Microstructural evidence for convection in high-silica granite

The separation of crystals and melt, leading to magmatic fractionation, is a key process in the chemical evolution of continental crust (e.g., Lee and Bachmann, 2014; Keller et al., 2015). However, our understanding of the mechanisms involved in crystal-melt segregation comes primarily from microstructural observations from mafic intrusions (e.g., Holness et al., 2007; Namur and Humphreys, 2018) and geochemical studies on volcanic arcs (Gelman et al., 2014; Lee and Morton, 2015; Barnes et al., 2016), both of which account for a volumetrically minor component of continental crust. While progress has been made in identifying the products of crystal-melt segregation in some granitic complexes (Gelman et al., 2014; Lee and Morton, 2015), the internal dynamics of crystallizing silicic intrusions are not well understood. In part, this lack of understanding may be because silicic systems do not behave in the same way as mafic systems. The high viscosity of silicic melt and the similar densities of melt and the main crystallizing phases (quartz and feldspar) make it difficult to separate liquid and solid to form large crystal accumulations in granitic bodies (Martin and Nokes, 1989; Bea, 2010). The fact that highly silicic granitic melt is commonly close to a multiphase eutectic also restricts the formation of modal layering, a key record of the internal dynamics of mafic intrusions (e.g., Holness et al., 2017). We present microstructural evidence from quartz clusters that supports the hypothesis that crystals remain suspended in cooling, liquid-rich bodies of high-silica granite (HSG; >70 wt% SiO₂) during periods of sustained convection.

The Bushveld granites are a suite of anorogenic (A-type) HSGs located near the top of the ca. 2.0 Ga Bushveld Complex, South Africa. This suite primarily consists of sheeted sills that postdate the crystallization of the Rustenberg Suite (the layered mafic complex; Kleemann and Twist, 1989). The sills are dominated by coarse-grained quartz and perthite, with minor hornblende, biotite, and cassiterite. The thicknesses of individual sills are on the order of tens to hundreds of meters (de Beer et al., 1987). One such sill, the Bobbejaankop granite, was the focus of this study (Fig. 1).

The Bobbejaankop granite forms an ~300-m-thick sill exposed in the Zaaiplaats tin mine in the northern lobe of the Bushveld Complex. A historical source of tin ore (cassiterite), the Bobbejaankop sill contains an ~40-m-thick central horizon that is enriched in the incompatible elements tin, tungsten, and copper (Coetzee and Twist, 1989). The enrichment of incompatible elements in a central horizon suggests that the body simultaneously crystallized inward from the floor and roof of the intrusion, a process common in layered mafic intrusions but only seldom observed in granitic sills (e.g., Sawka et al., 1990; Nishimura and Yanagi, 2000).

The crystallization sequence of the six HSG samples used in this study was calculated to predict, and provide context for, the relative timing of microstructure formation. Calculations were made with Theriak–Domino software (de Capitani and Petrakakis, 2010), a Gibbs free-energy minimization program, using the Holland and Powell (1998) thermodynamic database and the activity composition relations employed by Dyck et al. (2020).

We used cathodoluminescence (CL) imaging to study the internal microstructure (i.e., compositional zonation) of quartz grains from Bobbejaankop granite samples, and collected quartz crystal orientation data by electron backscatter diffraction (EBSD). CL and EBSD images were generated on a FEI Quanta 650 scanning electron microscope. To prevent detector saturation, CL images were first acquired under high vacuum with an accelerating voltage of 20 kV, a working distance of 12.5 mm, and 80 μs dwell time. The thin (5 nm) carbon coat used for CL was then removed prior to collecting EBSD data, which were acquired under low-vacuum mode with an accelerating voltage of 30 kV and a working distance of 15 mm. All processing and plotting of EBSD data were carried out using the MTEX toolbox in Matlab.

Following the procedure outlined by Beane and Wiebe (2012), we calculated mean orientations for each quartz grain. We set the minimum misorientation angles for a grain and subgrain boundaries at 10° and 2°, respectively. To account for the presence of growth twinning, quartz-quartz pairs separated by 60° about the c axis (Dauphine twin law) were identified and treated as single grains. We then used processed orientation maps to investigate the orientation relationships between all quartz-quartz pairs.

Detailed studies of the relative orientation of grains in quartz clusters have demonstrated that, in many cases, crystal pairs exhibit coincident-site lattice orientation relationships (CSL-ORs; Vance and Gilreath, 1967; Wiebe et al., 2007; Beane and Wiebe, 2012; Graeter et al., 2015), whereby the crystal lattices of neighboring grains are systematically aligned with a high degree of symmetry across the grain boundary. Notably, CSL-ORs represent the lowest grain-boundary energy configuration. In the context of melt crystallization, the conjoining of two or more grains of the same phase, particularly if those grains are of a similar size, is often a consequence of synneusis (Wieser et al., 2019). Synneusis may result in the formation of CSL-ORs between crystal pairs if at least one of the crystals maintains freedom of movement in the melt. However, the probability of two crystals joined during synneusis with a CSL-OR will depend on the crystal morphology and lattice symmetry: higher symmetry and planar crystal facets increase the probability of forming CSL-ORs (German et al., 2009).

For dipyramidal quartz-quartz pairs, there are five common CSL-ORs, described by the angle formed between the c axes of the two crystals and the crystal face on which they are conjoined: parallel rhomb, parallel prism, rhomb 90°, prism 90°, and rhomb 27° (i.e., Esterel twin; Fig. 2A). EBSD data can be used to distinguish Esterel twins, rhomb 90°, and prism 90° relationships by identifying grain boundaries characterized by a misorientation with a defined rotation axis and angle (e.g., 90° ± 10° rotation about the axis for rhomb 90°; Fig. 2B). As demonstrated by Wiebe et al. (2007), with the addition of CL images, parallel CSL-ORs (rhomb and prism) can be identified for oscillatory-zoned grains with parallel crystallographic orientations (Fig. 2B).

With the exception of finely disseminated interstitial cassiterite, which is restricted to a central horizon of the sill, there is only minor stratigraphic variation in the modal mineralogy of the Bobbejaankop sill. Quartz forms well-faceted idiomorphic grains (primocrysts). Perthite forms subhedral to anhedral grains that, together with interstitial biotite and cassiterite, fill the space around the quartz grains. Apart from the late appearance of cassiterite in the central horizon, the apparent order of crystallization does not change through the sill. There is, however, notable stratigraphic microstructural variability, with an increase of both the grain size and the extent of clustering of quartz toward the center of the sill (Fig. 3A). Individual quartz grains exhibit concentric oscillatory zonation under CL (Fig. 3B), indicating growth as isolated crystals suspended in melt (Holness et al., 2018). The grains within the chain-like clusters are of similar size and are joined on well-developed (i.e., prismatic) crystal facets (Fig. 3B), again indicative of a prolonged period of growth as isolated grains in a liquid-rich environment before cluster formation.

Phase equilibria models calculated for a liquid composition equivalent to the bulk composition of the Bobbejaankop granite predict that quartz was the first major crystallizing phase (Fig. S1 in the Supplemental Material1). This prediction corroborates the microstructural observations of quartz oscillatory zonation and well-developed crystal facets, both of which require crystal growth to be unimpeded by other crystals. Accordingly, we are confident that the Bobbejaankop quartz microstructures preserve a record of the earliest stages of magma chamber dynamics prior to the formation of a rigid crystal mush.

The crystallographic orientation of quartz grains in two HSG samples (ZP99 and ZP131) collected from the center of the Bobbejaankop sill were analyzed by EBSD. Figure 2C shows the frequency of CSL-ORs for sample ZP131 relative to a simulated random orientation relationship for a population of uncorrelated quartz grains with the same orientations. Comparing our results to an uncorrelated population corrects for the influence of lattice symmetry on the distribution of possible misorientation angles. The quartz clusters found in the two samples from the center of the Bobbejaankop sill contain a high frequency of CSL-ORs, with most quartz grains oriented either parallel to, or at 90° to, neighboring grains and conjoined on their rhomb and prism faces (Fig. 2C). The concentric zoning of individual quartz grains (Fig. 3B) rules out cluster formation by growth twinning or epitaxial nucleation (the latter is, in any case, energetically unfavorable compared to continuous growth of existing grains; Kelton and Greer, 2010). Instead, the high frequency of CSL-ORs indicates that individual quartz crystals moved independently in the early stages of magma solidification.

The initial alignment of quartz crystals in low-energy orientations is likely a stochastic process that occurs during flow of a crystal-laden melt (DiBenedetto et al., 2020). The absence of randomly oriented pairs in the clusters suggests that quartz crystals only formed grain boundaries when they met already aligned in, or close to, low-energy grain-boundary orientations. Experiments on gold nanoparticles demonstrated that at very short distances (<50 nm), crystals will rotate toward a CSL-OR during synneusis (Chen et al., 2006). It is unclear whether a similar process occurs in magmatic environments, in which the crystals are larger. Instead, crystals that meet along certain faces (e.g., prism-prism) may remain together and undergo textural modification resulting in the rotation of one or both crystals toward a CSL-OR (Fig. 4B).

There is a growing consensus, based on microstructural data, that the crystallization of HSGs is not a static process (Huber et al., 2009; Bea, 2010; Bachmann and Huber, 2016). As indicators of the dynamics operating soon after magma chamber inflation, constraints on the magmatic environment in which quartz clusters form are key to understanding how HSGs crystallize. The formation of similar quartz clusters in the high-silica Vinalhaven complex in Maine (USA) has been interpreted as a consequence of crystal accumulation by settling (Beane and Wiebe, 2012). Conversely, the inward increase in number and size of clusters of well-faceted, concentrically zoned quartz crystals in CSL-ORs and the increasing grain size of the constituent grains (Fig. 4) toward the central horizon of the Bobbejaankop sill can only be explained by their formation from crystals suspended in a convecting magma, where they were able to move freely relative to other quartz crystals. The interpretation of the concentrically zoned quartz being magmatic rather than hydrothermal in origin is supported by the near-uniform major-element bulk composition throughout the sill and the enrichment of incompatible elements in a central horizon, implying that the melt was efficiently mixed during limited progressive fractionation during solidification. The modal proportion of the clusters in the sill center suggests that convection continued even as the crystal fraction reached 0.2–0.3. Taken together, these observations support a model of inward growth of solidification fronts from the top and bottom margins, accompanied by sustained mixing by convection of a crystal-bearing melt, in contrast to the commonly accepted view of crustal magma bodies as mush-dominated features (e.g., Cashman et al., 2017).

Numerical simulations of granite crystallization indicate that convection, driven by the increased density of crystal-laden melt at the roof of the intrusion, is the rule rather than the exception in the early stages of solidification of granite sills thicker than 200 m (Bea, 2010). Because plausible estimates of convection velocities are about one order of magnitude larger than the terminal Stokes’ velocity for quartz and feldspar phenocrysts, crystal-melt segregation is unlikely in HSGs. With limited segregation, the liquidus temperature remains near constant. As crystallization progresses to the more central horizon within the sill, a thermal buffer zone effectively minimizes the temperature contrasts across the body, preventing thermally driven convection once the magma has reached the temperature at which nucleation occurs (Marsh, 1989; Huber et al., 2009). Instead, the negatively buoyant, crystal-laden melt at the intrusion roof may initiate Rayleigh-Taylor instabilities, and hence convection, for any intrusion geometry with a vertical component >200 m (Bea, 2010). The simulations presented by Bea (2010) assumed the sill was emplaced in crust with an elevated stable geotherm, which would promote crystal nucleation at the roof of the intrusion. Studies of the Bushveld Complex aureole indicate the thermal gradient was unusually high at the time of granite magmatism (Waters and Lovegrove, 2002).

Unlike basaltic intrusions, in which gravitationally driven crystal-liquid segregation drives the chemical evolution of the bulk liquid (Sparks et al., 1984), our observations suggest the early crystallizing phases in HSGs remain in suspension during much of the solidification history. An apparent absence of crystal-liquid segregation in highly silicic intrusions means that, in many cases, crystal cumulates should not be expected to form. This is one way to explain the apparent lack of crystal cumulates in the upper crust, and it also explains the across-body homogeneity in major-element concentrations that many HSGs exhibit (Lee and Morton, 2015). As the solidification front advances from the margins inward, the crystal-laden melt in the center of a highly silicic body will become enriched in incompatible elements, like Sn, W, and Cu.

In the context of granite Sn/W ore mineralization, a “sandwich horizon” enriched in cassiterite, wolframite, and other mineralizing phases is expected to form within the interior of high-silica granites, provided that (1) the intrusions are large enough to support convection during crystallization, and (2) the intrusions remain closed chemical systems. We demonstrated here that HSGs can undergo continued convection during crystallization and that this behavior provides a means to concentrate highly incompatible and economic elements without crystal-melt segregation and the formation of crystal cumulates.

We thank Gerald Dickens, Chris Huber, Joseph Boro, and an anonymous reviewer for feedback that improved this manuscript. Laurence Robb is thanked for providing samples. This work was supported by a Natural Sciences and Engineering Research Council of Canada Discovery Grant awarded to B. Dyck.