Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.arcetri.astro.it/science/k20/papers/0411768.ps Äàòà èçìåíåíèÿ: Tue Nov 30 19:23:32 2004 Äàòà èíäåêñèðîâàíèÿ: Fri Feb 28 06:57:55 2014 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: ring galaxy

arXiv:astro­ph/0411768
v1
29
Nov
2004
Mon. Not. R. Astron. Soc. 000, 1{19 (2004) Printed 30 November 2004 (MN L A T E X style le v2.2)
The evolution of the galaxy B-band rest-frame morphology
to z  2: new clues from the K20/GOODS sample
P. Cassata 1? , A. Cimatti 2 , A. Franceschini 1 , E. Daddi 5 , E. Pignatelli 3 ,
G. Fasano 3 , G. Rodighiero 1 , L. Pozzetti 4 , M. Mignoli 4 and A. Renzini 5
1 Dipartimento di Astronomia, Vicolo Osservatorio 2, I-35122, Padova, Italy
2 INAF, Osservatorio Astronomico di Arcetri, Largo E. Fermi 5, I-50125 Firenze, Italy
3 INAF, Osservatorio Astronomico di Padova, Vicolo Osservatorio 2, I-35122, Padova, Italy
4 INAF, Osservatorio Astronomico di Bologna, Via Ranzani 1, I-40124, Bologna, Italy
5 European Southern Observatory, Karl-Schwarzschild-Str. 2, D-85748, Garching, Germany
ABSTRACT
We present a detailed analysis of the evolution of the rest-frame B-band morphology
of K-selected galaxies with 0 < z < 2:5. This work is based on the K20 spectro-
scopic sample (Ks < 20) located within the Chandra Deep Field South area, coupled
with the public deep GOODS HST+ACS multi-band optical imaging available in that
eld. Thanks to the spectroscopic completeness of this catalog reaching 94%, we can
compare the morphological and spectroscopic properties of galaxies with unprece-
dented detail. Our morphological analysis includes visual inspection and automatic
procedures using both parametric (e.g. the Sersic indices treated by the GALFIT
and GASPHOT packages) and non-parametric (the Concentration, Asymmetry and
clumpineSs, CAS) methods. By exploiting the 4-band deep ACS imaging we account
in detail for the morphological K-correction as a function of the redshift and show
that, while the parametric methods do not eôciently separate early- and late-type
galaxies, non-parametric ones prove more eôcient and reliable. Our analysis classies
the K20 galaxies as: 60/300 (20%, class 1) normal ellipticals/S0; 14/300 (4%, class 2)
perturbed or peculiar ellipticals; 80/300 (27%, class 3) normal spirals; 48/300 (16%,
class 4) perturbed or actively star-forming spirals; 98/300 (33%, class 5) irregulars.
The morphological and spectroscopic classications are compatible with each other
for more than 90% of the sample galaxies, while 7 class-1 E/S0's show emission lines
and 11 spirals and irregulars (class 3+4+5) have purely absorption-line spectra. The
evolution of the merging fraction is constrained up to z  2, by carefully account-
ing eects of morphological k-correction: both asymmetry criterion and pair statistic
show an increasing merging fraction as a function of the redshift. We nally analyse
the redshift-dependence of the eective radii for early- and late-type galaxies and nd
some mild evidence for a decrease with z of the early-type galaxy sizes, while disks
and irregulars remain constant. Altogether, this analysis of the K20 sample indicates
the large predominance of spirals and irregulars at 0:5 < z < 1:5 in K-band selected
samples at even moderate depths.
Key words: galaxies: evolution { galaxies: interactions { galaxies: structure
1 INTRODUCTION
Galaxies in the local universe can be organized in a sequence
of morphologies (e.g. the Hubble sequence) which must be
the result of the specic processes having originated them.
Yet, the relative roles over cosmic time of processes such as
? E-mail: cassata@pd.astro.it
merging of dark matter halos, dissipation, starburst, feed-
back, AGN activity, etc. remain largely conjectural, in par-
ticular concerning the establishment of the galaxy morpho-
logical dierentiation. Therefore, morphological analyses of
faint high-z galaxies and studies of the evolution of galaxy
sizes with cosmic time give an important insight on how the
matter aggregated into the structures that we see today. The
combination of a morphological investigation for ux limited
c
2004 RAS

2 P.Cassata et al.
samples of faint galaxies with complete redshift information
provides decisive constraints on the formation epoch and the
pattern for the galaxy build-up.
While in the local universe approximately 65% of all
galaxies are spirals, 32% are ellipticals and 4% irregu-
lars/peculiars (Marzke et al. 1998), there are indications
that in the distant universe the irregular/peculiar fraction
becomes predominant (i.e. Glazebrook et al. 1995, Abra-
ham 1996). Van den Bergh et al. (2000, 2001) concluded
that most of the galaxies with z & 0:5 can be hardly classi-
ed within the Hubble scheme: irregular/peculiar/merging
objects become more common beyond z  0:3, spiral struc-
tures at z & 0:6 are more chaotic than locally and spirals and
ellipticals become extremely rare at z & 1:5. Conselice et al.
(2004) reached similar conclusions, adding that at z & 2
over 80% of the stellar mass is stored in peculiar galaxies.
Exploiting the same data set used in this work Cimatti
et al. (2003) investigated the morphology of the K20/CDFS
EROs, nding that only the  30% of these objects were
early-type galaxies.
As a continuation of the K20 survey, Daddi et al. (2004)
identied 9 actively star-forming/merging galaxies in the
K20/GOODS eld with 1:7 < z < 2:3 and stellar mass
M > 10 11 M , while Cimatti et al. (2004) found over the
same eld 4 passive, early-type galaxies with 1:6 < z < 1:9
with similar stellar masses. Thus, while at z = 0 most so
massive galaxies are elliptical/S0 galaxies, by z  2 star-
burst and passive galaxies appear to be in nearly equal num-
bers.
The Cold Dark Matter (CDM) models of galaxy forma-
tion predict that dark matter halos formed by the merging
of smaller units in the past. Thus, the evolution of the merg-
ing fraction with redshift gives also important cosmogonic
information (see for a review Abraham 1998). Several stud-
ies report an increasing merging fraction with redshift, both
at moderate (z < 0:3, Patton et al. 1997), and at higher
redshifts (Le Fevre et al. 2000, by pair counts; Conselice et
al. 2003, by the Asymmetry measurements).
However, in order to study the evolution with redshift
of galaxy morphologies and sizes, the dependence of mor-
phological properties on the wave-band and the eects of
the changing rest-frame wavelength as a function of redshift
(the so-called morphological K-correction) have to be taken
into account. Galaxies observed in their blue rest-frame ap-
pear with later-type morphologies than observed in the red
part of the spectrum (e.g. Windhorst et. al 2002; Papovich
et al. 2003).
The problem of quantifying morphological properties
of high redshift galaxies is still quite open. In the literature,
parametric and non-parametric methods are usually applied.
The former attempt to model the light distributions with a
combination of analytic laws, like de Vaucouleurs or expo-
nential proles, or the Sersic model (GALFIT, Peng et al.
2002 and GIM2D, Simard et al. 2002). Various parameters
(i.e. bulge/disk B/D ratios or Sersic indices) are then de-
rived, which are known to correlate with qualitative Hubble
classications. This approach suers however a number of
problems. It needs to assume that the galaxy light distri-
bution is well reproduced by a symmetric prole, hence is
not suitable to treat merging structures, spiral arms, dust
lanes and so on. Moreover, there are degeneracies in the so-
lutions for multi-component ts (e.g. needed to retrieve the
B/D ratio), because of the large number of free parameters.
This becomes quickly unmanageable at decreasing galaxy
brightness and number of galaxy pixels.
An often used representation of galaxy morphological
types is the parameter n associated to the Sersic prole
 / exp( 1=n), with some clear advantages over multi-
component ts (Pignatelli et al. 2004). Unfortunately, there
are even local ellipticals with a roughly exponential pro-
le, while only the most luminous and massive objects have
n  4 (i.e. Caon et al. 1993). This obviously implies an
intrinsic level of degeneracy.
Alternatively, non-parametric approaches to quantita-
tive morphology have been developed in the last few years
by several authors. The most used ones are the Concen-
tration (C) parameter (Abraham et al. 1996), that roughly
correlates with the Sersic index and with the B/D ratio, and
the Asymmetry (A) parameter (Abraham et al. 1996, Con-
selice et al. 2000), able in principle to distinguish irregulars
or mergers from more symmetric galaxies (E/S0/Sa).
More recently another parameter has been introduced,
the clumpiness (S), measuring the degree of \patchiness" of
a galaxy (i.e. the light in the high spatial frequencies). The
clumpiness is expected to correlate with the star-formation
rate (Conselice 2003). The advantage of the non-parametric
approach is that no a-priory assumption is made about the
distribution of light. Some problems however still remain,
like the in uence of the noise on the asymmetry and clumpi-
ness measures, or the choice of the spatial scale used to com-
pute clumpiness, that must be adapted to the distance and
the size of the object, or again the choice of the center of
rotation for the computation of the asymmetry.
One aspect that must be taken in consideration when
dealing with a morphological analysis is the degree of human
interaction needed to obtain the tting parameters, both for
non-parametric and for parametric procedures, in particular
when treating large amounts of data, as those made avail-
able by the new Advanced Camera for Surveys on HST (e.g.
the GOODS, COSMOS, UDF survey projects to mention a
few). The GALFIT tool (Peng et al. 2002), for example, has
the advantage to allow masking out bad-pixel zones of the
image (e.g. dust lanes or spiral arms), in order to improve
the t. However, this appears unfeasible object by object
over very large samples. GASPHOT, an automated tool re-
cently developed by Pignatelli, Fasano and Cassata (2004),
tting the galaxy light-prole with a Sersic model, has been
instead dened to retrieve in fully automatic mode funda-
mental photometric and morphological parameters (magni-
tudes, radii, axial ratios, Sersic indices) for all the objects
in a given image.
In this paper we exploit the very deep high resolu-
tion imaging recently obtained over the CDFS area with
ACS/HST taken in the GOODS/HST treasury program (Gi-
avalisco et al. 2004) to make a careful morphological study
of galaxies in the K20 sample within this area. In partic-
ular, we have exploited the multi-band coverage provided
by HST and the redshift information available for each ob-
ject to minimize the eects of morphological K-correction:
we have studied each object in the ACS band closer to the
B-band rest-frame. We also take advantage of the K-band
selection, that collects the light from low-mass stars domi-
nating the baryonic content of galaxies, and thus providing
a better mass completeness level. The K-band also allows
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 3
Figure 1. Some examples of galaxies representing morphological classes. Each box measures 4:5 00  4:5 00 . From top-left to bottom-right
we report: an isolated elliptical (class 1); a perturbed elliptical in which a little distortion of isophotes can be appreciated (class 2a); a
couple in which main object is an elliptical (class 2b); a normal spiral (class 3); a peculiar spiral, in which signs of interaction with a
little companion at bottom-right can be noted (class 4); an irregular galaxy (class 5).
to minimize the eects of the spectral K-correction, dust
absorption, and evolution.
We are particularly motivated to such an extensive ana-
lysis by the excellent optical spectroscopic follow-up ob-
tained for this sample with the ESO Very Large Telescope
in the last several years. This paper is devoted to the study
of galaxy morphologies and sizes as a function of redshift.
A forthcoming paper will expand on statistical analyses of
the galaxy distribution as a function of morphology and will
attempt to interpret these data with modellistic representa-
tions.
The galaxy sample is presented in Sect. 2. The
HST/ACS data for the morphological analysis are presented
in Sect. 3. In Sect. 4 we use parametric and non-parametric
analyses, in particular with GASPHOT and GALFIT and
the Concentration-Asymmetry-Clumpiness set, and com-
pare their results with those of a visual inspection. In Sect.
5 we discuss the evolution with the redshift of the inferred
fractions of each morphological class, including the merging
fraction. The high level of spectroscopic coverage ( 95%) is
exploited for comparison with the morphological classica-
tion in Sect. 6, while our constraints on the redshift evolution
of sizes for various morphological types are reported in Sect.
7. Our conclusion are drawn in Sect. 8.
2 THE SAMPLE
The original sample was selected in the K-band by the K20
team into two independent sky regions, one centered in the
Chandra Deep Field South (CDFS) covering 32.2 arcmin 2 ,
the second centered around the QSO 0055-269, and cov-
ers 19.8 arcmin 2 . Detailed informations can be found in
Cimatti et al. (2002).
The complete sample consists of 546 objects (including
a few stars) down to K = 20. In this work, we restrict our-
selves to the 346 objects lying in the CDFS area, for which
high resolution imaging by ACS/HST has recently become
available as a result of the GOODS HST Treasury program
(Giavalisco et al. 2004). After rejecting stars and quasars,
our nal sample includes 300 galaxies.
The available data include deep spectroscopy obtained
with VLT+FORS1 and VLT+FORS2, and has been re-
cently complemented with ESO/GOODS public data. In
the CDFS area the spectroscopic coverage is 92% with
K20 data only, and it reaches the 94% including spectra
by the ESO/GOODS public spectroscopic survey (Vanzella
et al. 2004). For the remaining objects we used through
out the paper the photometric redshifts derived using the
ESO/GOODS VLT+FORS1 BV IRz and VLT+ISAAC
JHKs public imaging (see Cimatti et al. 2002a).
c
2004 RAS, MNRAS 000, 1{19

4 P.Cassata et al.
Figure 2. A sample of galaxies with z  1 are shown in their z-band, corresponding to 4500  A rest-frame (left side of each panel),
and in their V-band, corresponding to 3000  A (right side of each panel). These examples are representative of morphological the
K-correction problem aecting galaxies observed in their U-band (that is galaxies having z & 1:5 observed in z-band) rather than in
their B-band rest-frame. The panels labelled with 1, 2 and 3 contains elliptical/S0 galaxies, panels with 4, 5 and 6 normal spirals, panels
with 7, 8 and 9 perturbed spirals and panels with 10, 11 and 12 irregular/merger galaxies. The size in arcsec of the images is reported
in each panel.
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 5
The sample includes objects belonging to two clusters
at 0:665 < z < 0:672 and 0:732 < z < 0:740 (see Cimatti et
al. 2002b), including respectively 14 and 32 galaxies.
Galaxy spectra have been automatically classied ac-
cording to their features into three main classes by Mignoli
et al. (2004): 1. early type; 2. early type + emission lines; 3.
pure emission line.
3 THE HST/ACS DATA
Ground based photometry and spectroscopy was comple-
mented with ACS imaging in the BV iz bands taken in the
GOODS=HST Treasury Program (Giavalisco et al. 2003).
We used the released version 1.0 of the images. The GOODS
ACS/HST Treasury Program has surveyed two separate
elds (the Chandra Deep Field South and the Hubble Deep
Field North) with four broad band lters: F435W (B),
F606W(V), F775W(i) and F850LP(z). Observations in the
V, i and z lters have been split into 5 epochs, separated by
about 45 days, in order to detect transient objects. Observa-
tions in the B band are taken during epoch 1 for both elds.
Images taken at consecutive epochs have position angles in-
creasing of 45 degrees. Total exposure times are 2.5, 2.5, 5
orbits in the V, i and z bands respectively. The exposure
time in the B-band is 3 orbits. In August 2003 the GOODS
team released the version 1.0 of the reduced, stacked and
mosaiced images for all the data acquired during the ve
epochs of observation. To improve the PSF sampling, the
original images, which have a scale of 0.05 arcsec/pixel, have
been drizzled onto images with a scale of 0.03 arcsec/pixel.
We have exploited the multi-band high resolution imag-
ing to study each objects in the ACS band closer to the B
rest-frame, in order to minimize eects of morphological K-
correction: the F435W lter is used in the range 0 < z < 0:2,
F606W in the range 0:2 < z < 0:55, F775W in the range
0:55 < z < 0:85 and F850LP for z > 0:85. The z band
is only a poor approximation of the B-band rest-frame for
objects with z & 1:2, approaching towards to the U-band
rest-frame. We will come back later to this problem.
4 DATA ANALYSIS
The morphological analysis has been performed in three
steps: (a) a visual inspection, in order to assign each ob-
ject to morphological classes based on the detected features
(spiral arms, tails, double nuclei...); (b) a surface-brightness
prole analysis performed with GASPHOT and GALFIT, in
order to quantify morphology and in particular to extract
Sersic indices; (c) a non-parametric analysis of the distribu-
tion of the galaxy light, using the measures of asymmetry,
concentration and clumpiness (or smoothness) to separate
dierent galaxy types. In the following we will use only the
results obtained by the visual inspection, reinforced by those
from automatic procedures. However, we are also interested
in identifying a completely automatic procedure able to seg-
regate at least early- from late-type galaxies with a low level
of interaction. This procedure will be useful for the incoming
new very large and deep surveys like GOODS and COSMOS.
4.1 Visual inspection and morphological
classication
The galaxies have been separated into ve morphological
classes (see Fig.1): 1. Early-type galaxies (ellipticals and
S0); 2. Peculiar early-types, that is either spheroidals objects
showing some amount of isophotal asymmetry, or galaxy
pairs in which the main component is clearly elliptical;
3. Normal Spirals with regular disk structure and no evi-
dence for luminous high-star formation regions; this kind of
galaxies have usually a luminous dominant bulge; 4. Per-
turbed Spirals, that is disk-dominated galaxies (with a cen-
tral bulge) showing an asymmetric structure due to inter-
actions or to zones of strongly enhanced star formation; 5.
Irregular galaxies, that is objects with a high degree of asym-
metry showing no evidence for a disk component.
Figure 1 illustrates some galaxies extracted from the 5
morphological classes.
The visual inspection has been performed indepen-
dently by three authors, PC, GR and GF, in order to mini-
mize systematic trends. For the cases on which the classi-
cations did not match, we have chosen the median morpho-
logical value.
Eventually, we have assigned 60 objects to class 1, 14
to class 2, 80 to class 3, 48 to class 4 and 98 to class 5.
The classes with peculiar disks and irregular galaxies are
the most populated ( 50% of the galaxies fall in class 4
or 5). It was not possible to systematically separate ellip-
ticals from S0, given the reduced isophotal area covered by
the sources. Among the 46 galaxies belonging to the above
mentioned clusters, 20 are normal ellipticals, 3 are peculiar
ellipticals, 10 are normal spirals, 5 are peculiar spirals and
8 are irregulars.
4.1.1 The eect of K-correction on the objects with
z > 1:2
We analyze here the eects of K-correction on the morpho-
logical classication of objects having z & 1:2. The reddest
ACS available band (the F850LP lter) does not match for
these galaxies the B-band rest-frame, that is used for the
morphological analysis of the remaining objects at lower z,
but rather matches the U-band rest-frame.
To this end we have selected a subsample of galaxies
with z  1, and we have compared their morphologies in
the V (6060  A, corresponding to the U-band rest-frame) and
z (8500  A, corresponding to the B-band rest-frame) ACS
bands. In particular, 3 Elliptical/S0 galaxies (1 3 in Fig.
2), 3 Normal Spirals (4 6 in Fig. 2), 3 Perturbed Spirals
(7 9 in Fig. 2) and 3 Irregulars (10 12 are shown as
examples in Fig. 2).
Ellipticals. The galaxies in panel 1 and 3 have a clear ellip-
tical morphology both in U as in B rest-frame. The galaxy
in panel 2 represents a border case: according to its B rest-
frame image has been classied as S0 and assigned to class
2; it must be noted that this is the only case in which the
classiers have been able to distinguish S0 from E or Sa mor-
phologies. In the U rest-frame the S0 morphology is roughly
preserved, tending perhaps to resemble to a Sa Spiral.
Normal Spirals. The galaxies in the panels 4, 5 and 6 have
been classied as normal spirals according to the presence
c
2004 RAS, MNRAS 000, 1{19

6 P.Cassata et al.
Figure 3. Sersic index retrieved by Gasphot versus those retrieved by Galt for K20 galaxies in four bins of asymmetries (A). Dashed
lines mark the region where
n=n 6 0:5. Filled squares are visually classied ellipticals/S0 galaxies (class 1 and 2), open diamonds are
spirals (class 3 and 4) and crosses are irregulars.
in their B-band rest-frame images of a central bulge on top
of a spiral-dominated disk. It turns out that for the objects
in panel 4 and 6 the morphology is preserved going from B
to U rest-frame; in the remaining case the bulge becomes
much less luminous in U and the star formation regions of
the disk make the morphology looking more irregular.
Perturbed Spirals. The galaxies in the panels 7, 8 and 9 have
been classied as spirals according to the presence in their
B-band rest-frame of a disk structure with a central bulge
(less luminous that in the above case). The peculiarity de-
pends on the presence of asymmetric structures (zones of
high star formation or interactions). Also in this case, the U
rest-frame luminosity of the central bulge decreases, making
the asymmetries more evident. In particular, in two cases
(panels 7 and 8) the dominant morphology becomes irreg-
ular, whereas perhaps in the latter case (panel 9) the un-
derlying disk structure and the spiral arms structure are
preserved.
Irregulars. In all the three cases the irregular morphology is
preserved, as expected.
Observing the Fig. 2, it turns out that the most impor-
tant eect of the morphological K-correction concerns ob-
jects with a B rest-frame spiral-like morphology, which may
become irregulars in the U rest-frame imaging. In order to
quantify the percentage of mis-classied spirals at z & 1:2,
we have performed the comparison between B and U rest-
frame imaging over a sample of 27 disk galaxies at z  1.
We found that the morphology moved to irregular only in
5/27 cases (18%). Since we do not know if the z  1 galax-
ies are representative of the universe at higher z, this result
suggests that at z & 1:2 the fraction of irregular galaxies
could be slightly over-estimated, as the fraction of spirals
could be slightly under-estimated.
We have also performed the above analysis over a sam-
ple of 10 elliptical galaxies at z  1, seeking cases similar
to panel 2 of Fig. 2. We concluded that the the cited case
is the only in which elliptical morphology is not preserved
moving from B to U rest-frame. Then the number of ellipti-
cal galaxies at z > 1:2 seems to be rather solid.
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 7
4.2 Surface brightness prole analysis
4.2.1 The GASPHOT automatic analysis tool
GASPHOT is a package for fully automatic surface photom-
etry of galaxies, with a very low level of visual interaction,
hence particularly suitable for very large imaging datasets
(Pignatelli et al. 2004). However, its validity for morpho-
logical classication is to be considered only in a statistical
sense, rather than on a single object basis.
The tool uses a modied version of SExtractor (Bertin
& Arnouts 1996) to identify galaxies in the image and to ex-
tract isophotes. Then it analyzes the proles and derives the
main photometric parameters of each identied object. The
program determines the light growth-curves along the ma-
jor and minor axes. These are then tted with a Sersic law
( / r 1=n ), convolved with the PSF, with ve free param-
eters: the total magnitude M tot , the half-luminosity radius
in arcsec re , the Sersic index n, the axial ratio b=a and the
value of the local background.
This one-dimensional approach provides a more robust
estimate than a more complex 2-D t of the surface bright-
ness image, which is more sensitive to the presence of fea-
tures like spiral arms, double nuclei or dust lanes. It is also
less aected by instrumental artifacts in the real image, par-
ticularly for very faint galaxies like our own.
Proles with high values of the Sersic index imply
an early-type morphology (the de Vaucouleurs prole has
n=4), while low values usually indicate later-type morpholo-
gies (the exponential prole has n=1). Unfortunately, many
bulge-dominated objects have intermediate Sersic indices,
or even an exponential prole, so a residual degeneracy still
remains in assigning the morphological class.
The situation gets more complicated when treating ir-
regular galaxies and when dealing with blended objects. The
isophotes agged by SExtractor as blended are not used in
the t, in order to minimize distortions. This makes the
tting procedure much more certain. In this framework, ir-
regular galaxies can only be identied by very large  2 or
failed ts. In these cases, however, the photometric param-
eters obtained are obviously meaningless.
4.2.2 The automatic analysis package GALFIT
GALFIT is an automatic tool to extract structural param-
eters from galaxy images. At variance with GASPHOT,
GALFIT ts the whole 2-D sky-projected light distribu-
tion. It combines together dierent parametric models (the
Nuker law, the Sersic-de Vaucouleurs prole, the exponential
disk, the Gaussian or Moat functions) and allows multi-
component tting (useful to calculate e.g. Bulge/Total light
ratios) and provides measures of the diskyness/boxyness of
the examined galaxy. If available, the PSF is used to con-
volve the model before tting. It is also possible to mask out
of the t peculiar regions (dust lanes, nearby companions,
spiral arms, etc.) that the user wants to exclude.
Even if GALFIT allows to reach a high level of detail
in modeling galaxy light, it needs substantially more inter-
action for each individual object and it works well for bright
galaxies with good sampling, rather than for the fainter ones
closer to the sensitivity limit.
We have built a tool automatically running GALFIT
Figure 4. The distributions of Sersic indices retrieved by GAS-
PHOT (continuous line) and GALFIT (dotted line) for the four
morphological classes. The mean value of n Sersic for the two tools
is reported in each panel.
over all galaxies of interest, making use of SExtractor to
determine the required initial guess of the model parameters
(magnitudes, scale radii, axial ratios and position angles).
4.2.3 Comparison between GALFIT and GASPHOT
Although GALFIT can in principle use multi-component
models, in order to compare with each other the results from
GALFIT and GASPHOT, we have forced GALFIT to the
single component, Sersic model.
We must also rst clear the sample from the objects
for which the two tools were not able to produce a t. In
fact, because of the numerical approximations involved and
in order to avoid unrealistic run-away solutions, GASPHOT
allows the tting parameters to move during the tting pro-
cess in the region limited by Re > 0:7 pixels and 0:5 < n < 8
only. Solutions whose best t-parameters are beyond these
boundaries are rejected, and the related galaxies are agged
as "failed ts". GALFIT has no such limits: however, in or-
der to perform a consistent comparison between the results
of the two tools, we put the same limits on the best-t pa-
rameters of both tools and removed from the sample those
for which one of the two tools was not able to nd a solution.
In addition, due to the very low S/N ratio, GASPHOT
and GALFIT were not able to produce any t for 16 and
14 galaxies, respectively. Once we removed these objects,
mainly irregular or high asymmetric objects, the galaxies
left in the sample are respectively 254 (GASPHOT) and
236 (GALFIT), of which 211 are in common.
In Fig. 3 we have reported the Sersic indices retrieved
by GASPHOT against those obtained by GALFIT, nd-
ing that 73% of the sample is included in the region
jnGASPHOT nGALFIT j=hni 6 0:5, where hni is the average
of the two results, while 9% have
n=hni > 1. It is worth
stressing that, even though for toy galaxies the results from
GASPHOT and GALFIT have been found to dier less than
10% from one another (Pignatelli et al. 2004), the large scat-
ter in Figure 3 is not surprising when dealing with real fea-
tured galaxies, with light distributions often asymmetric and
far from being stick to the models. In particular, due to the
c
2004 RAS, MNRAS 000, 1{19

8 P.Cassata et al.
dierent tting approaches adopted (1D vs. 2D), we expect
that GASPHOT and GALFIT provide rather dierent re-
sults at increasing asymmetry. Thus, we split the Figure 3
in dierent bins of the asymmetry parameter A. Actually,
apart from the rst asymmetry bin, the scatter around the
1:1 relation increases with the asymmetry of the galaxies
analyzed, and a similar trend is found for the number of
failed ts as a function of the asymmetry: more asymmetric
galaxies are obviously harder to model with a spherically
symmetric Sersic law. The large scatter observed in the rst
bin of A is due to the predominance of early-type galaxies
(high values of Sersic index) and to the fact that the un-
certainty of the retrieved values of n intrinsically increases
with the Sersic index itself (Pignatelli et al. 2004).
A similar trend is visible in Fig. 5, where we show the
comparison between the optical radii retrieved by the two
tools. Even now 73% of the sample is comprised in the region
jrGASPHOT rGALFIT j=hri 6 0:5 In this case, however, the
scatter around the 1:1 relation is not symmetric: there is a
large amount of galaxies for which the GALFIT scale radius
is considerably larger than that obtained by GASPHOT. A
similar eect was also found by Pignatelli et al. (2004) in a
systematic comparison between the two tools.
We checked visually the objects for which this dierence
was more noticeable. We found that, for compact objects
with a small isophotal radius and high Sersic index, GALFIT
can sometimes recover unrealistic optical radii, often many
times larger than the isophotal area itself.
Finally, we want to investigate further whether a cri-
terion based on the Sersic index could be able to separate
early- from late-type galaxies. Ravindranath et al. (2004) for
example classied as bulge-dominated and disk-dominated
galaxies objects with nSersic 6 2 and nSersic > 2, respec-
tively.
In Fig. 4 the distributions of the Sersic indices ob-
tained by GASPHOT and GALFIT for the ve morpho-
logical classes are reported (normal E's/S0 and perturbed
ellipticals are plotted together). It is worth noticing that,
while the results of the two tools for individual objects may
dier, the statistical distributions are quite similar.
The number of elliptical/S0 galaxies (morphological
class 1 and 2) which have Sersic indices lower than 2 are
only 6 and 3 according to GASPHOT and GALFIT respec-
tively. Instead, the contamination of late type galaxies with
Sersic indices greater than 2 is larger (56/254 and 50/237,
according to GASPHOT and GALFIT respectively). From
Figure 4 we must conclude that the Sersic index alone just
provides a broad, not univocal indication of the morpholog-
ical type. Since this index is a measure of the concentration
of light, this is in agreement with the results of Abraham
et al. (1996), who found that, besides the Concentration, at
least one more parameter is needed to classify galaxies.
4.3 The CAS parameter set
The non-parametric methods for morphological estimates
are those which do not need to assume a parametrized an-
alytic function to model the galaxy light distribution. They
constitute an important complement to the problem of quan-
titative morphology (Abraham et al. 1996, Conselice et al.
2000, Conselice 2003). We use the classical Concentration,
Asymmetry and clumpineSs (or smoothness) parameters.
The Concentration correlates with the Sersic index: high
Concentration values correspond to early-type morphology,
while lower values are suggestive of a disk-dominated or
irregular galaxy. The Asymmetry can distinguish irregular
galaxies or perturbed spirals from relaxed systems as E/S0
and normal spirals. The Clumpiness quanties the degree of
structure on small scales, and roughly correlates with the
rate of star formation.
4.3.1 Denitions
The operational denitions of the three parameters may dif-
fer from author to author. We refer to the Conselice et al.
(2003) denitions, but we modied them in order to make
simpler and faster the computations. The Concentration is
a logarithmic ratio of the apertures containing 80% and 20%
of the total ux:
C = 5 log
 r80
r20

: (1)
operatively, we interpolate the ux growth curve obtained
using 12 aperture radii within which SExtractor calculates
the ux.
The Asymmetry measures how much the galaxy light
is symmetric with respect to a rotation of 180 ô around the
galaxy centroid. So the Asymmetry is qualitatively the resid-
ual of the dierence between the original I0 and the rotated
I180 galaxy image:
A =
P j I0 I180 j
2
P j I0 j
: (2)
In order to reduce the in uence of the background noise,
we subtract from the galaxy image the background given
by SExtractor, and we remove those pixels having values
lower than 1.5 times the rms of the background. The results
strongly depend on the adopted galaxy centroid. We solve
this problem by computing the Asymmetry on a grid of 100
100 points around the center calculated by SExtractor. The
distance between contiguous points in this articial grid is
0.4 pixels, so the Asymmetry is computed for points up to
20 pixels far from the initial center. The Asymmetry is
dened as the minimum of the values computed on the whole
grid.
The Clumpiness measures the fraction of galaxy light
lying in the high spatial frequencies. It is computed by sub-
tracting from the original an image smoothed by a box of a
given size:
S =
X (I0 I)
I0
(3)
where I0 is the original image and I is that convolved
with a box of scale . The value of the retrieved clumpi-
ness depends strongly by the scale used for the convolu-
tion. We use  = 2=3rSEx and  = rSEx (where rSEx is
the FLUX RADIUS parameter given by SExtractor) respec-
tively for galaxies with rSEx lower and greater than 5 pixels.
Again, in order to reduce the in uence of the background
noise we remove those pixels having values lower than 2
times the rms of the background.
To automatically compute the CAS parameters for all
the galaxies in our image we have used specically a de-
signed IDL tool which uses SExtractor to produce the ux
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 9
Figure 5. Scale radius retrieved by GASPHOT versus those retrieved by GALFIT for K20 galaxies in four bins of asymmetry. Dashed
lines mark the region where
re=re 6 0:5. Filled squares are ellipticals/S0 galaxies (class 1 and 2), open diamonds are spirals (class 3
and 4) and crosses are irregulars.
growth curves of galaxies, then IDL calculates the concen-
tration indices, subtracts the background and lter noisy
pixels. Finally a postage cut for each galaxy of interest is
produced and these postages are suitably rotated and con-
volved to calculate Asymmetry and Clumpiness.
We compute the CAS parameters in the ACS bands
roughly corresponding to the B-band rest frame for each
galaxy.
4.3.2 Simulation of CAS parameters for high-z galaxies
The poor sampling of high z galaxies is likely to aect the
measures of the CAS parameters. To check and quantify this
point, we have simulated the eects of moving local galaxies
with known CAS values to higher redshifts. We have used
as test objects all the galaxies in the sample having z < 0:3.
Following Conselice (2003), the angular size of galaxies is
reduced by a rebinning factor b given by the ratios of the
angular diameter distances:
b = dA(z2)
dA(z1)
(4)
where z1 and z2 are respectively the initial and nal red-
shifts. Then, pixel uxes are reduced by the cosmological
(1 + z) 4 factor and are K-corrected. Operatively, we rst l-
ter out the noise of the test galaxies with a simple smoothing
algorithm, and, after rebinning, changing the angular sizes
and correcting the pixel uxes, we convolved the images with
a model of the PSF. Finally the noise is re-added and the
CAS parameters are computed.
In Fig. 6 we report the results of these simulations for
4 ellipticals, 7 spirals (including normal and perturbed spi-
rals) and 11 irregulars galaxies. For comparison, the ranges
of values that we measured for the 300 galaxies in our sam-
ple are reported as vertical bars. Although the number of
simulated objects is not large, we note that the Asymme-
try A does not depend on z for spirals and irregulars, apart
from an increase of the scatter at increasing redshift. On the
other hand, a small increment of the mean value with z is
present for the ellipticals.
There is a moderate dependence on redshift in the esti-
mate of Concentration C for the ellipticals: values at z > 0:5
are systematically lower than local values. Again, the scatter
increases with the redshift for all classes.
As for the Clumpiness S, a bias appears to be introduced
c
2004 RAS, MNRAS 000, 1{19

10 P.Cassata et al.
Figure 6. Evolution of the CAS parameters with z for a sample of articially redshifted galaxies. 4 ellipticals, 7 spirals and 11 irregular
galaxies are simulated at increasing redshift to isolate systematic eects. The resulting parameters for the entire sample in the four
redshift bin is shown for comparison (vertical bars).
for all morphological classes, since the mean value increases
articially with redshift, while the scatter remains moderate.
Spirals and irregulars display a large spread in the val-
ues of the Concentration and Clumpiness which are not re-
produced by our simulations, likely due to the small size of
the local reference sample.
Accortding to the measures of non-parametric indica-
tors on real galaxies, up to redshift z ' 1:5, we have high-
lighted some biases in the CAS structure of galaxies, intro-
duced by instrumental eects.
In conclusion, our simulations suggest that CAS param-
eters provide an eective tool to analyse and discriminate
galaxy morphologies in the z-interval of the K20 sample.
4.3.3 Results of the CAS analysis
Figures 7, 8 and 9 show the distribution of our galaxies sam-
ple in the CAS space in various redshift bins. The dierent
symbols refer to our visual morphological classication as
discussed in Sect. 4.1.
Objects in the morphological class 1 (ellipticals and S0s)
separate signicantly from the other classes. We have identi-
ed, independently from the redshift bin, the domains pop-
ulated mainly by early-type galaxies, which are bounded by
the continuous lines in the gures. These boundaries are
given by the following conditions:
A < 0:2; C > 2:9; 0:05 < S < 0:27 A < 0:45S+0:085:
(5)
We stress that these boundaries are independent of the
galaxy redshift. Only 1 elliptical galaxy lies outside from
these regions (it is in fact a very small galaxy with an un-
certain classication).
There are however 20 late-type galaxies out of 226 (9%),
as judged from visual inspection, which fall within the early-
type galaxy CAS domain, and so would be misclassied by
such criterion. It is worth to note that using dierent bound-
aries for each redshift bin reduces the contamination of late-
type in the ellipticals domain of a small amount. This con-
tamination is due to the following objects: 8 galaxies domi-
nated by a luminous and symmetric bulge, but with evidence
for a faint disk; 3 ellipticals with a small companion in in-
teraction; 5 very compact objects, for which CAS estimate
is uncertain; 4 further symmetric and concentrated galaxies,
but clearly patchy, for which the measure of the clumpiness
essentially fails because of the small area and high ellipticity.
The datapoints corresponding to elliptical galaxies
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 11
Figure 7. Evolution of the distribution of galaxies in the Concentration-Asymmetry plane in four redshift bins. The parameters are
retrieved in the ACS band nearest to the B-rest frame: galaxies with z < 0:2 are analyzed in the F435W lter, objects with 0:2 < z < 0:55
in the F606W lter, objects with 0:55 < z < 0:85 in the F775W lter and those with z > 0:85 in the F850LP lter. The rectangle marks
the region populated by ellipticals/S0 galaxies.
within the boundaries dened by eq.(5) show a scatter which
sensibly depends on redshift: while it stays small up to
z = 0:85, it tends to explode above this limit.
On the other hand, galaxies belonging to class 3, 4 and
5 do not show a tendency to segregate in the CAS plots, but
they occupy almost the same regions.
In conclusion, the application of the CAS analysis to the
K20 galaxies suggests this technique to be very eôcient in
disentangling early-type from late-type and irregular galax-
ies, with few percent of contamination. On the contrary, we
failed in identifying algorithms able to resolve the dierent
classes contributing to the late-type category.
4.4 The Clumpiness-Asymmetry Relation for
Galaxies
Conselice (2003) noticed that the Asymmetry and Clumpi-
ness parameters for normal galaxies in the local universe
are correlated, populating a strip in the S-A plane. On the
other hand, merging systems or irregular galaxies have typ-
ically the same clumpiness of the interacting components
but higher asymmetry, hence they deviate from the relation.
Fig. 10 conrms this trend in our sample: the two parame-
ters are clearly correlated for normal ellipticals and normal
spirals. Because of the dierent operative denitions of the
CAS parameters with respect to those of Conselice (2003),
we needed to recalibrate the relation. To this end we used
the simulations described in Sect. 4.3.2 using data on 4 el-
lipticals and 3 normal spirals and changing their redshift:
our simulated A-S relation for normal (E/Sp) galaxies is
reported in the inset of Fig. 10, together with the strip con-
taining the 90% of the data points. The best-t relation and
its 90% boundaries are:
A = (0:44  0:10)S + (0:08  0:08): (6)
Fig. 10 reports the distribution of the galaxies in the S-A
plane and the strip in which we expect to nd only normal
galaxies (dened as the strip that contain 90% of the sim-
ulated objects). The plot conrms that the large majority
of visually inspected normal spirals and ellipticals fall in-
deed within the boundaries. Outside them only irregulars,
peculiar spirals and some normal spiral are found.
c
2004 RAS, MNRAS 000, 1{19

12 P.Cassata et al.
Figure 8. Evolution of the distribution of galaxies in the clumpineSs-Asymmetryplane in four redshift bins. The parameters are retrieved
in the ACS band nearest to the B-rest frame: galaxies with z < 0:2 are analyzed in the F435W lter, objects with 0:2 < z < 0:55 in the
F606W lter, objects with 0:55 < z < 0:85 in the F775W lter and those with z > 0:85 in the F850LP lter. The lines marks the region
typically populated by ellipticals/S0 galaxies.
However, this criterion appears of limited usefulness to
disentangle normal spirals from irregulars because a large
number of visually classied irregulars also fall in the normal
galaxy region.
5 THE REDSHIFT DISTRIBUTIONS OF THE
MORPHOLOGICAL CLASSES
In Fig. 11 the redshift distribution for all the ve morpho-
logical classes are reported. It should be noted the excess
of early type galaxies and spirals at redshift z  0:75, sig-
nature of the cluster of galaxies at  z = 0:737. The highest
redshift elliptical is is at z = 1:903. No disk galaxies lie at
z > 1:8. Irregular galaxies instead are numerically relevant
from z = 0:2 up to z = 2:5.
Figure 12 shows the evolution of the fraction of mor-
phological classes with the redshift. For clarity we used here
only three main classes: early types (including type 1 and
2), disks (type 3 and 4) and irregular/peculiar (type 5). The
shaded regions account for the eects of morphological K-
correction, as described in Sect. 4.1.1: the upper limit for the
irregulars (and the lower for the spirals) corresponds to the
(unrealistic) case in which the K-correction does not aect
the classication of objects with z & 1:2; the lower limit for
the irregulars (and the upper for the spirals) shows instead
the case in which  20% of the morphologically classied
irregulars at z & 1:2 are spirals.
The more evident result is the fast growth of the fraction
of irregulars above z  0:8, where they are the dominant
population of the sample, being at z & 1:5 more than the
60% of the entire population. The Elliptical fraction remains
near to 20% up to z = 1:5 (beyond this redshift the statistic
is too poor), except for the point at z  0:75, where the
fraction is higher due to presence of the two clusters. The
fraction of disk galaxies remains rather constant to the 50%
up to z  1, then it decreases rapidly.
Conselice et al. (2004), studying the B-band rest-frame
morphology of an I-selected sample of galaxies in the HDFs
up to z=3, reach similar conclusions (see Fig. 9 therein):
they also found a rapid decrease of ellipticals and spirals
above z  1 and a contemporary increase in the number
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 13
Figure 9. Evolution of the distribution of galaxies in the Concentration-clumpineSs plane in four redshift bins. The parameters are
retrieved in the ACS band nearest to the B-rest frame: galaxies with z < 0:2 are analyzed in the F435W lter, objects with 0:2 < z < 0:55
in the F606W lter, objects with 0:55 < z < 0:85 in the F775W lter and those with z > 0:85 in the F850LP lter. The rectangle marks
the region typically populated by ellipticals/S0 galaxies.
of peculiar galaxies. Our result however seems to be more
robust at least up to z  2: on the one hand, the K-band
selection ensures a better mass sampling than the optical
selection, and on the other hand, the larger area minimizes
eects of cosmic variance. The accurate investigation of the
eects of morphological K-correction conducted in previous
sections avoids biases in the morphological fraction up to
z  2.
In Fig. 13 a comparison of the morphological fractions
(early- and late-type only) obtained with various methods
explored in this paper is proposed. In the top two panels
the fractions of objects with nSersic ? 2 are shown against
the redshift according to GASPHOT and GALFIT. In the
bottom-left panel the early-type galaxies are selected ac-
cording to their CAS parameters (see Eq. 5). Finally, in the
bottom-right panel the results by the visual inspection are
reported. Among the automatic methods, only the CAS pa-
rameters one is able to reproduce the results by the visual
inspection. Among the  250 objects for which GASPHOT
and GALFIT give acceptable ts,  50% have nSersic > 2.
The late-type fraction obtained with the nSersic parame-
ter only suers of two eects: the large number of irregular
high asymmetric objects for which GASPHOT and GAL-
FIT do not obtain an acceptable t and the contamination
of the Bulge dominated objects. The discrepancy between
GASPHOT and GALFIT for objects with z & 2 (classied
by the former as late and by the latter as early-type) must
be considered rather marginal, due to the small number of
galaxies in that redshift range for which GASPHOT and
GALFIT found acceptable ts.
5.1 The Evolution of the Merging Fraction
The high resolution very deep imaging in the K20 eld gives
us also the opportunity to investigate the evolution of the
merging fraction with the redshift for this K-band selected
sample. Up to date, studies of the merging fraction against
redshift have been done for optically selected (Patton et al.
1997, Le Fevre et al. 2000) and NIR selected (Conselice et
al. 2003) samples. In this paragraph, we investigate the evo-
lutionary merging fraction using both pairs statistics (as i.e.
c
2004 RAS, MNRAS 000, 1{19

14 P.Cassata et al.
Figure 10. The distribution of the sample galaxies in the S-A plane. The strip between dotted lines shows the region containing normal
galaxies, calibrated using the normal galaxies of the sample simulated at various redshifts, shown in the top-right panel.
Figure 11. Redshift distributions for the ve morphological
classes.
Le Fevre 2000) and asymmetry technique (Conselice et al.
2003).
Following Le Fevre et al. (2000), we have visually identi-
ed in the K20 sample objects having a companion brighter
than i=24.5 within 20 kpc from the center, independently on
the morphology of the main object, after removing cluster
objects.
Given that in almost all cases we have the redshift only
for the main object and we do not know if the observed pair
consists of truly interacting objects or not. Then, we need
to apply a correction factor accounting for this projection
contamination. This correction is calculated integrating the
galaxy number counts published by Driver et al. (1995) up to
I=24.5 within a circle of physical radius of 20 kpc. Given that
at z < 0:5 the 20 kpc radius projects over a large area, the
high contamination of spurious pairs makes not statistically
signicant the measures at such low redshifts.
A second correction is needed to estimate how many
of these physical pairs are going to merge. Here we use the
correction proposed by Patton et al. (1997), suggesting that
in the local universe half of the pairs with relative velocities
less than 350Kms 1 are expected to merge within 0.5 Gyr.
Given that the velocity is expected to vary with redshift as
(1+z) 1 , the merging fraction is obtained by multiplying the
pairs fraction (corrected for eects of projection) by 0:5(1 +
z).
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 15
Figure 13. Morphological fractions for early- and late-type galaxies as a function of the redshift obtained by the dierent methods
proposed in this paper: in the top two panels the criterion based on n Sersic ? 2 is used for GASPHOT and GALFIT; in the bottom-left
panel early-galaxies are selected according to Eq. 5; in the bottom-right panel the visual classication is used.
For comparison, we have recomputed the merging frac-
tion in a completely dierent way, based on asymmetry com-
putation (Conselice et al. 2003): an object is identied as
merger if it has an asymmetry greater than a certain value
(in this case A > 0:4). Also in this case cluster objects
have been removed. This technique selects a dierent class
of objects compared with those identied by pairs statis-
tics. Given that it is computed in a small area around the
object, the highest asymmetry is measured for strongly in-
teracting objects, rather than for objects in a pair often
separated by many kiloparsecs, and that therefore do not
fall together in the area for the asymmetry computation.
Moreover, the merging fraction obtained by the asymmetry
technique strongly depends on the border value of A used
for merger identication.
Figure 14 reports the inferred merging fraction for the
K20 sample as a function of the redshift both from pair
statistic (lled circles) and from asymmetry criterion (lled
squares). The results from the two techniques are in good
agreement with each other, and with those obtained by other
authors. Our data, together with those by Le Fevre et al.
(2000), Patton et al. (1997) and Conselice et al. (2003) pro-
vide the evolution of the merging fraction as a(1+z) m , where
a and m are free parameters:
fmerg = (0:022  0:002) (1 + z) 2:20:3 : (7)
The value of m is in quite good agreement with m = 2:80:9
obtained by Patton et al. (1997) using only its data at
z < 0:3, but is smaller than the value obtained by Le Fevre
et. al 2000 (m = 3:4  0:6). We can conclude that the frac-
tion of merging evaluated for a K-band selected sample in-
creases up to z  1:5, according with previous works, but,
due to the limited statistic, we can not constrain the evolu-
tion for higher redshift. There are however indications that
it remains lower than 20% up to z=2.
It must be noted that if it is true that the asymme-
try and pairs techniques select dierent merging events (ob-
served respectively in late and early phases), the merging
fraction could be the sum of the two, then the numbers in
Fig. 14 must be multiplied by a factor 2.
6 COMPARING THE MORPHOLOGICAL
AND SPECTROSCOPIC CLASSIFICATION
In Table 1 we report a comparison between morphologi-
cal and spectroscopic classications for our sample galaxies.
c
2004 RAS, MNRAS 000, 1{19

16 P.Cassata et al.
spectral perturbed normal perturbed all
class ell. ellipticals spirals spirals irr/mer types
Early type 51 5 6 3 1 66
Early+EmLines 5 3 20 3 7 38
Emission lines 2 6 53 41 76 178
Unobs./Unid. 2 0 1 1 14 18
Total 60 14 80 48 98 300
Table 1. Comparison between morphological and spectroscopical classication.
Figure 12. Morphological fractions as a function of the redshift.
The results for z & 1:2 are likely to be aected by morphological
K-correction, given that at those redshifts z-band does not match
more the B rest-frame, but the U rest-frame. We showed in Sect.
4.1.1 that the  20% of the galaxies with z  1 classied as spirals
in their B rest-frame appear as irregulars in their U rest-frame.
The shaded regions show the condence interval for irregulars
and spirals galaxies. The datapoint at z=0.7 is aected by the
presence of the two clusters.
The two classications have been made completely indepen-
dently.
Among the 18 spectroscopically unobserved or uniden-
tied objects, 15 fall in the irregular class. The bulk of them
(9/15) are low surface brightness objects very close to the
optical detection limits. For the remaining object with both
spectroscopic and morphological classication, the agree-
ment is quite good: 51/60 normal ellipticals have an early
type spectrum, 118/128 spirals (morphological class 3 and
4) and 97/98 irregulars have late type or early+emission-line
spectra.
Among the galaxies classied as normal ellipticals, there
are 5 objects with early spectra, but showing some emis-
sion lines, and 2 galaxies with emission lines spectra. Among
these 7 objects, 4 are small and with diôcult morphological
classication (Sersic indices n > 2:3 and re < 0:25 00 ); one
has been classied as S0/Sa, because of the axial ratio near
to 0.5, even if a clear disk component is not so evident; the
remaining 2 have clearly an elliptical morphology (n > 4).
Among the 14 galaxies that we have morphologically
classied as peculiar ellipticals (galaxies with a dominant
Figure 14. The inferred merging fraction for the K20 sample
as a function of the redshift, both from pairs statistics (lled
circles) and from asymmetry criterion (lled squares). Errors bar
are calculated by a Poisson statistic. Results of previous works
are also reported, together with the best-t relation to the data
(dotted line).
elliptical component, but with signs for distortion of the
isophotes), 9 have emission lines in their spectrum, show-
ing that a certain degree of activity is really present in the
galaxies. In this morphological class however 5 galaxies that
show no signs of emission lines fall.
Ten late-type galaxies (belonging to the morphological
classes 3, 4 and 5) show early-type spectra with no signs
for emission lines. In 6 cases a luminous and concentrated
bulge dominates the spectrum, probably overwhelming the
contributions from the star-forming regions. In one case the
classication is uncertain due to the small area covered by
the galaxy. The agreement between morphological and spec-
troscopic properties is good also for the galaxies belonging to
the two clusters: only 1 object classied as elliptical shows
signs of emission lines, whereas 4 late-type galaxies have
early-type spectra.
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 17
Figure 15. The evolution with the redshift of the galaxy sizes
of elliptical class. Upper panel: mean angular size in four redshift
bins compared with three models of evolution for the physical
galaxy size. For each point error bars represent the standard de-
viation divided by the square root of the number of galaxies in
that bin. Continuous and dotted lines are two model in which
galaxy size evolves with the redshift respectively as (H=H 0 ) 1
and as (1 + z) 1 , whereas dashed line is a model of constant size.
The three curves are normalized to the observed mean value at
redshift 0:5 < z < 1. Lower panel: the physical sizes of early type
galaxies are reported as a function of the redshift. The three mod-
els of evolution and no-evolution of galaxy sizes are also reported.
7 GALAXY SIZES VERSUS REDSHIFT
A critical prediction of galaxy formation models concerns
the evolution of the physical sizes. The disk scale-lengths, in
particular, are predicted to decrease proportionally to the
inverse of the Hubble parameter (Fall & Efstathiou 1980).
We have investigated the evolution with the redshift of the
galaxy sizes for early/type (class 1 and 2), disks (class 3 and
4) and irregulars (class 5) up to z  1:5.
As a measure of the physical size of galaxies we use the
eective radius re retrieved by GASPHOT by tting the
galaxy light prole with a Sersic model.
In the upper panels of Figs. 15, 16 and 17 we report
the mean angular size in four redshift bins (z < 0:5, 0:5 <
z < 1, 1 < z < 1:5 and z > 1:5) respectively for ellipticals,
spirals and irregulars. Cluster galaxies can be recognized at

z = 0:736 and  z = 0:668.
For comparison, the curves corresponding to no-
evolution of the physical size, evolution with the inverse of
Hubble parameter H(t) and evolution with (1 + z) 1 are
also plotted, each normalized to the observed datapoint at
z = 0:75. In the lower panels, the data on the physical sizes
for individual galaxies are reported.
Our results do not show much evidence for size evolu-
tion in the explored redshift range. In particular, spirals and
irregulars are entirely consistent with no evolution, in good
agreement with the results of Ravindranath et al. (2004),
Figure 16. Same as Fig. 15 for disk galaxies.
Figure 17. Same as Fig. 15 for irregular galaxies.
who payed particular attention in discussing the observa-
tional biases.
A small decrease of Re may be seen for the ellipti-
cal population, for which our results are consistent with
Re / (H=H0) 1 . The statistical signicance of this result
is however not at all marginal: we point out that at high z,
K-band selects only the most massive and luminous galax-
ies, while at low z the explored mass interval is much more
wide.
c
2004 RAS, MNRAS 000, 1{19

18 P.Cassata et al.
8 DISCUSSION AND CONCLUSIONS
We have exploited very deep and high resolution ACS/HST
imaging recently become available in the CDFS area and
covering a major portion of the K20 survey, to conduct an
accurate morphological analysis of the K20 sample high-
redshift galaxies. For each object, the analysis has been per-
formed in the ACS band closer to the B-band rest-frame in
order to minimize eects of morphological k-correction.
Our main results are hereafter summarized.
(i) From visual classication (performed by three au-
thors), we nd that: 60 galaxies are normal ellipticals or
S0; 14 are peculiar ellipticals; 80 are normal spirals; 48 spi-
rals showing signs of interaction or of regions of excess star
formation; 98 are irregular galaxies. We have carefully in-
vestigated the eects of the morphological K-correction on
the objects at z & 1:2, for which the B-band rest-frame is
not available, establishing robust upper and lower limits to
the fraction of each morphological class.
(ii) We have investigated the capabilities of parametric
and non-parametric techniques in disentangling morpholog-
ical classes with a low level of interaction. In particular, we
used the GALFIT and GASPHOT packages to t the light
distribution with the Sersic parametric model, to exploit the
correlation between Sersic indices and B/D ratios.
We have shown that parametric tools like GASPHOT and
GALFIT are not completely eôcient in distinguishing be-
tween early- and late-type objects. They do not nd accept-
able ts for  50=300 objects, mainly those with high asym-
metry or low surface brightness. The general agreement of
these two tools for the remaining objects is quite good (while
for  10% they turn out to be completely inconsistent). The
agreement worsens at increased asymmetry of the objects, as
expected (see Figs. 3 and 5). The distributions of the Sersic
indices for the dierent visual morphological classes turn out
to be quite similar for the two tools (see Fig. 4). It is inter-
esting to note that a quite substantial number ( 50=250) of
visually classied late-type galaxies have a high Sersic index
(n > 2): this case may be produced by a prominent bulge
on top of a tiny disk structure.
(iii) We have complemented the morphological analysis
with the calculation of non-parametric quantities like Con-
centration, Asymmetry and Clumpiness for the objects in
the sample. We have shown that there is a precise region of
the CAS space (dened by Eq. 5) occupied mostly by early
type galaxies: ellipticals are the objects with higher Concen-
tration and smaller Clumpiness and Asymmetry. Only one
elliptical falls out, while 20 late type lye within this region.
Simulations have been performed to investigate the reli-
ability of the CAS values retrieved at high redshifts. It has
been shown that some biases in the measures are introduced
by the degradation of the resolution at high z, somehow pre-
venting an assessment of the intrinsic evolution of the CAS
parmeters.
(iv) We have shown that a criterion based on the Sersic
index only does not completely segregate early- by late-
type galaxies (see Fig. 13; we have already mentioned the
many low surface brightness or highly asymmetric objects
for which the t procedures do not converge, and the number
of late-type galaxies with a Sersic index nSersic > 2).
The CAS criterion instead better reproduces the classi-
cation between early- and late-type obtained by visual in-
spection (see Fig. 13).
(v) Over the 300 objects of the sample, only 74 turn out
to be early type galaxies (our class 1 and 2). Their red-
shift distribution is dominated by the cluster at z  0:737,
and the number decreases rapidly for z > 1:5 following the
general trend for a rapid decrease at such z for the whole
population.
The most numerous galaxies are the irregulars: 1/3 of the
objects in the sample belongs this class. Their fraction in-
creases from  20% at low-z to  60% at z > 1:2 and is dom-
inant at higher redshifts. The simultaneous decrease of early
and disk galaxies suggests intuitively that some amounts of
high redshift irregular galaxies may progressively transform
into the local elliptical and spiral galaxy population. A con-
clusion on this important issue would however require much
improved statistics in the critical redshift interval of z  1:5
to 2. In any case our analysis indicates a predominance of
spirals and irregulars in K-band selected samples at even
moderate depths.
We stress that these results turn out to be rather robust,
thanks to the use of the B-band rest-frame up to z  1:2,
(so minimizing eects of morphological K-correction), and
to the accurate assessment of biases introduced by the K-
correction at higher z.
(vi) We also analyse the evolution with the redshift of the
merging fraction. We use both the pair statistics technique
and the asymmetry criterion A > 0:4, in order to measure
merging fraction. Although the two techniques likely select
two dierent kinds of objects (corresponding to early and
late phases of a merger), we found that the results of each
criterion are in good agreement with each other and with
previous works. The inferred merging fraction remains in
any case lower than 20% up to z = 2. If we consider that
the total merging fraction is the sum of the two, then the
numbers in Fig. 14 are to be multiplied by a factor  2.
(vii) We have performed a systematic comparison be-
tween morphological and spectroscopic classication, show-
ing quite good agreement: 91% of the sample have spectral
characteristic compatible with morphological ones. There
are some ellipticals (7/60) with a very relaxed morphology
showing emission lines in their spectra. The late-type ob-
jects showing early-type passive spectra are mainly bulge-
dominated spirals, for which probably the signal from the
disk component is overwhelmed by the concentrated and lu-
minous bulge.
(viii) Finally the redshift dependence of galaxy sizes is
investigated for elliptical (class 1+2), disk (class 3+4) and
irregular (class 5) galaxies separately. Our results show no
evolution for the sizes of disks and irregulars, whereas a
small decrease of Re has been found for elliptical galaxies:
their sizes seems to vary as Re / (H=H0) 1 . These results
are particulalry important if we consider that we are select-
ing at the high redshifts the most luminous, massive, hence
the largest, of the galaxy population at that cosmic epoch.
ACNOWLEDGEMENTS:We are grateful to the referee,
Roberto Abraham, for careful reading and interesting com-
ments on the manuscript.
c
2004 RAS, MNRAS 000, 1{19

The evolution of the galaxy B-band rest-frame morphology to z  2 19
REFERENCES
Abraham, R. G., Tanvir, N. R., Santiago B. X. et al., 1996,
MNRAS, 279, 47
Abraham, C., 1998, IAU Symp. 186, Galaxy Interactions
at Low and High Redshift, astro-ph/9802033
Bertin, E. and Arnouts, S., 1996, A&AS, 117, 393
Bundy, K., Fukugita, M., Ellis, R. S. et al., 2004, ApJ,
601, L123
Caon, N., Capaccioli, M. and D'Onofrio, M., 1993, MN-
RAS, 265, 1013
Cimatti, A., Mignoli, M., Daddi, E., et al., 2002, A&A,
392, 395
Cimatti, A., Pozzetti, L., Mignoli, M., et al., 2002, A&A,
391, 1L
Cimatti, A., Daddi, E., Cassata, P., et al., 2003, A&A,
412, 1L
Cimatti, A., Daddi, E., Renzini, A., et al., 2004, Nature ,
430, 184
Conselice, C., Bershady, M. and Jangren, A., 2000, ApJ,
529, 886
Conselice, C., 2003, ApJS, 147, 1
Conselice, C., Bershady, M., Dickinson, M. and Papovich,
C., 2003, AJ, 126, 1183
Conselice, C., Blackburne J. A. and Papovich, C., 2004,
astro-ph/0405001
Daddi, E., Cimatti A., Renzini, A., et al., 2004, ApJ, 600,
127L
Driver, S. P., Windhorst, R. A., Ostander, E. J. et al., 1995,
ApJ, 449, L23
Fall, S. M. and Efstathiou, G., 1980, MNRAS, 193, 189
Giavalisco, M., Ferguson, H. C., Koekemoer, A. M. et al.,
R., 2004, ApJ, 600, L93
Glazebrook, K., Ellis, R., Santiago, B. and Griôths, R.,
1995, MNRAS, 275, L19
Le Fevre, O., Abraham, R., Lilly, R. S. et al., 2000, MN-
RAS, 311, 565
Marzke, R. O., Da Costa, L. N., Pellegrini, P. S. et al.,
1998, ApJ, 503, 617
Mignoli, M., et al., 2004, in prep.
Patton, D. R., Pritchet, C. L., Yee, H. K. C. et al., 1997
ApJ, 475, 29
Papovich, C., Giavalisco, M., Dickinson, M. et al., 2003
ApJ, 598, 827
Peng, C. Y., Ho, L. C., Impey, C. D., Rix, H., 2002 AJ,
124, 266
Pignatelli, E., Fasano, G. and Cassata, P., 2004, submitted
Ravindranath, S., Ferguson, H. C., Conselice, C. et al.,
2004, ApJ, 604, L9
Simard, L., Willmer, C. N., Vogt, N. P. et al., 2002, ApJS,
142, 1
van den Bergh, S., Cohen, J. G., Hogg, D. W. and Bland-
ford, R., 2000, AJ, 120, 2190
van den Bergh, S., Cohen, J. G. and Crabbe, C., 2001, AJ,
122, 611
Vanzella, E., Cristiani, S., Dickinson, M. et al., 2004,
astro-ph/0406591
Windhorst, R. A., Taylor, V. A., Jansen, R. A. et al., 2002,
ApJS, 143, 113
This paper has been typeset from a T E X/ L A T E X le pre-
pared by the author.
c
2004 RAS, MNRAS 000, 1{19