The Bushveld Complex is located in South Africa and is the largest mafic layered intrusion in the world, covering 65 000 km² (Fig. 1). It was emplaced between 2.5 and 2.06 Ga within sedimentary host rocks (Zeh et al., 2015) and consists of several suites, but the V–Ti deposits are mainly found in the mafic Rustenburg Layered Suite (RLS, Fig. 1). The RLS comprises (from bottom to top) the Lower, Critical, Main, and Upper zones (Cawthorn, 2015). The Lower and Critical zones crystallized from primitive, Mg-rich magmas, and the Main Zone crystallized from a more tholeiitic, Fe-rich magma, while the Upper Zone is derived from residuals of these earlier liquids (Eales and Cawthorn, 1996). Magnetite is mainly found near the top of the RLS, and its first appearance is often used to define the boundary between the Main and Upper zones (Molyneux, 1974; Scoon and Mitchell, 2012; Yuan et al., 2017). The Main and Upper zones transition from gabbronorites to Fe–Ti-oxide gabbros and host up to 30 magnetite- and ilmenite ± apatite-rich layers. These magnetite layers are laterally continuous over large distances, potentially tens or hundreds of kilometres, and, as a result, V–Ti deposits are found in most areas where the Upper zone is exposed or occur near the surface. Due to lateral extent of the magnetite-rich layers, as well as the lateral homogeneity in terms of thickness, texture, and likely mineral compositions (e.g. Reynolds, 1985), single samples from such layers are representative of a large extent of the Bushveld Complex. We therefore use seven samples of magnetite-rich rocks (Table S1 in the Supplement) from a drillcore in the western limb of the complex to study Bushveld mineral textures (BK3; Fig. 1). The whole-rock composition and the compositions of silicate minerals have been described previously (Yuan et al., 2017).

In the Bushveld Complex, magnetite and ilmenite are the dominant V and Ti ore minerals, respectively. Their compositions are dominated by Ti and Fe, but they also contain V, Al, Mg, and Mn. Ilmenite has the general formula AxB2-xO3, where Fe²⁺, Fe³⁺, and Ti act as substitutes on the A (Fe²⁺, Fe³⁺), and B (Fe³⁺, Ti⁴⁺) crystallographic sites. This forms a solid-solution series (ilm_SS) between pure ilmenite (ilm: FeTiO₃; all Fe as Fe²⁺) and hematite (hem: Fe₂O₃; all Fe as Fe³⁺) endmembers (Fig. 2), where the relative contributions of each endmember are usually expressed as mol fractions Xi, with ∑X=1. As a result, ilm_SS can have a continuous range of compositions, varying from Ti-rich (maximum 35.86 wt % Fe for stoichiometric FeTiO₃) to Fe-rich (maximum 69.94 wt % Fe for stoichiometric Fe₂O₃) with decreasing Xilm and increasing Xhem. While near-endmember compositions (Xhem or Xilm>0.80) are stable down to low temperatures (<500 °C), more equal mixtures (0.30<Xhem, Xilm<0.7) are unstable below ∼900 °C (Buddington and Lindsley, 1964; Lindsley, 1991). As a result, cooling below this temperature leads to exsolution into near-endmember compositions becoming increasingly closer to the endmember compositions with decreasing temperature (Buddington and Lindsley, 1964; Carmichael, 1961).

Magnetite has the general formula AxB3-xO4, which mainly forms a solid solution series (mtss) between Ti-rich ulvöspinel (usp: Fe₂TiO₄, with all Fe as Fe²⁺) and Fe-rich magnetite (mt: Fe₃O₄, with one Fe²⁺ and two Fe³⁺ per formula unit) endmembers (Fig. 2). However, natural magnetite can incorporate other elements like Al and Mg, potentially forming more complex solid solutions with additional components, e.g. hercynite (FeAl₂O₄) or spinel (MgAl₂O₄; Turnock and Eugster, 1962; Charlier et al., 2015; Arguin et al., 2018). During progressive cooling, mt solid solutions become unstable at low temperatures and exsolve (Buddington and Lindsley, 1964). Exsolution temperatures depend on the components involved and vary from <600 °C for magnetite–ulvöspinel solid solution (Buddington and Lindsley, 1964) to ∼860 °C for hercynite–magnetite solid solution (Turnock and Eugster, 1962). In addition to exsolution, ulvöspinel components in magnetite can oxidize and produce ilmenite + magnetite according to the following reaction (Buddington and Lindsley, 1964; Duchesne, 1970; Tan et al., 2016): 16Fe2TiO4usp+O2→6FeTiO3ilm+2Fe3O4mt. This oxidation therefore leads to the formation of near-endmember mt and ilm, i.e. mineral purification (Charlier et al., 2015b), and can be expressed as an intermediate between mt and ilm.

Ilmenite and magnetite exsolutions create intricate mineral textures made up of intergrowths of the near-endmembers of their respective solid-solution series. For magnetite, these textures may be complicated further by ulvöspinel oxidation. As a result, different mineral textures form depending on (1) the extent of exsolution and oxidation, (2) the bulk composition of the host minerals, and (3) the potential occurrence of subsequent processes like diffusive re-equilibration. Note that rates of exsolution and oxidation are also diffusion controlled and depend on the diffusion coefficients of all involved elements. If ilmenite formed through ion exchange, this would show as replacement textures at ulvöspinel grain boundaries (Charlier et al., 2015b, their Fig. 15). We do not observe such textures in Bushveld (see below) and have only found “classical” sandwich and trellis textures. We therefore do not think that ion exchange has contributed significantly to ilmenite formation.

Magnetites with cloth textures in the Bushveld Complex show a 3D network of ulvöspinel lamellae along the (100), (010), and (001) crystallographic planes (Figs. 3 and 4; Price, 1980; Ramdohr, 1953; Tan et al., 2016). When crystals are cut parallel to one of the exsolution planes they intersect at 90° in cross-section (Figs. 3c, f and 4b), but when crystals are cut along the [111] crystallographic axis they show 60–120° intersections (Figs. 3d, e and 4c). These lamellae are a few micrometres to nanometres thick, and their spacing ranges from ca. 0.1 to 3 µm. Granular and needle-like Al-rich (hercynitic) spinel [(Mg,Fe)Al₂O₄] of up to 3 µm in length is sometimes associated with these lamellae (Fig. 3e; Spencer and Lindsley, 1981; Charlier et al., 2015b). This may indicate evolution along the magnetite–ulvöspinel–Al–spinel solid-solution series (e.g. Charlier et al., 2015b) and implies solvus temperatures higher than those of magnetite–ilmenite solid solution (see Sect. 1.2). However, the modal abundances of spinel in cloth-textured magnetite are very low (≪1 %), and its impact is therefore expected to be minor. At Bushveld, cloth textures are often observed in interstitial magnetite within grains with sandwich textures (see below) but never in grains with trellis textures (see below). Identical textures are found in VTM deposits worldwide and they form via exsolution as magnetite cools below the solvus (Arguin et al., 2018; Duchesne, 1972; Taner et al., 1998; Namur et al., 2010).

In the Bushveld Complex, magnetite sandwich or trellis texture are characterized by relatively thick, ∼10–300 µm ilmenite lamellae inside magnetite grains, spaced tens to hundreds of micrometres apart. In the sandwich textures, lamellae are all roughly parallel (Fig. 5a and b), and the interstitial magnetite has pervasive cloth textures made up of thin ulvöspinel lamellae (see Sect. 2.2, Fig. 5a–c). Furthermore, granular ilmenite is sometimes found along grain boundaries of sandwich textured magnetite. Contrastingly, in the Bushveld trellis textures, ilmenite lamellae form along 2–3 different planes (Fig. 5a), with 60–120° angles between them. Here, the interstitial magnetite (mt) contains very thin (<1 µm) and short (∼10 µm), needle-like exsolutions of ilmenite and spinel (Fig. 5d). However, often a border array of spinel of ca. 10 µm is present around the trellis lamellae where these needles are mostly absent (Fig. 5b–d).

The ilmenite lamellae and needles form via oxidation of the ulvöspinel component in titanomagnetite according to Reaction (R1). In sandwich textures, they form along a single (111) plane (Fig. 4a and b), while they form along multiple sets of the {111} planes in trellis textures (Fig. 4a; Buddington and Lindsley, 1964; Haggerty, 1991; Mücke, 2003; Tan et al., 2016). Notably, the presence of ulvöspinel in the cloth texture of interstitial magnetite in sandwich textures indicates (1) that these magnetites were not completely oxidized and (2) that the sandwich ilmenite lamellae formed before the cloth texture.

The ilmenite-free zone around ilmenite trellis lamellae indicates that diffusion of Ti through titanomagnetite has taken place after oxidation. Through this process, the needle-like ilmenite is effectively transported towards the trellis lamellae, leaving behind an ilmenite-free border zone (Duchesne, 1970; Von Gruenewaldt et al., 1985). However, diffusion rates in the trellis textures were not high enough to have all ilmenite and spinel components migrate to the ilmenite lamellae, and, in more internal parts, the needles are therefore still present. Similarly, the observed granular ilmenites likely formed through diffusion near grain borders, expelling ilmenite from the magnetite grains (Duchesne, 1970; Namur et al., 2010).

Analyses of the ilmenite and magnetite parts of trellis and sandwich textures show that they have near-endmember mt and ilm compositions (Fig. 7). This indicates that all ulvöspinel has been converted according to Reaction (R1) and that the magnetite has been fully oxidized into mt and ilm near-endmembers. Here, ilmenite exsolutions have V₂O₅ contents of 0.27 wt %–0.41 wt %, while, in the magnetite host, they are considerably higher at 0.72 wt %–1.14 wt % (Fig. 7). Before the formation of the trellis texture, these components were in solid solution, and the oxidation of the precursor magnetite therefore enriched the remaining magnetite in vanadium (Fig. 6). The downside of this process is that the formation of ilmenite trellis also reduces the degree of magnetite liberation. Since the grains with a trellis texture are fully oxidized, the amount of ilmenite lamellae reflects the composition of the precursor magnetite, where more lamellae indicate lower magnetite components formed at magmatic temperature. This means that trellis grains with fewer lamellae have higher overall V₂O₅ contents in addition to having more liberated magnetite.

V₂O₅ concentrations in trellis-textured magnetite are overall lower than those in cloth-textured host magnetite. As indicated above, the stratigraphic positions of trellis samples are higher than those with cloth textures. The samples with trellis textures therefore crystallized later, likely from magmas with overall lower vanadium contents (e.g. Yuan et al, 2017; Namur and Humphreys, 2018). The lower vanadium concentrations in the trellis-textured magnetites are therefore a consequence of the overall lower V ore grade of these samples and not of element redistributions related to texture formation. Indeed, for magmatic mtss with the same bulk V₂O₅, the process of trellis formation is expected to enrich the mt host more than the process of cloth formation (Fig. 6). In trellis grains, TiO₂ concentrations in ilmenite lamellae are 52.2 wt %–52.8 wt % (i.e. near stoichiometric ilm), much higher than in any of the analysed magnetites (<16 wt %). This shows that titanium effectively gets extracted from the mtss into ilm lamellae. In samples with sandwich lamellae, ilmenite exsolutions have comparatively low V₂O₅ contents of 0.3 wt %–0.46 wt % but high TiO₂ concentrations of 53.3 wt %–55.6 wt %.

Chemical compositions of cloth lamellae and host magnetite can only be measured as mixtures of both as analytical spot sizes of currently available techniques for in situ chemical analyses are greater than the lamellae thickness and spacing. Measured compositions therefore depend not only on the magnetite_SS compositions but also on the density of the exsolution network. Indeed, results from Bushveld show mixtures of endmember magnetite and ulvöspinel (i.e. compositions along the magnetite–ulvöspinel solid solution; Fig. 7). This is because there is no oxidation of ulvöspinel to ilmenite in Bushveld, unlike at some other VTM deposits (e.g. Rogaland, Norway: Duchesne, 1970; Emeishan, China: Tan et al., 2016). The mixed analyses show that V₂O₅ contents increase with increasing magnetite components from ∼1.19 wt % V₂O₅ at Xmt=0.27 wt % to 2.4 wt % at Xmt=0.58 (Fig. 7). Overall, cloth grains with lower-density ulvöspinel exsolution networks have higher V₂O₅ contents. Vanadium is therefore preferentially hosted in the magnetite component, but we believe that the compositional evolution we observe represents an evolution of the magma composition rather than a heterogeneous distribution of the ulvöspinel exsolution (Yuan et al., 2017) In any case, due to the pervasive cloth texture its degree of liberation is comparatively low. TiO₂ in magnetite increases with increasing mol fractions of ulvöspinel (Xusp), but, since the ulvöspinel lamellae are very thin, their liberation degree is low.

While the processing of the sandwich structure is partially more feasible compared to the cloth structure, it is still a challenging microstructure to process. The larger lamellas may be liberated through conventional grinding methods. However, new developments in ore processing are promising and could increase the degree of liberation. For example, use of high-pressure grinding mills (HPGR) (Guo et al., 2019) and microwave treatments (Wang et al., 2019) have been studied in an effort to break the ore structure along grain boundaries. According to Wang et al. (2019) the use of microwave treatment resulted in more cracks between grain boundaries, thus further enhancing grinding procedures and resulting in better liberation with coarser particle sizes. It is noteworthy that, in the processing of the sandwich and, especially, the trellis structures, the benefits of using HPGR or the microwave treatment will likely improve the efficiency of the downstream processing. While such novel methods improve liberation and can theoretically also expose the thin Ti lamellas in the sandwich structure, their efficient enriching relies heavily on their grain properties and the selected processing method, as discussed below.

Well-liberated larger ilmenite-containing grains from sandwich-textured magnetite can, theoretically, be separated through gravity separation. It can be assumed that the novel comminution methods mentioned above expose ilmenite surfaces, which also enables the use of froth flotation, especially for ilmenite. As a surface and adsorption-based separation technique, its performance depends heavily on proper liberation of valuable minerals (Bulatovic, 2007). In particular, the larger ilmenite inclusions can occupy most of the volume of a single 30–40 µm grain (depending on orientation). This, in turn, facilitates selective separation through flotation, while some gangue recovery is also to be expected depending on the actual grain size. However, another equally important aspect is the state of the surface minerals. For example, in the case of ilmenite flotation, the ratio between Fe²⁺ and Fe³⁺ determines the degree of interaction between the mineral surface and typical anionic collectors (such as sodium oleate) used in ilmenite flotation (Li et al., 2024). A further enhancement in the collector adsorption can be done by activating the ilmenite surfaces with sodium persulfate (Na₂S₂O₈), which reportedly prompts a conversion from Fe²⁺ into Fe³⁺ (Dai et al., 2025).

The trellis structure is expected to show the most promise in terms of enriching Ti and V in their own pre-concentrates. The larger grains can be exposed even through conventional comminution, while the novel methods further improve liberation and reduce energy consumption. When properly liberated, the larger ilmenite inclusions can be separated from the V-bearing magnetite through magnetic separation with relative ease compared to the more fine-grained structures. The ilmenite-rich non-magnetic fraction can be further enriched through froth flotation as the sizes of the ilmenite-bearing particles in the trellis structure are better suited for flotation (Wills and Finch, 2015).

As discussed above, the cloth structure contains thin needle-like exsolutions, which are notoriously difficult to liberate efficiently using conventional comminution techniques (Xu et al., 2017). With these methods, even a theoretical complete liberation of the cloth structure would result in overly fine particle sizes that are difficult or impossible to enrich – not to mention the immense energy demand of such operations (Baldassarre et al., 2020; Parapari, 2021). While some of the lamellas may be exposed when using HPGR or microwave treatment, a continuous cloth structure inhibits efficient separation in conventional enrichment unit operations. Structural changes occur in a sub-micrometre spacing, which results in a near-uniform structure throughout the cloth mineralization when considering processing methods such as gravity or magnetic separation. In addition, even with optimal reagents, flotation of the fine-grained ilmenite exsolutions is extremely difficult. In the typical size range where flotation is most effective (100–150 µm), the available surface area consisting of ilmenite is too small for the collectors to render the mineral surface sufficiently hydrophobic for them to attach to the bubbles in flotation. Conversely, if the particle size is very small (<20 µm), gangue minerals can form slime coatings on top of the valuable minerals, effectively acting as depressants, while smaller particles also report to the concentrate more readily through the non-selective recovery of entrainment (Wills and Finch, 2015). Micro-flotation technologies of fine fractions could be a solution for a better recovery (Rulyov, 2016; Rulyov et al., 2020), but this still remains to be fully investigated.

While Sect. 4.1 and 4.2 described the properties and typical processing behaviour of the distinct mineral structures present in the Bushveld Complex, in any realistic scenario it is impossible to segregate the processed ores based on their mineral structures to individual processing flow sheets. Accordingly, the efficient utilization and processing of this ore requires detailed planning throughout the beneficiation chain that simultaneously considers all structure types present in the complex. The resulting flow sheet should contain novel concepts, such as selective blasting and fragmentation combined with already existing technologies, like the HPGR, to maximize both the liberation and the energy efficiency of the comminution circuit. The subsequent downstream processing design should consider a series of varying unit operations, each designed for separate particle size ranges, from coarse to fine, with strategically positioned re-grinding stages, resulting in optimized energy consumption and separation efficiency.