Fluorine, chlorine, zinc, niobium, zirconium, yttrium, and rubidium have been deter-mined on fifteen obsidians from Eburru volcano (Kenya Rift Valley), spanning the range from pantel-leritic trachyte to pantellerite. All pairs of elements show positive correlation coefficients, ranging between 0·769 and 0·998, but with most values better than 0·900. In spite of some very high correlations, only two of the twenty-one best-fit lines pass near the origin of the Cartesian coordinates. Linear distributions are found within two separate groups of elements: F, Zr, Rb; and Cl, Nb, Yt. Zn behaves in general as a member of the second group but seems to be subject to an additional variation. When an element from the fluorine group is plotted against one from the chlorine group the resulting pattern is non-linear. Therefore, although the elements in both groups would generally be considered ‘residual’ (partition coefficients between crystals and liquid approaching zero) there are clearly detectable differences in their variation, and hence their behaviour.

Major-element variations in the obsidians are such that a vapour (fluid) phase would be needed to account for any magma evolution. The trace-element patterns are also impossible by closed-system crystal fractionation and suggest that this fluid may have been rich in halogens, with the metallic elements forming preferred ‘complexes’ with either F or Cl. The F-Zr-Rb ‘complex’ also varies quite independently of the important major oxides (e.g. A12O3) in the rocks. In the case of Rb this is but one aspect of a more significant anomaly, in which there is no sign of any influence of alkali feldspar (which partitions Rb) in the variation. This is remarkable because trachytes and rhyolites have normative ab+or > 50 %, and any evolutionary process controlled by crystal ⇋ liquid interactions must be dominated by the melting or crystallization of alkali feldspar. The results on the Eburru obsidians show that if they are an evolutionary series then either, the process was not crystal ⇋ liquid controlled, or that any such process has been overriden (or buffered) by other processes that have superimposed the observed trace-element patterns. In the latter event, the buffering phase may have been a halogen-bearing vapour.

The same considerations must apply to other pantellerite provinces where Rb appears to have behaved as a ‘residual’ element.