A stochastic model of the chemical enrichment of metal-poor systems by core collapse supernovae is used to study the scatter in stellar abundance ratios .
Large-scale mixing of the enriched material by turbulent motions and cloud collisions in the interstellar medium , and infall of pristine matter are taken into account .
The resulting scatter in abundance ratios , e.g .
as functions of the overall metallicity , is demonstrated to be crucially dependent on the as yet uncertain supernovae yields .
The observed abundance ratios and their scatters therefore have diagnostic power as regards the yields .
The relatively small star-to-star scatter observed in many chemical abundance ratios , e.g .
by Cayrel et al .
( 2004 ) for stars down to [ \mathrm { Fe } / \mathrm { H } ] = -4 , is tentatively explained by the averaging of a large number of contributing supernovae and by the cosmic selection effects favoring contributions from supernovae in a certain mass range for the most metal-poor stars .
The scatter in observed abundances of \alpha -elements is understood in terms of observational errors only , while additional spread in yields or sites of nucleosynthesis may affect the odd-even elements Na and Al .
For the iron-group elements we find some systematic deviations from observations in abundance ratios , such as systematically too high predicted Cr/Fe and Cr/Mg ratios , as well as differences between the different sets of yields , both in terms of predicted abundance ratios and scatter .
The semi-empirical yields recently suggested by Francois et al .
( 2004 ) are found to lead to scatter in abundance ratios significantly greater than observed , when applied in the inhomogeneous models .
” Spurs ” , very narrow sequences in abundance-ratio diagrams , may disclose a single-supernova origin of the elements of the stars on the sequence .
Verification of the existence of such features , called single supernova sequences ( SSSs ) , is challenging .
This will require samples of several hundred stars with abundance ratios observed to accuracies of 0.05 dex or better .