Skip to main content

Full text of "Loss of star forming gas in SDSS galaxies"

See other formats

Evidence for progressive loss of star-forming gas in SDSS galaxies 

F. Calura 1 , R. Jimenez 2,3 , B. Panter 4 , F. Matteucci 1 ' 5 , A. F. Heavens 4 

1 INAF - Osservatorio Astronomico di Trieste, via G.B. Tiepolo 11, 34131 Trieste, Italy 
2 Institute of Space Sciences (CSIC-IEEC) /ICREA, campus UAB, Bellaterra-08193, Spain 
3 Dept. of Astrophysical Sciences, Peyton Hall, Princeton University, Princeton, NJ 08544, 


4 SUPA, Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford 

Hill, Edinburgh EH9-3HJ, UK 
5 Dipartimento di Astronomia-Universitd di Trieste, Via G. B. Tiepolo 11, 34131 Trieste, 


f caluraOoats . inaf . it 

Using the star formation rates from the SDSS galaxy sample, extracted using 
the MOPED algorithm, and the empirical Kennicutt law relating star formation 
rate to gas density, we calculate the time evolution of the gas fraction as a function 
of the present stellar mass. We show how the gas-to-stars ratio varies with stellar 
mass, finding good agreement with previous results for smaller samples at the 
present epoch. For the first time we show clear evidence for progressive gas loss 
with cosmic epoch, especially in low-mass systems. We find that galaxies with 
small stellar masses have lost almost all of their cold baryons over time, whereas 
the most massive galaxies have lost little. Our results also show that the most 
massive galaxies have evolved faster and turned most of their gas into stars at an 
early time, thus strongly supporting a downsizing scenario for galaxy evolution. 

Subject headings: Galaxies: evolution; Galaxies: fundamental parameters 

1. Introduction 

In galaxies, the star formation rates are known to depend on the the cold (T < 200 K) 
gas densities via the well-known empirical Kennicutt (1998, hereafter K98) relation. Accord- 
ing to this relation, the star formation rate surface density E* is proportional to a certain 
power of the gas surface density S 9as , according to the relation £* oc Xjj as , with k ~ 1.4. This 
relation expresses the natural condition that the star formation rate depends on the amount 

- 2 - 

of gas available; galaxies dominated by young stellar populations have large reservoirs of 
gas, which can be turned into stars. On the other hand, gas-poor galaxies have consumed 
their gas in the past and/or have ejected it from its cold state, perhaps into the intergalactic 
medium, and at the present time they do not exhibit significant star formation. In dense 
environments, gas can also be stripped, owing to interactions with the intergalactic medium 
or to encounters with other systems. 

A fundamental quantity related to the gas accretion history of galaxies is the gas to stellar 
fraction (G/S). An important study of this quantity is the one of Kannappan (2004, here- 
inafter K04), who determined G/S for ~ 35000 galaxies from the Sloan Digital Sky Survey 
(SDSS) and 2MASS databases on the basis of photometric techniques. A similar study on 
a smaller sample of low mass galaxies has been performed by Geha et al. (2006, hereinafter 
G06). Previous important studies of the evolution of the star formation and gas content in 
galaxies at low and high redshift are the works by Erb et al. (2006a, 2006b) and Erb (2008). 
By means of analytic chemical evolution models, Dalcanton (2007) studied the chemical 
properties of local galaxies and put some quantitative constraints on their amount of gas 
infall and outflow. However, none of these studies was able to address how infall and outflow 
vary as a function of the galactic mass. 

In the present paper, we start from the SDSS galaxy sample and by means of the K98 relation 
and other simple physical assumptions, we aim at computing how the gas accretion history 
of local galaxies depend on their stellar masses. The sample we are considering includes 
~ 310000 galaxies, larger than the one considered by K04 by a factor of ~ 9. The strength 
of our method is that it is based on very few free parameters, whose range of values can 
be tested very easily, allowing us to draw robust conclusions on some fundamental physical 
properties of galaxies and on their evolution with cosmic time. We study how the gas mass 
of these galaxies has varied throughout their evolution and compare our predictions with 
observational data available in the literature. We focus on the present-day gas fractions 
and study how these quantities depend on the galactic stellar mass, investigating how the 
'downsizing' character of the galaxy populations affects their gas accretion history. 
The paper is organized as follows: in section 2, we present the galaxy sample and the MOPED 
algorithm used for the analysis. In section 3, we describe the methods used to derive the gas 
masses for the galaxies of the SDSS catalogue. In section 4, we presents our results and in 
section 5 we draw some conclusions. We assume a concordance cosmology with Q m = 0.24, 
tt A = 0.76 (Spergel et al. 2007) and h = 0.71. 

- 3- 

2. SDSS DR3 analysis with MOPED 

Following Panter et al. (2007a), we select galaxies from the third data release (DR3) 
main galaxy catalogue of the SDSS. In order to remove any bias and simplify our V max 
criterion (where V max is the maximum volume of the survey in which the galaxy could be 
observed in the SDSS sample), we have cut our sample at /i r < 23.0. The surface brightness 
cutoff of 23 mag / arcsec 2 excludes gas-rich, low surface brightness galaxies. However, very 
gas-rich galaxies are present in our sample, as we will see later in Sect. 4.1. 
At low redshifts, a small number of Sloan galaxies are subject to shredding - where a nearby 
large galaxy is split by the photometric pipeline into several smaller sources. 
Shredding affects a negligible number of galaxies for 0.01 < z < 0.25, for z < 0.01 it could be 
as many as 10% of galaxies. (SDSS collaboration, private communication). We have chosen 
to make a conservative cut at z — 0.02 to avoid this problem. This also removes the problems 
of non Hubble-flow peculiar velocities giving erroneous distances based on redshift, which 
can have a significant effect on recovered stellar mass and metal return. The total number of 
galaxies that satisfy our cuts is 299571. 

In order to estimate the completeness of the survey we have used the ratio of target galaxies 
to those that have observed redshifts (P. Norberg, Priv. Comm.). This does not allow for 
galaxies which are too close for the targetting algorithm, and we estimate this fraction at 
a 6% level from the discussion in Strauss et al. (2002). As a result of both these cuts, our 
effective sky coverage is 2947 square degrees. We also removed from our analysis a set of 
wavelengths which may be affected by emission lines, not modelled by the stellar population 

We use the 3 A single stellar population models of Bruzual & Chariot (2003) as the basis for 
this study, but we emphasize that the star formation histories allowed are very flexible. We 
assume an Initial Mass Function (IMF) given by Chabrier (2003), which is very successful 
at reproducing current observations in our galaxy. We chose to apply a uniform velocity 
dispersion of 170 kms -1 to the 3A models, reflecting a typical value for the Main Galaxy 
Sample. Although a velocity dspersion of 170 km/s is too large to represent the smallest 
mass galaxies in the sample, this discrepancy does not affect any of our results. 
Physical parameters of galaxies have been derived from their observed spectra by means of 
the MOPED algorithm (Heavens, Jimenez & Lahav 2000; Reichardt, Jimenez & Heavens 
2001). As with previous MOPED studies, for this analysis we have used a single foreground 
dust screen. We chose to use an LMC extinction law for the main analysis as given in Gordon 
et al. (2003). 


2.1. Recovering star formation and metallicity history using MOPED 

In the past, the SFH of galaxies was typically modelled by an exponential decay with 
a single parameter - for more complex models one or two bursts of formation were allowed. 
In fact, it would be better not to put any such constraints on star formation, particularly 
considering that each galaxy may have (as a result of mergers) several distinctly different 
aged populations. Star formation takes place in giant molecular clouds, which have a lifetime 
of around 10 7 years. Splitting the history of the Universe into the lifetimes of these clouds 
give a natural unit of time for star formation analysis, but unfortunately it would require 
several thousands of these units to map the age of galaxies formed 13 billion years ago, and 
the lack of sensitivity of the final spectrum to the detailed history would make any estimate 
of star formation history extremely degenerate. We choose a compromise solution, where 
we allow 11 time bins in which the star formation rate (SFR) can vary independently. This 
is still a difficult computational problem, but the MOPED (Heavens et al. 2000) algorithm 
reduces processing time to approximately 1 minute per spectrum on a standard workstation. 
The boundaries between the 11 different bins used are determined by considering bursts of 
star formation at the beginning and end of each period (at a fixed metallicity) and set the 
boundaries such that the fractional difference in the final spectrum is the same for each bin. 
This leads to a set of timebins tj which are almost equally spaced in log(lookback time). 
Nine timebins are spaced with a ratio of log(lookback time) of 2.07 in this application of 
MOPED, plus a pair of high-redshift bins to improve resolution at z > 1. The central ages 
for the timebins U are 0.0139, 0.0288, 0.0596, 0.123, 0.256, 0.529, 1.01, 2.27, 4.70, 8.50 and 
12.0 Gyr. Single-value decomposition analysis of SDSS spectra, as suggested by Ocvirk et al 
(2006), indicates that only 2-5 components can be reliably recovered, not 11, but MOPED 
in fact recovers < 5 significant components in 98% of cases (Tojeiro et al 2007). 
The gas which forms stars in each time bin is also allowed to have a metallicity which can vary 
independently. Extensive testing (Panter et al. 2003, Panter et al. 2007) shows that for large 
samples the average star formation history is recovered with good accuracy. The Bruzual 
& Chariot (2003) models allow metallicities between 0.02 < (Z/Z Q ) < 1.5. In order to 
investigate metallicity evolution (Panter et al. 2007) no regularization or other constraint is 
applied to the metallicity of the populations - each different age can have whatever metallicity 
fits best. A further complexity to the parametrization to deal with post-merger galaxies 
which contain gas which has followed dramatically different enrichment processes would be 
to have several populations with the same age but independent metallicities. It is possible 
to consider a more complex parametrization, but again one risks degeneracies in solution. 
With 11 ages, 11 metallicities and the dust parameter, the model has 23 parameters. The 
23 dimensional likelihood surface is explored by a Markov-Chain-Monte-Carlo technique 
outlines in Panter et al. (2003). Further information on the MOPED algorithm is contained 

- 5 - 

in Panter (2005). 

We take account of the 3 arcsec fibre aperture by correcting the spectrum upwards by the 
ratio of the r band Petrosian flux in the photometric survey to the r band flux received 
by the fibre. Although for individual galaxies this is likely to fail, for the population as a 
whole there is evidence from the colours (Glazebrook et al. 2003) that there is no overall 
bias between the sampling of stellar populations by the spectroscopy and photometry. 

3. Derivation of the gas masses 

The MOPED algorithm allows us to determine the star formation history of each galaxy, 
i.e. the star formation rate values at the end of the 11 timebins described above. For each 
timebin, we aim at determining the cold gas mass of any galaxy on the basis of its SFR, 
by inverting the K98 relation, which links the gas surface density to the SFR per unit 
area. A similar technique was used by Erb et al. (2006a) and Erb (2008) to derive the gas 
fractions for a sample of star forming galaxies at z ~ 2 and to study the implications of 
the mass-metallicity relation observed at high redshift on the galactic gas accretion history, 

Following K98, for any galaxy and at a given time, the gas surface density S 9as , expressed 
in M Q pc~ 2 , depends on the SFR surface density £*, expressed in M Q yr~ l kpc~ 2 according 

£ -( ) a714 M Q pc~ 2 (1) 

9 v 2.5 x 10" 4; oy y ' 

The MOPED catalogue provides the deprojected SFRs in units of M Q /yr, hence to 
derive the corresponding surface densities we need an estimate of the galactic scale-length 
radius as a function of the baryonic mass. We calculate this quantity by means of the scaling 
relations by Mo, Mao & White (1998) and we follow its evolution with lookback time. A 
major difference between the works by Erb et al. (2006a) and ours concern mainly our 
detailed study of the scaling radius as a function of mass and redshift. 
For each galaxy, we know the present day stellar mass M„, whereas the scale length is likely 
to be set principally by the total mass. We assume that each galaxy is embedded inside a 
dark matter halo of mass M = M*/ /, where / is a parameter which we take to be a constant. 
The value of / and its dependence on mass and time are, in a sense, part of what we are 
trying to obtain, but here we use / only to set the scale-length of the galaxy. As we will see 
later, the assumption for / has a minor impact on our results. 

Initially, we will make the assumption (relaxed later) that the baryons collapse into a disk, 

- 6- 

which is assumed to be thin and in centrifugal balance, with an exponential surface density 
profile Ti tot (R) = S exp(— R/R d ). S = T, gas + is the central density of the disk, having 
an angular momentum which is a fraction / of the angular momentum of the dark matter 

For the halo, we assume a Navarro, Frenk & White (1997) mass profile and we take into 
account the effects of the self-gravity of the disk. For each halo of mass M, the virial radius 
i? 2 oo is defined as 

R20 ° = \M{z) (2) 


V c = [10 GMH(z)] 1/3 (3) 

where G is the gravitational constant and H(z) is the Hubble parameter as a function of the 
redshift z, given by 

h(z) = h [q a + (i - n A - n m )(i + zf + n m (i + zf} 1 ' 2 (4) 

where Q m , Qa and H are the matter density parameter, the vacuum energy parameter and 
the Hubble constant at z — 0. 

The scaling radius Rd can be calculated as (Mo et al. 1998): 

Ai?2oo/c t , I 1 / 2 /fi(A, c vir , fb) 
R d ^ • (5) 

A is the spin parameter of the halo and depends on the total energy of the halo E, its angular 
momentum J and its mass M according to: 

\ = J\E\ l/2 G- l M- b/2 . (6) 

The quantity A is likely to assume values in the range 0.01 < A < 0.1 (Heavens & Peacock 
(1988), Barnes & Efstathiou (1987), Jimenez et al. 1998). In this paper, we assume a typical 
value of A = 0.05. Scatter in A would propagate as scatter in the gas fractions, but we 
average over many galaxies to obtain the average gas fractions. 

c V i r is the halo concentration factor, and is calculated following Bullock et al. (2001) and 
Somerville et al. (2006) i.e. for each halo, by defining a collapse redshift z c as M*(z^) = F-M. 
c vir is given by c vir (M, z) = K(l + z c )/(l + z), where F and K are two adjustable parameters. 
Following Somerville et al. (2006), we assume K = 3.4 and F = 0.01. 

To compute the quantities f Cmr and /r, we use analytic fitting functions presented in Mo, 
Mao & White (1998). 


The use of these scaling relations described above is physically motivated and, as shown by 
Somerville et al. (2006), accounts for the observed redshift evolution of the disk size-stellar 
mass relation from the GEMS (Galaxy Evolution from Mophology and SED; Rix et al. 2004) 
survey out to z ~ 1. 

For each galaxy, if if) is the SFR in units of M Q /yr and if we assume for the SFR surface 
density profile £*(i?) = X*,o exp(—lAR/Rd), the central SFR surface density is then given 

* 1> 



The gas surface density £ 9as is then determined by the K98 relation, and the gas mass M gas 
(in Mq) is given by: 

M gas = Y, gas x 2nR 2 d . (8) 

In section 4.1, we relax the assumption of an exponential disk profile, and show how the 
assumptions of a different surface density profile and of different scaling relations affect the 
calculated galactic gas masses. 

At this stage, we can see how the calculated gas mass M gas depends on the baryonic fraction 
/. The quantity Rd/R2oo as a function of / has been calculated by Somerville et al. (2006) 
and can be fitted by an exponential function R d /R 2 oo — 0.77 exp(— 4.7/). The dependence 
of M gas on / is then approximately 

M gas oc exp(-2.7f)/f - 19 (9) 

The dependence of the resulting gas masses on / is hence very weak, as a result of which 
any reasonable estimate for / will suffice. 

In setting the scale length, we take / = 0.06, in agreement with the recent results of Hoekstra 
et al. (2005), based on the weak-lensing signals of isolated galaxies at low redshift. This is 
also consistent with most of the cold baryons being in stellar form. Several recent cosmic 
baryon budget calculations (Calura & Matteucci 2004, Fukugita & Peebles 2004, Shankar et 
al. 2006) support this hypothesis, in particular for spiral disks and spheroids. 

A potential difficulty of our study is that the analysis above assumes that the galaxy 
remains a single structure during its history, whereas the star formation history deduced from 
the fossil record makes no statement about where the stars were when they were formed. 
Hence a natural question to ask is how the possible presence of merging affects these results. 


However, if we assume the scaling relation R d oc MJ holds for any sub-units before merging, 
the final gas-to-star ratio is almost independent of the amount of merging which has taken 
place. To see this, let us assume that at some lookback time the observed galaxy was in 
several pieces i = 1, . . . n, each containing a fraction of the final stellar mass. If we 
assume the star formation rate is proportional to f iy then the gas mass in each sub-unit is 

- 8- 

proportional to (ipi/ ' R 2 dj )°' 114: R 2 di oc / 4 9 , so the total gas mass is modified to 


M gas ^M 9as J2ff 9 - (10) 


Since Yli fi = 1j we see that for any reasonable merging history, the gas mass is virtually 
unchanged. The assumptions we make here are certainly challengeable, but this calculation 
gives us confidence that the results are likely to be robust to the merger history. 

4. Results 

4.1. The evolution of the gas to stellar mass fractions 

In this section, we present how the gas accrection history (i.e. the infall history) of our 
galaxies evolved with cosmic time. In section |4"72"1 we will present how the outflow history of 
our galaxies evolved with cosmic and with galactic mass. 

One way to study the evolution of the gas accretion history is to analyze the evolution with 
mass and with cosmic time of the gas to stellar mass ratio (G/S). In figured], we show the 
evolution of G/S calculated as described in section 3, as a function of the stellar mass for the 
MOPED galaxies at various lookback times. The dispersion of the calculated G/S is relatively 
small at a lookback time of 12 Gyr. The dispersion however remains large (the individual 
G/S span 6 orders of magnitude) up to the preset time. At any lookback time, the average 
G/S ratio decreases as the stellar mass increases in the range 1O 6 M < M* < 1O 12 M . Note 
that the highest mass bins (M* > 10 M Q ) are characterised by very few systems with a 
peculiar behaviour. In the last Gyr of evolution, our predictions indicate an increase in the 
average G/S ratios for the galaxies with stellar masses M* < 1O 1O M , due to the fact that 
these galaxies must have accreted gas at late times. 

The anti-correlation between (G/S) and stellar mass found at any lookback time can be 
interpreted in the following way. In general, higher mass galaxies are more efficient at 
turning gas into stars, and lower mass galaxies are less efficient and so retain a large amount 
of their primordial gas. This is clearly visible also in Fig. [2j where we show the evolution of 
(G/S) as a function of lookback time for several mass bins. In this figure, it is clear that the 
slope of the curves steepens from low mass to high mass systems, indicating decreasing gas 
consumption timescales from dwarf to giant galaxies. This behaviour is in agreement with 
the downsizing picture of galaxy evolution, according to which the least massive systems 
formed the bulk of their stars (consuming most of their gas reservoirs) at recent times, 
whereas in the most massive galaxies the buildup of the stellar mass was complete several 

- 9- 

Gyrs ago. 

At the present time, we predict individual \og(G/S) values between —4 and ~ 1.5. This 
range of values is in agreement with existing local observational estimates for galaxies of 
various morphological types. The lowest estimates of G/S have been derived for local dwarf 
spheroidal galaxies and for large ellipticals. The gas-poorest dwarf spheroidal galaxies of the 
Local Group have upper limits on the atomic H to blue luminosity ratio of Mhi/Lb ~ 0.001 
(Mateo 1998). These values correspond to log(G/5') ~ —3.67, assuming a helium correction 
factor of 1.4 and a stellar mass-to-light ratio of (M/L) BjE = 6.5M /L (Fukugita, Hogan & 
Peebles 1998). In local SO and E galaxies, Sansom et al. (2000) observed values of M H i/L B 
down to ~ O.OOO1M /L , corresponding to \og(G/S) ~ -4.67. 

On the other hand, the largest gas reservoirs are observed in local irregular galaxies. Hunter 
& Elmegreen (2004) find for local irregulars Mhj/Lb up to ~ 5M /L , corresponding to 
\og(G/S) ~ +0.8, assuming a helium correction factor of 1.4 and a stellar mass to light 
ratio of (M/L) B,»rr ~ 1Mq/Lq (FHP98). These observational estimates are in very good 
agreement with the lower and upper extremes of the G/S values we derive for present-day 

All our calculations have been performed by assuming the same scaling relations for 
all galaxies, i.e. we have assumed that all our galaxies are self-gravitating disks, embedded 
in dark matter halos. This assumption may seem unrealistic, since spheroids are known to 
follow scaling relations different from disk galaxies. To tackle this problem, we have divided 
our galaxy sample into two categories, the first composed by disk galaxies, the second by 
spheroids. Disk galaxies follow the scaling relations described in section 3. For spheroids, 
instead of an exponential surface mass density profile, we have assumed a Sersic law, given 

Z sph (R) = S[,exp {-b n [(R/R eff ) l / n - 1]} (11) 

(Sersic 1968), where b n ~ 2n — 0.324 (Ciotti & Bertin 1999). We want to test two possible 
values for the index n: n = 4, corresponding to the de Vaucouleurs (1948) law, and n — 2, 
since most galaxies do not fit into the binary "de Vaucouleurs" or "exponential disk" cate- 
gories but instead span some range of Sersic parameters (Blanton et al. 2003). 
In this case, for the SFR surface density profile, we have assumed that S* iSp fe(i?) oc [^^(i?)] 1 ' 4 
To compute the effective radius R e ff, we have used the Shen et al. (2003) relation between 
R e ff and M*: 

/ M \ a56 

R >»= iM {w*k) kpc (12) 

(see also Boylan-Klochin et al. 2005, Robertson et al. 2007), valid at any lookback time. 

The distinction between disks and spheroids has been performed by means of two crite- 
ria: (1) the stellar mass of the galaxy and (2) its present day SFR. The first criterion seems 


H — — I — I — I — — I — I — I — — I — h 

ee e ee © 

2.27 Gyr - 

e q 

i- ■ i 


8 10 12 

log(M./M @ ) 

H — — I — I — I — — I — I — I — — I — h 



=l I- . I 



10 12 
log(M./M ) 

Fig. 1. — Evolution of the gas to stellar mass ratios as a function of the stellar mass for the 
MOPED galaxies at various lookback times. The small grey squares represent the individual 
values, whereas the black open circles with error bars represent the arithmetic means in 
various mass bins, plotted with their uncertainties. From the upper right panel, in clockwise 
sense, we show the values computed at lookback times of 12 Gyr, 8.5 Gyr, 4.7 Gyr, 2.27 
Gyr, 1.01 Gyr and Gyr. 

5 10 

lookback time (Gyr) 

Fig. 2. — The gas to stellar mass ratio as as a function of lookback time for several stellar mass 
M* bins. Solid black thin line: M* < 6.5M . Dotted blue thin line: 6.5 < M*/M < 7.5. 
Short-dashed cyan thin line: 7.5 < M*/M < 8.5. Long-dashed green thin line: 8.5 < 
MJMq < 9.5. Dot-short-dashed yellow thin line: 9.5 < MJM Q < 10.5. Dot-long-dashed 
magenta thin line: 10.5 < MJM Q < 11.5. Solid black thick line: M* > 11.5M . 


i 1 r 

i 1 r 

i 1 r 

"i r 


_ # default scaling 

M. (n = 4) 



SFR (n = 4) 

k SFR (n = 2) 

J L 

J L 


8 10 

log(M./M ) 


Fig. 3. — The gas to stellar mass ratio as a function of the stellar mass for the MOPED 
galaxies, computed with the same default scaling relations for all galaxies (solid circles), 
using different scaling relations for spheroids and spirals, first disentangling by means of the 
stellar mass (solid squares, see text for details) and secondly by means of the current SFR, 
assuming Sersic index n = 4 (open circles) and n = 2 (solid triangles). 


plausible since it is known that at low redshift, early type galaxies have stellar masses higher 
than late types and that the mass function of galaxies at stellar masses M* < lO n M and 
M* > 1O 11 M is dominated by late types (i.e. disks) and early types (i.e. possibly spheroids, 
Pannella et al. 2006, Shankar et al. 2006), respectively. On the basis of this, we have com- 
puted the gas masses assuming that all the galaxies with M* < 10 11 M Q obey the scaling 
relations described in section 3, whereas all the galaxies with stellar mass > 10 11 M Q 
are characterized by a Sersic surface density profile, as given by equation [IT], and follow the 
scaling relation of eq. [T^J We are aware that this criterion prevents us from adopting the 
appropriate profile for low-mass early type galaxies, such as dwarf spheroidals and dwarf 
ellipticals. For this reason, we consider an alternative criterion for the distinction between 
disks and spheroids, based on the SFR. 

In the second case, we have assumed that the galaxies with a present-day SFR lower than a 
given threshold (we have used a threshold value of 0.5 M & /yr) and having assembled more 
than 50% of their total stellar mass during the first timestep, belong to the category of 
spheroids, whereas the rest are spiral disks. In Fig. [3j we show the present-day gas to 
stellar mass ratio as a function of the stellar mass, computed assuming the same scaling laws 
for all galaxies (solid circles), different scaling laws for spheroids and disks distinguished 
by mass (solid squares) and by current SFR. While distinguishing by SFR, we plot our re- 
sults assuming two different values for the Sersic index n of the density profile assumed for 
spheroids: n = 4 (open circles) and n = 2 (solid triangles). 

In any case, we obtain gas masses and gas fractions very similar to the ones obtained assum- 
ing the same scaling relations for all the galaxy population. The same situation occurs at 
larger lookback times, i.e. at any redshift, the gas masses computed in the three cases are 
very similar. This indicates that our results are fairly independent of the adopted profile. 
Motivated by this fact, all the results presented in the remainder of the paper have been 
obtained by adopting the same disk scaling relations for the whole galaxy sample. 
At this point, our aim is to understand how our estimates of the scaling radii performed 
by means of the exponential profile compare with the optical radii of the SDSS galaxies, 
determined from the observed light profiles. In Fig. HI left panel, we show the computed 
scaling radii as a function of the r-band Petrosian radii Rp for the SDSS sample, overplotted 
with the mean are ± 1 a errors. In Fig. HI the darkest regions correspond to the densest 
areas, where most of our galaxies lie. This figure shows that for the bulk of the galaxies and 
for radii < 8kpc, there is almost a 1:1 correspondence between scaling radii and Petrosian 
radii. The scaling radii underestimate the optical radii for Rp > 8kpc. The reason for this 
discrepancy may be found in Fig. HI right panel, where we show the expected angular size, 
computed from the scaling radii, against actual angular size on the sky for the SDSS galax- 
ies. From this plot, we can see that the deviations are largest when the size on sky is very 
small or very large, and that the discrepancy concerns the minority of the galaxies, hence it 


is not a major reason of concern. The match between scaling radii and Petrosian radii seem 
problematic when the galaxy radius is < 4 arcsec and > 18 arc sec, otherwise there is a very 
good agreement. For a limited number of galaxies, the underestimation of the actual size 
may imply an underestimation of the gas fractions. 

Galaxies of the SDSS sample have been shown to split into two different populations, one 
composed of blue galaxies dominated by recent star formation episodes and another com- 
posed of red, passively-evolving early types (Strateva et al. 2001, Blanton et al. 2003). In 
figure El we show the calculated present day G/S values as a function of the stellar mass for 
these two different galactic populations, compared to the corresponding observational esti- 
mates by Kannapan (2004). The estimates by Kannapan (2004) are based on photometric 
techniques. The atomic G/S are estimated using the u — K colours of ~ 35000 galaxies 
from SDSS DR2 and 2MASS databases. This technique is calibrated by means of HI data 
from the HyperLeda HI catalogue. By means of her technique, Kannapan (2004) estimated 
individual log(G/S') values between ~ —2 and ~ 0.5. The G/S estimates are very reliable 
in the lowest mass bins but tend to represent overestimations in the largest mass bins (S. 
Kannapan, private communication). For this reason, here we compare the average G/S val- 
ues computed considering only the galaxies with \og{G/S) > —2. 

By comparing our results with the ones of Kannapan (2004), we are able to obtain constraints 
on the epoch at which galaxies of the blue and red populations assembled the bulk of their 
stars. In particular, we can reproduce very well the data obtained by K04 if we assume that 
the red population consists of all the galaxies having formed > 60% of their present-day 
stellar mass during the first 4 timesteps, i.e. in ~ 11.4 Gyr. The overall agreement between 
our estimations and the observational estimates by Kannapan (2004) is good for all galaxies 
with stellar masses log(M*/M & ) < 11. For the most massive galaxies of the red population, 
we tend to underestimate the gas fractions. This is not surprising, since as stated above, the 
estimates by Kannapan (2004) are likely to represent overestimations of the actual values in 
particular in the highest mass bins. For the most massive galaxies of the blue population, 
we predict mean (G/S) higher than the ones determinded by K04 because of the fact that 
highest mass bins contain very few galaxies, some of which present very high gas fractions, 
considerably increasing the average G/S in these bins. 

The study by Panter et al. (2007b, in preparation) shows that it is possible to split 
the SDSS galaxies into two populations on the basis of their colour excess E(B — V), which 
expresses the amount of dust present in any galaxy. To disentangle between the blue and 
the red galactic populations, we tested also this criterion. In this case, the red and blue 
components are the galaxies with the E(B — V) < 0.01 and E(B — V) > 0.01, respectively. In 
figure El we present the predicted mean (G/S) values for the red and blue galaxy populations 
as compared to the observational data by K04, having determined the distinction between 

- 15 - 

the two populations by means of their colour excess E(B — V). The separation between blue 
and red appears less clear than in the former case. Furthermore, the agreement between 
our predictions and the data by K04 is worse than the one shown in Fig. In fact, in the 
case of Fig. El the observed (G/S) are systematically underestimated by our predictions, in 
particular the data for the blue galaxy population. This shows that the colour excess is not 
a totally reliable diagnostic to disentangle blue and red galaxies, being strongly dependent 
on the inclination causing overlapping between the two different populations (Calzetti 2001). 
To conclude this section, we compare our results for the estimated gas fractions with some 
individual cases of the sample studied by Geha et al. (2006). 

G06 studied the gas content for a sample of low luminosity dwarf galaxies selected from the 
Sloan Digital Sky Survey. This sample consits of 101 galaxies, for which they have obtained 
follow-up Hj observations using the Arecibo Observatory and Green Bank Telescope. Using 
the measured masses of Hj and the optical properties, G06 derived the i7/-to-stellar mass 
ratio Mhj/M* for the galaxies of their sample. 

We have indentified 23 galaxies from our sample that are also present in the G06 sample. 
For this 23 galaxies, we compare our values for the stellar masses and MhJM* to the values 
found by Geha et al. (2006) by means of a different technique. The results of this comparison 
are shown in table 1. For each galaxy, we derive the quantities by multiplying the gas 
mass M g by the solar photospheric H mass fraction Xh = 0.75 (Lodders 2003). 
In some cases, the stellar mass values estimated by means of the MOPED algorithm differ 
from the values reported by G06. These differences may be due to several factors, such as the 
assumption of a different IMF (G06 use the IMF by Kroupa et al. 1993) and uncertainties 
in the distances because of peculiar velocities. G06 do not present any uncertainty for 
their determined stellar masses. For 11 out of 24 galaxies, the discrepancies between our 
Mhj/M* values and the ones found by G06 are within 0.7 dex, corresponding to a factor 
of 5. In 10 cases, we severely underestimate the M^/M* and in the few remaining cases, 
we overestimate the results by G06 by a factor larger than 5. The very low M^/M* we 
obtain in some cases may be due to underestimated galaxy dimensions or to the presence 
of infalling gas not taking part to star formation, as observed for the local Sculptor dwarf 
spheroidal galaxy (Carignan et al. 1998). However, most likely it is an indication of the fact 
that MOPED cannot determine precisely the star formation history of individual galaxies 
but of ensambles of them, and only for large numbers the errors are very small (see also 
Fig. 8). There is however another trend worth pointing out: those galaxies for which we find 
the best agreement in the Geha sample are the ones that show strongest absorption lines, 
thus providing MOPED a better handle to recover the star formation history, while those 
that give the worse agreement do have only emission lines (recall that we only model the 
continuum and absorption lines with MOPED. 


Table 1: M Hl /M* calculated for single galaxies and compared to the results by Geha et al. 

Name This paper Geha et al. (2006) 


\og(M Hl /M*) 

log(M, ) log(M Hl /M.) 


o n c o o 






f .oo /by 

n ii 071 
U. llz ( 1 




lyzy i 1 








-z.Uo /oo 





I (UUz 


o 7/1 on 
-I. /4oyo 




















































































































4.2. Fractions of lost baryons in SDSS galaxies 

In galaxies, gas can be both transformed into stellar form and ejected into the external 
environment, i.e. the intergalactic or intracluster media (ICM, IGM). Gas can be ejected by 
galaxies by means of various processes. Supernova (SN) explosions heat the interstellar gas 
and, as soon as the binding energy of the gas exceeds its thermal energy, some fraction of 
the ISM can be removed through SN-driven galactic winds (e.g., Larson 1974). In general, 
isolated galaxies eject gas mostly by means of galactic winds. Galaxies in dense environments 
(i.e. clusters and groups) undergo gas loss via additional mechanisms, which in general are 
bound to environmental effects and can be of various types: tidal interactions, ram pressure 
stripping, viscous stripping, starvation and thermal evaporation (see Boselli & Gavazzi 2006 
and references therein). In this section, we aim at studying how the total amount of cold gas 
ejected by means of any mechanism evolved with cosmic time. Our approach does not allow 
us to infer which are the main mechanisms driving gas ejection, for which more complex 
dynamical simulations or galactic chemical evolution models would be suited, where SN 
feedback is taken into account (Matteucci 1994; Recchi et al. 2002). 

To perform this task, once again we use the 11 timebins described in sect. 12.11 For each 
galaxy, at each timebin U we know the total baryonic mass M to t{U) = M gas (ti) + M*(£j). We 
can compute the net ejected (or accreted) mass as the difference between the total baryonic 
mass computed at two following timesteps and t i: M e j = M tot (ti) — M tot {ti_i). This 
quantity may be positive or negative, with the negative results indicating mass loss. Note 
that gas which is heated to the extent that it does not take part in star formation is counted 
as ejected, regardless of whether it is physically removed from the galaxy or not. We can 
use this information to compute the complete mass outflow histories for the galaxies of our 
sample. When baryons are lost or accreted, the total mass of the galaxy goes down, which 
puts it in a different mass bin. The current stellar mass of a galaxy depends on the star 
formation hostory of that galaxy, which is used here to determine the infall and ouflow 

In Figure El we show the computed evolution of the mean fraction of lost baryons fi ost , 
defined as fi ost = , /f — , as a function of the stellar mass for the SDSS catalogue. With 
one exception, these curves show a very consistent picture of progressive baryon loss over 
time for all but the most massive galaxies, with the lowest-mass systems losing much more 
baryon mass than the high-mass systems. The exception to this is the curve at a lookback 
time of 1.1 Gyr, which is anomalous. The reason for this is almost certainly connected with 
the difficulty of determining the star formation rate for this population, as there is nowhere 
in the spectrum where the population is dominant (Mathis et al, 2006). 

The mean fi ost values computed for large galaxies (log(M* / 'M®) > 11) do not show 
a strong evolution with time, remaining at a low level of ~ 10 — 20%. Combined with 


the G/S ratio for these systems, we conclude that early star formation is very efficient in 
these galaxies. On the other hand, the mean fi ost values computed for galaxies in the range 
7 < log (M*/M Q ) < 11, undergo a strong evolution with lookback time, and this indicates 
that the galaxies of lowest mass are continually losing baryonic mass from the cold phase 
over a long period. This result is a confirmation of the predictions from chemical evolution 
models for elliptical galaxies (Matteucci 1994), which showed that, in order to explain the 
increase of the [Mg/Fe] ratio with galactic mass, one has to assume that the efficiency of star 
formation is an increasing function of the galactic mass. One possibility for this is an "Inverse 
Wind" scenario of galaxy evolution, in which the most massive galaxies are the first ones 
to experience the galactic winds and to complete their outflow and star formation history, 
whereas the dwarf galaxies undergo substantial outflows and continuous star formation until 
most recent times. 

As one would expect from the depths of the potential wells, we find very clear evidence 
that at the present day, the lower-mass galaxies have lost a larger fraction of their cold 
baryons, as much as 90% for galaxies with a stellar mass of 1O 6 M . An analytical fit to the 
present day fi os t-M* relation is 

valid for 6 < log(M*/M Q ) < 12.5. An increasing trend of the ejected fractions with de- 
creasing galactic mass is also confirmed by analytical and numerical single-burst models for 
dwarf galaxies (Mac Low & Ferrara 1999). Our results are in good agreement with results 
from galactic chemical and chemo-dynamical models considering more complex star forma- 
tion histories. By means of chemo-dynamical simulations, Recchi et al. (2002) showed that 
a dwarf galaxy of baryonic mass of ~ 1O 7 M undergoing multiple starbursts can eject up 
to ~ 75% of its mass. Chemical evolution models for elliptical galaxies of masses between 
10 9 M Q and 1O 12 M predict decreasing ejected fractions, with values between 90% and 10 % 
(Gibson & Matteucci 1997, Pipino et al. 2002). 

The shape and normalization of the present day fi os t vs M* curve derived observationally 
in this study is in good agreement with theoretical predictions from chemo-dynamical models 
where only supernova feedback has been employed. While we cannot rule out that AGN 
feedback can be also at play with the present observations, it seems that supernova feedback 
alone is able to displace the observed amount of cold gas from the dark halo as a function 
of mass. 

This study does not exclude preheating (Mo et al. 2005, Crain et al. 2007), which prevents 
accretion of gas, but rather provides evidence of some process which has ejected gas from 
the cold phase after accretion. 



Finally, in fig. [8], we show the fi os t vs G/S for all the dwarf galaxies of our sample, 
in this case all the galaxies with M* < 1O 9 M . Here, we show the behaviour of all the 
individial galaxies and the mean values, plotted with the error bars. Fig. [8] shows a clear 
anticorrelation between the lost fraction and the present-day gas fraction. The two average 
values computed in the lowest G/S bins apparently show an anomalous behaviour, but these 
two bins contain very few systems, 8 total. The larger shaded area and the smaller dotted 
area indicate the regions where 100% and 85 % of the dwarf galaxies of the sample by GOG 
lie, respectively. Fig. [8] shows that our results are in substantial agreement with the ones of 
G06. According to Geha et al. (2006), the baryonic Tully-Fisher relation indicates that gas 
removal processes were not important for the majority of their sample. The Geha sample 
includes dwarf galaxies with extremely high gas fraction, which, according to our results, 
must have retained the majority (> 60% if we consider the 85% of the Geha galaxies) of 
their present mass. 

Our study strongly indicates that the main driver of the fraction of lost gas is the mass of the 
galaxy. Environmental effects could also play a role in determining the fraction of gas lost 
from galaxies, in particular for low mass galaxies (Geha et al. 2006). However, the SDSS 
galaxies span a large range of different environments and if the environmental effects were 
dominant, we should not expect such a clear correlation between lost gas and stellar mass. 
Furthermore, recent studies (Sheth et al. 2006, Blanton et al. 2006, Mateus et al. 2007, 
Berta et al. 2008) of the relation between galaxy colours, luminosities and star formation 
histories and environment for the SDSS sample have shown that such properties are more 
related to the mass of the host dark matter halo than to the large scale environment. The 
marked correlation studies in Mateus et al. (2007) and Berta et al. (2008) show that the 
environmental dependence is mostly driven by the mass, although other physical parameters 
like spin do play a secondary role. A more detailed study of lost gas fraction as a function 
of environment is deferred to a future paper. 

The results of this paper have been obtained self-consistently from the star formation history 
of the SDSS galaxies, in a way which should be robust to the degree of merging which has 
taken place. The star formation history is linked, via the Kennicutt law, to the gas mass 
present at any time, and hence allows us to trace the net gas accretion/outflow history. 
The metallicity history can be used to determine the metal outflow history, the lost metal 
fractions and the chemical enrichment of the inter galactic medium; this will the subject of 
a forthcoming paper. 


5. Conclusions 

In this paper, we used the Kennicutt (1998) relation, linking the surface star formation 
rate to the gas mass surface density, together with the scaling relations by Mo, Mao & 
White (1998), linking the galactic scale-length radius to the baryonic mass, to recover the 
gas masses for ~ 310000 galaxies of the SDSS DR3 sample. Our main conclusions can be 
summarised as follows. 

1) We studied the time evolution of the gas to stellar mass fractions for all the galaxies of 
the SDSS catalogue. In the last Gyr of evolution, our predictions indicate an increase in 
the average (G/S) for the galaxies with stellar masses M* < 1O 1O M , in agreement with the 
the downsizing picture of galaxy evolution, where the most massive galaxies form the bulk 
of their stars at early times. At the present time, we predict individual \og(G/S) values 
between —4 and ~ 1.5. These lower and upper limits are in agreement with independent 
estimates of the (G/S) in the local gas poorest and richest galaxies, respectively. 

2) We split the galaxy sample into a red and a blue population by means of two different 
criteria, i.e. the recent amount of star formation and the colour excess E(B — V). The 
(G/S) calculated by adopting the first criterion are in good agreement with independent 
estimates based on photometric techniques (Kannappan 2004), in particular for the galaxies 
with stellar masses log(M*/ 'M ) < 11. The adoption of the colour excess criterion implies 
instead a less clear distinction between the two galactic populations. 

3) We computed the time evolution of the average fraction of lost baryons fi os t, defined as 
fiost = u ~r% — as a function of the stellar mass for the SDSS catalogue. With the exception 
of an anomaly at a lookback time of lGyr (which we put down to the difficulty of determining 
star formation rates at this time; Mathis et al 2006), the results show clear signatures that 
the fraction of lost cold baryons increases with time for all but the highest-mass galaxies, and 
that the fraction lost is heavily dependent on the stellar mass of the galaxy. The lost fraction 
varies between about 10-20% for M* > 1O 12 M , to ~ 80 - 90% for low-mass (M* < 1O 7 M ) 
galaxies. The significant loss of gas from the low-mass systems is not surprising given the 
small potential wells involved, and may be responsible for the lack of observed low-mass 
galaxy satellites compared with simulations (Klypin et al 1999, Moore et al 1999), even 
including the latest dwarfs detected in SDSS (e.g. Belokurov et al 2007). Our results are in 
agreement with chemical evolution models for elliptical galaxies (Matteucci 1994), and could 
be explained by an "Inverse Wind" scenario for galaxy evolution, i.e. that the most massive 
systems are the first ones to undergo mass loss via galactic winds. In any event our results 
show clearly the "downsizing" character of the galaxy population, with the star formation 
in high-mass galaxies essentially complete at early times. 

-21 - 


We wish to thank Simone Recchi, Sheila Kannappan, Daniela Calzetti and Antonio 
Pipino for many interesting discussions. FC thanks the hospitality of the Department of 
Physics &; Astronomy of the University of Pennsylvania, where part of this work was carried 
out. We thank an anonymous referee for comments and suggestion that helped us improve 
the paper. 


Barnes, J., Efstathiou, G.P, 1987, ApJ, 319, 575 
Belokurov, V. et al. 2007, ApJ, 654, 897 

Berta, Z., Jimenez, R., Heavens, A. F., Panter, B., 2008, MNRAS, submitted, 

Blanton, M. R., et al. 2003, ApJ, 594, 186 

Blanton, M. R., Eisenstein, D., Hogg, D. W., Zehavi, I., 2006, ApJ, 645, 977 
Boselli, A., Gavazzi, G., 2006, PASP, 118, 517 

Boylan-Kolchin, M., Ma, C.-P, Quataert, E. 2005, MNRAS, 362, 184 
Bruzual G., Chariot S., 2003, MNRAS, 344, 1000 

Bullock, J. S., Dekel, A., Kolatt, T. S., Kravtsov, A. V., Klypin, A. A., Porciani, C, Primack, 
J. R. 2001, ApJ, 555, 240 

Calura F., Matteucci F., 2004, MNRAS, 350, 351 

Calzetti, D., 2001, PASP, 113, 1449 

Carignan, P., et al, 1998, AJ, 116, 1690 

Chabrier, G., 2003 ,PASP, 115, 763 

Ciotti, L., Bertin, G. 1999, A&A, 352, 447 

Crain, R. A., Eke, V. R., Frenk, C. S., Jenkins, A., McCarthy, I. G., Navarro J. F., Pearce, 
F. R., 2007, MNRAS, 377, 41 

-22 - 

Dalcanton, J. J., 2007, ApJ, 658, 941 

de Vaucouleurs, G. 1948, Ann. d'Astrophys., 11, 247 

Erb., D. K., 2008, ApJ, 674, 151 

Erb, D. K., Steidel, C. C, Shapley, A. E., Pettini, M., Reddy, N. A., Adelberger, K. L., 
2006a, ApJ, 646, 107 

Erb, D. K., Steidel, C. C, Shapley, A. E., Pettini, M., Reddy, N. A., Adelberger, K. L., 
2006b, ApJ, 647, 128 

Fukugita M., Peebles P. J. E., 2004, ApJ, 616, 643 

Fukugita, M., Hogan, C. J., Peebles, P. J. E., 1998, ApJ, 503, 518 (FHP98) 
Geha, M., Blanton, M. R., Masjedi, M., West, A. A., 2006, ApJ, 653, 240 
Gibson, B. K., Matteucci, F., 1997, ApJ, 475, 47 
Glazebrook, K., et al, 2003, ApJ, 587, 55 

Gordon, K. D., Clayton, G. C, Misselt, K. A., Landolt, A. U., Wolff, M. J., 2003, ApJ, 594, 

Heavens, A. F., Jimenez, R., k Lahav, O. 2000, MNRAS, 317, 965 
Heavens, A. F., Peacock, J., 1988, MNRAS, 232, 339 
Hunter, D. A., Elmegreen B. G., 2004, AJ, 128, 2170 

Jimenez, R., Padoan, P., Matteucci, F., Heavens, A. F., 1998, MNRAS, 299, 123 
Kannappan, S. J., 2004, ApJ, 611, L89 (K04) 
Kennicutt, R. C, 1998, ApJ, 498, 541 (K98) 

Klypin A., Kravtsov A. V., Valenzuela O., Prada, F., 1999, ApJ, 522, 82-92 
Kroupa, P., Tout, C. A., Gilmore, G. 1993, MNRAS, 262, 545 
Larson, R. B., 1974, MNRAS, 169, 229L 
Lodders, K., 2003, ApJ, 591, 1220 

Mac Low, M.-M., Ferrara A., 1999, ApJ, 1999, 513, 142 


Mateo, M.L., 1998, ARA&A, 36, 435 

Mateus, A., Sodre, L., Cid Fernandes, R., Stasinska, G., 2007, MNRAS, 374, 1457 
Mathis H., Chariot S., Brinchmann J., 2006, MNRAS, 365, 385 
Matteucci, F., 1994, A&A, 288, 57 

Mo, H. J., Mao, S., White, S. D. M., 1998, MNRAS, 295, 319 

Mo, H. J., Yang, X., van den Bosch, F. C, Katz, N., 2005, MNRAS, 363, 1155 

Moore, B. et al. 1999, ApJ, 524, L19 

Navarro, J. F., Frenk, C. S., White, S. D. M. 1997, ApJ, 490, 493 

Ocvirk, P., Pichon, C, Lanon, A., Thibaut, E., 2006, MNRAS, 365, 460 

Pannella, M., Hopp, U., Saglia, R. P., Bender, R., Drory, N., Salvato, M., Gabasch, A., 
Feulner, G., 2006, ApJ, 639, LI 

Panter, B., Heavens, A. F., Jimenez, R., 2003, MNRAS, 343, 1145 

Panter B., Thesis, 2005. Availiable at from the Edinburgh Research Archive at 

Panter, B., Jimenez, R., Heavens, A. F., & Chariot, S. 2007, MNRAS, 378, 1550 

Panter, B., et al. 2007b, in prep. 

Pipino, A., Matteucci, F., Borgani, S., Biviano, A., 2002, NewA, 7, 227 
Recchi, S., Matteucci, F., D'Ercole A., Tosi, M., 2002, A&A, 384, 799 
Reichardt, C, Jimenez, R., Heavens, A. F., 2001, MNRAS, 327, 849 
Rix, H.-W. et al. 2004, ApJS, 152, 163 

Robertson, B., Cox, T. J., Hernquist, L., Franx, M., Hopkins, P. F., Martini, P., Springel, 
V., 2006, ApJ, 641, 21 

Sansom, A. E., Hibbard, J. E., Schweizer, Franois, 2000, AJ, 120, 1946 

Sersic, J. L. 1968, Atlas de Galaxies Australes (Cordoba: Obs. Astron., Univ. Nac. Cordoba) 

Shankar, F., Lapi, A., Salucci, P., De Zotti, G., Danese, L., 2006, ApJ, 643, 14 

Sheth, R. K., Jimenez, R., Panter, B., Heavens, A. F. 2006, ApJ, 650, L25 

Somerville, R. S., et al., 2006, ApJ, in press, astro-ph/0612428 
Spergel, D. N., et al, 2007, ApJS, 170, 377 
Strateva, I., et al. 2001, AJ, 122, 1861 
Strauss, M. A., et al., 2002, AJ, 124, 1810 

Tojeiro, R., Heavens A. F., Jimenez R., Panter B., 2007, MNRAS, submitted, arXiv0704.0941 

This preprint was prepared with the AAS IATgX macros v5.2. 

- 25 - 

5 10 15 20 5 10 15 20 

r Band Petrosion Radius (kpc) r Band Petrosian Radius Angle (arcsec) 

Fig. 4. — Left panel: Scaling radii as a function of the r-band Petrosian radii Rp for the SDSS 
sample, overplotted with the means ± 1 a errors. The darkest regions correspond to the 
densest areas, where most of our galaxies lie. Right panel: expected angular size, computed 
from the scaling radii, against actual angular size on the sky for the SDSS galaxies. 


_l I I I I I I I I I I I I I I I I I I I I I I I I I L 

7 8 9 10 11 12 

log(M./M ) 

Fig. 5. — G/S as a function of the stellar mass for the blue (solid pentagons) and red (open 
pentagons) galactic components as computed by Kannapan (2004). The contour encloses the 
area where 92% of the galaxies with stellar masses 9 < log(M*/M ) < 12 fall, after weighting 
each one by 1/V max , according to K04. The typical systematic error on the observational 
points is of 0.05 dex. The solid and open diamonds are the mean values computed in this 
paper for the red and blue galactic components, respectively. In this case, the red and blue 
populations consist of all the galaxies having formed more than the 60% and less than the 
60% of their present-day stellar mass during the first 4 timesteps (i.e. in ~ 11.4 Gyr), 

J I I I I I I I I I I I I I I I I I I I I I I I L 

7 8 9 10 11 12 

log(M./M s ) 

Fig. 6. — G/S as a function of the stellar mass for the blue (solid pentagons) and red (open 
pentagons) galactic components as computed by Kannapan (2004). Symbols are as in Fig. [5J 
with the difference that in this case the red and blue components are the galaxies with the 
E{B -V) < 0.01 and E(B - V) > 0.01, respectively. 


6 8 10 12 

log ( M./M ) 

Fig. 7. — Time evolution of the mean fraction of lost baryons fi os t, as a function of the stellar 
mass for the SDSS catalogue. Various colours correspond to different lookback times. The 
green, cyan, blue, red, magenta and black sqaues with error bars are the fi ost as a function 
of the stellar mass computed at lookback times of 8.5 Gyr, 4.7 Gyr, 1.1 Gyr, 0.5 Gyr, 0.25 
Gyr and Gyr (i.e. at the present day). The green, cyan, blue, red and magenta lines are 
drawn to guide the eye through the points. The solid black line is an analytical fit to the 
present day values (see section 4.2). The anomalous behaviour at 1.1 Gyr is almost certainly 
a result of the difficulty of estimating the star formation rate at this age (see text for more 


log ( M gas /M.) 

Fig. 8. — fi os t vs G/S for all the dwarf galaxies of our sample, i.e. all the galaxies with 
M* < 1O 9 M . The small solid squares are the points for all the individial galaxies, whereas 
the large squares are the mean values, plotted with the error bars. The larger shaded area 
and the smaller dotted area indicate the regions where 100% and 85 % of the dwarf galaxies 
of the sample by Geha et al. (2006) lie, respectively.