The Astrophysical Journal, 698:L116–L120, 2009 June 20 doi:10.1088/0004-637X/698/2/L116
C
2009. The American Astronomical Society. All rights reserved. Printed in the U.S.A.
STAR FORMATION AND DUST OBSCURATION AT z ≈ 2: GALAXIES AT THE DAWN OF DOWNSIZING
∗
M. Pannella
1
, C. L. Carilli
1
, E. Daddi
2
, H. J. McCracken
3
,F.N.Owen
1
, A. Renzini
4
, V. Strazzullo
1
, F. Civano
5
,
A. M. Koekemoer
6
, E. Schinnerer
7
, N. Scoville
8
,V.Smol
ˇ
ci
´
c
8
, Y. Taniguchi
9
, H. Aussel
2
,J.P.Kneib
10
, O. Ilbert
11,12
,
Y. Mellier
3
, M. Salvato
8
, D. Thompson
13
, and C. J. Willott
14
1
National Radio Astronomy Observatory, P.O. Box 0, Socorro, NM 87801-0387, USA; mpannell@nrao.edu
2
CEA, Laboratoire AIM-CNRS-Universit
´
e Paris Diderot, Irfu/SAp, Orme des Merisiers, F-91191 Gif-sur-Yvette, France
3
Institut d’Astrophysique de Paris, 98 bis Boulevard Arago, 75014 Paris, France
4
INAF-Osservatorio Astronomico di Padova, Vicolo dell’Osservatorio 5, I-35122 Padova, Italy
5
Harvard Smithsonian Center for Astrophysics 60 Garden Street, MS 67 Cambridge, MA 02138, USA
6
Space Telescope Science Institute 3700 San Martin Drive, Baltimore MD 21218, USA
7
Max Planck Institut f
¨
ur Astronomie, K
¨
onigstuhl 17, Heidelberg D-69117, Germany
8
California Institute of Technology, MS 105-24, Pasadena, CA 91125, USA
9
Graduate School of Science and Engineering, Ehime University, Bunkyo-cho, Matsuyama 790-8577, Japan
10
Laboratoire d’Astrophysique de Marseille, Technop
ˆ
ole de Marseille-Etoile 38, rue Frdric Joliot-Curie, 13388 Marseille Cedex 13, France
11
Institute for Astronomy, 2680 Woodlawn Dr., University of Hawaii, Honolulu, Hawaii 96822, USA
12
Laboratoire d’Astrophysique de Marseille, BP 8, Traverse du Siphon, 13376 Marseille Cedex 12, France
13
Large Binocular Telescope Observatory University of Arizona, 933 North Cherry Avenue Tucson, AZ, USA
14
Herzberg Institute of Astrophysics, National Research Council, 5071 West Saanich Rd, Victoria, BC V9E 2E7, Canada
Received 2009 February 27; accepted 2009 May 8; published 2009 June 1
ABSTRACT
We present first results of a study aimed to constrain the star formation rate (SFR) and dust content of galaxies at
z ≈ 2. We use a sample of BzK-selected star-forming galaxies, drawn from the Cosmic Evolution Survey, to perform
a stacking analysis of their 1.4 GHz radio continuum as a function of different stellar population properties, after
cleaning the sample from contamination by active galactic nuclei. Dust unbiased SFRs are derived from radio fluxes
assuming the local radio-IR correlation. The main results of this work are: (1) specific star formation rate (SSFR)s are
constant over about 1 dex in stellar mass and up to the highest stellar mass probed, (2) the dust attenuation is a strong
function of galaxy stellar mass with more massive galaxies being more obscured than lower mass objects, (3) a single
value of the UV extinction applied to all galaxies would lead to a gross underestimate of the SFR in massive galaxies,
(4) correcting the observed UV luminosities for dust attenuation based on the Calzetti recipe provides results in very
good agreement with the radio derived ones, (5) the mean SSFR of our sample steadily decreases by a factor of ∼ 4
with decreasing redshift from z = 2.3 to 1.4 and a factor of ∼ 40 down the local universe. These empirical SFRs
would cause galaxies to dramatically overgrow in mass if maintained all the way to low redshifts; we suggest that this
does not happen because star formation is progressively quenched, likely starting from the most massive galaxies.
Key words: galaxies: evolution – galaxies: fundamental parameters – galaxies: ISM – galaxies: luminosity
function, mass function – galaxies: statistics – surveys
1. INTRODUCTION
How and when galaxies build up their stellar mass is still
a major question in observational cosmology. While a general
consensus has been reached in the last few years on the evolution
of the galaxy stellar mass function (e.g., Dickinson et al.
2003; Drory et al. 2004; Bundy et al. 2005; Pannella et al.
2006; Fontana et al. 2006; Marchesini et al. 2008), the redshift
evolution of the star formation rate (SFR) as a function of stellar
mass SFR(M, z) still remains unclear. Several studies, mainly
based on UV derived SFRs, have found the persistence of the
locally well established anticorrelation between specific star
formation rate (SSFR) and stellar mass up to high redshift
(e.g., Juneau et al. 2005; Feulner et al. 2005; Bauer et al.
2005; Erb et al. 2006; Noeske et al. 2007; Zheng et al. 2007;
Cowie & Barger 2008; Damen et al. 2009; Davies et al. 2009).
This anticorrelation is often regarded in the literature as a
manifestation of the “downsizing” scenario (Cowie et al. 1996),
whereby more massive galaxies form at higher redshift. Part of
∗
Based on observations collected, within the COSMOS Legacy Survey, at the
HST, Chandra, XMM, Keck, NRAO-VLA, Subaru, KPNO, CTIO, CFHT, and
ESO observatories. The National Radio Astronomy Observatory is a facility of
the National Science Foundation operated under cooperative agreement by
Associated Universities, Inc.
this effect is certainly real, as for z<2 the most massive galaxies
tend to be passively evolving ellipticals with no or little ongoing
star formation. However, when referring to actively star-forming
galaxies alone whether this anticorrelation exists or not depends
on the way SFRs are estimated. As already warned by Cowie
et al. (1996), one important and poorly known ingredient in
deriving SFRs from rest-frame UV fluxes is the amount of
dust attenuation suffered by the UV light in the interstellar
medium. Lacking spectral information for large samples, the
dust attenuation factor is the result of a multiparameter, and
highly degenerate, fitting to the multiwavelength photometry
available or, when the photometric coverage is not sufficient, a
median factor is applied to the whole galaxy sample.
An independent estimate of the SFR in a galaxy, not biased
by the galaxy’s dust content, is provided by its radio continuum
emission. This is due to processes such as the free–free emission
from H ii regions and the synchrotron radiation from relativistic
electrons, dominated by young massive stars. By mean of
the well established (but not as well understood) radio-FIR
correlation (e.g., Condon 1992; Kennicutt 1998; Yun et al.
2001) it is possible to estimate the total SFR in a galaxy from
its radio luminosity. Thanks to their arcsecond resolution and
relatively wide field of view, radio interferometric observations
offer several advantages over present-day FIR facilities which
L116
No. 2, 2009 STAR FORMATION AND DUST OBSCURATION AT z ≈ 2 L117
Figure 1. Left: the selection diagram for sBzK star-forming galaxies at z ≈ 2.
Right: photometric redshift distribution of the COSMOS sBzK sample. The
bulk of the sample spans the redshift range [1.3–2.5].
are limited by their ∼10
resolution and narrow field of view.
For this very reason radio continuum observations turn out to be
an excellent tool for tracing the dust-unobscured star formation
in the high-redshift universe.
However, radio emission is not only produced by star forma-
tion but also by active galactic nuclei (AGNs), and therefore
a major challenge in deriving dust-unbiased SFRs from radio
fluxes is to remove the AGN contamination (e.g., Smol
ˇ
ci
´
cetal.
2008). At z>1, even in the deepest present-day surveys, ra-
dio detections are likely to include a substantial population of
AGNs, although extreme ULIRG/SMG-like starbursting galax-
ies do exist. Therefore, the best way to explore, with existing
radio facilities, the dust-unbiased SFRs of normal galaxy pop-
ulations is to use a stacking analysis of the radio data, which
allows the investigation of large galaxy samples drawn from
optical–NIR surveys that are individually undetected in the ra-
dio. This technique already has been used in a number of radio
studies (e.g., Daddi et al. 2007a; White et al. 2007; Carilli et al.
2007, 2008; Dunne et al. 2009; Garn & Alexander 2009).
In this context the Cosmic Evolution Survey (COSMOS;
Scoville et al. 2007), with its state-of-the-art multiwavelength
coverage all the way from X-rays to radio of a 2
◦
field provides
an ideal opportunity to build large high-redshift galaxy samples
with well characterized spectral properties. We take advantage
of the COSMOS database to select a large sample of 1 <z<3
star-forming galaxies, and derive dust-unbiased SFRs from
stacking the 1.4 GHz radio data.
Throughout this paper we use AB magnitudes and adopt
a Λ cosmology with Ω
M
= 0.3, Ω
Λ
= 0.7 and H
0
=
70 km s
−1
Mpc
−1
.
2. RADIO, OPTICAL, NEAR INFRARED, AND X-RAY
DATA
Here, we use the VLA medium-deep 1.4 GHz imaging
covering the whole COSMOS field with a fairly uniform rms
(≈10 μJy) and an angular resolution of 1.
5 (see Schinnerer
et al. 2007).
Deep SUBARU B,z imaging (Capak et al. 2007), and CFHT
K
s
-band data (McCracken et al. 2009) were used to select a dust-
unbiased sample of about 34,000 star-forming BzK galaxies
(sBzK, see the left panel of Figure 1) with K
s
< 23. We refer to
McCracken et al. for a detailed description of the K-selected BzK
sample. Following Daddi et al. (2004) galaxy stellar masses were
estimated assuming a Salpeter (1955) initial mass function from
0.1 to 100 M
. A photometric redshift was assigned to more
than 80% of the sBzK sample, by cross-correlating with the
COSMOS photometric redshift catalog by Ilbert et al. (2009).
The median photo-z is ∼1.7, with less than 2% of the sample
at redshift lower than 1 or higher than 3 (see right panel of
Figure 1), confirming the effectiveness of the BzK selection
technique.
A small fraction (≈2%) of the sBzK sample has a 1.4 GHz
counterpart. The minimum flux density of the radio counterparts,
corresponding to a 3σ detection, is about 30 μJy which, at a
median redshift of 1.7, corresponds to a radio luminosity of
about 5 ×10
23
W/Hz at 1.4 GHz.
In the local universe it is usually assumed, based on radio
luminosity function studies (e.g., Sadler et al. 2002; Condon
et al. 2002), that 1.4 GHz radio luminosities greater than
≈2 × 10
23
W/Hz are mostly produced by AGNs, while below
this luminosity star formation has a dominant role in producing
the observed radio emission, maybe still in concurrence with a
low-luminosity AGN. Even though recent studies (e.g., Smol
ˇ
ci
´
c
et al. 2009a, 2009b; V. Strazzullo et al. 2009, in preparation)
suggest that such a characteristic luminosity was brighter at
higher redshift, the sBzK radio detections are mostly extreme
objects: AGN-dominated galaxies or SMG-like starbursts.
In order to study star formation and dust content for the
normal galaxy population, mostly undetected in radio, we
removed from the sample all the objects with radio counterparts.
In doing so we are likely removing, along with AGN-dominated
sources, the tail of extreme star-forming objects. Nonetheless we
prefer this conservative approach in order to derive more robust
conclusions.
In a further attempt to remove AGN-contributed radio emis-
sion, we restrict our analysis to the inner central 0.9 deg
2
of
the COSMOS field, which is covered by deep Chandra obser-
vations (Elvis et al. 2009). The depth of the Chandra survey
reaches 1.9 × 10
−16
erg cm
2
s
−1
, in the soft band (0.5–2 keV),
which at the median redshift of our sample allows an important
census of the AGN luminosity function. We cross-correlated the
sBzK sample with the catalog of Chandra counterparts (Civano
et al. 2009), removing all matched sources (575) from the final
catalog.
Excluding individually detected X-ray sources may not com-
pletely eliminate AGN sources from the sample. Indeed, sBzK
galaxies with a mid-IR (MIR) excess (∼20%–30% of the total)
are likely to contain heavily obscured (Compton thick) AGNs,
as indicated by their X-ray stacking (Daddi et al. 2007b). How-
ever, MIR excess and nonexcess (normal) galaxies exhibit quite
similar 1.4 GHz radio properties (Daddi et al. 2007a), suggesting
that such Compton-thick AGNs do not contribute substantially
to the radio flux at 1.4 GHz. Moreover, our median stacking
technique automatically reduces the impact of a minority of
galaxies with radio emission in excess from what is expected
from star formation, if they exist.
We ended up with a reduced sample of 11798 objects, over
an area of ≈ 0.9 deg
2
, with a photometric redshift in the range
1 <z<3 and having removed both the radio (248) and the
Chandra (575) detections. In the following, we will present
results based on this subsample but even analyzing the full
sBzK sample our conclusions would remain substantially the
same.
3. DATA ANALYSIS AND RADIO-DERIVED SFR
For each of these sBzK sources, we produced a cutout in the
radio mosaic of 173 × 173 pixel
2
(60.5 × 60.5 arcsec
2
). These
cutouts were then stacked to create median images. Median
stacking is more robust than mean against the tails of the
L118 PANNELLA ET AL. Vol. 698
Figure 2. Left: median stacking result of all the 34000 sBzK galaxies. Middle:
best-fit dirty beam convolved Gaussian to the stacked data. The total flux
recovered is 8.8 ± 0.1 μJy. Right: residual image.
Figure 3. Total radio-derived SFR vs. B band (left), B–z color (middle), and
stellar mass (right). The solid line is the best-fit line: Log(SFR)∝ 0.95 Log M
∗
.
distribution, while the rms still goes down by ≈
√
N. Stacking
images allows an easy way to treat the Bandwidth Smearing
(BWS) effect. This is a well known instrumental effect in
aperture synthesis astronomy (see Thompson 1999), consisting
of a radial stretching of the sources in the image due to the
finite width of the frequency response of the receiver. The BWS
does not affect the total flux of the source though. Total fluxes
are retrieved by fitting a dirty beam convolved with a Gaussian
function to the stacked data. The Galfit code (Peng et al. 2002)
was used for this purpose, but very similar results were obtained
using the AIPS/CLEAN algorithm. As an example, in Figure 2
we show the stacked data, model and residual image for the
original whole (34,000 galaxies) sBzK sample. Given the large
number of objects in our sample, we were able to stack the
radio continuum in bins of different galaxy properties, such
as magnitude, color, and mass. Measured radio fluxes were
converted to SFRs using the median redshift of each stacked
sample (1.7 for the whole population), a synchrotron emission
spectral index of −0.8, and the conversion factor between radio
luminosity and SFR from Yun et al. (2001), i.e.,
SFR = 5.9 ±1.8 × 10
−22
L
1.4 GHz
(M
yr
−1
), (1)
where L
1.4 GHz
is in W Hz
−1
. Errors on SFRs are the squared
sum of the uncertainties coming from the off-source rms in
the stacked images, the fitting to recover total fluxes, and the
uncertainty in Equation (1).
In Figure 3, we show our results for the radio stacking
of the AGN-cleaned sBzK sample as a function of: (1) the
observed B-band magnitude, which is related to the rest-frame
dust uncorrected UV luminosity; (2) the (B − z) color, which
for galaxies at z ∼ 2 is a proxy for the UV slope of the spectral
energy distribution and hence it relates to dust extinction;
and (3) the galaxy stellar mass. We conclude that (1) overall,
the emerging UV light is poorly correlated with the ongoing
SFR, and—somewhat counter intuitively—the highest SFRs are
found among the UV-faintest galaxies; (2) this happens because
galaxies with higher SFRs are more extinguished in the UV;
and (3) the SFR increases with stellar mass almost linearly, as
the slope of the Log(SFR)-Log M
∗
relation is 0.95 ± 0.07, in
Figure 4. Left: radio derived SSFR (solid symbols) at z ≈ 1.6 (squares) and
≈ 2.1 (pentagons) are compared to the uncorrected UV derived SSFR (empty
symbols) as a function of Log M
∗
. Right: radio derived SSFRs from this work
(solid dots), for star-forming galaxies with M
∗
∼ 3 ×10
10
M
, as a function of
redshift at z ≈ 1.4, 1.6, 1.9, 2.3. Literature data are plotted as empty circles. The
dotted curve shows the SSFR as a function of redshift described by Equation (2).
agreement (within the small errors) with the relation found by
Daddi et al. (2007a).
3.1. The Specific Star Formation Rate
In Figure 4 (left panel) we present radio derived SSFRs (SSFR
= SFR/M
∗
) for the reduced sBzK sample, divided into two
redshift bins centered at z ≈ 1.6 (solid squares) and 2.1 (solid
pentagons). From the observed B-band magnitudes we also
derive UV
1500
luminosities, uncorrected for dust attenuation,
then estimating uncorrected UV-derived SSFRs, which are also
plotted in the same figure with empty symbols. Some striking
features are worth noting in the plot: (1) the UV-derived SSFR
drops dramatically with increasing mass whereas dust free
SSFRs show no such effect, the SSFR being constant over almost
one dex in mass (see also Dunne et al. 2009); (2) correcting the
UV light with a single value of extinction A
1500
at all masses (an
approximation often adopted in the literature, see e.g., Gabasch
et al. 2004; Juneau et al. 2005; Bauer et al. 2005) would result in
an artificial decreasing SSFR with increasing mass; and (3) the
mean dust attenuation is a function of the galaxy stellar mass,
with more massive galaxies being more dust-extinguished.
By taking advantage of the available photometric redshifts,
we can split our sBzK sample in four redshift bins centered at
z ≈ [1.4, 1.6, 1.9, 2.3] and look for the redshift evolution of the
SSFR, which is almost independent of stellar mass. On the right
panel of Figure 4 we show how SSFRs are steadily increasing
with redshift, by a factor ∼ 4 in the explored redshift range.
We also overplot three lower redshift realizations
(Brinchmann et al. 2004; Noeske et al. 2007; Elbaz et al. 2007)
and the z ∼ 2 estimate by Daddi et al. (2007a), by computing
the SSFR predicted from these studies for star-forming galaxies
with M
∗
∼ 3 × 10
10
M
, and show how the SSFRs have de-
creased by a factor of 40, for galaxies of this mass, from z ≈ 2.3
all the way down to the local universe.
3.2. The Dust Attenuation at 1500 Å
By forcing the dust-corrected UV-SFRs to agree with the
radio-SFRs, as both a function of galaxy stellar mass and
(B − z) color, we obtain how the UV light attenuation A
1500
at z ∼ 2 relates to these quantities. The result is shown in
the inserts of Figure 5. Meurer et al. (1999) found a similar
relation for a sample of local starburst galaxies. Our relation
naturally extends their results to higher redshifts, and also nicely
No. 2, 2009 STAR FORMATION AND DUST OBSCURATION AT z ≈ 2 L119
Figure 5. Left: UV light attenuation (A
1500
= 2.5× Log(SFR
1.4 GHz
/SFR
1500
)
as a function of galaxy stellar mass. Right: UV light attenuation as a function
of B–z color (UV slope). The dotted line shows the attenuation law derived in
Daddi et al. (2004) as described in the text.
shows that the sBzK selection is much less biased against highly
obscured objects than UV-selected samples. The latter ones are
indeed limited to moderate extinctions, such as A
1500
<3.6 mag
(Meurer et al. 1999).
In the explored redshift interval the dust attenuation, stellar
mass, and SFR are all tightly correlated with each other. The
left panel of Figure 5 shows that the dust extinction A
1500
tightly
correlates with galaxy mass. Therefore, assuming a constant
value for A
1500
(independent of galaxy mass) introduces a
systematic bias and the resulting SSFR(M
∗
) relation decreases
with increasing stellar mass.
We emphasize the excellent agreement of the dust-attenuation
correction derived here using the radio data with that derived
from the UV continuum slope: the dotted line in the right panel
of Figure 5 shows the relation between attenuation and (B − z)
color predicted by the Calzetti et al. (1994) law, as calibrated in
Daddi et al. (2004).
3.3. The Mass Growth of Galaxies
The present results confirm that, within the explored mass
range, the SSFR of z ∼ 2 star-forming galaxies is almost
independent of stellar mass (Daddi et al. 2007a; Dunne et al.
2009). A tight correlation between SFR and stellar mass was
also found at z ∼ 1 (Elbaz et al. 2007), z = 0.2–0.7 (Noeske
et al. 2007) and z ∼ 0 (Brinchmann et al. 2004). Other studies
extensively discussed by Dunne et al. (2009) find instead a
SSFR that declines appreciably with increasing stellar mass
(see also Cowie & Barger 2008). In this respect, we concur with
the arguments put forward by Dunne et al. that appear to be
strengthened by our findings.
The SSFR secular decline and the mentioned results can be
represented roughly by the relation
SFR 270 (M
∗
/10
11
M
)(t/3.4 × 10
9
yr)
−2.5
(M
yr
−1
),
(2)
where t is the cosmic time.
We stress here that the exponent of M
∗
in Equation (2)
may not be strictly 1, and may depend on redshift (compare
with Dunne et al. 2009), hence this relation is best valid for
M
∗
∼ 3 × 10
10
M
and z<2.4(T>2.7 Gyr) for which
if was derived. Still it represents a fair approximation for the
star-forming galaxy population. In Figure 4, we show that such
a relation does not hold for the z ∼ 4 galaxies belonging to the
Daddi et al. (2009) sample, this suggests that it has an important
flattening above a certain redshift, qualitatively resembling in
its behavior the redshift evolution of the cosmic star formation
history.
Integrating Equation (2)fromz = 2toz = 0, i.e., assuming
that individual galaxies continued making stars and growing in
mass all the way to low redshift, they would increase in mass by
a factor ∼250, a clear overgrow even neglecting the contribution
of merging events.
On the other hand, star-forming galaxies do form stars at these
high rates, so either Equation (2) is grossly erroneous, or at some
point it ceases to apply to individual galaxies. Indeed, between
z ∼ 2.4 and z ∼ 0 a major transformation takes place in the
population of galaxies. While at z ∼ 2.4 only a small fraction of
the stellar mass is in passively evolving galaxies (elliptical and
bulges), this fraction grows up to ∼60% by z = 0 (Baldry et al.
2004). Therefore, we argue that Equation (2) does indeed apply
to star-forming galaxies all the way to z = 0, but star formation
turns off in a growing fraction of galaxies, which progressively
turn into passive ellipticals and bulges. Mapping quantitatively
this transformation goes beyond the scope of the present
Letter.
4. CONCLUSIONS
We have presented first results of a study aimed to investigate
the dust-unbiased star formation properties of high-redshift
galaxies, by focusing on their stacked radio properties. We use
a sample of sBzK galaxies, drawn from the COSMOS survey,
with a median redshift of 1.7, rejecting known AGN identified in
both deep X-ray Chandra data and the 1.4 GHz radio imaging.
We demonstrate that a universal dust-attenuation correction
cannot be applied to our sample. For instance, the generic
factor of 5 often used to correct the UV light of Lyman-break
galaxies (LBG) is applicable in our sample only for objects with
M
∗
∼ 3 × 10
10
M
—which incidentally is very close to the
median stellar mass of LBG galaxies (Shapley et al. 2001)—but
would grossly underestimate the correction for more massive
galaxies.
We extend the results of Daddi et al. (2007a) that UV light,
appropriately corrected, is a reliable tracer of SFR at z ∼ 2.
We find that the SFR of star-forming galaxies increases almost
linearly with stellar mass at all explored redshifts.
It appears that we are witnessing an evolution era when almost
all star-forming galaxies had the same evolutionary timescales
and a nearly exponential growth, independent of mass. This
is consistent with Dunne et al. (2009) results. They argue that
the discrepancy found with literature studies might be due to
selection biases present in UV and optically selected studies.
While we agree with their statement, we point out that an
underestimate of the dust attenuation correction could also
explain such discrepancy.
We also find that the mass-independent SSFRs decrease by a
factor of 4 in the redshift range from z = 2.3 to 1.4, a trend that
continues all the way to the local Universe. Individual galaxies
would enormously overgrow in mass if these empirical SFRs
were maintained down to low redshifts. We suggest that this
does not happen because many galaxies turn passive, and do
so in a downsized fashion, because massive galaxies are first to
reach unsustainable SFR levels. Thus, in massive star-forming
galaxies at z ∼ 2 downsizing has not started yet, but it soon
will: we are just at the dawn of downsizing.
Constraining the nature of the physical processes by which
SSFRs are kept approximately constant in star-forming galaxies
of wildly different mass, and the mechanisms that contribute
to discontinuing the star formation activity in massive high-
L120 PANNELLA ET AL. Vol. 698
redshift galaxies, are both substantial challenges for theoretical
models to reproduce and for observers to investigate in full
detail. New ideas on gas accretion modes (Dekel et al. 2009) and
recent observations of widespread large molecular gas reservoirs
(Daddi et al. 2008; Tacconi et al. 2008) in distant massive
galaxies will likely provide crucial paths to understand these
issues.
We thank the anonymous referee for constructive com-
ments that improved the presentation of our results. M.P., V.S.,
and C.L.C. acknowledge partial support from the Max-Planck
Forschungspreise 2005. E.D. and H.J.Mc.C. acknowledge sup-
port from the French grants ANR-07-BLAN-0228-03 and ANR-
08-JCJC-0008. A.R. acknowledges support from the ASI grant
COFIS. F. C. acknowledges support from NASA-Chandra grant
G07-8136A. This work is based in part on data products pro-
duced at TERAPIX. The HST COSMOS Treasury program
was supported through NASA grant HST-GO-09822. We grate-
fully acknowledge the contributions of the entire COSMOS
collaboration.
REFERENCES
Baldry, I. K., et al. 2004, ApJ, 600, 681
Bauer, A. E., Drory, N., Hill, G. J., & Feulner, G. 2005, ApJ, 621, L89
Brinchmann, J., et al. 2004, MNRAS, 351, 1151
Bundy, K., Ellis, R. S., & Conselice, C. J. 2005, ApJ, 625, 621
Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1994, ApJ, 429, 582
Capak, P., et al. 2007, ApJS, 172, 99
Carilli, C. L., et al. 2007, ApJS, 172, 518
Carilli, C. L., et al. 2008, ApJ, 689, 883
Civano, et al. 2009, submitted
Condon, J. J. 1992, ARA&A, 30, 575
Condon, J. J., Cotton, W. D., & Broderick, J. J. 2002, AJ, 124, 675
Cowie, L. L., & Barger, A. J. 2008, ApJ, 686, 72
Cowie, L. L., Songaila, A., Hu, E. M., & Cohen, J. G. 1996, AJ, 112, 839
Daddi, E., et al. 2004, ApJ, 617, 746
Daddi, E., et al. 2007a, ApJ, 670, 156
Daddi, E., et al. 2007b, ApJ, 670, 173
Daddi, E., et al. 2008, ApJ, 673, L21
Daddi, E., et al. 2009, ApJ, 694, 1517
Damen, M., et al. 2009, ApJ, 690, 937
Davies, G. T., et al. 2009, MNRAS, 395, L76
Dekel, A., et al. 2009, Nature, 457, 451
Dickinson, M., Papovich, C., Ferguson, H. C., & Budav
´
ari, T. 2003, ApJ, 587,
25
Drory, N., et al. 2004, ApJ, 608, 742
Dunne, L., et al. 2009, MNRAS, 394, 3
Elbaz, D., et al. 2007, A&A, 468, 33
Elvis, M., et al. 2009, arXiv:0903.2062
Erb, D. K., et al. 2006, ApJ, 647, 128
Feulner, G., et al. 2005, ApJ, 633, L9
Fontana, A., et al. 2006, A&A, 459, 745
Gabasch, A., et al. 2004, ApJ, 616, L83
Garn, T., & Alexander, P. 2009, MNRAS, 394, 105
Ilbert, O., et al. 2009, ApJ, 690, 1236
Juneau, S., et al. 2005, ApJ, 619, L135
Kennicutt, Jr., R. C. 1998, ARA&A, 36, 189
Marchesini, D., et al. 2008, arXiv:0811.1773
McCracken, et al. 2009, submitted
Meurer, G. R., Heckman, T. M., & Calzetti, D. 1999, ApJ, 521, 64
Noeske, K. G., et al. 2007, ApJ, 660, L43
Pannella, M., et al. 2006, ApJ, 639, L1
Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ, 124, 266
Sadler, E. M., et al. 2002, MNRAS, 329, 227
Salpeter, E. E. 1955, ApJ, 121, 161
Schinnerer, E., et al. 2007, ApJS, 172, 46
Scoville, N., et al. 2007, ApJS, 172, 1
Shapley, A. E., et al. 2001, ApJ, 562, 95
Smol
ˇ
ci
´
c, V., et al. 2008, ApJS, 177, 14
Smol
ˇ
ci
´
c, V., et al. 2009a, ApJ, 690, 610
Smol
ˇ
ci
´
c, V., et al. 2009b, ApJ, 696, 24
Tacconi, L. J., et al. 2008, ApJ, 680, 246
Thompson, A. R. 1999, in ASP Conf. Ser. 180, Synthesis Imaging in Radio
Astronomy II, ed. G. B. Taylor, C. L. Carilli, & R. A. Perley (San Francisco,
CA: ASP)
White, R. L., Helfand, D. J., Becker, R. H., Glikman, E., & de Vries, W.
2007, ApJ, 654, 99
Yun, M. S., Reddy, N. A., & Condon, J. J. 2001, ApJ, 554, 803
Zheng, X. Z., et al. 2007, ApJ, 661, L41
![Figure 1. Left: the selection diagram for sBzK star-forming galaxies at z ≈ 2. Right: photometric redshift distribution of the COSMOS sBzK sample. The bulk of the sample spans the redshift range [1.3–2.5].](/figures/figure-1-left-the-selection-diagram-for-sbzk-star-forming-v0vetv8u.webp)
