Constraints on the Nd-isotopic composition and nature of the last major influx of magma into the Bushveld Complex

The Pyroxenite Marker, a thin, orthopyroxene-dominated marker horizon, is observed towards the top of the Main Zone of the Bushveld Complex, where the last voluminous influx of magma into the Bushveld Complex is thought to have occurred. In an attempt to constrain the Nd-isotopic composition of the magma added at the level of the Pyroxenite Marker, a total of 21 whole-rock samples from a borehole (BH7771) drilled on the Central Sector of the Eastern Limb of the Bushveld Complex were analysed for their Sr–Nd isotopic ratios. Modelling suggests that the added magma had a unique Sr (87Sr/86Sri = 0.7063–0.7067) and Nd (ƐNdi on the order of − 5.9) isotopic composition, distinct from any of the rocks constituting the layered sequence below the Pyroxenite Marker. Dispersion of data points around the modelled isotopic (melt–melt) mixing curves is interpreted to reflect the incorporation of minerals derived from either the incoming or resident magmas into individual rock layers occurring across the Pyroxenite Marker interval, either in response to the mixing of minerals settling through a stratified magma column, or potentially through the intrusion and mixing of crystal-laden magmas with unique isotopic compositions from a sub-Bushveld staging chamber.


Introduction
The Pyroxenite Marker is a thin, orthopyroxene-dominated marker horizon occurring towards the top of the Main Zone of the Bushveld Complex. It crops out in the Eastern Limb of the Bushveld Complex, where it was first described by Von Gruenewaldt (1973) and Molyneux (1974). The Pyroxenite Marker is sandwiched between a succession of rocks exhibiting a sustained reversal in mineral compositions (i.e. increasing plagioclase An% and mafic silicate Mg# upwards), which starts > 100 m below the layer, and an overlying sequence of rocks exhibiting a normal differentiation trend (i.e. decreasing plagioclase An% and mafic silicate Mg# upwards). The layer also marks the level within the Main Zone where primary orthopyroxene replaces inverted pigeonite as low-Ca pyroxene. The Pyroxenite Marker appears to be present in all three of the major exposed limbs of the Bushveld Complex, having been described in the Bierkraal (BK2) drill core in the Western Limb of the complex (Cawthorn et al. 1991) and in the Bellevue (BV1) drill core in the Northern Limb of the complex (Cawthorn 2020).
The Pyroxenite Marker interval is commonly interpreted as having formed in response to the addition of a large volume of a less-differentiated melt and subsequent mixing with the resident melt, a view which is strongly supported by Srisotopic investigations (Kruger et al. 1987;Cawthorn et al. 1991;Kruger 1994). Sharpe (1985) suggested an alternative model, proposing that a dense magma, added at the level of the Merensky Reef, underplated the resident magma without undergoing significant mixing. The underplated magma then partially crystallised to form the Main Zone below the Pyroxenite Marker. Mixing occurring at the top of the underplated magma and the upwards displaced magma resulted in the development of the Pyroxenite Marker interval. This model, however, received criticism from several authors on the basis of thermal, density and compositional considerations (Kruger et al. 1987;Cawthorn et al. 1991). Gabbronoritic rocks below the Pyroxenite Marker typically exhibit initial 87 Sr/ 86 Sr ratios of > 0.708, with the sequence of rocks above the Pyroxenite Marker displaying a near-constant initial 87 Sr/ 86 Sr ratio of 0.7073 all the way to the top of the layered sequence (Kruger 1994). The melt that was added to the chamber at the level of the Pyroxenite Marker is, therefore, regarded as the last major influx of melt into the Bushveld Complex, a view that not only finds support when considering the isotopic evidence (Kruger et al. 1987;Kruger 1994), but also mineral compositions that vary regularly towards lower temperature compositions above the level of the Pyroxenite Marker (e.g. Mangwegape et al. 2016).
The composition of the melt added at the level of the Pyroxenite Marker has been rather well-constrained, particularly its trace element and Sr-isotopic compositions. Cawthorn et al. (1991) suggested that the added melt had an initial 87 Sr/ 86 Sr ratio of 0.7063 and showed, on the assumption, that the resident and incoming melts had similar Sr contents, that the volume of melt added should have been approximately equal to the volume of the resident melt. Vantongeren and Mathez (2013) calculated an initial 87 Sr/ 86 Sr ratio of 0.7066-0.7068 for the incoming melt, using equilibrium Sr contents calculated for the melts from the trace element contents of the minerals present across the Pyroxenite Marker interval. Their calculations suggest that the added melt volume was ~ 33-44% of the volume of the resident melt, depending on whether melt was lost from the chamber or not.
In this contribution, we attempt to constrain the Nd-isotopic composition and nature of the magma that was added at the level of the Pyroxenite Marker through analysis of the whole-rock isotopic compositions of rocks across the Pyroxenite Marker interval as sampled from borehole BH7771 on the Eastern Limb of the Bushveld Complex.

Driekop drill core: geology and sampling
Borehole BH7771 is a 2696.38 m deep hole drilled at Driekop ( Fig. 1), on the Central Sector of the Eastern Limb, by Impala Platinum in 2008. It is collared in the Main Zone and extends down to the footwall of the Merensky Reef in the Upper Critical Zone. A total of 45 samples were collected as part of this study, covering the depth interval between 474.09 m and 665.10 m. The Pyroxenite Marker is represented by a 1.91 m-thick orthopyroxenite layer with a sharp lower contact and a gradational upper contact, occurring at a depth of 488.26 m. The base of the Pyroxenite Marker was used as the reference datum for the profiles presented in this study.

Petrography
Polished thin sections were prepared and petrographically examined using transmitted light microscopy. The Olympus BX53M/DP74 microscope at the University of the Witwatersrand was used to take high-resolution scans of each thin section. These scans were then used for point counting by placing a digitised point grid on the scan and assessing 300 points. Minerals were assigned concentrations of less than  Scoon and Mitchell 2009and Roelofse and Ashwal 2012. "f " indicating major faults 0.5% by volume if they were observed under the microscope, but not encountered during point counting.

Whole-rock trace element geochemistry
Trace element analyses were performed on 45 samples in the Earth Lab of the University of the Witwatersrand on a Thermo Scientific iCAP RQ Inductively Coupled Plasma Mass Spectrometer (ICP-MS). A portion of each of the 45 samples collected was crushed and milled using a jawcrusher and swing mill, respectively. 50 mg of sample was weighed and digested in either a microwave digester (MARS from CEM) or using the open beaker/hotplate method depending on the sample type and digestibility. For the microwave digester method, the sample powder was added to a Teflon vessel along with 6 ml of ultra-high purity 2:1 HF:HNO 3 . Digestion vessels were then microwaved for 40 min at 180 °C and 400 PSI. The mixtures were transferred to Savillex beakers, capped and placed on a hotplate for 24 h at 70 °C. The acid was thereafter allowed to evaporate. 2 ml of HNO 3 was then added to the Savillex beakers, which were capped and placed on a hotplate for 24 h at 70 °C, before being allowed to dry down. This was repeated once more, before the samples were removed from the hotplate and 300 µl HNO 3 was added. Samples were stored in this state until they were ready for analysis. For the open beaker/hotplate method, 50 mg sample was placed into a Savillex beaker with 3 ml of 2:1 HF:HNO 3 and placed on a hotplate at 70 °C and evaporated off. Once dried, 3 ml of HF:HNO 3 was added and capped and placed on the hotplate at 70 °C for 72 h. The sample was then dried down at 70 °C and 3 ml HNO 3 added and evaporated off. This was repeated three times.The prepared samples were diluted to 50 ml (dilution factor 1:1000) with 5% HNO 3 with 100 ppb Re and Rh, as well as 50 ppb In and Bi as internal standards.
The Thermo Scientific iCAP RQ was optimised for maximum counts on In and both oxide and doubly charged ion levels set to < 2%. All measurements were done in triplicate and averaged. Certified Reference Materials (BCR-1 and BHVO-2) were digested and analysed along with all unknowns for quality control. The Certified Reference Materials had to return less than 10% deviation from known concentrations for all elements to pass the quality controls. Deviations were usually within 5% (see Electronic Appendix A). A Total Procedural Blank (TPB) was analysed along with all unknowns.

Whole-rock major element geochemistry
A portion of each of the 45 samples collected was crushed and milled using a jaw-crusher and swing mill, respectively. An amount of 10 g of sample powder was dried overnight at 110 °C and then subjected to 1000 °C for 4 h, to gravimetrically determine the loss on ignition (LOI). An amount of 0.28 g of powder was mixed with 1.5 g Alfa Aesar spectroflux (La 2 O 3 (47%), Li 2 B 4 O 7 (37%), Li 2 CO 3 (16%)) and 0.02 g NaNO 3 . This mixture was then fluxed at 1000 °C (± 5 min) in a Pt crucible. The melt was moulded and pressed to produce fusion disks. The prepared fusion disks were analysed by wavelength dispersive X-ray Fluorescence (XRF) spectrometry using a Rigaku Primus IV spectrometer equipped with a 4 kW Rh tube. Certified Reference Materials were used to monitor the accuracy and precision of the analyses (see Electronic Appendix A). NIM-N (SARM 4) and NIM-P (SARM 5) were prepared following the same sample preparation method as mentioned above and analysed during subsequent analytical batches. Repeatability for the major elements obtained on replicates and expressed as percentage relative standard deviations (%RSDs) were all less than ~ 3% for SARM 5 (except for K 2 O, 10.2%) and less than 5% for SARM 4 (except for K 2 O, 6.9% and P 2 O 5 , 17.4%). Relative uncertainties for all oxides compared to the accepted values for the SARM 5 standard were all below 3%, except for Na 2 O (+ 9.5%), K 2 O (− 11.1%), and P 2 O 5 (− 7.5%). For SARM 4, relative errors were also all below 3%, except for TiO 2 (− 5.5%), Na 2 O (− 10.6%) and P 2 O 5 (− 17.3%).

Mineral chemistry
Twenty-one samples were used for quantitative mineral chemical analyses of feldspar and pyroxene using four wavelength dispersive spectrometers on a JEOL JXA-8230 electron probe micro-analyser at Rhodes University. Analyses were performed at an accelerating potential of 15 kV and a beam current of 20 nA. A 10 µm beam was used on feldspar to avoid Na migration, with a 1 µm beam used for the analysis of pyroxenes. All elements were measured on K-alpha peaks except for Sr and Ba, which were measured on L-alpha peaks. Counting times on peak were 10 s, and 10 s total on background for all elements. Commercial "SPI" standards were used for intensity calibration. The standards for feldspar were orthoclase (Al, Si, K), albite (Na), rhodonite (Mn), MgF 2 (F), plagioclase (Ca), barite (Ba), SrTiO 3 (Sr) and kaersutite (Mg, Ti). Further standards for pyroxene included pyrope (Mg, Si), Ni-metal (Ni), Cr 2 O 3 (Cr), and kaersutite (Al). Unknown acquisitions were peaked on the calibration standards. The data were collected with JEOL software (PC EPMA 1.9.2.0), and its ZAF matrix correction algorithm (Heinrich/Duncumb-Reed with FFAST-2005 MACs) was applied to correct for differential matrix effects. Oxygen was calculated by stoichiometry. Potential analytical drift was monitored for by interspersing secondary standards during the analyses.

Isotopic determinations
A total of 21 samples were chosen for whole-rock Rb-Sr and Sm-Nd isotopic determinations. Approximately 100-350 mg of sample was weighed and different volumes of 85 Rb-84 Sr and 149 Sm-150 Nd spikes were added and then dissolved in a mixture of HF-HNO 3 . Sr and Rb separation took place in 2 ml BioRad AG50 × 8, 200-400 mesh resin columns. Sm and Nd were separated in 1 ml Eichrom Ln Spec 100-150 µm resin columns. The isotopic compositions of samples were measured in low resolution mode on a Nu Instruments Plasma II MC-ICP-MS, housed at the Spectrum Analytical Facility (University of Johannesburg). Backgrounds were measured and subtracted from the measurements. Rb analyses were collected for 20 cycles and Sr for 60 cycles, giving an internal precision of ˂ 0.001% (1 standard error) on fractionation-corrected 87 Sr/ 86 Sr. Sm measurements were collected for 20 cycles and Nd for 60 cycles, giving an internal precision of ˂ ± 0.002% (1 standard error). Total procedural blanks and two Certified Reference Materials (BCR-2 and BHVO-2) were included in the analyses to ensure quality and reproducibility (see Electronic Appendix B). Initial 87 Sr/ 86 Sr and 143 Nd/ 144 Nd ratios were calculated using decay constants of 1.42 × 10 -11 y −1 (Steiger and Jäger 1977) and 6.54 × 10 -12 y −1 (Begemann et al. 2001), respectively, for an age of 2.055 Ga (Zeh et al. 2015). Ɛ Ndi values were calculated using values for CHUR of 143 Nd/ 144 Nd = 0.512638 (Goldstein et al. 1984) and 147 Sm/ 144 Nd = 0.1967 (Jacobsen and Wasserburg 1980).

Petrography
A summary of the mineral modes, grain sizes and additional petrographic observations is provided in Electronic Appendix C. The variation in the modal mineralogical compositions of samples with depth is graphically depicted in Fig. 2. The petrography of samples from the Main Zone within the Eastern Limb has been extensively documented (e.g. von Gruenewaldt 1970Gruenewaldt , 1971Gruenewaldt and 1973von Gruenewaldt and Weber-Diefenbach 1977;Molyneux 1974;Marais 1977;Klemm et al. 1985). Microscopic observations reveal that the studied samples consist of 26 gabbronorites, 5 pyroxenites, 5 leucogabbronorites, 3 melagabbronorites, 3 anorthosites and 1 leucogabbro (Fig. 2). Two samples of iron-rich ultramafic pegmatoid (IRUP) were also encountered at depths of − 40.74 m and − 43.69 m, respectively. Lithologies in the Driekop drill core (BH7771) are dominantly composed of plagioclase, high-Ca and low-Ca pyroxene (orthopyroxene and /or inverted pigeonite). Accessory phases include mica (biotite and phlogopite), quartz and amphibole throughout the studied interval. Phlogopite and amphibole are especially abundant within the pyroxenites. Other minerals encountered include talc (as an alteration product of pyroxene) and chlorite (as an alteration product of plagioclase and pyroxene).

Gabbroic and anorthositic rocks
The majority of the rocks in the drill core are gabbroic to anorthositic as per mineral mode. These rocks generally have a spotted appearance resulting from the presence of subhedral to anhedral crystals of pyroxene in an intergranular texture ( Fig. 3a) or sub-ophitically enclosing plagioclase. Plagioclase is the main phase in these rocks with a modal abundance of 16-91%. It occurs as abundantly twinned laths ( Fig. 3b) frequently displaying preferential orientation. Wedge-shaped and bent plagioclase twin-lamellae ( Fig. 3b), indicative of post-crystallisation deformation (von Gruenewaldt 1971) are common. Plagioclase grain sizes vary from 1.2 mm to 5.9 mm. Ophitic to sub-ophitic textures are present in some samples, where plagioclase occurs enclosed by either orthopyroxene or clinopyroxene. In some samples, plagioclase chadacrysts are finer grained than unenclosed plagioclase crystals. Myrmekite is very common in the gabbronoritic lithologies and increases to up to 5 modal % in some parts. It generally occurs along plagioclase grain boundaries and in some cases as patches within plagioclase crystals (Fig. 3c).
Orthopyroxene modally constitutes 1-38% of samples below the Pyroxenite Marker and 14-37% of samples above the Pyroxenite Marker. It occurs as subhedral to anhedral crystals ( Fig. 3d) throughout most of the studied interval. Orthopyroxene grain sizes vary considerably within the gabbronorites, with an average grain size of ~ 2 mm below the Pyroxenite Marker and ~ 4 mm above the Pyroxenite Marker. Orthopyroxene normally contains an abundance of thin and closely spaced clinopyroxene exsolution lamellae. Below the Pyroxenite Marker, orthopyroxene containing thick, bleb-like exsolution lamellae of clinopyroxene is abundant (Fig. 3d). This inverted pigeonite (Poldervaart and Hess 1951) is abundant in the rocks below the Pyroxenite Marker and is replaced by primary orthopyroxene as the low-Ca pyroxene at approximately 2 m above the base of the Pyroxenite Marker. Its modal abundance within the Pyroxenite Marker itself is ~ 1%.
Clinopyroxene modally constitutes between 0 and 38% of the samples studied and occurs as anhedral to subhedral crystals between plagioclase laths. In some samples, ophitic textures were observed, with clinopyroxene oikocrysts enclosing plagioclase. Clinopyroxene displays the presence of orthopyroxene exsolution lamellae and is frequently twinned (Fig. 3d). Clinopyroxene exhibits distinct schiller structures throughout the investigated interval (Fig. 3d).
Pleochroic (brown to colourless) amphibole modally constitutes 0 to 21% of the samples studied and occurs mostly as patches around or near pyroxene crystals. It is particularly abundant within two altered samples of gabbronorite at depths of − 16.69 m and − 143.25 m. The amphibole seems to replace the pyroxene in these two samples. Amphibole also occurs in the vicinity of minerals such as quartz and biotite.
Quartz constitutes up to 13 modal % and occurs mostly interstitially. Biotite and phlogopite constitute up to 9 modal % and also tends to occur interstitially. A detailed study of the opaque minerals was not done, due to their limited occurrence. Opaque minerals constitute up to 1 modal % of the gabbronoritic samples. Other minerals encountered in the gabbroic samples include chlorite (as alteration product

Pyroxenitic rocks
Three prominent pyroxene-enriched horizons were observed in the studied interval. The lower occurrence (at depths 574.96-576.75 m) is an ~ 1.8 m-thick zone surrounded by gabbronorite. The second (514.70-518.53 m) is an ~ 3.8 m-thick zone also surrounded by gabbronorite. The Pyroxenite Marker will be discussed separately below. The sampled pyroxenite horizon (the lower occurrence at depth − 86.85 m) is dominated by poikilitic textures. Orthopyroxene modally constitutes 8% of this sample. Some of the orthopyroxene contains an abundance of thick clinopyroxene exsolution lamellae consistent with it being inverted pigeonite, modally constituting about 45% (Fig. 4a). The unit consist of orthopyroxene crystals that are up to ~ 6 mm in size. Orthopyroxene locally encloses small chadacrysts of clinopyroxene (Fig. 4b). Clinopyroxene modally constitutes about 20% of the sample and occurs as subhedral to anhedral crystals (Fig. 4b). Clinopyroxene displays the presence of orthopyroxene exsolution lamellae and is frequently twinned (Fig. 4a). Greenish pleochroic amphibole modally constitutes about 6% of the pyroxenite horizon and occurs mostly as patches in or as alteration rims on pyroxene crystals. Intercumulus plagioclase (Fig. 4b) modally constitutes about 11% of the pyroxenite sample. Brown to colourless pleochroic phlogopite (Fig. 4b) modally constitutes ~ 1% of the pyroxenite horizon and usually occurs as patches along the edges of orthopyroxene crystal boundaries.

Mineral chemistry
Compositional data for plagioclase, clinopyroxene and orthopyroxene are reported in Electronic Appendix D. Between three and four analyses of plagioclase, orthopyroxene and clinopyroxene were performed on each sample, with one analysis spot per crystal centred on the apparent core of each crystal. Due to the absence of visible zonation during petrographic examination of the samples, no attempt was made to study the presence of zonation through the analysis of multiple spots on individual crystals.

Plagioclase
The composition of plagioclase varies across the studied interval ( Fig. 2), with an average An% [An% = 100 × molar Ca/ (Ca + Na + K)] of 64 below the Pyroxenite Marker, 61 in the Pyroxenite Marker and 70 above the Pyroxenite Marker. The average anorthite contents for individual samples, with within-sample variation expressed as 1 σ-standard deviations, are between 55.1 ± 0.7 and 73.9 ± 0.2 below the Pyroxenite Marker, 59.7 ± 0.9-61.9 ± 1.3 in the Pyroxenite Marker and 64.2 ± 2.6%-72.8 ± 2.68% above the Pyroxenite Marker. Below the Pyroxenite Marker, an overall increase in the An% can be seen from 57. The Mg# of clinopyroxene parallels the orthopyroxene trend and is on average ~ 0.08 higher than that of the co-existing orthopyroxene. The average Mg# for clinopyroxene below, within and above the Pyroxenite Marker is 0.74 ± 0.03, 0.78 ± 0.01 and 0.77 ± 0.01, respectively. Selected incompatible and compatible trace element concentrations and ratios are presented in Fig. 6 as a function of stratigraphic height. Zr is fairly constant, at an average value of 9.0 ppm, in the gabbronoritic and anorthositic rocks covering the depth interval − 176.84 to + 14.17 m. A localised maximum in the Zr concentration, with a value of 23.9 ppm, is observed in the pyroxenite layer covering the depth interval − 86.72 to − 86.85 m. Fluctuations in the Zr concentrations are associated with the pyroxenite layers as well as the IRUP samples. In the Pyroxenite Marker, the Zr is fairly constant with an average value of 11.8 ppm.
The Sr/Al 2 O 3 ratio as a proxy for differentiation of plagioclase shows a step-wise decrease towards the Pyroxenite Marker, from a value of ~ 0.0015 at the base of the studied interval, to an average of ~ 0.0013 within and above the Pyroxenite Marker. The Cr/Sc ratio as a proxy for clinopyroxene differentiation shows very little variation from the base of the studied interval, with values << 5 and up to ~ 40 m below the base of the Pyroxenite Marker. It then increases rather abruptly to values > 20 towards the Pyroxenite Marker, a value which is also retained in samples above the Pyroxenite Marker. The abrupt increase in Cr/Sc ratios occurs across the level where inverted pigeonite is replaced by primary orthopyroxene as the dominant low-Ca pyroxene.

Isotope geochemistry
87 Sr/ 86 Sr i ratios and Ɛ Ndi values are presented in Table 1, and their variation with depth is displayed graphically in  Chondrite-normalised REE patterns of rocks across the Pyroxenite Marker interval as exposed in BH7771. Normalisation factors from Lodders (2003). The inset shows the range in REE concentrations for the Main Zone of the western Bushveld Complex at Union Section (Maier and Barnes 1998) and the range in REE concentrations of B3 magma (Harmer and Sharpe 1985).
Values of Ɛ Ndi show relatively little variation from the base of the studied succession to a depth of − 50.39 m, varying between − 6.2 and − 6.7. The next sample, at depth − 26.44 m, returned an Ɛ Ndi value of − 4.9. This is followed by a sustained decrease in the Ɛ Ndi values towards depth − 2.77 m, where Ɛ Ndi is − 6.8. From − 1.82 m to the lowermost sample of the Pyroxenite Marker, Ɛ Ndi increases from − 6.7 to − 6.0. From the base of the Pyroxenite Marker to the top of the studied interval, significant inter-sample variations appear to be the norm, with values fluctuating between − 5.8 and − 6.7.

Discussion
Three main models have been proposed to explain the observed compositional and isotopic variations across the Pyroxenite Marker interval. Sharpe (1985) attributed the Pyroxenite Marker interval to mixing occurring between underplated Main Zone magma and upwards displaced resident magma of Upper Critical Zone lineage. Cawthorn et al. (1991) suggested that the Pyroxenite Marker interval formed in response to mixing between approximately equal  Vantongeren and Mathez (2013) proposed that the melt that was added at the level of the Pyroxenite Marker was added as numerous, "small" injections that became homogenised, cooled and partially crystallised between successive injections. The implication of all of the cited models is that the rocks occurring across the Pyroxenite Marker interval, which is thought to represent the zone of mixing between resident and incoming melts, should exhibit isotopic compositions that may be modelled through simple twocomponent melt-melt mixing curves. As will become clear from the subsequent section, in which our data across the Pyroxenite Marker interval are compared to several meltmelt mixing models based on different assumptions from the published literature, none of the mixing models provide a particularly good fit to our data, necessitating a reassessment of the mechanism by which mixing took place and the nature of the end-members involved. Fig. 7 Variations in 87 Sr/ 86 Sr i and Ɛ Ndi (at 2.055 Ga) with depth across the Pyroxenite Marker interval in BH7771. *Numbers refer to depth in metres in borehole relative to the base of the Pyroxenite Marker. Data for the Western Limb from Cawthorn et al. (1991) and for the Eastern Limb from Sharpe (1985). Insets show close up of the area around the Pyroxenite Marker

Constraints on the Nd-isotopic composition and amount of melt/magma added at the level of the Pyroxenite Marker
To constrain the Nd-isotopic composition of the magma added at the level of the Pyroxenite Marker, the following additional parameters need to be assessed: (i) the Nd-isotopic compositions of the resident and mixed magmas, (ii) the Nd concentration of the resident and incoming or mixed magmas and (iii) the ratio in which the two magmas mixed, defined as: For the purposes of our calculations, the Nd-isotopic composition of the resident magma is taken as Ɛ Ndi = − 7.01 (the average value for the lower Main Zone as reported by Maier et al. 2000) and that of the mixed magma as Ɛ Ndi = − 6.43 (Tanaka and Masuda 1982). It should be noted that the latter value is not well constrained and is based on the sole available whole-rock Nd-isotopic composition currently available for the upper Main Zone (Vantongeren et al. 2010). As shown by our own data across the Pyroxenite Marker interval (Fig. 5), the REE contents of the resident and incoming magmas were likely rather similar. Similar Nd contents for the resident and incoming magmas allow for a simplification of the two-component mixing equation from: to The mixing ratio may be constrained with reference to the available Sr-isotopic data for the Pyroxenite Marker interval. Cawthorn et al. (1991) argued that the magma added at the level of the Pyroxenite Marker had an 87 Sr/ 86 Sr i ratio of 0.7063, similar to that of a magma intruded into the lower reaches of the western Bushveld Complex (Davies and Cawthorn 1984). On this basis, and on the assumption that the resident magma and the incoming magma had similar Sr contents, these authors could show that the volume of magma added at the level of the Pyroxenite Marker was approximately equal to that of the resident magma. If it is assumed that the Sr contents of the resident and incoming magmas were indeed similar, and setting 87 Sr/ 86 Sr Mix = 0.7073 (Kruger et al. 1987), 87 Sr/ 86 Sr Res = 0.7084 (Maier et al. 2000) ( (1 − f ) and 87 Sr/ 86 Sr Inc = 0.7063 (Cawthorn et al. 1991), equation 3 can be applied to the Rb-Sr system to show that the mass of the incoming magma was approximately equal to that of the resident magma ( f = 0.48) . The Nd-isotopic composition of the incoming magma is then constrained to Ɛ Ndi = − 5.9 (Fig. 8a). Vantongeren and Mathez (2013), using equilibrium Sr contents calculated from trace element contents of the minerals present across the Pyroxenite Marker interval, estimated an 87 Sr/ 86 Sr i ratio of 0.7066-0.7068 for the magma that was introduced at the level of the Pyroxenite Marker. Their calculations, performed on the assumption that no magma was lost from the chamber, show that the Sr content of the incoming magma (319 ppm) was likely significantly higher than that of the resident magma (181 ppm), such that the incoming magma had a much greater effect on the 87 Sr/ 86 Sr i ratio of the mixed magma than the resident magma. Using 87 Sr/ 86 Sr i ratios of 0.7067 (Vantongeren and Mathez 2013) and 0.7084 (Maier et al. 2000) for the incoming and resident magmas, respectively, and the aforementioned Sr contents a mixing ratio ( f ) of 0.49 is required to produce an 87 Sr/ 86 Sr i ratio of 0.7073 for the mixed magma, a result that is not significantly different from that produced when relying on the values of Cawthorn et al. (1991) as presented in the preceding paragraph. In the present scenario, the Nd-isotopic composition of the incoming magma is still constrained to Ɛ Ndi = − 5.9 when assuming similar Nd contents for the resident and incoming magmas (Fig. 8b). When using the calculated Nd contents for the resident (16 ppm) and incoming (31 ppm) magmas as calculated by Vantongeren and Mathez (2013), an Ɛ Ndi value of − 6.1 is obtained for the incoming magma (Fig. 8c). This value does not change significantly when using the equilibrium liquid compositions calculated by Vantongeren and Mathez (2013) for the resident and incoming magmas in the involving where magma loss.

Mechanism of mixing: mixing of minerals or melts?
As is clear from Fig. 8, none of the mixing models provide a particularly good fit to our data across the Pyroxenite Marker interval, with the majority of data points scattered about the mixing curves. We propose that this is the result of minerals having crystallised from the resident and incoming magmas, respectively, and having been incorporated into individual rock layers across the Pyroxenite Marker interval. The notion that rocks occurring within the Bushveld Complex might represent mixtures of minerals accumulated from different parts of a stratified magma column is not entirely novel. This was shown to be the case in the Merensky Reef (Prevec et al. 2005;Seabrook et al. 2006), in noritic blocks enveloped by anorthosite in the upper Main Zone (Bourdeau et al. 2022) and also across the Pyroxenite Marker interval, where Cawthorn et al. (1991) reported the existence of plagioclase primocrysts from both magmas occurring within single samples and envisaged that "composite packets of liquid plus crystals" from the incoming magma (intruded at an intermediate level) plunged through the column of resident magma to accumulate at the base of the chamber. It seems feasible that pyroxenes that crystallised from the resident and incoming magmas might have co-accumulated in a similar manner, although compositional evidence for this is likely to have been destroyed due to the rapidity of solid-state diffusion of Fe and Mg in pyroxene relative to that of plagioclase, in which coupled substitution of CaAl and NaSi has to take place (Cawthorn et al. 1991). Figure 9 shows the mixing model as presented in Fig. 8a with two additional mixing curves that constrain all of our data apart from two data points that we consider to be aberrant. The assumptions on which the upper mixing curve (curve C) is based are as follows: (i) rocks represent mixtures of plagioclase crystallised from the resident magma and clinopyroxene crystallised from the incoming magma, (ii) plagioclase and clinopyroxene have Sr concentrations of 457 ppm and 11.6 ppm, respectively and (iii) plagioclase and clinopyroxene have Nd concentrations of 1.4 ppm and 9.2 ppm, respectively. The Sr and Nd concentrations are averages from the work of Vantongeren and Mathez (2013) for rocks across the Pyroxenite Marker interval. The lower Fig. 8 Whole-rock isotopic mixing models for the Pyroxenite Marker interval. a This mixing scenario is based on the assumption that the resident and incoming magmas had similar Sr contents. A is the resident magma: 87 Sr/ 86 Sr i = 0.7084 and Ɛ Ndi = − 7.01 (Maier et al. 2000); and B is the incoming magma: 87 Sr/ 86 Sr i = 0.7063 (Cawthorn et al. 1991) and Ɛ Ndi = − 5.9. b This mixing scenario is based on the assumption that the Sr content of the incoming magma (319 ppm) was higher than that of the resident magma (181 ppm) as per Vantongeren and Mathez (2013). A is the resident magma: 87 Sr/ 86 Sr i = 0.7084 and Ɛ Ndi = -7.01 (Maier et al. 2000); and B is the incoming magma: 87 Sr/ 86 Sr i = 0.7067 (Vantongeren & Mathez 2013) and Ɛ Ndi = − 5.9. c This mixing scenario is based on the assumption that the Sr content of the incoming magma (319 ppm) was higher than that of the resident magma (181 ppm) and that the Nd content for the resident magma (16 ppm) is lower than the incoming magma (31 ppm), as per Vantongeren and Mathez (2013). A is the resident magma: 87 Sr/ 86 Sr i = 0.7084 and Ɛ Ndi = − 7.01 (Maier et al. 2000); and B is the incoming magma: 87 Sr/ 86 Sr i = 0.7067 (Vantongeren & Mathez 2013) and Ɛ Ndi = − 6.1 mixing curve is based on the assumption that rocks represent mixtures of plagioclase crystallised from the incoming magma and clinopyroxene crystallised from the resident magma. Orthopyroxene was omitted from our calculations for the sake of simplicity, but would have little effect on both the Sr and Nd isotopic compositions of the mixtures due to the low abundance (< 0.2 ppm) of both elements in it. Our data across the Pyroxenite Marker interval are not fully constrained if the incoming magma had an Ɛ Ndi value of -6.1 (Fig. 8c), and for that reason we propose that the incoming magma had an Ɛ Ndi value of -5.9 (Fig. 8a, b). This would imply that the Nd contents of the resident and incoming magmas must have been rather similar, in contrast to what has been proposed by Vantongeren and Mathez (2013).

Where and how did mixing take place?
Most models for the development of the Pyroxenite Marker interval assume that mixing of the resident and incoming magmas happened within the confines of the presently exposed Bushveld Complex magma chamber and that the incoming magma was added gradually to the chamber, either forming a layer that was stabilised at an intermediate level within a stratified magma column before being thoroughly mixed with the resident magma (Cawthorn et al. 1991) or as numerous, "small" injections that became homogenised, cooled and partially crystallised between successive injections (Vantongeren and Mathez 2013). A possibility that has not been explored is that the rocks occurring across the Pyroxenite Marker interval may represent mixtures of minerals and melts that underwent mixing during ascent and subsequent emplacement from a sub-compartmentalised staging chamber as envisaged by Roelofse and Ashwal (2012). In this scenario, the Main Zone beneath the Pyroxenite Marker interval could represent the product of the repeated intrusion of variably contaminated crystal mushes with an average 87 Sr/ 86 Sr i and Ɛ Ndi on the order of 0.7084 and − 7.01, respectively. Input and mixing during ascent between this magma and another crystal-laden magma with fundamentally different isotopic composition (e.g. 87 Sr/ 86 Sr i = 0.7063; Ɛ Ndi = − 5.9), resident in a different sub-compartment, could equally account for the present dataset, a finding which is consistent with that of Yao et al. (2021), who argued that the Rustenburg Layered Suite below the level of the Pyroxenite Marker was emplaced as numerous crystal slurries that obtained a variety of isotopic compositions as a result of interaction with chemically and isotopically distinct assimilants as they turbulently rose through transcrustal conduits before entering the magma chamber and being deposited on the chamber floor as masses of cumulates when magma velocity slowed.
The distribution of rocks dominated by plagioclase from the incoming magma and pyroxene from the resident magma versus plagioclase from the resident magma and pyroxene from the incoming magma appears difficult to reconcile with straightforward gravity-driven processes. Rocks occurring more than 10 m below the Pyroxenite Marker appear to be dominated by plagioclase from the resident magma and pyroxene from the incoming magma, whereas rocks within 10 m below the Pyroxenite Marker appear to be dominated by pyroxene from the resident magma and plagioclase from the incoming magma. Above the Pyroxenite Marker, rocks again appear to be dominated by plagioclase from the resident magma and pyroxene from the incoming magma. This Fig. 9 Mixing models for the Pyroxenite Marker interval in BH7771, showing that data may be constrained through the incorporation of minerals derived from the resident and incoming magmas, respectively, into rocks occurring across this interval, as opposed to only through the mixing of aphyric magmas. The central mixing curve is as per Fig. 8a. Mixing curve C represents mixtures of plagioclase crystallised from the resident magma and clinopyroxene crystallised from the incoming magma. Mixing curve D represents mixtures of plagioclase crystallised from the incoming magma and clinopyroxene crystallised from the resident magma. See text for details apparently random distribution of minerals constituting the different layers across the Pyroxenite Marker interval is entirely consistent with the model of Yao et al. (2021) as explained above, which would also explain the lateral and vertical differences in isotopic composition across the Pyroxenite Marker interval as is apparent from Fig. 7, where our own data across the interval are compared with that of Sharpe (1985) and Cawthorn et al. (1991).
Irrespective of the veracity of these arguments, which would need additional analytical work (e.g. in situ isotopic determinations or isotopic determinations on mineral separates) to further evaluate, the magma that was responsible for the observed isotopic shifts across the Pyroxenite Marker interval had a unique Sr and Nd isotopic composition ( Fig. 10) not exhibited by any of the rocks occurring below this level of the intrusion. The similarity in Ɛ Ndi values for the Upper Critical and Main zones, as is apparent from Fig. 10, along with the observed Nd-isotopic variations across the Pyroxenite Marker interval as presented in this study, argue against the model of Sharpe (1985), who attributed the Pyroxenite Marker interval to mixing between underplated Main Zone magma and displaced resident magma of the Upper Critical Zone.

Conclusions
The results of our modelling suggest that: • The magma added at the level of the Pyroxenite Marker had a unique Sr and Nd isotopic composition not seen in any of the layered rocks occurring below the level of the Pyroxenite Marker, with an 87 Sr/ 86 Sr i ratio of 0.7063-0.7067 and an Ɛ Ndi value on the order of − 5.9. The unique isotopic composition of the added magma argues against the model of Sharpe (1985), who attributed the Pyroxenite Marker interval as having formed in response to mixing between displaced Upper Critical Zone and underplated Main Zone magma. • The magma added at the level of the Pyroxenite Marker records evidence for a lower degree of crustal contamination compared to the resident magma. • The Pyroxenite Marker interval records evidence for the accumulation of minerals derived from both the resident and incoming magmas, which is inconsistent with the injection, homogenisation, cooling and crystallisation of numerous small pulses of magma as envisaged by Vantongeren and Mathez (2013). • The inferred presence of minerals with isotopic compositions pointing to derivation from either the resident or incoming magma co-existing in rocks of the Pyroxenite Marker interval could be explained either by the mixing of minerals settling through a stratified magma column, as envisaged by Cawthorn et al. (1991), or potentially through the intrusion and mixing of crystal-laden magmas with unique isotopic compositions from a sub-compartmentalised staging chamber during ascent, as envisaged by Roelofse and Ashwal (2012). Detailed in situ isotopic investigations, which fall beyond the scope of the present investigation, may assist in differentiating between these fundamentally different alternatives.

Supplementary Information
The online version contains supplementary material available at https:// doi. org/ 10. 1007/ s00410-023-01996-z. Funding Open access funding provided by University of the Free State. This study was funded by the NRF (National Research Foundation) and CIMERA (DSI-NRF Centre of Excellence for Integrated Mineral and Energy Resource Analysis).
Data availability All data generated over the course of this study are included in the published article and the accompanying supplementary information files.

Conflict of interest
The authors have no competing interest to declare that are relevant to the content of this article.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.