LITHIUM ABUNDANCES OF EXTREMELY METAL-POOR TURNOFF STARS^*

The effective temperatures (T_eff's) of our program stars were determined from profile fits to hydrogen Balmer lines, a technique used as well by Asplund et al. (2006) and Bonifacio et al. (2007). We employed the Hα and Hβ lines (T_eff(Hα) and T_eff(Hβ)). Although Hγ is also covered by our observations, it lies at the edge of the CCD, making continuum rectification difficult, and thus has not been used. The method for continuum rectification of the observed spectra and analysis used follows exactly that described in Barklem et al. (2002), which may be consulted for details. The most important aspects of the analysis are that the synthetic profiles are computed assuming LTE line formation using one-dimensional LTE plane-parallel MARCS models (Asplund et al. 1997), with convection described by mixing length theory with parameters α = 0.5 and y = 0.5. The most important line-broadening mechanisms for the wings are Stark broadening and self-broadening, which are described by calculations of Stehlé & Hutcheon (1999) and Barklem et al. (2000), respectively. The fitting is accomplished by minimization of the χ² statistic, comparing the observed and synthetic profiles. Essentially the same method was used by Asplund et al. (2006), although with some differences in the rectification and profile comparison techniques.

We note that the assumption of LTE for formation of the Balmer line wings in cool stars has recently been shown to be questionable on the basis of theoretical non-LTE (NLTE) calculations (Barklem 2007), the role of hydrogen collisions being a major uncertainty. Those calculations admit that the temperatures from LTE Balmer line wings could be systematically too cool by of order 100 K if hydrogen collisions are inefficient, although LTE is not ruled out. Due to the fact that there is no strong evidence favoring any particular hydrogen collision model, we chose to calculate in LTE as this temperature scale is well studied, and it is computationally most practical. However, we emphasize that LTE is not a safe middle ground, and will lead to temperatures systematically too cool should departures from LTE exist in reality.

Often, Hα is given higher weight than higher series lines such as Hβ for reasons discussed by Fuhrmann et al. (1993). However, in EMP turnoff stars this is no longer obvious since blending by metal lines becomes unimportant, and Hβ becomes in fact almost insensitive to gravity, while Hα is quite gravity sensitive (see Table 4 of Barklem et al. 2002). Further, the calculations by Barklem (2007) suggest that NLTE effects, if they exist, will be largest in Hα. Thus, in combining the temperatures from Hα and Hβ, we have in fact given Hβ double the weight of Hα.

The derived effective temperatures are dependent on the gravity assumed in the calculation. The analysis is iterated for the gravity, which is determined from the analysis of the high-resolution spectrum (Section 3) for each object. The results and uncertainties of the T_eff determinations are listed in Table 2. The uncertainties listed in the tables are estimated from the quality of the spectrum and the fit.

For comparison purposes, we also estimated the effective temperatures from the (V − K)₀ colors (T_eff(V − K)), using the effective temperature scales of Alonso et al. (1996), Ramírez & Meléndez (2005b), and González Hernández & Bonifacio (2009). For the estimates using the scale of Alonso et al. (1996), we assume [Fe/H] =−3.0 for stars having [Fe/H]<−3.0 following Ryan et al. (1999). The photometry data were collected from Beers et al. (2007), the Two Micron All Sky Survey (2MASS) catalog (Skrutskie et al. 2006), and the SIMBAD database. The V magnitudes of the SDSS stars are derived from the g magnitude and g − r color, as in Aoki et al. (2008). Interstellar reddening is estimated from the dust maps of Schlegel et al. (1998) and from the interstellar Na i D line, using the scale of Munari & Zwitter (1997); results are listed in Table 3. We adopt the reddening estimate obtained from the Na D line for the brightest four stars (HD 84937, BD+26°3578, G 64–12, and G 64–37) because these are nearby stars, and the reddening might be overestimated from the dust maps. The interstellar Na D lines blend with the stellar lines in the spectrum of HE 1148–0037, so the reddening value from the dust map is adopted for this object. For the other six objects, the averages of the reddening values derived from the two methods are adopted. The differences between the E(B − V) values based on the two methods are generally 0.01 mag or smaller. The exception is for CS 22965–054, which exhibits the largest E(B − V) among the sample, E(B − V) = 0.09 from the dust map and 0.13 from the Na D line. The extinction in each band is obtained from the reddening relation given by Schlegel et al. (1998).

The photometry data and the derived effective temperatures are listed in Table 3. Figure 2 shows the differences between the effective temperatures from Balmer lines and those from V − K colors for the three temperature scales. The effective temperatures from the scales of Ramírez & Meléndez (2005b) and González Hernández & Bonifacio (2009) are systematically higher than those from Alonso et al. (1996). The differences between the results from González Hernández & Bonifacio (2009) and from Alonso et al. (1996) are 100–150 K. The results from Ramírez & Meléndez (2005b) are 100–200 K higher still than those from González Hernández & Bonifacio (2009) for stars with [Fe/H] <−3, while the agreement is fairly good for the three stars having higher metallicity (shown by smaller symbols in the figure). These differences between the effective temperature scales were already reported by Ramírez & Meléndez (2005b) and González Hernández & Bonifacio (2009).

Zoom In Zoom Out Reset image size

The comparisons with the results from Balmer line analyses indicate that the effective temperatures from the scale of Alonso et al. (1996) agree in general with those from the Balmer lines. The star showing the largest discrepancy is SDSS 0040+16. This star has a (V − K)₀ smaller than the color range (1.1 < (V − K)₀ < 1.6) for which the formulae of Alonso et al. are applicable. The results from the scales of González Hernández & Bonifacio (2009) and Ramírez & Meléndez (2005b) for stars with [Fe/H] <−3 are systematically higher than those from Balmer lines by 150 K and 250 K, respectively.

Determination of the Li abundance is sensitive to the effective temperatures adopted in the analysis. We discuss the impact of effective temperatures on the metallicity dependence of Li abundances in Section 4.

The estimate of effective temperatures for EMP stars from these temperature scales would be rather uncertain, because of the small number of stars in the sample used to produce the scales. Alonso et al. (1996) includes a very small number of stars for the ranges of the color and metallicity. Ramírez & Meléndez (2005a) increased the number of such stars, which are used to produce the scale of Ramírez & Meléndez (2005b). However, the sample is still not large (10 stars) and includes one carbon-enhanced star and two known double-lined spectroscopic binaries. Further measurements of effective temperatures by the infrared flux method are desired for this metallicity range.

The abundance analyses were made using the grid of ATLAS9 NEWODF model atmospheres (Kurucz 1993; Castelli & Kurucz 2003), with enhancement of the α elements. Calculations of synthetic spectra for abundance analyses were made employing a one-dimensional LTE spectral synthesis code that is based on the same assumptions as the model atmosphere program of Tsuji (1978). The surface gravities (log g) are estimated from the effective temperatures and Y² isochrones (Kim et al. 2002) for [Fe/H]  = −2.5 (for HD 84937 and BD+26°3578) and [Fe/H] =−3.5 (for the others) assuming the ages of these stars to be 12 Gyr. For the T_eff range of our sample, two possibilities of log g exist: the subgiant case (log g ∼3.8) and the main-sequence case (log g∼4.4). We performed abundance analyses for Fe i and Fe ii (see below for details), and selected the log g that provides better agreement of iron abundances from the two species for each star. The derived log g value is not sensitive to the assumption of the age for selecting isochrones. The results for log g and the derived iron abundances from the two species are listed in Table 2. We redetermined effective temperature when the gravity assumed in the first estimate of effective temperature is different from the derived one. We note that the Fe abundance from Fe i lines and the Li abundance are not sensitive to the gravity adopted in the analysis.

The analysis, when adopting the microturbulent velocity (v_micro) of 1.5 km s⁻¹, results in no statistically significant dependence of the iron abundances on the strengths of the Fe i lines used for the analysis for BS 16545–089, HE 1148–0037, G 64–12, G 64–37, and BD+26°3578, while 1.2 km s⁻¹ is preferable for HD 84937. Since the number of iron lines available in the abundance analyses of the other four stars is too small to estimate their microturbulent velocity, we assume v_micro of 1.5 km s⁻¹ for these objects. We note that the Li abundance derived from the weak Li i λ 6708 line is insensitive to the microturbulent velocity adopted in the analysis.

In the following discussion, we adopt the iron abundances from Fe i lines (Table 2) as the metallicity. Although the iron abundances from Fe i lines are considered to be possibly affected by NLTE effects (Asplund 2005, and references therein), our results are insensitive to the gravity adopted in the analysis. We refer to a solar Fe abundance of log _⊙(Fe) = 7.45 (Asplund et al. 2005) to derive the [Fe/H] and [X/Fe] values.

In order to investigate correlations between the abundances of Li and other elements, Na, Mg, and Sr abundances are determined by a standard LTE analysis from the Na i 5890 and 5896 Å lines (D lines), Mg i 5172, 5183, and/or 5528 Å lines, and Sr ii 4078 and/or 4215 Å lines, respectively. The agreement of the abundances from two or three lines for each element is fairly good. The results are listed in Table 4. Correlations between the abundances of Li and these elements are examined in Section 4. The Na D lines are known to suffer a significant NLTE effect that is dependent on the line strengths. The effect is, however, not large (Δ[Na/Fe]_NLTE −  [Na/Fe]_LTE ≲  0.1 dex) for the EMP stars that have equivalent widths of the D lines less than 50 mÅ (Takeda et al. 2003; Andrievsky et al. 2007). The effect could be larger (∼−0.3 dex) for the less metal-poor stars whose equivalent widths are as large as 100 mÅ.

The Li abundances are derived from the Li i 6708 Å line by fitting the observed spectra with synthetic spectra. Doppler corrections for the observed spectra are determined by the Fe i line positions used for the above abundance measurements. In the fitting procedure for the four bright stars (HD 84937, BD+26°3578, G 64–12, and G 64–37), a small wavelength shift is included as a fitting parameter (see below). The continuum level is estimated from the wavelength range around the Li line (6706.9–6707.4 Å and 6708.2–6708.7 Å). The list of the Li lines provided by Smith et al. (1998) is adopted, neglecting the contribution of ⁶Li. We assume a Gaussian profile to account for the broadening by macroturbulence, including rotation, and by the instrument. We assume a broadening of 7 km s⁻¹ for the stars observed with R ⩾ 90, 000, and 8 km s⁻¹ for the others, values that sufficiently explain the widths of weak Fe lines. We performed the analyses changing the value by ±1 km s⁻¹, and confirmed that the effect of the assumed line broadening on the derived ⁷Li abundance is small (∼0.01 dex).

We then search for the Li abundance that yields the minimum χ². We found that the fit is slightly improved by allowing a small wavelength shift for the four brightest stars, while the effect is negligible for the others. Hence, in our procedure for fitting of the synthetic spectra to the observations, the free parameter was the Li abundance, and for the four brightest stars also a wavelength shift. The number of data points to which the fitting is applied is 18 and 36 for the cases of two pixel binning and of no binning, respectively. The 2σ range is adopted as the fitting error (σ[A(Li)]_fit in Table 4). While the fitting errors for the four bright stars (G 64–12, G 64–37, HD 84937, and BD+26°3578) are on the order of 0.02 dex and are smaller than the other errors (see below), those for the other fainter stars are 0.06–0.18 dex, depending on the S/N of the spectra.

The equivalent widths of the Li doublet are determined by integrating the flux density of the synthetic spectra calculated for the final Li abundance as mentioned in Section 2. The values are listed in Table 4.

In addition to the above fitting errors, the uncertainties in the adopted continuum levels, macroturbulence, and atmospheric parameters should be considered. The continuum level is estimated for the wavelength range around the Li doublet, in which 25 (in the case of two pixel binning) or 50 (in the case of no binning) data points exist. The uncertainty is 0.1% or smaller for the four bright stars (S/N ≳ 300), while it is approximately 0.5% for the spectra with S/N ∼ 50. The effects of continuum level shifts of 0.1% and 0.5% are estimated to be 0.01 dex and 0.04 dex, respectively, by means of an analysis of the spectrum of G 64–12, in which the continuum level was changed artificially. The errors due to the uncertainty of the macroturbulent velocity are on the order of 0.01 dex, as mentioned above.

The errors due to the uncertainties in atmospheric parameters are estimated by changing the parameters by δ(T_eff) =+100 K, δ(log g) =−0.3 dex, and δ(v_micro) = +0.3 km s⁻¹ for G 64–12. The effect of metallicity changes of 0.2 dex on the abundance analyses is negligible for EMP stars. We confirmed that the effects of changes of log g and v_micro on the derived Li abundance are negligible. The Li abundance changes by +0.07 dex when the effective temperature is changed by +100 K. Such error estimates are scaled for the uncertainty for T_eff for each object listed in Table 2.

The total abundance errors (σ[A(Li)]_tot) are obtained by adding, in quadrature, the fitting errors and the errors due to the uncertainty of the continuum placement, the macroturbulence, and the effective temperature for each star (see Table 4).

Table 5 compares the results of the present study for stars in common with other recent studies. The T_eff and A(Li) of CS 22948–093 determined by this work and by Bonifacio et al. (2007) agree fairly well, while a significant discrepancy of A(Li) is found for CS 22965–054. This is clearly caused by the different effective temperatures adopted in the two analyses.

The abundances of BD+26°3578 were determined by Asplund et al. (2006, where the star is referred to as HD 338529), adopting very similar atmospheric parameters (T_eff, log g) = (6335 K, 4.04) to those of the present work (6340 K, 3.9). Their Fe abundance ([Fe/H] =−2.26 and log (Fe) =5.24) from Fe ii lines as well as the Li abundance agrees well with our result.

The T_eff's of G 64–12 and G 64–37 determined by Boesgaard et al. (2005) are about 200 K lower than ours, which result in their lower A(Li) values. This is, however, partially compensated by the difference of model atmospheres used in the analysis. Boesgaard et al. adopted the grid of Kurucz (1993), which assumes convective overshooting. We found that the resulting A(Li) from this grid is systematically higher by 0.06 dex than that obtained from the no overshooting models by conducting analysis with both models. The discrepancy of A(Li) derived by us and Boesgaard et al. (2005) is explained by the differences of T_eff and the model atmosphere grid. The T_eff of Boesgaard et al. (2005) is spectroscopically determined by the analysis of Fe i lines. A systematic difference between T_eff's from the excitation equilibrium and those from other methods (such as photometric colors) have been reported by previous studies (e.g., Norris et al. 2001; Barklem et al. 2005), which might be attributed to NLTE effects on Fe line formation, although the recent work by Hosford et al. (2009) reported no discrepancy between the T_eff from LTE analyses of Fe i lines and T_eff from the Hα profile analysis.

The T_eff's determined by Ryan et al. (1999) are systematically lower than ours. The differences in T_eff for G 64–12, G 64–37, and CD −24°17504 are smaller than 100 K, while the differences are as large as 200 K for HD 84937 and BD+26°3578. The discrepancy of A(Li)'s between the two studies is basically explained by the differences in the adopted effective temperatures.

The above comparisons demonstrate that the differences of derived Li abundances are due to differences of the effective temperature and/or the model atmosphere grid adopted in the analysis. These effects are systematic, and do not essentially affect the discussion of the slope or scatter of Li abundances if the same technique of effective temperature estimates and similar model atmosphere grids are used.

Figure 3 shows the Li abundances as a function of [Fe/H] for our sample and others.¹² We confirm that the Li abundance of the "reference" star BD+26°3578 ([Fe/H]∼−2.3) determined by our analysis agrees well with the measurement by Asplund et al. (2006). By contrast, stars with [Fe/H] <−3 (the EMP stars) appear to have lower Li abundances on average, and also exhibit some scatter. We discuss this point further below.

Zoom In Zoom Out Reset image size

The average of A(Li) (i.e., 〈A(Li)〉) of the eight stars with [Fe/H]<−3 is 2.03, with a sample standard deviation (σ) of 0.09 dex. The standard deviation is comparable with, or slightly smaller than, the measurement errors of our analysis (0.07–0.23 dex). Hence, we do not detect any significant scatter of the Li abundances in our sample of EMP stars. The 〈A(Li)〉 is lower by 0.24 dex and 0.20 dex than the average of our two reference stars (2.27) and the average of the results of the LTE analysis by Asplund et al. (2006) for six stars in −2.5< [Fe/H] <−2.0 (2.23), respectively. The difference of 0.2 dex is significant, compared to the σN^−1/2 = 0.09/$\sqrt{8}$ = 0.03 dex, where N is the number of objects in the EMP sample. We note that the Li abundances for stars in −2.5< [Fe/H]<−2.0 are well determined by Asplund et al. (2006), and the scatter is small (σ =  0.04 dex and σN^−1/2 = 0.02 dex). We conclude that the Li abundances for stars with [Fe/H]<−3 are 0.2 dex lower than those of stars with higher metallicity on average, while no significant scatter or trend with metallicity is detected in our EMP sample. A similar conclusion was reached by Bonifacio et al. (2007); our new measurements for stars in the lowest metallicity range support their results.

Such a difference can obviously be produced by a decreasing trend (slope) of A(Li) as a function of metallicity. The possible slope of A(Li) was discussed by Bonifacio et al. (2007) in detail. However, no clear physical reason for the slope, which appears only at the lowest metallicity range, has yet been identified. Another possibility is that scatter of A(Li) increases in the range [Fe/H] <−3, and, as a result, the average decreases. This case would be relatively easily explained by depletion of Li, although some reason for the metallicity dependence of the depletion is also required.

If the effective temperatures estimated from the (V − K)₀ colors using the scales of Ramírez & Meléndez (2005b) and González Hernández & Bonifacio (2009) are employed, the Li abundances of our stars would be systematically higher. The effective temperatures derived using the scale of González Hernández & Bonifacio (2009) are systematically higher by 220 K and 150 K than the values from the Balmer line analysis for EMP stars and for stars with [Fe/H]∼−2.3, respectively, resulting in 0.15 dex and 0.10 dex higher Li abundances. Although the difference of Li abundances between the EMP stars and less metal-poor stars becomes slightly smaller than the result derived adopting the effective temperature from the Balmer line analysis, the difference is still significant. By contrast, if the effective temperatures estimated using the scales of Ramírez & Meléndez (2005b) are adopted, the values are 290 K and 150 K higher than those from the Balmer lines, resulting in 0.20 dex and 0.10 dex higher Li abundances. In this case, the difference of Li abundances between the stars with [Fe/H]<−3 and >−2.5 is only 0.1 dex, which is no longer significant compared to our measurement errors.

Although no statistically significant scatter of A(Li) is found for our full EMP sample, within our measurement errors, the existence of some star-to-star differences in A(Li) is suggested. For instance, even if the stars with the largest measurement errors are excluded from the evaluation, a similar scatter of A(Li) remains. A difference of 0.14 dex is found in A(Li) for the two bright stars G 64–12 and G 64–37, as found by Boesgaard et al. (2005) and Nissen et al. (2005). In this subsection, we investigate correlations between A(Li) and the adopted stellar parameters, in order to search for a hint for understanding the lower Li abundances among the EMP stars.

Figure 4 shows A(Li) as a function of T_eff. No clear correlation can be seen in this figure. It should be noted that the random error of the effective temperature propagates into the derived Li abundance, which is represented by the arrows shown in the diagram. An increasing trend of Li abundance with decreasing T_eff is potentially influenced by this error.

Zoom In Zoom Out Reset image size

Our sample includes stars that have already evolved to the subgiant branch. Their effective temperatures during the main-sequence stage should be higher than the current values, and might be as high as the hottest stars among the main-sequence sample. The stars having log g < 4.0 are overplotted by large circles representing candidate subgiants in the figure. If these stars are excluded, we find that stars cooler than 6150 K exhibit higher and almost constant Li abundances, while the warmer stars show some scatter. However, this probably reflects a metallicity effect, as the cooler stars (excluding subgiants) are objects having [Fe/H] > − 2 studied by Asplund et al. (2006), while the warmer stars have [Fe/H] <−2. That is, the sample of very metal-poor stars ([Fe/H] <−2) have T_eff >6150 K, or are subgiants. This could be due to a bias in the sample selection caused by the fact that distant stars must be observed to cover lower metallicity ranges, and intrinsically faint stars are not sampled. Other than this point, no clear correlation appears between the Li abundances and effective temperature, even if subgiants are removed from the plot, or if the highest temperature among the sample (∼6500 K) is assigned for them as their T_eff during their main-sequence phase.

Figure 5 shows Li abundances as functions of [Na/Fe], [Mg/Fe], and [Sr/Fe]. An anticorrelation between the Li abundance and the [Na/Fe] ratio was reported for the globular cluster NGC 6752 by Pasquini et al. (2005). Such a trend is, however, not seen in our sample (Figure 5): the two low Li stars HE 1148–0037 and SDSS 0040+16 have comparatively high [Na/Fe] ratios, while that of the other low Li star, CS 22948–093, is low. If the NLTE effects on the Na abundances are taken into consideration, the abundances of less metal-poor stars become lower by about 0.3 dex. However, this correction does not result in any correlation between the Li and Na abundances.

Zoom In Zoom Out Reset image size

No significant scatter is found for [Mg/Fe] in our sample. Also, no correlation is found between the abundances of Li and the Mg abundances. In contrast, a large scatter of measured Sr abundances exists, which could reflect the contribution of a neutron-capture process that is efficient in the very early Galaxy (Truran et al. 2002; Travaglio et al. 2004; Aoki et al. 2005), as well as the contribution of the main r-process to higher metallicity objects. No correlation is found between the Li and Sr abundances, indicating that the Li production or depletion is not likely to be related to neutron-capture nucleosynthesis processes.

We are also interested in exploring whether the kinematics of the stars with lower A(Li) exhibit any peculiarities that distinguish them from the rest of the stars. For this exercise, we combined our present sample with that of Bonifacio et al. (2007). Proper motions of SDSS stars are listed in the public DR6 database, while those of other stars are taken from the NOMAD database (Zacharias et al. 2004). Distances were estimated from the luminosity classifications given by either the present paper or in Bonifacio et al. (2007), applying the methods described by Beers et al. (2000). Stars with log g > 4.0 were considered dwarfs, those with 3.5 ⩽ log g ⩽ 4.0 were considered main-sequence turnoff stars, and those with log g < 3.5 were considered subgiants. The adopted surface gravities, metallicities, photometry, distances, radial velocities, and proper motions are listed in Table 6. These data were used to derive the full space motions and other orbital information, following the procedures described by Carollo et al. (2007); results are listed in Table 7.

Figure 6 shows the derived rotational velocity with respect to the Galactic center, V_ϕ, as a function of [Fe/H]. The stars with low values of A(Li) are labeled by filled circles. We note that three stars with "normal" A(Li) are on highly prograde orbits (an additional star, CS 31061–0032, with V_ϕ = 205 km s⁻¹, may be a member of the metal-weak thick disk). This is somewhat unexpected, given the essentially zero net rotation of the inner halo, and net retrograde rotation of the outer halo. We plan to study these stars in more detail in the near future.

Zoom In Zoom Out Reset image size

Based on the derived space velocities and orbital parameters, an attempt was made to assign approximate population memberships for these stars, listed in the second column of Table 7. The assignments take into account the values of the velocity ellipsoids derived for the inner and outer halo, and the thick disk (including the metal-weak thick disk; D. Carollo et al. 2009, in preparation), as well as the derived Z_max (the maximum distance from the Galactic plane in the vertical direction) for each star. We did not attempt to assign a membership to the four highly prograde stars, because, as was mentioned above, they require further investigation (FI in the second column of Table 7). Note that in some cases it was not possible to uniquely distinguish a single population assignment, so multiple assignments are given.

Inspection of these assignments indicates no tendency for the stars with low A(Li) to be associated preferentially with either the inner- or outer-halo populations. This is similar to the conclusions drawn by Bonifacio et al. (2007) based on inspection of radial velocities alone. However, the sample size (especially of low A(Li) stars) and the existing errors on the Li abundance determinations preclude a final determination.

Our measurements of the Li doublet for EMP stars based on the effective temperatures from the Balmer lines showed that the Li abundances in the metallicity range [Fe/H] <−3 are lower, on average, than for stars with higher metallicity. The same conclusion is also reached by adopting the effective temperatures from (V − K)₀ using the scales of Alonso et al. (1996) and González Hernández & Bonifacio (2009), although the Li abundances are systematically higher if the latter is adopted. If the temperature scale of Ramírez & Meléndez (2005a) for (V − K)₀ is adopted, the dependence of the Li abundances on metallicity becomes marginal.

Although no scatter or trend of Li abundances is detected within the sample of EMP stars, the metallicity dependence of the Li abundance could be a key to understanding the Li problems found for metal-poor stars. Observations to obtain better quality spectra for such EMP stars, to improve measurements of the Li feature itself, as well as for improved determination of T_eff from the Hα and Hβ profiles, will provide a useful constraint on the possible scenarios proposed to explain the Li problems. Moreover, measurements of Li abundances for even lower metallicity stars ([Fe/H] <−3.5) are also vital. Measurements of Li abundances in this metallicity range have been reported so far only for one system, a double-lined spectroscopic binary (CS 22876–032: Norris et al. 2000; González Hernández et al. 2008). We plan additional investigations of this metallicity range by high-resolution spectroscopic studies of EMP star candidates discovered by the Hamburg/ESO survey and SDSS/SEGUE in the near future.