The Fraction of Quiescent Massive Galaxies in the Early Universe

Abstract

Aims: The aim of this work is to collect a complete, mass--selected sample of galaxies with very low specific star formation rate, for a comparison with the prediction of recent theoretical models. Method: We use the 24/K flux ratio, complemented by the SED fitting to the full 0.35-8.0 mum spectral distribution, to select quiescent galaxies from z~0.4 to z~4 in the GOODS--MUSIC sample. Our observational selection can be translated into thresholds on the specific star formation rate SFR/M_*, that can be used to compare with the theoretical predictions. Results: We find that, in the framework of the well known global decline of the quiescent fraction with redshift, a non-negligible fraction ~15-20% of massive galaxies with very low specific star formation rate exists up to z~4, including a tail of "Red&Dead" galaxies with SFR/M_*<10^{-11}/yr. Recent theoretical models vary to a large extent in the prediction of the fraction of galaxies with very low specific star formation rates, and are unable to provide a global match to our data.

Full text

Page 1

arXiv:0901.2898v2 [astro-ph.GA] 29 Apr 2009
Astronomy & Astrophysics manuscript no. 11650
April 29, 2009
c ? ESO 2009
The fraction of quiescent massive galaxies in the early Universe
A. Fontana1, P. Santini1,2, A. Grazian1, L. Pentericci1, F. Fiore1, M. Castellano1, E. Giallongo1, N. Menci1, S.
Salimbeni1,3, S. Cristiani4, M. Nonino4, and E. Vanzella4
1INAF - Osservatorio Astronomico di Roma, Via Frascati 33, 00040 Monteporzio (RM), Italy
2Dipartimento di Fisica, Universit` a di Roma “La Sapienza”, P.le A. Moro 2, 00185 Roma, Italy
3Department of Astronomy, University of Massachusetts, 710 North Pleasant Street, Amherst, MA 01003
4INAF - Osservatorio Astronomico di Trieste, Via G.B. Tiepolo 11, 34131 Trieste, Italy
Received .... ; accepted ....
ABSTRACT
Aims. We attempt to compile a complete, mass–selected sample of galaxies with low specific star–formation rates, and compare their
properties with theoretical model predictions.
Methods. We use the f(24µm)/f(K) flux ratio and the SED fitting to the 0.35 − 8.0µm spectral distribution, to select quiescent
galaxies from z ≃ 0.4 to z ≃ 4 in the GOODS–MUSIC sample. Our observational selection can be translated into thresholds in
specific star–formation rate ˙ M/M∗, which can be compared with theoretical predictions.
Results. In the framework of the well-known global decline in quiescent galaxy fraction with redshift, we find that a non-negligible
fraction ≃ 15 − 20% of massive galaxies with low specific star–formation rate exists up to z ≃ 4, including a tail of “red and dead”
galaxies with ˙ M/M∗< 10−11yr−1. Theoretical models vary to a large extent in their predictions for the fraction of galaxies with low
specific star–formation rates, but are unable to provide a global match to our data.
Conclusions.
Key words. Galaxies:distances and redshift - Galaxies: evolution - Galaxies: high redshift
1. Introduction
Understanding the formation and evolution of early–type galax-
ies is a major goal of present-daycosmology,as well as a funda-
mental benchmark for “ab-initio” theoretical models of galaxy
evolution.
According to several independentlines of evidence, the pop-
ulation of massive galaxies has undergone major evolution dur-
ing the epoch correspondingto the redshift range 1.5− 3, where
the galaxy stellar mass function evolved significantly at high
masses (Berta et al. 2007; Fontana et al. 2006) (F06 in the fol-
lowing),massivegalaxiessettledontotheHubblesequence(e.g.,
Abraham et al. (2007) and Franceschini et al. (2006)), and the
red sequence appears in high z clusters (e.g., Zirm et al. 2008).
The nature of the physical processes responsible for this
rapid rise remains unclear. A large number of massive (≃
1011M⊙) actively star–forming galaxies is clearly in place at
z ≃ 2 (Daddi et al. 2004; Papovich et al. 2007). Within this pop-
ulation, the more massive galaxies tend to be the more actively
star–forming (Daddi et al. 2007b), at variance with trends mea-
sured in the local Universe.
At the same time, galaxies with low levels of star–formation
rates at z ≃ 1.5 − 2 have been detected by imaging surveys
based on color criteria (e.g., Daddi et al. 2004) or SED fit-
ting (Grazian et al. 2007), and by spectroscopic observations
of red galaxy samples (Cimatti et al. 2004; Saracco et al. 2005;
Kriek et al. 2006). These results have motivated the inclusion
of efficient methods for providing a rapid assembly of mas-
sive galaxies at high z (such as starburst during interactions) as
Send
fontana@oa-roma.inaf.it
offprint requeststo: A.Fontana, e-mail:
well as quenching of the SFR, most notably via AGN feedback
(Menci et al. 2006; Bower et al. 2006; Hopkins et al. 2008).
Unfortunately, a detailed validation of the prediction of the-
oretical models has been hampered so far by the lack of a sta-
tistically well defined sample of early–type galaxies at high red-
shift, and by the difficulty in defining a common criterium to
identify early–type galaxies in the data. It is difficult to iso-
late passively evolving galaxies from the wider population of
intrinsically red galaxies at high redshift, which include also
a (probably larger) fraction of dust-enshrouded star–forming
galaxies. The two classes are indeed indistinguishable when
selected by means of a single color criterium, such as the
“ERO” classification (R − K > 4) (Daddi et al. 2000; McCarthy
2004) or the “DRG” one (J − K > 2) (Franx et al. 2003;
van Dokkum et al. 2003). The corresponding SEDs, however,
are not degenerate and can be distinguished with more com-
plex criteria, even when spectroscopy is not feasible. Some of
these criteria adopt more colors, such as the I − J/J − K method
proposed by Pozzetti & Mannucci (2000) and spectroscopically
validated by Cimatti et al. (2003), or the BzK method proposed
by Daddi et al. (2004). Other methods rely on the spectral fit-
ting of the overall SED, either by making use of the resulting
rest–frame colors, as in the case of the U − V/V − I criteria pro-
posed by Wuyts et al. (2007), or directly using the output of the
SED fitting process (Arnouts et al. 2007; Salimbeni et al. 2008;
Grazian et al. 2007).
We used themethodof usingSED fittingoutputin ouranaly-
sis oftheGOODS-S dataset,to disentanglethe differentcontrib-
utors to the mass density at high redshift (Grazian et al. 2007)
and to describe the evolution in the luminosity function of red
galaxies up to z ≃ 3 (Salimbeni et al. 2008).

Page 2

2 A. Fontana et al.: The fraction of quiescent massive galaxies in the early Universe
In this work, we use additional information available from
observations in the mid-IR 24µm band, to allow a more careful
selection of high redshift passively evolving galaxies, with the
specific aim of comparing their number density with theoretical
expectations.
We use a revised version of our GOODS-MUSIC sample
(Grazian et al. 2006), a 15 band multicolor (U–to–24µm ) cat-
alog extracted from an area of 143.2 arcminutes squared within
the GOODS–South public survey. The key improvements com-
pared to our previous work are a revised IRAC photometry
and the addition of 24µm flux data for all objects in the cata-
log. Full details are given in a companion paper (Santini et al.
2009, S09 in the following). Another important difference is
that we include all objects detected in the 4.5µm image, down to
m45< 23.5. In the following, we adopt a mass-selected sample,
obtained by applying a mass threshold at M∗ ≥ 7 × 1010M⊙to
ourphotometriccatalogbasedona combinedselection K < 23.5
or m45< 23.5. This photometric sample is complete at this mass
limit to z ≃ 4, also for dust–absorbed star–forming galaxies (see
F06).
Observed and rest–frame magnitudes are in the AB sys-
tem, and we adopt the Λ-CDM concordance model (H0 =
70km/s/Mpc, ΩM= 0.3 and ΩΛ= 0.7).
2. Quiescent galaxies
2.1. The role of mid-IR observations
In principle, the observed mid–IR flux is a powerful tool for dis-
tinguishing between the two classes of red galaxies at high red-
shift. Dust–absorbed star–forming galaxies are expected to be
bright in the mid–IR, where most of the UV–light absorbed by
dust is re–emitted. In contrast, passively evolving galaxies are
expected to be far dimmer, since ordinary stars have low emis-
sion at IR wavelengths. To quantify this criterium, we use the
ratio f(24µm)/f(K) between the observed flux at 24µm and the
K–band flux. We show the f(24µm)/f(K) for all galaxies in our
sampleinFig.1,infourredshiftbins.Toshowhowthiscolorcan
help us to differentiate between the galaxy types, we computed
the same f(24µm)/f(K) to z ≃ 4 for a set of local templates in-
cluding early–type, spiral, and starburst galaxies (Polletta et al.
2007). All templates were k–corrected at the different redshifts.
As shown in Fig. 1, the loci populated by the different galaxy
types in the U − V versus f(24µm)/f(K) plane are separated
well, allowing us to distinguish between actively star–forming
galaxies and those with moderate–to–low star formation.
In Fig. 1, we show the observed f(24µm)/f(K) for the com-
plete M∗ ≥ 7 × 1010M⊙sample, both for individual galaxies
as well as its general distribution. Upper limits were assumed
to be equal to the 1σ upper limits in the 24µm photometry. It
is clearly shown in Fig. 1 that the f(24µm)/f(K) ratio of star–
forming galaxies increases with redshift, and that a significant
population of starburst exists at z > 1, with colors similar to
those of local, rarer LIRGs. On the other hand, few bright galax-
ies at low and intermediate redshifts, with low star–formation
rates are detected at low levels in the 24µm image.
The most striking result is that f(24µm)/f(K) distribution
is clearly bimodal, its minimum corresponding to a gap be-
tween the loci of active and passive galaxies. The observed
minima in the distribution of the f(24µm)/f(K) ratio occurs at
f(24µm)/f(K) = 1,2,5 and 6 in the four redshift bins adopted
in Fig. 1. Two warnings are requiredabout this bimodality.First,
it is enhanced by the upper limits at faint 24µm levels, the true
distribution probably being wider even at lower f(24µm)/f(K).
Fig.1. Ratio of the observed fluxes at 24µm to that in the K
band as a function of the rest-frame U −V color, in four redshift
bins, for the M∗≥ 7 × 1010M⊙sample. Upper limits correspond
to objects fainter than 20µJy in the 24µm image. Galaxies with
age/τ > 6 (as discussed in Sect. 3) are shown in magenta. Large
blue circles are the “obscured AGN” candidates of Fiore et al.
(2008). Closed areas represent the range of f(24µm)/f(K) ob-
tained by redshiftinga set of local samples (Polletta et al. 2007):
frombottomtotop,early–typegalaxies(blackline),spirals(blue
line), starbursts (red line).
More importantly, the bimodality disappears if a deeper cata-
log is used (for instance, the z ≤ 26 selected version of the
GOODS–MUSIC catalog). The high mass-limit of our sample,
indeed, preferentially selects extremely red galaxies, as shown
by their significantly red U −V color, which are those for which
the bimodality is clearly evident. This is a key feature of our ap-
proach, since the f(24µm)/f(K) ratio provides a natural way of
distinguishing between active and quiescent galaxies within the
population of high–redshift, red galaxies.
2.2. The selection on rest–frame quantities
Thecriteriumthat we adoptedtoidentifyquiescentgalaxiescan-
not be directly applied to the output of current theoretical mod-
els for galaxy formation and evolution, which typically do not
predict the mid–IR flux. The comparison becomes more natural
when we convert the data into the quantities immediately pro-

Page 3

A. Fontana et al.: The fraction of quiescent massive galaxies in the early Universe3
Fig.2.
the K band as a function of the Specific Star–formation Rate
(˙ M/M∗), for the sample with M∗ ≥ 7 × 1010M⊙. Galaxies with
˙ M/M∗< 10−12yr−1have been arbitrarily set to ˙ M/M∗= 10−12
yr−1. Upper limits refer to objects fainter than 20µJy in the
24µm image. Large blue circles are the “obscured AGN” can-
didates of Fiore et al. (2008). The solid vertical line corresponds
to the inverse of the age of the Universe at the centre of the red-
shift bin. The dashed vertical line shows the threshold on ˙ M/M∗
adoptedto classify “red and dead” galaxies. The horizontalsolid
line refers to the minimum in the observed distribution of the
f(24µm)/f(K) ratio in the same redshift range.
Ratio between the observed fluxes at 24µm and in
vided by these models, i.e., stellar mass or star–formation rates.
We use the stellar masses M∗providedby the SED fitting, of ac-
curacy described at length in F06, and briefly in Sect. 3. To de-
rivestar–formationrates forgalaxiesdetectedat 24µm,we apply
the method of Papovich et al. (2007) in converting the 24µm +
rest–frame UV luminosity into a total star–formation rate, using
the Dale & Helou (2002) models with a (lowering) correction at
high mid–IR fluxes. For galaxies below 20µJy at 24µm, we use
the SFR derived from the SED fitting. As we show in S09 (see
also Daddi et al. 2007b), these two SFR estimators agree rela-
tively well, especially at low SFR levels.
In Fig.2,we showthe
f(24µm)/f(K) ratio and ˙ M/M∗ in our sample. For simplicity,
we only plot the two redshift ranges that are more populated in
our sample: the same trend holds at higher and lower redshifts.
This correlation is largely expected given the relationship
between ˙ M and the 24µm flux on the one hand, and between
M∗and the K magnitude on the other. The key point, however,
is that the correlation is so tight that it allows us to translate
the (observational) criterium based on the f(24µm)/f(K) ratio
(horizontal line in Fig. 2) into a (model–oriented) cut to the
correlation betweenthe
Fig.3. Probability distribution of the ˙ M/M∗ratio in the sample
of “red and dead” galaxies at 1.5 ≤ z ≤ 2.5, averaged over the
whole sample. The thick curve shows the probability distribu-
tion of the SED fitting to the 14 bands SED, from the U band
to the 8µm band. The thin curve shows the probability distribu-
tion removing models that over predict the SFR with respect to
the upper limit provided by the constraint at 24µm. The dashed
vertical line shows the limit ˙ M/M∗< 10−11yr−1that we adopt to
define “red and dead” galaxies.
estimated˙ M/M∗(vertical line in Fig. 2). As shown in Fig. 2, the
two samples in practice coincide.
Based on this evidence, we use the specific star–formation
rate˙ M/M∗toselect quiescentgalaxies,whichallowsdirectcom-
parison of the data with theoretical models. Given that ˙ M/M∗is
dimensionally the inverse of a timescale, a natural threshold of
˙ M/M∗is the inverseofthe ageof theUniverseat the correspond-
ing redshift (tU(z))−1. We define galaxies with ˙ M/M∗< (tU(z))−1
as “quiescent” in the following. Such a name is motivated by the
fact that - by definition- these galaxieshaveexperienceda major
episode of star–formation prior to the observations1.
3. “Red and dead” galaxies
3.1. Basic definitions
Galaxies definedas “quiescent”by the criteriumpresentedin the
previous Sect. may still have a measurable amount of ongoing
SFR (e.g., ˙ M ≃ 10M⊙/yr for a M∗ ≃ 1011M⊙ galaxy). It is
thereforeinterestingto estimate the fractionof galaxieswith low
or negligible levels of SFR. As we show in the following, the
physicalpropertiesoftheseobjects,thatwelabel“redanddead”,
constitute a sterner test to theoretical models.
To select galaxies at high redshift on the basis of low star–
formationrates, weneedto complementour f(24µm)/f(K)ratio
data with output from the fitting analysis of the optical–IR ob-
served SED. The lack of detection in the 24µm image provides
only an upper limit to the ongoing SFR (see for instance Fig. 2),
and its exact level can be estimated from the SED fitting only.
We applied the SED fitting technique following the recipe de-
scribed in several papers (see Fontana et al. (2004) and F06 for
details). The U-to-8µm photometry was compared with a grid
of models from the Bruzual & Charlot (2003) (BC03) spectral
synthesis code, characterized by exponentially declining star–
formation histories of timescale τ for a set of ages, metallicities,
and dust extinctions. For comparison with our previous work,
and most of the literature, we used the standard Salpeter IMF
1Indeed, if M∗= ?˙ M?past×tU(z), where ?˙ M?pastis the star–formation
rate averaged over the whole age of the Universe at the corresponding
z, the requirement ˙ M/M∗< (tU(z))−1implies ˙ M ≤ ?˙ M?past.

Page 4

4 A. Fontana et al.: The fraction of quiescent massive galaxies in the early Universe
and the BC03 models. We verified however that adoption of the
most recent version of the code incorporating the treatment of
the post–AGB stars (Bruzual A 2007) does not change our re-
sults significantly (see Salimbeni et al. 2009 for a preliminary
analysis). In particular, only 1 (out of 144) objects classified as
“red and dead” would not be classified as such with the new ver-
sion.
To select “red and dead” galaxies, we use a threshold of
˙ M/M∗< 10−11yr−1(vertical dashed line in Fig. 2). To some ex-
tent, this threshold is arbitrary, the important point being that
we adopt the same cut in theoretical models, to ensure proper
comparison. However, we note that the same threshold was also
used previously (Brinchmann et al. 2004) to separate active and
quiescent galaxies. We also note that it corresponds to a thresh-
old age/τ > 6 between the fitted age and the star–formation
exponential timescale τ. This follows from the fact that, assum-
ing exponentiallydeclining star–formationhistories, the specific
star–formation rate can be computed analytically to be ˙ M/M∗=
(τ(e(age/τ)− 1))−1, yielding ˙ M/M∗ ≃ 10−11yr−1for age/τ ≃ 6,
for small values of τ. This coincidence allows us to compare
the pure SED fitting approach with the additional analysis pro-
videdbythe24µm data.A thresholdto age/τwas appliedbefore
(Arnouts et al. 2007; Salimbeni et al. 2008; Grazian et al. 2007)
to broadly identify passively evolving galaxies, although with
a less conservative value of age/τ > 4. In Fig. 1, we plot the
location in the U − V vs. f(24µm)/f(K) plane of all galaxies
with age/τ > 6 (magenta points). The classification scheme
based purely on the optical–near IR SED fitting agrees well
with the f(24µm)/f(K) criterium, providing an important con-
sistency check. The main exception are a number of “Compton
thick AGN” candidates at 1.5 < z < 2.5, selected as described in
Fiore et al. (2008), where the 24µm emission was attributed to
AGN activity. These objects are discussed further in Sect. 3.4.
We select “red and dead” galaxies by requiringthat˙ M/M∗<
10−11yr−1, not only that age/τ > 6. This is because when es-
timating SFR, we assign the SED–derived values only to ob-
jects undetected in the mid–IR, i.e., those with 24µm flux below
20µJy. For this reason, the few objects fitted with age/τ > 6 but
with large f(24µm)/f(K) are not included in our sample of “red
and dead” galaxies. We now discuss the accuracy of this method
for the most interesting subsample, i.e., “red and dead” galaxies
at z ≥ 1.5.
3.2. “Red and dead” galaxies at 1.5 ≤ z ≤ 2.5
A sizeable fraction of the sample with ˙ M/M∗ < 10−11yr−1
have redshifts in the range 1.5 ≤ z ≤ 2.5, where we detect
32 galaxies. The reliability of their ˙ M/M∗measure can be as-
sessed using the error analysis already widely adopted in similar
cases (Papovich et al. 2001, F06). Briefly, the full synthetic li-
brary used to determine the best-fit model spectrum is compared
with the observed SED of each galaxy. For each spectral model
(i.e., for each combination of the free parameters age, τ, Z and
E(B − V)) the probability P of the resulting χ2is computed and
retained, along with the associated ˙ M and M∗. The ensemble of
˙ M/M∗values correspondingto probabilities above a given mini-
mum thresholddefines the range of acceptablevalues for˙ M/M∗.
Since mostof our“redanddead”galaxieshaveonlya photomet-
ric redshift, our analysis for these objects has been completedby
allowing the redshift parameterto be free. In Fig. 3, we show the
resulting distribution of the probability P, averaged over the en-
tire sample of “red and dead” galaxies at 1.5 ≤ z ≤ 2.5. The tail
at large values of ˙ M/M∗is due to 2 objects with a nearly equal
probability of being fitted accurately by a dusty starburst model
solution. The correspondingly large values of ˙ M that would be
required would easily overpredictthe observed flux (which is an
upper limit) at 24µm. Taking into account this additional con-
straint, we display in Fig. 3 the distribution of P after removing
the models that overpredict the SFR with respect to the value
provided by the upper limit in the emission at 24µm. It is clear
from Fig. 3 that the SED fitting for these galaxies is well con-
strained at levels of ˙ M/M∗< 10−11yr−1, since the probability of
having˙ M/M∗has only a small tail above 10−11yr−1.
This result is unsurprising. Since the galaxies in our sam-
ple are by selection massive (M∗ ≥ 7 × 1010M⊙), the thresh-
old
˙ M/M∗ = 10−11yr−1is equivalent to small but measur-
able amounts of SFR, of approximately 1M⊙yr−1. These lev-
els correspond to detectable fluxes in the deep B-z optical im-
ages of GOODS (the magnitude computed from a galaxy with
≃ 1M⊙yr−1, for small values of E(B − V), is in the range 26-28
mags) and can then be measured by the full SED fitting.
We can also check the small subset of our “red and dead”
sample with spectroscopic observations in this redshift range.
Using both the K20 and the public GOODS data set, we found
6 objects that have a spectroscopic redshift. Four of these ob-
jects areclassifiedas earlyspectraltype,withnodetectable[OII]
emission line. Two other objects have both absorption features,
typical of evolved systems, and a weak [OII] line. A weak [OII]
line is not necessarily an indicator of ongoing star–formation
(Yan et al. 2006).AdoptingthestandardconversionofKennicutt
(1998),wederivestarformationratesoftheorderof2-3 M⊙yr−1,
consistent with the SED fitting estimates, confirming that these
objects have ˙ M/M∗ ? 10−11yr−1. We conclude that our analy-
sis is sufficiently robust at 1.5 < z < 2.5, and allows to build a
mass–selected sample of “red and dead” galaxies.
3.3. “Red and dead” galaxies at z > 2.5
As we move to higher redshifts, 2.5 ≤ z ≤ 4, our analysis be-
comes more uncertain. First, the constraints on the 24µm emis-
sion are weaker or non–existent, since this band is shifted out-
side the range dominated by dust emission. Objects also become
even fainter and redder, and in general lack any spectroscopic
confirmation. In this redshift range, we detect 12 “red and dead”
candidates, 9 located at 2.5 < zphot < 3. We carefully exam-
ined the data for each object to assess the detection reliability.
First, we verified that the photometry was not contaminated by
nearby companions, and that the overall SED was smooth and
not biased obviously by photometric fakes. All galaxies selected
in this redshift range are very red with z−B > 4, which is redder
than typical EROs at lower redshifts. Their optical–IR SED is
dominated by a break between the K and the IRAC bands, and
a change of slope that is a signature of an evolved galaxy pop-
ulation (star–forming, dusty objects exhibit a more featureless
slope). The redshift probability distribution is broad, but typi-
cally implies that zphot > 1.5 − 2 with a relatively large spread
∆zphot≃ 0.5.
Quantitatively, the error analysis described above indicated
that 5 of the 9 objects at 2.5 < zphot< 3 are constrained to have
˙ M/M∗ < 10−11yr−1, even for such a broad redshift range. An
example of these objects is shown in the upper panel of Fig. 4.
The four other objects have a wider distribution of˙ M/M (Fig. 4,
middle panel), as well as redshift.
In a similar way, the physical propertiesand the nature of the
threeobjectsdetectedatz > 3areweaklyconstrained.Theyhave
a clear minimumin the χ2distributionaroundzphot≃ 3.5−4,but

Page 5

A. Fontana et al.: The fraction of quiescent massive galaxies in the early Universe5
Fig.4. Examples of “red and dead” galaxies at z > 2.5.For each
object, from left to right: the observed flux in the GOODS band
andthe best-fit SED; the redshiftprobabilityfunction;the˙ M/M∗
probability distribution. From top to bottom, the three galaxies
shown represents three different categories: a robust candidateat
2.5 < z < 3; a less secure candidate in the same redshift range; a
typical example of the candidates at z > 3.
have a tail to z ≃ 2. The distribution of acceptable˙ M/M∗values
also extends significantly beyond 10−11yr−1.
We conclude that at least 55% of our “red and dead” candi-
dates at 2.5 < zphot< 3 are robust candidates, while the remain-
ing 45% and the three candidates at z > 3 are more uncertain.
It would be natural to convert this finding into an upper limit to
the true number density of “red and dead” galaxies. However,
we note that a comparable number of galaxies are inferred to
have ˙ M/M∗slightly above 10−11yr−1and are not included in our
sample, but have a range of acceptable model fits extendingwell
below ˙ M/M∗ = 10−11yr−1. We conclude that at z > 2.5 the es-
timate of ˙ M/M∗becomes unreliable, due mainly to the limita-
tions in the depth of existing observations, and that the resulting
statistical analysis of “red and dead” galaxies must be treated
with caution. As we show, the statistical error (including both
Poisson and cosmic–variance components) is so large for these
small samples that a more rigorous treatment of these uncertain-
ties is probably unnecessary.
3.4. Hosts of Compton thick AGNs
A potential source of uncertainties in many statistical analysis
of high redshift galaxies is contamination by AGNs. As we de-
scribe in S09, we removed from our catalog both spectroscopi-
cally confirmedAGNs and X-raysources with an optically dom-
inant point–like source. We also identified galaxies that likely
harbor Compton thick AGNs using the criterium defined by
Fiore et al. (2008), which is based on the detection of a mid–IR
excess in very red galaxies. This criterium is similar to the one
adoptedby Daddi et al. (2007a) but also includesobjects that are
muchredderin the optical–IRrangethan the BzK–selected sam-
ple of Daddi et al. (2007a). A full discussion of the SED of these
objects will be presented elsewhere: for the moment, we note
that9ofthese“ComptonthickAGN”candidatesat1.5 < z < 2.5
are classified as “red and dead” galaxies, as shown in Fig. 1.
Unfortunately, the nature of these objects remains elusive. Most
of these objects are among the reddest in our sample, as indi-
cated by their extreme rest-frame U − V. They are extremely
faint or even undetected in the deep z-band ACS images, and
are included in our sample only because of their detection in ei-
ther K or 4.5µm images. As a result, the probabilitydistributions
of their photometric redshifts are often broad, although they are
typically constrained to be at zphot > 2. Only three (of nine)
galaxies have the distribution of ˙ M/M∗constrained to be below
10−11yr−1. In the following, we do not rely on this classifica-
tion, and we therefore compute the fraction of “red and dead”
galaxies at 1.5 < z < 2.5 in three different ways (labelled a),
b), c) in the caption of Fig. 5), to explore all possible cases.
First (option a)), we remove all “Compton thick AGN” candi-
dates from our sample, irrespective of their SED classification.
Alternatively (option b)), we assume that none of them is a “red
and dead” galaxy, which provides a lower limit to their fraction.
Finally (option c)), we include them in our statistical analysis,
assigning their formal SED classification. As we show in the
following Sect., the fraction computed with these different as-
sumptions changes its value by ≃20%, still providinginteresting
constraints on theoretical models.
4. Results and discussion
We finally compute the fraction of “quiescent” and “red and
dead” galaxies in different redshift ranges, following the def-
initions provided in the last two Sections. These fractions are
shown in Fig.5 as a function of redshift. The different numbers
of “red and dead” galaxies at 1.5 < z < 2.5 reflect the differ-
ent accounting methods of hosts of Compton thick AGNs. Error
bars were computed by summing (in quadrature) the Poisson
and cosmic variance error. The latter was computed by measur-
ing the relative variance within 200 samples bootstrapped from
the Millennium Simulation (Kitzbichler & White 2007), using
an area as large as GOODS-S and applying the same selection
criteria.
We recall that these fractions are computed with respect to
the total number of galaxies with M∗≥ 7 × 1010M⊙in our sam-
ple. Since we are interested in the ratio between galaxy classes,
the impact of the exact choice of this thresholdis measurablebut
not dramatic. We verified that by either increasing or decreasing
the mass limit cut by a factor of 2, the fraction of both “qui-
escent” and “red and dead” galaxies in different redshift ranges
changes by about 0.1, in the data (where applicable) as well as
in the theoretical models. This factor of 2 variation may also be
produced by adopting a different IMF, and is similar to the un-
certainty in the stellar mass estimate.
Ouranalysis confirmsthe cosmologicaldecreasein the num-
ber density of massive early–typegalaxies at high redshifts: qui-
escent galaxies dominate the population of massive galaxies at
z << 1, and become progressively less common at higher z.
However,we note that a significant fraction of galaxies with low
levels of SFR is in place even at the highest redshifts sampled
here (z ≃ 3.5) with a fraction of about 10-15% at z > 2.
By definition, “quiescent”,massive galaxies assembled most
of their stellar mass in previousepochs, implyingthat they expe-
riencedanactive starburstphase orimportantmergingprocessat
higher redshifts. The large fraction of “quiescent” galaxies that
we observe at z ≃ 2 implies that these processes must have been

Page 6

6 A. Fontana et al.: The fraction of quiescent massive galaxies in the early Universe
Fig.5. Lower Fraction of “red and dead” galaxies (defined by
˙ M/M∗ < 10−11yr−1) as a function of redshift, in the M∗ ≥
7×1010M⊙mass–selected sample. Points representthe observed
values. Filled, open and starred points (slightly offset for clarity)
refer to the three different strategies a,b,c discussed in the text to
accountfor obscuredAGNs. Error bars include Poisson and cos-
mic variance errors. Lines refer to the predictions of theoretical
models, as described by the legend: Menci et al. (2006) (M06),
Fontanot et al. (2007) (F07), Kitzbichler & White (2007)(K07),
Nagamine et al. (2006), (N06); Upper Fraction of “quiescent”
galaxies as a function of redshift, defined by ˙ M/M∗< (tU(z))−1,
in the same mass–selected sample. Points and lines as in the
lower panel.
frequent at very high redshifts. The upper panel of Fig. 5 can
also be interpreted in terms of “active” galaxies, i.e., the com-
plementary galaxy fraction with ˙ M/M∗ ≥ (tU(z))−1. According
to the duty-cycle argument presented in S09, these actively star
forminggalaxiesexperienceda majorepisode ofstar–formation,
potentially building up a substantial fraction of their stellar mass
in this episode. Their fraction increases with redshift, constitut-
ing more than 50% of the massive galaxy population at z ≥ 2.
Finally, the sizeable number of “red and dead” galaxies is
already in place at z > 1.5, implying that the star–formation
episodes must be quenched either by efficient feedback mecha-
nism and/or by the stochastic nature of the hierarchical merging
process.
It is interesting to determine whether theoretical models
agree with these observational results. In Fig. 5, we plot the pre-
dictions of several models, applying the same criteria based on
the ˙ M/M∗ values. We consider purely semi-analytical models
(Menci et al. (2006), M06, Fontanot et al. (2007), F07), a semi-
analytical rendition of the Millennium N-body dark matter sim-
ulation (Kitzbichler & White (2007), K07), and purely hydrody-
namicalsimulations(Nagamine et al.(2006),N06).Thefinalare
presented for three different timescales τ of the star–formation
rate (ranging from 2 × 107yrs to 2 × 108yrs), and represented
with a shaded area. All these models agree in predicting a grad-
ual decline with redshift in the fraction of galaxies with a low
SFR.
As far as the “quiescent” fraction is concerned, we note that
most models agree quantitatively in their predictions at all red-
shifts. There is a general tendency to slightly overpredict the
fraction of “quiescent” galaxies, although the relatively large
error bars of our sample prevent firm conclusions from being
drawn. As we show in more detail in S09 (see also Daddi et al.
2007b), this is the result of an overall underestimate of the me-
dian SSFR of massive galaxies, which increases the number of
mildly star–forming galaxies that we detect within our selection
criteria ˙ M/M∗ < (tU(z))−1. The F07 models provides a notice-
able exception to this process. A main success of this model,
indeed, is its capability of reproducing the Scuba counts and
the high associated SFR (Fontanot et al. 2007): it is unsurpris-
ing that it also predicts a large fraction of active galaxies, and
hence a smaller fraction of “quiescent” ones.
Conversely,the predictedfractionof “red and dead” galaxies
variessignificantlyat all redshifts.This revealsthatthepredicted
fraction of galaxies with very low levels of SFR is a particularly
sensitive quantity, and provides a powerful way of highlighting
the differences between the models. Some models (M06, F07)
underpredict the fraction of “red and dead” galaxies at all red-
shifts, and in particular predict virtually no object at z > 2, in
contrast to what observed. The Millennium-based model agrees
with the observed quantities, while the hydro model appears to
overpredict them.
It is beyond the scope of the present paper to discuss the ori-
gins of these differences. They are difficult to ascertain, because
of the complex interplay between all the physical processes in-
volved in these models, the different physical process imple-
mented - most notablythose related to AGN feedback- and their
different technical implementations. The failure of most models
to reproducesimultaneouslythe fraction of “quiescent”and “red
and dead” massive galaxies in the early Universe probably im-
plies that the balance between the amount of cool gas and the
star–formation efficiency on the one side, and the different feed-
back mechanisms on the other, is still poorly understood.
Acknowledgements. We are grateful to Mark Dickinson, Roberto Maiolino and
Pierluigi Monaco for the useful discussions. We also thank K. Nagamine for
providing the output of his models. We are also in debt with the two referees for
useful and prompt suggestions, that improved the presentation of the work. This
work is based on observations carried out with the Very Large Telescope at the
ESO Paranal Observatory under Program ID LP168.A-0485 and ID 170.A-0788
and the ESO Science Archive under Program IDs 64.O-0643, 66.A-0572, 68.A-
0544, 164.O-0561, 163.N-0210, and 60.A-9120. The Millennium Simulation
databases used in this paper and the web application providing online access
to them were constructed as part of the activities of the German Astrophysical
Virtual Observatory. We acknowledge financial contribution from contract ASI
I/016/07/0 (COFIS).
References
Abraham, R. G., Nair, P., McCarthy, P. J., et al. 2007, ApJ, 669, 184
Arnouts, S., Walcher, C. J., Le F` evre, O., et al. 2007, A&A, 476, 137
Berta, S., Lonsdale, C. J., Polletta, M., et al. 2007, A&A, 476, 151
Bower, R. G., Benson, A. J., Malbon, R., et al. 2006, MNRAS, 370, 645
Brinchmann, J., Charlot, S., White, S. D. M., et al. 2004, MNRAS, 351, 1151
Bruzual, G. & Charlot, S. 2003, Royal Astronomical Society, Monthly Notices,
344, 1000
Bruzual A, G. 2007, ArXiv Astrophysics e-prints
Cimatti, A., Daddi, E., Cassata, P., et al. 2003, A&A, 412, L1
Cimatti, A., Daddi, E., Renzini, A., et al. 2004, Nature, 430, 184
Daddi, E., Alexander, D. M., Dickinson, M., et al. 2007a, ApJ, 670, 173
Daddi, E., Cimatti, A., Pozzetti, L., et al. 2000, A&A, 361, 535
Daddi, E., Cimatti, A., Renzini, A., et al. 2004, ApJ, 617, 746
Daddi, E., Dickinson, M., Morrison, G., et al. 2007b, ApJ, 670, 156
Dale, D. A. & Helou, G. 2002, The Astrophysical Journal, 576, 159

Page 7

A. Fontana et al.: The fraction of quiescent massive galaxies in the early Universe7
Fiore, F., Grazian, A., Santini, P., et al. 2008, ApJ, 672, 94
Fontana, A., Pozzetti, L., Donnarumma, I., et al. 2004, A&A, 424, 23
Fontana, A., Salimbeni, S., Grazian, A., et al. 2006, Astronomy & Astrophysics,
459, 745
Fontanot, F., Monaco, P., Silva, L., & Grazian, A. 2007, MNRAS, 382, 903
Franceschini, A., Rodighiero, G., Cassata, P., et al. 2006, A&A, 453, 397
Franx, M., Labb´ e, I., Rudnick, G., et al. 2003, ApJ, 587, L79
Grazian, A., Fontana, A., De Santis, C., et al. 2006, Astronomy & Astrophysics,
449, 951
Grazian, A., Salimbeni, S., Pentericci, L., et al. 2007, Astronomy &
Astrophysics, 465, 393
Hopkins, P. F., Cox, T. J., Kereˇ s, D., & Hernquist, L. 2008, ApJS, 175, 390
Kennicutt, Jr., R. C. 1998, ARA&A, 36, 189
Kitzbichler, M. G. & White, S. D. M. 2007, Royal Astronomical Society,
Monthly Notices, 376, 2
Kriek, M., van Dokkum, P. G., Franx, M., et al. 2006, ApJ, 649, L71
McCarthy, P. J. 2004, ARA&A, 42, 477
Menci, N., Fontana, A., Giallongo, E., Grazian, A., & Salimbeni, S. 2006, The
Astrophysical Journal, 647, 753
Nagamine, K., Ostriker, J. P., Fukugita, M., & Cen, R. 2006, ApJ, 653, 881
Papovich, C., Dickinson, M., & Ferguson, H. C. 2001, ApJ, 559, 620
Papovich, C., Rudnick, G., Le Floc’h, E., et al. 2007, ApJ, 668, 45
Polletta, M., Tajer, M., Maraschi, L., et al. 2007, The Astrophysical Journal, 663,
81
Pozzetti, L. & Mannucci, F. 2000, MNRAS, 317, L17
Salimbeni, S., Giallongo, E., Menci, N., et al. 2008, A&A, 477, 763
Santini, P., Fontana, A., Grazian, A., et al. 2009, Submitted to A&A
Saracco, P., Longhetti, M., Severgnini, P., et al. 2005, MNRAS, 357, L40
van Dokkum, P. G., F¨ orster Schreiber, N. M., Franx, M., et al. 2003, ApJ, 587,
L83
Wuyts, S., Labb´ e, I., Franx, M., et al. 2007, ApJ, 655, 51
Yan, R., Newman, J. A., Faber, S. M., et al. 2006, ApJ, 648, 281
Zirm, A. W., Stanford, S. A., Postman, M., et al. 2008, ApJ, 680, 224