0.07
8.18−0.08
jλrest and rest-frame colors calculated for galaxies with Lrest
> 1.4 × 1010 h−2
V
70 L¯,V .
(U − B)rest
(B − V )rest
0.14+0.02
−0.02
0.21+0.02
−0.02
0.16+0.01
−0.01
0.06+0.01
−0.01
−0.04+0.004
−0.01
−0.04+0.03
−0.03
−0.34+0.04
−0.03
−0.44+0.04
−0.03
0.75+0.02
−0.02
0.84+0.01
−0.01
0.69+0.005
−0.01
0.64+0.01
−0.01
0.55+0.005
−0.01
0.61+0.02
−0.02
0.38+0.04
−0.04
0.29+0.04
−0.03
4.4 The Properties of the Massive Galaxy Population
Table 1 — Rest-Frame Optical Luminosity Density and Integrated Color
z
a
SDSS
COMBO-17
c
FIRES
b
77
4 The rest-frame optical luminosity density, color, and stellar mass density of the
78
universe from z=0 to z=3
Figure 2 — The rest-frame optical luminosity density vs. cosmic age and redshift from galaxies
tot
. For comparison we plot jλrest determinations from other
>Lthresh
< 25 and Lrest
with Ks,AB
V
V
surveys down to our Lrest
limits.
The
squares are those from our data, the triangles are from
λ
the Combo-17 survey (Wolf et al. 2003), the circle is that at z = 0.1 from the SDSS (B03),
and the pentagon is that from Shapley et al. (2001). The dotted errorbars on the COMBO-17
data indicate the rms field-to-field variation derived from the three spatially distinct COMBO-17
fields. The solid line is a power law fit to the FIRES, COMBO-17, and SDSS data of the form
jλrest (z) = jλrest (0) ∗ (1 + z)β .
plied the appropriate AB to Vega correction tabulated in Bessell (1990). Because
the difference in λef f is small between the two filters in each of these colors, the
shifts between the systems are less than 5%. The luminosity density in the V -band
rest
= 1.53 ± 0.26 × 108 h70 L¯ Mpc−3 . Using the V -band
at 2.41 < z < 3.2 is jV
Schechter function parameters for our cosmology, φSDSS
= 5.11×10−3 h370 Mpc−3 ,
∗
10 −2
αSDSS = −1.05, and LSDSS
=
2.53
×
10
h
L
,
we
can
match the increase in
¯
70
∗,V
rest
jV if L∗,V brightens by a factor of 1.7 out to 2.41 < z < 3.2.
4.2
The Cosmic Color
Using our measures of jλrest we estimated the cosmic rest-frame color of all the
visible stars which lie in galaxies with Lrest
> 1.4 × 1010 h−2
70 L¯ . We derived the
V
mean cosmic (U − B)rest and (B − V )rest by using the jλrest estimates from
4.4 The Properties of the Massive Galaxy Population
79
4 The rest-frame optical luminosity density, color, and stellar mass density of the
80
universe from z=0 to z=3
the previous section with the appropriate magnitude zeropoints. The measured
colors for the FIRES data, the COMBO-17 data, and the SDSS data are given
in Table 1. Emission line corrected (U − B)rest colors may be calculated by
applying Equation 1 to the values in the table. For the FIRES and COMBO-17
data, uncertainty estimates are derived from the same bootstrapping simulation
used in § 4.1. In this case, however, the COMBO-17 and SDSS errorbars do
not include an extra component from errors in the transformation to rest-frame
luminosities, since these transformation errors may be correlated in a non-trivial
way.
The bluing with increasing redshift which could have been inferred from Figure 2 is seen explicitly in Figure 3. The color change towards higher redshift occurs
more smoothly than the evolution in jλrest , with our FIRES data meshing nicely
with the COMBO-17 data. It is immediately apparent that the rms field-to-field
errors for the COMBO-17 data are much less than the observed trend in color,
in contrast to Figure 2. This explicitly shows that the integrated color is much
less sensitive than jλrest to field-to-field variations, even when such variations may
dominate the error in the luminosity density. The COMBO-17 data at z . 0.4
are also redder than the local SDSS data, possibly owing to the very small central
apertures used to measure colors in the COMBO-17 survey. The colors in the
COMBO-17 data were measured with the package MPIAPHOT using using the peak
surface brightness in images smoothed all to identical seeing (1.00 5). Such small
apertures were chosen to measure very precise colors, not to obtain global color
estimates. Because of color gradients, these small apertures can overestimate the
global colors in nearby well resolved galaxies, while providing more accurate global
color estimates for the more distant objects. Following the estimates of this bias
provided by Bell et al. (2003b), we increased the color errorbars on the blue side
to 0.1 for the z < 0.4 COMBO-17 data. It is encouraging to see that the color
evolution is roughly consistent with a rather simple model and that it is much
smoother than the luminosity density evolution, which is more strongly affected
by cosmic variance.
We interpreted the color evolution as being primarily driven by a decrease in
the stellar age with increasing redshift. Applying the (B − V )rest dependent
emission line corrections inferred from local samples (See §3.2.1), we see that the
effect of the emission lines on the color is much less than the magnitude of the
observed trend. We can also interpret this change in color as a change in mean
cosmic M/L∗ with redshift. In this picture, which is true for a variety of monotonic
SFHs and extinctions, the points at high redshift have lower M/L∗ than those at
rest
low redshift. At the same time, however, the evolution in jV
with redshift is
quite weak. Taken together this would imply that the stellar mass density ρ ∗ is
also decreasing with increasing redshift. We will quantify this in §4.3.
To show how our mean cosmic (U − B)rest and (B − V )rest colors compare
to those of morphologically normal nearby galaxies, we overplot them in Figure 4
on the locus of nearby galaxies from Larson & Tinsley (1978). The integrated
4.4 The Properties of the Massive Galaxy Population
81
colors, at all redshifts, lie very close to the local track, which Larson & Tinsley
(1978) demonstrated is easily reproduceable with simple monotonically declining
SFHs and which is preserved in the presence of modest amounts of reddening,
which moves galaxies roughly parallel to the locus. In fact, correcting our data for
emission lines moved them even closer to the local track. While we have suggested
that M/L∗ decreases with decreasing color, if we wish to actually quantify the
M/L∗ evolution from our data we must first attempt to find a set of models which
can match our observed colors and which we will later use to convert between
the color and M/L∗V . We overplot in Figure 4 two model tracks corresponding
to an exponential SFH with τ = 6 Gyr and with E(B − V ) = 0, 0.15, and 0.35
(assuming a Calzetti et al. (2000) reddening law). These tracks were calculated
using the 2000 version of the Bruzual A. & Charlot (1993) models and have Z = Z ¯
and a Salpeter (1955) IMF with a mass range of 0.1-120M¯ . Other exponentially
declining models and even a constant star forming track all yield similar colors
to the τ = 6 Gyr track. The measured cosmic colors at z < 1.6 are fairly well
approximated either of the reddening models. At z > 1.6, however, only the
E(B − V ) = 0.35 track can reproduce the data. This high extinction is in contrast
to the results of P01 and Shapley et al. (2001) who found a mean reddening for
LBGs of E(B − V ) ∼ 0.15. SY98 and Thompson, Weymann, & Storrie-Lombardi
(2001), however, measured extinctions on this order for galaxies in the HDF-N,
although the mean extinction from Thompson, Weymann, & Storrie-Lombardi
(2001) was lower at z > 2. The amount of reddening in our sample is one of the
largest uncertainty in deriving the M/L∗ values, nonetheless, our choice of a high
extinction is the only allowable possibility given the integrated colors of our high
redshift data.
Although this figure demonstrates that the measured colors can be matched, at
some age, by this simple E(B − V ) = 0.35 model, we must nevertheless investigate
whether the evolution of our model colors are also compatible with the evolution
in the measured colors. This is shown by the track in Figure 3. We have tried
different combinations of τ , E(B − V ), and zstart , but have not been able to find
a model which fits the data well at all redshifts. The parameterized SFR(z) curve
of Cole et al. (2001) also provided a poor fit to the data. Given the large range of
possible parameters, our data may not be sufficient to well constrain the SFH.
4.3
Estimating M/L∗V and The Stellar Mass Density
In this subsection we describe the use of our mean cosmic color estimates to derive
the mean cosmic M/L∗V and the evolution in ρ∗ .
The main strength of considering the luminosity density and integrated colors
of the galaxy population, as opposed to those of individual galaxies, lies in the
simple and robust ways in which these global values can be modeled. When attempting to derive the SFHs and stellar masses of individual high-redshift galaxies,
the state-of-the-art models for the broad-band colors only consider stellar populations with at most two separate components (SY98; P01; Shapley et al. 2001).
4 The rest-frame optical luminosity density, color, and stellar mass density of the
82
universe from z=0 to z=3
Figure 3 — The evolution of the cosmic color plotted against redshift and cosmic time for our
data in addition to data from other z . 1 surveys. The squares are those from our data, the
triangles are from the Combo-17 survey (Wolf et al. 2003), and the circle is that at z = 0.1
from the SDSS (B03). The open symbols indicate the empirical emission line correction to the
integrated colors derived using the spectroscopic and photometric data from the NFGS (Jansen
et al. 2000b). The dotted errorbars on the COMBO-17 data indicate the field-to-field variation.
Note that the integrated rest-frame color is much more stable than jλrest against field-to-field
variations. The COMBO-17 data point at z = 0.3 has been given a color errorbar of 0.1 in the
blueward direction and an open symbol to reflect the possible systematic biases resulting from
their very small central apertures. We also overplot a model with an exponentially declining SFH
with τ = 6Gyr, E(B − V ) = 0.35, and zstart = 4.0 assuming a Calzetti et al. (2000) extinction
law.
Using their stellar population synthesis modeling, Shapley et al. (2001) proposes
a model in which LBGs likely have smooth SFHs. On the other hand, SY98 concluded that they may only be seeing the most recent episode of star formation and
that LBGs may indeed have bursting SFHs. This same idea was supported by P01
and Ferguson, Dickinson, & Papovich (2002) using much deeper NIR data. When
using similar simple SFHs to model the cosmic average of the galaxy population,
4.4 The Properties of the Massive Galaxy Population
83
Figure 4 — The (U − B)rest vs. (B − V )rest at z =1.12, 2.01, and 2.8 of all the relatively
unobscured stars in galaxies with Lrest
> 1.4 × 1010 h−2
70 L¯ . The thick solid black line is the
V
local relation derived by Larson & Tinsley (1978) from nearby morphologically normal galaxies.
The large symbols are identical to those in Figure 3. For clarity we do not plot the field-to-field
errorbars for the COMBO-17 data. The small solid points are the colors of nearby galaxies
from the NFGS (Jansen et al. 2000a), which have been corrected for emission lines. The small
crosses are the NFGS galaxies which harbor AGN. The thin tracks correspond to an exponentially
declining SFH with a timescale of 6 Gyr. The tracks were created using a Salpeter (1955) IMF
and the 2000 version of the Bruzual A. & Charlot (1993) models. The dotted track has no
extinction, the dashed track has been reddened by E(B − V ) = 0.15, and the thin solid track
has been reddened by E(B − V ) = 0.35, using the Calzetti et al. (2000) extinction law. The
black arrow indicates the reddening vector applied to the solid model track. The emission line
corrected data lie very close the track defined by observations of local galaxies and the agreement
with the models demonstrates that simple SFHs can be used to reproduce the integrated colors
from massive galaxies at all redshifts.
a more self-consistent approach is possible. While individual galaxies may, and
probably do, have complex SFHs, the mean SFH of all galaxies is much smoother
than that of individual ones.
Encouraged by the general agreement between the measured colors and the
simple models, we attempted to use this model to constrain the stellar mass-to-
4 The rest-frame optical luminosity density, color, and stellar mass density of the
84
universe from z=0 to z=3
light ratio M/L∗V in the rest-frame V -band, by taking advantage of the relation
between color and log10 M/L∗ found by Bell & de Jong (2001). For monotonic
SFHs, the scatter of this relation remains small in the presence of modest variations
in the reddening and metallicity because these effects run roughly parallel to the
mean relation. Using the τ = 6 Gyr exponentially declining model, we plot in
Figure 5 the relation between (U − V )rest and M/L∗V for the E(B − V ) = 0,
0.15, and 0.35 models. It is seen that dust extinction moves objects roughly
parallel to the model tracks, reddening their colors, but making them dimmer
as well and hence increasing M/L∗V . Nonetheless, extinction uncertainties are a
major contributor to our errors in the determination of M/L∗V . We chose to derive
M/L∗V from the (U − V )rest color instead of from the (B − V )rest color because at
blue colors, where our high redshift points lie, (B − V )rest derived M/L∗V values
are much more sensitive to the exact value of the extinction. This behavior likely
stems from the fact that the (U − V )rest color spans the Balmer/4000Å break
and hence is more age sensitive than (B − V )rest . At the same time, while
(U − B)rest colors are even less sensitive to extinction than (U − V )rest , they
are more susceptible to the effects of bursts.
We constructed our relation using a Salpeter (1955) IMF5 . The adoption of a
different IMF would simply change the zeropoint of this curve, leaving the relative
M/L∗ as a function of color, however, unchanged. As discussed in §4.2, this model
does not fit the redshift evolution of the cosmic color very well. Nonetheless, the
impact on our M/L∗V estimates should not be very large, since most smooth SFHs
occupy very similar positions in the M/L∗V vs. U − V plane.
This relation breaks down in the presence of more complex SFHs. We demonstrate this in Figure 6 where we plot the τ = 6 Gyr track and a second track
whose SFH is comprised of a 50 Myr burst at t = 0, a gap of 2 Gyr, and a constant SFR rate for 1 Gyr thereafter, where 50% of the mass is formed in the burst.
It is obvious from this figure that using a smooth model will cause errors in the
M/L∗V estimate if the galaxy has a SFR which has an early peak in the SFH. At
blue colors, such a early burst of SFR will cause an underestimate of M/L∗V , a
result similar to that of P01 and D03. At red colors, however, M/L∗V would be
overestimated with the exact systematic offset as a function of color depending
strongly on the detailed SFH, i.e. the fraction of mass formed in the burst, the
length of the gap, and the final age of the stellar population.
The models show that the method may over- or underestimate the true stellar
mass-to-light ratio if the galaxies have complex SFHs. It is important to quantify
the errors on the global M/L∗V based on the mean (U − V )rest color and how
these errors compare to those when determining the global M/L∗V value from
individual M/L∗V estimates. To make this comparison we constructed a model
whose SFH consist of a set of 10 Myr duration bursts separated by 90 Myr gaps.
We drew galaxies at random from this model by randomly varying the phase and
5 We do not attempt to model an evolving IMF although evidence for a top-heavy IMF at
high redshifts has been presented by Ferguson, Dickinson, & Papovich (2002)
4.4 The Properties of the Massive Galaxy Population
85
Figure 5 — The relation between (U − V ) and M/L∗V for a model track with an exponential
timescale of 6 Gyr. The dotted line is for a model with E(B − V ) = 0, the dashed line for a
model with E(B − V ) = 0.15, and the solid line is for a model reddened by E(B − V ) = 0.35
(using a Calzetti extinction law), which we adopt for our M/L∗V conversions. The vertical solid
arrows indicate the colors of the three FIRES data points, the vertical dotted arrow indicates
the color of the SDSS data, and the diagonal solid arrow indicates the vector used to redden the
E(B − V ) = 0 model to E(B − V ) = 0.35. The labels above the vertical arrows correspond to
the redshifts of the FIRES and SDSS data.
age of the burst sequence, where the maximum age was 4 Gyr. Next we estimated
the total mass-to-light ratios of this sample by two different methods; first we
determined the M/L∗V for the galaxies individually assuming the simple relation
between color and mass-to-light ratio, and we took the luminosity weighted mean
of the individual estimates to obtain the total M/L∗V . This point is indicated by
a large square in Figure 7 and overestimates the total M/L∗V by ∼ 70%. Next
we first add the light of all the galaxies in both U and V , then use the simple
relation between color and M/L∗V to convert the integrated (U − V )rest into a
mass-to-light ratio. This method overestimates M/L∗V by much less, ∼ 35%. This
4 The rest-frame optical luminosity density, color, and stellar mass density of the
86
universe from z=0 to z=3
Figure 6 — The effect of an early burst of star formation on the relation between (U − V ) and
M/L∗V . The relation between (U − V ) and M/L∗V for a model track with an exponential
timescale of 6 Gyr is show by the solid line. We also show a track for a SFH which includes a 50
Myr burst at t = 0 followed by a gap of 2 Gyr and then a constant SFR rate for 1 Gyr thereafter,
where the fraction of mass formed in the burst is 0.5. The track continues for a total time of 4.5
Gyr. The dots are placed at 100 Myr intervals and the dotted section of the line indicates the
very rapid transition in color caused by the onset of the second period of star formation. Both
tracks have the same extinction.
comparison shows clearly that it is best to estimate the mass using the integrated
light. This is not unexpected; the star formation history of the universe as a whole
is more regular than the star formation history of individual galaxies. If enough
galaxies are averaged, the mean star formation history is naturally fairly smooth.
Using the relationship between color and M/L∗V we convert our (U − V )rest and
measurements to stellar mass density estimates ρ∗ . The resulting ρ∗ values
are plotted vs. cosmic time in Figure 8. We have included points for the SDSS survey created in an analogous way to those from this work, i.e. using the M/L ∗V derived from the rest-frame color and multiplying it by jVSDSS for all galaxies with
rest
jV
4.4 The Properties of the Massive Galaxy Population
87
Figure 7 — A comparison of different measures of the global M/L∗V for a mock catalog of
galaxies with bursting SFHs. The solid line represents the relation between (U − V ) and
M/L∗V for a model track with an exponential timescale of 6 Gyr. The black dots show the true
M/L’s of the model starbursting galaxies, as described in the text; the open circle shows the true
luminosity weighted Mtot /Ltot of the mock galaxies. The square shows the luminosity weighted
M/L∗V derived by applying the simple model to the individual galaxies - in this case, the mean
Mtot /Ltot is overestimated by 70%. The triangle is the Mtot /Ltot derived from the luminosity
weighted mean color (or (U − V )tot ) of the model galaxies. It overestimates Mtot /Ltot by only
35%.
. The ρ∗ estimates are listed in Table 2. We have derived the statis>Lthresh
Lrest
V
V
tical errorbars on the ρ∗ estimates by creating a Monte-Carlo simulation where we
rest
allowed jV
and (U − V )rest (and hence M/L∗V ) to vary within their errorbars.
We then took the resulting distribution of ρ∗ values and determined the 68% confidence limits. As an estimate of our systematic uncertainties corresponding to the
method we also determined M/L∗V from the (U − B)rest and (B − V )rest data
using an identical relation as for the (U − V )rest to M/L∗V conversion. The
(U − B)rest derived M/L∗V values were different from the (U − V )rest values
4 The rest-frame optical luminosity density, color, and stellar mass density of the
88
universe from z=0 to z=3
Table 2 — M/L∗V and Stellar Mass Density Estimates
z
0.1 ± 0.1 a
b
1.12+0.48
−1.12
+0.40 b
2.01−0.41
b
2.80+0.40
−0.39
a
b
log M/L∗V
¯
[M
L¯ ]
+0.03
0.54−0.03
0.13+0.07
−0.06
−0.42+0.09
−0.10
−0.70+0.11
−0.12
log ρ∗
[h70 M¯ Mpc−3 ]
8.49+0.04
−0.05
8.14+0.11
−0.10
7.48+0.12
−0.16
7.49+0.12
−0.14
SDSS
FIRES
by a factor of 1.02, 0.80, 0.95, and 1.12 for the z =0.1, 1.12, 2.01, and 2.8 redshift
bins respectively. Likewise the (B − V )rest determined M/L∗V values changed by
a factor of 0.99, 1.11, 1.15, and 0.80 with respect to the (U − V )rest values. While
the (U − V )rest values are very similar to those derived from the other colors, the
(U − V )rest color is less susceptible to dust uncertainties than the (B − V )rest data
and less susceptible to the effects of bursts than the (U − B)rest data.
The derived mass density rises monotonically by a factor of ∼ 10 all the way
to z ∼ 0.1, with our low redshift point meshing nicely with the local SDSS point.
5
5.1
Discussion
Comparison with other Work
Figure 8 shows a consistent picture of the build-up of stellar mass, both for the
luminous galaxies and the total galaxy population. It is remarkable that the results
from different authors appear to agree well given that the methods to derive the
densities were different and that the fields are very small.
We compared our results to the total mass estimates of other authors in Figure 8. In doing this we must remember, because of our Lthresh
cut, that we are
λ
missing significant amounts of light, and hence, mass. Assuming the SDSS luminosity function parameters, we lose 46% of the light at z = 0. At z = 2.8, however,
we inferred a brightening of L∗,V by a factor of 1.7, implying that we go further
down the luminosity function at high redshift, sampling a larger fraction of the
total starlight. If we apply this brightening to the SDSS L∗,V we miss 30% of
the light below our luminosity threshold at z = 2.8. Hence, the fraction of the
total starlight contained in our sample is rather stable as a function of redshift.
To graphically compare our data to other authors we have scaled the two different
axes in Figure 8 so that our derivation of the SDSS ρ∗ is at the same height of the
total ρ∗ estimate of Cole et al. (2001). At z < 1 we compared our mass estimates
to those of Brinchmann & Ellis (2000). Following D03, we have corrected their
4.5 Discussion
89
Figure 8 — The build-up of the stellar mass density as a function of redshift. The solid points
> 1.4×1010 h−2
are for galaxies with Lrest
70 L¯ and were derived by applying the E(B−V ) = 0.35
V
relation in Figure 5 to the (U − V )rest colors and jλrest measurements from the FIRES (solid
squares) and SDSS data (solid circle). The y-axis scale on the left side corresponds to the
ρ∗ values for these points. The hollow points show the total stellar mass density measurements
from the one-component models in the HDF-N (D03; hollow stars; calculated assuming solar
metallicity), the CFRS (Brinchmann & Ellis 2000; hollow circles), and the 2dFGRS + 2MASS
(Cole et al. 2001; hollow hexagon). The dotted errorbars on the D03 points reflect the systematic
mass uncertainties resulting from metallicity and SFH changes. The y-axis scale on the right
hand side corresponds to the ρ∗ estimates for these points. The relative scaling of the two axes
was adjusted so that our SDSS ρ∗ estimate was at the same height as the total ρ∗ estimate of
Cole et al. The solid curve is an integral of the SFR(z) from Cole et al. (2001) which has been
fit to extinction corrected data at z . 4.
published points to total masses by correcting them upwards by 20% to account for
their mass incompleteness. The fraction of the total stars formed at z < 1 agrees
well between our data and that of SDSS and Brinchmann & Ellis. At z > 0.5, we
compared our results to those of D03. D03 calculated the total mass density, using the integrated luminosity density in the rest-frame B-band coupled with M/L
4 The rest-frame optical luminosity density, color, and stellar mass density of the
90
universe from z=0 to z=3
measurements of individual galaxies. The fractions of the total stars formed in our
sample (60%, 13%, and 9% at z =1.12, 2.01, and 2.8) are almost twice as high as
those of D03. The results, however, are consistent within the errors.
We explored whether field-to-field variations may play a role in the discrepancy between the two datasets. D03 studied the HDF-N, which has far fewer
”red”galaxies than HDF-S (e.g., Labbé et al. 2003, Franx et al, 2003). If we
omit the J − K selected galaxies found by Franx et al. (2003) in the HDF-S, the
formal M/L∗V decreases to 45% and 43% of the total values and the mass density
decreases to 57% and 56% of the total values in the z = 2.01 and z = 2.8 bins
respectively, bringing our data into better agreement with that from D03. This
reinforces the earlier suggestion by Franx et al. (2003) that the J − K selected
galaxies contribute significantly to the stellar mass budget.
The errors in both determinations are dominated by systematic uncertainties,
although our method should be less sensitive to bursts than that of D03 as it uses
the light integrated integrated over the galaxy population.
We note that Fontana et al. (2003) have also measured the stellar mass density in the HDF-S using a catalog derived from data in common with our own.
They find a similar, although slightly smaller evolution in the stellar mass density,
consistent with our result to within the uncertainties.
5.2
Comparison with SFR(z)
We can compare the derived stellar mass to the mass expected from determinations
of the SFR as a function of redshift. We use the curve by Cole et al. (2001), who
fitted the observed SFR as determined from various sources at z . 4. To obtain
the curve in Figure 8 we integrated the SFR(z) curve taking into account the time
dependent stellar mass loss derived from the 2000 version of the Bruzual A. &
Charlot (1993) population synthesis models.
We calculated a reduced χ2 of 4.3 when comparing all the data to the model. If,
however, we omit the Brinchmann & Ellis (2000) data, the reduced χ2 decreases to
1.8, although the results at z > 2 lie systematically below the curve. This suggests
that some systematic errors may play a role, or that the curve is not quite correct.
The following errors can influence our mass density determinations:
-Dusty, evolved populations: it is assumed that the dust is mixed in a simple
way with the stars, leading to a Calzetti extinction curve. If the dust is distributed
differently, e.g., by having a very extincted underlying evolved population, or by
having a larger R value, the current assumptions lead to a systematic underestimate of the mass. If an underlying, extincted evolved population exists, it would
naturally explain the fact that the ages of the Lyman-break galaxies are much
younger than expected (e.g., P01, Ferguson et al. 2002). There may also be galaxies which contribute significantly to the mass density but are so heavily extincted
that they are undetected, even with our very deep Ks -band data. If such objects
4.5 Discussion
91
are also actively forming stars, they may be detectable with deep submillimeter
observations or with rest-frame NIR observations from SIRTF.
-Cosmic variance: the two fields which have been studied are very small. Although we use a consistent estimate of clustering from Daddi et al. (2003), red
galaxies make up a large fraction of the mass density in our highest redshift bins.
Since red galaxies have a very high measured clustering from z ∼ 1 (e.g., Daddi
et al. 2000, Mccarthy et al. 2001) up to possibly z ∼ 3 (Daddi et al. 2003), large
uncertainties remain.
-Evolving Initial Mass Function: the light which we see is mostly coming from
the most massive stars present, whereas the stellar mass is dominated by low mass
stars. Changes in the IMF would immediately lead to different mass estimates
but if the IMF everywhere is identical (as we assume), then the relative masses
should be robust. If the IMF evolves with redshift, however, systematic errors in
the mass estimate will occur.
-A steep galaxy mass function at high redshift: if much of the UV light which
is used to measure the SFR at high redshifts comes from small galaxies which
would fall below our rest-frame luminosity threshold then we may be missing
significant amounts of stellar mass. Even the mass estimates of D03, which were
obtained by integrating the luminosity function, are very sensitive to the faint end
extrapolation in their highest redshift bin.
5.3
The Build-up of the Stellar Mass
The primary goal of measuring the stellar mass density is to determine how rapidly
the universe assembled its stars. At z ∼ 2−3, our results indicate that the universe
only contained ∼ 10% of the current stellar mass, regardless of whether we refer
> 1.4 × 1010 h−2
only to galaxies at Lrest
70 L¯ or whether we use the total mass
V
estimates of other authors. The galaxy population in the HDF-S was rich and
diverse at z > 2, but even so it was far from finished in its build-up of stellar
mass. By z ∼ 1, however, the total mass density had increased to roughly half its
local value, indicating that the epoch of 1 < z < 2 was an important period in the
stellar mass build-up of the universe.
A successful model of galaxy formation must not only explain our global results,
but also reconcile them with the observed properties of individual galaxies at all
redshifts. For example, a population of galaxies at z ∼ 1 − 1.5 has been discovered
(the so called extremely red objects or EROs), roughly half of which can be fit
with formation redshifts higher than 2.4 (Cimatti et al. 2002) and nearly passive
stellar evolution thereafter. Our results, which show that the universe contained
only ∼ 10% as many stars at z ∼ 2 − 3 as today would seem to indicate that
any population of galaxies which formed most of its mass at z & 2 can at most
contribute ∼ 10% of the present day stellar mass density. At z ∼ 1 − 1.5, where
the EROs reside, the universe had assembled roughly half of its current stars.
Therefore, this would imply that the old EROs contribute about ∼ 20% of the
4 The rest-frame optical luminosity density, color, and stellar mass density of the
92
universe from z=0 to z=3
mass budget at their epoch. Likewise, it should be true that a large fraction of the
stellar mass at low redshift should reside in objects with mass weighted stellar ages
corresponding to a formation redshift of 1 < z < 2. In support of this, Hogg et al.
(2002) recently have shown that ∼ 40% of the local luminosity density at 0.7µm,
and perhaps ∼ 50% of the stellar mass comes from centrally concentrated, high
surface brightness galaxies which have red colors. In agreement with the Hogg et
al. (2002) results, Bell et al. (2003a) and Kauffmann et al. (2003) also found that
∼ 50 − 75% of the local stellar mass density resides in early type galaxies. Hogg et
al. (2002) suggest that their red galaxies would have been formed at z & 1, fully
consistent with our results for the rapid mass growth of the universe during this
period.
6
Summary & Conclusions
In this paper we presented the globally averaged rest-frame optical properties
of a Ks -band selected sample of galaxies with z < 3.2 in the HDF-S. Using
our very deep 0.3 − 2.2µm, seven band photometry taken as part of the FIRE
Survey we estimated accurate photometric redshifts and rest-frame luminosities
tot
< 25 and used these luminosity estimates to measure
for all galaxies with Ks,AB
the rest-frame optical luminosity density jλrest , the globally averaged rest-frame
>
optical color, and the stellar mass density for all galaxies at z < 3.2 with L rest
V
1.4 × 1010 h−2
70 L¯ . By selecting galaxies in the rest-frame V -band, we selected
them in a way much less biased by star formation and dust than the traditional
selection in the rest-frame UV and much closer to a selection by stellar mass.
We have shown that jλrest in all three bands rises out to z ∼ 3 by factors of
4.9±1.0, 2.9±0.6, and 1.9±0.4 in the U , B , and V -bands respectively. Modeling
this increase in jλrest as an increase in L∗ of the local luminosity function, we derive
that L∗ must have brightened by a factor of 1.7 in the rest-frame V -band.
Using our jλrest estimates we calculate the (U − B)rest and (B − V )rest colors
of all the visible stars in galaxies with Lrest
> 1.4 × 1010 h−2
70 L¯ . Using the
V
COMBO-17 data we have shown that the mean color is much less sensitive to
density fluctuations and field-to-field variations than either jλrest or ρ∗ . Because
of their stability, integrated color measurements are ideal for constraining galaxy
evolution models. The luminosity weighted mean colors lie close to the locus of
morphologically normal local galaxy colors defined by Larson & Tinsley (1978).
The mean colors monotonically bluen with increasing redshift by 0.55 and 0.46
magnitudes in (U − B)rest and (B − V )rest respectively out to z ∼ 3. We interpret
this color change primarily as a change in the mean stellar age. The joint colors can
be roughly matched by simple SFH models if modest amounts of reddening (E(B −
V ) < 0.35) are applied. In detail, the redshift dependence of (U − B)rest and
(B − V )rest cannot be matched exactly by the simple models, assuming a constant
reddening and constant metallicity. However, we show that the models can still be
used, even in the face of these small disagreements, to robustly predict the stellar
4.6 Summary & Conclusions
93
mass-to-light ratios M/L∗V of the integrated cosmic stellar population implied by
our mean rest-frame colors. Variations in the metallicity does not strongly affect
this relation and it holds for a variety of smooth SFHs. Even the IMF only affects
the normalization of this relation, not its slope, assuming that the IMF everywhere
is the same. The reddening, which moves objects roughly along this relation is,
however, a large source of uncertainty. Using these M/L∗V estimates coupled with
our jλrest measurements, we derive the stellar mass density ρ∗ . These globally
averaged estimates of the mass density are more reliable than those obtained from
the mean of individual galaxies determined using smooth SFHs, primarily because
the cosmic mean SFH is plausibly much better approximated as being smooth,
whereas the SFHs of individual galaxies are almost definitely not.
The stellar mass density, ρ∗ , increases monotonically with increasing cosmic
time to come into good agreement with the other measured values at z . 1 with
a factor of ∼ 10 increase from z ∼ 3 to the present day. Within the random
uncertainties, our results agree well with those of Dickinson, Papovich, Ferguson,
& Budavári (2003) in the HDF-N although our ρ∗ estimates are systematically
higher than in the HDF-N. Taken together, the HDF-N and HDF-S paint a picture
in which only ∼ 5 − 15% of the present day stellar mass was formed by z ∼
2. By z ∼ 1, however, the stellar mass density had increased to ∼ 50% of its
present value, implying that a large fraction of the stellar mass in the universe
today was assembled at 1 < z < 2. Our ρ∗ estimates slightly underpredict the
stellar mass density derived from the integral of the SFR(z) curve at z > 2. A
resolution of the small apparent discrepancy between different fields, and between
the predictions from optical observations will in part require deeper NIR data, to
probe further down the mass function, and wider fields with multiple pointings
to control the effects of cosmic variance. In addition, large amounts of followup optical/NIR spectroscopy are required to help control systematic effects in
the zphot estimates. The 25 square arcminute MS1054-03 data taken as part of
FIRES and the ACS/ISAAC GOODS observations of the CDF-S region will be
very helpful for such studies. Observations with SIRTF will also improve the
situation by accessing the rest-frame NIR, where obscuration by dust becomes
much less important. Finally, systematics in the M/L∗ estimates may exist because
of a lack of constraint on the faint end slope of the stellar IMF.
We still have to reconcile global measurements of the galaxy population with
what we know about the ages and SFHs of individual galaxies. Our globally
determined quantities are quite stable and may serve as robust constraints on
theoretical models, which must correctly model the global build-up of stellar mass
in addition to matching the detailed properties of the galaxy population.
Acknowledgments
GR would like to thank Jarle Brinchmann and Frank van den Bosch for useful
discussions in the process of writing this paper, Christian Wolf for providing ad-
4 The rest-frame optical luminosity density, color, and stellar mass density of the
94
universe from z=0 to z=3
ditional COMBO-17 data products, and Eric Bell and Marcin Sawicki for giving
comments on an earlier version of the paper. GR would also like to acknowledge
the generous travel support of the Lorentz center and the Leids Kerkhoven-Bosscha
Fonds, and the financial support of Sonderforschungsbereich 375.
Funding for the creation and distribution of the SDSS Archive has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Aeronautics and Space Administration, the National Science Foundation,
the U.S. Department of Energy, the Japanese Monbukagakusho, and the Max
Planck Society. The SDSS Web site is http://www.sdss.org/.
The SDSS is managed by the Astrophysical Research Consortium (ARC) for
the Participating Institutions. The Participating Institutions are The University
of Chicago, Fermilab, the Institute for Advanced Study, the Japan Participation Group, The Johns Hopkins University, Los Alamos National Laboratory, the
Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, University of Pittsburgh, Princeton
University, the United States Naval Observatory, and the University of Washington.
References
Arnouts, S. et al. 2002, MNRAS, 329, 355
Baldry, I. K. et al. 2002, ApJ, 569, 582
Barger, A. J., Cowie, L. L., & Richards, E. A. 2000, AJ, 119, 2092
Bell, E. F. & de Jong, R. S. 2001, ApJ, 550, 212
Bell, E. F., McIntosh, D. H., Katz, N., & Weinberg, M. D. 2003, ApJ, 585, L117
Bell, E. F., Wolf, C., Meisenheimer, K., Rix, H.-W., Borch, A., Dye, S., Kleinheinrich,
K., McIntosh, D. H. 2003, submitted to ApJ, astro-ph/0303394
Benı́tez, N., Broadhurst, T., Bouwens, R., Silk, J., & Rosati, P. 1999, ApJ, 515, L65
Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393
Bessell, M. S. & Brett, J. M. 1988, PASP, 100, 1134
Bessell, M. S. 1990, PASP, 102, 1181
Blanton, M. R. et al. 2001, AJ, 121, 2358
Blanton, M. R. et al. 2003a, ApJ, in press (astro-ph/0209479)
Blanton, M. R. et al. 2003b, AJ, 125, 2348
Blanton, M. R. et al. 2003c, ApJ, 592, 819
Bolzonella, M., Pelló, R., & Maccagni, D. 2002, A&A, 395, 443
Brinchmann, J. & Ellis, R. S. 2000, ApJ, 536, L77
Bruzual A., G. & Charlot, S. 1993, ApJ, 405, 538
Calzetti, D., Armus, L., Bohlin, R. C., Kinney, A. L., Koornneef, J., & Storchi-Bergmann,
T. 2000, ApJ, 533, 682
Casertano, S. et al. 2000, AJ, 120, 2747
Charlot, S. & Longhetti, M. 2001, MNRAS, 323, 887
Chen, H. et al. 2003, ApJ, 586, 745
Cimatti, A. et al. 2002, A&A, 381, L68
Cohen, J. G., Hogg, D. W., Blandford, R., Cowie, L. L., Hu, E., Songaila, A., Shopbell,
P., & Richberg, K. 2000, ApJ, 538, 29
4.6 Summary & Conclusions
95
Cole, S. et al. 2001, MNRAS, 326, 255
Coleman, G. D., Wu, C. -C., & Weedman, D. W. 1980, ApJS, 43, 393
Colless, M. et al. 2001, MNRAS, 328, 1039
Cowie, L. L., Songaila, A., & Barger, A. J. 1999, AJ, 118, 603
Cox, A. N. 2000, Allen’s astrophysical quantities, (4th ed. New York: AIP Press; Springer)
Cristiani, S. et al. 2000, A&A, 359, 489
Daddi, E., Cimatti, A., Pozzetti, L., Hoekstra, H., Röttgering, H. J. A., Renzini, A.,
Zamorani, G., & Mannucci, F. 2000, A&A, 361, 535
Daddi, E. et al. 2003, ApJ, 588, 50
Dickinson, M., Papovich, C., Ferguson, H. C., & Budavári, T. 2003, ApJ, 587, 25
Eisenhardt, P. & Dickinson, M. 1992, ApJ, 399, L47
Erb, D. K., Shapley, A. E., Steidel, C. C., Pettini, M., Adelberger, K. L., Hunt, M. P.,
Moorwood, A. F. M., & Cuby, J. 2003, ApJ, 591, 101
Ferguson, H. C., Dickinson, M., & Papovich, C. 2002, ApJ, 569, L65
Fernández-Soto, A., Lanzetta, K. M., & Yahil, A. 1999, ApJ, 513, 34
Folkes, S. et al. 1999, MNRAS, 308, 459
Förster Schreiber, N. et al. in preparation
Fontana, A., D’Odorico, S., Poli, F., Giallongo, E., Arnouts, S., Cristiani, S., Moorwood,
A., & Saracco, P. 2000, AJ, 120, 2206
Fontana, A. et al. 2003, ApJ, 594, L9
Franx, M. et al. 2003, ApJ, 587, L79
Frenk, C. S., White, S. D. M., Efstathiou, G., & Davis, M. 1985, Nature, 317, 595
Fried, J. W. et al. 2001, A&A, 367, 788
Fukugita, M., Shimasaku, K., & Ichikawa, T. 1995, PASP, 107, 945
Fukugita, M., Ichikawa, T., Gunn, J. E., Doi, M., Shimasaku, K., & Schneider, D. P.
1996, AJ, 111, 1748
Giavalisco, M., Steidel, C. C., Adelberger, K. L., Dickinson, M. E., Pettini, M., & Kellogg,
M. 1998, ApJ, 503, 543
Giavalisco, M. & Dickinson, M. 2001, ApJ, 550, 177
Glazebrook, K. et al. 2003, in preparation
Glazebrook, K. et al. 2003, ApJ, 587, 55
Hauschildt, P. H., Allard, F., & Baron, E. 1999, ApJ, 512, 377
Hogg, D. W. et al. 2002, AJ, 124, 646
Huchra, J. P. 1977, ApJ, 217, 928
Jansen, R. A., Franx, M., Fabricant, D., & Caldwell, N. 2000, ApJS, 126, 271
Jansen, R. A., Fabricant, D., Franx, M., & Caldwell, N. 2000, ApJS, 126, 331
Kashikawa, N. et al. 2003, AJ, 125, 53
Kauffmann, G., White, S. D. M., & Guiderdoni, B. 1993, MNRAS, 264, 201
Kauffmann, G., Colberg, J. M., Diaferio, A., & White, S. D. M. 1999, MNRAS, 303, 188
Kauffmann, G. et al. 2003, MNRAS, 341, 33
Kinney, A. L., Calzetti, D., Bohlin, R. C., McQuade, K., Storchi-Bergmann, T., &
Schmitt, H. R. 1996, ApJ, 467, 38
Labbé, I. et al. 2003, AJ, 125, 1107
Lanzetta, K. M., Yahata, N., Pascarelle, S., Chen, H., & Fernández-Soto, A. 2002, ApJ,
570, 492
Larson, R. B. & Tinsley, B. M. 1978, ApJ, 219, 46
Lilly, S. J., Le Fevre, O., Crampton, D., Hammer, F., & Tresse, L. 1995, ApJ, 455, 50
Lilly, S. J., Le Fevre, O., Hammer, F., & Crampton, D. 1996, ApJ, 460, L1
Madau, P. 1995, ApJ, 441, 18
4 The rest-frame optical luminosity density, color, and stellar mass density of the
96
universe from z=0 to z=3
Madau, P., Pozzetti, L., & Dickinson, M. 1998, ApJ, 498, 106
Madau, P., Ferguson, H. C., Dickinson, M. E., Giavalisco, M., Steidel, C. C., & Fruchter,
A. 1996, MNRAS, 283, 1388
McCarthy, P. J. et al. 2001, ApJ, 560, L131
Moorwood, A. F. 1997, Proc. SPIE, 2871, 1146
Norberg, P. et al. 2002, MNRAS, 336, 907
Papovich, C., Dickinson, M., Ferguson, H. C., 2001, ApJ, 559, 620
Pei, Y. C., Fall, S. M., & Hauser, M. G. 1999, ApJ, 522, 604
Pettini, M., Steidel, C. C., Adelberger, K. L., Dickinson, M., & Giavalisco, M. 2000, ApJ,
528, 96
Pettini, M., Shapley, A. E., Steidel, C. C., Cuby, J., Dickinson, M., Moorwood, A. F.
M., Adelberger, K. L., & Giavalisco, M. 2001, ApJ, 554, 981
Pettini, M., Rix, S. A., Steidel, C. C., Adelberger, K. L., Hunt, M. P., & Shapley, A. E.
2002, ApJ, 569, 742
Poli, F., Menci, N., Giallongo, E., Fontana, A., Cristiani, S., & D’Odorico, S. 2001, ApJ,
551, L45
Poli, F. et al. 2003, ApJ, 593, L1
Rigopoulou, D. et al. 2000, ApJ, 537, L85
Rudnick, G. et al. 2003, in preparation
Rudnick, G., Franx, M., Rix, H.-W., Moorwood, A., Kuijken, K., van Starkenburg, L.,
van der Werf, P., Röttgering, H., Labbé, I. 2001, AJ, 122, 2205
Salpeter, E. E. 1955, ApJ, 121, 161
Sawicki, M. J., Lin, H., & Yee, H. K. C. 1997, AJ, 113, 1
Scalo, J. M. 1986, Fundamentals of Cosmic Physics, 11, 1
Schade, D. et al. 1999, ApJ, 525, 31
Seaton, M. J. 1979, MNRAS, 187, 73P
Shapley, A. E., Steidel, C. C., Adelberger, K. L., Dickinson, M., Giavalisco, M., & Pettini,
M. 2001, ApJ, 562, 95
Skrutskie, M. F. et al. 1997, ASSL Vol. 210: The Impact of Large Scale Near-IR Sky
Surveys, 25
Somerville, R. S., Primack, J. R., & Faber, S. M. 2001, MNRAS, 320, 504
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K. L. 1996, ApJ,
462, L17
Steidel, C. C., Adqelberger, K. L., Giavalisco, M., Dickinson, M., & Pettini, M. 1999,
ApJ, 519, 1
Stoughton, C. et al. 2002, AJ, 123, 485
Strauss, M. A. et al. 2002, AJ, 124, 1810
Thompson, R. I., Weymann, R. J., & Storrie-Lombardi, L. J. 2001, ApJ, 546, 694
Thompson, R. I. 2003, ApJin press, astro-ph/0306594
van de Ven, G., van Dokkum, P. G., & Franx, M. 2003, submitted to MNRAS, astroph/0211566
van Dokkum, P. G., Franx, M., Kelson, D. D., & Illingworth, G. D. 2001, ApJ, 553, L39
van Dokkum, P. G. et al. 2003, ApJ, 587, L83
Vanzella, E. et al. 2002, A&A, 396, 847
Williams, R. E. et al. 1996, AJ, 112, 1335
Wilson, G., Cowie, L. L., Barger, A. J., & Burke, D. J. 2002, AJ, 124, 1258
Wolf, C., Meisenheimer, K., Rix, H.-W., Borch, A., Dye, S., & Kleinheinrich, M. 2003,
A&A, 401, 73
York, D. G. et al. 2000, AJ, 120, 1579
Appendix A. Derivation of zphot Uncertainty
97
Zepf, S. E. 1997, Nature, 390, 377
Appendix A.
Derivation of zphot Uncertainty
Given a set of formal flux errors, one way to broaden the redshift confidence interval
without degrading the accuracy (as noticed in R01) is to lower the absolute χ 2 of
every χ2 (z) curve without changing its shape (or the location of the minimum).
By scaling up all the flux errors by a constant factor, we can retain the relative
weights of the points in the χ2 without changing the best fit redshift and SED, but
we do enlarge the redshift interval over which the templates can satisfactorily fit
the flux points. Since we believe the disagreement between zspec and zphot is due
to our finite and incomplete template set, this factor should reflect the degree of
template mismatch in our sample, i.e., the degree by which our models fail to fit
the flux points. To estimate this factor we first compute the fractional difference
between the model and the data ∆i,j for the jth galaxy in the ith filter,
∆i,j =
dat
mod
− fi,j
)
(fi,j
dat
fi,j
(A.1)
where f mod are the predicted fluxes of the best-fit template combination and f dat
are our actual data. For each galaxy we calculated
v
u
N
filt
u
X
1
∆2
(A.2)
∆j = t
Nfilt − 1 i=2 i,j
where we have ignored the contributions of the U -band. While the U -band is
important in finding breaks in the SEDs, the exact shapes of the templates are
poorly constrained blueward of the rest-frame U -band and the U -data often
deviates significantly from the best-fit model fluxes.
To determine the mean deviation of all of the flux points from the model ∆ dev
tot
< 22
we then averaged over all galaxies in our complete FIRES sample with Ks,AB
(for which the systematic zphot errors should dominate over those resulting from
photometric errors) to obtain
∆dev =
Ngal
1 X
|∆j | .
Ngal j=1
(A.3)
We find ∆dev = 0.08, which includes both random and systematic deviations from
the model. We modified the Monte-Carlo simulation of R01 by calculating, for
each object j,
v
³ ´
uP
u Nfilt fi 2
¿ À
t i=2 δfi
S
=
(A.4)
N j
Nfilt − 1
4 The rest-frame optical luminosity density, color, and stellar mass density of the
98
universe from z=0 to z=3
again excluding the U -band. We then scaled the flux errors, for each object, using
the following criteria:




δfi
0
À
¿
δfi =
S


 δfi ∆dev
N j
:
:
¿ À
1
S
≤
∆dev
¿ N Àj
.
S
1
>
N j
∆dev
(A.5)
The photometric redshift error probability distribution is computed using the δf i0 ’s.
Note that this procedure will not modify the zphot errors of the objects with low
S/N where the zphot errors are dominated by the formal photometric errors. The
resulting probability distribution is highly non-Gaussian and using it we calculate
hi
low
the upper and lower 68% confidence limits on the redshift zphot
and zphot
respectively. As a single number which encodes the total range of acceptable z phot ’s, we
hi
low
define δzphot ≡ 0.5 ∗ (zphot
− zphot
).
Figure 6 from L03 shows the comparison of zphot to zspec . For these bright
galaxies, it is remarkable that our new photometric redshift errorbars come so close
to predicting the difference between zphot and the true value. Some galaxies have
large δzphot values even when the local χ2 minimum is well defined because there
is another χ2 minimum of comparable depth that is contained in the 68% redshift
confidence limits. There are galaxies with δzphot < 0.05. Some of these are bright
low redshift galaxies with large rest-frame optical breaks, which presumably place
a strong constraint on the allowed redshift. Many of these galaxies, however, are
faint
δzphot is unrealistically low. Even though these faint galaxies have
­ S ® and the
1
,
≤
N j
∆dev they still can have high S/N in the B450 or V606 bandpasses and
hence have steep χ2 curves and small inferred redshift uncertainties. In addition,
many of these galaxies have zphot > 2 and very blue continuum longward of Lyα .
The imposed sharp discontinuity in the template SEDs at the onset of HI absorption causes a very narrow minimum in the χ2 (z) curve, and hence a small δzphot ,
but likely differs from the true shape of the discontinuity because we use the mean
opacity values of Madau (1995), neglecting its variance among different lines of
sight.
It is difficult to develop a scheme for measuring realistic photometric redshift
uncertainties over all regimes. The δzphot estimate derives the zphot uncertainties
individually for each object, but can underpredict the uncertainties in some cases.
Compared to the technique of R01 however, a method based completely on the
Monte-Carlo technique is preferable because it has a straightforwardly computed
redshift probability function. This trait is desirable for estimating the errors in
the rest-frame luminosities and colors and for this reason we will use δz phot as our
uncertainty estimate in this paper.
Appendix B. Rest-Frame Photometric System
Appendix B.
99
Rest-Frame Photometric System
To define the rest-frame U , B , and V fluxes we use the filter transmission curves
and zeropoints tabulated in Bessell (1990), specifically his U X, B, and V filters.
The Bessell zeropoints are given as magnitude offsets with respect to a source
which has constant fν and AB = 0. The AB magnitude is defined as
ABν = −2.5 ∗ log10 hfν i − 48.58
(B.1)
where hfν i is the flux fν (ν) observed through a filter T (ν) and in units of
ergs s−1 cm−2 Hz −1 . Given the zeropoint offset ZPν for a given filter, the Vega
magnitude mν is then
mν = ABν − ZPν = −2.5 ∗ log10 hfν i − 48.58 − ZPν .
(B.2)
All of our observed fluxes and rest-frame template fluxes are expressed in f λ . To
obtain rest-frame magnitudes in the Bessell (1990) system, we must calculate the
conversion from fλ to fν for the redshifted rest-frame filter set. The flux density
of an SED with fλ (λ) integrated through a given filter with transmission curve
T (λ) is
R
fλ (λ0 )T 0 (λ0 )dλ0
R
(B.3)
hfλ i =
T 0 (λ0 )dλ0
or
R
hfν i =
Since
Z
0
0
0
fν (ν 0 )T 0 (ν 0 )dν 0
R
.
T 0 (ν 0 )dν 0
0
fλ (λ )T (λ )dλ =
Z
fν (ν 0 )T 0 (ν 0 )dν 0
(B.4)
(B.5)
we can convert to hfν i through
R 0 0
T (λ )dλ0
hfν i = hfλ i ∗ R 0 0
T (ν )dν 0
(B.6)
and use hfν i to calculate the apparent rest-frame Vega magnitude through the
redshifted filter via Eq. B.2.
Appendix C.
Estimating Rest-Frame Luminosities
We derive for any given redshift, the relation between the apparent AB magnitude
mλz of a galaxy through a redshifted rest-frame filter, its observed fluxes hfλi ,obs i
in the different filters i, and the colors of the spectral templates. At redshift z, the
rest-frame filter with effective wavelength λrest has been shifted to an observed
wavelength
λz = λrest × (1 + z)
(C.1)
4 The rest-frame optical luminosity density, color, and stellar mass density of the
100
universe from z=0 to z=3
and we define the adjacent observed bandpasses with effective wavelengths λ l and
λh which satisfy
λl < λ z ≤ λ h .
(C.2)
We now define
Cobs ≡ mobs,λl − mobs,λh
(C.3)
Ctempl ≡ mtempl,λl − mtempl,λh ,
(C.4)
where mobs,λl and mobs,λh are the AB magnitudes which correspond to the fluxes
hfλl ,obs i and hfλh ,obs i respectively. We then shift each template in wavelength to
the redshift z and compute,
where mtempl,λl and mtempl,λh are the AB magnitudes through the λl and λh
observed bandpasses (including the atmospheric and instrument throughputs).
We sort the templates by their Ctempl values, Ctempl,a , Ctempl,b , etc., and find the
two templates such that
Ctempl,a ≤ Cobs < Ctempl,b .
(C.5)
We then define for the ath template
(C.6)
Cλl ,z,a ≡ mtempl,λl − mtempl,λz
th
where mtempl,λz is the apparent AB magnitude of the redshifted a template
through the redshifted λrest filter. We point out that because our computations
always involve colors, they are not dependent on the actual template normalization
(which cancels out in the difference). Taking our observed color Cobs and the
templates with adjacent “observed” colors Ctempl,a and Ctempl,b , we can interpolate
between Cλl ,z,a and Cλl ,z,b
¶
µ
Cλl ,z,b − Cλl ,z,a
(C.7)
mobs,λl − mλz = Cλl ,z,a + (Cobs − Ctempl,a ) ×
Ctempl,b − Ctempl,a
and solve for mλz .
When Cobs lies outside the range of the Ctempl ’s, we simply take the two nearest
templates in observed Ctempl space and extrapolate Eq. C.7 to compute mλz .
Equation C.7 has the feature that mλz ≈ mobs,λl when λz = λl (and hence
when Cλl ,z,a and Cλl ,z,b ≈ 0). While this method still assumes that the templates
are reasonably good approximations to the true shape of the SEDs it has the
advantage that it does not rely on exact agreement. Galaxies whose observed
colors fall outside the range of the templates can also be easily flagged. A final
advantage of this method is that the uncertainty in mλz can be readily calculated
from the errors in the observed fluxes.
From mλz , we compute the rest-frame luminosity by applying the K-correction
and converting to luminosity units
µ
¶2
Lrest
DL
= 10−0.4(mλz − M¯,λrest − ZPλrest ) ×
× (1 + z)−1 × h−2 (C.8)
L¯
10pc
Appendix C. Estimating Rest-Frame Luminosities
101
where M¯,λrest is the absolute magnitude of the sun in the λrest filter (M¯,U = +5.66,
M¯,B = +5.47, and M¯,V = +4.82 in Vega magnitudes; Cox 2000), ZPλrest is the
zeropoint in that filter (as in Eq. B.2 but expressed at λ and not at ν), and DL
is the distance modulus in parsecs. Following R01, we correct this luminosity by
the ratio of the Kstot flux to the modified isophotal aperture flux (see L03). This
adjustment factor, which accounts both for the larger size of the total aperture and
the aperture correction, changes with apparent magnitude and it ranges from 1.23,
tot
tot
at 20 <Ks,AB
≤ 24, to 1.69, at 24 <Ks,AB
≤ 25, and it has an RMS dispersions
of 0.17 and 0.49 in the two magnitude bins respectively.
The uncertainty in the derived Lrest
has contributions both from the observaλ
tional flux errors and from the redshift uncertainty, which causes λz to move with
respect to the observed filters. The first effect is estimated by propagating the observed flux errors through Eq. C.7. As an example, object 531 at zphot =2.20 has
tot
Ks,AB
= 24.91 and signal-to-noise in the Ks -band of 8.99 and 5.43 in our modified isophotal and total apertures respectively. The resultant error in Lrest
purely
V
tot
≈ 24, the typical signal-to-noise in the
from flux errors is then 26%. At Ks,AB
Ks -band increases to ≈ 13 and ≈ 6.3 in our modified isophotal and total apertures
respectively and the error Lrest
decreases accordingly.
λ
To account for the redshift dependent error in the calculated luminosity, we
use the Monte-Carlo simulation described first in R01 and updated in §Appendix
A.. For each Monte-Carlo iteration we calculate the rest-frame luminosities and
determine the 68% confidence limits of the resulting distribution. The 68% concan be highly asymmetric, just as for zphot . For objects
fidence limits in Lrest
λ
tot
with Ks,AB
. 25 we find that the contributions to the total Lrest
error budget are
λ
dominated by the redshift errors rather than by the flux errors.
CHAPTER
FIVE
Field Galaxies at 1 < z < 3
The Color Magnitude Distribution of
the evolution and modeling of the blue sequence
ABSTRACT
We use deep near-infrared VLT/ISAAC imaging to study the rest-frame
color-magnitude distribution of infrared selected galaxies in the redshift
range 1 < z < 3. We find a well-defined blue peak of star-forming galaxies at
all redshifts. The blue galaxies populate a color-magnitude relation (CMR),
such that more luminous galaxies in the rest-frame V -band tend to have
redder ultraviolet-to-optical colors. The slope of the CMR does not evolve
with time, and is similar to the slope of blue, late-type galaxies in the local
universe. Analysis of spectra of nearby late-type galaxies suggests that the
steepness of the slope can be fully explained by the observed correlation of
dust content with optical luminosity. The zeropoint of the blue CMR at a
given magnitude reddens smoothly from z = 3 to z = 0, likely reflecting an
increase of the mean stellar age and an increase in the mean dust opacity of
blue-sequence galaxies. A distinct feature is that the color scatter around the
z ∼ 3 CMR is asymmetric, with a blue “ridge” and a skew towards red colors. We have explored which types of star formation histories can reproduce
the scatter and the skewed shape of the color distribution. These included
models with constant star formation rates and sudden cutoffs, exponentially
declining star formation rates, burst models, and models with episodic star
formation. The episodic models reproduced the color distribution best, with
quiescent periods lasting 30-50% of the length of an active period, and duration of the duty cycle between 150 Myr to 1 Gyr. The episodic star formation
in these models rejuvenate the galaxies during each episode, making it significantly bluer than a galaxy with constant star formation of the same age.
This could be a solution of the enigmatic observation that z = 3 galaxies are
much bluer than expected if they were as old as the universe. Finally, the
color distribution has a strong tail of very red galaxies. The relative number
of red galaxies increases sharply from z ∼ 3 to z ∼ 1. The rest-frame V band luminosity density in luminous blue-sequence galaxies is constant, or
decreases, whereas that in red galaxies rises with time. We are viewing the
progressive formation of red, passively evolving galaxies.
Ivo Labbé, Marijn Franx, Gregory Rudnick, Natascha M. Förster Schreiber,
Emanuele Daddi, Pieter G. van Dokkum, Konrad Kuijken, Alan Moorwood,
Hans-Walter Rix, Huub Röttgering, Ignacio Trujillo, Arjen van der Wel, Paul van
der Werf, & Lottie van Starkenburg
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
104
and modeling of the blue sequence
1
Introduction
T
he formation and evolution of galaxies is a complex process, involving the
growth of dark-matter structure from the gradual hierarchical merging of
smaller fragments (e.g., White & Frenk 1991; Kauffmann & White 1993), the
accretion of gas, the formation of stars, and feedback from supernovae and black
holes. Presently, the theories describing the growth of large-scale dark-matter
structure are thought to be well-constrained (Freedman et al. 2001; Efstathiou et
al. 2002; Spergel et al. 2003), but the formation history of the stars inside the
dark-matter halos is still poorly understood. Neither hydrodynamical simulations
(e.g., Katz & Gunn 1991; Springel et.al. 2001; Steinmetz & Navarro 2002) nor
state-of-the-art semianalytic models (e.g., Kauffmann, White, & Guiderdoni 1993;
Somerville & Primack 1999; Cole, Lacey, Baugh, & Frenk 2000) provide unique
predictions for the formation of the stars in galaxies, given the large parameter
space available to these models. Direct observations are critical to constrain them.
Specifically the observations of massive galaxies provide strong tests for pictures
of galaxy formation, as their build-up can be directly observed from high redshift
to the present epoch. The largest samples of high-redshift galaxies to date have
been selected by their rest-frame UV light, through the Lyman Break technique
(LBGs; Steidel et al. 1996a,b, 2003). Unfortunately, the rest-frame UV light is
highly susceptible to dust and not a good measure of the number of intermediate
and low mass stars, which may dominate the stellar mass. In fact, rest-frame
optical observations have shown LBGs to be relatively low mass (M ∼ 1010 M¯ ),
unobscured, star-forming galaxies (e.g., Papovich, Dickinson, & Ferguson 2001;
Shapley et al. 2001).
The rest-frame optical light is already much less sensitive to the effects of dust
obscuration and on-going star formation than the UV, and it is expected to be a
better tracer of stellar mass. Recent advances in near-infrared (NIR) capabilities
on large telescopes are now making it possible to select statistically meaningful
samples of galaxies by their rest-frame optical light out to z ∼ 3. In this context
we started the Faint Infrared Extragalactic Survey (FIRES; Franx et al. 2000), a
deep optical-to-infrared multicolor survey of NIR-selected galaxies. The rest-frame
optical observations are also useful as much of our knowledge in the local universe
is based on studies at these wavelengths.
A particular diagnostic of galaxy formation is the shape of the galaxy luminosity and mass functions as a function of the spectral type. In the local universe
these distributions have been determined in great detail (e.g., Strateva et al. 2001;
Norberg et al. 2002; Blanton et al. 2003; Kauffmann et al 2003). While a precise characterisation of the local distributions can strongly constrain the models,
powerful tests can also be made using their evolution as a function of redshift.
For example, Bell et al. (2004a,2004b) show that the V -band luminosity density of photometrically selected early-type galaxies is nearly constant in the range
0.2 < z < 1.1, suggesting an increase in stellar mass of passively evolving galaxies
5.1 Introduction
105
over this range. This is not expected from the simplest collapse models, where all
galaxies form at high redshift and evolve passively to the present-day.
Different constraints on galaxy formation come from correlations between integrated galaxy properties. For example, morphologically early-type galaxies in the
local universe and distant clusters populate a well-defined color-magnitude relation (CMR), which is thought to reflect a sequence of increasing metallicity with
stellar mass (e.g., Bower, Lucey, & Ellis 1992; Schweizer & Seitzer 1992; Kodama
& Arimoto 1997). The small scatter of this relation implies that early type galaxies
formed most of their stars at high redshift (e.g., van Dokkum et al. 1998). Blue,
star forming, late-type galaxies also display a color-magnitude relation (Chester &
Roberts 1964; Visvanathan 1981; Tully, Mould, & Aaronson 1982), but the scatter
is larger (Griersmith 1980), and its origin is probably more complex; it has been
interpreted as a sequence in the mean stellar age (Peletier & de Grijs 1998), dust
attenuation (Tully et al. 1998), and/or metallicity (Zaritsky, Kennicutt, & Huchra
1994, Bell & De Jong 2000).
Interestingly, Papovich, Dickinson, & Ferguson (2001) found a color-magnitude
relation for blue, star-forming galaxies at z ∼ 3, from a sample of NIR-selected
galaxies in the Hubble Deep Field North (HDFN). The trend, which is also seen in
the similar field of the Hubble Deep Field South (Labbé et al. 2003), is such that
galaxies more luminous in the rest-frame V −band, tend to have redder ultravioletto-optical colors. A distinct aspect of the high-redshift CMR, is that the galaxies
are asymmetrically distributed around the relation, with a well-defined blue envelope (Papovich et al. 2004).
Studies of colors and magnitudes of star-forming galaxies at z > 2 also raised
questions. A particular puzzle was presented by modeling of their stellar populations, which implied average luminosity weighted ages of a few 100 Myr, much
younger than the age of the universe at these redshifts (e.g., Papovich, Dickinson,
& Ferguson 2001; Shapley et al. 2001). From this, and from the relative absence of
candidates for red, non star-forming galaxies in the HDFN it was suggested that
star formation in LBGs occurs with short duty cycles and a timescale between star
formation events of . 1 Gyr.
In this paper, we investigate the evolution of the rest-frame colors of galaxies
as a function of redshift in the range 1 . z . 3. The deep optical-to-NIR imaging
and the homogeneous photometry of the FIRES project is excellently suited for
such studies, and we use it to analyze rest-frame ultraviolet-to-optical colors and
magnitudes of a sample of 1475 Ks -band selected galaxies.
We focus on the galaxies that populate the blue peak of galaxies at low and
high redshift. We wish to understand the nature of the blue color-magnitude
relation, the evolution of the colors towards high redshift, and the origin of the
conspicious skewed color distribution around the CMR at z ∼ 3. For the latter we
explore models with different types of star formation histories that might produce
such distributions. Finally, we chart the evolution of the relative number of red
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
106
and modeling of the blue sequence
galaxies over the redshift range 1 . z . 3.
This paper is organized as follows. We present the data in §2, describe the color
magnitude distribution of FIRES galaxies in §3, analyze the blue color-magnitude
relation in §4, and model the scatter of galaxies around the blue CMR in §5.
Finally, §6 presents the evolution of the red galaxy fraction. Where necessary, we
adopt an ΩM = 0.3, ΩΛ = 0.7, and H0 = 70 km s−1 Mpc−1 cosmology. We use
magnitudes calibrated to models for Vega throughout.
2
2.1
The Data
The Observations and Sample Selection
The observations were obtained as part of the public Faint Infrared Extragalactic
Survey (FIRES; Franx et al. 2000) the deepest groundbased NIR survey to date.
We cover two fields with existing deep optical WFPC2 imaging from the Hubble
Space Telescope (HST): the WPFC2-field of HDFS, and the field around the z =
0.83 cluster MS1054-03. The observations, data reduction, and assembly of the
catalog source catalogs are presented in detail by Labbé et al. (2003) for the HDFS
and Förster Schreiber et al. (2004a) for the MS1054 field.
Briefly, we observed in the NIR Js , H, and Ks bands with the Infrared Spectrometer and Array Camera (ISAAC; Moorwood 1997) at the Very Large Telescope
(VLT). In the HDFS, a total of 101.5 hours was invested in a single 2.50 × 2.50
pointing, resulting in the deepest groundbased NIR imaging, and the deepest
K−band to date, even from space. We complemented the existing deep optical
HST WFPC2 imaging in the U300 , B450 , V606 , I814 bands (Casertano et al. 2000).
A further 77 hours of NIR imaging was spent on a mosaic of four ISAAC pointings centered on the z = 0.83 foreground cluster MS1054-03, reaching somewhat
shallower depths. We complemented the data with WFPC2 mosaics in the V 606
and I814 bands (van Dokkum et al. 2000), and collected additional imaging with
the VLT FORS1 instrument in the U, B, and V bands (Förster Schreiber et al.
2004a). In both surveyed fields the effective seeing in the final NIR images was
≈ 0.00 45 − 0.00 55 FWHM.
We detect objects in the Ks -band using version 2.2.2 of the SExtractor software
((Bertin & Arnouts 1996). For consistent photometry accross all bands, all images
were aligned, and accurately PSF-matched to the filter in which the image quality
was worst. Stellar curve of growth analysis indicates that the fraction of enclosed
flux agrees to better than 3% for the apertures relevant to our color measurements.
The color measurements were done in a customized isophotal aperture defined
from the Ks −image. The estimate of total flux in the Ks -band was computed
using SExtractors AU T O aperture for isolated sources, and in an adaptive circular
aperture for blended sources. In both cases, a minimal aperture correction for the
light lost by a point source was applied. Photometric uncertainties were derived
empirically from the flux distribution in apertures placed on empty parts of the
5.2 The Data
107
maps. For details concerning all aspects of the photometric measurements, see
Labbé et al. 2003. The total 5−σ limiting depth for point sources are Kstots = 23.8
for the HDFS, and 23.1 for the MS1054-field.
2.2
Photometric Redshifts and Rest-Frame Colors
Photometric redshifts were estimated by fitting a linear combination of redshifted
empirical galaxy spectra, and a 10 Myr old simple stellar population model (1999
version of Bruzual A. & Charlot 1993) to the observed flux points. The algorithm
is described in detail by Rudnick et al. (2001, 2003). We adopt a minimux flux
error of 5% for all bands to account for zeropoint uncertainties and for mismatches
between the observations and the photo-z template set.
Monte-Carlo simulations were used to estimate the errors δzph,M C on the photometric redshifts. These errors reflect the photometric uncertainties, template
mismatch, and the possibility of secondary solutions. We determined the accuray of the technique from comparisons to the available spectroscopy in each field.
We find δz =< |zspec − zphot |/(1 + zspec ) >= 0.07, and δz = 0.05 for sources at
z ≥ 2. The errors calculated from simulations δzph,M C are consistent with this.
We identified and removed stars using the method described in Rudnick et al.
(2003).
We combine the observed SEDs and photometric redshifts to derive rest-frame
rest
luminosties Lrest
that interpolates diλ . We used a method of estimating Lλ
rectly between the observed fluxes, using the templates as a guide. The rest-frame
are described extensively by
phometric system, and details on estimating Lrest
λ
(Rudnick et al. 2003). Throughout we will use the rest-frame U X, B, and V filters
Beers et al. (1990) and the HST/FOC F 140W, F 170W, and F 220W filters, which
we will call 1400, 1700, and 2200 throughout. We adopted the photometric system of Bessell (1990) for the optical filters, which was calibrated to the Dreiling
and Bell (1980) model spectrum for Vega. The HST/FOC UV zeropoints were
calibrated to the Kurucz (1992) model for Vega.
The rest-frame luminosities are sensitive to the uncertainties in the photometric
redshifts. Therefore we only analyze the sample of galaxies with δzph,M C /(1 +
zph ) < 0.2, keeping 1354 out of 1475 galaxies. The median δzph,M C /(1 + zph )
for the remaining sample is 0.05. We checked that the color distribution of the
rejected galaxies was consistent with that of the galaxies we kept.
The reduced images, photometric catalog, redshifts, and rest-frame luminosities
are all available on-line through the FIRES homepage1 .
1 http://www.strw.leidenuniv.nl/˜fires
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
108
and modeling of the blue sequence
Figure 1 — The rest-frame 2200 − V colors against absolute magnitude in the V −band for
galaxies in the field of the HDFS (a) and in the field of MS1054 (c). The two samples are
split into three redshift bins. The errorbars represent the 1σ uncertainties on the rest-frame
colors. The dotted line roughly marks a conservative color limit, corresponding to the 2σ flux
uncertainty in the observed filter that is closest to 2200-band at the maximum redshift of the
bin. Galaxies redward of this line have uncertain rest-frame colors, but can be observed. The
solid line shows a fit of a linear relation with a fixed slope of -0.17 to the galaxies in the blue peak
of the color-magnitude distribution. Panels b (HDFS) and d (MS1054) show histograms of the
colors after the slope is subtracted. The peak of this distribution is normalized to the intercept
of the CMR at MV = −21. We show the color distributions of all detected galaxies in each bin
(dark gray histograms) and those to a limiting magnitude of MV ≤ −19.5 and MV ≤ −20.5 in
the field of the HDFS and MS1054 respectively (light gray histograms).
3
The rest-frame Color-Magnitude Distribution of Galaxies
from z ∼ 1 to z ∼ 3
A well-studied diagnostic in the local universe is the optical color-magnitude distribution (e.g., Strateva et al. 2001; Hogg et al. 2002; Baldry et al. 2004). Quantifying this distribution out to higher redshifts, will provide strong constraints on
models of galaxy formation, which must reproduce these observations.
To quantify the evolution of galaxy colors and magnitude with redshift, we
compare the distribution at the rest-frame 2200 − V color versus absolute V −band
magnitude. We present evolution in terms of the 2200 − V color in panels a and
c of Figure 1. These colors are sensitive to variations in stellar population age,
as they span the Balmer and 4000 Å breaks, and attenuation by dust, by virtue
of the long wavelength baseline extending into the UV. We prefer the rest-frame
2200-band to the more commonly used U -band, because at z = 2 − 3 the 2200band is probed by our deepest optical observations. In addition, the uncertainties
in our photometric redshifts tend to increase the scatter more in the U and B
colors than in the 2200 colors as the former bands directly straddle the Balmer
5.4 The Color-Magnitude Relation of Blue Field Galaxies
109
and 4000 Å breaks.
We divide the galaxies in three redshift bins centered at z=1.0, 1.8, and z=2.7.
The redshift range is defined so that the rest-frame 2200 and V -band always lies
within the range of our observed filters. The number of redshift bins is a compromise between keeping a statistically meaningful sample at each redshift, while
reducing the effects of evolution over the time span in each bin. The limits of
the bins are defined so that galaxies in the smaller field of the HDFS are equally
divided in number. The redshift bins of the larger MS1054 field are matched to
those of the HDFS.
Clearly, in both fields and in all redshift bins there is a well-defined ridge of
blue galaxies. Furthermore, the ridge is tilted: the more luminous galaxies along
the ridge tend to have redder 2200 − V colors. This ridge is what we define as the
color magnitude relation of blue galaxies.
There is no reason to think that this color-magnitude relation is due to some
selection effect; we can easily detect galaxies that are bright in MV and blue in
2200 − V , and our photometry is sufficiently deep to ensure that the trend is not
caused by a lack of faint red galaxies. Only at at low redshifts (z ∼ 1) and faint
magnitudes may we miss some of the bluest galaxies, as the sample was selected
the observed Ks -filter, which is significantly redder than the redshifted V −band
at z ∼ 1.
It is striking that the blue color magnitude relation has a very well- defined
boundary to the blue, and a much more extended tail to the red. This red tail
extends up to 4 magnitudes, and defines a clear red color magnitude relation in
the lowest redshift bin in the MS1054 field. This is mainly caused by the colormagnitude relation of the cluster galaxies at z = 0.83, and is also observed in the
field up to z = 1 (e.g., Bell et al. 2004a)
We now proceed to analyze the properties of the blue color magnitude in sections 4 and 5, and will turn to the red galaxies in section 6.
4
The Color-Magnitude Relation of Blue Field Galaxies
Two key features of the blue CMR in Fig. 1 are that the peak of the distribution,
and hence the zeropoint of the blue sequence, reddens from z ∼ 3 to z ∼ 1, while on
first sight the slope does not seem to vary appreciably. We proceed by quantifying
the slope and zeropoint of the relation and their evolution with redshift.
4.1
The Slope and its Evolution
We assume that the distribution of colors and magnitudes can be characterized
by a linear relation. We obtain the best fitting slope as follows: for each value
of the slope we obtain a distribution of color residuals. We select the slope that
maximizes the peak of the histogram of the residuals. This so-called “mode re-
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
110
and modeling of the blue sequence
Figure 2 — The evolution of the slope of the blue color-magnitude relation from linear fits to
the galaxy distributions in the HDFS (filled circles) and MS1054 (diamonds). The results are
plotted at the median redshift of the galaxies in the bin. The uncertainties correspond the 68%
confidence interval obtained with bootstrap resampling. The data are consistent with a constant
slope with redshift. Also drawn are values for blue late type galaxies from the Nearby Field
Galaxy Survey (Jansen, Franx, & Fabricant 2000a), where we assume the CMR is caused by a
systematic trend with dust reddening (star), or stellar age (triangle). Finally, we show the value
for the metallicity-luminosity relation of local early type galaxies (square; Bower, Lucey, & Ellis
1992).
gression” technique is very insensitive to outliers. The histogram is not calculated
in discrete bins, but using a kernel density esimator with a small gaussian kernel
of σ = 0.2, comparable to the photometric uncertainty in the colors of individual galaxies. The results depend slightly on the smoothing parameter used but
these effects are small compared to the intrinsic uncertainty caused by the limited
number of galaxies in the sample.
These uncertainties were calculated by bootstrap resampling the color-magnitude
distributions 200 times, and repeating the fitting procedures. We selected the central 68% of the best-fit slopes as our confidence interval.
Figure 2 shows the values of the slope δ(2200 − V )/δMV versus redshift in
fields of the HDFS and MS1054. We have included an additional low-redshift bin
centered at z ∼ 0.5, also determined from our data. To compare to observations
at z = 0, we derived the 2200 − V slope from the U − V slope under three different
assumptions:
In the first case we assume the slope to result from a metallicity-luminosity sequence identical to that of nearby early-type galaxies, where more luminous galaxies are more metal-rich. We used observations of Coma cluster galaxies (Bower,
Lucey, & Ellis 1992) and converted the observed slope in U − V colors to 2200 − V
colors with the help of Bruzual & Charlot (2003) models. These models were constructed to match observed colors of the cluster galaxies on the red CM-relation,
5.4 The Color-Magnitude Relation of Blue Field Galaxies
111
Figure 3 — The relation between 2200 − V and U − V colors for Bruzual & Charlot(2003)
stellar population models with declining star formation rates and a range of timescales τ . Solar
metallicity (solid lines) and 1/3 solar models (dashed lines) are shown. The square indicates stellar ages of ∼1 Gyr, the diamonds indicate ages of ∼10 Gyr. Over a range of stellar population
ages and metallicities the relationship is tight for blue galaxies, allowing a fairly accurate transformation of their colors. Also drawn is a Calzetti et al. (2000) attenuation vector of A V = 1. In
the presence of dust, the transformation of U − V to 2200 − V colors is done with stellar tracks
that already include reddening.
assuming the expected star formation history, i.e., high formation redshift and
passive evolution.
The other two are derived from a linear fit to the U − V colors of nearby
blue-sequence galaxies from the Nearby Field Galaxy Survey (NGFS; Jansen et
al. 2000a). Here we synthesized the 2200 − V slope under two assumptions. First,
we assumed the CMR to reflect a systematic age trend, where higher-luminosity
galaxies have higher stellar ages, and we used Bruzual & Charlot (2003) models
to transform U − V colors to 2200 − V colors (see Figure 3) For galaxies with blue
colors, the models give ∆(2200−V ) ≈ 1.5∆(U −V ) for a range of metallicities and
stellar ages. Second, we assume the slope is the result of increasing dust opacity
with luminosity. Adopting the Calzetti et al. (2000) dust law yields ∆(2200−V ) ≈
2.3∆(U − V ). An SMC extinction law (Gordon et al. 2003) would yield similar
values.
Fig. 2 shows that the slope of the CM relation does not depend on redshift.
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
112
and modeling of the blue sequence
Only if we use the z = 0 slope derived for very red early-type galaxies in Coma
is there any hint of evolution. This model is rather extreme, however, and will
not be considered further. The measured evolution of the slope is 0.01z ± 0.03,
consistent with zero. The error-weigthed mean value of all FIRES measurements
is
δ(2200 − V )/δMV = −0.17 ± 0.021
If the slope were due to an age gradient as a function of magnitude, one might
expect the slope to steepen with redshift, but we see no significant effect in the
data presented here. We analyze the cause of the relation later in §4.3.
We note that Barmby et al. (2004) found a linear CMR for faint galaxies, and
an upturn of the CMR at bright magnitudes. Figure 1 shows some evidence for an
upturn at the bright end, but obviously our sample is too small to describe this
properly, and we hence focus entirely on the linear part of the relation.
4.2
The Zeropoint and its Evolution
We determine the zeropoint by assuming that the slope of the CMR does not
evolve and remains at the mean value of δ(2200 − V )/δMV = −0.17. We subtract
the slope so that all galaxy colors are normalized to the color at MV = −21.
The histogram of normalized colors was determined as described before, and the
zeropoint was determined from the location of the peak.
The evolution of the CMR zeropoint at MV = −21 with redshift is shown in
figure 4. The errorbars reflect the central 68% confidence interval of the zeropoint,
obtained with the bootstrapping technique. We also present the CMR zeropoint
of a local galaxy sample from the Nearby Field Galaxy Survey (Jansen, Fabricant,
Franx, & Caldwell 2000). This value was derived from a fit to the U − V colors of
blue-sequence galaxies. We used tracks in Figure 3 to transform from U − V colors
to 2200 − V , applying a small correction for a dust reddening of E(B − V ) = 0.15.
The color excess was obtained from the spectra of the galaxies (see §4.3), and
adopting a Calzetti et al. (2000) dust law.
Clearly, the color of the blue CMR at fixed absolute magnitude reddens monotonically from z ∼ 3 to z ∼ 0.5. The galaxies become redder in 2200 − V by ≈ 1.2
mag from z ∼ 2.7 to z ∼ 0.5 (from the FIRES data alone), or by ≈ 1.45 from
z ∼ 2.7 to z = 0 Surprisingly a straight line describes the points very well, and we
find
2200 − V = 0.24 ± 0.06 − (0.46 ± 0.03)z
The colors in the independent fields of the HDFS and MS1054 agree very well,
and if extrapolated to z = 0, are consistent with the z = 0 point from the NFGS.
5.4 The Color-Magnitude Relation of Blue Field Galaxies
113
Figure 4 — The intercepts of fits to the blue CMR at fixed MV = −21, marking the color
evolution of the blue sequence as a function of redshift. We show the results in the field of the
HDFS field (filled circles) and in the field of MS1054 (diamonds). The errorbars correspond to the
68% confidence interval derived from bootstrap resampling. The star indicates the z = 0 relation
from the NFGS (Jansen, Franx, & Fabricant 2000a). The lines represent tracks of Bruzual &
Charlot (2003) stellar population models. We show a model with formation redshift z f = 3.2,
a star formation timescale τ = 10 Gyr, and fixed reddening of E(B − V ) = 0.15 (dashed line);
one with zf = 10, constant star formation, and fixed E(B − V ) = 0.15 (solid line); a model with
zf = 10, τ = 10 Gyr, and E(B − V ) evolving linearly in time from 0 at z = 10 to 0.15 at z = 0
(dotted line).
This is encouraging considering that absolute calibration between such different
surveys is difficult, and that the NFGS data has been transformed to 2200 − V
colors from other passbands.
Next we compare the observed color evolution to predictions from simple stellar
populations models. We assume that the galaxies remain on the ridge of the CMR
throughout their life. This assumption may very well be wrong, but more extensive
modeling is beyond the scope of this paper. Obviously for such simple models,
the galaxies have all the same color, and these follow directly from the colors of
the stellar population model, depending on star formation history, and the dust
absorption only.
We calculate rest-frame colors using Bruzual & Charlot (2003) stellar population synthesis models with solar metallicity and exponentially declining star
formation rates. We added dust reddening to the model colors using a Calzetti et
al. (2000) dust law. To generate colors at constant absolute magnitude we apply
a small color correction to account for V −band luminosity evolution of the stellar
population. It reflects the fact that galaxies that were brighter in the past populated a different, redder part of the CMR. Similarly, when we allow varying levels
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
114
and modeling of the blue sequence
of dust attenuation, we apply a correction to account for dimming of the V -band
light. We use the measured slope to apply the correction, the total amplitude of
the effect is less than ∆(2200 − V ) . 0.1 for most models.
The simple model tracks are shown in Figure 4. A model with constant star
formation and formation redshift zf = 10 fits remarkably bad: the evolution is
much slower than the observed evolution. When we explore models with exponentially declining star formation rates, we find that a decline time scale τ = 10Gyr
and formation redshift zf = 3.2 fits best. We note that we restricted the fit to
zf ≥ 3.2 as the color-magnitude relation is already in place at redshifts lower than
that.
Naturally, the models can be made to fit perfectly by allowing variable reddening by dust. As an example we show a τ = 10 Gyr model with formation redshift
zf = 10 with reddening evolving linearly with time from E(B − V ) = 0 at zf = 10
to E(B − V ) = 0.15 at z = 0.
We conclude that the relatively strong color evolution in the interval 0 < z < 3
is likely caused by both aging of the stellar population and increasing levels of
dust attenuation with time.
4.3
The Origin of the Blue Sequence in the Local Universe
It is impossible to determine the cause of the color-magnitude relation of blue
galaxies from broad band photometry alone. Fits of models to the photometry
produce age and dust estimates which are very uncertain (e.g., Shapley et al.
2001, Papovich et al. 2001, Förster-Schreiber et al. 2004). Spectroscopy is needed
for more direct estimates of the reddening and ages. Unfortunately, the restframe
optical spectroscopy of distant galaxies is generally not deep enough for the detection of Hβ, and the balmer decrement cannot be determined to high enough
accuracy. Hence we can only analyze the relation for nearby galaxies which have
spectroscopy with high signal-to-noise ratio.
We use the local sample of galaxies in the Nearby Field Galaxy Survey (Jansen,
Fabricant, Franx, & Caldwell 2000; Jansen, Franx, Fabricant, & Caldwell 2000),
a spectrophotometric survey of a subsample of 196 galaxies in the CfA redshift
survey (Huchra et al 1993). It was carefully selected from ∼ 2400 galaxies to
closely match the distributions of morphology and magnitude of the nearby galaxy
population, and covers a large range of magnitudes −15 < MB < −23 . The
strength of the NFGS is the simultaneous availability of integrated broadband
photometry, and integrated spectrophotometry of all galaxies, including line fluxes
and equivalent widths of [OII] Hα , and Hβ . Integrated spectra and photometry
are essential to enable a fair comparison to the integrated photometry of our highredsfhift galaxy sample.
The trends of U − V colors, dust absorption, metallicity, and Hα equivalent
width for normal nearby galaxies are illustrated in Figure 5. The distribution
5.4 The Color-Magnitude Relation of Blue Field Galaxies
115
Figure 5 — U − V colors and nebular emission lines properties versus absolute V -band magnitude of nearby normal galaxies from the Nearby Field Galaxy Survey (Jansen et al. 2000a). (a)
The U − V color versus absolute V magnitude. A linear fit to the blue sequence is shown (solid
line). In the other panels we plot only the galaxies that are within 0.25 mag of the blue colormagnitude rellation (dashed lines). (b) The E(B − V )HII color excess versus absolute V -band
magnitude, derived from the observed ratio of integrated fluxes of Hα and Hβ. The fluxes are
corrected for Balmer absorption and Galactic reddening. The line shows a linear fit to the data
(d). The metallicity sensitive [N II]λ6584/Hα ratio. The dashed line indicates Solar metallicity.
(c) The Hα equivalent width EW[Hα], corrected for Balmer absorption and attenuation by dust.
The correction for dust absorption was obtained from the measured E(B − V ) HII and assuming
an absorption of the stellar continuum E(B − V )cont = rE(B − V )HII , where r = 0.7. The
solid line is a linear fit to the data. Also shown are linear fits to the data in the case that r = 1
(dashed line) and r = 0.2 (dotted line).
U − V colors versus absolute V magnitudes is shown in (a). A blue sequence and
a red sequence of galaxies are visible. A linear fit to the blue sequence using the
same technique as described in §4.1 gives a slope of δ(U − V )/δMV = −0.08. We
isolate the blue galaxies and only consider the spectral properties of galaxies with
colors ∆(U − V ) < 0.25 from the linear fit.
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
116
and modeling of the blue sequence
In (b) we show the reddening E(B −V )HII towards HII regions versus absolute
V -band magnitude. The reddening was computed by Jansen et al. (2000) from
the observed Balmer decrement Hα/Hβ, assuming the intrinsic ration of 2.85 (case
B recombination). As noted by Jansen et al., the sample shows a clear correlation;
more luminous galaxies tend to have higher dust opacities.
The slope of the reddening-luminosity relation is δE(B − V )HII /δMV = 0.06.
This reddening estimate applies to the H II regions in the galaxies, and it is
generally thought that the mean reddening towards the stars is lower by a factor
r: E(B − V )cont = rE(B − V )HII with 0.5 ≤ r ≤ 1 (Kennicutt 1998; Calzetti,
Kinney, & Storchi-Bergmann 1996; Erb et al. 2003). For a Calzetti et al. (2000)
dust law, the implied range of U − V slope is 0.06 - 0.11, close to the observed
value. Other forms of the extinction law, appropriate for the Milky Way (MW;
Allen 1976) or the Small Magellanic Cloud (SMC; Gordon et al. 2003), yield similar
values.
In (b) we show the relation between metallicity and absolute magnitude, as
traced by nebular emission from H II regions. The correlation of the [N II]λ6584/Hα
line ratio with integrated MV shows that brighter galaxies are more metal-rich.
To estimate the implications of this relation on the broadband colors, we explore
two possibilities. Firstly, we use BC03 models to calculate the expected U − V
color variations for stellar populations of a wide variety of metallicities. We explored a grid of models with a range of exponentially declining star formation rates
(τ = 0 − ∞) and ages (t = 0 − 13 Gyr). We find that, regardless of age, models
with blue colors U − V < 0.5 show variations with metallicities of ∆(U − V ) . 0.1,
very small compared to the observed trend.
Finally, we show in (d) the relation between the Hα equivalent width and absolute magnitude, after correction for Balmer absorption and dust absorption. The
EW[Hα] measures the instantaneous star formation rate per unit optical luminosity, and can be interpreted as the ratio of present to past averaged star formation
rate or the time since the onset of star formation.
The correlation of EW[Hα] with MV depends on the value of r used for the
correction for dust extinction. It is straightforward to see that the extinction
correction for log EW[Hα] is proportional to (1 − r): if r = 1, the H II regions and
continuum are equally extincted, whereas for lower values of r, the Hα is more
extincted than the continuum.
We illustrate the effect in (d). The filled circles and the solid line show the data
and a linear fit after correction with r = 0.7. There is no relation with absolute
magnitude, implying that all the galaxies have the same age. The dashed line
shows a linear fit to the data in the case of equal absorption r = 1. Now more
luminous galaxies have slightly higher mean stellar ages. The dotted line shows
r = 0.2, leading to a decreasing ages with increasing luminosity.
We can use the derived slopes to calculate the dependence of age and reddening
as a function of magnitude. If metallicity variations are ignored, an exact solution
5.4 The Color-Magnitude Relation of Blue Field Galaxies
117
can be derived: The reddening of the H II regions follows immediately from the
data, r and the slope of log age versus magnitude follow directly from the observed
slopes of U − V versus magnitude and EW[Hα] versus magnitude. The solution is
r = 0.78
and
δ log(age)/δMV = 0.027
This age gradient is very small, and produces a very shallow color magnitude
slope of 0.01, opposite to the observed slope. The main cause of the color magnitude relation is the variation of dust reddening with magnitude. These results
does not change significantly if metallicity contributes -0.01 to the color magnitude
relation.
We conclude that the data on local galaxies imply that the CMR for blue
galaxies is mostly caused by variations in reddening, with only small contributions
from age and metallicity variations.
4.4
Comparison to z ∼ 3 Galaxies
We turn to our broadband observations at z ∼ 3 and compare the situation in the
local universe to that at high redshift. We select the 186 galaxies in the FIRES
sample at 2.2 < z < 3.2, and inspect the properties of the galaxies that lie within
∆(2200 − V ) < 1 mag of the blue CMR.
4.4.1
The Models
Following previous studies of high redshift galaxies (Papovich, Dickinson, & Ferguson 2001; Shapley et al. 2001; Förster Schreiber et al. 2004), we fit stellar population models to the broadband SEDs of individual galaxies, and interpret the
distribution of best-fit ages and extinctions. We use the publicly availably HYPERZ fitting code (Bolzonella, Miralles, & Pello 2000), updated with the synthetic
template spectra from the latest version of the Bruzual & Charlot (2003) stellar
population synthesis code. We use the Basel 3.1 library (Westera et al. 2002) of
theoretical stellar spectra, selected the Padova1994 stellar evolutionary tracks, and
adopted a Salpeter IMF with upper and lower mass cut-offs of 0.1 and 100 M¯.
For simplicity, we only discuss the results for models with constant star formation rates and solar metallicity. Metallicities of blue, star-forming galaxies z ∼ 2−3
are not well constrained,and may be solar (Shapley et al. 2004) or somewhat lower
(Pettini et al. 2001). We prefer solar metallicity models as they have been directly
calibrated against empirical stellar spectra. We refer to earlier studies for detailed
discussions how the model assumptions affect the distribution of best-fit parameters Papovich, Dickinson, & Ferguson (2001); Shapley et al. (2001). Finally, we
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
118
and modeling of the blue sequence
Figure 6 — The result of Bruzual & Charlot stellar population fits to the broadband SEDs
FIRES fields at 2.2 < z3.2. We show the distribution of best-fit stellar ages and extinctions as
a function of absolute V −band magnitude for solar metallicity, constant star forming models,
reddened with according to a Calzetti et al.(2000) dust law. Only the results for the galaxies
within 1 mag in 2200-V of the blue CMR are drawn. (a) Best-fit ages and (b) best fit color
excesses E(B-V) versus absolute magnitude. The solid lines shows linear fits to the data. and
the inset shows the distribution of residuals around the best fit relation for the color excess.
include reddening by dust by adopting the Calzetti et al. (2000) starburst attenuation law. Hence, the models are characterized by 3 free parameters: age (time since
onset of star formation), star formation rate (SFR), and the level of extinction.
4.4.2
Results
We kept 152 out of a total 186 galaxies for which the models could find an acceptable fit, based on the value of χ2 per degree of freedom. The galaxies that did
not fit well, could have had SEDs that are not well-described by our (incomplete)
template set, or suffer from contamination by emission lines. The distribution of
colors and magnitudes of the rejected galaxies is similar to that of the galaxies
with good fits, and their exclusion is not likely to affect the results.
In Fig 6(a) we show the between best-fit age against Mv . A weak trend is
indicated, in the sense that the faint galaxies are slightly older than the bright
ones. A least squares fit indicates a slope of log(age) versus MV 0.05 ± 0.06,
significant at the 0.8 σ level.
In contrast, the reddening E(B − V ) shows a much stronger correlation with
MV in (c). We fitted a linear relation to the data finding a δE(B − V )/δMV =0.05
± 0.01. Remarkably, this is almost the same as the slope implied for the continuum
reddening of nearby galaxies δE(B − V )/δMV =0.047 (see §4.3).
5.4 The Color-Magnitude Relation of Blue Field Galaxies
119
Figure 7 — The steepness of the CMR slope versus wavelength in the field of the HDFS.
The data show the slope in the rest-frame λ − V color versus MV as a function of the filter
λ. Overplotted are expectations for three extinction laws: the Calzetti et al. (2000) dust law
(dashed line), the MW extinction law (Allen 1976; dotted), and the SMC extinction law (Gordon
et al. (2003); dash-dot line). The normalization is different for each law: δE(B − V )/δM V =0.04
for the Calzetti, 0.05 for the MW, and 0.02 for the SMC. The thick solid line represents the
color-dependence of the CMR slope in the case that stellar population age correlates with M V .
We show the track for a solar-metallicity, constant star-forming model. The normalization is
log(age) ∝ −0.30 MV .
4.4.3
The Blue CMR in Various Rest-Frame Colors
As another way to show why the data imply dust as the main cause of the CMR
slope, we derived the slope of the CMR for many different restframe filters, and
show the results in Figure 7. We described our rest-frame luminosities and colors
earlier in §2.2. We overplot the expected relation as a function of wavelength for
several extinction curves. These curves were scaled with an arbitrary constant to
provide the best fit. As we can see, the Calzetti curve provides the best fit, whereas
the SMC and MW curve fit progressively worse. We also show the expected
dependence if the CMR is caused by age variations. Again, the amplitude of this
curve is fitted to the data points. It does not fit at all to the bluest point, which
is the CMR slope in 1400 − V color versus MV .
We conclude that, models fits to SEDs at z = 3 are in agreement with the
picture in local universe. Reddening by dust explaines the slope of the CMR
consistently.
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
120
and modeling of the blue sequence
5
Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
In this section we explore models to produce the narrow and asymmetric color
distribution of galaxies along the CMR. We focus on the redshift range 2.2 <
z < 3.2, where the fraction of red galaxies is the lowest, and where perhaps the
color distribution may be described by simpler models than at lower redshift. The
distribution (Figure 1) is characterized by a blue peak and a skewness to much
redder colors. The color distribution has a sharp cutoff at the blue side of the
peak, and the scatter of the colors around the peak is quite low: most of the high
redshift galaxies occupy a narrow locus in color space.
Here, our basic assumption is that the color scatter around the CMR is caused
by age variations. We explore 4 different scenarios. In each scenario, we generate
the complete star formation history of model galaxies, typically characterized by
2 or 3 free parameters. We then compile a large library of Monte-Carlo realizations, and generate the expected galaxy color distributions, taking into account
observational errors and biases. We compare the model distributions to the observed color distribution to identify the best fitting parameterizations of the star
formation histories. Kauffmann et al. (2003) used a similar method to constrain
the star formation histories of local galaxies from the Sloan Digital Sky Survey.
A desirable property of this method, as will become apparent below, is that by
using the information contained in the galaxy color distribution, we can resolve
some of the degeneracies produced by fitting SEDs to the broadband colors of
individual galaxies (e.g., Papovich, Dickinson, & Ferguson 2001), in particular the
degeneracies in prior star formation history.
With this in mind, we will now discuss the model ingredients, the generation
of the library, the fitting methodology, and the results for several different parameterizations of the star formation histories.
5.1
A library of Star Formation Histories
As the basic ingredient, we used the solar metallicy BC03 models as discussed in
§4.4, and we fixed the BC03 model parameters except the star formation history:
we only explore the effect of aging on the galaxy color distributions. We will
compare the predictions of the models directly to our rest-frame luminosities and
colors (see §2.2).
We generated a large library of Monte-Carlo realizations of stellar populations
with different star formation histories. We classify the formation histories into 4
seperate scenarios, characterized by distinct parameterizations of the star formation rates. They range from simple constant star formation, to complex models
including bursts. Details of the parameterizations are given together with the results in §5.3. We do not introduce additional complexity, such as a distribution of
metallicities, or distributions in the star formation rate parameters, as we do not
5.5 Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
121
have sufficient observational constraints to justify more freedom in the models.
For each scenario, and for each SFH parameter combination, we generated between 100-200 realizations of individual star formation histories. For every instance
we saved a selection of parameters, including colors, magnitudes, star formation
rates, and masses. The final libraries of the scenarios contain on the order of ∼ 10 5
unique star formation histories each.
5.2
Fitting Method
Briefly, we draw samples from the libraries, generate distributions of colors and
magnitudes, and compare them to rest-frame luminosity and colors of FIRES
galaxies in the redshift range 2.2 < z < 3.2. We fit the models seperately to the
color distributions of the HDFS and MS1054 fields, as these have different depths
and areas.
5.2.1
Creating Mock Observations
We need to account for three essential aspects of the data: the observations are
magnitude limited, contain photometric errors, and have additional color scatter
by variations in dust content.
It is evident that magnitude limits can substantially alter the resulting color
distribution if galaxies in the models evolve strongly in luminosity. For example,
an otherwise undetected galaxy undergoing a massive burst will temporarily increase in luminosity, and can enter a magnitude-limited sample, changing the color
distribution. This effect is enhanced if there are many more galaxies below the
magnitude limit than above. Hence, the steepness of the faint end slope of the
luminosity function also plays a role.
We adopt a faint-end slope of α = −1.6 according to the rest-frame far-U V
luminosity function of Steidel et al. (1999). Shapley et al. (2001) found a steeper
slope, but it resulted from a positive correlation of observed R and R − Ks photometry, which we do not see in our data. We applied the luminosity function in
the following way.
From the models we generate sets of galaxies with the same redshift distribution
as the observed distribution. We then draw a luminosity from a luminosity function
with α = −1.6 and scale the model galaxy to that luminosity. Instead of using the
instantaneous luminosity of the mock galaxy at the time of observation to compute
the scaling, we use its median luminosity over the redshift range 2.2 < z < 3.2.
Furthermore, we add photometric errors to the model colors as a function of
model luminosity. The standard deviations of the errors were determined from a
linear fit to the errors on the rest-frame luminosities and colors as a function of
rest-frame MV . Hence, these include the photometric redshift uncertainties. The
fit gives a mean error in the rest-frame 2200 − V color of 0.15 in the HDFS and
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
122
and modeling of the blue sequence
0.19 for the MS1054.
Finally, we include reddening by dust in the models, adopting the Calzetti et
al. (2000) starburst attenuation law. We add a distribution of color excesses
E(B − V ) = 0.1 ± 0.05(1σ)
to the model colors and magnitudes, where E(B − V ) is required to be greater
than 0. The mean value is appropriate for MV = −21 galaxies from the bestfit models in §4.3, and we used a scatter of 0.05 equal to that found in local
galaxies (see §4.2) and similar to the distribution of E(B − V ) in §4.3. Using
this distribution of extinctions, the scatter added to the model colors is σ(2200 −
V )dust ∼ 0.2. Adopting an SMC extinction law would result in σ(2200 − V )dust ∼
0.15. We remark that the extinction variations of the FIRES galaxies at z = 2 − 3
are well-constrained by the low scatter in the rest-frame far-UV colors.
Summing up, the average (2200 − V ) scatter introduced into the models by
including both photometric uncertainties and dust variations is 0.25 mag for the
HDFS, and 0.28 mag for the MS1054 field. Note that the width of the observed
scatter at 2.2 < z < 3.2, which we characterize by the central 32% of the color
distribution, is 0.58 and 0.83 (for the HDFS and MS1054, respectively; see also
Table 2).
Finally, to complete the mock color distributions, we imposed a magnitude
limit of MV = −19.5 for the HDFS field, and MV = −20.5 for the field of MS1054.
We are complete for all SED types down to these magnitude limits.
5.2.2
The Fitting
First, we subtracted the color-magnitude relation δ(2200 − V )/δMV from the observed colors as in §4.2, and we normalize the 2200 − V color distribution to the
color of the CMR intercept at MV = −21. Then, for each scenario and for each
field, we found the best-fit parameters by performing the two-sided KolmogorovSmirnov (KS) test on the unbinned color distributions of the models and the data.
Next, we multiplied the KS-test probabilities of the individual fields, and we selected the parameter combination that yielded the highest probability.
A property of the KS-test is that it tends to be more sensitive around the
median values of the distribution than at the extreme ends (Press et al. 1992).
Given this behaviour, it is easy to see that our assumptions for the stellar population model, such as IMF and metallicity, may impose strong, yet undesirable,
constraints on the fits. For example, a steeper IMF slope less rich in high-mass
stars (e.g., Scalo instead of Salpeter) results in inherently redder spectra, whereas
lower metallicity of the stellar populations leads to intrinsically bluer spectra. The
KS-test would be sensitive to these systematic color changes in color, violation our
basic assumption that the variations in color are caused by variations in age alone.
5.5 Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
123
Furthermore, both the FIRES fields contain some very red galaxies (Franx et
al. 2003; van Dokkum et al. 2003). These galaxies are thought to have much larger
dust extinctions, higher ages, and perhaps different star formation histories than
the z ∼ 3 blue sequence galaxies we model here (see, e.g., van Dokkum et al.
2004; Förster Schreiber et al. 2004a). In principle, we do not expect our simple
models to account for them, and while the KS-test is relatively insensitive to a
small numbers of outliers, we might consider excluding them from the fit.
For these reasons we perform 3 different KS-tests, which we will refer to as
KS1, KS2, and KS3. Firstly, we do a straightforward two-sided KS-test on the
model and observed distributions. For the second, we normalize both the model
and the observed color distribution to the median, and hence we only test for the
shape of the color distribution, and not the absolute value of the color. Lastly,
we compare the median normalized distributions after excluding galaxies that are
∆(2200 − V ) > 1.75 redder than the mode of the distributions, thereby removing
the reddest outliers. The mode is calculated as in §4.1.
In the subsequent presentation of the results we focus on the second test (KS2),
which is sensitive to the shape only, but we discuss the outcome of the other tests
where appropriate.
We note in advance, that due to inherent differences between the fields the
combined KS probabilities will never reach high values. For example, a twosided KS-test comparing the MS1054 data to the observations of the HDFS (to a
magnitude limit of MV = −20.5) gives a probability of 70%. We do not expect
any model that fits both data sets to exceed this probability.
5.3
5.3.1
Results
Constant Star Formation
In scenario 1, we assume galaxies start forming stars at random redshifts z f . We
take zf to be distributed uniformly in time from z = 2.2 up to a certain certain
maximum redshift zmax , where zmax is between 3.2 < zmax < 10 in steps of 0.2.
The star formation rate of each individual galaxy is constant for a certain time
tsf after which star formation ceases. We construct predictions for values of t sf
sampled logarithmically in 10 steps from 0.05 to 3 Gyr. Hence, this model is
characterized by two parameters: zmax and tsf .
In Figure 8(a) we show the color evolution as a function of time for a characteristic star formation history in this model, together with a schematic representation
of the star formation rate. We also overplot the color intercept of the observed
CMR in the HDFS and MS1054 fields at the median redshift of observation. In
(b) we show the corresponding color-magnitude evolution. Figure 8 (c) shows
the entire color-magnitude distribution of the best-fit models in the fields of the
HDFS and MS1054. The observations are overplotted after subtracting the colormagnitude relation −0.17(MV + 21) (see also §4.1). In (d) we show the absolute
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
124
and modeling of the blue sequence
Figure 8 — The results of model 1. Galaxies in this model start off at random redshifts
z0 < zmax , and form stars at a constant rate for a duration tsf before stopping. The maximum
formation redshift zmax and tsf are the free parameters. (a) The stellar population track of
2200 − V color against age for a characteristic galaxy (z0 = 4, tsf = 1 Gyr). The star formation
rate is illustrated schematically by the gray line. Also drawn is the observed 2200 − V color of
the blue CMR at fixed MV = −21 in the HDFS (star) and MS1054 (diamond). A mean dust
reddening of E(B − V ) = 0.1 ≈ E(2200 − V ) = 0.4 (Calzetti et al 2000) is added to the model
color. (b) The track of 2200 − V color against absolute V -band magnitude for the same galaxy
in steps of 100 Myr (filled circles) (c) The full color-magnitude diagram of the best-fit model
(black points) in the redshift range 2.2 < z < 3.2. The filled gray circles are the data from the
HDFS (top) and MS1054 (bottom). The color distribution of the model is broadened to account
for photometric errors and scatter in the dust properties, where we used σ(2200 − V ) dust = 0.2.
Only galaxies brighter than the absolute magnitude cut-off (dashed line) are included in the fit.
(d) The histograms of residual 2200 − V colors show the data (gray histograms) and the best-fit
model (hatched histograms). The best-fit parameters are zmax = 4.6 and tsf = 1 Gyr.
color histograms of the best-fit model together with the data. We note that the
“shape-sensitive” KS2 test was used to find the best fit.
The best-fit model in the KS2 test has a maximum formation redshift z max =
4.6 and a constant star formation timescale tsf = 1 Gyr. Some characteristics are
reproduced, such as the narrow blue peak, but the asymmetric profile in the blue
peak is not, although there is a low-level tail to red colors containing passively
evolving galaxies. The fit is rather poor; the KS2-test assigns a 0.11 probability
to the fit, the KS1 gives a maximum 0.01, while the KS3 gives 0.04.
In order to understand the behaviour of this model, we explored the zmax and
tsf parameter space. We found this model cannot produce the distinct skewness
towards red colors in the blue peak, even if it cuts off star formation in a substantial
fraction of the galaxies around the epoch of observation. This would be only
mechanism to produce a red skew in this model, but the reddening of the passively
5.5 Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
125
Figure 9 — Same as Figure 8 for model 2. Galaxies start forming stars at random redshifts
z0 < zmax , and form stars at an exponentially declining rate with e-folding time τ . The maximum
formation redshift zmax and τ are the free parameters in this model. A characteristic galaxy
shown in (a,b) has z0 = 3.5 and τ = 0.5 Gyr. The best-fit parameters of the in model (c,d) are
zmax = 4.2 and τ = 0.5 Gyr.
evolving galaxies is so rapid (see Fig. 8a) that instead of a red wing, it produces
a prominent second red peak of passively evolving systems.
5.3.2
Exponentially Declining Star Formation
Scenario 2 is almost identical to the first, but now each individual galaxy has a
single exponentially declining star formation rate with timescale τ . This model is
characterized by two parameters: τ and zmax , where τ is sampled logarithmically
in 10 steps from 0.05 to 3 Gyr.
Figure 9 shows a characteristic star formation history, and the fitting results
of scenario 2. The KS2 test yields a best-fit model with a maximum formation
redshift zmax = 4.2 and a constant star formation timescale τ = 0.5 Gyr. It fits
very poorly however, as the distribution is too broad and symmetric around the
median. The KS2 test rules out this model at the 99% confidence level. The KS1
and KS3 test give similar answers.
Exploring the zmax and τ parameter space, we can understand why the fit is
always poor. The only way for this model to produce a red skew in the color
distribution, is by having a relatively small value for the star formation timescale
τ compared to the mean age of the galaxies at 2.2 < z < 3.2. In that case,
a substantial number of galaxies is entering a “post-starburst” phase, where the
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
126
and modeling of the blue sequence
Figure 10 — Same as Figure 8 for model 3. Repeated burst models, with a multicompenent
star formation history: an underlying constant star formation rate, and superimposed bursts of
a certain strength r = Mburst /Mtot and frequency n. The formation redshift is fixed to z = 10.
A characteristic galaxy shown in (a,b) has n = 1 and r = 1 Gyr−1 . The best-fit parameters of
the model in (c,d) are n = 0.3 Gyr−1 and r = 4. The dashed line in (a) represents a constant
star forming model (Bruzual & Charlot 2003).
mean stellar age is t > τ and the instantaneous SFR becomes much smaller than
the past average. These galaxies gradually move away from the blue CMR, creating
a skewness to red colors.
However, τ = 0.5 Gyr models redden more quickly than CSF models at all
ages, as can be seen comparing Fig 9a with Fig 8a. This results in somewhat
redder mean colors, and a broader spread on the blue side of the peak, created by
the newly formed galaxies that are continuously added to the sample.
5.3.3
Repeated Bursts
In scenario 3, all galaxies start forming stars at a fixed z = 10. The stars form
in two modes: a mode of underlying constant star formation, and superimposed
on this, random star bursts. The bursts are distributed uniformly in time with
frequency n: the average number of bursts per Gyr. The amplitude of the burst
is parameterized as the mass fraction r = Mburst /Mtot where Mburst is the stellar
mass formed in the burst and Mtot is the total mass formed by the constant star
formation and any previous bursts combined. During a burst, stars form at a
constant, elevated rate for a fixed time tburst = 100 Myr. Thus, there are two free
parameters in this model: the burst frequency n, and the burst strength r. We
sample n as n1/2 from 0.1 to 6 Gyr−1 , and r as r 1/3 from 0.01 to 4.
5.5 Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
127
Figure 11 — The map of KS-probabilities versus model parameter repeated burst models of
scenario 3. The free parameter were the burst frequency, and the strength r = Mburst /Mtot of
the bursts. The grayscale linearly encodes the probability of the fit. It shows the degeneracies
of the model parameters, and regions of parameter space that are clearly excluded. The panels
correspond to the model fits to the HDFS data, MS1054 data, and the data sets combined.
Figure 10 shows a characteristic star formation history and the best-fit results
of scenario 3. The KS2 test yields a best-fit model with an average burst frequency
of 0.3 Gyr−1 and a mass fraction formed in each burst of r = Mburst /Mtot = 4
(= 400%): extremely massive, but relatively infrequent bursts. This solution
reproduces the correct shape of the color distribution, i.e., the blue cut-off and red
skew, but not the absolute colors. The median 2200 − V color is 0.4 mag too red,
reflecting the z = 10 formation redshift. The KS2 test probability for this model
is 0.35, with the KS3 test giving a similar value. The KS1 test rejects the model
at the 99% confidence level, as a result of the wrong median color.
We explore in Figure 11 the KS2-fit probability over the relevant part of n, r
parameter space. In both fields, only models with infrequent massive bursts are
allowed, although the exact strength is relatively unconstrained. There is notable
difference between the HDFS and MS1054 fields. Specifically, the broader red wing
in MS1054 observations compared to the HDFS favors a higher burst frequency.
As illustrated in Figure 10(a), the skewness towards red colors is produced by
post-starburst galaxies. These galaxies have just formed large numbers of A-stars
in the previous burst that outshine for some time the O- and B-stars formed in the
underlying mode of constant star formation, hence biasing the integrated colors
to the red. Interestingly, the color of a galaxy after a starburst always remain
redder compared to constant star formation (or no bursts). Concluding, recurrent
bursts do not “rejuvenate” the ensemble of galaxies. Rather, after a very brief
blue period during the burst, the galaxy is redder forever after.
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
128
and modeling of the blue sequence
Figure 12 — Same as Figure 8 for model 4. Episodic star formation model with active periods
of star formation and times of quiescence. The free parameter are the star formation duration
tsf , the fractional duration of quiescence rq , and the remaining fraction of star formation rsf r in
the non-active period. The formation redshift is fixed to z = 10. A characteristic galaxy shown
in (a,b) has tsf = 500 Myr, rtq = 0.6 Gyr−1 ), and rsf r = 0.02. The best-fit parameters of the
model in (c,d) are star formation duration tsf r = 200 Myr, rtq = 0.4, and rsf r = 0.02. Note
that the stellar population after resuming star formation is bluer than before it stopped.
5.3.4
Episodic Star Formation: The Duty Cycle
In the final scenario, galaxies again start at a fixed z = 10, and subsequently
form stars at a constant rate for an “active” period of length tsf . Then star
formation suddenly drops to some fraction rsf r of the nominal value, while the
galaxy passes through a quiescent period that takes a fraction rtq of the “active”
episode rtq = tq /tsf . This constitutes one duty cycle, and these star formation
histories are characterized by repeating cycles of fixed length. We randomize only
the phases of the cycles. There are three free parameters in this model: tsf , rtq ,
and rsf r . We sampled tsf logarithmically in 10 steps from 0.05 to 3.0 Gyr, rtq
logarithmically in 16 steps from 0.04 to 40, and rsf r logarithmically in 10 steps
from 0.01 to 0.8.
Figure 12 shows a characteristic star formation history and the fitting results
of scenario 4. Using the KS2 test, the best-fit model galaxies have active periods
of tsf = 200 Myrs, and quiescent periods of 50 Myr (rtq = 0.4), during which star
formation drops to rsf r = 2% of original rate. Figure 12(d) shows that the key
features of the observed color distribution are correctly reproduced: the blue mean
colors and the asymmetric shape of the observed distribution. The fit probability
of the KS2 test is 0.58, and the KS1 test yields comparable probabilities, indicating
that the model also produces the correct blue absolute colors. The KS3 test, which
5.5 Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
129
Figure 13 — The map of KS-probabilities versus model parameters in the duty-cycle models
of scenario 4. The free parameter were the star formation duration tsf , the fractional duration
of quiescence rq , and the remaining fraction of star formation rsf r in the non-active period. The
seven panels in each row correspond to 2-dimensional slices of the 3-dimensional parameter cube.
Each slice is taken at a fixed rsf r , which is indicated in the lower-right corner. The three rows
correspond to the model fits to the HDFS data, MS1054 data, and the data sets combined. The
grayscale encodes linearly the probability of the fit.
excludes the reddest galaxies, yields 0.89, reflecting that the mismatch with the
model, but also between the two fields mutually, is mainly the enhanced number
of very red galaxies in the field of MS1054 (see also Förster Schreiber et al. 2004).
We discussed in §5.2 that our models are not necessarily expected to describe the
small number of very red galaxies, and we conclude that this model provides the
best description of the shape of the blue peak.
In Figure 13 we show the distribution of the (KS2) fit probabilities over the
entire 3-dimensional parameter space. We present 2-dimensional slices of parameter space at steps in fraction of residual star formation rate rsf r . The parameter
space allowed by the two datasets are somewhat different. The scatter in the
HDFS is intrinsically smaller, leading to smaller values of rq and allowing longer
star formation durations tsf . Also noticable is the “plume” of reasonably high
probabilities in tsf r , rtq . This region of parameter space resembles the model discussed in §5.3.1, where star formation is constant and stops exactly at the age
of observation. The broader red wing of MS1054 clearly favors longer quiescent
periods and short duty-cycles. It can also be seen that the length of the cycles is
not well-constrained, and depends on the level of residual star formation rate r sf r .
If the intra-burst star formation rates are high, then the quiescent periods can be
longer.
A striking aspect of the color evolution of a galaxy in this model is that when
star formation resumes after the quiescent period, then the color of the population
is at least as blue as, or even bluer than before quiescence (see Fig 12a). The high
probabilities of the KS1 test already suggested that, in contrast to the bursting
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
130
and modeling of the blue sequence
model, galaxies with cycling star formation histories manage to maintain their
extremely blue, colors despite the high formation redshift z = 10. The repeating
“rejuvenating” of the colors leads to a substantial slower color evolution with time
than a constant star forming model, as evidenced from the blue “ridge” of the
color evolution track in Fig 12(a).
5.4
Discussion
The results presented here show that if we use the information contained in the
observed color distribution of the galaxies, we can effectively constrain the simple
models of the star formation history.
The first two scenarios (constant SFR and cut-off, and declining SFRs) were
selected for their simplicity, and because these type of star formation histories are
often assumed in SED modeling of high redshift galaxies (e.g., Shapley et al. 2001;
Papovich, Dickinson, & Ferguson 2001; van Dokkum et al. 2004; Förster Schreiber
et al. 2004a).
It is worrying that these scenarios generally fail to reproduce the observed color
distribution. In addition, the parameter values for the best fit are not realistic.
The maximum formation redshift is generally low z = 4 − 4.5, and the timescale
of star formation (tsf or τ ) is comparable to the mean age of the galaxies at
the 2.2 < z < 3.2 redshift of observation. Hence, this is a special moment in
the formation history, with many galaxies switching from active star formation to
passive evolution or much lower SFRs. Such models predict profound evolution
in the color distribution from z = 4 through z = 2, in contrast to the modest
changes in the observed color distribution (Papovich, Dickinson, & Ferguson 2001;
Papovich et al. 2004, this work).
The more complex models we considered are characterized by a fixed, high
formation redshift (z = 10) and constant star formation, which is modulated by
random events, such as starbursts.
The repeated burst model (scenario 3) with underlying constant star formation,
does explain the shape of the observed color distribution, but we disfavore it for
two reasons. Firstly, the adding of bursts does not rejuvenate the galaxy colors but
leads to exceedingly red mean colors, mismatching the observations. We note that
part of if can be resolved by tuning model assumptions, e.g., modifying metallicity,
IMF, formation redshift, etc.
The second, more suspicious aspect is that the model needs to produce bursts
with a high mass fraction, so that the longer-lived, but intrinsically fainter stars
produced in the burst, outshine the luminous O- and B-stars created in the underlying mode of constant star formation. High mass fractions imply extreme instant
star formation rates. Given a burst duration of 100 Myr, an r = Mburst /Mtot ≈ 4
burst at z = 3 leads to a 60−fold increase of the instant star formation rate, transforming an ordinary L∗ Lyman break galaxy into a monster with SFRs of 2000
5.5 Constraints of the Color-Magnitude Relation on the
Star Formation Histories of Blue Galaxies at z ∼ 3
131
M¯ yr−1 . Such galaxies are not observed, unless they are also temporarily highly
abscured, and show up at sub-mm wavelengths as “SCUBA” sources. Slightly
longer burst durations do not alter this picture.
It is evident that the underlying constant star formation artificially introduced
the need for massive bursts. Should it temporarily cease, then the bursts would
not need to involve such high mass fractions to produce a red wing in the color
distribution. In fact, if the underlying star formation temporarily ceased, one does
not need bursts at all: hence the episodic model.
The episodic models (scenario 4) reproduced the color distribution best of the
scenarios explored here. The best-fit parameters correspond to quiescent periods
lasting 30-50% of the time of an active period, and typical duration of the total
duty cycle of 150 Myr to 1 Gyr. The relative fraction of the quiescent period is
better constrained than the length of the duty cycle, which correlates with the
level of residual star formation in the passive periods.
The episodic star formation in these models rejuvenate the galaxies during each
episode, making it significantly bluer than a galaxy with constant star formation of
the same age. This also demonstrates the dependence of the derived ages on prior
star formation history. If the broadband SEDs of our mock galaxies at z ∼ 3 were
fit with constant star forming stellar populations, then the best-fit ages would be
∼ 500 Myr, or a “formation redshift” of the galaxies of zf ∼ 4, instead of the true
zf = 10.
The episodic model could be a solution to the enigma presented by studies
of the broadband SEDs of z ∼ 3 Lyman Break Galaxies (Papovich, Dickinson,
& Ferguson 2001; Shapley et al. 2001). When the SEDs of LBGs were fit with
constant or declining star forming models, the resulting best-fit ages (100-300 Myr)
were much smaller than the cosmic time span of the 2 < z < 3.5 observation epoch
or the age of the universe (1 Gyr, and 2 Gyr, respectively). Papovich, Dickinson,
& Ferguson (2001) already cautioned that constant or declining SFH are probably
wrong for most galaxies, and suggested that episodic star formation could provide a
mechanism to rejuvenate the appearance of the galaxies. Here we have formulated
such a model, subjected it to a quantitative test, and constrained its parameters.
Even more constraints can be placed by taking into account the observed evolution of the color distribution as a function of redshift. The models described here
may be tested directly by the data at lower redshift. However, one reason why we
focused on the redshift range 2.2 < z < 3.2 is that here the fraction of red galaxies
is the lowest, and our simple models are perhaps appropriate. At lower redshift
the color distribution is more complex with a prominent blue and red peak (see
Figure 1), and such simple models might not apply.
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
132
and modeling of the blue sequence
6
The Onset of the Red galaxies
We could already see in Figure 1 that the color distribution in the FIRES fields
evolves strongly from z ∼ 3 to z ∼ 1. This trend is particularly clear in the
lightgray color histograms in (b) and (d). Most notably, at z ∼ 3 most galaxies
are on the blue sequence, and there is no evidence for a well-populated red peak.
The onset of a red peak is tentatively observed in the histograms at z . 2, in
particular in the field of MS1054. The red peak in the z ∼ 1 bin of the MS1054
field contains a major contribution of the cluster of galaxies at z = 0.83. A
prominent red sequence is also observed in photometrically selected samples in
the field up to z = 1 (e.g., Bell et al. 2004b, Kodama et al. 2004).
We calculate the red galaxy fraction as a function of redshift using a simple color criterion to separate red from blue galaxies. Classically, color-limited
fractions are defined relative to the red color-magnitude relation (Butcher & Oemler 1984), but such a definition is unusable here, as the red sequence is virtually absent at z ∼ 3 in our sample. We therefore define red galaxies as all
galaxies more than 1.5 magnitudes redder than the blue color-magnitude relation: (2200 − V ) + 0.17MV > 1.5. The threshold was set to be 2 − 3 times the
scatter in 2200 − V color around the blue sequence. There is no evidence from the
data that the scatter is a function of redshift (see Table 2).
Figure 14(a) shows the evolution of the fraction of red galaxies by number
(Nred /Ntot ). We also show the absolute luminosity density contributed by the red
and blue galaxies seperately in (b), and the fraction of total luminosity contributed
by red galaxies (c). We calculated the fractions to a fixed rest-frame magnitude
limit to which we are complete for all SED types at z ∼ 3 . The limits are
MV = −19.5 in the field of the HDFS, and MV = −20.5 in the field of MS1054.
The luminosity densities were computed by adding the luminosities of the galaxies
above the magnitude limit, and dividing it by the cosmic volume of the redshift
bin. The uncertainties in all estimates were obtained by bootstrapping the colormagnitude distributions.
In both fields, we find a sharp increase from z ∼ 3 to z ∼ 1 in the relative number, the absolute rest-frame V −band luminosity density, and the relative V −band
luminosity density of red galaxies. At the same time V -band luminosity density
in luminous blue-sequence galaxies remains constant, or decreases.
The differences between the fractions in the fields are substantial. In the lowest
redshift bin (z ∼ 1), part of it can be explained by the contribution of the cluster.
At higher redshift however, small number statistics and the variations in space
density of the red galaxies due to large scale structure are the probable cause
(see, e.g., Daddi et al. 2003). The FIRES fields are still small; the total surveyed
area is less than 30 arcmin2 . We note that effect of the 2200 − V color scatter
introduced by the limited accuracy of our photometric redshift technique, 0.32 in
the median for the red galaxies, is minor. The dominant error is the limited number
of galaxies in the sample, which is accounted for in the bootstrap calculation of
5.6 The Onset of the Red galaxies
133
Figure 14 — Evolution of the fraction of red galaxies as a function of redshift. in the HDFS
(filled circles) and MS1054 (diamonds). In (a) we show the number fraction of galaxies that are
(2200 − V ) > 1.5 mag redder than the blue sequence. (b) The absolute luminosity density in
luminous blue (solid line) and red galaxies (dashed line). (c) The relative fraction of the total
luminosity density in red galaxies. There is a sharp increase in the luminosity density in red
galaxies between z=3 and z=1. We note that the field of MS1054 contains a massive cluster at
z = 0.83.
the uncertainties.
The evolution of the red galaxies at high redshift appears to be different from
the evolution between z = 1 and z = 0. Classic studies of the colors of galaxies
at z < 1 found no evidence luminosity evolution in red galaxies, although the
errorbars were fairly large (e.g., Lilly et al. 1995; Lin et al. 1999; Pozzetti et
al. 2003). Recently, Bell et al.(2004b) found a constant luminosity density in
photometrically selected red galaxies in the range 0.2 < z < 1.1, and interpreted
this as an increase in stellar mass in the early-type galaxy population, in apparent
agreement with the hierarchical models of galaxy formation of Cole, Lacey, Baugh,
& Frenk (2000).
We must defer such interpretations for our sample, until we better understand
the nature of the red galaxies at z > 1. The foremost questions are whether all
red galaxies are truly early-types with passively evolving stellar populations, or do
other factors, such as reddening by dust play a role. Furthermore, when does the
narrow red sequence establish it self?
In the local universe, ∼ 70% of the galaxies in the narrow red sequence are morphologically early-type (Strateva et al. 2001; Hogg et al. 2002). The morphologies
of z ∼ 0.7 red peak galaxies in the GEMS survey (Rix et al. 2004), indicate that
85% of their rest-frame V-band luminosity density comes from visually classified
E/S0/Sa (Bell et al. 2004a). In contrast, 70% of the Extremely red objects (EROs)
at 1 < z < 2 are believed to be dust reddened star-forming objects (e.g., Yan &
Thompson 2003; Moustakas et al. 2004), while many of the z ∼ 3 red galaxies in
the FIRES sample also show signs of highly reddened star formation (van Dokkum
et al. 2004; Förster Schreiber et al. 2004). Recent mid-IR imaging with IRAC on
the Spitzer Space Telescope in the HDFS, indicates that only 25% of them are
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
134
and modeling of the blue sequence
dominated by passively evolving stellar populations (Labbé et al. 2004). Clearly,
future study through spectra and MIR-imaging is necessary to give us a better
understanding of the nature of the red colors
The next step in this kind of analysis would be to establish at what redshift the
narrow red sequence establishes itself. Unfortunately, our photometric redshifts
are too uncertain to allow an determination rest-frame colors with an accuracy
better than 0.04 mag, typical of the scatter in the red color-magnitude relation
(Bower, Lucey, & Ellis 1992). Hence spectroscopy is needed to establish the onset
of the red color-magnitude relation.
7
Summary and Conclusions
We used deep near-infrared VLT/ISAAC imaging to study the rest-frame colormagnitude distribution of infrared selected galaxies in the redshift range 1 < z < 3.
We found a well-defined blue peak of star-forming galaxies at all redshifts. The
blue galaxies populate a color-magnitude relation (CMR), such that more luminous
galaxies in the rest-frame V -band tend to have redder ultraviolet-to-optical colors.
The slope of the CMR does not evolve with time, and is similar to the slope
of blue, late-type galaxies in the local universe. Analysis of the spectra of nearby
late-type galaxies from the NFGS suggests that the slope can be fully explained
by the observed correlation of dust content with optical luminosity. The zeropoint
of the blue peak at a given magnitude reddens smoothly from z = 3 to z = 0,
likely reflecting an increase of the mean stellar age and an increase in the mean
dust opacity of blue-sequence galaxies.
A key feature of the blue CMR relation is that the color distribution around it
is asymmetric, with a blue “ridge” and a skew towards red colors. Assuming the
scatter is caused by variations in mean stellar age, we have constructed models to
explore the constraints that these observations place on the star formation history
of blue field galaxies at z = 2 − 3. In the best-fitting models, galaxies form stars in
short “duty-cycles”, characterized by alternating episodes of active star formation
and quiescence. In these particular models, the best constrained parameter is the
relative duration of the quiescent period, which is 30-50% of the length of an active
period. The best-fit total length of the duty-cycle is uncertain, as it correlates with
the amount of residual star formation during quiescence; the data allow a range
of 150 Myr to 1 Gyr.
The models can explain the color distribution well, and, surprisingly, can also
produce the very blue colors of z = 2 − 3 galaxies. The colors are bluer than the
colors of galaxies with the same age with constant star formation, because the
galaxies are rejuvenated by each burst after each quiescent period. These models
can therefor solve the paradox that z ∼ 3 galaxies appear much younger than the
age of the universe when fit with constant star formation models (e.g., Papovich,
Dickinson, & Ferguson 2001). The solution is simply that the luminosity weighted
5.7 Summary and Conclusions
135
ages of the stars derived from model fitting should not be interpreted as the time
since the galaxy first started forming stars.
Finally, we find a sharp increase from z ∼ 3 to z ∼ 1 in the relative number,
the absolute rest-frame V −band luminosity density, and the relative luminosity
density of luminous red galaxies, while the V -band luminosity density in luminous
blue-sequence galaxies is constant, or decreases. Studies at redshifts z ≤ 1 imply
very little evolution, suggesting that the bulk of the evolution takes place round
z ∼ 2. Obviously, our fields are very small, and the variations between the fields
are large enough to suggest that the uncertainties in the red galaxy fraction are
dominated by large scale structure (see e.g., Daddi et al. 2003.
On the other hand, the scatter and evolution of the blue galaxy sequence is
similar between the fields, indicating that field-to-field variations do not play a
large role there. Here the challenge is to confirm directly the cause of the relation
between color and magnitude, and the cause of the skewness to red colors of the
scatter around the relation. Very high signal-to-noise spectroscopy in the Near-IR
will be needed for this purpose, to measure the balmer emission lines, in order to
estimate ages and reddening. Furthermore, these studies need to be performed at
higher redshifts. The advent of multi-object NIR spectrographs on 8-10m class
telescopes will make such studies feasible in the near future.
Acknowledgments
We thank the staff at ESO for their dedicated work in taking these data and making
them available. This research was supported by grants from the Netherlands
Foundation for Research (NWO), the Leids Kerkhoven-Bosscha Fonds, and the
Lorentz Center.
References
Allen, D. A. 1976, MNRAS, 174, 29P
Baldry, I. K., Glazebrook, K., Brinkmann, J., Ivezić, Ž., Lupton, R. H., Nichol, R. C., &
Szalay, A. S. 2004, ApJ, 600, 681
Bell, E. F., et al. 2004, ApJ, 600, L11
Bell, E. F., et al. 2004, ApJ, 608, 752
Bell, E. F. & de Jong, R. S. 2000, MNRAS, 312, 497
Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393
Bessel, M. S. 1990, A&AS, 83, 357
Blanton, M. R., et al. 2003, ApJ, 592, 819
Bower, R. G., Lucey, J. R., & Ellis, R. S. 1992, MNRAS, 254, 601
Bolzonella, M., Miralles, J.-M., & Pelló, R. 2000, A&A, 363, 476
Bruzual A., G. & Charlot, S. 1993, ApJ, 405, 538
Bruzual et Charlot et al. 2003 (BC03)
Butcher, H. & Oemler, A. 1984, ApJ, 285, 426
Casertano, S. et al., AJ, 120, pp. 2747–2824, 2000
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
136
and modeling of the blue sequence
Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1996, ApJ, 458, 132
Calzetti, D., Armus, L., Bohlin, R. C., Kinney, A. L., Koornneef, J., & Storchi-Bergmann,
T. 2000, ApJ, 533, 682
Cole, S., Lacey, C. G., Baugh, C. M., & Frenk, C. S. 2000, MNRAS, 319, 168
Chester, C. & Roberts, M. S. 1964, AJ, 69, 635
Daddi, E., et al. 2003, ApJ, 588, 50
Dreiling, L. A. & Bell, R. A. 1980, ApJ, 241, 736
Efstathiou, G., et al. 2002, MNRAS, 330, L29
Erb, D. K., Shapley, A. E., Steidel, C. C., Pettini, M., Adelberger, K. L., Hunt, M. P.,
Moorwood, A. F. M., & Cuby, J. 2003, ApJ, 591, 101
Franx, M. et al., The Messenger 99, pp. 20–22, 2000
Franx, M. et al. 2003, ApJ, 587, L79
Forster Schreiber et al. 2004,ApJ, accepted
Forster Schreiber et al. 2004, in preparation
Freedman, W. L., et al. 2001, ApJ, 553, 47
Gordon, K. D., Clayton, G. C., Misselt, K. A., Landolt, A. U., & Wolff, M. J. 2003, ApJ,
594, 279
Griersmith, D. 1980, AJ, 85, 1295
Hogg, D. W., et al. 2003, ApJ, 585, L5
Jansen, R. A., Fabricant, D., Franx, M., & Caldwell, N. 2000, ApJS, 126, 331
Jansen, R. A., Franx, M., Fabricant, D., & Caldwell, N. 2000, ApJS, 126, 271
Jansen, R. A., Franx, M., & Fabricant, D. 2001, ApJ, 551, 825
Katz, N. & Gunn, J. E. 1991, ApJ, 377, 365
Kauffmann, G. & White, S. D. M. 1993, MNRAS, 261, 921
Kauffmann, G., White, S. D. M., & Guiderdoni, B. 1993, MNRAS, 264, 201
Kauffmann, G., Charlot, S., & White, S. D. M. 1996, MNRAS, 283, L117
Kauffmann, G., et al. 2003, MNRAS, 341, 33
Kennicutt, R. C. 1998, ApJ, 498, 541
Kodama, T. & Arimoto, N. 1997, A&A, 320, 41
Labbé, I., et al. 2003, AJ, 125, 1107
Labbé et al 2004, in preparation
Le Borgne, J.-F., et al. 2003, A&A, 402, 433
Moorwood, A. F. 1997, Proc. SPIE, 2871, 1146
Papovich, C., Dickinson, M., & Ferguson, H. C. 2001, ApJ, 559, 620
Papovich, C., et al. 2004, ApJ, 600, L111
Peletier, R. F. & de Grijs, R. 1998, MNRAS, 300, L3
Pettini, M., Shapley, A. E., Steidel, C. C., Cuby, J., Dickinson, M., Moorwood, A. F. M.,
Adelberger, K. L., & Giavalisco, M. 2001, ApJ, 554, 981
Press et al. 1992, Numerical Recipes in C
Rix, H., et al. 2004, ApJS, 152, 163
Rudnick, G. et al. 2001, AJ, 122, 2205
Rudnick, G., et al. 2003, ApJ, 599, 847
Schweizer, F. & Seitzer, P. 1992, AJ, 104, 1039
Shapley, A. E., Steidel, C. C., Adelberger, K. L., Dickinson, M., Giavalisco, M., & Pettini,
M. 2001, ApJ, 562, 95
Shapley et al., 2004, ApJ, accepted
Spergel, D. N., et al. 2003, ApJS, 148, 175
Springel, V., White, S. D. M., Tormen, G., & Kauffmann, G. 2001, MNRAS, 328, 726
Somerville, R. S. & Primack, J. R. 1999, MNRAS, 310, 1087
5.7 Summary and Conclusions
137
Steidel, C. C., Giavalisco, M., Dickinson, M., & Adelberger, K. L. 1996, AJ, 112, 352
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K. L. 1996, ApJ,
462, L17
Steidel, C. C., Adelberger, K. L., Giavalisco, M., Dickinson, M., & Pettini, M. 1999, ApJ,
519, 1
Steidel, C. C., Adelberger, K. L., Shapley, A. E., Pettini, M., Dickinson, M., & Giavalisco,
M. 2003, ApJ, 592, 72
Steinmetz, M. & Navarro, J. F. 2002, New Astronomy, 7, 155
Tully, R. B., Mould, J. R., & Aaronson, M. 1982, ApJ, 257, 527
Tully, R. B., Pierce, M. J., Huang, J., Saunders, W., Verheijen, M. A. W., & Witchalls,
P. L. 1998, AJ, 115, 2264
van Dokkum, P. G., Franx, M., Kelson, D. D., Illingworth, G. D., Fisher, D., & Fabricant,
D. 1998, ApJ, 500, 714
van Dokkum, P. G., Franx, M., Fabricant, D., Illingworth, G. D., & Kelson, D. D. 2000,
ApJ, 541, 95
van Dokkum, P. G. et al. 2003, ApJ, 587, L83
van Dokkum, P. G., et al. 2004, ApJ, 611, 703
Visvanathan, N. 1981, A&A, 100, L20
Westera, P., Lejeune, T., Buser, R., Cuisinier, F., & Bruzual, G. 2002, A&A, 381, 524
White, S. D. M. & Frenk, C. S. 1991, ApJ, 379, 52
Zaritsky, D., Kennicutt, R. C., & Huchra, J. P. 1994, ApJ, 420, 87
5 The color magnitude distribution of field galaxies at 1 < z < 3: the evolution
138
and modeling of the blue sequence
Table 1 — The Blue Color Magnitude Relation
1
z
0.5 − 0.74
0.7 − 1.4
1.4 − 2.2
2.2 − 3.2
HDFS
a2
-0.176
-0.206
-0.140
-0.139
3
b
0.226
-0.186
-0.705
-0.950
MS1054
a2
-0.262
-0.144
-0.234
-0.324
b3
0.098
-0.190
-0.516
-0.892
1
The redshift range of the subsample
The best-fit slope of the blue CMR.
3
The intercept of the blue CMR, after
fitting with a slope fixed at −0.17.
4
The lower redshift bound was extended to z=0.35 for the HDFS,
as the F300W filter is bluer than the classical U-band.
2
Table 2 — The Scatter around the Blue CM
1
z
0.5 − 0.74
0.7 − 1.4
1.4 − 2.2
2.2 − 3.2
1
HDFS
2
σobs
0.53
0.66
0.37
0.57
3
σtrue
0.46
0.61
0.27
0.51
MS1054
2
σobs
0.79
0.72
0.73
0.83
3
σtrue
0.73
0.65
0.66
0.77
The redshift range of the subsample.
The measured central 32% of the color distribution after rejecting
galaxies with ∆(2200 − V ) + 0.17MV > 1.5.
3
The intrinsic color distribution after subtracting the observational errors.
4
The lower redshift bound was extended to z=0.35 for the HDFS,
as the F300W filter is bluer than the classical U-band.
2
CHAPTER
SIX
IRAC Mid-Infrared Imaging of
Red Galaxies at z > 2
new constraints on age, dust, and mass
ABSTRACT
We present deep 3.6 − 8 micron imaging with IRAC on the Spitzer Space
Telescope of a population of galaxies with red rest-frame optical colors at
z > 2. The 13 distant red galaxies (DRGs) were selected in the field of the
Hubble Deep Field South on the simple color criterion Js − Ks > 2.3 and
we compare their properties to those of 23 Lyman Break Galaxies (LBGs or
U-dropouts) at z ∼ 2.5 in the same field. The new IRAC data reaches restframe NIR wavelengths, which are crucial in determining the nature of these
galaxies. We are able to uniquely identify 3 out of 11 DRGs as old passively
evolving systems at z ∼ 2.5. The others are heavily reddened star-forming
galaxies, for which we are now better able to distinguish between the effects
of age and dust. Furthermore, the rest-frame NIR data allow more robust
estimates of the stellar mass and stellar mass-to-light ratios (M/LK ). We
find that in a mass-selected sample DRGs contribute 1.5−2× as much as the
LBGs to the cosmic stellar mass density at 2 < z < 3.5. Also, at a given restframe K luminosity the red galaxies are twice as massive with average stellar
masses ∼ 1011 M¯ , and their M/LK mass-to-light ratios exhibit only 1/3 of
the scatter compared to the U-dropouts. This is consistent with a picture
where DRGs are more massive, more evolved, and have started forming at
higher redshift than most LBGs. We find no evidence for a substantial AGN
contribution to the observed optical-MIR SEDs.
Ivo Labbé, Jiasheng Huang, Marijn Franx, Gregory Rudnick, P. Barmby,
Emanuele Daddi, Pieter G. van Dokkum, Giovanni G. Fazio. Natascha M. Förster
Schreiber, Konrad Kuijken, Alan F. Moorwood, Hans-Walter Rix, Huub
Röttgering, Ignacio Trujillo, Arjen van der Wel, Paul van der Werf, & Lottie van
Starkenburg
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
140
dust, and mass
1
Introduction
M
apping the properties of massive galaxies as a function of redshift provides
strong tests for theories of galaxy formation, as their build-up can be directly
probed from high redshift to the present epoch. But while the theories describing
the growth of large-scale dark-matter structure are thought to be well-constrained
(Freedman et al. 2001; Efstathiou et al. 2002; Spergel et al. 2003), the formation
history of the stars inside the dark-matter halos is still poorly understood. Direct
observations are essential for progress on this front.
Until recently, it was difficult observe statistically meaningful samples of massive high-redshift galaxies. The best-studied samples are selected on the rest-frame
UV light through the Lyman Break technique (LBGs; Steidel et al. 1996a,b, 2003;
Madau et al. 1996; Giavalisco & Dickinson 2001), which yielded large numbers of
relatively low mass, unobscured, star-forming galaxies at z > 2 (Papovich, Dickinson, & Ferguson 2001; Shapley et al. 2001). Recent advances in near-infrared
(NIR) capabilities on large telescopes are now making it possible to access the
rest-frame optical for large numbers of galaxies to z ∼ 3. The rest-frame optical is
already much less sensitive to dust obscuration and on-going star formation than
the rest-frame UV, and is expected to be a better (yet imperfect) tracer of stellar
mass.
In this context, we started the Faint Infrared Extragalactic Survey (FIRES;
Franx et al. 2000): a deep optical-to-infrared multicolor survey of NIR-selected
galaxies in two fields. In the deepest field, the Hubble Deep Field South (HDFS),
we spent 102 hours of VLT/ISAAC imaging on one pointing in the Js , H, and
Ks -bands resulting in the deepest ground-based NIR imaging, and the deepest
K−band imaging to date, even from space (Labbé et al. 2003). We selected highredshift galaxies with the simple color criterion Js − Ks > 2.3, designed to isolate
galaxies at 2 < z < 4 with a prominent Balmer- or 4000 Å break (see Franx et al.
2003). We find these Distant Red Galaxies (DRGs) at high surface densities ∼ 3
arcmin−1 to K = 22.5, with space densities about half of that of LBGs selected
from ground-based imaging down to R = 25.5.
Previous studies using broadband optical-to-NIR SED fitting and optical/NIR
spectroscopy have suggested that DRGs, at a given rest-frame optical luminosity,
have higher ages, contain more dust, and are more massive than LBGs (Franx et
al. 2003; van Dokkum et al. 2004; Förster Schreiber et al. 2004), and they may contribute comparably to the cosmic stellar mass density (Franx et al. 2003; Rudnick
et al. 2003). However, the nature of their red colors is still poorly understood, and
the masses are somewhat uncertain, raising many questions. Are they all truly
old, or are some also very young and very dusty? What is the fraction of passively,
evolving “dead” systems? How much do the DRGs contribute to the stellar mass
density in a mass-selected sample? And how do they relate to the blue Lyman
break galaxies. Finally, what is their role in the formation and evolution of massive
galaxies?
6.2 The Observations, Photometry, and Sample Selection
141
To shed light on all of these questions, we need imaging in the rest-frame
NIR, including the rest-frame K−band. Here, we present deep MIR imaging with
IRAC on the Spitzer Space Telescope of a sample of distant red galaxies in the
field of the HDFS. In combination with the wealth of existing deep imaging from
HST/WFPC2 and VLT/ISAAC, the IRAC data promises to establish more robust
stellar masses and mass-to-light ratios, and may help to reduce the degeneracies
between age and dust in modeling of the broadband spectral energy distributions
(SEDs) (see, e.g., Papovich, Dickinson, & Ferguson 2001; Shapley et al. 2001).
Where necessary, we assume ΩM = 0.3, ΩΛ = 0.7, and H0 = 70 km s−1 Mpc−1 .
The magnitudes are given on the Vega system.
2
The Observations, Photometry, and Sample Selection
We have observed the HDFS WFPC2 field with the IRAC camera (Fazio et al.
2004) on the Spitzer Space Telescope, integrating 1 hour each in the 3.6, 4.5, 5.8,
and 8 micron bands. The full observations and data reduction are described in
Labbé et al. (2004), but we give an abbreviated outline here.
The IRAC images were taken in the 4 broadband MIR filters at 3.6, 4.5, 5.8,
and 8 microns over a 50 x 50 field of view. The pixels are ≈ 1.200 in size. The
observations in the field of the HDFS were taken on May 26 2004 (GTO program
214) and were split into dithered frames of 200s each, except for the 8µ band where
the frame time was 50s.
We used the Basic Calibrated Data (BCD) as provided by the Spitzer Science
Center pipeline, and refined the astrometry of the individual frames with 2MASS
sources. We rejected cosmic rays, corrected for known artifacts such as column
pulldown, and muxbleed, and accounted for the “first frame effect” by subtracting
a median stacked image from the individual frames. Finally, we corrected the
frames for geometric distortion, projected them on the existing ISAAC K−band
image (Labbé et al. 2003), and average-combined them. We compared the result
to a median combined image to make sure that we excluded all cosmic rays.
The pixel scale of the final images is 0.00 36 per pixel, about 1/3 of the original
IRAC scale and equal to 3× the scale of the ISAAC image. The limiting depths at
3.6, 4.5, 5.8, and 8 micron are 22.2, 21.3, 18.9, and 18.2 mag, respectively (derived
from the 5σ effective flux dispersion in 3 arcsec diameter apertures). The image
quality ranges from 2 - 2.4 arcseconds FWHM and is best in the 4.5µ band. The
positional accuracy with respect to sources in the ISAAC K−band is better than
0.00 2 across the field. From hereon, we only consider the deepest central part of the
MIR imaging overlapping the 2.50 × 2.50 ISAAC field.
To achieve consistent photometry across the MIR IRAC bands and the existing NIR catalog1 , we carefully matched the point-spread-function (PSF) of the
Ks , 3.6, 4.5, and 5.8 band to the 8µ band, where the image quality was worst. We
1 NIR
images and catalogs are available from http://www.strw.leidenuniv.nl/˜fires
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
142
dust, and mass
deconvolved a selection of bright stars in the 8 micron image with a selection of
bright stars in the other maps, providing us with the required PSF-match kernels. We convolved the maps to the 8 micron PSF, and checked the quality of
the match by dividing stellar growth curves, normalized to 1 at the IRAC zeropoint radius of R = 12.2 arcseconds. The agreement is better than 3% at radii
R > 2.2 arcseconds; hence we used a diameter of D = 4.00 4 as our fixed aperture
for photometry.
To reduce the effects of confusion, we used the deep ISAAC K−band image
to model and subtract neighbouring sources. The image quality in the K−band
is ∼ 0.00 5, so confusion is less of an issue. We proceeded by convolving each Kband source in isolation to the IRAC PSF, and fitted them simultaneously to each
individual IRAC map, leaving only the fluxes as free parameters. We did not
use the model fluxes directly; instead, we use the models to subtract the confusing
neigbours next to the sources of interest. We then proceeded with normal aperture
photometry in a D = 4.00 4 diameter. We combined the IRAC fluxes with existing
optical/NIR photometry in D = 200 apertures from the catalog published by (Labbé
et al. 2003). Thus to obtain consistent colors, we applied an aperture correction
to the IRAC fluxes which was the ratio of the original K−band flux in the D = 2 00
aperture and the K−band flux in the PSF-matched D = 4.00 4 aperture. Finally,
we assumed a minimum photometric error of 10% reflecting various calibration
uncertainties. The end result is a fairly homogenous photometric catalog spanning
the observed optical-to-MIR in 11 filters.
The primary sample of interest is comprised of the DRGs in the field of the
HDFS which were selected on Js − Ks > 2.3 and K < 22.5 (see Franx et al. 2003).
The sample comprises 14 galaxies in the redshift range 1.9 < z < 3.8, which were
all detected at 3.6, and 4.5 micron. Two galaxies have only upper limits at 5.8
and 8.0 micron. One of these was excluded from further analysis as the deblending
procedure was unsuccessful; the source was also confused in the K−band. The
final sample contains 13 DRGs.
As a comparison sample, we selected Lyman Break Galaxies in the same field
from the WFPC2 imaging (Casertano et al. 2000) to the same limit in K, using
the criteria of Madau et al. (1996). We note that the photometric system of the
WFPC2 is somewhat different from the one adopted for the selection of groundbased LBGs (Steidel et al. 2003). The blue F 300W bandpass causes Lyman break
galaxies to enter the selection window already at redshifts z & 1.8 (Giavalisco &
Dickinson 2001). As a result the redshift distribution of space-based “U-dropouts”
is a better match to that of the DRGs than the groundbased LBGs, or the more
recent “BM/BX” objects at z ∼ 2 (Steidel et al. 2004).
We kept the 37 LBGs that fell in the same redshift range as the DRGs (1.9 <
z < 3.8). Because of their higher surface density, source confusion played a more
important role. Conservatively, we selected only the most isolated, and hence least
affected galaxies out of the 37, leaving a final sample of 23. We checked that the
median K-band magnitude and the median I − K color of this subsample was
6.3 Mid-Infrared Properties of Red Galaxies at z > 2
143
the same as that of the full sample; hence it should be representative. Whenever
calculating global averages, we simply scale the average properties of either sample
to the original numbers (14 and 37).
To conclude, the median K−band magnitude of the DRGs is 21.4 and that
of the space-based LBGs is 21.5. The redshift distributions are comparable, and
both have a median zphot = 2.5. Stars were excluded from the LBG sample using
a method detailed in Labbé et al. (2003) and Rudnick et al. (2003). The galaxy
identifications througout this paper refer to the published catalog of Labbé et al.
(2003), which includes photometric redshifts for all sources.
We determined the rest-frame luminosities and colors by direct interpolation in
AB magnitudes between the observed filters. The rest-frame luminosities are sensitive to the uncertainties in the photometric redshifts. We used photometric redshifts based on an algorithm developed by Rudnick et al. (2001, 2003) and spectroscopic redshifts where available. The photometric redshifts are in good agreement
with the spectroscopic redshifts, with an average |zphot − zspec |/(1 + zspec ) = 0.05
for sources at z > 2. We checked that the publicly available HYPERZ method
(Bolzonella, Miralles, & Pello 2000) gave consistent answers, and we checked that
adding the MIR did not change the photometric redshifts significantly. We decided
not to use these new photometric redshifts for now, as the differences were small.
3
Mid-Infrared Properties of Red Galaxies at z > 2
One of the primary motivations for obtaining the 3.6 − 8 micron IRAC imaging is
to extend the spectral energy distribution (SED) of the DRGs into the rest-frame
near-infrared. The rest-frame NIR is essential to understand why these galaxies
have colors that are much redder than Lyman Break Galaxies. Is it because they
are much more obscured, or because their stellar populations are more evolved?
The answer to this question is a crucial step in understanding their mutual relation.
Some insight has already been obtained from modeling of the observed optical
and near-NIR SED (Förster Schreiber et al. 2004), but it has not been possible to
uniquely distinguish old, passively evolving galaxies from very dusty, actively star
forming systems.
The rest-frame NIR is expected to seperate very young and extremely reddened
(AV > 3) galaxies from older galaxies with much less extinction (AV < 1.5), as
in the first case the spectrum is red througout the rest-frame UV and optical,
and peaks somewhere around the rest-frame J-band, while in the other case the
spectrum peaks in the rest-frame optical.
As a first step we can see if the nature of these objects can be elucidated
by a specific combination of two colors alone. We inspect the observed I − K
versus K − 4.5µ color-color diagram of the galaxies (see Figure 1). At a redshift of
z=2.5, these colors correspond to the rest-frame 2200 − V versus V − J colors. We
examine the three DRGs whose optical-to-NIR SEDs were fit significantly better
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
144
dust, and mass
Figure 1 — I − K versus K − 4.5 color-color diagram of two samples of z > 2 galaxies.
Diamonds show Lyman Break galaxies. Filled circles show galaxies selected on the color-criterion
J − K > 2.3 or Distant Red Galaxies (DRGs). Red galaxies whose optical-to-NIR SED were
best fit with unreddened single age bursts are marked with a star. Also shown are time-evolution
tracks of Bruzual & Charlot (2003) stellar population models, redshifted to z = 2.5. (solid line)
A single age burst, (dashed line) a constant star forming model with a reddening of A V = 1.5.
The reddening vector assumes a Calzetti et al. (2000) extinction law.
with unreddened single age bursts than with dusty constant star formation models
(indicated by stars in Figure 1); these are amongst the reddest in I − K color.
For these three galaxies the best fit single age models predict fairly blue K − 4.5µ
colors. On the other hand, the worse fitting dusty constant star formation models
for these galaxies predict them to be the reddest in K − 4.5µ, which they are not.
Had we only K − 4.5µ color, and hence lacked the apriori knowledge of the best-fit
SFH, this result implies that we would still have been able to crudely distinguish
old passively evolving galaxies from those that are heavily reddened and vigorously
forming stars.
To look in more detail at the three candidate old “dead” galaxies we show in
Figure 2 their broadband SEDs. Remarkably, single burst model predictions based
on the optical and NIR data alone, are directly confirmed by the IRAC 3.6 − 8
micron observations. For these galaxies we can definitively rule out reddened
6.3 Mid-Infrared Properties of Red Galaxies at z > 2
145
Figure 2 — Comparisons between the predicted IRAC fluxes from fits to the optical-to-NIR
SED and the actual IRAC 3.6 − 8 micron observations. The predictions (gray solid lines) are
based on Bruzual & Charlot (2003) stellar populations fitted to the optical-to-NIR SEDs (open
circles). The top row shows the 3 DRGs which, excluding the MIR data, were best-fit with
an unreddened single age burst. The observed MIR IRAC fluxes (filled circles) provide direct
confirmation on their nature. Galaxy 767 shows a flux excess at 8 micron, which cannot be
produced by constant star formation and reddening (black dashed lines); it is possibly related
to AGN activity. The bottom row shows the MIR predictions for galaxies using constant star
forming models with Calzetti et al. (2000) dust reddening. For heavily reddened galaxies, the
MIR observations were often different from the optical-to-NIR based predictions, reflecting the
degeneracy between age and dust in the models. For galaxies with lower dust content A V . 1,
IRAC confirms the predictions. Also drawn is the best-fit to the full SED (black dashed lines).
models with constant star formation. Before the MIR imaging, we could only
marginally exclude these models from the formal χ2 of the fit (Franx et al. 2003).
We also tried less extreme models with declining star formation rates (τ = 300 −
500 Myr) and dust reddening (Calzetti et al. 2000); we find that such models fit
significantly worse than the single age bursts.
One of the galaxies clearly has excess flux at 8 micron. Complex stellar populations with, for example, heavily reddened star formation cannot account for such
excess, because they also generate 5.8µ − 8µ colors that are too blue. It may be
related to an obscured active galactic nucleus (AGN).
Also in Figure 2 we show the MIR predictions for galaxies whose optical-toNIR fluxes were better fit with constant star forming models and reddening. In
general the observed MIR flux points do not confirm the predictions, reflecting the
well-known degeneracy between age and dust in the models which allows a wide
variety of predicted values(see, e.g., Papovich, Dickinson, & Ferguson 2001). This
effect is particularly severe for heavily reddened galaxies that are red throughout
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
146
dust, and mass
Figure 3 — The range in best-fit extinction and age values of various DRGs, for constant
star forming models (Bruzual & Charlot 2003) including dust Calzetti et al (2000). We show
separately the solutions obtained excluding, and including the mid-infrared data. The range in
values is derived from Monte-Carlo simulations, where the photometry is randomized within the
photometric errors, and the fitting procedure is repeated. The improvement of the constraints
depends on the nature of the galaxies.
the optical and NIR, but much less so for moderately obscured galaxies that peak
at observed λ < 2.2µ, e.g. object 176.
The immediate question is how much the MIR fluxes improve the constraints on
the models? To investigate this used the HYPERZ package (Bolzonella, Miralles,
& Pello 2000), updated with the latest Bruzual & Charlot (2003) templates, to fit
stellar population models to the full SED. We used solar metallicity models with a
Salpeter IMF ranging from 0.1 − 100 M¯ , and we adopted a Calzetti et al. (2000)
extinction law. In the fitting we kept the redshifts fixed to the values we derived
earlier with an different method (see §2), and we restricted the ages to the age of
the Universe at each redshift.
Next, we derived confidence limits on the best-fit values for age and dust
through Monte-Carlo simulations. For each galaxy, we generated 200 simulated
SEDs by randomizing the photometry within the photometric errors, and we fitted
them in the same way as the observed SEDs. We did the simulation twice: one
time excluding the MIR fluxes, and the other time including the MIR fluxes. We
derived the 1σ limits on each from the central 68% of the distribution. We crudely
accounted for systematic errors by multiplying the limits with the square root of
the reduced χ2 of the best-fitting model. We assume that the three free parameters in the model are age, dust, and star formation rate. We show in Figure 3
the simulation output for 4 DRGs spanning a wide range in properties. Three of
them were also present in Figure 2.
We can immediately learn several things from the impact of the new MIR
data. Firstly, it has become possible to distinguish extremely reddened, very
young models (AV > 3) from much older less reddened models (AV < 2) (see
galaxy 496 and 500 in Figure 3). Galaxy 496 is now uniquely fit with a very
6.4 Comparison to Lyman Break Galaxies
147
young, heavily reddened model, implying instant star formation rates of more
than 2000M¯ yr−1 , in the range of the sub-mm selected “SCUBA” sources (e.g.,
Ivison et al. 2000; Smail et al. 2000). Galaxy 656 is best fit with a very old,
very dusty model in both cases. Here the improvement is modest, although bestfit solution has shifted somewhat. It is clear though that even with deep IRAC
imaging the age-dust generacy is not completely resolved for all sources. We note
that the uncertainties on the IRAC fluxes are mostly systematic, hence better
calibration can improve the situation. But ultimately other observations, such as
NIR spectroscopy, are needed to place independent constraints. Finally, as already
suggested by Figure 2, contraints on models that are dominated by the blue stellar
continuum with moderate extinction levels AV < 1 do not improve particularly
from the MIR data: the parameters were already well-constrained from the optical
and NIR data alone.
In summary, we find for the whole sample of 13 DRGs that 3 galaxies are
uniquely identified as old and passively evolving, 7 galaxies are well-fit by dustreddened and star forming models with a range in extinctions and ages, 1 galaxy
can be fit with either model, and 2 galaxies were badly fit. One of the 2 galaxies
with a bad fit (galaxy 66) is known to have strong emission lines.
Current evidence suggests that for about half of the DRG sample the IRAC
MIR imaging improves the constraints on the ages and extinction levels. It is
therefore imperative to see whether this has changed the median best-fit properties
of the galaxies, most importantly estimates of their stellar masses.
4
Comparison to Lyman Break Galaxies
Our previous studies of the broadband SEDs (Franx et al. 2003; Förster Schreiber
et al. 2004), and emission lines (van Dokkum et al. 2004) have indicated that at
a given rest-frame optical luminosity DRGs are older, more obscured, and more
massive than LBGs.
The results are still somewhat preliminary however, as even at optical wavelengths the effects of dust and on-going star formation introduce uncertainties.
This may be particularly relevant when comparing the stellar masses of DRGs
to those of galaxies with unobscured star formation, such as LBGs. Papovich,
Dickinson, & Ferguson (2001) demonstrated that it is in principle possible to hide
5× the stellar mass under the glare of active, unobscured star-formation, in a
maximally old stellar population.
The IRAC mid-infrared imaging greatly reduces such uncertainties, as the restframe K−band light is a better indicator of the amount of stellar mass. In Figure 4
we present the result of stellar population modeling with the MIR fluxes included.
The main result is that the median properties of both samples are unchanged with
respect to earlier modeling based on the optical-NIR SEDs (Förster Schreiber et
al. 2004). In particular, we find no evidence for an old and massive population
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
148
dust, and mass
Figure 4 — (a) The best-fit values of Bruzual & Charlot (2003) model fits to the broadband
SEDs of DRGs (filled circles) and Lyman Break Galaxies (diamonds). The stars indicate galaxies
that are best-fit with an unreddened single age burst, others were fit with constant-star formation
and Calzetti et al. (2000) dust. The DRGs are on average older, dustier, and more massive than
LBGs.
hiding in the Lyman Break Galaxies. This is in line with IRAC studies of groundbased LBG samples Barmby et al. (2004), which are at somewhat higher redshift
(z ∼ 3) than our space-based U-dropout galaxies (z ∼ 2.5).
Using the same fitting method as in §3 with the full complement of data, we
find the following average values for the 11 out of 13 DRGs with good fits: a mean
stellar mass of < M∗ >= 6.8 × 1010 M¯ , a mean age of < t >= 1.8 Gyr, and a
mean visual extinction < AV >= 1.7, with a large dispersion in all properties.
To calculate the averages, we used the biweight estimator (Beers et al. 1990),
which is not sensitive to outliers. The 3 out 11 galaxies that were best-fit with
an unreddened single age burst have an median age of t = 2.6 Gyr, the maximum
age of the universe at z ∼ 2.5. In contrast, the median properties of LBGs in the
HDFS indicate that they are typically 4 times younger, are 4 times less obscured,
and have stellar masses that are ∼ 4 times lower.
6.5 The rest-frame K-band mass-to-light ratio
149
Other systematic uncertainties on the mass estimates remain, such as the shape
of the initial mass function (IMF) and the mass-limits, or the effects of metallicity,
but these can only be resolved with direct spectroscopic measurements of kinematics and emission line ratios, which is difficult for DRGs (see e.g., van Dokkum
et al. 2003, 2004).
Even without relying on SED modeling we can use simple arguments to estimate the relative contribution of two samples to the stellar mass density at z ∼ 3.
In the redshift range 2 < z < 3.5 we find 11 DRGs and 32 LBGs to the same
K = 22.5 magnitude limit. The DRGs emit about 60% of the light from LBGs at
8µ, but have redder K − 8µ colors, < K − 8µ >= 3.5 versus < K − 8µ >= 3.1
respectively. The K − 8µ color corresponds to the rest-frame V − K at z = 2.5.
Bruzual & Charlot (2003) stellar population models indicate that the redder restframe V − K colors of the DRGs translate into mass-to-light ratios M/LK that
are twice as high. These values are fairly insensitive to the effects of dust or star
formation history.
Hence at the high-mass end the K−selected DRGs contribute somewhat more
to the z ∼ 3 stellar mass density than LBGs, despite their lower number densities.
This contribution increases if the sample were selected in the 8µm band, which
is closer to a selection by stellar mass. To a limiting magnitude at 8µ of 18.2,
we recover 10 out of 13 DRGs, and 12 out of 23 LBGs, boosting the contribution
of DRGs. Hence, in a mass-selected sample at high mass the DRGs contribute
1.5 − 2× more to the stellar-mass density than LBGs.
The major uncertainties in these estimates are the low number statistics and
the variations in space density of the red galaxies due to large scale structure. We
note that the HDFS is known to contain more objects with red observed V − H
colors compared the HDFN (Labbé et al. 2003), although our second deep field
MS1054-03 shows a similar surface density in DRGs over a much larger 50 × 50 field
(see, e.g., van Dokkum et al. 2003; Förster Schreiber et al. 2004).
5
The rest-frame K-band mass-to-light ratio
We used the stellar population fits to the full SED to obtain estimates of the
mass-to-light ratios M/LK . We show the results in Figure 5. In (a) the M/LK is
plotted against extinction showing that the M/LK ratios are not sensitive to the
effects of dusts. This is expected as dust absorption at rest-frame K is small (cf.,
the visual mass-to-light ratios, Förster Schreiber et al. 2004).
A fundamental aspect of the rest-frame K−band mass-to-light ratios is that
they increase with aging of the stellar population. At the same time, complex star
formation histories, such as bursts, cause a spread in the luminosity weighted ages,
and thus in the mass-to-light ratios. Interestingly, the DRGs have systematically
higher M/LK than LBGs, as expected from their redder rest-frame V − K color,
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
150
dust, and mass
Figure 5 — The mass-to-light ratios in the rest-frame K−band (M/LK ) against visual extinction (a) and against age (b). The ages, masses and extinctions, were derived from Bruzual
& Charlot (2003) stellar population fits to the SEDs. We used the best-fitting star formation
history (single burst or constant) to calculate the masses. The galaxies best-fit with unreddened
single age bursts are marked with a star. The DRGs (filled circles) and Lyman Break Galaxies
(diamonds) are indicated. As expected, the observed K−band mass-to-light ratios are not sensitive to extinction effects, but do correlate with age. Overplotted in (b) is the evolution track
of a Bruzual & Charlot (2003) stellar population model for constant star formation. Declining
star formation histories follow similar tracks. The DRGs have higher mass-to-light ratios and
less scatter in the mass-to-light ratios than LBGs.
but more importantly, their scatter is lower. Using the biweight estimator (Beers
et al. 1990), we find log(M/LK ) = 0.17(±0.15) for the DRGs versus log(M/LK ) =
−0.22(±0.45) for the LBGs.
This is consistent with a picture where DRGs are more evolved, and started
forming at higher redshifts than LBGs. Similar conclusions were reached from
emission line studies of a small sample of DRGs in the field of MS1054-03 (van
Dokkum et al. 2004).
6
Discussion and Conclusions
The result presented here offer tantalizing insight into the nature of the red J s −
Ks > 2.3 galaxies at z > 2. The new mid-IR imaging with IRAC on Spitzer
allowed us to uniquely identify “dead” galaxies with old passively evolving stellar
populations. We placed better contraints on the relative role of age and dust for
the star forming DRGs, breaking the degeneracy for some, and reducing it for
many. Furthermore, we obtained more robust stellar masses, and mass-to-light
ratios M/LK in the rest-frame K−band.
6.6 Discussion and Conclusions
151
We confirm the earlier studies of DRGs that found them to be more evolved,
dustier, and more massive than LBGs (van Dokkum et al. 2004; Förster Schreiber
et al. 2004). Specifically, we find average masses for the DRGs of ∼ 1011 M¯ ,
and we excluded the existence of large amounts of stellar mass hiding under the
glare of active star formation in our sample of z ∼ 2.5 U-dropouts (cf., Papovich,
Dickinson, & Ferguson 2001; see also Barmby et al. 2004). We find that for samples
selected in the observed K−band, the DRGs contribute at least as much to the
stellar mass density at 2 < z < 3.5 as Lyman Break Galaxies. If the galaxies were
selected in the rest-frame K−band, as a proxy for selection by stellar mass, then
the DRGs would contribute 1.5 − 2× more than LBGs.
The systematically higher mass-to-light ratios of the DRGs and the lower scatter have possibly far-reaching implications for scenarios of their formation and
evolution. On the one hand, it strongly suggests that DRGs are more evolved
and started forming at higher redshift than LBGs. They may have started out
as Lyman Break Galaxies at z & 5, and then endured a prolonged period of star
formation which increased their stellar mass, metallicity, and dust content. This is
consistent with studies of their emission line properties (van Dokkum et al. 2004).
On the other hand, the comparable M/LK , masses, and stellar ages of the dusty
star forming and “dead” DRGs suggest that they are more closely related to each
other than to LBGs.
While it can not be excluded that DRGs undergo a renewed “Lyman Break
phase” after the addition of metal poor gas, the higher masses suggest that they
are not simply LBGs seen along more obscured lines of sight.
How DRGs relate to lower-redshift galaxies is still unclear. Given their large
masses, it is inevitable that they will evolve into massive galaxies locally, as they
can never lose appreciable amounts of stellar mass. However, we cannot exclude
that they evolve in complicated ways, and change their appearance after dramatic
events, such as a gas-rich mergers (see e.g., Steinmetz & Navarro 2002).
Finally, we have found little evidence in the broadband SEDs of the DRGs
that AGNs play a major role. We find a flux excess at 8 microns for 1 out of 13
galaxies, possibly related to an obscured AGN, although it did not affect the rest
of the SED. For the LBG sample we find evidence for AGN activity at 8 microns
in 3 out of 23 galaxies. The high fraction of AGNs found earlier by van Dokkum
et al. (2004) in a spectroscopic sample of DRGss might have been a selection
effect, or implies that most of the AGNs have low-luminosity. Chandra studies
of DRGs in the FIRES MS1054-03 field suggest a luminous AGN fraction of 5%
(Rubin et al. 2004), comparable to our findings. In this light it seems fair that we
have interpreted our broadband SEDs exclusively in terms of stellar population
properties.
The results presented here can clearly benefit from larger samples, and we
are undertaking more studies at optical-to-MIR wavelengths towards this goal.
In addition, large programs are underway to obtain follow-up optical and NIR
6 IRAC Mid-infrared imaging of red galaxies at z > 2: new constraints on age,
152
dust, and mass
spectroscopy, which are constraining their space densities, and are providing independent ways to measure the extinctions, star formation rates, and masses. While
such studies are now very hard, they will benefit tremendeously from the arrival
of multi-object NIR spectrographs at 8 − 10 meter class telescopes.
Acknowledgments
This research was supported by grants from the Netherlands Foundation for Research (NWO), the Leids Kerkhoven-Bosscha Fonds, the Lorentz Center, and
the Smithsonian Institution. GR would like to acknowledge the support of the
Deutsche Forschunggemeinschaft (DFG), SFB 375 (Astroteilchenphysik).
References
Barmby, P. et al. 2004, ApJS, in press
Beers, T. C., Flynn, K., & Gebhardt, K. 1990, AJ, 100, 32
Bolzonella, M., Miralles, J.-M., & Pelló, R. 2000, A&A, 363, 476
Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000 (BC03)
Casertano, S. et al. AJ, 120, pp. 2747–2824, 2000
Calzetti, D., Armus, L., Bohlin, R. C., Kinney, A. L., Koornneef, J., & Storchi-Bergmann,
T. 2000, ApJ, 533, 682
Efstathiou, G., et al. 2002, MNRAS, 330, L29
Fazio, G.G. et al., ApJS, in press
Franx, M. et al. The Messenger 99, pp. 20–22, 2000
Franx, M. et al. 2003, ApJ, 587, L79
Förster Schreiber et al. 2004, ApJ, accepted (astro-ph/0408077)
Freedman, W. L., et al. 2001, ApJ, 553, 47
Giavalisco, M. & Dickinson, M. 2001, ApJ, 550, 177
Ivison, R. J., Smail, I., Barger, A. J., Kneib, J.-P., Blain, A. W., Owen, F. N., Kerr,
T. H., & Cowie, L. L. 2000, MNRAS, 315, 209
Labbé, I., et al. 2003, AJ, 125, 1107
Labbé, I., et al, 2004, in preperation
Madau, P., Ferguson, H. C., Dickinson, M. E., Giavalisco, M., Steidel, C. C., & Fruchter,
A. 1996, MNRAS, 283, 1388
Papovich, C., Dickinson, M., & Ferguson, H. C. 2001, ApJ, 559, 620
Rudnick, G. et al. 2001, AJ, 122, 2205
Rudnick, G., et al. 2003, ApJ, 599, 847
Rubin, K., et al. 2004, ApJ, accepted (astro-ph/0408101)
Smail, I., Ivison, R. J., Owen, F. N., Blain, A. W., & Kneib, J.-P. 2000, ApJ, 528, 612
Steidel, C. C., Giavalisco, M., Dickinson, M., & Adelberger, K. L. 1996, AJ, 112, 352
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger
Steidel, C. C., Adelberger, K. L., Shapley, A. E., Pettini, M., Dickinson, M., & Giavalisco,
M. 2003, ApJ, 592, 728
Steidel, C. C., Shapley, A. E., Pettini, M., Adelberger, K. L., Erb, D. K., Reddy, N. A.,
& Hunt, M. P. 2004, ApJ, 604, 534
Steinmetz, M. & Navarro, J. F. 2002, New Astronomy, 7, 155
Shapley, A. E., Steidel, C. C., Adelberger, K. L., Dickinson, M., Giavalisco, M., & Pettini,
6.6 Discussion and Conclusions
M. 2001, ApJ, 562, 95
Shapley, A. E. et al. 2003, ApJ, accepted (astro-ph/0405187)
Spergel, D. N., et al. 2003, ApJS, 148, 175
van Dokkum, P. G. et al. 2003, ApJ, 587, L83
van Dokkum, P. G. et al. 2004, ApJ, accepted (astro-ph/0404471)
153
Nederlandse samenvatting
Achtergrond
S
terrenstelsels zijn de meest majestueuze en imponerende verschijningen
aan de hemel. Majestueus, maar tegelijkertijd een onbegrepen fenomeen.
Ze herbergen miljarden sterren en vormen lichtbakens in een uitgestrekt heelal
dat voornamelijk gevuld is met leegte. Sommigen zijn egaal, roodkleurig, en ellipsvormig, anderen zijn blauw, schijfvormig, met prachtige spiraalstructuren. Edwin Hubble was de eerste die sterrenstelsels op basis van deze vormen indeelde.
Hoe komt het dat sterren wel in melkwegstelsels ontstaan maar niet daarbuiten?
En wat veroorzaakt die verschillen in structuur?
Die vraag heeft diepere gronden dan men zou vermoeden. Het blijkt dat de
vorming en evolutie van sterrenstelsels nauw verbonden is met de eigenschappen
van het heelal als geheel. De astronoom Fritz Zwicky realiseerde zich al in 1937,
dat er iets mysterieus aan de hand was met de bewegingen van sterrenstelsels in
clusters. Clusters zijn enorme opeenhopingen van sterrenstelsels die, gevangen in
elkaars zwaartekracht, om elkaar heen wentelen. Zwicky mat de snelheden van de
sterrenstelsels en rekende uit hoeveel massa vereist was om sterrenstelsels zo te
laten bewegen. Dit bleek maar liefst meer dan 100 maal zo veel te zijn als het
gewicht van alle sterren in die stelsels bij elkaar. Blijkbaar heeft een sterrenstelsel
een hoop te verbergen. Het probleem van de “donkere materie” was geboren.
Tot op de dag van vandaag, bijna 70 jaar later, weten we niet wat donkere
materie is. Het enige dat we weten is dat normale lichtgevende materie, het materiaal waar u, ik, en onze Zon uit bestaan, een bijna verwaarloosbare fractie van
het heelal vormt, slechts een onbetekenende lichtvervuiling. Opeenhopingen van de
onderliggende donkere materie, die langzaam maar zeker groeien door met elkaar
te botsen en te versmelten, bepalen uiteindelijk het uiterlijk en de bewegingen van
sterrenstelsels. Het positieve is dat het licht van de sterren dus wel aangeeft waar
de donkere materie ophoopt. Zodoende geeft het ons een indirecte manier om de
donkere materie te bestuderen.
In de afgelopen decennia is onze kennis over de structuur en de inhoud van
het heelal op de grootste schaal in een stroomversnelling geraakt. Na een ware
stortvloed van waarnemingen uit het diepste heelal, is het nu vrijwel zeker dat dat
er ooit een tijd was zonder sterrenstelsels en zonder sterren. De donkere materie
was extreem gelijkmatig verdeeld, met slechts minieme rimpels, en voor de rest
was er niets dan heet gas. Toen het heelal vervolgens expandeerde en afkoelde,
155
156
Nederlandse samenvatting
groeiden de rimpels van donkere materie tot golven. De pieken van deze golven
vormden de kiemen van sterrenstelsels, waar afgekoeld gas naar toe stroomde en
de eerste sterren vormde. De vreemde situatie in de kosmologie heden ten dage,
is dat we meer lijken te weten over het allereerste begin van sterrenstelsels en het
eindresultaat, dan over wat er is gebeurd in de tussentijd. Eén van de drijfveren
achter dit proefschrift is om daar verandering in te brengen.
Zelfs deze relatief bescheiden taak – er zijn veel lastiger vragen te bedenken – is
een formidabele uitdaging. Stof, gas, stervorming, supernova’s, zwarte gaten; dit
alles speelt een rol in de evolutie van een sterrenstelsel. We weten nog te weinig van
sterrenstelsels om ze met formules te beschrijven. We kunnen dus niet simpelweg
uitrekenen wat er tussen de oerknal en nu is gebeurd. Daarnaast kunnen we niet
met sterrenstelsels experimenteren; we kunnen ze niet opnieuw laten onstaan.
De redding van de kosmologie is dat de snelheid van het licht niet zo snel is
als men denkt. Weliswaar reist het licht in een vingerknip om de aarde en in 8
minuten van hier naar de zon, maar naar de dichtstbijzijnde ster duurt de reis
alweer ruim 4 jaar, en van het ene uiteinde van ons sterrenstelsel – de Melkweg –
naar het andere, maar liefst 100.000 jaar. De implicatie is dat wanneer men kijkt
naar sterrenstelsels op nog veel grotere afstanden, het licht er soms wel miljoenen
jaren over heeft gedaan om ons te bereiken. We zien dan hoe het sterrenstelsel er
miljoenen jaren geleden uitzag en kijken effectief terug in de tijd.
Dit is de basis van de techniek die in dit proefschrift is toegepast. We maken
met de modernste reuzentelescopen de zeer gevoelige afbeeldingen van ver verwijderde sterrenstelsels. Vervolgens bepalen we de afstand tot de sterrenstelsels en
zo weten we wanneer het licht verzonden was. Tenslotte zetten we de afbeeldingen in de goede volgorde en het resultaat is een kosmische film die laat zien hoe
sterrenstelsels van hun geboorte tot nu zijn geëvolueerd.
Infrarood
In de praktijk is het niet zo eenvoudig. Als men sterrenstelsels vanaf hun jeugd wil
zien, moet men meer dan 10 miljard lichtjaar ver weg kijken, dus slechts een paar
miljard jaar na de oerknal. Sterrenstelsels op dermate grote afstanden zijn vanaf
de aarde bezien absurd zwak. Zelfs met de allerbeste instrumenten op de grootste
telescopen moet men tientallen uren licht verzamelen om ze te kunnen zien.
Daarnaast gebeurt er iets met het licht tijdens zijn reis. Door de expansie van
het heelal wordt het licht opgerekt en verschuift naar rodere golflengtes, evenredig
met de afstand die is afgelegd. Wat op grote afstanden ooit is verzonden op de
vertrouwde zichtbare golflengten komt nu bij ons aan in het infrarode deel van het
spectrum.
Voor de komst van de huidige generatie infraroodcamera’s keek men slechts
met instrumenten die gevoelig waren voor zichtbaar licht, zoals met de WFPC2
camera op de Hubble Space Telescope. Door de roodverschuiving keek men in
157
de werkelijkheid naar het ultraviolette (UV-) licht uit het verre heelal. Tot enkele
jaren geleden, grofweg tot het begin van dit proefschrift, werd de meeste informatie
over het jonge heelal verkregen via waarnemingen door deze “UV-bril”.
Nu kan het zo zijn dat het heelal door een UV-bril een totaal misvormd, of
erger, incompleet beeld geeft. Als we bijvoorbeeld op zo’n manier het nabije heelal
zouden bekijken, dan zouden de zwaarste sterrenstelsels, de elliptische stelsels,
vrijwel onzichtbaar zijn. Andere reuzenstelsels, zoals onze eigen Melkweg, zouden
er totaal anders uitzien.
Dus voordat we toekomen aan de vraag hoe sterrenstelsels zijn ontstaan, moeten
we ons afvragen of we wel een compleet beeld hebben van sterrenstelsels in het
vroege heelal. Infrarode waarnemingen spelen daarbij een cruciale rol.
Dit proefschrift
Diepste Infrarode Blik in het Heelal
In hoofdstuk 2 van dit proefschrift presenteren we de diepste en meest gevoelige
infraroodfoto’s genomen die ooit zijn gemaakt van de hemel. Gebruik makend van
een nieuwste infraroodcamera (ISAAC) op de Very Large Telescope (VLT) van
de European Southern Observatory (ESO) te Chili, en slechts observerend onder
optimale opstandigheden, bekeken we tussen oktober 1999 en oktober 2000 meer
dan 100 uur lang één speciale plek aan de hemel, het Hubble Deep Field South
(HDFS).
Het HDFS – een gebiedje aan de hemel dat honderd keer zo klein lijkt als de
volle maan – werd in 1998 speciaal geselecteerd omdat het schijnbaar leeg was, dus
zonder heldere sterren of sterrenstelsels. Een opname door de Hubble ruimtetelescoop met een extreem lange belichtingstijd (in totaal een week) onthulde toen
honderden sterrenstelsels op miljarden lichtjaren afstand. Afgezien van de tegenhanger uit 1995, de Hubble Deep Field North (HDF-N), was tot dan toe nog nooit
zo diep in het heelal gekeken.
Verrassend genoeg onthulden onze nieuwe infraroodwaarnemingen tal van sterrenstelsels die bij de vorige observaties verborgen bleven. Van al deze sterrenstelsels bepaalde we vervolgens de precieze helderheden, kleuren, en afstanden. Het
bleek dat veel van deze stelsels op zo’n grote afstand staan dat hun licht meer
dan 12 miljard jaar nodig heeft gehad om de aarde te bereiken. We zien ze dus
in hun prille ontwikkelingsfase, minder dan 2 miljard jaar na de oerknal waaruit
het heelal is ontstaan. Desondanks lijken sommige exemplaren dankzij hun rode
kleur al verrassend veel op “volwassen” sterrenstelsels, zoals de elliptische stelsels
in het lokale heelal. Dit is niet makkelijk te verklaren in gangbare modellen van
de vorming van sterrenstelsels.
Het gebruik van de infraroodcamera was essentieel, omdat de ontdekte stelsels
het meeste van hun licht uitzenden in het zichtbare licht. Zichtbaar licht, uit-
158
Nederlandse samenvatting
gezonden door sterren op 12 miljard lichtjaar afstand, wordt door ons op aarde
na de roodverschuiving waargenomen als nabij-infrarood licht (golflengte 0,001 0,002 millimeter). Dat is ook de voornaamste reden dat de Hubble ruimtetelescoop
destijds deze populatie primordiale sterrenstelsels heeft gemist.
In hoofdstuk 3 beschrijven we andere sterrenstelsels die nog nooit eerder
waren gezien zo vroeg in de evolutie van het heelal: spiraalstelsels van reusachtige
afmetingen. Het formaat van deze stelsels is praktisch hetzelfde als die van onze
eigen Melkweg, of onze naaste buur, het Andromeda stelsel. Theorieën voor de
vorming van deze schijfstelsels zeggen dat zulke reuzen pas vele miljarden later
onstaan. Het is nog onbekend waarom we er zo vroeg al zo veel zien.
De Kleur van de Kosmos
Hoofstuk 4 beschrijft de evolutie van de gemiddelde kosmische kleur, en daarmee
de gemiddelde leeftijd van de sterren in het heelal.
Met behulp van de kleuren, helderheden en de afstandsdata konden het roodverschuivingseffect in het licht van de sterrenstelsels ongedaan maken. Tegelijkertijd, gaf de afstand (in lichtjaren) tevens aan hoe ver we naar elk sterrenstelsel terug in de tijd keken. Door het licht van alle sterrenstelsels in een bepaald
afstandsinterval te sommeren, kon de ware kleur van het heelal in elk tijdperk
berekend worden.
De diepe waarnemingen in het HDFS van het verleden naar heden leverde zo
een unieke, historische volkstelling onder sterren op. De waarnemingen tonen aan
dat in de meeste sterrenstelsels – net als in vergrijzende westerse landen – de gemiddelde leeftijd van sterren stijgt, aangezien er onvoldoende geboortes plaatsvinden
om de overledenen op te volgen. Jonge sterren zijn heet en blauw, terwijl oude
sterren koel en rood zijn, zodat ook het heelal als geheel steeds roder wordt. 2,5
Miljard jaar na de oerknal was het heelal nog blauw, maar nu, 11 miljard jaar
later, is het beige.
Weliswaar onstaan in de meeste sterrenstelsels telkens nieuwe sterren uit het
gas en stof die de ruimte tussen de sterren vult – op elk moment zal een modaal
sterrenstelsel dus sterren van alle leeftijden bevatten – maar gemiddeld over alle
sterrenstelsels blijkt er een duidelijke, gelijkmatige verouderingstrend in de sterrenpopulatie op te treden.
Zoals gezegd, is de blauwe kleur van het vroege heelal het gevolg van de jeugdige
leeftijden van de sterren die toen de sterrenstelsels bevolkten. Omdat we weten
dat jonge, blauwe sterren veel meer licht uitstralen dan oude, rode sterren, kunnen we ook concluderen dat het aantal sterren in het jonge heelal veel lager was
dan nu. Immers de totale hoeveelheid licht is ongeveer constant gebleven. Onze
resultaten impliceren dat relatief veel sterren vrij laat in de geschiedenis van het
heelal gevormd zijn, niet lang voordat ook onze zon is ontstaan. Deze conclusie,
die goed aansluit bij wat al bekend was over leeftijden van sterren in ons deel van
159
het heelal, geeft voor het eerst direct zicht op de evolutie van sterrenpopulaties
gedurende een groot deel van de geschiedenis van het heelal.
Cyclische Stervorming
In hoofstuk 5 bestuderen we de verdeling van de kleuren van sterrenstelsels met
actieve stervorming zo’n 2.5 miljard jaar na de oerknal; hoeveel zijn er blauw, hoeveel zijn er rood? Omdat we er redelijk zeker van zijn dat we geen grote aantallen
sterrenstelsels meer over het hoofd zien, kunnen we nu de statistische verdeling van
de kleuren gebruiken om meer te leren over de gemiddelde ontstaansgeschiedenis.
De kleurverdeling van sterrenstelsels in het jonge heelal is echter paradoxaal.
Aan de ene kant was de spreiding in de kleurverdeling heel klein, wat gewoonlijk
duidt op gemiddeld vrij hoge leeftijd in de sterrenpopulaties (meer dan 1.5 miljard
jaar), maar de gemiddelde kleur was ook extreem blauw, wat weer duidt op een
jonge leeftijd (minder dan 0.5 miljard jaar).
We onderzochten verschillende modellen voor de ontstaansgeschiedenis om deze
verdeling te verklaren en zijn tot de conclusie gekomen dat een model met cyclische
stervorming de beste oplossing geeft. In dit scenario vormen de stelsels niet continu
sterren, maar zijn ze actief gedurende een bepaalde periode om vervolgens een
korte tijd inactief te zijn. Na de inactieve periode, hervat de stervorming weer.
De inactiviteit mag echter niet te lang duren, want tijdens een inactieve periode
wordt de kleur zeer snel roder en we weten inmiddels dat er 2.5 miljard jaar na de
oerknal niet heel veel rode stelsels zijn waargenomen.
Het blijkt dat de cyclische stervorming resulteert in gemiddeld zeer blauwe
kleuren, zelfs op relatief hoge leeftijden. Zodoende biedt het een oplossing voor de
paradox.
Rood, Roder, Roodst
Hoofdstuk 6 besluit met een eerste stap in het vervolgonderzoek van de eerder
ontdekte rode stelsels, zoals reeds beschreven in hoofstuk 2. Onze nabij infrarood
data (1 − 2 micrometer) alleen was echter nog niet genoeg om met zekerheid vast
te stellen wat de oorzaak is van de rode kleur. Is het daadwerkelijk de oude leeftijd
van de sterrenpopulatie? Of speelt een roodkleuring door stofabsorptie een rol?
Het antwoord werd gegeven door mid-IR afbeeldingen, verkregen met de IRAC
camera aan boord van de gloednieuwe Spitzer ruimtetelescoop.
Onze mid-IR waarnemingen geven zicht op het roodverschoven nabij-IR licht
van deze ver verwijderde stelsels. In het intrinsieke nabij-IR licht is het makkelijker
om een onderscheid te maken tussen de effecten van stof en leeftijd. We vonden
voor het eerst definitieve bewijzen voor zeer oude stelsels in het jonge heelal.
Daarnaast bleek dat de door ons gevonden rode stelsels systematisch zwaarder
waren dan de sterrenstelsels die voorheen in het zichtbare licht waren ontdekt.
160
Nederlandse samenvatting
Zelfs al zijn de rode stelsels kleiner in aantal dan de al bekende blauwe stelsels in
het jonge heelal, toch herbergen ze de meeste van alle sterren. Zij moeten dus een
cruciale rol spelen in de onstaansgeschiedenis van sterrenstelsels. Een belangrijk
deel van het toekomstige onderzoek aan sterrenstelsels zal er op gericht zijn om
deze stelsels beter in kaart te brengen.
Curriculum vitae
I
k kwam ter wereld op 11 november 1972 te Alphen a/d Rijn en bracht het
grootste deel van mijn jeugd door in Leiderdorp. In juni 1991 behaalde ik het
VWO diploma aan het Stedelijk Gymnasium te Leiden.
Ik studeerde van 1991 tot en met 1992 Cognitieve Kunstmatige Intelligentie
aan de Universiteit van Utrecht. In september 1993 begon ik aan de studie Sterrenkunde aan de Universiteit Leiden. Tijdens mijn doctoraalfase werkte ik met
prof. dr. M. Franx aan gravitationeel gelensde afbeeldingen van sterrenstelsels in
het verre heelal. Tegelijkertijd was ik freelance programmeur voor verschillende
bedrijven. Mijn studie resulteerde op 29 februari 2000 in het behalen van het
doctoraalexamen Sterrenkunde.
Op 1 maart 2000 trad ik in dienst als Onderzoeker in Opleiding aan de Universiteit Leiden op een subsidie van de Nederlandse Organisatie voor Wetenschappelijk Onderzoek. Onder leiding van prof. dr. M. Franx onderzocht ik de evolutie
van sterrenstelsels met behulp van zeer gevoelige infrarood afbeeldingen. Naast
mijn promotor waren mijn voornaamste medewerkers dr. G. Rudnick en dr. N.M.
Förster Schreiber.
Ik bracht werkbezoeken aan het California Institute of Technology (Pasadena,
VS) en het Harvard-Smithsonian Center for Astrophysics (Cambridge, MA, VS).
Ik nam deel aan de NOVA herfstschool in 2002, en presenteerde mijn onderzoeksresultaten op conferenties en workshops in Durham, Hawaii, Heidelberg, Manchester,
München, Durham, Sydney, en Venetie (2x). Ik heb gedurende mijn onderzoek
waarnemingen verricht met de 10 m W.M. Keck Telescope op Hawaii (1 keer) en
met de 4.2 m W. Herschel Telescope op La Palma (4 keer). Gedurende 3 semesters
assisteerde ik bij het sterrenkundig prakticum.
Tenslotte droeg ik bij aan de popularisering van de sterrenkunde, door middel
van openbare voordrachten, radio- en kranteninterviews en publicaties in populairwetenschappelijke tijdschriften. Als muzikant speelde ik jarenlang in een viermans
funk-formatie. Ik beoefen verscheidene sporten.
Na mijn promotie zal ik als Carnegie Fellow verbonden zijn aan de Observatories of the Carnegie Institution of Washington (OCIW) in Pasadena (CA,VS).
161
Nawoord / Acknowledgments
V
ele mensen ben ik dank verschuldigd voor de totstandkoming van dit proefschrift. Hier wil ik recht doen aan diegenen, die op welke manier dan ook,
zelfs als niet met naam genoemd, onmisbaar zijn geweest.
Voorop wil ik stellen dat ik geen betere plek had kunnen wensen voor deze
promotie. Over de hele linie, van het secretariaat en systeembeheer tot aan de
staf, heb ik niets dan ondersteuning en aanmoediging ondervonden. Een speciaal
woord van dank heb ik voor Kirsten Groen en Jan Lub.
The most important aspect, however, was being part of the FIRES team. Without the enthusiasm, inspiration, and guidance of its members, and the friendship
and collegiality of Greg Rudnick and Natascha Förster Schreiber in particular, this
work would not have been remotely possible.
I gratefully acknowledge the Leidsche Kerkhoven-Bosscha Fonds, the Nederlandse Organisatie voor Wetenschappelijk Onderzoek, the Leidse Sterrewacht, and
the Harvard-Smithsonian Center of Astrophysics for their generous financial support.
Los van het harde werk heb ik de afgelopen jaren genoten van de sociale activiteiten op de Sterrewacht. Het Forza voetbalteam, het jaarlijkse nacht-volleybal
toernooi, de vrijdagmiddagborrel: het leven op de Sterrewacht zou grijs zijn geweest zonder deze extracurriculariteiten.
Mijn vrienden hebben door hun afleiding en stelselmatige desinteresse voor de
“Astrologie” meer in positieve zin bijgedragen dan zij zich realiseren. Anna, Bas,
Flip & Tiets, Lou, Micha, Miesj, Pedro, Superleen en Thomas, zonder jullie was
hier niets van terecht gekomen.
Als laatste wil ik mijn ouders en Dymph bedanken voor de liefdevolle wereld
waarin ik ben opgegroeid. Met mijn aanstaande reis naar de VS voor de boeg is
het onmisbaar te weten dat er altijd een thuis is.
163
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement