A detailed geochemical study is presented of the uppermost Critical Zone, especially of the footwall and hanging wall to the Merensky Reef, at Impala Platinum Mines in the Bushveld Complex. The approximately 100 m-thick sequence below the Merensky Reef consists of 13 distinct layers which have sharp boundaries. They are adcumulates with varying proportions of cumulus plagioclase, orthopyroxene and chromite.

Experimental studies on the composition of coexisting orthopyroxene liquid indicate that the magma which produced this sequence contained between 4 and 6% MgO. The magma from which the Merensky Reef formed was more evolved than the footwall magma.

Significant variations exist for both the En content of orthopyroxene and mg# number of whole-rock analyses in short vertical sections. Pyroxenite and norite always have higher values than anorthosite. Extremely sharp breaks in these values correlate with changes in modal proportions, and argue against both significant fractionation within the studied interval, and infiltration metasomatism. Quantitative modelling shows that the entire footwall section could have contained pyroxene with a uniform primary composition of En82, and that all the variation now observed reflects the effect of reaction with trapped magma.

Two independent methods for determining the proportion of trapped liquid are presented, based on mg# number and incompatible element abundances. Both yield a uniform proportion in all samples of approximately 10%. Immiscible sulphide liquid from the Merensky Reef can be shown to have infiltrated downwards for <5 m, despite its high density contrast with silicate magma, very low viscosity and low crystallization temperature. Residual silicate magma would have had even more restricted mobility. The migration of residual liquid or fluid through pothole structures in the floor of the Merensky Reef is not supported by the present data.

The Upper Critical Zone of the Bushveld Complex consists of several cyclic units, each ranging from ultramafic to leucocratic. One view is that they were initiated by addition of, and crystallization from, relatively magnesian magma. Experimental studies on plausible parental magma compositions to this sequence show that plagioclase joins orthopyroxene in the crystallization sequence once the Mg# of the orthopyroxene has decreased below 83. Some norites from the Upper Critical Zone contain orthopyroxene with this Mg#, supporting the validity of the experimental studies. There are no orthopyroxene compositions with higher Mg# than 83 in the entire Upper Critical Zone in either the eastern or western limbs of the Bushveld Complex. This observation suggests that all these rocks formed from plagioclase-saturated magmas. If this interpretation is correct, pyroxenites in this succession are not the result of crystallization from a magnesian magma. Instead, they result from the mechanical separation of plagioclase and orthopyroxene, probably due to crystal sorting during settling, from a magma lying at the cotectic for these two minerals. Rocks exist in the Upper Critical Zone that contain non-cotectic proportions of cumulus plagioclase and orthopyroxene, again supporting models of crystal sorting during settling. In such a model, anorthosites result from the delayed accumulation of plagioclase relative to pyroxene, and not to formation from a magma saturated only in plagioclase.

When traced from northwest to southeast in the western limb, there is a change in the relative proportions of plagioclase to orthopyroxene in the Upper Group 2 chromitite cyclic unit. In the northwest the unit is dominated by ultramafic rocks. In the southeast the plagioclase to pyroxene ratio exceeds that of the cotectic proportion, relations that may result from the lateral transport of suspended plagioclase grains to the southeast.