2002ApJS..139..219J

We report the abundances of 30 elements in 23 metal-poor ([Fe/H]←1.7) giants. These are based on 7774 equivalent widths and spectral synthesis of 229 additional lines. Hyperfine splitting is taken into account when appropriate. Our choice of model atmospheres has the most influence on the accuracy of our abundances. We consider the effect of different model atmospheres on our results. In addition to the random errors in T_eff, logg, and microturbulent velocity, there are several sources of systematic error. These include using T_eff determined from Fe I lines rather than colors, ignoring non-LTE effects on the Fe I/Fe II ionization balance, using models with solar [α/Fe] ratios, and using Kurucz models with overshooting. Of these, only the use of models with solar [α/Fe] ratios had a negligible effect. However, while the absolute abundances can change by greater than 0.10 dex, the relative abundances, especially between closely allied atoms such as the rare earth group, often show only small (less than 0.03 dex) changes. We found that some strong lines of Fe I, Mn I, and Cr I consistently gave lower abundances by ∼0.2 dex, a number larger than the quoted errors in the gf-values. After considering a model with depth-dependent microturbulent velocity and a model with hotter temperatures in the upper layers, we conclude that the latter did a better job of resolving the problem and agreeing with observational evidence for the structure of stars. The error analysis includes the effects of correlation of T_eff, logg, and ξ errors, which is crucial for certain element ratios, such as [Mg/Fe]. The abundances presented here are being analyzed and discussed in a separate series of papers.