The Diamond-Poor Nature of Corundum-Bearing Eclogite and Karoo-Type Overprint of Deep Eclogite Sources Beneath the Kimberley Region (Kaapvaal Craton)

Abstract

Decades of studies have shown that the petrogenesis of kimberlite-borne cratonic eclogite and pyroxenite xenoliths reflects the endemics of their crustal protoliths and local lithosphere evolution. Detailed investigations of the origin and metasomatic history of individual eclogite xenolith suites are thus required to understand how cratonic eclogite reservoirs—and their diamond inventory—evolve in the regional tectonomagmatic context. Here, we investigate a little-studied eclogite and pyroxenite xenolith suite from the Balmoral kimberlite in the Kimberley area of the Kaapvaal craton, which, like eclogite suites in neighbouring kimberlites, likely originated as subducted Archaean oceanic crust. Detailed petrographic observations and mineral major- and trace-element analyses, combined with published data for eclogite xenoliths and eclogitic inclusions in diamond, show that this sample suite records at least two distinct episodes of metasomatic overprint: (1) Metasomatism by a kimberlite-like melt caused a decrease in clinopyroxene jadeite component and garnet grossular component and imparted high MgO and Cr₂O₃ contents, recorded dominantly by pyroxenite xenoliths. Comparison to Kaapvaal kimberlites and lamproites confirms that the geochemical trends cannot be reconciled with bulk kimberlite-eclogite mixing but require formation of diopside-rich clinopyroxene from the melt instead. This metasomatic style is recognised world-wide, and at Balmoral is notably restricted to the shallow lithosphere (150–200 km). (2) A distinct metasomatic event generated eclogites with extreme Y-HREE enrichment, at Balmoral restricted to the deep lithosphere (200–260 km). We propose that this enrichment style reflects garnet breakdown, with the liberation of these garnet-compatible elements to the metasomatic melt. This signature is identified in eclogite xenoliths both from early Cretaceous lamproite (Bellsbank) and late Cretaceous kimberlite (Balmoral, Kimberley) localities, tentatively ascribed to interaction with melts forming the Karoo large igneous province.

Balmoral corundum-bearing eclogites derive from depths overlapping high-Ca eclogites showing a Karoo-type overprint, which significantly diluted the Al₂O₃ content in the bulk rock and increased silica activity as gauged by the decrease in Al[IV] in clinopyroxene, thereby destabilising corundum. Craton-wide, preserved corundum-bearing eclogites record diamond-stable ƒO₂ and pressure conditions, yet show little compositional overlap with inclusions in eclogitic diamond. This may reflect the low propensity of COH fluids to reach carbon saturation in this lithology. The preserved corundum-bearing eclogites have reconstructed bulk major-element compositions, which, combined with small to absent Eu anomalies, suggest protoliths representing deep oceanic crustal (>0.5 GPa) cumulates of two pyroxenes plus only minor plagioclase that contained a significant melt component. The identification of such deep hybridised crustal rocks may reflect higher mantle potential temperatures and the formation of thicker oceanic crust in the Archaean.

INTRODUCTION

The mineralogy and chemical composition of eclogite xenolith suites from cratons globally have been thoroughly characterised, encompassing the Kaapvaal–Kalahari (Viljoen, 1995; Appleyard et al., 2007; Jacob et al., 2009; Aulbach & Viljoen, 2015; Shu et al., 2018; Radu et al., 2019; Le Roex et al., 2020; Aulbach et al., 2020a; Smart et al., 2021a, 2021b), Zimbabwe (Gurney et al., 1984; Deines et al., 1991; Viljoen et al., 1996; Aulbach et al., 2017a), Siberian (Jacob et al., 1994; Snyder et al., 1997; Misra et al., 2004; Agashev et al., 2018; Mikhailenko et al., 2020; Sun et al., 2020), West African (Barth et al., 2001, 2002), Slave (Heaman et al., 2006; Aulbach et al., 2007; Schmidberger et al., 2007) and Superior cratons (Smit et al., 2014). The petrogenesis of xenolithic eclogite, summarised in Aulbach & Smart (2023), is broadly understood, comprising, typically, generation in palaeo-spreading ridges, seawater alteration en route to the trench, metamorphism with dehydration and, for some, partial melting, followed by residence in the cratonic lithospheric mantle, which may be followed by multiple metasomatic episodes prior to entrainment in kimberlite magmas billions of years later (Barth et al., 2001; Barth et al., 2002; Jacob, 2004; Smart et al., 2009; Tappe et al., 2011; Aulbach & Jacob, 2016; Aulbach et al., 2020a; Korolev et al., 2021).

The multi-stage evolution and the wide range of possible protoliths of eclogite xenoliths, from melt-like to plagioclase-rich cumulates, result in a wide range of major and trace element compositions of mantle-derived eclogite xenoliths. This is further complicated by regional characteristics of the tectonomagmatic evolution of the cratonic mantle lithosphere hosting the eclogite, which is then sampled as eclogite xenoliths and within eclogitic diamonds (i.e. inclusions), or only the diamonds themselves. Thus, despite our broad understanding of eclogite petrogenesis, the study of individual eclogite xenolith suites adds rich detail to the variability of the processes occurring in the deep cratonic lithosphere and their link to the regional tectonomagmatic evolution. Moreover, some eclogite xenolith suites comprise samples with unusual mineralogy (e.g. zircon-, zoisite- or allanite-bearing specimens; Heaman et al., 2002; Shchukina et al., 2018; Radu et al., 2022; Mikhailenko et al., 2024), microstructures (e.g. exsolution features, marked mineral compositional zonation or secondary garnet rims; Alifirova et al., 2015; Korolev et al., 2021) or compositions (e.g. stepped REE patterns; Aulbach et al., 2011, 2020a) that are not adequately addressed in review papers that propose petrogenetic models based on broad common features of xenolithic eclogites globally.

Here, we investigated a suite of hitherto sparingly characterised eclogite and pyroxenite xenoliths from Balmoral in the Kaapvaal craton (Fig. 1), which is noteworthy for including corundum-bearing specimens. Schulze et al. (2003) previously reported an oxygen isotope datum for eclogitic garnet in an eclogite from Balmoral, suggesting that these eclogites originated from subducted oceanic crust. Mxinwa (2014) studied eclogites from the Balmoral region, interpreting them as products of subducted altered ancient oceanic crust that was subsequently metasomatised. Using new petrographic and mineral compositional data, in comparison with published data on both eclogite xenoliths and inclusions in eclogitic diamond, allows us to identify two distinct metasomatic events and link these to the regional tectonomagmatic evolution, to propose a more detailed model for the nature of the deep crustal protoliths of corundum-bearing eclogites and to assess the effect of their origin and evolution on the eclogitic diamond inventory of the subcratonic lithosphere.

GEOLOGICAL SETTING

The Phanerozoic evolution of the Kaapvaal craton is marked by three distinct episodes of magmatism. Firstly, comprising the Karoo large igneous province, which underwent eruptions during the mid to early Jurassic period, around 174–185 Ma (Jourdan et al., 2005, 2007). Following this, lamproites (or Group II kimberlites) intruded from ~115 to 200 Ma, while kimberlite magmas (or Group I or archetypal kimberlites) were predominantly emplaced between ~80 and 115 Ma (Allsopp & Barrett, 1975; Smith et al., 1985; Le Roex, 1986; Smith et al., 1994; Phillips et al., 1998, 1999). The eruption of Karoo magmas spanned a wide geographic area of southern Africa over a duration of approximately 5–6 million years, with peak volcanic activity occurring between 178 and 184 Ma. In contrast, lamproites and kimberlites represent shorter, punctuated magmatic events, each forming over a period of less than a few million years. Kimberlites within the Kaapvaal craton often share geographical proximity with lamproites.

The Balmoral kimberlite is located on the Kaapvaal craton (De Wit et al., 1992) in the Northern Cape Province of South Africa (Fig. 1a). It occurs approximately 80 km to the north-west of the town of Kimberley in close proximity (1 km) to the neighbouring Leicester kimberlite pipe (with a zircon U–Pb age of 93 Ma; Davis, 1977). These two kimberlite pipes were discovered in the early 1890’s and intermittently mined in the case of Leicester, with Balmoral having low diamond grade (De Wit et al., 2016). The Balmoral kimberlite contains a suite of mantle-derived xenoliths and xenocrysts, which include extremely altered peridotites, along with eclogites and megacrysts (Mathias et al., 1970; Schulze, 2001; Schulze et al., 2003; Schulze, 2023).

The Kaapvaal craton was formed by amalgamation of the 3.55 to 3.05 Ga Witwatersrand block in the east and the 3.15 to 2.9 Ga Kimberley block in the west (De Wit et al., 1992; Schmitz et al., 2004). These two terranes are separated by the 2.93 to 2.88 Ga Colesberg lineament, which represents a zone of subduction associated with craton assembly (Schmitz et al., 2004). The mantle lithosphere underlying the Kimberley region of the Kaapvaal craton has been variably metasomatised, with one metasomatic episode occurring at 180–190 Ma, which has been linked to the intrusion of the Karoo large igneous province (Giuliani et al., 2014).

SAMPLES AND PETROGRAPHY

A collection of 65 eclogite and pyroxenite xenoliths from the Balmoral kimberlite pipe was studied to understand the nature and evolution of the lithospheric mantle beneath the Kaapvaal craton in South Africa. These eclogites were categorised into bimineralic and corundum-bearing varieties based on the presence or absence of corundum. Bimineralic samples comprise 85% (55 out of 65) of the total xenoliths examined, while the corundum-bearing eclogites account for the remainder. The bimineralic eclogites are coarse-grained, granular rocks consisting of garnet and clinopyroxene plus minor secondary phlogopite and opaques (Appendix 1).

Under an optical microscope, garnet displays a variety of colours, ranging from dark brown to greyish brown, pale orange, orange, pale pink and pink (Fig. 2). Garnet grains range in size from 0.5 to 4 mm and are generally coarser than the coexisting clinopyroxene (Fig. 2). The grains are mostly subhedral to anhedral, with noticeable fracturing, and sometimes these fracture planes may be filled with serpentine or carbonate. Kelyphitic rims are commonly observed. Some grains show linear to curvilinear boundaries with shape-preferred orientation (Fig. 2b).

Clinopyroxene is typically dark green but may also range in colour from pale green to bright green, apple green and dull green (Fig. 2). The grain sizes vary from 0.3 to 2.5 mm (Fig. 2). Grain shapes are typically anhedral, often forming the matrix and intersected/filled by carbonate veins. They are extensively altered, with the degree of alteration in clinopyroxenes varying from 10 to 50% based on visual estimates.

Overall, bimineralic eclogites are characterised by a cumulate texture, though some samples feature an interlocking texture. As observed in eclogite studies at the Rietfontein kimberlite (Appleyard et al., 2007), the Kimberley area (Jacob et al., 2009), and other locations, exsolution lamellae of garnet within clinopyroxene are also present in a few samples of Balmoral bimineralic eclogites (Fig. 2b). The visually estimated average modal proportion of garnet to clinopyroxene is 60:40. However, we consider a ratio of 55:45 in these eclogites for reconstructing the whole rock compositions (rationalised and described below), as the higher garnet proportion might be due to the preferential survival of garnet-rich parts of the xenoliths during transport in the kimberlite and subsequent recovery processes at the ore treatment plant.

The corundum-bearing eclogites are characterised by relatively large subhedral garnet and clinopyroxene grains, with sizes ranging from 0.5 to 5 mm in diameter (Fig. 2e and f). Some of the specimens lack clinopyroxene and can be classified as ‘garnetite’ (Samples 8 and 83) (Fig. 2e; Appendices 1 and 2). The garnet is predominantly pale pink to honey brown in colour, while the clinopyroxene exhibits a pale green hue (Fig. 2e and f). These subhedral garnet grains are mostly embedded in a cloudy clinopyroxene matrix. Corundum appears in various shades of purple and pale pink and is smaller in size, with average diameters of up to 2.5 mm (Fig. 2e and f). The modal abundances of corundum in these samples range from 0.10 to 5 vol.%. Corundum typically exhibits a rounded subhedral shape and is at times also elongated in habit, and it is typically unaltered. In some samples, the orientation of corundum grains is noticeable, a characteristic also seen in corundum-bearing eclogites from sample suites across Southern Africa (Shee & Gurney, 1979; Jacob, 2004; Viljoen et al., 2005). Of all the eclogite samples collected at Balmoral, the corundum-bearing eclogites exhibit the most advanced alteration, particularly in the case of clinopyroxene. In some instances, clinopyroxene has undergone a complete transformation into an intensely dark green material, which is optically unidentifiable, akin to observations for some of the eclogites at the Koidu kimberlite complex in Sierra Leone (Hills & Haggerty, 1989) (Fig. 2f). Most of the corundum-bearing eclogites exhibit a cumulate texture, with corundum interlocked between the garnet grains. Appendix 1 provides a detailed petrographic description of each sample, and Appendix 2 lists the thick section scans for all samples.

The petrographic characteristics of Balmoral eclogites are broadly similar to those of other eclogites worldwide (Jacob, 2004). Kimberlite-borne eclogites typically host a diverse array of accessory phases. For instance, the Orapa kimberlite pipe features accessory phases such as diamond, corundum, coesite and kyanite (Shee & Gurney, 1979), while the off-craton Rietfontein kimberlite contains kyanite, ilmenite, rutile, sulphides, amphibole and phlogopite as accessory minerals (Appleyard et al., 2007). Eclogites and pyroxenites from the Kaalvallei kimberlite pipe exhibit accessory phases including diamond, spinel, corundum, and orthopyroxene (Viljoen et al., 2005). Additionally, corundum-bearing eclogite xenolith suites have been documented at Bellsbank, Roberts Victor and Doornkloof on the Kaapvaal craton (Bishop et al., 1978; Carswell et al., 1981; Smyth et al., 1991; Shu et al., 2016; Radu et al., 2019; Aulbach et al., 2020a; Smart et al., 2021a, 2021b).

SAMPLE PREPARATION AND ANALYTICAL TECHNIQUES

Sections of 30 mm thickness were prepared from these eclogite specimens for petrographic analysis (Appendix 1). Within the thin sections, specific minerals of interest, such as garnet, clinopyroxene and corundum, were identified and targeted for further analysis. In addition, grains of garnet, clinopyroxene and corundum were picked from crushed eclogite, with typically 3 grains selected, and mounted in resin blocks (Appendix 2). These resin blocks containing the extracted mineral grains were polished to create smooth surfaces suitable for insitu major and trace element analysis.

For the major and minor element oxides, the analyses were carried out using a Cameca SX100 electron microprobe housed at the University of Johannesburg’s central analytical facility (Spectrum). The following operating conditions were employed: Beam current of 20 kV, accelerating voltage of 20 nA and the magnitude of the incident electron beam was focused to 3 μm. On the peak, counting times were 22s for Na, 10s for Mg, 10s for Al, 8s for Si, 20s for K, 16s for Ca, 14s for Cr, 16s for Mn, 16s for Fe, and 50s for Ni. To measure the background, counting times were 44 s for Na, 20s for Mg, 20s for Al, 16 s for Si, 40s for K, 32s for Ca, 28s for Ti, 26s for Cr, 32s for Mn, 32s for Fe, and 50s for Ni. Detection limits for the measured oxides are as follows: 0.019 wt.% for Na, 0.019 wt.% for Mg, 0.016 wt.% for Al, 0.019 wt.% for Si, 0.014 wt.% for K, 0.022 wt.% for Ca, 0.024 wt.% for Ti, 0.034 wt.% for Cr, 0.036 wt.% for Mn, 0.035 wt.% for Fe and 0.013 wt.% for Ni.

At the spectrum facility of the University of Johannesburg, trace element analysis via laser ablation inductively coupled plasma mass spectrometry (LA-ICPMS) was conducted. This involved coupling a 193 nm ArF RESOlution SE Excimer Laser (Applied Spectra, Sacramento) with a Thermo Fisher iCAP RQ ICP-MS. Operating parameters included an 80 μm laser spot size, a 6-mJ beam energy with 25% attenuation, and a repetition rate of 10 Hz, achieving a fluence on the sample of 1.9 J/cm². The ablation took place within a Laurin Technic dual-volume cell (which keeps the sample volume small and constant) under a helium atmosphere, with a helium flow rate of 0.35 l/min, which was mixed with argon flowing at 0.8 l/min. Additionally, N₂ gas was introduced at a rate of 0.001 l/min into the laser signal before entering the ICPMS to enhance signal intensity. Analyte intensities were optimised before analysis while maintaining an oxide production rate < 0.8%. For standardisation, NIST612 standard glass served as the bracketing standard, while ²⁹Si and ²⁷Al were used as the internal standard for silicates (garnet and clinopyroxene) and corundum, respectively. Sets of 10 sample spots were bracketed by NIST612 analyses, supplemented by NIST610, USGS basalt glass BIR-1G and GHR1 (natural garnet standard; see Campbell et al., 1996; Viljoen et al., 2010) as secondary standards in order to monitor accuracy and instrumental reproducibility (Appendix 3).

Data processing was performed using the GLITTER software package (Van Achterbergh et al., 2001). The individual results, with comparison of BIR-1G to accepted values as reported on the GeoReM website (http://georem.mpch-mainz.gwdg.de in May 2024), are shown in Appendices 3–8. This suggests that accuracy is within 15% for all elements except Nb, La, Tb, Ho, Tm and Lu, whereas Ta, Th and U were not detected in BIR-1G, with correspondingly greater uncertainties attached to reported concentrations of these elements in the unknowns. Barium and Nb abundances were used to monitor for accidental co-ablation of minute amounts of kimberlite that can be present as an optically invisible component in intra-grain cracks, etc. This led to the exclusion of analyses where this component was detected, except in cases where only analyses with some kimberlite contribution were obtained. Following the rationale and mineral-kimberlite mixing calculations in Aulbach & Viljoen (2015; their Appendices B and E), the elements most affected by kimberlite admixing are Rb, Nb, Ta, Th and U in both clinopyroxene and garnet, as well as Sr, La, Ce, Pr, Nd and Pb in garnet, which are therefore discarded as likely contaminated.

SAMPLE CLASSIFICATION, NOMENCLATURE AND DATABASE

To understand the origin and evolution of Balmoral eclogite and pyroxenite xenoliths, it is essential to classify them. The Balmoral eclogites exhibit low concentrations of minor elements such as Na₂O in garnet and K₂O in clinopyroxene. Bimineralic eclogites from Balmoral typically have Na₂O contents in garnet up to 0.06 wt.% and K₂O contents in clinopyroxene below 0.1 wt.%. According to the classification by McCandless & Gurney (1989), these bimineralic eclogites are categorised as Type II eclogites. Similarly, corundum-bearing eclogites display Na₂O contents in garnet <0.09 wt.% and K₂O contents in clinopyroxene <0.08 wt.% also classifying them as Type II eclogites.

Following the classification scheme outlined by Aulbach & Jacob (2016), we first distinguished true eclogites (n = 56) from pyroxenites (n = 9) based on clinopyroxene Ca/(Ca + Na), which reflects the jadeite component and separates pyroxenites with jadeite-poor, diopside-rich clinopyroxene from true omphacite-bearing eclogites (Fig. 3a and b) (cf. Lee et al., 2006). Thereafter, the sum of the whole-rock heavy rare earth element (HREE) abundances (∑HREE; in μg/g, summed from Tb to Lu) against the whole-rock Eu/Eu^* (defined as chondrite-normalised Eu/(Sm*Gd)^0.5 > 1.05; chondrite of Sun & McDonough, 1989); Rudnick & Fountain, 1995; Fig. 3c and d) was used to identify eclogites with possible plagioclase-rich protoliths (gabbroic eclogites and gabbroic pyroxenites) from those having possible melt-like protoliths, given that REE are incompatible in accumulating plagioclase (and olivine) and typically anticorrelated with Eu/Eu^* as a result. We define gabbroic eclogites/pyroxenites as those having Eu/Eu^* >1.05, and all others as having melt-like or plagioclase-poor protoliths, which are then classified as high-Mg, low-Mg and high-Ca eclogites employing the Mg# (molar Mg/(Mg + Fe_total)) versus Ca# (molar Ca/(Ca + Mg + Fe_total + Mn)) in garnet (Fig. 3e and f). All new corundum-bearing eclogites also have high-Ca garnet, but two additionally have reconstructed bulk rocks with positive Eu anomalies qualifying them as gabbroic eclogites (Fig. 3e). Corundum-bearing eclogites are shown with distinct symbols in plots to track their composition relative to other eclogite and pyroxenite classes, in particular to corundum-free gabbroic eclogites and high-Ca eclogites.

For comparison purposes, we collated data for eclogite xenoliths, and classified them according to the scheme reviewed in the previous paragraph, from the following localities on the Kaapvaal craton from the literature, with a particular focus on, but not limited to, corundum-bearing suites: Bellsbank, Doornkloof, Kaalvallei, Kimberley, Lace and Roberts Victor (Carswell et al., 1981; McCandless & Gurney, 1989; Caporuscio & Smyth, 1990; O'Reilly & Griffin, 1995; Viljoen, 1995; Jacob et al., 2003; Viljoen et al., 2005; Jacob et al., 2009; Gréau et al., 2011; Huang et al., 2012; Aulbach & Viljoen, 2015; Shu et al., 2016; Radu et al., 2019; Aulbach et al., 2020a; Hardman et al., 2021) (some localities indicated in Fig. 1b). In addition, we compare xenolith data to inclusions in diamond from selected localities (Voorspoed, Finsch, Roberts Victor, Monastery, Bellsbank) on the Kaapvaal craton as published in the database of Stachel (2021). Many of the literature data do not have the trace element data needed for characterization according to Aulbach & Jacob (2016), in particular, REE that would allow to distinguish cumulate from melt-like protoliths, and they are therefore not further considered, with the exception of corundum-bearing eclogites and inclusions in diamond, which are shown with separate symbols in the diagrams. The trace-element database was purged of analyses showing Ba concentrations indicative of co-ablation of kimberlites (>1 μg/g in cpx, >0.5 μg/g in garnet; cf. arguments and rationalisation in Aulbach et al., 2020b). Although this does not lead to a perfect separation of contaminated vs. otherwise enriched samples, the step is taken to mitigate against the risk of interpreting potentially meaningless analyses.

RESULTS

Average major- and trace-element concentrations in garnet, clinopyroxene and corundum, are described below and listed in Tables 1–3.

Major element compositions

Garnet

Garnet compositions exhibit considerable variability, with Mg# ranging from 0.57 to 0.84, Ca# ranging from 0.10 to 0.50, and Cr# (molar Cr/(Cr + Al)) ranging from 0.001 to 0.04. Among these, garnet in pyroxenites has the highest median Mg# at 0.72, followed by gabbroic eclogites at 0.69 with the lowest Mg# in high-Ca eclogites (0.64). Pyroxenitic garnet also features relatively high Cr₂O₃ contents reaching up to 1.7 wt.%, with a median value of 0.28 wt.%. Garnet in corundum eclogites has the highest Ca# at 0.47, followed by gabbroic eclogites at 0.38, and lowest in pyroxenites at 0.12. Na₂O content ranges from 0.03 to 0.09 wt.%. Additionally, minor elements in garnet include TiO₂ (0.02 to 0.32 wt.%) and MnO (0.13 to 0.70 wt.%).

Compared to garnet from other Kaapvaal localities, Ca#–Mg# relationships of garnet in eclogite xenoliths from Balmoral show a more restricted compositional range, in particular lacking a population with low Mg#, while garnet with low Ca# is restricted to pyroxenites, similar to other localities (Fig. 3e). Garnet with intermediate Mg# and low Ca# that typifies high-Mg eclogites is absent at Balmoral save a single sample. It is conspicuous that garnet in corundum eclogites, both in this and other studies, shows scarce overlap in garnet inclusions in diamond from the Kaapvaal craton (Fig. 3f).

Clinopyroxene

Clinopyroxene exhibits a wide range of variability in Mg# (0.81–0.93), Cr# (0.002 to 0.19), and jadeite content (0.04–0.50). Clinopyroxene from gabbroic eclogites has the highest Mg# (median 0.89), followed by corundum eclogites (0.88), pyroxenites (0.87) and high-Ca eclogites (0.86). Clinopyroxene from corundum and gabbroic eclogites has the highest jadeite components (0.43 and 0.32, respectively), compared to pyroxenites (0.11). Clinopyroxene in pyroxenites contains the highest Cr₂O₃ content (median 0.21 wt.%), whereas clinopyroxene in corundum eclogites has the lowest content (0.05 wt.%), and contents are intermediate in high-Ca eclogites. FeO content ranges from 0.90 to 4.80 wt.%, with the highest median FeO in pyroxenites (4.12 wt.%) and the lowest in corundum eclogites (1.82 wt.%). K₂O content varies from b.d.l to 0.04 wt.%, with the lowest in clinopyroxene from high-Ca eclogites (b.d.l). TiO₂ content ranges from 0.03 to 0.65 wt.%, with the highest median TiO₂ in clinopyroxene from high-Ca and corundum eclogites (0.39 wt.%) compared to pyroxenites (0.20 wt.%). Similarly, Al₂O₃ content ranges from 1.50 to 17.17 wt.%, with the highest median Al₂O₃ in clinopyroxene from corundum eclogites (15.14 wt.%) and lowest in pyroxenites (3.01 wt.%).

When jadeite content in clinopyroxene from Balmoral eclogites and pyroxenites is compared against Ca# in garnet, high-Ca eclogites display a conspicuous linear trend, approximately connecting to corundum eclogite with high values and pyroxenites with low values for both ratios (Fig. 3a). When these two components are considered together, there is no overlap between corundum-bearing eclogites from this and other studies with inclusions in diamond (Fig. 3b).

Corundum

Average chemical analyses of corundum in corundum-bearing eclogites reveal dominant Al₂O₃ compositions. FeO content reaches up to 0.52 wt.%, while detectable levels of Cr₂O₃ range from 0.07 to 0.20 wt.%.

Trace element compositions

Garnet

The chondrite-normalised REE concentrations (denoted by subscript N) for garnet from Balmoral eclogites and pyroxenites vary by approximately one order of magnitude. Typically, the La_N values are subchondritic, with LREE_N concentrations that increase steeply up to Sm, then show weak positive slopes to Dy or Ho, then either continue to have a positive slope towards Lu (as in high-Ca eclogites), flatten out (as in pyroxenites), or decrease (as in gabbroic and corundum eclogites) (Fig. 4a). High-Ca eclogites thus have garnet with the highest MREE-HREE_N values. The REE_N patterns exhibit weak negative to positive Eu anomalies (Eu/Eu^* = 0.73 to 1.50) across all samples. Garnet in high-Ca and corundum eclogites shows the highest LREE_N values, with some (gabbroic and corundum eclogites) samples showing La_N > 1 while that in gabbroic eclogites from Balmoral displays the strongest positive Eu anomaly (Eu/Eu^* = 1.19). In gabbroic eclogites, garnet displays a distinct positive slope in its REE_N pattern from Sm to Lu, along with the highest median Ni content (44 μg/g) and lowest median Y (6 μg/g) and HREE abundances. Garnet from gabbroic pyroxenite (sample no. 52) is unique, with its pattern increasing steeply from La_N to Gd_N, then decreasing towards Lu. The median Cr concentration of garnet in all samples in this study is >100 μg/g; garnet in gabbroic eclogites collectively has lower Cr abundances (median 97 μg/g). Median V concentrations exceed 200 μg/g across all samples. Garnet contains significant amounts of Co, Ni, Sc and Y (with the exception of gabbroic eclogites, which have garnet with 6 μg/g median Y compared to 36 μg/g in garnet from high Ca-eclogites), as well as Zr (>10 μg/g), but <10 μg/g of other trace elements.

Clinopyroxene

Clinopyroxene in most samples in this study has La/Ce_N < 1, a maximum at Ce_N to Nd_N, and then a consistent decrease in HREE_N, which vary by about an order of magnitude between different samples (Fig. 4b). The Eu/Eu^* ranges from 0.86 to 1.26, with clinopyroxene in Balmoral gabbroic eclogites having the highest values overall (range 0.95–1.26). Clinopyroxene from pyroxenites has the highest REE_N concentrations, which also exhibits a maximum at a lighter REE (Ce or Pr) compared to clinopyroxene from other types of eclogites. Clinopyroxene in gabbroic eclogites shows the lowest Y and HREE_N abundances.

Corundum

The majority of the trace elements measured in corundum, including REE, are below the analytical detection limit, save for Ti (0.003 to 4 μg/g), V (0.001 to 1 μg/g), Cr (0.07 to 24 μg/g), Ni (0.001 to 0.19 μg/g) and Ga (0.001 to 1 μg/g).

Reconstructed whole-rock compositions

Rationale

Bulk rock compositions were reconstructed using the constituent mineral compositions, as is customary (Jerde et al., 1993; Barth et al., 2001; Heaman et al., 2006; Jacob et al., 2009). Reconstructed bulk rock compositions are typically employed to eliminate any contributions from secondary minerals, as well as garnet and clinopyroxene (largely filled fractures) altered by mantle metasomatism or infiltrated kimberlite melt. Although proportions of garnet and clinopyroxene may vary depending on factors such as sample size, mineralogical layering, accessory phases, compositional variability in protoliths, and pressure, bulk rock reconstruction was performed assuming a modal abundance of 55 and 45 wt.% for garnet and clinopyroxene, respectively, in all samples, which is considered appropriate for eclogites with picritic protoliths (Aulbach & Jacob, 2016). A fixed ratio was chosen because the small xenolith size relative to the coarse grain size precludes accurate determination on the section scale, even if this could be achieved with high precision. Moreover, a review of modal abundances in large eclogite xenoliths (>10 cm) and in experiments exploring the phase relationships of eclogite with MORB composition yields similar, subequal garnet-clinopyroxene proportions (Aulbach et al., 2020b). Eclogites classified as low-Mg and high-Mg (many of the latter actually pyroxenites on account of their Jd-poor clinopyroxene) have indistinguishable modal proportions, e.g. in the Koidu suite (Hills & Haggerty, 1989). Furthermore, variations in major element compositions relating to differences in protoliths are largely accommodated by both garnet and clinopyroxene upon subduction, not resulting in stark modal variability (Aulbach & Jacob, 2016). Thus, a single modal proportion was used for all samples in this study and from the literature. As the standard deviation in garnet-clinopyroxene modal abundances in large xenoliths reported in the literature is 10% (Aulbach et al., 2020b), we assign this as an uncertainty to the chosen modal abundances. The corresponding error bars are displayed in Fig. 5, illustrating that the compositional variation of reconstructed bulk rocks is much larger than this uncertainty.

Although the trace-element composition of corundum is difficult to quantify accurately using Si as an internal standard, the results do illustrate that this mineral is extremely trace-element-poor (most elements below detection limit, Appendix 3) and does not significantly contribute to the bulk-rock save to add Al₂O₃ and dilute the concentrations of all other elements. Hence, for corundum-bearing eclogites, the concentrations of all elements were adjusted downward using observed corundum modal abundances (Appendix 9), while Al₂O₃ concentration was increased accordingly. Abundance by weight was taken as equal to abundance by volume, which, given corundum abundances ≤5 vol.%, introduces insignificant uncertainty relative to the total uncertainty on reconstructed bulk compositions. Note that modal abundance estimates for corundum are typically not reported for literature data, which can generate offsets relative to Balmoral corundum-bearing eclogites for all components except Al₂O₃ where differences of several wt.% may reflect that corundum is not considered separately in bulk rock reconstruction. Because rutile was not considered in the bulk rock reconstruction, TiO₂ contents estimated from only garnet and clinopyroxene must be considered minima.

Major elements

Major-element compositions of reconstructed bulk rocks are displayed in Fig. 5. The MgO contents within the Balmoral eclogites and pyroxenites range from 8.16 to 19.20 wt.%. Notably, high-Ca, corundum-bearing, and gabbroic eclogites tend to have lower MgO (≤14.92 wt.%). Conversely, pyroxenites, including one gabbroic pyroxenite, have higher MgO contents (≥ 15.81 wt.%) and exhibit a distinct negative correlation with FeO, ranging from 4.94 to 10.33 wt.% in these samples, whereas in other samples (high-Ca, corundum-bearing, and gabbroic eclogites), FeO varies from 3.40 to 9.72 wt.% (Fig. 5a). The Al₂O₃ contents in pyroxenites, including one gabbroic pyroxenite, vary between 12.51 and 14.25 wt.%, while higher concentrations are observed in the remaining samples (high-Ca, corundum-bearing, and gabbroic eclogites), all forming a broad negative relationship with MgO (Fig. 5b). Concentrations of Na₂O show a similar distribution (Fig. 5c). Corundum-bearing eclogites exhibit the highest CaO contents, with a median of 15.53 wt.%, followed by gabbroic eclogites (15.42 wt.%), high-Ca eclogites (14.83 wt.%) and pyroxenites (11.67 wt.%) (Fig. 5d). In contrast, pyroxenites show the highest median SiO₂ contents at 47.04 wt.%, followed by gabbroic (46.68 wt.%), high-Ca (46.29 wt.%), and corundum-bearing eclogites (44.92 wt.%) (Fig. 5e). However, compared to modern mid-ocean ridge basalts (MORB) (Jenner & O'Neill, 2012), the Balmoral eclogites exhibit markedly lower average SiO₂ (46.28 vs. 53.74 wt.%), FeO (7.12 vs. 15.62 wt.%), and TiO_2min (0.20 vs. 3.01 wt.%), and higher MgO (12.18 vs. 9.66 wt.%) and CaO (14.28 vs. 9.66 wt.%) contents. Reconstructed whole-rocks have CaO/Al₂O₃ ratios (0.82–0.93) that overlap with MORB (CaO/Al₂O₃ = 0.35–2) and other bimineralic eclogites, e.g. from Siberian craton (CaO/Al₂O₃ = 0.4–1) (Ireland et al., 1994).

As described above, reconstructed bulk rocks of the main eclogite/pyroxenite classes at Balmoral—gabbroic eclogites and high-Ca eclogites, which comprise the corundum eclogites, and pyroxenites (Appendix 9)—show significant compositional differences, but their compositional distribution also differs from that in other eclogite suites. For example, craton-wide, low-Mg eclogites (absent at Balmoral) and high-Mg eclogites are much more abundant, while the Balmoral high-Ca suite occupies a sparsely populated space with respect to CaO vs. MgO content (Fig. S2 in Appendix 10), as also evident from grossular-in-garnet vs. jadeite-in-clinopyroxene relationships (Fig. 3a and b).

Trace elements

In terms of REE_N patterns, the reconstructed bulk compositions of eclogites and pyroxenites from Balmoral primarily correspond to two distinct groups. High-Ca eclogites and pyroxenites (Fig. 4d), tend to exhibit a characteristic low LREE_N/HREE_N ratio, manifesting a concave-down pattern. These samples generally have elevated MREE_N, with some high-Ca eclogites characterised by an exceptional enrichment in the HREE_N, displaying positive slopes between Tb_N and Lu_N. Conversely, corundum-bearing eclogites, two of which qualify as gabbroic eclogites on account of their positive Eu_N anomalies, and gabbroic eclogites, alongside the gabbroic pyroxenite (Fig. 4c), predominantly have sub-normal mid-ocean ridge basalt (NMORB) HREE_N concentrations and positive Eu anomalies typical of oceanic crustal cumulates. The other corundum-bearing eclogites from Balmoral do not have positive Eu anomalies and qualify as high-Ca eclogites, whereas the few corundum-bearing eclogites from the literature that have reported trace-element compositions, have variably strong positive Eu_N anomalies.

Certain samples within the corundum-bearing and gabbroic eclogite groups display steep increases between Er_N and Nd_N, with decreasing normalised abundances from Nd to La, resulting in a concave-down pattern (Fig. 4c). In the high-Ca eclogites, the abundance of Lu_N varies from roughly 3 to 100x chondrite, whereas for the other samples, it ranges from 1 to ~10x. Pyroxenites have the highest Ce_N abundances, followed by high Ca eclogites (median 10 μg/g) and corundum-bearing and gabbroic eclogites (median 7 μg/g each). Across all samples, the REE_N patterns exhibit relatively less variability in LREE_N and MREE_N, with high-Ca eclogites having a particularly wide range of HREE_N concentrations. Zr_N abundances are highest in the high-Ca eclogites (median 47 μg/g), followed by pyroxenites (34 μg/g), gabbroic eclogites (33 μg/g), and corundum-bearing eclogites (32 μg/g). With respect to chondrite-normalised reconstructed bulk-rock REE_N patterns, corundum-bearing and gabbroic eclogites (plus the gabbroic pyroxenite) have mostly sub-NMORB HREE_N concentrations, though in part showing steep increases between Er_N and Nd_N, but decreasing normalised abundances between Nd_N and La_N, creating a concave-down pattern (Fig. 4c). In contrast, high-Ca eclogites and pyroxenites, while also showing the concave-down pattern, have overall higher MREE_N and REE_N abundances and, for high-Ca eclogites, in part show an extreme enrichment in the HREE_N, with positive slopes between Tb_N and Lu_N (Fig. 4d). Comparison with other Kaapvaal eclogite xenolith localities reveals that this peculiar REE_N pattern is similarly observed at Kimberley and at Bellsbank (Fig. S3 in Appendix 10).

Trace element partitioning

The partitioning of trace elements between clinopyroxene and garnet can serve as a means to evaluate sample equilibrium. Trace element partitioning between clinopyroxene and garnet is determined by calculating the Nernst partition coefficients (^cpx/grtD_Element). From the experimentally derived partition coefficients (Green et al., 2000; Klemme et al., 2002), many incompatible elements (e.g. Sr, Ba, Rb) and LREE preferentially partition into clinopyroxene, and this is reflected by elevated D values for the partitioning between clinopyroxene and garnet in the eclogite and pyroxenite xenoliths under study. Fig. S1 in Appendix 10 shows that calculated ^cpx/grtD_Element values increase with decreasing garnet Ca#, consistent with prior observations (O'Reilly & Griffin, 1995; Harte & Kirkley, 1997). The highest ^cpx/grtD_Element are observed in clinopyroxene-garnet pairs within pyroxenites. As previously noted, pyroxenites are characterised by lower jadeite component in clinopyroxene and lower grossular component in garnet, decrease of the latter strongly inhibiting trace-element incorporation into garnet (Aulbach et al., 2020b). The partition coefficients in Fig. S1 in Appendix 10 identify samples 33, 81 and 97 as possibly not having achieved clinopyroxene-garnet trace element equilibrium.

Thermobarometry

Equilibrium temperatures and pressures were determined using the chemical compositions of coexisting garnet and clinopyroxene in the samples. Because tetrahedrally coordinated Al, the pressure-sensitive component of clinopyroxene (Beyer et al., 2015), limits accurate barometry on mantle eclogites, the equilibration temperatures determined for the eclogite were computed using the garnet-clinopyroxene Fe²⁺–Mg exchange geothermometer proposed by Krogh (1988), solved iteratively with the peridotite-derived regional steady-state conductive geotherm (rationale detailed in Aulbach et al., 2020b). Balmoral is close to the Leicester kimberlite, some 20 km NE of Barkley West in South Africa, which has been dated at 92 Ma (Becker et al., 2007), i.e. overlapping the timing of kimberlite magmatism on the Kaapvaal craton, suggesting a regional geotherm of 38 mW/m² according to the Hasterok & Chapman (2011) family of geotherms (which corresponds approximately to a 40 mW/m² geotherm according to the older Pollack & Chapman (1977) family of geotherms as published e.g. in Grütter, 2009).

The resultant equilibration temperatures are displayed in Fig. 6. The corresponding equilibration pressure estimates lie in the range of 6.0 GPa to 8.3 GPa for high-Ca eclogites, 6.8 to 7.4 GPa for corundum-bearing eclogites and 4.5 to 6.3 GPa for pyroxenites (Appendix 9). The distribution depths for the Balmoral eclogites, before their entrainment in the host kimberlite, were determined by using the equation [Z = –P−13.7) × 3.092 + 50] that was established for the Fennoscandian Shield by combining numerical models and peridotite xenolith data (Kukkonen et al., 2003), in which Z and P are the depth (km) and pressure (Kbar), respectively. Using this formula, the Balmoral eclogite and pyroxenite xenoliths equilibrated at a depth range of 148–265 km.

DISCUSSION

Balmoral eclogite and pyroxenite xenoliths: the case for an origin as variably metasomatised Archaean subducted oceanic crust

Although, overall, eclogite and pyroxenite xenoliths from Balmoral have no strong low-pressure signatures, Eu/Eu^* up to 1.4 are exhibited by a few corundum eclogites, and small Eu anomalies are shown (by definition) by all gabbroic eclogites (Fig. 4), thus pointing to plagioclase-rich oceanic cumulate protoliths. A partial compositional overlap with modern oceanic crustal cumulates is evident (Fig. 5). Plagioclase-rich precursors can explain why corundum-bearing eclogites from Balmoral, like those from other Kaapvaal localities, are characterised by mostly very high Al₂O₃ and Na₂O contents, high CaO contents, but low FeO and MgO contents (Fig. 5), and is consistent with the anticorrelation between Eu/Eu^* (increasing with increasing plagioclase mode) and REE abundances (excluded in accumulating plagioclase) (Fig. 3d; only HREE are summed to avoid effects of incompatible element enrichment). Gabbroic eclogites from Balmoral show some compositional overlap with corundum-bearing eclogites, and may sample a different portion of the intrusive part of oceanic crust. A similar conclusion was reached for eclogite xenoliths from Bellsbank, at a distance of just ~60 km from Balmoral, which also comprise corundum-bearing samples (Shu et al., 2016). Some of these share the peculiar HREE enrichment exhibited by many Balmoral samples (Fig. S3 in Appendix 10). In bivariate major-element diagrams (Fig. 5), high-Ca eclogites have high CaO and low FeO contents, and form trends originating from corundum-bearing eclogites, which are similarly CaO-rich, but FeO-poorer, and they therefore may be related to the latter via some igneous or metasomatic process, thus probably have oceanic crustal protoliths as well. Their possible origin is discussed in more detail below. Conversely, it is difficult to argue for an origin of eclogite xenoliths as mafic underplates (Aulbach & Smart, 2023), not least because many have non-mantle oxygen isotopic compositions that require interaction with seawater (Korolev et al., 2018). This includes eclogite xenoliths from Bellsbank, which feature low δ¹⁸O (+2.9 to +4.7; Neal et al., 1990) and a single sample from Balmoral with similarly low δ¹⁸O (+3.4; Schulze et al., 2003), typical of deep oceanic crust that experienced high-temperature seawater alteration (Gregory & Taylor Jr, 1981).

Balmoral eclogite xenoliths have not yet been directly dated. Eclogite xenoliths and eclogitic inclusions in diamonds from various kimberlites on the Kaapvaal craton yield Archaean ages, comprising clinopyroxene Pb–Pb, whole-rock Sm-Nd, whole-rock Re-Os and sulphide inclusion-in-diamond Re-Os isotope data (Jagoutz et al., 1984; Richardson et al., 2001; Menzies et al., 2003; Aulbach & Viljoen, 2015; Aulbach et al., 2019). Clinopyroxene specifically in some corundum-bearing eclogites from neighbouring Bellsbank has ⁸⁷Sr/⁸⁶Sr of 0.7008, requiring separation of the protolith from the convecting mantle in the Archaean (Shu et al., 2016). It is difficult to decrease the Sr isotope composition of a mantle rock during later melt-rock interaction, and such strongly unradiogenic Sr isotopic signatures are typically retained in the eclogites with the strongest protolith cumulate character (high Eu/Eu^*, low HREE abundances, therefore low initial abundances of the highly incompatible element Rb; Aulbach & Smart, 2023). On account of the mineralogical and geochemical similarity of the Balmoral and Bellsbank eclogite suites and the spatial proximity of their magmatic hosts, it is justifiable to propose that the Balmoral eclogite xenolith suite had Archaean oceanic deep crustal protoliths as well. If so, higher mantle potential temperatures, with formation of correspondingly thicker oceanic crust, may explain why corundum-bearing xenoliths have reconstructed major element compositions that do not fall on the main trends exhibited by modern oceanic crustal cumulates (Fig. 5).

The unusual REE patterns of eclogite and pyroxenite xenoliths from Balmoral, with common concave-down patterns in virtually all samples (Fig. 4e and f), are inconsistent with a simple origin as low-pressure oceanic crustal cumulates and melts, or as residues from partial melting, for example during subduction (Tappe et al., 2011). Indeed, given the Archaean age determined for Kaapvaal eclogite xenoliths, they may be susceptible to metasomatism due to prolonged residence in the cratonic lithospheric mantle, prior to entrainment in the host kimberlite and lamproite magmas, as previously discussed for other localities, notably Kimberley (Rehfeldt et al., 2008; Jacob et al., 2009; Shu et al., 2018), Roberts Victor (Huang et al., 2012, 2014), Doornkloof (Aulbach et al., 2020a) or Palmietfontein (Smart et al., 2021b).

Deep crustal protoliths of corundum-bearing eclogites: crystal mush-melt mixtures

Corundum-bearing eclogite xenolith suites on the Kaapvaal craton have been reported before, such as Bellsbank, Roberts Victor and Kronstadt/Kaalvallei (Bishop et al., 1978; Carswell et al., 1981; Smyth et al., 1991; Viljoen et al., 2005; Shu et al., 2016; Radu et al., 2019; Smart et al., 2021b) (Fig. 1b). Prior work has specifically identified deep crustal olivine plus plagioclase cumulates (troctolites) as the protoliths to corundum-bearing eclogites (Shu et al., 2016). Subchondritic Zr/Hf ratios in corundum-bearing eclogites, which are recognised also in high-Ca eclogites, but not in pyroxenites proposed to originate by mantle metasomatism of eclogite (see discussion below) (Fig. 7a), can be explained by accumulation of minerals with low D(Zr)/D(Hf), such as pyroxenes (Green et al., 2000) and plagioclase (Aigner-Torres et al., 2007). In contrast, strongly supra-MORB V/Sc at low REE contents, which appears to be a characteristic of some corundum-bearing eclogites (Fig. 7b), may require significant spinel accumulation in the protolith, as this characteristic, together with low MgO contents, is shared by some oxide gabbros in modern ridges (reported in Godard et al., 2009).

It is noteworthy that the SiO₂ concentrations of corundum-bearing eclogite xenoliths are partly lower than those of modern oceanic crustal cumulates (Fig. 5). Moreover, at Balmoral corundum-bearing eclogites lack marked positive Eu anomalies, which are, however, exhibited by those from Bellsbank. Metasomatism is known to mute crustal signatures, such as the Eu anomaly and non-mantle δ¹⁸O (Aulbach et al., 2020b), and all corundum-bearing eclogites from Balmoral are LREE-enriched. Nevertheless, samples that retained the characteristic low ∑HREE abundances (Fig. 4c and d) also do not display pronounced positive Eu anomalies, seemingly suggesting that plagioclase was not a major accumulating mineral. Some 25% of plagioclase may need to accumulate before a significant Eu anomaly appears (Schmickler et al., 2004). Therefore, we explore the viability of accumulation at a deep crustal pressure of 0.75 GPa, where orthopyroxene and clinopyroxene reach the liquidus before plagioclase, corresponding to a greater oceanic crustal thickness that is commensurate with higher Archaean mantle potential temperatures (Schmickler et al., 2004).

The compositions of two pyroxenes and plagioclase in equilibrium with a picritic melt undergoing fractional crystallisation at 0.75 GPa at an oxygen fugacity corresponding to the fayalite–magnetite–quartz (FMQ) oxygen buffer were modelled using the MELTs software (Ghiorso & Sack, 1995; Asimow & Ghiorso, 1998) (details on the modelling including starting composition in Aulbach & Jacob, 2016). The results are plotted against reconstructed bulk compositions of Balmoral eclogite and pyroxenite xenoliths and of corundum-bearing eclogites from the literature, bearing in mind that the concentrations of oxides other than Al₂O₃ in the latter may be overestimated by a percentage corresponding to that of the abundance of corundum (corundum modal abundances are not reported for literature samples). This shows that corundum-bearing eclogites can be described as mixtures of two pyroxenes and plagioclase precipitated at 0.75 GPa in terms of Al₂O₃ vs. CaO (Fig. 8a). However, their SiO₂ contents are too low and FeO contents in part too high to represent such mixtures (Fig. 8b and c), but could be explained by addition of melt to a deep crustal crystal mush, as suggested for modern deep oceanic crustal sections (Sanfilippo et al., 2015).

Extreme HREE enrichment in Kaapvaal cratonic eclogite: a distinct Karoo-type overprint?

High-Ca eclogites at Balmoral are clearly compositionally distinct from pyroxenites, which formed by interaction with kimberlite-like melts (representing a distinct metasomatic episode discussed in the next section), and they are furthermore derived from greater depths (~193–265 km), partially overlapping those of gabbroic (~208–2400 km) and corundum eclogites (~215–235 km). Given that their major element relationships are inconsistent with forward-modelled peridotite melt compositions (Fig. 5), the protoliths to high-Ca eclogites, which have no positive Eu anomalies, could be interpreted to have formed in plagioclase-poor parts of deep oceanic crust. Indeed, corundum-free high-Ca eclogites from Balmoral retain some compositional similarity to corundum-bearing eclogites, for example with respect to elevated CaO contents (Fig. 8a) and low Zr/Hf ratios (Fig. 7a), although they have significantly lower Al₂O₃ content in the reconstructed bulk rock (Fig. 8a). Increased silica activity compared to corundum eclogites, as gauged by the decrease in Al[IV] in clinopyroxene as Y abundances increase (Fig. 7c), may have caused destabilisation of corundum, which is consistently absent from high-Ca eclogite xenoliths from Balmoral.

High-Ca eclogites also show narrow trends in plots of Al₂O₃, Na₂O and CaO vs. MgO, whereas deep oceanic crustal sections show much more compositional scatter (Fig. 5b-d). Furthermore, many high-Ca eclogites show extreme HREE-enrichment with strong positive slopes in HREE_N (Fig. 4d), unlike examples of oceanic gabbro or troctolite (Fig. 4c). Although melt extraction in the eclogite facies, with some 50% of garnet present, will result in bulk distribution coefficients for the HREE >1 (Barth et al., 2002; Klemme et al., 2002), even 30% melt extraction from eclogite will not significantly enrich the residue in compatible elements (inset Fig. 4f), and so is not a feasible process for the observed HREE enrichment. Alternatively, the high HREE concentrations in the reconstructed bulk rocks might result from overestimated garnet modes, but this does not explain why the garnet themselves are extremely HREE-rich (Fig. 4a). Furthermore, the visual estimates of modal abundances for high-Ca eclogites are not consistently different from those for other samples (Appendix 1), and reconstruction of the bulk rock with as much as 80% clinopyroxene and only 20% garnet does not remove the strong positive slope in the HREE patterns of the most enriched samples (Fig. S4b). Besides, various estimates for garnet-bearing cumulates that could hypothetically have formed in deep continental arcs or in crustal underplates (Lee et al., 2006; Jagoutz & Schmidt, 2013; Chin, 2018), then foundered and trapped in the lithospheric mantle, are too MgO-rich or too FeO-rich (Fig. 5a) to explain the HREE-rich yet FeO-poor compositions of high-Ca eclogites via accumulation of garnet. Finally, clinopyroxene in both high-Mg and low-Mg cumulates in garnet-clinopyroxene rocks from continental arcs is jadeite-poor, making them pyroxenites, in contrast to the omphacite-bearing high-Ca eclogites in this study (Fig. 3a). These arguments lead us to consider a metasomatic origin of high-Ca eclogites.

Neither the range of HREE concentrations reported for Kaapvaal late Cretaceous kimberlites and early Cretaceous lamproites (Becker & Le Roex, 2006) nor that reported for basaltic lavas from the older, ~180 Ma Karoo continental flood basalt province (Jourdan et al., 2007), approach those of the most strongly HREE-enriched high-Ca eclogites from Balmoral, which are up to three times higher (Fig. 4d). Hypothetical melts in equilibrium with garnet in high-Ca eclogites exhibit REE patterns with a strong depression in the MREE that have no equivalent in natural melts (Fig. S4a in Appendix 10). Accordingly, bulk mixing of corundum-bearing and gabbroic eclogites with either kimberlites/lamproites or Karoo lavas cannot produce high-Ca eclogites because the magma compositions are both too high in FeO, and Karoo lavas are additionally too low in CaO and MgO and too high in SiO₂ content (Fig. 8). In sum, a mechanism is required that does not directly involve garnet precipitation from or equilibrium with common melts. We therefore tentatively describe this as due to garnet breakdown, with liberation of HREE and Y, elsewhere in the cratonic lithosphere and mobilisation with a metasomatic melt. A role of garnet is supported by the observed decrease of Y/Yb as HREE abundances increase (r² = 0.86, n = 31; Fig. 7d), due to D(Y) being only ~0.5 times D(Yb) (Green et al., 2000). Because HREE released via the hypothesised garnet breakdown would not be able to remain enriched in the melt as it travels through a garnet-bearing matrix, we envisage that the source must have been proximal, and/or that direct contact of melt with garnet in the HREE source region was minimised by formation of kelyphite rims in response to melt-mediated heating.

The anomalous HREE-enrichment is not only observed in eclogite xenoliths from late Cretaceous kimberlites over a distance of nearly 100 km (Balmoral, Kimberley), but also in those from the older Bellsbank lamproite (118 Ma; Smith et al., 1985), providing a minimum age estimate for this event. Metasomatism in Kimberley eclogite xenoliths has been linked to the same metasomatic overprint that is recorded in peridotites (Jacob et al., 2009). Titanates in peridotites from Bultfontein, in the Kimberley area, yield U–Pb ages of ~180–190 Ma and therefore provide evidence for strong interaction of the regional deep lithospheric mantle with Karoo-aged melts, which added HFSE (Ti, Zr, Hf, Nb), LILE (K, Ba, Sr) and LREE at a pressure interval of ~3–4 GPa (Giuliani et al., 2014). These element groups are also enriched in high-Ca eclogites (Fig. 4f), but the metasomatism generating high-Ca eclogites occurred deeper (>4.5 GPa for a 38 mW/m² conductive geotherm and temperatures displayed in Fig. 6), and it therefore may represent the deep lithospheric expression of interaction with Karoo-type melts.

A group of high-Ca eclogites with relatively high MgO contents (~12–15 wt.%) may have experienced superposition of two metasomatic styles, i.e. HREE enrichment followed by interaction with kimberlite melt. This becomes apparent from a closer look at their minor and trace element compositions. For example, they have higher Cr₂O₃ contents than high-Ca eclogites with lower MgO contents (Fig. 9b), and plot on trends of TiO₂ or MnO vs. ∑HREE defined by pyroxenites (Fig. 9c and d). Conversely, in bivariate plots involving minor and trace elements, high-Ca eclogites that do not have suggested superposed metasomatic signatures often show clearly distinct trends from pyroxenites. For example, TiO₂, MnO and Sc contents increase less strongly in high-Ca eclogites than in pyroxenites with increasing ∑HREE concentrations (Fig. 9a, c and d).

Precursory metasomatism by kimberlite-like melt

Typical signatures of kimberlite metasomatism in eclogite xenoliths, such as high MgO and Cr₂O₃ contents, and an anticorrelation of FeO and MgO at MgO content >15 wt.%, interpreted as a metasomatic clinopyroxene addition trend (Aulbach et al., 2020b), are observed in Balmoral xenoliths classified as pyroxenites (Fig. 5e). These enrichment trends cannot be mediated by bulk mixing of kimberlite with cratonic eclogite because, although kimberlites and lamproites are sufficiently MgO-rich to cause the observed increase in MgO in pyroxenites (and likely some eclogites), the Cr₂O₃ content in pyroxenites is almost 1.5 wt.%, whereas the median concentration in Kaapvaal kimberlites and lamproites is just 0.24 wt.% (Becker & Le Roex, 2006). At the same time, these kimberlite and lamproite melts are too FeO-rich to explain the low FeO content and observed anti-correlation with MgO contents in the pyroxenites. This is especially obvious when pyroxenites from other Kaapvaal kimberlites are included in the consideration (Fig. S2e in Appendix 10), which are suggested to also have formed by interaction with kimberlite-like melt (Aulbach et al., 2020b). Moreover, at Balmoral, SiO₂ concentrations tend to be highest in pyroxenites compared to eclogites, which is difficult to reconcile with the SiO₂-poor nature of kimberlites and lamproites (22.5–39.0 wt.%; Becker & Le Roex, 2006). Conversely, addition of metasomatic diopside-rich clinopyroxene from a kimberlite-like melt can satisfy these observations (Aulbach et al., 2020b). We note that high-temperature Cr-poor clinopyroxene megacrysts, such as those from Orapa, which have trace-element compositions indicating equilibrium with kimberlites, have major-element compositions that are too rich in silica and poor in alumina to represent the added component (Fig. 8), which reflects that they do not derive from metasomatised eclogite sources but rather from deep kimberlite intrusions (Nkere et al., 2021).

We suggest that (high-Ca) corundum-bearing and gabbroic eclogites represent the least metasomatically overprinted lithology in the Balmoral suite, on account of their preserved low HREE abundances (Fig. 4c) and low MgO contents (Fig. 5). In this case, it seems likely that the kimberlite-related metasomatic overprint post-dated the HREE-enrichment observed in high-Ca eclogites. This is because high-Ca eclogites plot in a major-element compositional space intermediate between pyroxenites—corresponding to rocks strongly metasomatised by kimberlite-like melts—and corundum-bearing and gabbroic eclogites (Fig. 5d). The inferred latest enrichment by kimberlite-like melt is consistent with suggestions that the successful eruption of magmas is preceded by metasomatism that conditions the initially cold and refractory continental lithosphere (Wass & Rogers, 1980; Griffin et al., 1999; Giuliani et al., 2013; Yaxley et al., 2017). The relatively young nature of this precursory metasomatism relative to eruption of the host magma is supported by occasional preservation of geochemical heterogeneity (zoned garnet) and disequilibrium microstructures (distinct high-Mg garnet mantles on low-Mg garnet cores; Hills & Haggerty, 1989; Korolev et al., 2021). Of note, the two distinctly metasomatised lithologies at Balmoral—pyroxenites and high-Ca eclogites—show virtually no overlap with respect to their temperatures of residence in the mantle (Fig. 6). That pyroxenite-related metasomatism at Balmoral occurred at relatively shallow depth (150–200 km) compared to the other lithologies is consistent with observations for eclogite and pyroxenite suites globally (Aulbach et al., 2020b).

Why are diamonds rare in corundum-bearing eclogites?

The estimated depths of formation for corundum-bearing eclogites at Balmoral are greater than the depth of the graphite-diamond boundary of Day (2012), which suggests that diamond should be stable from a pressure point-of-view (corresponding to high temperatures along steady-state conductive geotherms; Fig. 6). Oxygen fugacity (ƒO₂) has by now been estimated for a large number of eclogite xenoliths from the Kaapvaal craton, placing most samples firmly within the stability field of reduced carbon (Aulbach et al., 2017b; Burness et al., 2020; Smart et al., 2021a, 2021b). Moreover, the effect of silica-undersaturation, typical for corundum-bearing eclogites, has been assessed in Aulbach et al. (2017a) and Smart et al. (2021a), and suggests that, when silica activity is known and corrected for, corundum-bearing eclogites are not more oxidising than other eclogite varieties. In sum, corundum-bearing eclogites record pressures and ƒO₂ permissive of diamond stability. Nevertheless, there is almost no compositional overlap between garnet and clinopyroxene inclusions in diamond, or in diamondiferous eclogites and those forming corundum-bearing eclogites (Fig. 3).

Corundum inclusions in diamond are classified as ‘rare’ in both Smith et al. (2022) and in Stachel et al. (2022), whereas 33% of all inclusion-bearing diamonds are eclogitic (Stachel & Harris, 2008), far exceeding the expected abundance of eclogitic substrates in the lithospheric mantle. Combined, these observations imply that diamond formation is extremely efficient in eclogite in general, but highly inefficient in corundum-bearing eclogite in particular, notwithstanding isolated reports of corundum inclusions (Hutchison et al., 2004). Schulze et al. (2000) note that at Roberts Victor, there is almost no overlap of textures and compositions between coesite-bearing and corundum-bearing eclogites, whereby the latter all qualify as Group II eclogites in the textural and compositional classification of McCandless & Gurney (1989), which are typically barren, whereas coesite-bearing eclogites tend to be Group I and may be diamondiferous. The scarce reported overlap between corundum- and coesite-bearing eclogites likely reflects the low silica activity in the former. The lack of compositional overlap between Balmoral corundum-bearing eclogites and inclusions in diamond (Fig. 3) may reflect that diamond-forming fluids do not interact with deep subducted oceanic crustal sections (Stachel et al., 2022) and/or that the compositions are not favourable to carbon reaching saturation in the fluids. It is by now widely documented that diamond formation is a metasomatic process dominantly involving aqueous COH fluids near the H₂O-maximum (Aulbach & Stachel, 2022; Stachel et al., 2022). Although more experimental constraints are needed, thermodynamic models show that the position of the carbon saturation surface in COH fluids is strongly affected by the presence of solutes and in particular by silica (Tumiati et al., 2017). We therefore hypothesise that the low silica activity in corundum-bearing eclogites indicates conditions that are not generally conducive to diamond formation via saturation in a COH fluid. If our hypothesis—that many high-Ca eclogites originated as corundum eclogites that were severely melt-metasomatised during a Karoo-related event—is correct, then the original fraction of diamond-unfriendly corundum-bearing eclogites in the regional lithosphere may have originally been much larger than what constitutes the entrained eclogite xenolith suite. This may help explain why diamond grades both at Balmoral and at Bellsbank reported by De Wit et al. (2016) are low, despite eclogite representing an otherwise excellent diamond source rock.

SUMMARY AND CONCLUSIONS

We investigated the petrography, major, and trace element compositions of 65 eclogite xenoliths (both bimineralic and corundum bearing) retrieved from the Balmoral kimberlite within the Kaapvaal craton to understand their origin and evolution. These xenoliths are dominantly categorised into gabbroic eclogites, high-Ca eclogites, corundum-bearing eclogites, and pyroxenites. They exhibit absent to weak negative and positive Eu anomalies, with Eu/Eu^* values up to 1.4. By combining our findings with published data, we identify compositional distinctions and interpret them within the regional tectonomagmatic context, as follows:

• (1) Corundum-bearing eclogites classify as gabbroic (when Eu/Eu^* is ≥1.05) or high-Ca eclogites (Eu/Eu^* <1.05) and have high Al₂O₃ and Na₂O contents and the lowest SiO₂ contents within the suite. They have reconstructed bulk compositions possibly indicating protoliths of deep oceanic crustal cumulates (>0.5 GPa) consisting of two pyroxenes with only minor plagioclase (hence no or small positive Eu anomalies), but also require a melt component, similar to crystal-melt mushes in shallower modern deep oceanic crust. Such deep hybridised crustal rocks likely reflect the formation of thicker oceanic crust during the Archaean era facilitated by elevated mantle potential temperatures.

• (2) Gabbroic eclogites exhibit significantly higher SiO₂ and slightly higher MgO contents than corundum-bearing eclogites, though there is some overlap, suggesting the presence of corundum in certain samples that is not exposed at the section scale.

• (3) High-Ca eclogites display tight linear relationships in MgO–Al₂O₃–Na₂O–CaO space, and are characterised by in part extreme enrichment in HREE, with positive slopes between Tb and Lu. This metasomatic style is restricted to the deep lithospheric realm (~200–260 km depth), implying a distinct high-temperature metasomatic event. The HREE-rich signature is suggested to reflect garnet breakdown in proximal parts of eclogite bodies within the cratonic lithosphere, with consequent liberation of the garnet-compatible elements to the metasomatic melt. Extreme HREE enrichment is observed in eclogite xenoliths both from early Cretaceous lamproite (Bellsbank) and late Cretaceous kimberlite (Balmoral, Kimberley) localities, and is tentatively linked to deep (195–265 km) interactions with melts involved in the formation of the Karoo large igneous province.

• (4) High-Ca eclogites show compositional trends originating from the compositional field of corundum-bearing eclogites and derive from depths overlapping those of corundum-bearing eclogites. However, they exhibit decreased Al₂O₃ content and increased silica activity as evident by the decrease in Al[IV] in clinopyroxene, which may have caused destabilisation of corundum if originally present.

• (5) Pyroxenites are distinguished by high MgO and Cr₂O₃ contents and relatively high FeO content, which decreases as MgO increases. These are hallmarks of metasomatism by kimberlite-like melt as a typical precursor to host magma eruption and therefore likely post-date the HREE enrichment that characterises corundum-free high-Ca eclogites. Pyroxenites are confined to relatively shallow lithospheric depths (~150–200 km), as observed for xenolith suites globally.

• (6) Given that corundum-bearing eclogites record diamond-stable ƒO₂ and pressure conditions, the lack of compositional overlap with diamondiferous eclogites and eclogitic inclusions in diamond is striking. This may reflect inefficient carbon saturation in COH fluids within silica-poor and strongly aluminous lithologies and may explain poor diamond grades in the two neighbouring Kaapvaal localities, Balmoral and Bellsbank, where corundum-bearing eclogites form a conspicuous component of the eclogite xenolith suite.

Acknowledgements

We thank Matthew Hardman and Nester Korolev for their highly detailed, insightful and constructive feedback, which significantly improved this paper, and the editor Carl Spandler for consummate editorial handling and challenging comments. JP is thankful for receiving a postdoctoral fellowship from the South African National Research Foundation and extends thanks to the Kreitman School of Ben Gurion University of the Negev for awarding him a postdoctoral fellowship. SA gratefully acknowledges generous funding by the German Research Foundation (Deutsche Forschungsgemeinschaft) under project number 467591567. This study was also supported with funds granted to FV by the South African Department of Science and Innovation under their Research Chairs initiative (geometallurgy), as administered by the National Research Foundation.

DATA AVAILABILITY

The data underlying this article are available in the article and in its online supplementary material as well as in the Earthchem Library ( 10.60 520/IEDA/113547).

Handling Editor: Dr. Carl Spandler

REFERENCES