Extremely metal-poor stars .2. Elemental abundances and the early chemical enrichment of the galaxy

We have obtained high-resolution spectra of 23 very metal-poor stars and present an abundance analysis for 19 of these for elements between Mg and Eu. The sample comprises roughly equal numbers of dwarfs and giants. All stars have [Fe/H] < -2.5, and 10 have [Fe/H] < -3.0. In addition, for six stars with [Fe/H] < -3.0, we compile equivalent widths from the literature (including our own studies) and recompute abundances. Possible errors in the stellar atmospheric models are discussed in detail. Hyperfine-structure corrections are presented for Mn and Co. We use robust techniques to delineate the main trends in the [X/Fe] versus [Fe/H] plots and compare these with the Galactic chemical evolution computations of Timmes, Woosley, & Weaver. The main results are as follows: The lowest abundance we derive for a previously unobserved star is [Fe/H] = -3.57, for CS 22172-002. There are now six stars with [Fe/H] < -3.50 as determined from high-resolution analyses. The alpha-elements Mg, Si, Ca, and Ti possess almost uniform overabundances (relative to iron) down to at least [Fe/H] = -4, the current limit of observations. [Mg/Fe] increases slightly at [Fe/H] < -2.5, but the slope is only -0.15 dex dex-l and may be due to systematic errors. Ti behaves like the other ct-elements, contrary to stellar nucleosynthetic calculations. Stars with [Fe/H] < -2.5 reach a plateau in [Al/Fe] similar or equal to -O.8 that extends at least down to [Fe/H] = -4, but Baumueller & Gehren have advised of the need for +0.5 dex non-LTE corrections in the dwarfs, leading to a revised plateau nearer [Al/Fe] similar or equal to -0.3. This level is qualitatively consistent with an odd-even effect for Mg and Al, and in quantitative agreement with Timmes et al.'s Galactic chemical evolution model. We confirm the underabundance of [Cr/Fe] and [Mn/Fe], and the overabundance of [Co/Fe], in stars with [Fe/H] < -2.5 that was highlighted by McWilliam et al. In addition, we point to a mild overabundance in [Ni/Fe], though this may be dominated by a few stars that have higher than normal [Ni/Fe] but that do not reflect an overall trend. Like CS 22949 - 037, discussed by McWilliam et al., CS 22876-032 shows real variations in yields, with Al and Mg produced in their normal ratios to one another but with underabundances in [Si/Fe], [Ca/Fe] and possibly [Sc/Fe]. It is clear from several elements that real star-to-star abundance differences are common at the lowest metallicities. Sr is measured in most of our stars, but Y and Ba are generally strong enough only in the giants. [Sr/Fe] shows a spread greater than 2 dex (i.e., by a factor in excess of 100) at [Fe/H] < -3, greatly exceeding reasonable errors in the measurement. This spread exists for both dwarfs and giants. [Ba/Fe] decreases as [Fe/H] falls below -2.0, and it exhibits less scatter than [Sr/Fe]. A small number of stars with [BaFe] > 0 deserve further study. CS 22897 - 008 has high Sr, Y, and C abundances for its [Fe/H] but normal Ba. This signature may have arisen from the weak s-process in M > 15 M. stars or by r-processing. By combining an analytic description of gaseous supernova remnants with supernova yields, we show that enrichment of the interstellar medium is influenced more by supernova physics (explosive energy) than by environmental conditions (cloud density). If supernova iron-peak yields are correlated with explosion energy, we can accommodate the well-defined abundance trends with a chaotic picture for halo formation involving independently evolving clouds, as was envisaged by Searle & Zinn. We calculate that a typical enrichment in the protohalo will produce [Fe/H] = -2.7. This coincides with larger abundance variations in held stars of lower metallicity and the lower abundance limit for Galactic globular clusters.