To sink, swim, twin, or nucleate: A critical appraisal of crystal aggregation processes

Despite the abundance of crystal aggregates in a diverse range of igneous rocks, no consensus has been reached on their origin (Schwindinger and Anderson, 1989; Jerram et al., 2003; Day and Taylor, 2007). As early as 1921, Vogt suggested that crystal aggregates form by synneusis, the “swimming together” and attachment of previously isolated crystals in a fluid medium. More recently, aggregates have been attributed to growth twinning (Welsch et al., 2013), the nucleation and growth of a daughter crystal via epitaxial growth/heterogeneous nucleation (Kirkpatrick, 1977; Dowty, 1980a; Hammer et al., 2010; Špillar and Dolejš, 2015), or the textural maturation of dendrites (Welsch et al., 2013, 2014). Previous attempts at distinguishing these hypotheses have utilized optical observations of chemical zoning (Dowty, 1980a), assessments of crystallographic orientation using a universal stage (Ross, 1957), and observations of three-dimensional (3-D) aggregate morphology (Schwindinger and Anderson, 1989; Welsch et al., 2013). In this study, we quantified changes in crystallographic orientation across grain boundaries using electron backscatter diffraction (EBSD) maps (e.g., Prior et al., 1999). This allowed direct comparison between observed aggregate geometries and those predicted by each hypothesis, even in optically isotropic or chemically homogeneous minerals. To demonstrate the versatility of this method, we investigated olivine aggregates from rapidly quenched scoria from several eruptions at Kīlauea volcano, Hawaiʻi, USA (A.D. ca. 1700, ca. 1920, and 1969–1975; see DR1 in the GSA Data Repository¹; Fig. 1) and chromite crystals from the UG2 chromitite layer of the Bushveld Complex, South Africa (Fig. 2).

Unravelling the magmatic processes responsible for aggregation has important implications. At Kīlauea, olivine constitutes the principal record of pre-eruptive processes in primitive lavas. Understanding the formation and evolution of chromitite layers such as the Bushveld UG2 layer is crucial to advance our understanding of how platinum group elements are concentrated into economically viable ore deposits (Naldrett, 2013). The timing and mechanism of aggregate formation also influence crystal redistribution and accumulation in layered intrusions (Campbell, 1978; Holness et al., 2017a), and the interpretation of quantitative textural measurements such as crystal size distributions (CSDs), which frequently record kinked CSDs, and an overabundance of larger crystal sizes in volcanic rocks (Marsh, 1998).

EBSD maps were made for 93 Kīlauea aggregates (Fig. 1) and a thin section of the UG2 layer of the Bushveld Complex containing ∼3500 chromite grains (Fig. 2). The crystallographic relationships between touching grains in aggregates were quantified in terms of their misorientation. This describes the coordinate transform from one crystal orientation to that across the grain boundary, displayed as an axis-angle pair (Fig. 3A). Misorientations were calculated using MTEX V5.0.4 (Bachmann et al., 2010), an open-source toolbox for MATLAB. Electron microprobe transects across olivine aggregates were obtained using a Cameca SX100 electron probe microanalyzer (EPMA; see DR1).

Olivine aggregates, consisting of 2–17 individual crystals (Fig. 1), showed misorientation axes strongly clustered about <100> (Fig. 3B), and broad peaks in misorientation angles at <5°, ∼40°, and ∼80° (Fig. 3C). EPMA transects across attached olivine crystals revealed flat compositional profiles in the cores of crystals, with sharp gradients across grain boundaries (Fig. 1E, sections X-X′ and Z-Z′). The distribution of chromite misorientation angles was very close to the theoretical random distribution (Fig. 4C), with the only prominent peak at 60°. Chromite misorientation axes exhibited weak clustering about <111> and <011> (Fig. 4A). Removal of the {111} twins responsible for the 60° peak resulted in a nearly uniform distribution of misorientation axes and angles (Figs. 4B and 4C).

Hydrodynamic processes within magmatic systems can drive the aggregation of preformed crystals (McIntire et al., 2019). One such process is synneusis (Vogt, 1921; Vance, 1969). While the driving forces associated with interfacial energy minimization during synneusis are too localized to drive crystals together (Dowty, 1980a), crystals may be aligned while suspended in convecting magma (Holness et al., 2017b), or during settling (Schwindinger and Anderson, 1989; Schwindinger, 1999), resulting in their attachment in specific orientations. Once sufficiently close, reductions in interfacial energy associated with particular crystallographic orientations can drive small readjustments in aggregate geometry (Dowty, 1980a), before further crystal growth cements the crystals together.

To assess the synneusis hypothesis in Kīlauea olivine aggregates, the faces of attachment between adjacent crystals were assessed by superimposing 3-D olivine shapes onto EBSD maps using the crystalShape MTEX class (Enderlein, 1997; https://mtex-toolbox.github.io/files/doc/crystalShape_index.html; Fig. 1C), and these were compared to the predicted hydrodynamical alignments during settling. Oblate olivines align their axis of symmetry vertically during settling, while prolate olivines align their axis of symmetry horizontally (Leal, 1980). This creates mostly {010} attachments in oblate olivines, and {110} attachments in prolate olivines, with occasional {021} attachments in both cases (Schwindinger and Anderson, 1989). We anticipate that similar alignments would also occur during convection.

The {110}-{110} attachments were the most commonly observed geometry in Kīlauea aggregates (Fig. 1C), followed by {010}-{010} or parallel {021}-{021} attachments (difficult to distinguish in two dimensions). Crystals that are perfectly attached along parallel {110}-{110}, {010}-{010} and {021}-{021} faces would have identical orientations, while perfect attachments along {021}-{010} and perpendicular {021}-{021} faces would generate misorientations about <100> of 40.4° and 80.9°, respectively. The presence of broad misorientation angle peaks centered about these angles is strong evidence for synneusis (Fig. 3C). Unlike crystal growth processes, which favor specific crystallographic relationships, attachment by synneusis can easily generate misalignments from these ideal attachment geometries through the mechanical hindrance to reorientation driven by the reduction of interfacial energies (Vance, 1969; Nespolo and Ferraris, 2004), or near-random attachment under conditions of high supersaturation (Schwindinger and Anderson, 1989). Imperfect attachment by synneusis also accounts for the presence of discontinuous melt films between adjacent olivine crystals (Figs. 1A and 1B). If formed during growth, such films would impose high energy penalties.

Finally, the EPMA transects support the hypothesis that the olivine aggregates formed by synneusis (Fig. 1E). While synneusis can bring together crystals with different core compositions that formed in disparate environments (Schwindinger and Anderson, 1989), aggregates produced by twinning or dendritic growth share a common history, so the cores of attached crystals should have identical chemistry (Philpotts et al., 1998). The sharp gradients in forsterite contents (which dissipates rapidly by diffusion) suggest that the aggregation event occurred shortly before eruption. Crystal aggregation may occur following a disturbance to the magmatic system, triggering their eruption, or during transport of the magma to the eruption site.

Evaluating the role of physical aggregation in the formation of chromite aggregates is complicated because attachment along the only crystal face, {111}, produces misorientation axes indistinguishable from primary twinning (60° rotation about <111>; Deer, 1966). However, this geometric relationship is relatively uncommon, and the removal of misorientations corresponding to the chromite twin law (and effects induced by crystal symmetry) produces a near-uniform misorientation axis distribution, and an angle distribution resembling the theoretical random distribution (Figs. 4B and 4C). This implies that chromite grains are randomly aggregated by a mechanism that is not selective to crystallographic orientation; this is inconsistent with the attachment of {111} faces by synneusis, or various growth processes. We suggest that the randomly oriented touching frameworks of chromite grains formed during the settling of individual grains through the overlying magma column to the top of the cumulate pile (Jackson, 1961). This hypothesis is consistent with microstructures in the cumulates overlying UG2, in which individual chromite grains apparently rest on underlying pyroxene primocrysts (Fig. 2C).

Crystal aggregates attached along specific planes have also been attributed to primary twinning (Welsch et al., 2013). However, neighboring chromite grains displaying twin relations are relatively rare (<10% of neighboring grains, based on deviations from ideal twin laws; Brandon, 1966; see DR1). Similarly, <7% of neighboring olivines are related by the three olivine twin laws (twinning on {011}, {012}, and {031}, corresponding to misorientations about <100> of 60.7°, 32.6°, and 59.3°, respectively; Dodd and Calef, 1971; Azevedo and Nespolo, 2017). Thus, twinning was not the dominant cause of grain aggregation in our samples.

Dowty (1980a) suggested that the textural and chemical characteristics of crystal aggregates could result from epitaxial growth or heterogeneous nucleation. These processes, driven by the reduction of the activation barrier to nucleation by the presence of an existing crystalline substrate in the melt (Lofgren, 1983; Mithen and Sear, 2014), are well documented in multiphase aggregates (Kirkpatrick, 1977; Hammer et al., 2010). However, in monomineralic aggregates, nucleation and growth of a daughter crystal are associated with energy penalties. The increased surface area per unit volume (and increased interfacial energy) caused by the growth of a daughter crystal was assessed by comparing an individual crystal of a given volume with an aggregate with the same total volume composed of two identical crystals (volumes and areas calculated in SHAPE7.4; Dowty, 1980b; see DR1). As chromite crystals express only {111} faces, the increase in interfacial energy for aggregates compared to individual crystals (11%) is independent of the interfacial energy of the {111} face. For olivine, theoretical surface energies for the six different olivine faces in a vacuum at 0 K (Bruno et al., 2014) were used. Regardless of the attachment face, the growth of a daughter crystal in olivine is accompanied by significant energy penalties (13%–24%; see DR1).

Energy penalties for heterogenous nucleation are even greater. While epitaxial growth describes the nucleation of crystals in a specific crystallographic orientation with respect to the host crystal (identical or twin orientations in single-phase aggregates; Mithen and Sear, 2014), heterogeneous nucleation has no such orientation constraints. The large lattice mismatches, and associated energy penalties, generated by heterogeneous nucleation may drive the detachment of a poorly orientated unit, followed by reattachment in a more-favorable orientation (Buerger, 1945). In certain cases, it may even be more energetically favorable for the components of the misaligned daughter crystal to remain in the liquid state (Brugger and Hammer, 2015).

The larger energy penalties for heterogeneous nucleation compared to epitaxial growth, alongside lower nucleation rates for mismatching parent and daughter lattices (Turnbull and Vonnegut, 1952), mean that heterogeneous nucleation should occur less frequently than epitaxy in monomineralic aggregates. Olivine and chromite aggregates show little evidence for epitaxy; twin relationships are rare (Figs. 3 and 4), and no crystals with uniform orientations show external morphologies indicative of aggregation. Given the limited evidence for epitaxial growth, it is improbable that the energetically unfavorable process of heterogeneous nucleation was the dominant cause of grain aggregation in this study.

Welsch et al. (2013) proposed that polyhedral aggregates are produced during an early period of dendritic growth, followed by textural ripening. This hypothesis was based on observations that attached olivine crystals have identical orientations, are in continuous contact, and are superimposed into chains of decreasing crystal size elongated along <100> (Welsch et al., 2013). However, crystallographic branching in dendrites (Cody and Cody, 1995; Faure et al., 2003a, 2003b) is associated with misorientations lower than ∼3° (Faure et al., 2003b), and so it cannot account for the 86% of olivine aggregates with misorientations >3° (Fig. 3). Additionally, neighboring crystals in the subset of olivine aggregates displaying small misorientations (<5°) were often separated by pockets of melt, and they showed no preferential elongation direction or size ordering (Fig. 1A). While the textural maturation of noncrystallographic branching dendrites (Keith and Padden, 1963; Donaldson, 1976) could produce the 86% of aggregates with large misorientation angles, the crystallographic signatures of branching olivine dendrites from West Greenland picrites differ from those of olivine aggregates (Larsen and Pedersen, 2000; Fig. DR5). While olivine dendrites and aggregates both display a misorientation angle peak at <20°, dendrites show no peaks at higher angles (Fig. DR5). The differences between these distributions are statistically significant (p = 0.013; Kolmogorov-Smirnov test). Additionally, misorientation axes in branching dendrites form a bimodal distribution, with a primary maximum at <010> (Donaldson, 1976) and a minor secondary maximum at <100> (Fig. DR5). This contrasts with the strong clustering of olivine misorientation axes at <100> in Kīlauea aggregates (Fig. 3B). Overall, the crystallographic and textural signatures of olivine aggregation are inconsistent with an origin from noncrystallographic branching during dendritic growth.

The large range of misorientations in chromite and olivine aggregates cannot be explained by crystal growth processes, which favor specific crystallographic relationships. Observed facial attachments and compositional profiles across adjacent olivines are consistent with crystal aggregation by synneusis shortly before eruption. In crystal populations affected by synneusis, textural investigations such as crystal size distributions must be constructed from measurements of individual grains within an aggregate. Consideration of the total size of an aggregate is only justified if aggregation can be attributed to twinning or dendritic growth. To distinguish between these cases routinely, a workflow could be developed whereby the origin of grain aggregation is assessed by EBSD maps, which are subsequently used to generate “informed” CSDs. The random crystallographic relationships between adjacent chromite grains in our sample of the UG2 chromitite layer are best explained by the formation of touching networks following crystal settling. Settling as individual crystals rather than aggregates has implications for initial mush porosity and accumulation time scales, with relevance to the formation of economically viable platinum group element deposits.

We thank Giulio Lampronti and Iris Buisman for help with collecting and interpreting electron backscatter diffraction and electron probe microanalyzer data, and Eric Dowty for providing guidance on the use of SHAPE software. Isobel Sides (funded by a Natural Environment Research Council [NERC] studentship) and Don Swanson (U.S. Geological Survey) collected the Kīlauea scoria samples. Lotte Larsen and Ilya Veksler are thanked for providing the Greenland picrites and UG2 chromitite samples, respectively. We thank Chris Clark for editorial handling, and François Faure, Dougal Jerram, and one anonymous reviewer for their constructive comments, which helped to greatly improve this manuscript. Wieser was funded by NERC doctoral training partnership (DTP) studentship NE/L002507/1; Vukmanovic was funded by Marie Curie grant 708131-EFOX-H2020-MSCA-IF-2015.