Compositional convection in magma chambers is thought to be an important process in the fractionation of liquid from crystals during the differentiation of magmas. It has been tested for in this study by undertaking isothermal crystal growth experiments in a silicate melt at atmospheric pressure in air. The melt used is a synthetic basalt in which iron is replaced by cobalt to minimise redox problems. Co-Mg olivine rims were overgrown on forsteritic olivine seeds cemented to the floor of a 2.4 cm deep alumina crucible. Following quenching and sectioning, glasses were examined optically for colour variations and by EPMA for compositional variations. It had been expected that the colour intensity of the blue glass would diminish in the Co-depleted zone that develops around crystal overgrowths, whereas in fact little difference is normally found, except for a slight fading of colour in glass above the apex of a seed in a few experiments. By contrast EPMA revealed zones up to 50 μm wide around seeds that are depleted in Co and Mg by up to 25 % at the crystal-glass interface and in patches above some crystals. Contour maps of X-ray count-rate data obtained in grids of analytical points show Co- and Mg-depleted glass around the overgrowths and in patches above the highest point of each seed, demonstrating that convection in the melt does occur during growth of individual crystals. As the experiments were carried out in a stable temperature gradient and the crystal seeds had no contact with the melt meniscus, thermal and surface-tensional convection are both eliminated, and the convection is inferred to be caused by a density difference resulting from compositional variation across the chemical boundary layer around a growing crystal. The density difference between the inside and outside of a boundary layer is calculated to be approximately −1%.

Very large volume (>1000 km3 of magma) crystal-rich dacitic ignimbrites that lack pronounced evidence of fractional crystallization or vertical zonation erupt in some continental magmatic arcs (e.g. the Lund Tuff of the Great Basin and the Fish Canyon Tuff of Colorado in western USA). Apparently, their magma chambers were only modestly heterogeneous and not systematically zoned from top to bottom. These ignimbrites have 40 to 50% phenocrysts set in a high-silica rhyolite glass. Mineral assemblages and mineral compositions suggest pre-eruption temperatures were 730 to 820°C and water and oxygen fugacities were relatively high. We have speculated that these very large volume ignimbrites are unzoned because crystallization and convection in slab-shaped magma chambers inhibited separation of crystals from liquids and resulted in a chamber filled with compositionally heterogeneous magma that lacked systematic chemical zonation or strong fractionation. However, many other very large volume silicic ignimbrites are strongly fractionated and may be vertically zoned (e.g. tuffs related to the Yellowstone hotspot). These rhyolitic tuffs typically have few phenocrysts, anhydrous mineral assemblages, low oxygen fugacities, crystallization temperatures of 830 to 1050°C, and a strong imprint of fractional crystallization. Yet these Yellowstone-type rhyolites are derived from chambers 40 to 70 km across which have sill-like shapes (depth/diameter ratios much less than 1). Thus, factors other than chamber shape must be important for establishing the degree of evolution and nature of zonation in silicic magma chambers. Here, the role of crystallinity-dependent viscosity on the evolution of these two types of contrasting magmas is explored. Calculated magma viscosities for the hot, dry, crystal-poor rhyolites are significantly lower than for the cooler, wetter, crystal-rich dacites. Perhaps these hot rhyolites had low enough crystal contents and viscosities to allow efficient crystal–liquid separation, probably by a combination of unhindered crystal-settling, floor crystallization (including compaction), and crystallization on the walls of large chambers. Clean separation of melt from residual solids at their sources may have been promoted by their high temperatures and low viscosities (<104.5 Pa s). In contrast, monotonous dacitic magmas may never have been crystal-free near-liquidus magmas. Their large magma chambers may have developed by progressive growth at a shallow level with repeated input of intermediate to silicic magma. Crystallization of the water-enriched dacitic magmas occurred at lower temperatures (<800 °C) where crystallinity and hence magma viscosity (>106.5 Pa s) were significantly higher. These characteristics inhibited all forms of crystal–liquid separation, hindered development of systematic vertical zonation, and promoted quasi-equilibrium crystallization in small domains within large heterogeneous magma chambers. Eruptions of these crystal-rich dacites may only occur if the roof fails over a growing magma chamber that is becoming increasingly molten.

Plutonic complexes with interlayered mafic and silicic rocks commonly contain layers (1–50 m thick) with a chilled gabbroic base that grades upwards to dioritic or silicic cumulates. Each chilled base records the infusion of new basaltic magma into the chamber. Some layers preserve a record of double-diffusive convection with hotter, denser mafic magma beneath silicic magma. Processes of hybridisation include mechanical mixing of crystals and selective exchange of H2O, alkalis and isotopes. These effects are convected away from the boundary into the interiors of both magmas. Fractional crystallisation aad replenishment of the mafic magma can also generate intermediate magma layers highly enriched in incompatible elements.

Basaltic infusions into silicic magma chambers can significantly affect the thermal and chemical character of resident granitic magmas in shallow level chambers. In one Maine pluton, they converted resident I-type granitic magma into A-type granite and, in another, they produced a low-K (trondhjemitic) magma layer beneath normal granitic magma. If comparable interactions occur at deeper crustal levels, selective thermal, chemical and isotopic exchange should probably be even more effective. Because the mafic magmas crystallise first and relatively rapidly, silicic magmas that rise away from deep composite chambers may show little direct evidence (e.g. enclaves) of their prior involvement with mafic magma.

When basaltic magma is emplaced into continental crust, melting and generation of granitic magma can occur. We present experimental and theoretical investigations of the fluid dynamical and heat transfer processes at the roof and floor of a basaltic sill in which the wall rocks melt. At the floor, relatively low density crustal melt rises and mixes into the overlying magma, which would form hybrid andesitic magma. Below the roof the low-density melt forms a stable layer with negligible mixing between it and the underlying hotter, denser magma. Our calculations applied to basaltic sills in hot crust predict that sills from 10-1500 m thick require only 2-200 years to solidify, during which time large volumes of overlying layers of convecting silicic magma are formed. These time scales are very short compared with the lifetimes of large silicic magma systems of around 106 years, and also with the time scale of 107 years for thermal relaxation of the continental crust. An important feature of the process is that crystallisation and melting occur simultaneously, though in different spots of the source region. The granitic magmas formed are thus a mixture of igneous phenocrysts and lesser amounts of restite crystals. Several features of either plutonic or volcanic silicic systems can be explained without requiring large, high-level, long-lived magma chambers.