The geochemical investigation of equilibrium and disequilibrium melting processes accompanying fractionation of partial melts from residual crystal matrices with varying rates were studied using a powerful centrifuge and high-temperature furnace assembly. A series of centrifuge experiments on partially molten fine-grained pyroxene gabbro have been performed at different centrifuge accelerations. Initial samples of gabbro consisting of 60-65 vol% of plagioclase (Pl) - An 90-95, 25-30 vol% augite (Cpx), 5-7 vol% of titanomagnetite (Ti-Mt), <1 vol% of olivine (Ol) have been partially annealed at 1235 and 1290°C in air for 24-30 h. The resulting initial samples of partially molten gabbro contained 55-60 vol% (1235°C) and 65-70 vol% (1290°C) of partial melt plus Pl and Ti-Mt crystals (Px and Ol absent). Cylindrical samples were loaded in Al2O3 capsules (inner diameter 5 mm, height 20 mm) and centrifuged for 2 h at accelerations (a) 100, 200, 500, and 1000 g at 1235°C, and (b) 1000 g at 1290°C. The chemical compositions of fractionated liquids and crystals were determined by electron microprobe. The local proportions of melt and crystals were determined from image analysis of thin sections.
In case (a) the upper layer consisted under all conditions of Pl-crystals + melt, whereas the mode of the bottom layer, containing both Pl and Ti-Mt, is dependent on acceleration. At low centrifuge accelerations (100-200 g) the bottom layer is almost devoid of Pl whereas at higher accelerations the lower layer contains a significant proportion of Pl-crystals buried below the Pl-free cap of Ti-Mt layer. The calculated density contrast for the results of the slow fractionation rate is larger than for the results of fast centrifuging. The bottom layer of heavy crystals under conditions of slow fractionation has a significant chemical and density gradient. Under conditions of fast fractionation the density contrast is smeared out and the overall density gradient is negligible. In case (b) at 1290°C two layers of liquid of different compositions and densities, a layer of the Pl-crystals between them, and a layer of Ti-Mt-crystals on the bottom were observed. The top layer of melt, with a density lower than the density of Pl-crystals, overlies the layer of Pl-crystals. Between the layer of Ti-Mt on the bottom and the layer of Pl in the center of the capsule a layer of melt with significant gradients in chemical composition and density was observed.
The observed acceleration- and temperature-dependence of modal layering results from the relative contributions of collective settling/floating of crystals, as well as from their individual setting/floating according to size and composition. The higher the acceleration, the larger the difference in time scales for the fractionation of small and large crystals and as a result of this a density current of heavy Ti-Mt crystals is more efficient locking the smaller-sized Pl-crystals in a cumulus layer. In our experiments the centrifuging results in a stable density gradient of melt, implying that convection currents in the melt have been avoided. Fast fractionation and melting at 1290°C results in melt segregation in two distinct liquids which may be explained in terms of the continuous withdrawal model for a two component eutectic system.