4. Abundance analysis
4.1. Stellar parameters
The models used in the analysis of the stars have been interpolated in the grid of Edvardsson et al. (1993), computed with an updated version of the MARCS code of Gustafsson et al. (1975) with improved UV line blanketing (OSMARCS models, see also Edvardsson et al. 1993). To compute the parameters of the stars we made the hypothesis that the star is a normal halo star (not very young), and used the isochrone of VandenBerg et al. (2000) computed for 12 Gyr and a metallicity of [Fe/H] .
The derivation, from the composite spectrum, of the atmospheric parameters of the A and B components of CS 22873-139 has been performed by successive iterations. We assumed that the two components have been formed together and thus have the same age, the same metallicity and differ only by their mass. A first approximation of the temperature of the A component is found from the profile of the wings of the line. Then its mass is read on the table of VandenBerg et al. (2000) corresponding to the isochrone, the mass of the B component is computed from = 0.90 and its parameters (temperature, absolute magnitude) are read in VandenBerg's table. The temperatures given by VandenBerg et al. have been shifted by 0.01 in log Teff (as needed to fit high parallax subdwarfs following Cayrel et al. 1997). The ratio of the fluxes are then deduced from the difference in magnitude. The obtained parameters are used to compute a second approximation of the composite profile of the hydrogen lines. A fast convergence is obtained (the wings of the H lines are not very sensitive to the parameters of the companion). We find a flux ratio in the B band equal to 0.40 (to be compared to 0.33 following Preston).
Let us remark that this isochrone is the most metal deficient computed by VandenBerg (it corresponds to the most metal-deficient globular clusters), it has been computed using a He abundance Y = 0.235, and an enhancement of the elements of 0.3 dex. Since our object turns out to be without the usual enhancement of elements in Pop I stars (cf Sect. 5.2), it has to be considered that the isochrone, computed for [Fe/H] = -2.3 dex, is in fact computed for a higher global metallicity: the isochrone is equivalent (Salaris et al. 1993) to a non-enhanced isochrone with [Fe/H] = dex. Our stars are more metal deficient than this value, but they are far enough from the turn-off for keeping the same flux ratio as the ratio which would be measured on an isochrone computed for a lower metallicity.
In Fig. 3 we present an example of the fit of computed and observed profiles. To determine the observed profile on the échelle spectra, the Gehren method has been used (Gehren 1990). In this method, often used, the continuum is fitted, not by an arbitrary polynomial, but by the mean of the observed blaze functions of the preceding and following orders (a method extensively used by the group of Gehren and Fuhrmann, in several papers by the Spite group and in particular Spite et al. 1996). For a good S/N ratio, this method gives, for spectra not crowded with absorption lines, an accuracy better than 0.5%, which corresponds to a random error of about 100 K in the temperature of each component. In the particular (more difficult) case where two different stars contribute to the observed spectrum, an estimation of the temperature error of 200 K for each star seems reasonable. (Of course there still remains the question of the existence of a possible additional systematic error). The position of the two components in the HR diagram is shown in Fig. 4.
The colors of the stars, following VandenBerg et al. (2000), are B-V = 0.37 and 0.51, resulting in = 0.41 for the binary: in disagrement with the bluer color observed by Preston et al. (1994), leading in Preston's paper to a higher temperature and a lower age. We discuss this problem in Sect. 4.3: in absence of any definite solution for this discrepancy, we will compute the abundances in both cases: low temperature and age of 12 Gyr, higher temperature and age of 8 Gyr (isochrones of VandenBerg et al. 2000). We begin with the first solution.
4.2. Atomic parameters and results
For iron, the log gf values are taken from Nave et al. (1994). The log gf values for Ti II are from Wiese & Fuhr (1975) and those for Na I, Mg I, Ca I, Sc II and Sr II are as adopted by Ryan et al. (1991). The measurements have been made individually on each spectrum.
We found that the best agreement with our observed spectra was obtained with the atmospheric parameters listed in Table 4 (see also Fig. 3). The results of the elemental abundance computations are given in Table 5.
Table 4. Best atmospheric parameters for the A and B components
Table 5. Abundances computed for the different lines of elements, using the parameters given in Table 4. In Columns 4 to 8 the logarithmic abundances of the different elements, log , are given using the convention log =12. In the last column, the mean value for each line
The mean abundances we derive for individual elements are given in Table 6. Fig. 5 shows an example of the best fit in the region of the Mg I 5172.7 Å line. The major error source in [Fe/H] is due to the error in . A change in of +200 K corresponds to a change in [Fe/H] of typically +0.14 dex. Effects of errors in abundance ratios like [Mg/Fe] are negligible.
Table 6. Abundances of the different elements in the Sun and in CS 22873-139 for the model deduced from the profile of the H line, and the model deduced from the colors of Preston
4.3. Analysis with Preston's parameters (2nd solution)
We have noted hereabove that the profile of the H line leads to temperatures too low for an agreement with the observed colors. Preston has computed that the global color would correspond to for the primary, leading to an age of 8 Gyears. An independent color measurement (Schuster et al. 1996) also provides a hotter global temperature ( = 6423 K). The reddening estimation of a binary is probably somewhat uncertain, but the reddening value is in agreement with the modern reddening maps (Schlegel et al. 1998).
We have therefore computed the parameters of the stars which would correspond to bluer colors. Using the table of VandenBerg which corresponds to an isochrone of 8 Gyr, we deduced that the temperature of the primary would be 6760 K and the temperature of the secondary 6160 K. With these parameters we computed also the abundances of the different elements: these abundances are also given in the Table 6.
The metallicities of the stars are a little higher (since the stars are supposed to be hotter) but it is important to remark that the ratios of the different elements like [Mg/Fe] do not change significantly.
The discrepancy between the spectroscopic and photometric temperature indicators, remains a problem, and several hypotheses may be imagined. For example, owing to the relatively high value of the index, it cannot be completeley excluded that the primary is a subgiant (or even a red HB star): Schuster et al. (1996) consider such a possibility for a few stars with a high index (their Sect. 3.1), and Norris et al. (2000) discuss this point for the rather similar binary that they analyze (but their star has a low index). At first, we noted that the ionisation equilibrium is not as good, in the second solution (hotter temperature) as it is in the first one, but since only one Fe II line can be measured in this very metal poor star, this fact is not a strong argument in favor of the low temperature solution.
Rather than further speculate about more or less different solutions, a check of the spectroscopic temperature should be made first. Unfortunately, the line is registered, in our spectra, on the edge of the detector, no accurate measuremement can be made, and the most urgent thing to do is to observe again both and using a better suited spectrograph (we will do it in the future).
© European Southern Observatory (ESO) 2000
Online publication: August 23, 2000