A&A 488, 463-479 (2008)
DOI: 10.1051/0004-6361:200809678
R. Maiolino1 - T. Nagao2 - A. Grazian1 - F. Cocchia1 - A. Marconi3 - F. Mannucci4 - A. Cimatti5 - A. Pipino6 - S. Ballero7 - F. Calura7 - C. Chiappini10,11 - A. Fontana1 - G.L. Granato8 - F. Matteucci7 - G. Pastorini3 - L. Pentericci1 - G. Risaliti9 - M. Salvati9 - L. Silva10
1 - INAF - Osservatorio Astronomico di Roma, via di Frascati 33,
00040 Monte Porzio Catone, Italy
2 -
National Astronomical Observatory of Japan, 2-21-1 Osawa,
Mitaka, Tokyo 181-8588, Japan
3 -
Dipartimento di Astronomia, Università di Firenze, Largo E. Fermi 2, 50125 Firenze, Italy
4 -
INAF - Istituto di Radioastronomia, Largo E. Fermi 5, 50125 Firenze, Italy
5 -
Dipartimento di Astronomia, Università di Bologna, via Ranzani 1, 40127 Bologna,
Italy
6 -
Astrophysics, University of Oxford, Keble Road, Oxford OX1 3RH, UK
7 -
Dipartimento di Astronomia, Università di Trieste, via Tiepolo 11, 34131 Trieste, Italy
8 -
INAF - Osservatorio Astronomico di Padova, Vicolo Osservatorio 5, 35122 Padova, Italy
9 -
INAF - Osservatorio Astrofisico di Arcetri, Largo E. Fermi 5, 50125 Firenze, Italy
10 -
INAF - Osservatorio Astronomico di Trieste, via Tiepolo 11, 34131 Trieste, Italy
11 -
Geneva Observatory, Geneva University, 51 chemins des Mailletes,
1290 Sauverny, Switzerland
Received 28 February 2008 / Accepted 24 June 2008
Abstract
We present initial results of an ESO-VLT large programme (AMAZE) aimed at
determining the evolution of the mass-metallicity relation
at z>3 by means of deep near-IR spectroscopy.
Gas metallicities are measured, for an initial sample of nine
star forming galaxies at ,
by means of optical nebular lines
redshifted into the near-IR. Stellar masses are accurately determined
by using Spitzer-IRAC data, which sample the rest-frame near-IR stellar light
in these distant galaxies.
When compared with previous
surveys, the mass-metallicity relation inferred at
shows an evolution much stronger than observed at lower redshifts.
The evolution is prominent
even in massive galaxies, indicating that
is
an epoch of major action in terms of star formation and metal enrichment also
for massive systems.
There are also indications that the
metallicity evolution of low mass galaxies is stronger
relative to high mass systems, an
effect which can be considered
the chemical version of the galaxy downsizing.
The mass-metallicity relation observed at
is difficult
to reconcile with the predictions of some hierarchical evolutionary models.
Such discrepancies suggest that at z>3 galaxies are assembled mostly with
relatively un-evolved sub-units, i.e. small galaxies with low star formation efficiency.
The bulk of the star formation and metallicity evolution probably occurs
once small galaxies are already assembled into bigger systems.
Key words: ISM: abundances - galaxies: abundances - galaxies: evolution - galaxies: high-redshift - galaxies: starburst
The connection between galaxy mass and metallicity has been known
for a long time, starting with the seminal work of Lequeux et al. (1979).
Given the difficulty in obtaining reliable galaxy masses several authors have
resorted in using the (optical) luminosity. In particular, various works
have reported a clear correlation between blue luminosity and metallicity,
in the sense that more luminous galaxies are characterized by higher metallicities
(e.g. Brodie & Huchra 1991; Skillman et al. 1989; Zaritsky et al. 1994; Garnett & Shields 1987). A major step forward
has been recently achieved by Tremonti et al. (2004), who used optical photometric and
spectroscopic data of a sample of 53 000 galaxies from the SDSS to determine
the mass-metallicity relation of local galaxies (
).
Their work clearly showed that the primary physical parameter driving the correlation
with the gas metallicity is the (stellar) mass of galaxies and not their luminosity.
While Tremonti et al. (2004) focused on the gas metallicity,
a similar relation by using the SDSS survey
was found by Gallazzi et al. (2006) for what concerns the stellar metallicity.
Various physical processes may be responsible for the mass-metallicity relation. One possibility is that outflows, generated by starburst winds, eject metal-enriched gas into the IGM preferentially out of low-mass galaxies (due to the shallow gravitational potential well), making their enrichment less effective than in massive systems (e.g. Tremonti et al. 2004; De Lucia et al. 2004; Finlator & Davé 2008). An alternative scenario is that low mass systems are still at an early evolutionary stage and have still to convert most of their gas into stars, hence they are poorly metal-enriched relative massive galaxies (which are instead already evolved). This is the so-called ``galaxy downsizing'' scenario, supported by various observational evidences (e.g. Juneau et al. 2005; Perez-Gonzalez et al. 2007; Franceschini et al. 2006; Asari et al. 2007; Feulner et al. 2005), where massive galaxies formed most of their stars rapidly and at high redshift, while low mass systems are characterized by a slower evolution, which extends to low redshift. Finally Köppen et al. (2007) ascribes the mass-metallicity relation to variations of the IMF high-mass cutoff in different star forming environments.
Table 1: Photometric properties and integration times of the AMAZE sub-sample presented here.
The relative role of these processes in shaping the mass-metallicity relation is debated. It is likely that each of them contributes at least to some extent, since observational evidences have been found for all of them. Each of these factors (outflows/feedback, downsizing, IMF) has profound implications on the evolution of galaxies. Therefore, it is clear that the mass-metallicity relation contains a wealth of information useful to constrain models of galaxy formation and evolution. Indeed, any model of galaxy evolution is now required to match the mass-metallicity relation observed locally (e.g. Cid Fernandes et al. 2007; Köppen et al. 2007; de Rossi et al. 2007; Kobayashi et al. 2007; Tassis et al. 2008; Finlator & Davé 2008; Bouché et al. 2007; Davé & Oppenheimer 2007; Tissera et al. 2005; Bouché et al. 2006; De Lucia et al. 2004; Brooks et al. 2007; Dalcanton 2007). However, different models predict different evolutionary patterns of the mass-metallicity relation as a function of redshift, and observational data are required to test and discriminate among them.
Observational constraints of the mass-metallicity relation
have been obtained up to
thanks to various deep surveys
(Erb et al. 2006; Liang et al. 2006; Savaglio et al. 2005). Additional observational studies
have investigated the evolution of the luminosity-metallicity relation
or, more generally, the metallicity of high-z star forming galaxies
(Kobulnicky et al. 2003; Maier et al. 2005; Kobulnicky & Kewley 2004; Maier et al. 2004; Kobulnicky & Koo 2000; Förster Schreiber et al. 2006; Maier et al. 2006).
Note that all of these studies
refer to the gas metallicity, while limited work has been done on the stellar metallicity
of high redshift sources
(Mehlert et al. 2006; Rix et al. 2004; de Mello et al. 2004; Halliday et al. 2008) due to difficulties in obtaining high S/N spectra
on the stellar continuum.
The general observational result is that the mass-metallicity relation (as well
as the luminosity-metallicity relation) evolves, in the sense that on average
higher redshift galaxies are characterized by lower metallicities (at a given mass).
Whether the relation evolves also in terms of its shape is still matter of debate.
Theoretical models can generally cope with the observed evolution, within both
the theoretical and observational uncertainties (e.g. Brooks et al. 2007; de Rossi et al. 2007; Kobayashi et al. 2007; Finlator & Davé 2008).
At ,
except for a few individual studies, little work has been currently
done for what concerns the mass-metallicity relation. Pettini et al. (2001) have measured
the metallicity for a small sample of Lyman Break Galaxies (LBG) at
,
but
without investigating the mass-metallicity relation. The metallicity evolution has
been investigated in DLA systems (Akerman et al. 2005; Prochaska et al. 2003; Kulkarni et al. 2005). However, a
study of the mass-metallicity relation for high-z absorption systems has not been
properly performed, due to difficulties in measuring the associated
stellar masses. The closest result is the finding of a relation between metallicity
and velocity dispersion (probably related to the mass) inferred by the width of the
absorption systems (Ledoux et al. 2006; Prochaska et al. 2007).
We have undertaken a large observing programme at ESO-VLT aimed at carefully
determining the mass-metallicity relation at z>3 for a sizeable sample of galaxies.
The final goal is to obtain a detailed description of the evolution of the
mass-metallicity evolution through the cosmic epochs, and therefore constrain
galaxy evolutionary scenarios.
In this paper we present preliminary results obtained by such a program, and discuss
the relevant implications for our understanding of the galaxy evolution at high
redshift. Throughout the paper we adopt the following cosmological parameters:
,
and
(Spergel et al. 2003).
AMAZE (Assessing the Mass-Abundance redshift[-Z] Evolution) is an ESO large program
aimed at determining the mass-metallicity relation in the redshift range
3<z<5.
Observations are being
performed with SINFONI (Eisenhauer et al. 2003), the near-IR integral field spectrometer
at VLT, for a total of 180 h, distributed in three semesters. Observations are expected
to be completed in mid-2008. The target sample consists of about 30 Lyman Break Galaxies
(LBGs), most of which at 3<z<3.7, and only a few of them at
4.3<z<5.2. A more detailed discussion on the sample
selection is given in Sect. 2.1.
In this paper we present
preliminary results based on a first set of data for 9 galaxies,
and restricted to the redshift range 3<z<3.7
(Table 1).
The integration times range from 3 to 7.5 h on source.
The goal of the SINFONI observations is to determine
the gas metallicities by means of a combination of strong line diagnostics
based on H
and [OIII]5007 shifted into the K band, as well as
[OII]3727 and [NeIII]3870 shifted into the H band for sources at
3<z<3.7.
At 4.3<z<5.2 we only rely on the [OII]/[NeIII] ratio observed in the
K band
(but sources in this redshift interval will not be discussed here).
Details on the gas metallicity determination are given in
Sect. 5.
The two-dimensional spectroscopic capabilities of SINFONI are obviously exploited also to map the emission lines. However, the two-dimensional analysis of the spectra goes beyond the scope of this paper and will be discussed in a separate paper.
Table 2: Physical properties of the sample inferred from their SED.
Galaxies in our sample are selected among
z>3 galaxies primarily identified through the
Lyman-break technique, mostly from the Steidel et al. (2003) survey and from the deep
spectroscopic surveys in the Chandra Deep Field South (CDFS) (e.g. Vanzella et al. 2006),
but we also included some lensed galaxies (e.g. Frye et al. 2007,2002)
to better explore the low mass end (but none of the lensed galaxies is in
the preliminary sample presented in this paper).
Galaxies were selected only amongst those with highly
reliable spectroscopic redshift (e.g. flagged as ``A'' in Vanzella et al. 2006). We
required that the redshift is such that the emission lines of interest for the metallicity
determination
([OIII], H,
[OII], [NeIII]) are out of strong sky emission lines and out of deep
atmospheric absorption features. Actually, these requirement could not always be
fulfilled for all of the emission lines (also because sometimes the
redshift determined through optical spectra is not accurate, due to winds affecting UV-rest frame
features, or IGM absorption of the Ly
); however, in these cases
any contamination by OH lines or lower S/N due to atmospheric absorption
will be fully taken into account when estimating the metallicity.
The additional requirement is that the source has been observed with at least two of the Spitzer-IRAC bands, which at these redshifts sample the rest-frame near-IR light. IRAC data are required for a reliable determination of the stellar mass (see Sect. 6). Finally, we excluded sources whose optical spectrum shows indications for the presence of an AGN. Moreover, for most of the sources we also required that deep hard X-ray data and mid-IR Spitzer-MIPS data are available, to better exclude the presence of a hidden AGN, as discussed in Sect. 2.2.
In this paper we only present results for an initial sub-sample of nine sources at 3<z<3.7, for which data have been already obtained and reduced. The list of sources, along with some of their photometric properties, is given in Table 1. Table 2 lists some of the physical properties of these sources as inferred from their broad-band spectral energy distribution. A detailed discussion on the extraction of these parameters will be given in Sect. 6.
The presence of an AGN, contributing to the gas ionization, affects the observed emission line ratios. In this case the metallicity diagnostic diagrams calibrated on star forming galaxies are not usable, since the excitation mechanism is totally different. As a consequence, galaxies hosting AGNs must be carefully avoided.
A first step is to exclude galaxies whose optical spectrum (UV rest frame) shows
indications for the presence of an AGN (e.g. NV, CIV, HeII, or broad Ly).
However, the absence of optical-UV AGN-like lines is a required condition,
but not sufficient to
rule out the presence of an AGN. Indeed, even if an AGN is present, the associated
optical-UV emission lines may be undetected either because obscured by dust (either on
small scales, for the BLR, or on larger scales, for the NLR), or because
their Narrow Line Region is not developed (e.g. Martínez-Sansigre et al. 2006; Maiolino et al. 2003).
An additional constraint comes from hard X-ray data.
In many of the fields used by us, deep Chandra observations allow the detection of
obscured (Compton thin) AGNs up to ,
even at Seyfert-like luminosities.
Therefore, an additional requirement was that our sources are not detected in the
hard X-rays (2-8 keV). They should not be detected also in the soft X-rays (0.5-2 keV) at
a level higher than expected by strongly star forming galaxies (actually none of the
galaxies is detected even in the soft band). In the CDFS the deep X-ray data allow us
to exclude the presence of obscured, Compton thin AGNs with 2-10 keV luminosity
higher than about
(i.e. in the Seyfert range).
However, current X-ray surveys are not deep enough to detect Compton thick AGNs, whose
emission is strongly suppressed even in the hard X-rays. This issue was made clear by
recent Spitzer results
(Martínez-Sansigre et al. 2007; Fiore et al. 2007; Alonso-Herrero et al. 2006; Daddi et al. 2007; Polletta et al. 2006). Indeed,
Spitzer observations have revealed the presence of obscured AGNs, through the associated
mid-IR hot dust emission, even in high-z galaxies that are undetected in deep hard X-ray
observations. These are shown to be high-zCompton thick (or nearly Compton thick) AGNs, which remained
elusive to previous optical and X-ray surveys. Deep 24 m Spitzer-MIPS data were
found to
be particularly efficient to identify high-z obscured AGNs, even at relatively low
intrinsic luminosities. As a consequence, we requested that our sources have
deep MIPS data at 24
m.
Fiore et al. (2007) showed that the mid-IR excess relative to the optical emission
(
), is a good tracer of obscured AGNs at high redshift.
As listed in Table 1, most of our sources are undetected at 24
m.
More specifically, all our sources have a ratio
,
which is
significantly lower than expected for Sy2s and QSO2s at
(
),
therefore ruling out the presence of AGNs (both Compton thick and thin)
even at Seyfert-like luminosities.
The near-IR spectroscopic observations were obtained by means of
SINFONI, the integral field spectrometer at VLT. SINFONI was used in its seeing-limited
mode, with the 0.25
pixel scale and with the H+K grism, yielding a spectral resolution
over the spectral range 1.45-2.41
m.
![]() |
Figure 1:
[OIII]5007 line maps ( left) and near-IR spectra
( right)
of the galaxies in the AMAZE sample presented here. Each [OIII]5007 map has a size of
![]() |
Open with DEXTER |
![]() |
Figure 2: Same as Fig. 1 for four additional sources. |
Open with DEXTER |
![]() |
Figure 3: Same as Figs. 1, 2 for one additional source. |
Open with DEXTER |
![]() |
Figure 4: Composite spectrum of the nine sources presented in this paper. |
Open with DEXTER |
Table 3: Line fluxes and metallicities inferred from the near-IR spectra.
Each target was acquired through a blind offset from a nearby bright star.
Each observing block consisted of 10 integrations, 5 min each, obtained by nodding the
position of the source within the
SINFONI field of view (generally by locating the
source in two opposite corners). This observational procedure allows background subtraction
by using frames contiguous in time, but with the source in different locations.
Moreover, the source was never located at the same position in the FOV: a minimal
dithering of 0.5
was required, so that instrumental artifacts can be minimized when
the individual observations are aligned and combined together.
The (K-band) seeing during the observations was generally better than 0.8
.
Each source was observed with a number of observing blocks ranging
from 5 to 9. Some observing block was discarded because the seeing was much worse,
or the background much higher, with respect to the other observing blocks.
The total on-source integration times are listed in Table 1.
Data were reduced by using the ESO-SINFONI pipeline (version 3.6.1). The pipeline
subtracts the sky from the temporally contiguous frames,
flat-fields the images, spectrally calibrates each individual
slice and then reconstructs the cube. Residual sky emission was accounted for by
removing the median of each spectral plane; this is feasible because our source occupy only
a small part of the field of view.
In some cases we performed an additional step in the background subtraction (which
resulted imperfect with the previous method probably because of minor uncertainties in the flat-fielding)
by sampling the
sky in a region outside the source (either annular or another region in the field of view
observed with the same effective integration) and rescaling it to optimally subtract the sky lines
on the spectrum of the source.
Individual cubes where aligned in the spatial
direction by relying on the telescope offsets and then averaging them together by applying
a 2
clipping to remove bad pixels and cosmic rays.
The atmospheric absorption and instrumental response were corrected by dividing the spectrum of the scientific target by the spectrum of a star (spectral type OV-BV or GV) taken close to the source, both in time and in elevation. The intrinsic spectrum of the star was removed by dividing the observed stellar spectrum by the appropriate template given in Pickles (1998), or by the solar spectrum in the case of GV stars (Maiolino et al. 1996).
We extracted the spectra within a fixed aperture of 0.75
in
diameter (corresponding to
6 kpc projected on sources at
),
which in most cases encloses more than 70% of the emission line flux and generally
maximizes the S/N ratio. However, one should keep in mind that there are metallicity gradients
within each galaxy and therefore the aperture choice may introduce biases, especially
for what concerns the comparison with low redshift surveys. This issue
will be discussed more extensively in Sect. 7.2.
An exception is SSA22a-aug96M36, whose line emission extends significantly
beyond the 0.75
aperture; in this case we adopted an aperture of
1.25
.
The resulting spectra, smoothed with a 2 pixel boxcar to improve the signal-to-noise, are
shown in Figs. 1-3. The location of [OII]3727, [NeIII]3870
(when observable), H
,
[OIII]4959 and [OIII]5007 is indicated with vertical dashed lines.
In each spectrum the bottom panels show the sky spectrum. The shaded vertical regions
overlaid on each spectrum highlight spectral regions affected by strong sky emission lines.
In Figs. 1-3 we also show the [OIII]5007 line map of each source.
Figure 4 shows the composite spectrum of all nine sources, obtained by
shifting the spectra to the rest-frame, resampling them to a common wavelength scale,
normalizing them by the flux of H
and averaging them. We excluded spectral
regions strongly affected by atmospheric absorption within individual spectra.
The stellar continuum is detected only in a few cases, and even in these cases the continuum is only seen in the map produced by stacking the cube in the spectral direction in the K or H band.
The emission line fluxes were measured by fitting a single gaussian over a linearly
interpolated, underlying continuum (which may be some weak stellar continuum or, more often,
residual thermal background or residual bias subtraction).
The resulting line fluxes are given
in Table 3. Note that for what H
is concerned we do not
perform any subtraction of a stellar component, since the stellar continuum
is always very week and generally undetected, hence the correction for any putative stellar
H
is negligible. Some authors apply a fixed correction of
2 Å
for the EW of
a putative H
in absorption; in our case such a correction would
generally affect the inferred
metallicities by less than 0.03 dex.
The only method
to determine the gas metallicity in faint distant emission line
galaxies is to use strong
line metallicity diagnostics. Essentially, the ratio between various
strong, optical emission lines is found to depend on the gas metallicity,
either directly and/or through other
dependences (e.g. the metallicity dependence of the ionization parameter, gas density,
hardness of ionizing radiation, etc.).
Various strong line ratios have been calibrated against metallicity, either determined
``directly'' (e.g. through the electron temperature
method) or ``indirectly''
(e.g. through photoionization models).
However, such calibrations have often been
performed in relatively narrow metallicity intervals, not adequate to explore the
wide metallicity range spanned by galaxies through the cosmic epochs, as we shall see.
Another serious problem is that such calibrations are often inconsistent with each other:
the same galaxy is found to have significantly different metallicities if different
strong-line diagnostics are adopted.
This issue has been reviewed in detail by
Kewley & Ellison (2008). Obviously, a wrong intercalibration between
different metallicity diagnostics has dramatic implications for the investigation of the
metallicity evolution. Indeed, at different redshifts people have
observed different emission lines, depending on the adopted band, and therefore an incorrect
intercalibration between the various diagnostics may hamper the capability of
investigating evolutionary effects, or may even introduce artificial trends.
To minimize this issue it is recommended to use the same strong line calibration method for all objects at various redshifts and from different surveys, or to convert the strong line calibrations adopted by different authors to a common calibration scale (e.g. by using the conversion formulas given in Kewley & Ellison 2008). However, the problem that we face in this paper is that no single strong line calibration method exists over the wide metallicity range spanned by galaxies through the cosmic epochs (as we shall see). Some methods nominally span a somewhat wider metallicity range, but they are known to run into serious troubles in some metallicity intervals.
The electron temperature
method (e.g. by exploiting the intensity of
the [OIII]4636 auroral line) provides a good measure of the metallicity
below about 12 + log (O/H) < 8.3 (e.g. Pettini & Pagel 2004; Pilyugin 2001), and can be used to calibrate
the strong line ratios in this range. The reliability of the
method in the low
metallicity range is confirmed by the comparison with the stellar (OB) photospheric metallicity
measurements (Bresolin et al. 2007,2006).
The
method has been extended to higher metallicities by various authors
(Yin et al. 2007; Liang et al. 2007; Garnett et al. 2004; Kennicutt et al. 2003).
However, at high metallicities
the
method tends to saturate and to underestimate significantly the true metallicity, due
to temperature fluctuations and gradients, both within individual HII regions and over
the whole galaxy.
This issue is expected theoretically (Stasinska 2005) and verified observationally
by the comparison with the metallicities determined through recombination lines,
which are insensitive to temperature fluctuations (Bresolin 2007,2006).
Photoionization models are an alternative way of calibrating strong line ratios (e.g. Tremonti et al. 2004; Zaritsky et al. 1994; Kewley & Dopita 2002), especially at high metallicities, where most of these studies apply. However, all photoionization models are subject to significant uncertainties and possible systematic effects. The observed spread in calibration between different models highlights this problem (Kewley & Ellison 2008). The photoionization models presented in Kewley & Dopita (2002) are probably not free from the uncertainties and possible systematic effects discussed above, however they provide results which are intermediate among other photoionization models (Kewley & Ellison 2008), and therefore can be considered fairly representative of this class of calibrations. Moreover, independent ``direct'' determinations of the metallicity (by exploiting the temperature-insensitive method of recombination lines, Bresolin 2007,2006) are in fair agreement with the photoionization models provided by Kewley & Dopita (2002). The latter did not investigate photoionization models at metallicities 12 + log (O/H) < 8.4. Other studies attempt to extend photoionization models to 12 + log (O/H) < 8.3, but fail to reproduce the observed line ratios (e.g. Dopita et al. 2006).
Table 4: Coefficients for different strong-line metallicity diagnostics in Eq. (1).
![]() |
Figure 5:
Relations between strong emission line ratios and gas metallicity. Blue squares are
low metallicity galaxies (from Nagao et al. 2006) for which the metallicity is inferred through
the electron temperature ![]() |
Open with DEXTER |
![]() |
Figure 6:
Example of the diagnostic tools used to determine the metallicity
in the specific case of the composite spectrum. The upper left panel shows the
best solution (blue cross) and the 1![]() ![]() ![]() |
Open with DEXTER |
Since no single method is capable of providing a calibration of the
strong line diagnostics over the wide metallicity range required to sample the evolution
of galaxies through the cosmic epochs (
), in this
paper we have to combine two different methods depending on the metallicity range.
At low metallicities (
)
we use calibrations of
the strong-line diagnostics based on the
method, which is
regarded as reliable at low metallicities and free of problems
related to temperature fluctuations and gradients
(Stasinska 2005; Bresolin et al. 2007,2006). We use the sample
of 259 low metallicities galaxies gathered by Nagao et al. (2006) for
which measurement of the auroral line [OIII]4636 is available.
The metallicity determination based on the
method is
detailed in Nagao et al. (2006). The inferred metallicities for this sample
are in the range
.
The inferred empirical relations
between metallicity and various strong-line diagnostics
are shown in Fig. 5 with blue squares. Note that
the
method is only used to calibrate the strong line diagnostics in
local galaxies (and at low metallicities), but it will not be used to directly
measure metallicities in our high-z objects, since the [OIII]4636 line
is too faint to be detected.
At
,
where the
method is know to fail (as discussed
above), we have to rely on photoionization models. More specifically, we adopt the
calibrations provided by the models in Kewley & Dopita (2002), aware of the caveats discussed above,
and in particular that these models are valid only at
.
We then used
the data of star forming galaxies in SDSS DR4, by adopting the same
constraints on the signal-to-noise discussed in Nagao et al. (2006), and by
also considering only objects with
,
as recommended by Kewley & Dopita (2002). This selection results in a total of
22 482 objects. After determining the gas metallicity for each object
with the Kewley & Dopita (2002)
method, we derived the empirical relations with various strong-line diagnostics
as shown in Fig. 5 (black dots).
The relations obtained by combining both the low metallicity and the high
metallicity samples were fitted with a polynomial curve. To avoid
the fit to be dominated by the regions containing the largest number of
objects (i.e. by the SDSS sample at
), we
divided the relations in metallicity bins (generally spaced by 0.1 dex), we
derived the median and dispersion of the strong-line flux ratio and of the
stellar masses within each bin, and then
fitted the polynomial function to these medians.
In most cases
a second order or a third order polynomial is appropriate to describe the
relation over the full metallicity interval. However, in some cases
a fourth order polynomial is required. The general functional form for describing
the strong-line metallicity calibration is therefore:
In the redshift range 3<z<3.7 investigated in this paper [OII]3727 and (not always)
[NeIII]3870 are observable in the H band, while H
and [OIII]5007 are observable in
the K band. This allows us to use five metallicity diagnostics (not all of them independent of
each other), namely:
,
,
,
,
.
The relationship between these ratios and the gas metallicity,
along with their dispersion, are shown in Figs. 5
and 6.
Each of these diagnostics has advantages and disadvantages. For instance,
is essentially unaffected by dust reddening, but it has a double
metallicity solution for each value of this ratio.
has a monotonic
dependence on metallicity, but it is potentially affected by dust reddening and it is also
affected by a larger dispersion.
is both a monotonic function of metallicity and little affected
by dust reddening, nonetheless
[NeIII] is generally the faintest of all these lines and the most
difficult to detect. However, if these various diagnostics are used simultaneously, then
it is possible to both account for dust reddening and remove the ambiguity of double
solutions. Essentially, if all the diagnostics are used then only some
combinations of metallicity and dust reddening are allowed by the data.
More specifically, we selected the following independent metallicity diagnostics:
[OIII]5007/H,
[OIII]5007/[OII]3727 and (when available) [NeIII]3870/[OII]3727 (note that the ratio
[OIII]5007/H
is very similar to the ``classical'' R23 parameter at the low
metallicities investigated by us, since at low metallicities [OIII]/[OII] is high).
We then determined the best pair of metallicity
and extinction that minimizes the
in the three corresponding diagrams,
both by including the measurement errors and the dispersion of each calibration diagram.
In practice, since [NeIII] is often undetected or not observable,
the metallicity is mostly determined
through the [OIII]5007/H
ratio, while the [OIII]5007/[OII]3727 ratio is used
to discriminate which of the two metallicity solutions for the [OIII]5007/H
ratio
applies and also to provide some constraints on the dust extinction (although the latter has
generally negligible effect on the metallicity determination since [OIII]5007/H
is
insensitive to reddening)
.
As an example, we show the results of this method in
Fig. 6 in the case of the composite spectrum.
The upper left panel shows the
best solution (blue cross) and the 1
confidence level (red curve,
obtained from solutions with
)
in the
-metallicity plane. In the other panels the black solid line
(best fit) and the dashed lines (dispersion) show the empirical relations between
various line ratios and the gas metallicity (Fig. 5);
the green errorbars show the observed
ratios (along the Y-axis) and the best-fit metallicity with uncertainty
(along the X-axis); the blue cross shows
the de-reddened ratios, by adopting the best-fit extinction; the red line shows the projection
of the 1
uncertainty of the fit obtained in the top-left panel.
It can be noted that the
extinction is subject to a large uncertainty, but the metallicity is
relatively well constrained.
The metallicities resulting from the procedure discussed above are listed for all
objects and for the composite spectrum in
Table 3.
One of the main worries when using strong emission line diagnostics in high-z sources is that the empirical calibrations are obtained by using local sources. Since the dependence of the strong lines ratios on metallicity involves also other dependences (e.g. on the ionization parameter, on the shape of the ionizing radiation, on the gas density), the evolution of the average galaxy properties (e.g. SFR, compactness) with redshift may affect the calibration of the metallicity diagnostics. It is very difficult to investigate this issue, since in principle one would need to obtain an empirical calibration at high-z by observing primary metallicity tracers (e.g. [OIII]4636), which are however extremely faint. An alternative way is to construct diagrams which are sensitive to the excitation mechanism, by disentangling the dependence on metallicity, and verify whether the line excitation conditions change with redshift.
Such a test was performed, at lower redshifts, by
investigating the diagram [OIII]/H
versus [NII]/H
for sources where
all of these lines are observable (Erb et al. 2006; Shapley et al. 2005; Liu et al. 2008). It is found that a fraction of sources
at
-2 are offset with respect to the sequence described by local
HII galaxies,
and displaced towards the AGN locus. An interpretation is that at least some of these
sources are affected by some AGN contribution (Liu et al. 2008),
which were not excluded from the sample because elusive in the UV rest-frame spectra.
Other sources may be characterized by truly different physical conditions with respect
to local HII regions, and in particular higher ionization parameter.
However, even in the latter cases both Brinchmann et al. (2008) and Liu et al. (2008) find that the
calibration of the strong line metallicity diagnostics do not deviate by a large amount
with respect to local HII galaxies. In particular, they find deviations by only about 0.1 dex
(or less) in terms of metallicity calibration, depending on the specific diagnostic adopted.
For what concerns the sources at
in AMAZE,
the available emission lines allow us to construct the
so-called BPT diagram (Baldwin et al. 1981), i.e. [OIII]5007/H
versus [OIII]5007/[OII]3727. As discussed in Dopita et al. (2006) this diagram
is strongly degenerate in terms of
metallicity, but it is sensitive to both the ionization parameter
and the hardness of the ionizing source.
Within the observational uncertainties,
the sources in our AMAZE sample do not deviate from the sequence of local
HII galaxies on the
BPT diagram, suggesting that the excitation conditions do not differ significantly
from the local galaxies used to calibrate the strong metallicity diagnostics.
To derive the stellar masses for the LBGs in the AMAZE sample,
we used an approach based on broad-band spectral fitting technique.
Broad-band photometric data for the sources in the CDFS were collected
from the GOODS-MUSIC multiwavelength catalog (Grazian et al. 2006). This
catalog provides photometric data in 14 spectral bands (from UV to
the Spitzer-IRAC bands), and it has been recently updated to include
the Spitzer-MIPS data at 24 m. For the LBGs in
Steidel et al. (2003), optical photometric data (U, G, R, I) were extracted
from the publicly available images (Steidel et al. 2003), while
Spitzer IRAC and MIPS data were obtained from the Spitzer archive;
the photometry extraction was performed following the same methods
described in Grazian et al. (2006).
The SED fitting technique adopted here is the same as in previous
papers (Grazian et al. 2006; Fontana et al. 2006; Grazian et al. 2007), and
similar to those adopted by other groups in the literature
(e.g. Pozzetti et al. 2007; Dickinson et al. 2003; Drory et al. 2004). This technique is based on
comparing the observed multicolor distribution of each object and a
set of templates, computed with standard spectral synthesis models
(see below) and chosen to
broadly encompass the variety of star-formation histories, ages,
metallicities, and extinction of real galaxies. More specifically,
we considered exponentially decaying SFR with e-folding times ranging
from 0.1 to 15 Gyr. We used the Salpeter
IMF (
and
), ranging over a
set of metallicities (from
to
)
and
dust extinction (
0<E(B-V)<1.1, with a Calzetti et al. (2000)
attenuation curve, which
is generally more appropriate for the stellar component).
For each model of this grid, we computed the expected magnitudes in
our filters set and found the best-fitting template with a
standard
normalization. The stellar mass and other best-fit
parameters of the galaxy, like SFR, age, and dust extinction, are
fitted simultaneously to the actual SED of the observed galaxy.
The metallicity of each galaxy is
fixed to value closest to the one determined by us through the
nebular lines (Table 3).
![]() |
Figure 7:
Mass-metallicity relation observed at different redshifts.
The blue, solid and dotted lines indicate the
mass-metallicity relation and its dispersion observed at
![]() ![]() ![]() |
Open with DEXTER |
The stellar mass derived here is subject to uncertainties and biases
related to the synthetic libraries used to carry out the fitting of
the galaxy SEDs. In general, the stellar mass turns out to be
the least sensitive parameter to variations of the input model assumptions, and the
extension of the SEDs to mid-IR wavelengths (near-IR rest-frame)
with IRAC greatly reduces
the formal uncertainties on the derived stellar masses, as shown in
Fontana et al. (2006).
The uncertainties in the stellar mass are derived as follows: we
compute the 90% confidence level on the mass by scanning the levels, fixing the redshift and the metallicity for each galaxy but
allowing the other parameters (SFR, age, dust extinction) to change.
Age and star formation rate are more uncertain
parameters to derive. In some cases we formally obtain best-fit ages below 50 Myr, which
are below the dynamical timescales for the star forming regions in these systems
(Shapley et al. 2001). Moreover, the conversion between UV luminosity and SFR becomes highly
non-linear below this age. As a consequence, we decided to restrict the allowed
ages to >50 Myr. However, this choice may only affect the inferred
SFR, while the determination of the stellar mass is essentially unaffected,
as discussed above.
For what concerns the library of spectral synthesis models we adopt both those provided by Bruzual & Charlot (2003) (hereafter BC03) and those by Maraston (2005) (hereafter M05). The resulting stellar masses are tabulated for both cases in Table 2. The masses inferred by using the M05 models are preferred, since they take into account the contribution by TP-AGB stars, and may differ from the stellar masses obtained with BC03 by even a factor of two, especially in older stellar systems. However, previous works on the mass-metallicity relation at lower redshift adopted the BC03 templates. Therefore, when comparing our results with the previous works at lower redshifts we will adopt for consistency the masses obtained with the BC03 templates. In Table 2 we also list the SFR, age and reddening inferred by the SED fitting (adopting templates by BC03, again for a consistent comparison with previous works).
Table 5: Best fit parameters for analytical form of the mass-metallicity relation in Eq. (2) at different redshifts.
Different studies of the mass-metallicity relation at various redshifts have employed different diagnostic lines and different calibrations. As discussed in Sect. 5 and more extensively in Kewley & Ellison (2008), the mismatch between the different calibration scales may introduce artificial evolutionary effects of the mass-metallicity relation. Therefore, it is important that different strong-line diagnostics used in different surveys are cross-calibrated in a consistent way. The relations obtained in Sect. 5 provide such a common cross-calibration between different strong-line diagnostics on the same metallicity scale. In this section we apply (when required) the correction to the metallicities inferred by past surveys at lower redshift to match our metallicity scale. We also apply corrections to the mass scale to account for the different IMF's adopted by previous works.
As discussed in Sect. 1 the local ()
mass-metallicity relation
was derived by Tremonti et al. (2004) by using SDSS spectra from DR2. Kewley & Ellison (2008) re-determined
the local mass-metallicity relation by using SDSS spectra from DR4 by setting tighter limits on the
redshift range (
0.04<z<0.1) so that the projected SDSS fiber covering factor is
>20% of the total photometric g'-band light, and also to minimize
incompleteness effects at higher redshifts.
The resulting median redshift of their sample is
0.07.
Kewley & Ellison (2008) calibrate the metallicities with the Kewley & Dopita (2002) method, which is the same
adopted by us at
,
hence no additional correction is required to match
our metallicity scale. The only correction to apply
is for the stellar masses, since Tremonti et al. (2004) and Kewley & Ellison (2008)
adopt a different IMF (Kroupa 2001).
We calculate that the masses in Kewley & Ellison (2008) must be multiplied by a factor of 1.17 to comply
with our IMF (note that the IMF's differs not only in terms of shape but also
in terms integration limits).
The thin blue solid lines in Fig. 7 show the Kewley & Ellison (2008) mass-metallicity relation
corrected as discussed above. The blue, dotted lines indicates the 1
dispersion of the same
relation.
At 0.4<z<1 we use the results by Savaglio et al. (2005). For consistency with our calibration scale
we re-determine the metallicities for each object in their sample by applying the same procedure
described
in Sect. 5.2 to the line fluxes tabulated by them. We exclude from their sample objects
without K-band data, since in these cases the stellar mass uncertainties are too large.
We also have corrected the stellar masses in Savaglio et al. (2005) by a factor of 1.4 to comply
with the IMF adopted by us.
The resulting mass-metallicity relation at
is shown with red diamonds and errorbars in
Fig. 7a.
At
we use the results by Erb et al. (2006), who infer the metallicity of LBG's at this
redshift through the [NII]/H
ratio measured in stacked spectra. Also in this case
we re-determine the metallicity in each mass bin by using the [NII]/H
metallicity
calibration obtained in Sect. 5, to be consistent with the calibrations adopted by us.
The stellar masses in Erb et al. (2006) have to be corrected by a factor of 1.4
to comply
with the IMF adopted by us.
The resulting mass metallicity relation at
is shown with red diamonds and errorbars in
Fig. 7b.
Finally, Fig. 7c shows the mass-metallicity relation inferred from the initial sample of
nine AMAZE sources
at .
The red diamonds with solid errorbars are individual objects.
The green square with dashed errorbars is
the composite spectrum. In this plot, for consistency with the other
works at lower redshifts, we use the stellar masses inferred with the BC03 templates.
Previous versions of our mass-metallicity relation
,
presented in our
previous preliminary works (Maiolino et al. 2007a,b),
were slightly different because of lower S/N spectra and
also because we used different calibrations (both for metallicity and stellar masses).
For a more straightforward comparison of the mass-metallicity relation at different redshifts,
it is useful to describe these relations by fitting them with the same functional form.
In order to minimize the number of free parameters we find statistically satisfactory
an approach similar
to Savaglio et al. (2005): the quadratic function fitting the local mass-metallicity relation
is shifted in mass and in metallicity
to provide the best fit of the mass-metallicity relation at various redshifts.
More specifically we adopt the following description of the mass-metallicity relation:
Since galaxies, and especially disk galaxies, are often characterized by metallicity gradients (the metallicity decreasing towards the outer regions), a possible caveat when comparing metallicities at different redshifts is the different aperture projected on the source. In particular, at high redshift spectroscopic observations are likely to sample most of the galaxy, while at low redshift and in local galaxies the spectroscopic aperture samples mostly the central higher metallicity region. This effect may mimic a metallicity evolution.
When comparing surveys at high redshifts this should be a minor issue, since the projected
apertures on the sources are not very different.
At ,
the adopted aperture of 0.75-1.3
in Savaglio et al. (2005)
corresponds to about 5.6-9.3 kpc; at
Erb et al. (2006) adopt an aperture of 0.76
corresponding to about 7.2 kpc. Our aperture of 0.75
at
corresponds to about 6 kpc.
However, aperture effects may be more serious for the local sample.
At the median redshift of 0.07 the SDSS fiber size (3
)
has a median size of 4 kpc,
and a median covering factor
34% relative to the total g'-band light, in contrast
with a covering factor of
70% at
.
The aperture effect is stronger for local
high mass galaxies, which are generally bigger and for which the covering factor reach values
as low as 20%.
The problem of a differential aperture effect as a function of galaxy mass
may also affect the shape of the local mass-metallicity relation, making the observed relation
steeper than it actually is (Kewley & Ellison 2008). However, the absolute magnitude of this effect
in the local sample is estimated to be at most
0.1-0.15 dex (Kewley & Ellison 2008), which is
significantly lower than the metallicity evolution observed at high redshift, at least at
.
Additional issues related to aperture effects will be discussed in the context of the comparison with models in Sect. 7.5.
When comparing the mass-metallicity relation at
inferred from LBG's with that inferred from local (or lower redshift) samples of star forming
galaxies, one must be aware
that we are comparing different classes of objects, which are not necessarily
linked from an evolutionary point of view. As a consequence, the evolution of the mass-metallicity
relation inferred in this paper should be regarded as the evolution of the
mass-metallicity relation of
galaxies representative of (or contributing significantly to) the density of star formation at
each epoch, and not the evolutionary pattern of individual galaxies. This issue will be
further discussed in the next sections.
In this section we mostly investigate whether
galaxies in our sample are representative of star forming galaxies at
.
Stellar masses of LBGs at
were measured
by Shapley et al. (2001)
, but for a subsample of galaxies
about one magnitude brighter (in R-band) than the parent sample of Steidel et al. (2003).
Shapley et al. (2001) obtain a median stellar mass of
.
The galaxies in our sample (Table 2 and Sect. 6) have a median stellar mass of
,
which is close to the value obtained by Shapley et al. (2001)
for their large LBG sample.
In any case, the distribution of stellar masses is not a concern, since the
mass is one of the two variables
that we are mapping on the mass-metallicity relation: even if we had a bias in terms of stellar mass,
this would simply imply that we preferentially populate the diagram in a certain mass range, making
the estimation of the mass-metallicity relation more uncertain in other mass ranges (because
under-populated), but not biased.
Biases in terms of star formation rate are of a greater concern. Our selection requirement that sources
must have a highly reliable spectroscopic redshift may bias our sample towards sources with strong UV continuum or strong Ly,
hence higher than average SFR. Table 2 shows the SFR
inferred from the rest-frame UV continuum of the sources in our sample (see Sect. 6 for
details), from which we infer a median SFR of
.
This is similar
to the median SFR (
)
of LBGs at
obtained by
Shapley et al. (2001) (who adopted a similar approach as ours to estimate the SFR from the UV continuum).
However, since the latter work is biased towards slightly brighter optical
magnitudes, the median SFR of the LBGs at
in
the parent sample of Steidel et al. (2003) is probably somewhat lower.
Since the
median stellar masses are similar, then the specific star formation rate (SSFR) in our sample
may be somewhat higher (up to a factor of 2) than in the LBG sample of Steidel et al. (2003).
Ellison et al. (2008) investigated the effect of the SSFR on the metallicity in
local galaxies. They
found that, for a variation of the SSFR by a factor of two,
the metallicity varies by less than
0.1 dex
in low mass galaxies (
), while no metallicity variations are found
in massive galaxies (
). As a consequence, a possible bias of
our sample in terms of SSFR relative to the LBG sample of Steidel et al. (2003) should not affect
significantly the inferred mass-metallicity relation.
Another possible source of bias is that LBGs are selected through their
UV rest-frame colors,
hence they miss any population of heavily reddened star forming galaxies. Reddy et al. (2007)
estimate that, due to this selection effect, LBG's represent 47% of the population
of star forming galaxies at
with
R<25.5(see also Hopkins & Beacom 2006). Dust reddened galaxies
are naively expected to be more metal rich
(since metallicity and dust content correlate, Hunt et al. 2005). On the contrary,
Rupke et al. (2008) have shown that, at least at low-z,
dusty galaxies (IR-selected) are characterized by gas metallicities lower than optical- and
UV-selected galaxies (probably due to metal poor gas infalling from the outskirts in
merging/interacting systems). Recently Caputi et al. (2008) have found a similar effect
in dusty galaxies at intermediate redshift (i.e. metallicities lower than in optically
selected galaxies). High-z dusty objects may behave
differently. However, if the same phenomenon is also present at high-z, then
LBG's would provide an upper limit to the metallicity of galaxies at
.
However,
we do not speculate further on the properties of star forming galaxies that are not sampled by LBGs,
and we simply emphasize that the results presented in this paper
only apply to about half of the star forming galaxies at
,
i.e. those
UV-selected.
![]() |
Figure 8: Comparison of the mass-metallicity relation observed at different redshifts, as parametrized by the analytical function Eq. (2) and coefficients in Table 5. |
Open with DEXTER |
![]() |
Figure 9:
Average metallicity of star forming galaxies as a function of the cosmic age
of the Universe, relative to local galaxies,
for three different families of galaxies with different stellar masses
(
![]() |
Open with DEXTER |
The evolution of the mass-metallicity relation is summarized in Fig. 8, where we
plot the best fits resulting from Eq. (2) both at
from the AMAZE
survey, and at lower redshifts from previous surveys.
Figures 7, 8 highlights a clear evolution of the mass-metallicity relation of star forming galaxies through the cosmic epochs. As already discussed in Sect. 7.3, this evolution should not be seen as the evolutionary sequence of individual objects, since at each redshift the various surveys are sampling different classes of star forming galaxies, which are not necessarily each other progenitors. The trend observed in Figs. 7, 8 should be regarded as the evolution of the mass-metallicity relation of galaxies dominating (or contributing significantly) the star formation density at each epoch.
At
the metallicity at
is lower by a factor of about 2.5 with respect to local
galaxies. Even if highly significant, such metallicity
decrease is modest if one considers that from z=0 to
z=2.2 the elapsed time is
11 Gyr, i.e. about 75% of the age
of the universe. From
to
the average metallicity of galaxies decreases by
another factor of about 2.5. However, the latter evolution is much stronger, and faster.
Indeed, such a metallicity variation occurs on a much shorter time scale, only
1 Gyr.
This effect is shown in more clearly Fig. 9, where the the average metallicity
of star forming galaxies is plotted
as a function of the age of the universe, for different classes of galaxies with
different stellar masses, by exploiting the analytical function in Eq. (2)
(note that, as discussed above, this figure does not provide the evolution of individual galaxies).
We further note that the evolution is strong even in massive galaxies.
Clearly,
is an epoch of major action
for the evolution of galaxies, both in terms of star formation and chemical enrichment,
even for massive systems.
We note that a similar strong evolution of the metallicities at
z>3 was obtained by
Mehlert et al. (2006) by investigating the stellar metallicities of a few galaxies with
bright UV continuum. Their result was however affected by uncertainties on the absolute
calibration of the stellar metallicity tracers.
It is also interesting to note that the metallicities obtained by us at
are in fair agreement with those expected at the same redshift by Panter et al. (2008), who
inferred the metallicity evolution of galaxies by
modelling its ``fossil'' spectral signatures in local galaxies.
The additional interesting result is the indication of a differential, mass-dependent
evolution of the metallicity. In particular, the metallicity evolution in low mass
systems appears stronger than in massive galaxies (Figs. 7-9).
This finding requires more
statistics to be confirmed at
(to come with the completion of the AMAZE project).
However, the evolution of the slope of the mass-metallicity relation (relative to the local slope)
is significant even at
(as already noted by Kobulnicky et al. 2003; Savaglio et al. 2005)
and at
.
A detailed investigation
of differential selection effects as a function of stellar mass is required to rule out
that observational biases are not affecting the slope of the mass-metallicity relation at each epoch.
However, if confirmed, such mass-dependent evolution of the metallicity
can be regarded as the ``chemical'' version of the galaxy downsizing:
high mass galaxies reach high metallicities at high redshift, on short timescales, while
low mass systems enrich their ISM over a prolonged period of time, extending to the current epoch.
![]() |
Figure 10:
Comparison between models/simulations predictions for the mass-metallicity relation at
![]() |
Open with DEXTER |
As mentioned in the introduction there is an intense theoretical activity aimed at interpreting the nature and origin of the mass-metallicity relation, and also at providing predictions on the expected mass-metallicity relation at high redshift. Comparing the models predictions with the observational results in a consistent way is not simple. Indeed, theoretical models predict a variety of galaxy populations, spanning a wide range of properties (e.g. in terms of SFR), while observations are limited to samples matching the survey selection criteria. Moreover, generally theoretical works provide the metallicity integrated over large apertures, including most of the galaxy, while observational metallicity measurements are generally obtained within a smaller aperture. In the future a collaborative effort with various theoretical groups is planned to match the outcome of models and simulations to the observational selection effects. However, a preliminary comparison with the already published theoretical results is instructive to infer some initial constraints on galaxy evolutionary models.
Savaglio et al. (2005) interpreted the evolution of the mass-metallicity relation from to z= 0 through a closed-box model with an exponentially decaying
,
where the
-folding time
decreases as a function of galaxy mass. However the inclusion
of our data at
makes this model not suitable. It is difficult
to simultaneously fit the observed mass-metallicity relations at
z=0.1, z=0.7, z=2.2 and z=3.5with a simple closed-box model, unless more complex scenarios of the SF history are envisaged.
Moreover, an exponentially decaying SFR, with the e-folding times provided by
Savaglio et al. (2005) makes the SFR extrapolated to local massive systems well below
1
,
i.e. these should be local quiescent galaxies (probably massive
elliptical), which cannot
be representative of the local star forming galaxies used to derive the local mass-metallicity
relation in Tremonti et al. (2004) and Kewley & Ellison (2008). Finally, it is unlikely that the closed-box
scenario applies to LBGs, which are characterized by strong, unbound winds
(e.g. Pettini et al. 2002).
Within the framework of the hierarchical models of galaxy evolution, de Rossi et al. (2007) performed numerical hydrodynamical simulations enabling them to provide detailed predictions on the evolution of the mass-metallicity relation at various epochs. In the top-left panel of Fig. 10 the dashed green line shows the mass-metallicity relation predicted at z=3by the simulations of de Rossi et al. (2007) (therein Table 3), while the shaded areas give the dispersion inferred from the same simulations (note that de Rossi et al. 2007, use the same IMF adopted by us, therefore no further correction is required). Since we do not have the observational data at exactly the same epoch of the simulations (z=3), we interpolate the observed mass-metallicity to the same epoch of the simulations by using Eq. (2) and Table 5. The observed mass-metallicity relation at z=3is shown in Fig. 10 with a thin, black solid line. We also show the mass-metallicity relation inferred by adopting the stellar masses measured with the M05 templates (thick solid line). Figure 10 shows a significant discrepancy between simulations and observations. The discrepancy was also present at z=2 (although at a lower level), as noted by de Rossi et al. (2007). They suggest that the inconsistency between simulations and observations is due to the lack of significant SN feedback in their simulations, which would remove metal enriched gas and lower the global gas metallicity.
However, the discrepancy with observations is present also for simulations that include the effect
of SN feedback. This is the case of the hierarchical, three-dimensional chemodynamical simulations
presented in Kobayashi et al. (2007), which include the feedback from SNII and hypernovae.
They provide predictions of the mass-metallicity relation as a function of redshift.
The green points in the top-right panel of Fig. 10 show the results
of such simulations at z=3 (also Kobayashi et al. 2007, use
the same IMF adopted by us, therefore no further correction is required).
Even in this case there is a discrepancy between simulations and observations.
The discrepancy is much reduced with respect to the de Rossi et al. (2007) simulations, but still
significant at low masses (
). At high
masses (
)
simulations and observations are nearly consistent. However,
one should take into account that Kobayashi et al. (2007) extract the metallicities in the simulated
galaxies within a radius of
,
i.e. by including metal poor external regions,
while observations provide metallicities within a radius of
.
Aperture
effects are probably more important in large, massive galaxies. If the extraction
radius of the simulations is matched to the observations then the discrepancy probably increases
strongly also in the high mass region.
The main problem of the latter simulations seems to be that the bulk of the chemical enrichment
occurs in small galaxies, yielding a steep metallicity evolution at masses
below <
on
the mass-metallicity plane. Then such evolved sub-units merge to form massive systems,
with little additional star formation (close-to-dry merging),
implying little additional
enrichment (i.e. flatter evolution towards high masses).
The discrepancy with
the observations
seems to imply that the evolution of galaxies at high redshift occurs through
the assembly of little evolved small galaxies, so that high-z
objects with
can exist with relatively low metallicities. In these systems most of the star formation
and chemical evolution occurs once they are already assembled into bigger systems.
The latter scenario is well described by the simulations presented in
Governato et al. (2007), which model the evolution of disks within a
hierarchical framework.
In these simulations a strong feedback,
due to SNe and to gas heating by the UV radiation,
prevents small galaxies to evolve significantly before merging into a bigger galaxy,
while the bulk of the chemical evolution and star formation occurs in the gas which has already
settled into the proto-disk.
Brooks et al. (2007) inferred the evolution of the mass-metallicity relation from these
simulations, whose prediction at z=3 is shown in the bottom-left panel of
Fig. 10 (Brooks, priv. comm.). The agreement with the observations
is good, although at high masses the simulations
tend to under-reproduce the observed metallicity. However the metallicities provided by
Brooks et al. (2007) are obtained without any constraints on the aperture (i.e. all cold gas);
if the information from the simulated galaxies is
extracted within our observational aperture (
)
then metallicities are
expected to increase (especially in large massive galaxies) and better reproduce the observations.
Yet, a potential problem of the Governato et al. (2007) and Brooks et al. (2007) models is that
their forming disks are characterized by modest star formation rates,
never exceeding
,
while LBGs (except for a minority of them) are
characterized by significantly larger SFR, suggesting that they are in the process of rapidly
forming spheroids.
Detailed predictions on the mass-metallicity relation were also obtained by Finlator & Davé (2008)
who used three-dimensional hierarchical simulations along with detailed outflows models.
They show that the evolution of the mass-metallicity relation out to
can
be well reproduced if a ``momentum-driven wind'' model is incorporated. The predictions
of their model at
reproduce reasonably well also the metallicity in massive galaxies
(
)
observed by us.
However, their model predicts an
up-turn of the mass-metallicity slope at
which is not observed by us.
Yet, the slope of the mass-metallicity relation is still poorly determined in our
data, due to the shortage of low-mass galaxies; we should wait for the completion of the
AMAZE program before claiming any significant inconsistency with the model in terms of slope
of the relation.
In the bottom-right diagram of Fig. 10
we also show the mass-metallicity relation expected by the
double-infall models for the formation of galactic disks and dwarfs presented in Chiappini et al. (2001)
and Cescutti et al. (2007). The green points show the mass-metallicity relation of spirals
according to these models, by tracing back their evolution until
z=3.5.
The figure shows a significant discrepancy
between the model and the observations.
Such a discrepancy is not surprising, since LBG's at
z>2 are probably
spheroids in the process of rapidly forming stars, and not spirals. More specifically,
in the double-infall model for disks the star formation
rate never exceeds a few times
,
while the median SFR of LBGs
is
.
Summarizing, there are currently no models or simulations
that can satisfactory explain the mass-metallicity relation observed at
.
The closest match is probably with the simulations of Governato et al. (2007) and Brooks et al. (2007),
although even in these cases there are some discrepancies in terms of SFR.
The location of LBG's at
on the mass-metallicity plane, along with
comparison with these models, suggest that
galaxies have been assembled through low mass systems
whose star formation efficiency was suppressed, hence which were little evolved.
The bulk of the star formation and of the chemical enrichment occurred
once small galaxies were already assembled
into bigger systems. In other words, most of the merging occurred before
most of the star formation.
![]() |
Figure 11:
Evolutionary tracks of individual galaxies on the
mass-metallicity diagram according to some
models for the formation of spheroids presented in Granato et al. (2004) (green solid lines)
and in Pipino et al. (2006) (violet dashed lines),
compared with the mass-metallicity relation observed
at ![]() |
Open with DEXTER |
We conclude this section by comparing in Fig. 11
the location of galaxies on the mass-metallicity diagram with the evolutionary
tracks (as a function of time) expected for individual spheroidal galaxies,
according to the models
in Granato et al. (2004) and Pipino et al. (2006).
These models
prescribe a nearly monolithic formation of elliptical
galaxies, where pristine gas collapses from the halo. In these models star formation is gradually
quenched as the galaxy evolves due to the feedback introduced by star formation and/or AGN activity.
The mass-metallicity tracks of these models are shown in Fig. 11
for different final stellar masses:
green solid lines and violet dashed line are for the Granato et al. (2004) model
and for the Pipino et al. (2006) model, respectively. The
observational data from the Amaze project are shown with diamonds and errorbars.
Red points are for masses inferred by using the
BC03 templates, orange points are for masses inferred by using the
M05 templates.
The models can easily embrace the observed data points.
This comparison is not aimed at explaining the mass-metallicity
relation at
,
since models of individual galaxies at different masses
must be convolved with the evolution of cosmic structures to
obtain a prediction of the mass-metallicity relation at any epoch.
Nonetheless,
Fig. 11 shows that the combination of mass and metallicities observed
in individual star forming galaxies at
does not lie in a region difficult to populate
by models of individual galaxies;
on the contrary, individual observations can be easily explained in terms
of rapidly evolving, massive systems, through these simple models.
However, convolving these models
with the hierarchical growth of dark matter structures is required to verify whether they
can really explain the mass-metallicity relation at
.
We have presented initial results of the AMAZE project, an ESO large programme aimed at determining the mass-metallicity relation of star forming galaxies at z>3. Near-IR spectra are being obtained with SINFONI, the VLT near-IR integral field spectrometer, for a sample of Lyman Break Galaxies at 3<z<5.
Gas metallicities are inferred by using a combination of diagnostics involving nebular
lines observable in the H and K bands. To have a metallicity scale consistent with the results
obtained by previous surveys at lower redshifts, we derived new accurate calibrations
of various strong-line metallicity diagnostics spanning the wide range
.
AGNs (which would affect and make unusable the metallicity diagnostics) were carefully removed through a multiwavelength approach using X-ray, optical and mid-IR data. In particular, the use of mid-IR (Spitzer-MIPS) data allows us to discard even heavily obscured, Compton thick AGNs.
Stellar masses are inferred by fitting multi-band photometric data with galaxy templates. Within this context crucial is the use of Spitzer-IRAC data, which sample the rest-frame near-IR stellar light at 3<z<5.
In this paper we have presented results from an initial sample of 9 LBGs at 3.1<z<3.7.
Emission lines required to constrain the gas metallicity are detected in all sources.
From the mass-metallicity relation at
inferred from this initial sample we obtain
the following results:
The finding that galaxies at z>3 are mostly assembled with un-evolved sub-units is not necessarily in contrast with models of ``dry-merging'', i.e. models where galaxy assembly occurs through systems that are significantly evolved and with little residual gas. Indeed, dry merging may be the main mode of galaxy evolution at lower redshifts (z<3) without being in conflict with our findings at z>3.
Acknowledgements
We thank A. Modigliani for his assistance in using the SINFONI pipeline. We are grateful to C. Kobayashi, S. Savaglio and A. Brooks for providing us the electronic version of their simulations and data. We thank M. E. De Rossi for useful comments. Part of this work was supported by the Italian Institute for Astrophysics (INAF) and by the Italian Space Agency (ASI) through contract ASI-INAF I/016/07/0.