A&A 480, 233-246 (2008)
DOI: 10.1051/0004-6361:20078847
J. I. González Hernández1,2 - P. Bonifacio1,2,3 - H.-G. Ludwig1,2 - E. Caffau2 - M. Spite2 - F. Spite2 - R. Cayrel2 - P. Molaro2,3 - V. Hill2 - P. François2 - B. Plez5 - T. C. Beers4 - T. Sivarani4 - J. Andersen6,7 - B. Barbuy8 - E. Depagne9 - B. Nordström 6 - F. Primas10
1 -
CIFIST Marie Curie Excellence Team
2 -
GEPI, Observatoire de Paris, CNRS, Université Paris Diderot, Place
Jules Janssen, 92190
Meudon, France
3 -
Istituto Nazionale di Astrofisica - Osservatorio Astronomico di
Trieste, via Tiepolo 11, 34143 Trieste, Italy
4 -
Department of Physics and Astronomy, CSCE: Center for the Study of
Cosmic Evolution, and JINA: Joint Institute for Nuclear Astrophysics,
Michigan State University, E. Lansing, MI 48824, USA
5 -
GRAAL, Université de Montpellier II, 34095
Montpellier
Cedex 05, France
6 -
The Niels Bohr Institute, Astronomy, Juliane Maries Vej 30,
2100 Copenhagen, Denmark
7 -
Nordic Optical Telescope, Apartado 474, 38700 Santa Cruz de
La Palma, Spain
8 -
Universidade de Sao Paulo, Departamento de Astronomia,
Rua do Matao 1226, 05508-900 Sao Paulo, Brazil
9 -
Las Cumbres Observatory, Goleta, CA 93117, USA
10 -
European Southern Observatory (ESO),
Karl-Schwarschild-Str. 2, 85749 Garching b. München,
Germany
Received 12 October 2007 / Accepted 6 December 2007
Abstract
Context. Unevolved metal-poor stars constitute a fossil record of the early Galaxy, and can provide invaluable information on the properties of the first generations of stars. Binary systems also provide direct information on the stellar masses of their member stars.
Aims. The purpose of this investigation is a detailed abundance study of the double-lined spectroscopic binary CS 22876-032, which comprises the two most metal-poor dwarfs known.
Methods. We used high-resolution, high-S/N ratio spectra from the UVES spectrograph at the ESO VLT telescope. Long-term radial-velocity measurements and broad-band photometry allowed us to determine improved orbital elements and stellar parameters for both components. We used OSMARCS 1D models and the TURBOSPECTRUM spectral synthesis code to determine the abundances of Li, O, Na, Mg, Al, Si, Ca, Sc, Ti, Cr, Mn, Fe, Co and Ni. We also used the CO5BOLD model atmosphere code to compute the 3D abundance corrections, notably for Li and O.
Results. We find a metallicity of
for both stars, using 1D models with 3D corrections of
-0.1 dex from averaged 3D models. We determine the oxygen abundance from the near-UV OH bands; the 3D corrections are large, -1 and -1.5 dex for the secondary and primary respectively, and yield [O/Fe
,
close to the high-quality results obtained from the [OI] 630 nm line in metal-poor giants. Other [
/Fe] ratios are consistent with those measured in other dwarfs and giants with similar [Fe/H], although Ca and Si are somewhat low (
). Other element ratios follow those of other halo stars. The Li abundance of the primary star is consistent with the Spite plateau, but the secondary shows a lower abundance; 3D corrections are small.
Conclusions. The Li abundance in the primary star supports the extension of the Spite Plateau value at the lowest metallicities, without any decrease. The low abundance in the secondary star could be explained by endogenic Li depletion, due to its cooler temperature. If this is not the case, another, yet unknown mechanism may be causing increased scatter in A(Li) at the lowest metallicities.
Key words: nuclear reactions, nucleosynthesis, abundances - Galaxy: halo - Galaxy: abundances - cosmology: observations - stars: Population II
Extremely metal-poor (EMP) stars formed with the chemical composition of the gas in the early Galaxy, and constitute a unique source of information on the first generations of stars. Among EMP stars, a special place is held by the dwarfs, which are not subject to the mixing episodes experienced by giants, thus enhancing their value as cosmological probes.
In fact, among these stars the Li abundance appears to be constant whatever the stellar temperature or metallicity (Spite & Spite 1982a,b), the Spite plateau. The simplest interpretation of the plateau is that it represents the primordial Li abundance, i.e., it reflects the amount of Li formed in the first minutes of the existence of the Universe. If so, Li can be used as a ``baryometer'', a tool to measure the baryonic density of the Universe, since this is the only cosmological parameter upon which the primordial Li abundance depends.
The independent determination of the baryonic density from the fluctuations of the cosmic microwave background (CMB) by the WMAP satellite (Spergel et al. 2003,2007) and other CMB experiments measuring fluctuations on smaller angular scales, such as the VSA (Rebolo et al. 2004; Grainge et al. 2003), ACBAR (Kuo et al. 2004) and CBI (Pearson et al. 2003) experiments, implies a primordial Li abundance which is at least a factor of 3-4 larger than that observed on the Spite plateau, creating a conflict with the traditional interpretation of the plateau.
In Paper VII in this series (Bonifacio et al. 2007) we investigated the Spite plateau at the lowest metallicities (down to [Fe/H]=-3.3) and found marginal evidence that at these low metallicities there could be an increased scatter or even a sharp drop in the Li abundance. It is therefore of great interest to explore the Li abundance in stars of even lower metallicity.
The star CS 22876-032 was identified in the first paper reporting results of
the HK objective-prism survey by Beers et al. (1985), who noted that it had the
weakest Ca II K line in the low-metallicity sample, suggesting that it could be
as metal-deficient as the record holder at that time, the giant CD -38245 (Bessell & Norris 1984). CS 22876-032 had already been observed in the
objective-prism survey of Slettebak & Brundage (1971), who classified it as an
A-type peculiar star and noted its weak and diffuse Balmer
lines. Having assigned to this star a much earlier spectral type, they
did not conclude that the weakness of the Ca II K line was indeed due
to an extremely low metallicity.
At the conference ``Chemical and Dynamical Evolution of Galaxies'' in 1989
(Bonifacio et al. 1990), P. Molaro announced that high-resolution spectra from the CASPEC
spectrograph at the ESO 3.6 m telescope indicated [Fe/H
for CS 22876-032. However, just afterwards Nissen (1989) discovered, from simlar
resolution spectra, that the star is a double-lined spectroscopic binary. The
spectra acquired by Molaro were obtained at a single-lined phase, and the
abundance analysis of CS 22876-032 by Molaro & Castelli (1990) assumed that it was a
single star. Thus, veiling was neglected, the adopted temperature was too low,
and the measured [Fe/H] was therefore a lower limit to the metallicity of the
system.
Although CS 22876-032 is relatively bright (V=12.84) for an EMP star,
it took another ten years before a sufficient number of high-resolution
spectra had been accumulated to allow determination of the orbital parameters
of this system, and to perform a consistent chemical analysis. Norris et al. (2000)
found the orbital period to be
424.7 days and the metallicity of the system [Fe/H]=-3.71. In spite of the
upward revision of the metallicity, partly due to the different solar
Fe abundance assumed (
instead
of 7.63 in Molaro & Castelli 1990), the two stars in CS 22876-032 remain
the most metal-poor dwarfs known.
Thus, the CS 22876-032 system constitutes a unique fossil, recording the chemical composition of the early Galaxy. Moreover, it allows a measurement of the Li abundance which probes the Spite plateau at a lower metallicity than any other known dwarfs. Note that, despite the extremely low iron abundance ([Fe/H]=-5.4), the star HE 1327-2326 (Frebel et al. 2005) has very high C, N and O abundances, so its global metallicity, Z, is considerably higher than that of CS 22876-032. It has also been shown recently that this star is most likely a slightly evolved subgiant, not a dwarf.
In this paper we use high-resolution, high-S/N ratio spectra from the ESO Kueyen 8.2 m telescope and the UVES spectrograph to improve the orbital solution and perform a complete chemical analysis of the two stars that comprise CS 22876-032. With respect to the Norris et al. (2000) analysis, our superior S/N ratio and larger spectral coverage permit measurement of abundances for many more elements, and, most importantly, for both components; the Norris et al. (2000) analysis of the secondary star was limited to Fe.
Spectroscopic observations of the CS 22876-032 were carried out with the
UV-Visual Echelle Spectrograph (UVES, Dekker et al. 2000) at the
European Southern Observatory (ESO), Observatorio Cerro
Paranal, using the 8.2 m VLT-Kuyen telescope on 2000 July 19, 20,
August 3, 11, and October 17, 20, and 2001 November 7, 8 and 9,
covering the spectral region from 300.0 nm to 1040.0 nm. Most
of the observations were made with a projected slit width of 1
at a resolving power
.
The
spectra were reduced in a standard manner using the UVES reduction
package within the MIDAS environment. The signal-to-noise
ratio per pixel varies from 25 at 312.0 nm, 50 at 330.0 nm up to 150 or higher above 410.0 nm.
We derived radial velocities from the UVES spectra by fitting a Gaussian
to several unblended spectral lines within the IRAF context. Table 1 shows the radial
velocities and 1-
errors estimated from the dispersion of
the measurements of different stellar lines. We also list other velocity data
given by Norris et al. (2000) and references therein, or which we have measured from
previously unpublished CASPEC or EMMI spectra of this system.
Here after, we denote the more massive and luminous primary star as Star A, the secondary as Star B.
Our new radial-velocity measurements of CS 22876-032 extend the time coverage of its orbit considerably and permit improvement of the orbital elements relative to those published by Norris et al. (2000). The computed orbital parameters are given in Table 2; Fig. 1 compares the observed radial velocities of both stars with the curves predicted from these orbital elements.
Note especially the improved mass ratio,
,
which provides stringent constraints on the stellar parameters for the
two components of the binary as discussed below. Note also that the
orbital eccentricity is the lowest found among halo spectroscopic
binaries with periods in the range 100-2000 days (see
Latham et al. 2002; Goldberg et al. 2002). While this might be a hint that tidal
interaction has been strong in this system, perhaps in the
pre-main-sequence phase, the separation of the stars has been so large
throughout their main-sequence life that this is unlikely to be the
cause of the apparent Li depletion we find in star B (see Sect. 6.1).
Table 2: Orbital elements of CS 22876-032.
![]() |
Figure 1: Upper panel: radial velocities of CS 22876-032 (filled circles: A; open circles: B). Open diamonds: data from Norris et al. (2000). The curves show the orbital solution (P = 425 d, e = 0.14) for Star A (solid) and B (dashed). Dot-dashed horizontal line: centre-of-mass velocity of the system. Lower panel: residuals from the fit. |
Open with DEXTER |
Our analysis used OSMARCS 1D LTE model
atmospheres (Asplund et al. 1997; Gustafsson et al. 1975; Plez et al. 1992; Gustafsson et al. 2003; Edvardsson et al. 1993) and the
TURBOSPECTRUM spectral synthesis code (Alvarez & Plez 1998). Models were
interpolated in pre-computed grids for a metallicity of
,
since this was the final iron abundance (see
Sect. 5.2), and with an
-element enhancement of
dex. We adopted solar abundances from
Grevesse & Sauval (2000), with the exception of O,
for which we adopted
,
based on 3D model atmospheres
(Caffau et al. 2007b; Ludwig & Steffen 2007).
The code TURBOSPECTRUM is used to determine 1D element
abundances in each component of the binary, either via spectrum
synthesis or by comparing the observed equivalent widths of different
stellar lines with the theoretical curves of growth (see
Sect. 5.2).
In this work we also consider two 3D model atmospheres, which have been
computed with the
code (Wedemeyer et al. 2003; Freytag et al. 2002), one for each star. The
atmospheric parameters are close to those observed for the two stars:
/
/[Fe/H]: 6550/4.50/-3.0 (A) and 5920/4.50/-3.0 (B). Each model
consists of a representative set of snapshots sampling the temporal evolution
of the photospheric flow at equal intervals in time. The total time intervals
were 2400 s for the warmer star A, and 9500 s for the cooler star B.
These time intervals should be compared to the convective turn-over
time scales. From the hydrodynamical point of view, typical time scales in
the models for both components are not much different from that in a solar
model, where the convective turn-over time scale amounts to about 500 s.
Thus, we sample about five turn-over time scales for the hotter and almost 20 for the cooler component.
The comparison of 3D vs. 1D models depends on which particular 1D model
is chosen. We compared each of our 3D models (hereafter denoted as
,
obtained from the mean temperature and pressure structure
of the full 3D model), to a corresponding standard
hydrostatic 1D model atmosphere (hereafter denoted as 1D
). The
model is a temporal and horizontal average of the 3D structure over surfaces of equal (Rosseland) optical depth. It is only dependent
on the particular way the 3D model is averaged.
The 1D
model is calculated with a 1D atmosphere code called
LHD. It assumes plane-parallel geometry and employs the same micro-physics
(equation-of-state, opacities) as
.
Convection is described by
mixing-length theory. Somewhat arbitrary choices to be made
relate to the value of the mixing-length parameter, which formulation
of mixing-length theory to use, and how turbulent pressure is
treated in the momentum equation; see Caffau et al. (2007a) for further
details. As usual, in the spectral synthesis of the 1D models, a value
of the micro-turbulence has to be adopted. For 1D as well as 3D models
the spectral synthesis calculations were performed with the spectrum
synthesis code Linfor3D
.
There are two main effects that distinguish 3D from 1D models,
the average temperature profile and the horizontal temperature fluctuations.
We quantify the contribution of both effects by introducing
the 3D correction as: 3D - 1D
.
The average temperature profile provided by a hydrodynamical simulation is
different from that of a 1D atmosphere assuming radiative equilibrium. This
effect is shown in Fig. 2, where the 3D average temperature profile,
plotted as a function of the pressure, is compared to the profiles from
1D
and MARCS models. As is evident from the plot, the
temperature profile is cooler than both 1D models in the
outer photospheric layers for both of the stars. This often-encountered
effect in metal-poor atmospheres (Asplund et al. 1999) is particularly important
for the oxygen abundances derived from OH molecules, as the difference is
largest precisely in the layers where these lines are formed. The result is
that the oxygen abundances become lower in the 3D formulation than in the 1D;
we quantify this effect through the 3D correction as defined above.
![]() |
Figure 2:
Average temperature profile of 3D
![]() ![]() |
Open with DEXTER |
![]() |
Figure 3: The two components of CS 22876-032 on a 14-Gyr isochrone by Chieffi & Limongi (priv. comm.) for Z=10-6 (solid line). For comparison, isochrones for 14 Gyr and Z=10-5 (dashed) and 12 Gyr and Z=10-6 (dotted) are also shown. |
Open with DEXTER |
The atmospheric parameters of each star in the CS 22876-032 system were estimated from the
photometric data available in Norris et al. (1993) and Preston et al. (1991), from whom we
adopt V=12.84, B-V=0.397,
U-B=-0.255 with uncertainties of 0.01, 0.02 and 0.01 respectively. We adopt
from Norris et al. (2000) and
Schuster et al. (1996,1993). We also extracted, from the 2MASS
database,
and
.
The
equations derived by Carpenter (2001) to transform from
2MASS magnitudes to the homogenised photometric system of
Bessell & Brett (1988) were then applied.
From the above information it is possible to estimate the reddening-corrected colours (U-B)0, (B-V)0, (V-K)0 and (J-K)0, which we use to derive the parameters of both components of the binary by comparing with theoretical isochrones. We have chosen the isochrone of Chieffi & Limongi (priv. commun.) for 14 Gyr and metallicity Z=10-6, from which one can compute composite colours from pairs of two models that lie on that isochrone. Thus, the stellar parameters can be derived from the best fit to the observed colours which also satisfy the mass ratio determined from the orbital solution.
We note that the isochrones of Chieffi & Limongi use the colour transformations based on ATLAS model atmospheres, for this temperature range, and the synthetic colours of Bessell et al. (1998).
The result is shown in Fig. 3, which corresponds
to a primary star with
K and
and a
secondary star with
K and
.
We
checked that a change of
Age
Gyr translates into a
change of +25 K and -0.01 dex for the
and
respectively, whereas a variation of the metallicity of
dex dsoes not have any impact on the resulting
stellar parameters (see Fig. 3).
The uncertainties of the stellar parameters were estimated using Monte Carlo
techniques. We injected noise in the seven observed quantities, V, B-V,
U-B, K, J, E(B-V), and
following Gaussian distributions with
standard deviations equal to the errors of these quantities. From these
distributions we computed a set of five variables, (U-B)0, (B-V)0, (V-K)
0, (J-K)0, and
for the 10 000 samples. We then found the best fit
to each of these set of variables via a
minimisation, defining
,
being
the ``observed'' value for each Monte Carlo simulation and
the
value extracted from two pairs of points in the theoretical isochrone.
The results of these simulations for the effective temperature and surface
gravity of both components are shown in Figs. 4 and 5
respectively. The lowest contour encloses roughly 95.4% of the 10 000
Monte Carlo events, analogous to
for a normal distribution. From these
simulations we adopted an error, at the
level, of
K and
K for the effective temperature, and
dex for the surface gravity.
For single stars, the wings of H
is also a very good
temperature indicator (Barklem et al. 2002; van't Veer-Menneret & Mégessier 1996; Cayrel 1988; Fuhrmann et al. 1993). Adopting the
broadening theory of Barklem et al. (2000), we computed H
profiles for
several effective temperatures, using TURBOSPECTRUM.
Unfortunately, all our UVES spectra were obtained near maximum line
separation, and the velocity difference is of the order of
30
,
which precludes separation of the individual H
profiles.
Therefore, we had to compute a composite spectrum in order to match
the observed profile.
![]() |
Figure 4: Distribution of effective temperatures for CS 22876-032 A and B obtained by Monte Carlo simulations, comparing the observed colours with those from the isochrone in Fig. 3 for the observed mass ratio. The lowest contour encloses 95.4% of the 10 000 simulations. |
Open with DEXTER |
![]() |
Figure 5: Same as Fig. 4, for the surface gravities. |
Open with DEXTER |
![]() |
Figure 6:
Computed H![]() |
Open with DEXTER |
Figure 6 compares this synthetic H
profile with the
observed profile for several combinations of effective temperatures
+
.
We did not vary
because
the H
absorption line of the cooler star B is weaker and
severely veiled by the flux of star A, so the combined profile is not
sensitive to changes in
.
This comparison seems to
confirm our estimate of the effective temperature from the colours,
suggesting also that it is on the same scale as the H
-based
temperatures.
Most of the elemental abundances were determined from equivalent width measurements of selected unblended lines. These were made with an automatic line-fitting procedure based on the algorithms of Charbonneau (1995), which performs both line detection and Gaussian fits to unblended lines. The implementation is the same as described in François et al. (2003). The equivalent widths were then corrected for the appropriate veiling factors and provided as input to TURBOSPECTRUM to determine the abundances. The detailed line-by-line abundances, together with observed EWs, veiling factors, and atomic data for both components can be found in Table A.1.
The mean abundances for each element, listed in
Table 3, are computed using the adopted stellar
parameters derived in Sect. 5.1 and a microturbulence of
.
We performed full 3D computations with Linfor3D for the Li
doublet and the OH lines. A full 3D analysis for the hundreds of lines
involved in this work is a considerable computational task, well
beyond the scope of the present paper. However, we used the
3D
models as input to TURBOSPECTRUM to
estimate the expected corrections due to the different average
temperature profiles of the 3D models.
From the full 3D spectrum synthesis performed for Li and OH, we expect
this to be close to the true 3D correction for star B, while we
expect significant contributions from the
temperature fluctuations in star A. In the following we
refer to (
3D
- 1D
)
as representing the difference between the abundance
found by TURBOSPECTRUM using a
3D
model
and that found using a MARCS 1D model. This is to distinguish them from
the true 3D corrections.
Table 3:
Element abundances of CS 22876-032. The Solar O abundance is
adopted from the 3D measurement of atomic lines of Ludwig & Steffen (2007); the other Solar
abundances from Grevesse & Sauval (2000). [X/H] and [X/Fe] are LTE values; [X/Fe]
refers to Fe I for OH and neutral species, to Fe II for
ionised species. The 3D abundance corrections
were
determined with TURBOSPECTRUM from the
3D
and 1D models, except for O and Li, where a full 3D analysis was performed
and we give two values: the abundance correction 3D -
3D
(``
3D
'') and ``1D'' = 3D - 1D
(see text).
is the standard deviation of the results from the n lines
(next column; if n=1, the wavelength of the line in nm is given).
See text for details on the measurement of Li, O, Sc, and Co.
In a double-lined spectrum the strength of each spectral line, in
particular those of the fainter component, is reduced by veiling from
the continuum flux of the other star. Thus, the measured equivalent widths
for each spectral component must be corrected for this veiling effect,
in order to obtain the intrinsic
values. The corrected equivalent widths can be estimated by
multiplying the observed EWs by veiling factors,
,
which solve the equation
and
where
is the primary-to-secondary
luminosity ratio. The values
are wavelength dependent
and can be estimated theoretically by computing the flux of each
stellar component, taking into account the ratio of the stellar radii.
Thus, the luminosity ratio can be expressed as
,
where
and Ri are the flux and radius of each star.
For consistency with the isochrone colours, we use
version 9 of the ATLAS code
(Kurucz 1993a,2005a) in its Linux version
(Sbordone 2005; Sbordone et al. 2004) to compute model
atmospheres and fluxes for each star by adopting the
and
derived in Sect. 5.1. We used the ``NEW'' Opacity Distribution
Functions (Castelli & Kurucz 2003), with 1
micro-turbulence, a
mixing-length parameter
of 1.25, and no overshooting. The
formulation of the mixing length is different between MARCS and ATLAS;
that used in our MARCS models roughly corresponds to
in the ATLAS formulation.
Table 4:
Abundance errors in CS 22876-032.
,
,
and
are the abundance changes caused
by changes in
of 100 K (A) or 150 K (B), in
by
0.1 dex, and by 0.5
in the microturbulence velocity
.
Other column headings and comments as in Table 3.
The ratio of the stellar radii was extracted from the theoretical
isochrone, being
.
The derived veiling factors lie in the range
(
)
in the
spectral region
300.0-700.0 nm. These estimates also
compare well with those used by Norris et al. (2000), although they adopted
fixed values for large spectral regions. In particular, they used
and
for all Fe lines between 370 and 440 nm, whereas we used
and
in that spectral region (see
Table A.1).
In addition to the 1D veiling factors computed for the effective
temperatures of both stars, we also calculated the
3D
veiling factors using the
3D
atmospheric models (whose temperature structure is different from that
of the OSMARCS 1D models). These veiling factors were adopted to
properly correct the EWs given as input to the
TURBOSPECTRUM code to estimate the (
3D
- 1D
)
abundance corrections (see
Sect. 5.2). In addition, we note that the 3D - 1D
correction (see Sect. 4.2), computed only for Li and OH lines, does
not consider different veiling factors for 1D and 3D models. In this
case, we computed only the 3D veiling factors using the continuum flux
provided by the full 3D model of each star in the spectral region
close to these lines. Thus, these 3D veiling factors were applied to
the observed EWs, and the resulting EWs were used to compute the
3D - 1D
corrections reported in Table 3.
The abundance measurements are dependent on the model parameters, i.e.
effective temperature, surface gravity and microturbulence.
However, in the analysis of a spectroscopic binary, it is not possible
to avoid the influence of the veiling factors on the error estimates.
The veiling factors depend on the effective temperatures and surface
gravities of both stars. Therefore, in order to estimate the sensitivity of an
element's abundance to a given stellar parameter, one should also estimate
how veiling factors change when one of the stellar parameters of each
star is modified. Thus, we also computed 1D veiling factors for
four pairs of models
/
,
/
according to the errors of the stellar parameters (see Sect. 5.1), by changing
one stellar parameter and fixing the three remaining parameters. The
uncertainties on the elemental abundances due to the errors of the
different model parameters are listed in Table 4. The
uncertainty of the microturbulence was assumed to be 0.5
.
The errors computed from the dispersion of the line measurements and the
signal-to-noise ratios are listed in Table 3. We used the
Cayrel formula (Cayrel 1988) to estimate the errors of the
observed EWs. Due to the high S/N of the spectra we obtained, these errors are
typically 0.1 pm in most of the spectral region covered,
except for the blue spectrum at 310.0-320.0 nm where the S/N ratio
drops significantly. Thus, the dispersion of the measurements for elements with
lines in this range is larger than for lines above 400 nm. In Table 3,
we give the larger of these two estimates for each element in each star. For Li,
the error due to the uncertainty in the continuum level was computed from a
Monte Carlo simulation, by injecting noise corresponding to the actual S/N ratio
near the Li line in the best-fit synthetic spectrum. In each case, the S/N ratio
was estimated taking into account the corresponding veiling factors.
We made a careful selection of 38 reliable Fe I lines in star A and 37 in B, taking into account the radial velocity separation of the two sets of lines in the double-lined spectrum. It is reassuring that we find the same [Fe/H] for both stars, within the errors. Note that in our analysis, in contrast with Norris et al. (2000), this was not imposed a priori, and thus supports our determination of atmospheric parameters. Our value of [Fe/H] is very close to that by Norris et al. (2000) despite the very different effective temperatures adopted. This results from the fact that different effective temperatures also imply different veiling factors, which must be factored in.
Reviewing the (3D
- 1D
)
corrections, we see that [Fe/H] is reduced in 3D, and the difference
between the two stars increases. We expect the true 3D corrections of the
primary star to be largely due to horizontal temperature fluctuations,
therefore probably larger than our (
3D
- 1D
)
correction, and we predict that a
full 3D-LTE synthesis will not improve the agreement of [Fe/H] between the
two stars. Our best estimate of [Fe/H] of the system is still the (
3D
- 1D
)
corrected value for star A: [Fe/H]=-3.78, which might be further reduced
by a full 3D-LTE synthesis. This confirms that the stars in CS 22876-032
are the most metal-poor dwarfs known to date.
We note that ionisation equilibrium is not achieved in either star. In both, the abundance derived from the Fe II lines is larger, for star B by a factor of two. We note, however, that the Fe II abundance shows a very large scatter in both A and B (0.25 and 0.31 dex, respectively) so that, within errors, the Fe I and Fe II abundances remain compatible. The number of Fe II lines measured is very large for stars of this metallicity (13 in star A, 9 in B). However, all the Fe II lines are weak and the majority of them are in the UV range, where the S/N ratio of our spectra drops dramatically.
The (3D
- 1D
)
corrections for Fe II are in the opposite direction to those
for Fe I, making the ionisation imbalance worse. The different
signs of the (
3D
- 1D
)
corrections for neutral and ionised species are
due to the different ionisation structure of the
and
1D models. We note that for the metal-poor subgiant HD 140283,
Shchukina et al. (2005) have performed NLTE computations for Fe I and
Fe II, using a single snapshot of a hydrodynamical simulation,
and found 3D-NLTE corrections of +0.6 for Fe I and +0.4 for
Fe II. To the extent that these computations can be considered
representative of the stars in CS 22876-032, we expect that a full
3D-NLTE analysis might achieve a better ionisation balance for iron.
![]() |
Figure 7: Synthetic spectral fits to the Li line in the co-added UVES spectrum of star A ( top) and B ( bottom). The observed spectra have been corrected for veiling (see Table 5), so the lines appear with their intrinsic strength in each star. |
Open with DEXTER |
Table 5:
Li abundances in CS 22876-032.
includes
corrections for depletion by diffusion (see text).
The high quality of our spectra allowed us to measure the Li doublet in both components of CS 22876-032 for the first time. Figure 7 shows our mean spectra of the Li region, where our well-resolved spectra (only) have been co-aligned on the lines of star A (top) and B (bottom), respectively. The superposed lines of the other star appear only slightly diffuse, because the velocity difference between the two stars is nearly the same in all our UVES spectra (see Table 1). The spectra in Fig. 7 have been corrected for veiling, as discussed in Sect. 5.2.1.
The Li abundance of each star was computed from two sets of spectra taken on different nights. We found differences of 0.01 and 0.05 dex between star A and B and adopted the average value of the two measurements. Table 5 lists the observed equivalent widths and the average Li abundances, with and without the NLTE corrections obtained from the tables of Carlsson et al. (1994). We further correct for the effect of depletion as predicted by the standard isochrones of Deliyannis et al. (1990). The correction is negligible for the hot primary, but larger for the secondary (see Table 5).
It is certainly surprising that the two components appear to have a different lithium content. The Li abundance of the primary component seems to be consistent with that observed in other metal-poor stars, i.e., the Spite plateau (Ryan et al. 1999; Charbonnel & Primas 2005; Spite & Spite 1982a; Asplund et al. 2006; Meléndez & Ramírez 2004; Bonifacio et al. 2007; Spite & Spite 1982b; Bonifacio & Molaro 1997), in spite of the uncertainties on the temperature scales adopted over the years by the various authors, but the secondary star definitely seems to exhibit a lower Li content.
According to the Li depletion
isochrones of Deliyannis et al. (1990), if star B were 350 K cooler than we assume,
the correction for A(Li) would be 0.6 dex. Such a change in
would
also imply a slightly higher
from the isochrone, so the model
dependencies given in Table 4 would imply that A(Li) should be
reduced by 0.17, giving a ``corrected'' Li abundance of A(Li)=2.20,
in essential agreement with A(Li) of the primary (we do not consider
variations in the temperature of star A). By changing simultaneously
the effective temperatures of
the pair (within the range allowed by photometry), we would derive
different Li abundances for each of the components. However, we would
also change the veiling factors, and the final abundance
difference between the two stars would be roughly the same.
We have checked whether inaccurate veiling corrections,
especially for the fainter lines of star B, could be the cause of the
different Li abundances. It turns out that this is impossible: bringing
into consistency with the Spite plateau would
require a doubling of the veiling correction at 670 nm; given that we
find consistent abundances for Na at 590 nm and Mg I at 880 nm, the
veiling correction cannot be off by a factor two at the intermediate
wavelength.
For the Li lines we also performed a full 3D-LTE synthesis using Linfor3D. The resulting corrections are listed in Table 5. Since the 3D computation was performed in LTE, the Li abundances must not be taken as definitive, as shown by Cayrel & Steffen (2000) and Asplund et al. (2003). It is, however, interesting to notice that while the 3D effect in star A is almost entirely due to the horizontal temperature fluctuations, for star B it is almost entirely due to the cooler average temperature profile of the 3D model. A full 3D-NLTE synthesis of Li in CS 22876-032 is beyond the scope of this paper. Full 3D-NLTE synthesis of the Li profile in HD74000 was recently addressed by Cayrel et al. (2007).
Table 6:
3D abundance corrections for the OH lines. [O/H] and [O/Fe] are relative to
.
The oxygen abundances have been derived from UV OH lines of the (0-0)
vibrational band of the
electronic system. The
use of these lines for oxygen measurements in metal-poor stars was
pioneered by Bessell et al. (1991). We were able to measure four lines
in the primary and nine lines in the secondary. Following the extensive
surveys of OH lines in metal-poor stars by Boesgaard et al. (1999b); Israelian et al. (1998,2001),
and the controversial finding of strongly increasing [O/Fe] with
decreasing metallicity, Asplund & García Pérez (2001) warned of the
possible role of 3D effects on the formation of these lines. For this
reason we decided to compute ad hoc
hydrodynamical
simulations for this system to correctly evaluate the 3D
effects. Our results are summarised in Table 6. Some of
the analysed lines are shown in
Figs. 8-10.
The 3D effects are clearly large and considerably different between
the two stars. Asplund & García Pérez (2001) attributed the 3D corrections
primarily to the different average temperature profile of the 3D models, in particular to their extremely cool outer layers. However,
we see that for the primary star the 3D correction is almost evenly
shared between average temperature profile and horizontal temperature
fluctuations.
![]() |
Figure 8: Synthetic spectral fits to OH lines in the co-added UVES spectrum of star A ( top) and B ( bottom) in CS 22876-032. The observed spectra have been corrected for veiling factors of 1.30 (A) and 4.36 (B). |
Open with DEXTER |
![]() |
Figure 9: Same as Fig. 8 for another OH line. |
Open with DEXTER |
![]() |
Figure 10: Synthetic spectral fits to OH lines in the co-added UVES spectrum of star B. The observed spectra have been corrected for a veiling factor of 4.36. |
Open with DEXTER |
Our cooler 3D model atmosphere exhibits much smaller temperature
fluctuations around the mean than the hotter one. In the cool
component the average temperature becomes so low that a substantial
amount of H2 is formed in the higher photospheric layers. The
associated increase of the specific heat makes it much harder for
pressure fluctuations to introduce temperature fluctuations. Due to
the smaller temperature fluctuations the cooler 3D and
3D
models provide essentially the same abundances.
The model of the primary star is hotter on average, H2 molecules
are much less abundant, and temperature fluctuations are much more
pronounced. Consequently, the resulting abundances differ between 3D
and
3D
models.
Figure 11 illustrates the
situation. The 3D model exhibits a stronger cooling with
respect to the 1D
model at 5900 K than is the
case at 6500 K. Over a wide pressure range, the structure is
almost adiabatic and passes through a region of substantial H2 molecule formation, indicated by the rather high values of the specific
heat, suppressing temperature fluctuations in that region. The almost
adiabatic structure of the cool model also indicates that the
convective overshooting is very efficient compared to radiative
heating, and has driven the thermal structure into almost adiabatic
equilibrium.
The large difference in the behaviour of our two models warns us that
to measure reliable abundances from OH one needs a grid of 3D models
which is fairly dense in temperature, to capture, for any metallicity,
the
at which H2 formation sets in.
Finally, we note that the O abundance derived from OH lines is rather sensitive to the surface gravity, as can be seen from Table 7. An increase of gravity of 0.3 dex introduces a decrease in O abundance of about 0.1 dex. This gravity dependence is larger than that of Fe I, however it is advisable to use Fe I to derive [O/Fe], since the gravity dependence of Fe II is in the opposite direction. Note that the values reported in Table 7 have been estimated without revising the veiling factor when changing the surface gravity of the models, contrary to the error estimates described in Sect. 5.2.2.
![]() |
Figure 11:
Entropy profiles of the ![]() ![]() ![]() |
Open with DEXTER |
The Be II resonance doublet at 313.0 nm is within the
wavelength range covered by our spectra. The S/N ratio in
that region is 25, yet none of the Be lines are detected, which
implies an upper limit for the Be abundance log(Be/H) < -13.0.
As this is an order of magnitude higher than the Be abundance expected
for these stars from the trend of Be abundance with metallicity
(Ryan et al. 1992; Boesgaard & Novicki 2006; Boesgaard et al. 1999a; Gilmore et al. 1992; Molaro et al. 1997; Primas 2002), this
result is not significant.
For the other elements, abundances were determined directly from the EWs, except for Sc and Co for which we used spectrum synthesis to take hyperfine splitting (HFS) into account. For Co we used the A and B factors measured by Pickering (1996). For the Sc II 361.3 nm line we used the A and B factors measured by Gangrsky et al. (2006). In both cases we used the code LINESTRUC of Wahlgren (2005) to compute HFS components. For the Sc II 424.6 nm line we used the HFS components given in Table 5 of McWilliam et al. (1995). The detailed atomic data of HFS for Sc and Co are provided in Table A.2. For Mn we used a single line, the strongest of the Mn II lines of Mult. 3, for which HFS is negligible according to Castelli & Hubrig (2004).
Silicon was only measured from the Si I line at 390.5 nm. In metal-poor cool giants, the silicon abundance is derived from a line at 410.3 nm, since the line Si I 390.5 nm is severely blended with CH lines (Cayrel et al. 2004). The Si I at 410.3 nm is very weak in metal-poor dwarfs, but CH lines are so weak that the line at 390.5 nm can be used.
Aluminium was measured from the resonance lines
Al I 394.4 nm and Al I 396.1 nm. The Al I 394.4 nm is blended with CH lines, which are also extremely weak
in metal-poor dwarfs and were not taken into account. We
computed synthetic spectra of the CH lines at 390.5 nm and 394.4 nm
for the stellar parameters of both dwarfs, and the CH lines are not
visible even for [C/Fe] as large as +2 dex. However,
the 394.4 nm line provided a significantly larger abundance than that
derived from the 396.1 nm line in both stars. For this reason, in
Table 3 we give the average Al abundance derived from both
lines with a large dispersion of 0.2 dex.
Table 7: 1D gravity effects on the Fe and O abundances.
Using 3D
models, we determine the (
3D
- 1D
)
corrections
listed in Table 3. We note that the (
3D
- 1D
)
corrections also
take into account different veiling factors estimated using
3D
models and 1D OSMARCS models. In general, the
veiling factors estimated from
3D
models are higher
for star B and lower for star A than those
obtained using 1D models. This effect is especially important for the
secondary star, and becomes more significant at shorter wavelengths. Taking into
account this effect, it is interesting to note that (
3D
- 1D
)
corrections are in general negative for neutral species and positive
for ionised species, at least for the primary star, for which the
veiling factors are not significantly different from those estimated
using 1D models, as already noted for iron. This reflects the
different ionisation structure of 1D and
3D
models.
The (
3D
- 1D
)
corrections do not help to achieve ionisation
equilibrium for Ca, Cr and Fe, the only elements for which abundances
were determined from both neutral and ionised species.
![]() |
Figure 12: 1D-NLTE Li abundances vs. [Fe/H] for the stars in CS 22876-032 (circles) and in other metal-poor dwarfs as reported by Asplund et al. (2006, triangles) and Bonifacio et al. (2007, diamonds). The Li abundances by Asplund et al. (2006) were recomputed using TURBOSPECTRUM as reported by Bonifacio et al. (2007). No 3D-LTE or 3D-NLTE was considered. |
Open with DEXTER |
In Fig. 12 we show the lithium abundances for the two
components of CS 22876-032, together with our data from Paper VII,
including the data of Asplund et al. (2006) rescaled in [Fe/H] and A(Li) to be homogeneous with our own.
Star A appears to have a Li abundance at the same level as the
majority of stars with metallicity below -2.5, if anything slightly
higher. Star B appears to be far below any of the
other measured stars. It should be noted that all the other stars in
Fig. 12 have effective temperatures determined from the wings
of H
using the broadening theory of Barklem et al. (2000), while for
CS 22876-032 they have been determined independently from
colours and isochrones. However, the reasonable agreement
between our computed H
profile and the observed profile shown
in Fig. 6 suggests that the two temperature scales are
fairly close.
The fact that at the lowest observable metallicity, star A
remains at the level of the plateau suggests that there is no downturn
or decrease in Li abundance at the lowest metallicities. This suggests
that the slope of A(Li) with [Fe/H] which is detectable in the sample
of Asplund et al. (2006) (but not in that of Paper VII alone) is not real,
but rather an artifact due to the H
temperature scale. It is
possible that it is ultimately due to our inability to correctly model
the atmospheres of extremely metal-poor stars and the wings of Balmer
line profiles. In Paper VII we argued that the data could suggest
either a vertical drop or an increased scatter in A(Li) at the lowest
metallicities. The drop now seems to be ruled out by the A(Li)
measured in star A. The measurement in star B, taken at face value,
may support the idea that at metallicities below -2.5, the Spite
plateau displays a sizeable scatter.
In Sect. 5.2.4 we pointed out how the difference of the
Li abundance between the two stars could be resolved by assuming that
the temperature of star B was 5550 K. Star B would thus be subject
to considerable Li depletion, according to the standard
Li depletion isochrones of Deliyannis et al. (1990). Such isochrones are available
only for metallicities considerably higher than that of CS 22876-032.
If, for any reason, either the lower metallicity of our system, or
inclusion of other physical phenomena, the dependence of Li depletion
on
is steeper than predicted by purely diffusive standard
isochrones, then to reconcile the Li abundances of the two stars, the
temperature of star B could be higher than 5550 K. Considering that
our estimated error on the effective temperature of star B is 150 K
(
), such cooler temperatures are not totally implausible.
In our view, the existence of a real scatter in Li abundances at the lowest metallicities remains to be established beyond any reasonable doubt. It is, nevertheless, worthwhile discussing the possible implications of such a scatter, if real.
![]() |
Figure 13:
Same as Fig. 12, but in the Li-
![]() |
Open with DEXTER |
From their Li isochrones in the presence of pure atomic diffusion
(Fig. 5 of Richard et al. 2005) it is obvious that one should expect Li
abundances in the range
Li
.
The sample
constituted by the 8 stars from Paper VII with metallicity below -3.0 and the two stars in CS 22876-032 spans the range 1.91-2.20 in A(Li). One could therefore suspect that in extremely
metal-poor stars, turbulence is lower and atomic diffusion more
efficient, causing the increased scatter in Li abundance.
In Fig. 13
we see no clear trend of A(Li) vs.
.
When comparing this figure
with Fig. 5 of Richard et al. (2005) we note that the ``high'' Li
abundances of the hottest stars in the sample (among which is the primary
of CS 22876-032), preclude any clear resemblance between the two pictures.
Therefore, current pure diffusion models are unable to explain the
behaviour of Li abundances with [Fe/H] and
.
It is possible, however, that they may do so, after an ad hoc
parametrisation of turbulence with metallicity.
Recently Korn et al. (2007,2006) found that the models of Richard et al. (2005), with a suitable value for the turbulent diffusion coefficient, can explain the 0.12 dex difference in A(Li) they find between stars at the turn-off and on the subgiant branch stars of the globular cluster NGC 6397. The same authors, however point out that assigning temperatures for the TO stars, hotter by 170 K (therefore close to the temperatures adopted by Bonifacio et al. 2002, for the TO stars of this cluster), such a difference would vanish.
Though suggestive, the applicability of such turbulent diffusive models remains to be proven. The main cause of concern is the parametrisation of the turbulent diffusive coefficient, which is linked to a fixed temperature, and not to the bottom of the convective zone (Richard et al. 2002).
In order to strengthen the observational constraints on such models and refine the estimates of the scatter and slope (or lack thereof) of the extreme metal-poor end of the Spite Plateau, further high-quality spectroscopy of EMP stars and additional accurate constraints on the effective temperatures of the whole sample are highly desirable.
![]() |
Figure 14: 3D [O/Fe] ratios of the stars in CS 22876-032 (circles) and in the metal-poor giants of Paper V. Triangles: ``unmixed'', diamonds ``mixed'' stars; downward triangle with arrow: upper limit for CS 22172-002. |
Open with DEXTER |
Figure 14 compares our 3D-LTE O abundances obtained from OH UV bands
in CS 22876-032 with the high-quality measurements of Paper V for giant
stars (mixed and unmixed), obtained from the [OI] 630 nm line. We note that
we computed the 3D-LTE [O/Fe] ratios using
the true 3D-LTE oxygen abundances and the 3D
Fe I abundances. We chose Fe I rather than
Fe II abundances as the reference, because
their sensitivity to the surface gravity is similar to that of the OH
bands. The reader might wonder why we find different O abundances for
stars A and B, but they are in fact consistent within the
error bars, which mainly reflect the uncertainties in the effective
temperature (see Table 4).
The abundances derived from [OI] are the raw values obtained from 1D-LTE model atmospheres. One worry is the possible effects of granulation on the abundances derived from the [OI] line in giants. In a recent paper, Collet et al. (2007) investigated the 3D effects in giant stars and found very small corrections for the [OI] lines for metallicities down to -2.0, but sizeable (almost 0.2 dex) downward corrections for [O/Fe]) at metallicity -3.0 (their Fig. 13 and Table 3). If we apply the corrections interpolated and extrapolated from Table 3 of Collet et al. (2007) to the giants of Paper V, the [O/Fe] decreases in all the giants and the good agreement between CS 22876-032 and the giants no longer holds. The mean [O/Fe] of the giants would be 0.51, while that of CS 22876-032 is 0.87. If instead we take the measurements of Paper V at face value, the mean [O/Fe] is 0.72.
At present we have no full 3D
models for giant stars, however, from a few snapshots for models of
= 4900,
and metallicity -2.0 and -3.0 we find
little difference at the two metallicities, and very small 3D
corrections. The 3D - 1D
abundance correction amounts to -0.006 dex at
metallicity -2.0 and -0.037 dex at -3.0, which is negligible. For our
models of giants the difference between the mean 3D temperature
structure and a corresponding 1D
(
= 1.0) model is not large. In
particular we do not find the substantial cooling of the highest photospheric
layers at metallicity -3.0, which
Collet et al. (2007) find (see their Fig. 1). Moreover, the mean 3D temperature
profile is slightly hotter than the 1D
temperature profile in the
[OI] 630 nm line-forming layers around log
.
The differences between our assessment of the impact of granulation on
the abundances derived from the [OI] 630 nm line in giants, and that of
Collet et al. (2007), will be further investigated in the future. This might
be rooted in the different binning schemes adopted by the two codes
for the opacity (4 opacity bins for the Stein & Nordlund code and 6 opacity bins for
). In any case we believe that, at present, it
is safer not to apply any 3D correction to the O abundances derived
for giants from the [OI] 630 nm line.
In Paper V we found that the mean value of
in the
range
was about
.
The
weighted average value in CS22876-032,
,
agrees with this determination. This
mean value does not exclude a slight increase of the ratio [O/Fe] in
the range
as seen in the theoretical
predictions of the chemical evolution models presented by
François et al. (2004). A more detailed discussion of the general behaviour of
oxygen with metallicity will be made in forthcoming investigations,
hopefully after 3D corrections will be determined for all the stars,
and for different lines of oxygen, Fe I and Fe II.
The LTE [Na/Fe] abundance ratios of both stars appear consistent
with the Galactic trend of this element in EMP giants
(Cayrel et al. 2004), subgiants and turn-off stars of similar metallicity
(Andrievsky et al. 2007). NLTE corrections are expected to be larger for the
metal-poor giants than for dwarfs. We have estimated the NLTE
corrections for Na to be
dex, according to the NLTE corrections
reported in Table 2 of Andrievsky et al. (2007). After applying these
corrections, Na abundances of the CS 22876-032 dwarfs remain
compatible with those of metal-poor giants and dwarfs, which exhibit
almost a constant ratio [Na/Fe
in the metallicity range
.
The LTE [Al/Fe] abundance ratios of the dwarfs in this system are 0.5 dex larger than those in metal-poor giants with
similar iron content. However, aluminium is also expected to exhibit
significant NLTE corrections (Baumüller & Gehren 1997), which might explain this
difference, as was the case for Na. This element also exhibits
an almost constant ratio [Al/Fe
for giants in the
metallicity range
when a fixed NLTE
correction of +0.65 dex is considered.
Within the errors, [Sc/Fe] is consistent with, although 0.2-0.3 dex lower
than, the [Sc/Fe] ratio in EMP giants which shows an almost constant
ratio [Sc/Fe.
Note that the [X/Fe] ratios given in Table 3
were computed relative to Fe II for ionised species. As noted
in Sect. 5.2.3, the Fe II abundances are less reliable than
those for Fe I because the Fe II lines in EMP dwarfs are
very weak, especially for star B in CS 22876-032. Therefore, [X/Fe]
for ionised species in star B should be regarded with caution.
The [Mg/Fe] ratios in both stars of CS 22876-032 are approximately
consistent with those found in EMP giants
(Cayrel et al. 2004), subgiants and dwarfs (Cohen et al. (2004), Bonifacio et al.
2007, in prep.), at the level of 0.3-0.4 dex.
[Ca/Fe] was derived from both Ca I and Ca II lines, which
yield similar [Ca/H] abundances, at least for
the primary star. However, although the [Ca/Fe] ratios differ by 0.2 dex, they seem to be slightly lower than those measured in
metal-poor giants and dwarfs, where a constant [Ca/Fe
is
seen.
[Si/Fe] also seems to be low, at [Si/Fe
for stars A and
+0.1 for the secondary, compared to the constant
[Si/Fe
for metal-poor giants. However, other metal-poor
dwarfs show similar abundances, which might be related to the
different Si lines used in giants and dwarfs (Bonifacio et al.
2007, in prep.).
Finally, [Ti/Fe] is constant at 0.3 in EMP giants. While we find
a similar result for star A, [Ti/Fe] in star B is completely different
due to its high Fe II abundance. We note that the standard deviation
in [Ti/Fe] is relatively high in both stars (
), although
the mean abundances are better determined when averaging the results from
the 19 lines in star A and 12 in B.
Chromium was derived from both Cr I and Cr II lines, and
we find a difference of 0.2 dex for the [Cr/Fe] ratios.
[Cr I/Fe I] in CS 22876-032 appears to agree with that
in other EMP dwarfs, but is slightly higher than seen in EMP giants.
The ratio [Mn/Fe] is consistent with the other EMP giants and dwarfs, at
[Mn/Fe
.
[Ni/Fe] also agrees with the values reported for
EMP giants, which show a constant [Ni/Fe
.
Cobalt is found to be slightly enhanced in CS 22876-032 relative to the gradually increasing trend of [Co/Fe] with decreasing [Fe/H] observed in giants, although marginally compatible within the errors.
One could ask whether a third star in CS 22876-032 might contribute
significantly to the total light and the veiling of the lines of the two
main components we have discussed so far. Such a star would need to have
a mass above 0.5
in order to have any significant effect on the
observed spectrum. The presence of such a third star can be ruled out by
two independent pieces of evidence.
First, we have individual spectra of CS 22876-032 with S/N ratios 100 in the region of the Mg I b triplet. These are among
the strongest stellar lines seen in these EMP stars and would be
at least as strong in the third star. The Cayrel formula predicts that
any line with an EW above
0.15 pm would be detected at the
level. Assuming
K and
(from the isochrone) plus
[Fe/H]= -3.6 and [Mg/Fe] = +0.2 for the hypothetical third star, we need
to dilute the strongest line of the Mg I b triplet at 518.36 nm
which would show an intrinsic EW of
15.0 pm. In Appendix A
we define the veiling factors for a triple system and conclude that the
non-detection of the 518.36 nm line requires f3 > 100. We can therefore
conclude that any third star contributes negligibly to the total light of CS 22876-032.
A second line of evidence is available from the radial velocities, which
are accurate to 1
and cover a period of 16 years. Any third star
of mass comparable to A and B should leave significant trends in the
velocity residuals from the orbital solution for periods shorter than
several decades. We have therefore attempted sinusoidal fits to the velocity
residuals for both stars and find periods of the order of 1300 days in both cases.
First, the standard deviations around these fits are 0.83 and 0.51
,
well below the purely observational errors, which shows that these
results cannot be statistically significant. Second, a period ratio of
3 between the outer and inner orbits is far too small for a triple
system to be dynamically stable. A white dwarf in an orbit of much longer period
is a possibility, but would not be detectable in our spectra.
In summary, we conclude that the abundance results reported here cannot be significantly affected by light from a third star in the system - certainly not the discrepant Li abundances of the two stars.
Our high-resolution VLT/UVES observations of the double-lined spectroscopic
binary CS 22876-032 confirm that it harbours the most metal-poor
dwarfs reported to date. Our improved orbital elements, together with published
photometry and theoretical isochrones, enable us to determine stellar
parameters of
K and
for the
primary (star A) and
K and
for the secondary (star B).
Using 1D OSMARCS models and the TURBOSPECTRUM code, we
determine abundances of Li, O, Na, Mg, Al, Si, Ca, Sc, Ti, Cr, Mn, Fe, Co,
and Ni, correcting the observed spectra for the veiling from the continuum
flux of the other star. We find [Fe/H
and
[Fe/H
for star A and B, respectively. Using CO5BOLD model atmospheres to estimate 3D abundance corrections, we compute
full 3D spectrum synthesis using the Linfor3D code for Li and O to estimate
the 3D - 1D
corrections, while we use a horizontal and temporal average of
the 3D model to compute
3D
abundances with TURBOSPECTRUM for the rest of elements. In general, we find (
3D
- 1D
)
corrections to be
0.1 dex, negative for neutral species and positive
for ionised species; for Fe in particular, we find corrections to be
-0.12 (A) and
-0.07 (B).
The [/Fe] ratios are consistent with our earlier results for
EMP giants (Cayrel et al. 2004, First Stars V), although Ca and Si are slightly
low ([X/Fe
), but actually consistent with our results for other EMP dwarfs
(Bonifacio et al. 2007, in prep.). [Na/Fe] appears consistent with
both EMP giants and dwarfs when NLTE corrections are considered. The
LTE value of [Al/Fe] is not consistent with those in EMP giants, but Al
is severely affected by NLTE effects, which may solve this discrepancy. The
iron-peak elements follow the established trends in EMP giants and dwarfs.
Our high-quality spectra allowed us to measure the Li doublet in both stars
of CS 22876-032 for the first time. We find NLTE Li abundances of
and
for stars A and B, respectively. While the
Li abundance of star A corresponds to the level of the Spite plateau, the
secondary star has a significantly lower abundance. This discrepancy
may be resolved by assuming that the secondary star has been subject
to significant Li depletion, which, according to standard Li depletion
isochrones, would have been the case if the star were 350 K cooler than
assumed by our analysis. Full 3D corrections for Li are estimated to be
-0.3 (A) and
-0.2 (B); however, these computations were
performed in LTE, and 3D NLTE corrections are needed to confirm the sign
and value of these corrections.
The near-UV part of our VLT/UVES spectra enabled us to measure oxygen
abundances from the OH bands. We find 1D [O/Fe] values of
(A) and
(B) and compute full 3D corrections for the OH lines,
which are -1.5 (A) and -1.0 dex (B). Using these
corrections and the
3D
Fe I abundances, we
determine 3D [O/Fe] ratios of
(A) and
(B).
These 3D [O/Fe] ratios are consistent with those derived from the [OI] line in EMP giants of similar metallicity, where 3D corrections should not
be significant.
Acknowledgements
We are grateful to A. Chieffi and M. Limongi, for computing, at our request, isochrones appropriate for the metallicity of CS 22876-032 and for many interesting discussions on the evolution of extremely low metallicity dwarfs. J.I., P.B. and H.-G.L. acknowledge support from the EU contract MEXT-CT-2004-014265 (CIFIST). P.B. also acknowledges support from MIUR - PRIN grant 2004025729. B.N. and J.A. thank the Carlsberg Foundation and the Danish Natural Science Research Council for support for this work. T.C.B. and T.S. acknowledge partial support from the US National Science Foundation under grants AST 04-06784, AST 07-07776, as well as from grant PHY 02-15783; Physics Frontier Center/Joint Institute for Nuclear Astrophysics (JINA). B.N. and J.A. acknowledge support from the Carlsberg Foundation and the Danish Natural Science Research Council. This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics Ans Space Administration and the National Science Foundation. This work has also made use of the IRAF facilities and the SIMBAD database, operated at CDS, Strasbourg, France.
Table 1:
Radial-velocity observations of CS 22876-032. For each
velocity,
is the estimated error and (O-C) the residual
from the orbital fit.
: H
velocities; omitted from
solution.
We derive here the expression for the veiling factors of a triple system which are a trivial extension of those for a double system, however since they are not readily found in any paper or book we know of, we provide them here for the reader's convienience.
We shall use the following notation:
![]() |
wavelength |
si (i=1,2,3) | the flux spectrum ![]() |
the ith component | |
ci (i=1,2,3) | the continuum flux spectrum ![]() |
the ith component | |
di (i=1,2,4) | the line depression = ci - si |
![]() |
the intrinsic equivalent width of a |
spectral line of the ith component | |
![]() |
the observed equivalent width of a |
spectral line of the ith component | |
![]() |
the observed equivalent width of the |
three components. |
Table A.1: Line data, equivalent widths, veiling factors and 1D abundances of CS 22876-032A, B.
Table A.2: Hyperfine structure of Sc and Co.