3 Stellar parameters and abundances

3.1 Methods

The abundance analysis was performed with the LTE spectral analysis code "turbospectrum'' in conjunction with OSMARCS atmosphere models. The OSMARCS models were originally developed by Gustafsson et al. (1975) and later improved by Plez et al. (1992), Edvardsson et al. (1993), and Asplund et al. (1997). Turbospectrum is described by Alvarez & Plez (1998) and Hill et al. (2002), and has recently been further improved by B. Plez.

The temperature of the star was estimated from the colour indices (Table 2) using the Alonso et al. (1999) calibration for giants. There is good agreement between the temperature deduced from B-V, V-R and V-K. However, Aoki et al. (2002) have shown that, in metal-deficient carbon-enhanced stars, the temperature determination from a comparison of broad-band colours with temperature scales computed with standard model atmospheres is sometimes problematic, because of the strong absorption bands from carbon-bearing molecules. In the case of CS 22949-037, although carbon and nitrogen are strongly enhanced relative to iron, the molecular bands are in fact never strong because the iron content of the star is so extremely low (ten times below that of stars analysed by Aoki et al. 2000). Neither the red CN system nor the C₂ Swan system are visible, and the blue CH and CN bands remain weak and hardly affect the blanketing in this region (cf. Sect. 3.2.1). Moreover, an independent Balmer-line index analysis yields $T_{\rm eff}= 4900$ K$~\pm125$ K, in excellent agreement with the result from the colour indices and with the value adopted by Norris et al. (2001).

With our final adopted temperature, $T_{\rm eff}= 4900$ K, the abundance derived from individual Fe I lines is almost independent of excitation potential (Fig. 1), at least for excitation potentials larger than 1.0 eV. Low-excitation lines are more sensitive to non-LTE effects, and the slight overabundance found from the lowest-excitation lines is generally explained by this effect (see also Norris et al. 2001).

Figure 1: Iron abundance as a function of excitation potential of the line. Symbols indicate different line strengths (in mÅ).

The microturbulent velocity was determined by the requirement that the abundances derived from individual Fe I lines be independent of equivalent width. Finally, the surface gravity was determined by demanding that lines of the neutral and first-ionized species of Fe and Ti yield identical abundances of iron and titanium, respectively.

Table 2 compares the resulting atmospheric parameters to those adopted by McWilliam et al. (1995) and Norris et al. (2001).

 

 

Table 2: Colour indices and adopted stellar parameters for CS 22949-037. Temperatures have been computed from the Alonso et al. (1999) relations for $E(B-V) \approx 0.03$ (Burstein & Heiles 1982). The reddening for different colours has been computed following Bessell & Brett (1988).

CS 22949-037			Colour
Magnitude or			corrected for	$T_ {\rm eff}$
Colour		Ref	reddening	(Alonso)
V	14.36	1
(B-V)	0.730	1	0.700	4920
(V-R)J	0.715	2	0.695	4910
(V-K)	2.298	3	2.215	4880

Adopted parameters for CS 22949-037
$T_ {\rm eff}$	$\log g$	[Fe/H]	$v_ {\rm t}$	Ref
4810	2.1	-3.5	2.1	1
4900	1.7	-3.8	2.0	4
4900	1.5	-3.9	1.8	5

References:
1 Beers et al., in preparation.
2 McWilliam et al. (1995).
3 Point source catalog, 2MASS survey.
4 Norris et al. (2001).
5 Present investigation.

  
3.2 Abundance determinations

The measured equivalent widths of all the lines are given in the appendix, together with the adopted atomic transition probabilities and the logarithmic abundance of the element deduced from each line. The error on the equivalent width of the line depends on the S/N ratio of the spectrum and thus on the wavelength of the line (Table 1). Following Cayrel (1988), the error of the equivalent width should be about 1 mÅ  in the blue part of the spectrum and less than 0.5 mÅ  in the red. Since all the lines are weak, the error on the abundance of the element depends linearly on the error of the equivalent width. In some cases (where complications due to hyper-fine structure, molecular bands, or blends are present) the abundance of the element has been determined by a direct fit of the computed spectrum to the observations.

Table 3 lists the derived [Fe/H] and individual elemental abundance ratios, [X/Fe]. The iron abundance measured here is in good agreement with the results by McWilliam et al. (1995) and Norris et al. (2001): with an iron abundance ten thousand times below that of the Sun, CS 22949-037 is one of the most metal-poor stars known today.

As expected for a giant star, the lithium line is not detected.

 

 

Table 3: Individual element abundances. For each element X, Col. 2 gives the mean abundance $\log \epsilon({\rm X})=\log N_{\rm X}/N_{\rm H} + 12$, Col. 3 the number of lines measured, Col. 4 the standard deviation of the results, and Cols.  5 and 6 [X/H] and [X/Fe], respectively (where $[{\rm X}/{\rm H}] =\log \epsilon({\rm X}) - \log \epsilon({\rm X})_{\odot}$).

Element	log $\epsilon(X)$	n	$\sigma$	[X/H]	[X/Fe]
Fe I	3.51	64	0.11	-3.99
Fe II	3.56	6	0.11	-3.94
C (CH)	5.72		synth	-2.80	+1.17
N (CN)	6.52		synth	-1.40	+2.57
O I	6.84	1	-	-1.99	+1.98
Na I	3.80	2	0.03	-1.88	+2.09
Mg I	5.17	4	0.19	-2.41	+1.58
Al I	2.34	2	0.03	-4.13	-0.16
Si I	4.05	2	-	-3.25	+0.72
K I	<0.89	2	-	<-4.06	<-0.09
S I	<5.09	1	-	<-2.12	<+1.78
Ca I	2.73	10	0.17	-3.63	+0.35
Sc II	-0.70	5	0.14	-3.87	+0.10
Ti I	1.40	8	0.09	-3.62	+0.35
Ti II	1.41	21	0.15	-3.61	+0.36
Cr I	1.29	5	0.10	-4.38	-0.41
Mn I	0.61	2	0.01	-4.78	-0.81
Co I	1.28	4	0.07	-3.64	+0.33
Ni I	2.19	3	0.02	-4.06	-0.07
Zn I	1.29	1	-	-3.41	+0.70
Sr II	-0.72	2	0.06	-3.64	+0.33
Y II	-1.80	3	0.11	-4.04	-0.07
Ba II	-2.42	4	0.10	-4.55	-0.58
Sm II	<-1.82	1	-	<-2.83	<+1.14
Eu II	<-3.42	1	-	<-3.93	<+0.04

  
3.2.1 CNO abundances and the ¹²C/¹³C ratio

The C and N abundances are based on spectrum synthesis of molecular features due to CH and CN. In cool giants, a significant amount of CN and CO molecules are formed and thus, in principle, the C, N, and O abundances cannot be determined independently. However, since CS 22949-037 is relatively warm, little CO is formed, and the abundance of C is not greatly dependent on the O abundance. Nevertheless, we have determined the C, N and O abundances by successive iterations and, in particular, the final iteration has been performed with a model that takes the observed anomalous abundances of these species into account in a self-consistent manner, notably in the opacity calculations.

Carbon and nitrogen

The carbon abundance of CS 22949-037 has been deduced from the $A^{2}\Delta - X^{2}\Pi$ G band of CH (bandhead at 4323 Å), and the nitrogen abundance from the $B^{2}\Sigma - X^{2}\Sigma$ CN violet system (bandhead at 3883 Å). Neither the $A^{3}\Pi_{\rm g} - X^{3}\Pi_{\rm u}$ C₂ Swan band nor the $A^{2}\Pi- X^{2}\Sigma$ red CN band, which are often used for abundance determinations, are visible in this star. Line lists for ${\rm ^{12}CH}$, ${\rm ^{13}CH}$, ${\rm ^{12}C^{14}N}$, and ${\rm ^{13}C^{14}N}$ were included in the synthesis. The CN line lists were prepared in a similar manner as the TiO line lists of Plez (1998), using data from Cerny et al. (1978), Kotlar et al. (1980), Larsson et al. (1983), Bauschlicher et al. (1988), Ito et al. (1988), Prasad & Bernath (1992), Prasad et al. (1992), and Rehfuss et al. (1992). Programs by Kotlar were used to compute wavenumbers of transitions in the red bands studied by Kotlar et al. (1980). For CH, the LIFBASE program of Luque & Crosley (1999) was used to compute line positions and gf-values. Excitation energies and isotopic shifts were taken from the line list of Jörgensen et al. (1996), as LIFBASE only provides line positions for ¹²CH. This procedure yielded a good fit of the CH lines, except for a very few lines which were removed from the list. Figure 2 shows the fit of the CN blue system.

Figure 2: Comparison of the observed spectrum (crosses) and synthetic spectra (thin lines) computed for $\rm [C/H] = -2.80$, and $\rm [N/Fe] = 2.56$ and +2.26., respectively.

Figure 3: Comparison of the observed spectrum (crosses) and synthetic profiles (thin lines) for $A^{2}\Delta - X^{2}\Pi$ ${\rm ^{12}CH}$ and ${\rm ^{13}CH}$lines computed for $^{12}\rm C/^{13}C= 4$ and 10, and with no ${\rm ^{13}C}$.

Our analysis confirms the large overabundances of carbon and nitrogen in CS 22949-037 ( $\rm [C/Fe] = +1.17 \pm 0.1$, $\rm [N/Fe] = +2.56 \pm 0.2$, see Table 3 and Fig. 2), in good agreement with the results of Norris et al. (2001). But as an important new result, we have also been able to measure the ${\rm ^{12}C/^{13}C~}$ ratio from ${\rm ^{13}CH}$ lines of both the $A^{2}\Delta - X^{2}\Pi$ and $B^{2}\Sigma - X^{2}\Pi$ systems. Using a total of 9 lines of ${\rm ^{13}CH}$ (6 from the $A^{2}\Delta - X^{2}\Pi$ system and 3 from the $B^{2}\Sigma - X^{2}\Pi$ system), we find ${\rm ^{12}C/^{13}C~}= 4.0\pm2$ (Fig. 3). We note that the wavelengths for the ${\rm ^{13}CH}$ lines arising from the $A^{2}\Delta - X^{2}\Pi$ system were systematically $\sim$0.2 Å larger than observed, so the wavelengths for the whole set of lines was corrected by this amount.

Our derived ${\rm ^{12}C/^{13}C~}$ ratio is much smaller than that found in metal-poor dwarfs: ${\rm ^{12}C/^{13}C~}= 40$ (Gratton et al. 2000), and is close to the equilibrium isotope ratio reached in the CN cycle ( ${\rm ^{12}C/^{13}C~}= 3.0$). Note also that the observed ¹³C/¹⁴N ratio (assuming that the N abundance is solely ¹⁴N) is about 0.03, close to the equilibirum value of 0.01 (Arnould et al. 1999). We conclude that the initial CNO abundances of CS 22949-037 have probably been modified by material processed in the equilibrium CN-cycle operating in the interior of the star, and later mixed into the envelope.

The very large nitrogen abundance of CS 22949-037 ( $\rm [N/Fe] = +2.6$) has been recently confirmed by Norris et al. (2002) from the NH band at 336-337 nm. Such large overabundances of nitrogen have previously been noted in two other carbon-enhanced metal-poor stars, CS 22947-028 and CS 22949-034 (Hill et al. 2000); ( $\rm [N/Fe] =+1.8$ and $\rm [N/Fe] =+2.3$, respectively), but in these stars the carbon overabundance was much larger than in CS 22949-037 ( $\rm [C/Fe] =+2.0$ and $\rm [C/Fe] =+2.5$, respectively). Norris et al. (1997a) also found a very nitrogen-rich star, CS 22957-027, with $\rm [N/Fe] =+2$. Note, however, that Bonifacio et al. (1998) obtained $\rm [N/Fe] =+1$ for this star, but this discrepancy can be probably accounted for by differences in oscillator strenghs adopted for the CN band.

Oxygen

The most significant result of this study is that we have detected the forbidden [O I] line at 630.031 nm - the first time that the oxygen abundance has been measured in a star as metal deficient as CS 22949-037. The equivalent width of this feature, measured directly from the spectrum, is about 6 mÅ. The [O I] line occurs in a wavelength range plagued by telluric bands of O₂, but the radial velocity of CS 22949-037 shifts the oxygen line to a location that is far away from the strongest telluric lines in all our spectra. Moreover, the position of the [O I] line relative to the telluric lines is different in the spectra obtained in August 2000 and September 2001 (the heliocentric correction varies from 3 to 16 km s^-1), which provides valuable redundancy in our analysis.

In the August 2000 spectra a weak telluric H₂O line ( $\lambda=629.726$ nm) is superimposed on the stellar [O I] line (Fig. 4). We have accounted for this H₂O line in two different ways. First, we have estimated its intensity using that of another line from the same molecular band system (R1 113), a feature observed at 629.465 nm, which should be twice as strong as the line at 629.726 nm. Secondly, we have observed the spectrum of a blue star just before that of CS 22949-037, and at about the same airmass. Figure 5 shows the spectrum of the comparison star, and Fig. 6 the result of dividing our spectra of CS 22949-037 by it.

Figure 4: Mean spectrum of CS 22949-037 from August 2000. The two telluric H2O lines are indicated (shifted in wavelength by about 3 Å due to the radial velocity of the star).

Figure 5: Mean spectrum of the comparison star HR 5881 (A0 V, $v\sin$ i = 87 km s-1), observed just before CS 22949-037.

Figure 6: The spectrum of CS 22949-037 divided by that of HR 5881 to eliminate the telluric lines (heavy line: mean of the three spectra obtained in August 2000 spectrum; thin line: September 2001 spectrum). The scale of both axes is the same as in Fig. 4. The measured equivalent width of the corrected [O I] line is 5 mÅ.

From the telluric line at 629.465 nm in the spectrum of CS 22949-037 (Fig. 4), we estimate the equivalent width of the line at 629.726 nm to be $\sim$1.5 mÅ. The equivalent width of the [O I] line itself should thus be about $4.5\pm1.5$ mÅ. From Fig. 6, the equivalent width of the O I line is 5 mÅ.

In our September 2001 spectrum of CS 22949-037 (Fig. 6) the telluric H₂O line falls outside the region of the stellar [O I] line, but the S/N ratio of that spectrum is much lower (Table 1). The measured equivalent width of the [O I] line from this spectrum is $5\pm2$ mÅ, again in good agreement with the previous result. Our final value for the equivalent width of the [O I] line at 630.031 nm is then $5.0\pm1.0$ mÅ, corresponding to $\rm [O/H] =-1.98\pm 0.1$ and $\rm [O/Fe] =+1.99 \pm 0.1$.

Figure 7: Oxygen abundance from the [O I] forbidden line in extremely metal-poor stars: CS 22949-037 (black dot, this paper); measured values (open circles) and upper limits (arrows) in an extended sample (paper in preparation). At low metallicities, the ratio [O/Fe] does not seem to depend on [Fe/H], and is about +0.65. CS 22949-037 is thus very oxygen rich compared to "normal" extremely metal-poor stars. The error bar in the upper right corner is the typical uncertainty associated with the abundance determination in the sample. CS 22949-037 is plotted with its own associated uncertainties.

According to Kiselman (2001) and Lambert (2002), the [O I] feature is not significantly affected by non-LTE effects because (a) the line is weak, (b) the transition is a forbidden one (with collisional rates largely dominating over radiative rates), and (c) the upper level is collisionally excited. Oxygen abundances derived from the [O I] line are therefore less prone to systematic errors, but the line is very weak in metal-poor stars, hence high resolution and S/N ratio are both required.

Our result that $\rm [O/Fe] \approx +2.0$ in CS 22949-037, the most metal-poor star with a measured O abundance, raises the question whether such a large overabundance of O is representative of the most metal-poor stars in general. This would be an argument in favour of a continued increase of [O/Fe] at the lowest metallicities (e.g., Israelian et al. 2001a, 2001b).

Our VLT programme includes a sample of XMP giants which were observed and analysed in exactly the same way as CS 22949-037 (Depagne et al., in preparation). To provide a meaningful comparison to CS 22949-037, the oxygen abundances derived for this sample (from the [O I] line) are shown in Fig. 7. We stress once again that these abundance measurements have been obtained using exactly the same analysis as described in this paper, applied to 11 stars with effective temperatures between 4700 K and 4900 K, surface gravities between 0.8 to 1.8 dex, and metallicities in the range -2.6 to -3.8. The comparison between the oxygen abundance in CS 22949-037 and the rest of the sample is therefore straighforward, free from systematics arising from the method itself (e.g., determination of stellar parameters, atmospheric models, choice of oxygen indicator). In the metallicity range $-3.5 \le {\rm [Fe/H]} \le -2.5$ we find a mean $\rm [O/Fe]\approx +0.65$, with surprisingly little scatter. We therefore expect that at $\rm [Fe/H] =-4.0$, the mean oxygen abundance should also be around $\rm [O/Fe]\approx +0.65$, as is also hinted at by the upper limit obtained on the $\rm [Fe/H] = -3.8$ giant plotted in Fig. 7. (This point will be discussed in detail, and with a larger sample of stars, in a subsequent paper in this series). Note that CS 22949-037 is the only one of the giants in our program with $\rm [Fe/H]\approx-4.0$ in which we could detect the [O I] line: for $\rm [O/Fe] =+0.65$ the predicted equivalent width is $\approx$0.2 mÅ, well below the normal detection threshold at this wavelength (0.5-1.0 mÅ, depending on the S/N of the spectra).

We thus conclude that the O abundance in CS 22949-037 is not typical for XMP stars; [O/Fe] appears to be $\sim$1.3 dex higher than the expected abundance ratio for stars of this low metallicity (Fig. 7).

We return to the discussion of the origin of these remarkable CNO abundances in Sect. 4.

  
3.2.2 The $\alpha$ elements Mg, Si, Ca, and Ti

In Fig. 8 we compare the light-element abundances of CS 22949-037 with those of the well-established XMP giants (the "classical'' sample) studied by Norris et al. (2001).

The Si abundance in CS 22949-037 has been determined from two lines at 390.553 nm and 410.274 nm. The first is severely blended by a CH feature, while the latter falls in the wing of the H$_{\delta}$ line. These blends have been taken into account in the analysis, and both lines yield similar abundances. The 869.5 nm and 866.8 nm lines of sulphur are not detected, and yield a fairly mild upper limit of $\rm [S/Fe]\leq +1.78$.

The even-Z ($\alpha$-) elements Mg, Si, Ca, and Ti are expected to be mainly produced during hydrostatic burning in stars, and are generally observed to be mildly enhanced in metal-poor halo stars. It is thus remarkable that, in CS 22949-037, the magnitude of the $\alpha$-enhancement decreases with the atomic number of the element: [Mg/Fe] is far greater than normal, while the enhancement of the heaviest $\alpha$-elements, like Ca and Ti, is practically the same as in the classical metal-poor sample (a point noted as well by Norris et al. 2001). The [Si/Fe] ratio has an intermediate value.

  
3.2.3 The light odd-Z elements Na, Al, and K

The abundances of the odd-Z elements Na, Al, and K in XMP stars are all derived from resonance lines that are sensitive to non-LTE effects. The Al abundance is based on the resonance doublet at 394.4 and 396.2 nm. Due to the high resolution and S/N of our spectra, both lines can be used, and the CH contribution to the Al I 394.4 nm line is easily taken into account. The Na I D lines are used for the Na abundance determination, while the K resonance doublet is not detected in CS 22949-037.

Derived [Na/Fe] and [Al/Fe] ratios are usually underabundant in XMP field stars (Fig. 8), while we find [K/Fe] to be generally overabundant in the classical XMP sample (Depagne et al., in preparation), in agreement with Takeda et al. (2002). As found already by McWilliam et al. (1995), Na is strongly enhanced in CS 22949-037 ( $\rm [Na/Fe] =+2.08$), while Al is less deficient than normal: $\rm [Al/Fe] =-0.13$ in CS 22949-037, while the mean value for the comparison sample is $\rm [Al/Fe] =-0.8$. The K doublet is undetected in both CS 22949-037 and CD -38 245, corresponding to upper limits of $\rm [K/Fe]\leq -0.1$ and $\rm [K/Fe]\leq -0.14$, respectively.

Several authors have pointed out that, in LTE analyses, the Na abundance may be overestimated (Baumüller et al. 1998), and the Al and K abundances underestimated (Baumüller & Gehren 1997; Ivanova & Shimanskii 2000; Norris et al. 2001). Following Baumüller et al. (1998) and Baumüller & Gehren (1997), for dwarfs with $\rm [Fe/H] =-3.0$, LTE analysis leads to an offset of $\Delta {\rm [Al/H] \approx -0.65}$ and $\Delta {\rm [Na/H] \approx 0.6}$. Under the hypothesis of LTE, Al and Na behave similarly in dwarfs and giants, and as a first approximation we can therefore assume that the correction is the same for giants as for dwarfs (Norris et al. 2001).

However, the atmospheric parameters of all the stars shown in Fig. 8 are very similar ( ${4850<T_{\rm eff}<5050}$ K, $1.7<{\rm log}~g<2$, and $\rm [Fe/H]\approx -4$), so the NLTE effects must also be very similar for all four stars. Accordingly, the difference between the Na, Al, and K abundances in CS 22949-037 and in the classical metal-poor sample must be real and independent of the non-LTE effects. Figure 9 shows the differences between the light-element abundances in CS 22949-037 and the mean of the three XMP giants CD-38 245, CS 22172-02, and CS 22885-96. For potassium, the upper limit derived above for CD-38 245 was taken as the best approximation to the K abundance of this star when forming the mean. Figure 9 highlights the dramatic decrease in the enhancement of the light elements from Na through Si in CS 22949-037; beyond silicon, the abundance ratios in CS 22949-037 are similar to those in other XMP stars.

Figure 8: Abundance of the elements from C to Zn in CS 22949-037 (filled circles) and in classical XMP stars with $\rm [Fe/H]\approx -4$ (open symbols). N, Mg, and Al are strongly enhanced in CS 22949-037, while the behaviour of the subsequent elements is normal.

Figure 9: ${\rm [X/Fe]_{22949-037} - [X/Fe]_{XMP}}$, where "XMP'' represents the mean abundances of the light elements of the three XMP giants CD -38$^{\circ }$245, CS 22172-02, and CS 22885-96.

  
3.2.4 The Si-burning elements

The distribution of the abundances of the Si-burning elements in CS 22949-037 is different from that observed in the Sun, but is rather similar to the distribution observed in three other very metal-poor stars (see Fig. 8). Furthermore, this pattern is rather well represented by the Z35C zero-metal supernova yields obtained by Woosley & Weaver (1995), as modified by Woosley and Heger (model Z35Z, A. Heger, private communication). Note that Cr and Mn are underabundant while Co and Zn are overabundant. The present analysis of CS 22949-037 is the first case in which a Zn abundance has been derived in such an extremely metal-poor star. We find that $\rm [Zn/Fe] = +0.7\pm0.1$, in agreement with the increasing trend suggested in other very metal-poor stars (e.g., Primas et al. 2000; Blake et al. 2001), but of course it should be kept in mind that this star may not reflect the general behavior of the most metal-deficient stars.

  
3.2.5 The neutron-capture elements

CS 22949-037 is a carbon-rich XMP giant, and the neutron-capture elements are often (though not always) enhanced in such stars (e.g., Hill et al. 2000). Sr is indeed enhanced in CS 22949-037 ([Sr/Fe] =+0.33), but the [Y/Fe] ratio is about solar ([Y/Fe] = -0.07). Ba is underabundant relative to iron ([Ba/Fe] =-0.58) while Sm and Eu are not detected at all ([Eu/Fe] $\leq 0.04$). In seeking an explanation for the origin of this pattern, we compare the distribution of heavy elements in CS 22949-037 to three well-studied groups of stars: (i) the classical XMP sample; (ii) the so-called CH stars, a well-known class of carbon-rich metal-poor stars, and (iii) the new class of mildly carbon-rich metal-poor stars without neutron- capture excess (Aoki et al. 2002). We have excluded the r-process enhanced XMP star CS 22892-052, as its mild carbon enhancement ([C/Fe] $\approx +1$) appears to be unique among the presently known examples of this class.

(i) The classical XMP stars. In the three classical XMP giants (Norris et al. 2001), both Sr and Ba are very deficient with respect to iron, and by about the same factor: $\rm [Sr/Fe]_{mean}= -1.16$, $\rm [Ba/Fe]_{mean}= -1.15$ and $\rm [Ba/Sr]_{mean}\approx 0.0$. In CS 22949-037, Sr is actually enhanced and $\rm [Ba/Sr] = -0.91$, which is a rather different pattern.

(ii) The CH stars.

As a group, the CH stars are moderately metal-poor ( $\rm [Fe/H] = -1.5$) yet strongly enriched in neutron-capture elements; most of them presumably formed as the result of the s-process inside an AGB star (Vanture 1992). The CH stars are members of long-period binary systems with orbital characteristics consistent with the presence of a fainter companion, and it is generally assumed that the abundance anomalies in these stars are the result of mass transfer from the AGB companion, which has now evolved into a white dwarf. At least one very metal-poor star is known to share many of these properties (LP 625-44; Norris et al. 1997a), and it too is a member of a binary system. More recently, Aoki et al. (2000) have confirmed the earlier work that suggested the level of [Sr/Ba] increases with atomic number (at contrast with our star), as expected from an s-process at high neutron exposure (e.g. Wallerstein et al. 1997; Gallino et al. 1998). For example, in some of these very metal-poor CH stars, ²⁰⁸Pb is extremely overabundant (Van Eck et al. 2001); this element is not even detected in our star. In CS 22949-037 the situation is the opposite: the heavy-element abundances decrease with atomic number and even turn into a deficit ( $\rm [Sr/Fe] = +0.33$, $\rm [Y/Fe] =-0.07$, and [Ba/Fe] =-0.58). Accordingly, the neutron-exposure processes in CS 22949-037 and the CH stars appear to have been completely different. Moreover, there is no indication that CS 22949-037 belongs to a binary system (cf. Table 1 and Sect. 2), but since the CH stars are generally long-period, low-amplitude binaries we cannot exclude that our star has been enriched by a companion. Accurate longer-term monitoring of the radial velocity of CS 22949-037 will be needed to settle this point.

(iii) The carbon-rich metal-poor stars without neutron-capture excess

Norris et al. (1997b), Bonifacio et al. (1998) and Aoki et al. (2002) observed five very metal-poor, carbon-rich giants ( ${\rm -3.4 <[Fe/H]< -2.7}$) without neutron-capture element excess. Like CS 22949-037, these stars exhibit carbon excesses of [C/Fe] $\approx +1$, but ${\rm ^{12}C/^{13}C~}\approx 10$. the nitrogen excesses in these stars is much smaller ( ${\rm0.0 <[N/Fe]< 1.2}$), and the neutron-capture abundances are about the same as in normal halo giants with $\rm [Fe/H]\approx -3$. Sr and Ba are underabundant, but the [Ba/Sr] ratio is close to solar, again completely unlike CS 22949-037. Aoki et al. (2002) have suggested that this class of stars could be the result of helium shell flashes near the base of the AGB in very low-metallicity, low-mass stars, but this hypothesis is not yet confirmed.

In summary, the detailed abundance patterns in CS 22949-037 appear to require a different origin from that of the currently known groups of carbon-rich XMP giants. This point is discussed further in the next section.