The frequency of carbon stars among extremely metal-poor stars

We demonstrate that there are systematic scale errors in the [Fe/H] values determined by the Hamburg/ESO Survey (HES; and by inference by the HK survey in the past) for certain extremely metal-poor, highly C-enhanced giants. The consequences of these scale errors are that (1) the fraction of carbon stars at extremely low metallicities has been overestimated in several papers in the recent literature, (2) the number of extremely metal-poor stars known is somewhat lower than has been quoted in the recent literature, and (3) the yield for extremely metal-poor stars by the HES is somewhat lower than is stated in the recent literature. A preliminary estimate for the frequency of carbon stars among the giants in the HES sample with -4 < [Fe/H] < -2.0 dex is 7.4% +/- 2.9%; adding an estimate for the C-enhanced giants with [C/Fe] > 1.0 dex without detectable C-2 bands raises the fraction to 14% +/- 4%. We rely on the results of an extensive set of homogeneous, detailed abundance analyses of stars expected to have [Fe/H] <= -3.0 dex selected from the HES to establish these claims. We have found that the Fe metallicity of the cooler (T-eff less than or similar to 5200 K) C stars as derived from spectra taken with HIRES at Keck are a factor of similar to 10 higher than those obtained via the algorithm used by the HES project to analyze the moderate-resolution follow-up spectra, and this algorithm is identical to that used until very recently by the HK survey. This error in the Fe abundance estimate for C stars arises from a lowering of the emitted flux in the continuum bandpasses of the KP index (Ca II lambda 3933) and, particularly, the HP2 (H delta) index used to estimate [Fe/H] due to absorption from strong molecular bands.