Controls on Pt/Pd ratios in Bushveld magmas and cumulates: a review complemented by new W isotope data

The Bushveld Complex of South Africa is underlain by a fine-grained sill complex which most workers interpret to represent the quenched parent magmas to the intrusion. The sills have unusually high Pt contents (up to ~ 25 ppb) and Pt/Pd ratios (average 1.50) exceeding those in most other mantle magmas globally. Unusually high Pt/Pd is also found in many Bushveld cumulates. Understanding the origin of the high Pt/Pd is important for exploration, in view of the contrasting monetary value of the metals, but also for unravelling the petrogenesis of the intrusion. Here, we review existing platinum-group element (PGE) data and present the first radiogenic W isotope data on a Bushveld rock, to evaluate a range of potential models, including PGE fractionation prior to final magma emplacement and within the Bushveld magma chamber, magma derivation from the sub-continental lithospheric mantle (SCLM), contamination of Bushveld magma with Pt-rich continental crust, and a meteoritic component in the mantle source to the magmas or in the crust with which the magmas interacted. We identify three key processes causing fractionation of metals prior to final magma emplacement and within the Bushveld chamber, namely crystallisation of Pt alloys, partial melting of cumulus sulfides triggered by flux of volatiles followed by sulfide melt percolation, and mobilisation of PGE by percolation of volatiles through the cumulate pile. The currently available W and Ru isotope data are inconsistent with derivation of the Bushveld magmas from mantle or crustal sources containing an enhanced meteoritic component relative to normal post-Hadean mantle.


Introduction
The Bushveld Complex represents one of the largest magmatic events on Earth, comprising the world's largest mafic-ultramafic layered intrusion as well as vast felsic and mafic lava flows and granite sheets. The Complex hosts the bulk of global platinum-group element (PGE) resources occurring in so-called reefs, i.e., broadly stratiform layers enriched in sulfide and sulfide and platinum-group minerals (PGM) (Smith and Maier 2021). Despite a century of research, the petrogenesis of the reefs remains debated. In part, this is because, for most layered intrusions, the nature of the parent magma(s) is unknown. The Bushveld Complex is an exception in that it is underlain by a suite of fine-grained sills that are considered to represent the parent magmas of the mafic-ultramafic portion of the Complex. Compared to global mafic-ultramafic magmas, the Bushveld magmas have unusually high Pt/Pd ratios averaging 1.5 . The specific reasons remain under debate (e.g., von Gruenewaldt and Merkle 1995;Cawthorn et al. 2002;Kinnaird and McDonald 2018). An improved understanding is important for exploration and mining as Pd is currently more than twice as valuable as Pt (~ 2100 USD/ oz Pd vs. ~ 900 USD/oz Pt, September 2022). Also, understanding the origin of the variation in Pt/Pd can potentially place added constraints on petrogenetic models for layered intrusions and on the compositional evolution of the Earth's mantle.

Pt/Pd and Au/Pt ratios in Bushveld magmas
Global komatiites and basalts that crystallised while remaining undersaturated in sulfide melt typically have up to ~20 ppb Pd and Pt (Fig. 1). Pt/Pd ratios cluster around or just below unity, except for relatively evolved lavas many of which tend to have somewhat lower Pt/Pd. The magmas show a subtle trend of increasing Pd and Pt with falling MgO resulting from incompatibility of Pd and Pt into silicate and oxide minerals (Barnes and Ripley 2016). The observed scatter may be due to a combination of mobility of PGE in hydrothermal and low-T fluids, and equilibration with immiscible sulfide melt (notably in some of the relatively evolved samples having < 10 wt.% MgO). In addition, there is a temporal control, in that Archean komatiites tend to have lower PGE than post-Archean lavas (Maier et al. 2009;Puchtel et al. 2020).
The Pd contents of Mg-basaltic to komatiitic Bushveld magmas (so-called Bushveld 1, or B1, magmas; Sharpe 1981) overlap with the Pd-rich samples of the global lavas, but the Bushveld magmas have markedly higher Pt and, somewhat less pronounced, Pt/Pd (Maier and Barnes 2004; Barnes et al. 2010Barnes et al. , 2015Maier et al. 2016) (Fig. 1). Other suites with elevated Pt and Pt/Pd include ~ 2.05 Ga Ti-rich komatiites of the Lapland Greenstone belt (Karasjok and Jeesiovaara, Fiorentini et al. 2011;Puchtel et al. 2020), 2.9 Ga komatiites of Western Australia (Fiorentini et al. 2011), the ~ 260 Ma Emeishan picrites of China (Li et al. 2014), the 120 Ma-recent Kerguelen basalts and picrites (Chazey and Neal 2005), and the ~ 1.27 Ga Coppermine Mg-basalts of northern Canada (Day et al. 2013). For most of these suites, there is no clear trend of PGE content or Pt/Pd against MgO, except for the Jeesiovaara suite analysed by Puchtel et al. (2020) which shows a good negative correlation between Pt and Pd with MgO, and the Bushveld magmas which show a subtle increase in Pt/Pd with falling MgO.

Pt/Pd (and Au/Pt) ratios in Bushveld cumulates
Bushveld cumulates show much more pronounced variation in Pt/Pd than the Bushveld magmas (von Gruenewaldt and Merkle 1995;Cawthorn et al. 2002;Naldrett 2004;Maier and Barnes 2004;Kinnaird 2005;Hutchinson and McDonald 2008;Kinnaird and McDonald 2018;Maier et al. 2021) pointing to the importance of magma chamber processes in controlling at least some of the observed variation. Variation in Pt/Pd occurs at several scales: 1. Regional-scale variation between the western (WBC), eastern (EBC) and northern Bushveld lobes (NBC): This is particularly evident in the well-characterised reef horizons (Naldrett et al. 2009;Fig. 1 in Grobler et al. 2019). For example, the Merensky Reef (MR) of the WBC tends to have Pt/Pd ratios between 2 and 2.5, the MR in the EBC has slightly lower ratios (~ 1.5-2), whereas most of its correlatives in the northern lobe (Platreef, Flatreef and Waterberg deposit) have Pt/Pd between unity and 1.6. 2. Regional variation within individual lobes: For example, in the contact-style mineralisation of the northern Bushveld lobe (Platreef and Flatreef), Pd/Pt is markedly higher in the north than in the south Kinnaird and McDonald 2018;McDonald et al. 2017).
To some degree, this may be due to the northern prospects representing stratigraphically higher intervals than the southern prospects, as will be discussed in a later section. 3. Large-scale stratigraphic variation: At the base of the WBC layered suite, Pt/Pd is typically relatively low (average of 0.35 in the lower 250 m of the WBC, Fig. 2, Maier et al. 2013). Across the remainder of the Lower Zone (LZ) and much of the Critical Zone (CZ), Pt/Pd shows a progressive increase with height, to peak in the UG1 footwall unit (Pt/Pd > 10) which is characterised by very low Pd values (< 5 ppb) at Pt contents of several tens of parts per billion (Maier et al. 2013 Fig. 4), the MG3 chromitite (Pt/Pd from < 1 to ~ 8, increasing upwards) and the MG4a chromitite of the EBC (Pt/Pd < 1 to 14) (Naldrett et al. 2012). The Middling unit at the top of the Flatreef shows a bell-shaped Pt/Pd trend, with an upward increase in Pt/Pd in the lower portion (from ~ 1 to ~ 5) and a decrease (from 5 to ~ 1) through the upper portion ( Fig. 5) ).
Analogous to Pt/Pd, ratios of Au/Pt show marked variation in Bushveld cumulates (Figs. 5,6a). In a broad sense,  Barnes et al. 2010;Maier et al. 2016;Fiorentini et al. 2011;Puchtel et al. 2020;Li et al. 2014;Chazey and Neal 2005;Day et al. 2013) Au/Pt increases with height: Many of the lowest Au/Pt ratios occur in the LZ and CZ, whereas the highest Au/Pt is found in the UZ. In addition, there is pronounced local-scale variation. A particularly marked increase in Au/Pt is seen across the reef interval of the upper CZ (Fig. 6b), and above individual PGE reefs (e.g. UG1 and UG1, Merensky and Bastard Reefs, Maier et al. 2021).
Smaller-scale variation can be studied, e.g., in the Flatreef of the NBC where continuous assays are available over the reef intervals. Au/Pt ratios show little systematic stratigraphic  (Maier et al. , 2022 except for a marked increase above the Merensky Reef, i.e. in the Middling and Bastard units (Fig. 5), correlating with an analogous increase in Pt/Pd, and a decrease in Pt + Pd and Pt/Cu. In the upper portion of the Middling unit, the trend of increasing Au/Pt and Pt/Pd, and decreasing Pt/Cu and Pt + Pd, is reversed, resulting in a bell-shaped pattern for the entire interval.
High Au/Pt ratios, at variable Pt/Pd, are also characteristic of many of the MZ-associated mineralisations in the northern lobe, e.g., in the Waterberg T zone (Kinnaird et al. 2017), the basal mineralisation at Nonnenwerth/Aurora (Manyeruke 2007;McDonald et al. 2017) and the Pyroxenite Marker mineralisation at Moorddrift Holwell et al. 2013).

Tungsten isotope analysis
Tungsten concentrations were measured by isotope dilution using the Element 2 single-collector ICP-MS at the University of Maryland, USA. Tungsten isotopic compositions were determined using a four-step ion exchange chromatography method described in Peters et al. (2019). Isotope compositions   Maier et al. 2021) were measured by thermal ionisation mass spectrometry in negative ionisation mode using a Thermo Fisher Triton at the University of Maryland following the method of Archer et al. (2017). Radiogenic W isotope compositions are presented using the µ-notation which defines the parts-per-million deviation of a sample's 182 W/ 184 W ratio from that of repeated measurements of a laboratory solution standard (Alfa Aesar) that is representative of bulk silicate Earth composition (µ 182 W = 0; e.g. Walker 2016).
Tungsten isotope compositions of sample 774.11, representing the komatiitic chilled margin of the LZ at Union section, exposed in drill core NG2 (Maier et al. 2016) are presented in Table 1 and show no deviations in µ 182 W normalised to 183 W/ 186 W (N6/3) and µ 183 W normalised to 184 W/ 186 W (N6/4), respectively, from the average of seven laboratory standard analyses (2SD = 4.5 and 8.0, respectively).

Melting of sub-continental lithospheric mantle (SCLM)
Based on the relatively high Pt/Pd in Bushveld magmas (Pt/ Pd average ~ 1.5) and Kaapvaal SCLM (average Pt/Pd ~ 2; Maier et al. 2012 and references therein) together with unusually high contents of many incompatible trace elements, pronounced negative Nb, Ta and Ti anomalies in multi-element variation diagrams, relatively high SiO 2 contents (up to 55 wt.% SiO 2 at MgO 12-14 wt.%), the predominance of orthopyroxene over clinopyroxene, and elevated Sr and low Nd isotope compositions, Maier and Barnes (2004) suggested that the Bushveld magmas represent partial melts of the SCLM. Richardson and Shirey (2008) and Mungall and Brenan (2014) concurred with this model, based on Os isotopes in diamond inclusions, experimental data and thermodynamic modelling. However, Maier et al. (2016) used MELTS modelling to show that partial melts of fertile Kaapvaal SCLM are Si depleted and K enriched relative to Bushveld komatiite as intersected in a chilled margin at the base of the Complex. The authors concluded that the Bushveld parent magmas likely represent asthenospheric partial melts modified by contamination with the continental crust (see discussion below). A small SCLM component remains a possibility, being that the Bushveld magmas likely ascended through the thick SCLM keel underlying the Kaapvaal craton. However, in view of the generally relatively low PGE contents of the Kaapvaal SCLM (average of ~ 4 ppb Pt, 2 ppb Pd, maximum of 18 ppb Pt and 10 ppb Pd; Maier et al. 2012), this would likely not have significantly influenced the PGE budget of the magmas, unless there are very PGE-rich segments in the SCLM that remain undiscovered or have been consumed by the melting.

Assimilation of Pt-rich crustal sediments and cumulates
Apart from SCLM derivation, the main alternative model to explain the enriched nature of the Bushveld Complex has been crustal contamination of asthenospheric magma (Kruger and Marsh 1982;McCandless and Ruiz 1991;Maier et al. 2000;Harris et al. 2005). However, the available data on Kaapvaal sedimentary rocks provide no strong support for the idea that crustal sedimentary contamination added significant Pt to the magmas. Stephenson (2018) found Pt/ Pd mostly above unity in the floor rocks to the northern lobe, but at mostly low PGE contents (< 10 ppb Pt and Pd), consistent with previous studies on Kaapvaal sediments (Siebert et al. 2005;Nwaila and Frimmel 2019;Wille et al. 2007).

Pd mobility through alteration
Based on empirical (Barnes and Liu 2012;Holwell et al. 2017) and experimental (Hanley 2005;Sullivan et al. 2021) evidence, it is normally argued that at typically magmatic conditions (sulfide present, ƒO 2 near QFM and near neutral pH), Pd is more mobile in fluids than Pt. One could thus argue that the Bushveld fine-grained floor sills were affected by alteration from percolating late-magmatic or metamorphic fluids resulting in loss of Pd and relatively high Pt/Pd. This model was proposed to explain the high Pt/Pd of the Kerguelen lavas ( Fig. 1) (Chazey and Neal 2005). However, in the case of the Bushveld magmas, the model is rejected as the fine-grained floor sills have mantle like S/Se and show no petrographic evidence of significant alteration.
The sills are not parental to the cumulates but instead represent residual liquids expelled from the cumulates Sharpe and Hulbert (1985) and Yao et al. (2021) argued that the medium-grained pyroxenitic and peridotitic floor sills (B1u) represent ejections of crystal mushes from the LZ, perhaps triggered by tectonic adjustment of the Bushveld magma chamber. This would imply that the sills are not representative of the parent magmas to the Bushveld LZ and LCZ, but instead represent residual melts from the cumulates. A similar model was proposed by Helz (1987) for the Stillwater floor dykes. Thus, liquid ejection from the magma chamber was clearly part of the thinking at the time. However, this model is not consistent with PGE systematics. Much of the LZ contains elevated PGE (Maier et al. 2013) suggesting either that the rocks crystallised from sulfidesaturated magma, that sulfide melt percolated downwards from above or that PGE were introduced by reactive flow of volatiles from the interior of the intrusion. If the sills were ejected residual liquids from the cumulates, they should mirror the elevated and heterogeneous PGE contents of the cumulates, yet the sills are undersaturated in sulfide melt and have fairly homogeneous PGE patterns . One could argue that the sills lost Pd (and S) during metamorphism, but the S/Se of the sills is mantle-like and not suggestive of significant element mobility, unless S and Se were removed together at high T (Wulf et al. 1995). Another argument against the Yao et al. model is that the Dullstrom lavas which cannot represent residual liquids expelled from the cumulates also have the characteristic high Pt/Pd of the B1-B3 Bushveld basalts (albeit at somewhat lower total PGE contents; unpublished data of Hannah Hughes).

Melting of PGE-rich mantle domains
The PGE contents of the Earth's modern mantle are largely the result of processes occurring during the Hadean and Archean eras: (1) The vast bulk of the Earth accreted ~ 4.5 Ga ago from impactors that remain incompletely known and appear not to be represented in our meteorite collections (Fischer Goedde et al. 2020).
(2) Next, the Earth's core formed via segregation of Fe-Ni alloy (and some sulfide) ~ 30 m.y. after Earth's initial accretion (Kleine et al. 2002). The segregation of the core resulted in efficient scavenging of the highly siderophile and chalcophile PGE and Au from the Earth's mantle (Stevenson 1981). (3) Puchtel et al. (2020) suggested that portions of the mantle remained poorly equilibrated with LV until ~ 2 Ga, based on W and Pt isotopes as well as PGE systematics in komatiites of the Lapland Greenstone belt. Heterogeneous inmixing of late veneer is consistent with models of modern mantle dynamics; for example, Barry et al. (2017) showed that whole-mantle convection may preserve long-term global patterns of upper-mantle geochemistry. Based on these data, one could hypothesise that the Pt-rich Bushveld magmas formed from a mantle that was relatively enriched in late veneer.
The idea is potentially testable using the short-lived radiogenic 182 Hf-182 W isotope system, where Hf is lithophile and W a moderately siderophile element. This system is only alive during the first ~ 60 Ma into Earth's history (Vockenhuber et al. 2004). Hence, variations in Earth's rock record must be attributed to Hadean processes, such as metallic core segregation and/or silicate differentiation. μ 182 W in terrestrial reservoirs shows considerable variation, with Earth's core being less radiogenic (μ 182 W = − 220 ppm) than bulk Earth (μ 182 W = − 200 ppm), bulk silicate Earth and modern upper mantle (μ 182 W = 0 ppm) (Fig. 7A) (Walker 2016 and references therein). The assumption is that during Hadean core formation, most of the Earth's Hf remained in the mantle due to the incompatibility of Hf with regard to Fe-Ni alloy. Thus, little 182 W could be produced via radiogenic decay of 182 Hf in the core, resulting in relatively low 182 W/ 184 W. In contrast, the Hadean mantle became enriched in 182 W, the daughter isotope of 182 Hf. The relative enrichment of the Hadean (and early Archean) mantle in 182 W is reflected in high μ 182 W of early-mid Archean lavas and cumulates (Tusch et al. 2021a, b). It is assumed that prior to late accretion, the early Earth's mantle was about 25-30 ppm more radiogenic in terms of μ 182 W than it is now. Late accretion of largely chondritic material (μ 182 W = ~ 180-200) shifted the tungsten isotopic composition of the mantle to the present value of μ 182 W = 0. The near-zero μ 182 W of post-Archean rocks is thus normally interpreted to result from partial melting of mantle refertilised with LV, whereas the higher μ 182 W of the older rocks is interpreted to reflect melting of mantle that had failed to equilibrate with LV. One could thus suggest that portions of the mantle containing a relatively high component of LV have lower μ 182 W than bulk silicate Earth (μ 182 W = 0).
To test whether this model could apply to Bushveld, we analysed the komatiitic chilled margin of the Complex (sample 774.11 of the NG2 drill core at Union section; Maier et al. 2016). This yielded μ 182 W + 2.1 ± 3.6 (Table 1), suggesting that the sample comes from normal mantle without LV-enriched or -depleted domains. However, field relationships (e.g. the presence of several quartzite xenoliths in the basal interval) and trace element systematics suggest that the sample has undergone some pre-emplacement and syn-emplacement contamination (Maier et al. 2016). The relatively high W/Th ratio of ~1 -2.2 suggests that the rock could have been affected by percolation of a W-rich volatile phase. The contaminants could have had "normal" μ 182 W, thereby potentially altering an originally negative μ 182 W signature. Another possibility is that the Bushveld source could have contained a component of unequilibrated iron meteorite. Iron meteorites have even lower µ 182 W and higher W concentrations than chondritic material. Hence, if the source was affected by an iron meteorite, then the 'original,' as in precontaminated, W signature could have been even more negative.
Another tool that could be used to evaluate the presence of mantle domains in the Bushveld source is nucleosynthetic Ru isotopes (Fig. 7B). Fischer Goedde et al. (2020) have established that the early (pre-LV) Earth had positive ε 100 Ru. The authors applied mass balance to argue that the late veneer must have had negative ε 100 Ru and that efficient mixing of this LV with the Hadean mantle resulted in ε 100 Ru = 0 (i.e. the value of bulk silicate Earth). Magmas derived from hypothetical LV-enriched mantle domains should thus have negative ε 100 Ru. The first Ru isotope data on Bushveld (on UG2 and LG6 chromitite; Fischer-Goedde et al. 2020) overlap with the modern mantle estimate, similar to our W isotope data.
We conclude that the currently available data provide no support for the idea that the Bushveld was sourced from LV-enriched mantle. However, the amount of data generated so far is relatively small and more work is clearly needed.

Contamination of mantle magma by a crustal meteoritic component (e.g. impact structures)
Based on presumed coevality between the Bushveld Complex and the Vredefort impact structure, Dietz (1963), Hamilton (1970) and Rhodes (1975) suggested that the Bushveld formed via a meteorite impact. It is now known that Vredefort is ~ 33 Ma younger than the Bushveld (Kamo et al. 1996). No evidence of shatter cones or high-P minerals has been found in the floor rocks of the Bushveld Complex (French and Hargraves 1971). However, it is conceivable that the Bushveld intrusion could have cannibalised much of the structural and mineralogical evidence of a major impact (Fig. 8). As demonstrated above, W and Ru isotope data as well as PGE contents of Bushveld ultramafic cumulates and magmas provide no clear evidence of a meteoritic component, possibly because the composition of ultramafic rocks shows some overlap with mantle rocks. A possible additional test could be to look for the compositional signature of an asteroid in the felsic lavas and granites which occur on top of the Rustenburg Layered Suite (RLS) and which must have ascended through any putative impact melt sheet at depth. As yet, no PGE data are available for the granites or rhyolites, but we note that the Dullstroom andesites have elevated Pt/Pd analogous to the RLS (unpublished data of Hughes). We conclude that there is currently no evidence for interaction of the lavas with a crustal meteoritic component, but due to paucity of data, this cannot be ruled out at present.

Origin of Pt/Pd variation in Bushveld cumulates
Bushveld cumulates show enormous variation in Pt/Pd (and Au/Pt), in stark contrast to the fine-grained sills with their relatively homogeneous PGE contents and Pt/Pd (and Au/Pt). This suggests that, compared to the sills, in the main Bushveld magma chamber, several additional processes operated that fractionated the PGE and Au from each other. Four main models are considered in the following. All likely applied at some stage during the solidification of the complex.

Primary magmatic fractionation of Pt, Pd and Au due to variable compatibility into crystallising sulfide melt and oxide crystals
The variation in Pt/Pd in the Bushveld cumulates could be the result of differences in D sulfide melt/silicate melt for Pt and Pd during crystallisation. As noted previously, some of the available experimental data show a somewhat higher compatibility of Pd than Pt with regard to sulfide melt (see summary in Barnes and Ripley 2016). One could propose that this is responsible for the relatively low Pt/Pd at the base of many layered intrusions, including Duluth (Barnes et al. 1997), Muskox (Barnes and Francis 1995) and intrusions in the Cape Smith belt (Barnes et al. 1992). For example, in the Cape Smith intrusions, sulfides have Pd/Pt ratios from 1 to 4, whereas the basaltic komatiites interpreted to represent the parent magmas have Pt/Pd at approximately unity. In the Bushveld magmatic province, one of the most notable examples of relatively low Pt/Pd at the base of an intrusion is the Uitkomst satellite intrusion which contains economically important sulfide mineralisation in its basal portion having Pt/Pd around 0.5, whereas (uneconomic) reef-style sulfides in the upper portion of the intrusion have Pt/Pd ~ 1.2 (Maier et al. 2022).
Within the Bushveld Complex itself, Maier et al. (2022) have proposed that differences in D between Pt and Pd can explain intra-layer variation in Pt/Pd occurring on a scale of meters to tens of meters (e.g. Figures 2-5). In the uppermost portion of the Flatreef (MU and BU), trends in metal contents and metal ratios resemble those of classical offset patterns described from, e.g., Munni Munni (Barnes 1993) or Great Dyke (Wilson 2001). Thus, Rh tenors in the upper portion of the main mineralised interval (M2 and BAR) are 67-76% of those in the lower portion (M1), Pd tenors in the upper portion are 79-87% of M1, Cu tenors are 90-92% of M1, Pt tenors are 107-120% of M1 and Au tenors are 149-202% of M1. These data could suggest that the offset patterns result at least partly from sulfide melt fractionation, with D Pd > D Pt > D Au , consistent with some of the published data (Barnes and Ripley 2016). Also, Au/Pt shows a marked increase above the UG1, UG2, Merensky and Bastard Reefs (Fig. 6) (Barnes and Maier 2002; and Au/Pt is markedly elevated in many of the reefs in the Northern Lobe which occur at stratigraphically elevated levels, e.g., the Waterberg T zone (Kinnaird et al. 2017), the Platreef at Nonnenwerth/Aurora (Manyeruke 2007;McDonald et al. 2017) and the Pyroxenite Marker mineralisation at Moordrift .
However, modelling suggests that variation in D sulfide melt-silicate melt is unlikely to be the only reason for the observed trends. At the relatively low R factors (mass ratio of silicate to sulfide melt; Campbell and Naldrett 1979) applicable to the basal mineralisation in the Flatreef (~ 1000; Maier et al. 2022), and assuming that B1 magma has Pt = 19 ppb and Pd = 14 ppb , and D sulfide melt/silicate melt is 10,000-1,000,000 (Barnes and Ripley 2016), modelled Pt tenors of sulfide are higher than Pd tenors because the tenors are more sensitive to the difference in PGE content (i.e. the higher Pt than Pd content) of the magmas than to potential differences in D. For example, to explain the observed, relatively high Pd/Pt of up to 2 in the basal Flatreef mineralisation, a relatively low D of around 2000 for Pt is required, but none of the experiments give such low values. Similarly, we cannot model the high Pt/Pd (~ 5) in the Middling unit (Fig. 5) by applying realistic D values, nor can we explain the extremely high Pt/Pd in some chromitites (e.g. at the base of the Merensky Reef; Barnes and Maier 2002).
With regard to PGE fractionation triggered by oxide crystallisation, Finnigan et al. (2008) proposed that the commonly observed spatial association of laurite with chromite may be due to localised reduction of the magma within micrometric melt domains surrounding the growing chromite crystals in response to scavenging of Fe 3+ by the crystallising chromite. Chromitites tend to be highly enriched in Pt and have high Pt/Pd (Barnes and Maier 2002, Fig. 4), suggesting that local-scale magma reduction during chromite crystallisation could also trigger formation of Pt alloys.

Pt and Pd fractionation caused by percolation of sulfide melt through cumulate mushes
Magmatic sulfide melt is relatively dense (4.0 and 4.5 g cm 3 ; Mungall and Su 2005) and solidifies at relatively low T (~ 700 °C; Helmy et al. 2021). Thus, it could potentially percolate through partially solidified mafic-ultramafic cumulates as long as the silicate grain size is relatively large and sulfide droplets are either smaller than the nodes, or large enough to have sufficient gravimetric power (Chung and Mungall 2009). Percolation of sulfide melt through Bushveld cumulates has been proposed previously for the Merensky Reef (Leeb-du Toit 1986;Godel et al. 2006;Smith et al. 2020) and in the Flatreef (Maier et al. 2022) where downward decreasing PGE and sulfide contents are accompanied by subdued PGE fractionation. In addition, the peculiar bell-shaped PGE and Pt/Pd distribution in the barren Middling unit, interleaved between BAR and M2 (Fig. 5), was interpreted to result from a combination of PGE fractionation (from relatively low Pt/Pd in the basal M2 reef to high Pt/Pd in the PGE depleted rocks above M2) and downward percolation of BAR sulfides into the upper portion of the Middling.
As has been explained above, fractionation of primary magmatic sulfide melt is unlikely to cause Pt-Pd fractionation on the scale observed. However, if solid sulfides undergo partial melting, e.g., in response to volatile fluxing, Pt and Pd have been shown to fractionate (Peregoedova et al. 2004), by generating a Pd-Cu-Ni-rich sulfide melt and refractory Pt-rich PGM. In the context of the Flatreef, we propose that volatiles ascending from the metamorphosed floor and/or the cooling cumulates pile caused desulfidation and melting of sulfides in the upper Flatreef (M1, M2 and BAR) resulting in downward percolation of a Pd-Cu-Ni sulfide melt contributing to progressively lower Pt/Pd of the Flatreef with depth (Fig. 3), at broadly constant Cu/Ni. The model could also explain the presence of Pd-rich sulfides in calcsilicate interlayers within the lower Flatreef and the sedimentary floor rocks (unpublished data of authors) and, via restitic Pt-rich sulfides and PGM, the high Pt/Pd (up to 5) in the Middling unit (Fig. 5).

Mobilisation of PGE (particularly Pd and potentially Au) in a volatile phase
Experimental data indicate that under certain conditions (acidic pH, high salinity, no sulfide in rock), Pt and Pd can be mobile in magmatic and hydrothermal volatiles, and Pd tends to be more mobile than Pt (Peregoedova et al. 2006;Hanley et al. 2005;Boudreau and McCallum 1992;Boudreau 2019;Sullivan et al. 2021). A volatile phase could be generated during advanced crystallisation of the cumulates, or via metamorphism of the floor rocks. In the Bushveld, there is empirical evidence for both Pt and Pd mobility, e.g., in the form of Pt-enriched hydrothermal veins at Waterberg (McDonald et al. 1999) and extremely Pd-depleted UG1 chromitite intersections . Volatiles could also have played a role in causing the unusually high Pt/Pd of the Merensky Reef Cr stringers; Naldrett and Lehmann (1988) and Li et al. (2005) have suggested that this is due to reaction of sulfide with chromite leading to desulfidation and associated mobility of Pd whereas Pt alloys behaved in a refractory manner. One could go further to suggest that the relatively high Pt/Pd in much of the Bushveld Complex is due to mobilisation of Pd from the centre to the margins of the complex, as has been suggested by Boudreau (2019) for Stillwater. In order to test this model, detailed traverses have to be sampled and analysed across all lobes of the RLS.
Another possibility is that Pd could be derived from the sedimentary floor rocks from where it was transported by metamorphic fluids into the basal portion of intrusions. Stephenson (2018) has shown that the sedimentary floor rocks of the Bushveld northern lobe tend to be relatively enriched in Pt relative to Pd, possibly reflecting selective Pd removal. As the PGE content of sediments is mostly quite low (typically < 10 ppb, Wille et al. 2007;Siebert et al. 2005), large volumes of the floor would need to be processed to affect the Pd budget of the cumulates significantly. For example, assuming that a 100-m-thick cumulate package at the base of the northern limb contains 100 ppb extra Pd over the normal Pd budget, mobilisation of 10 ppb Pd from 1000 m of sedimentary rock would be required. In conclusion, local hydrothermal Pd mobility in the Bushveld Complex is supported by empirical evidence and experimental data, but it remains unclear whether this is important on a large scale.
Crystallisation of the cumulates from parent magmas having variable Pt, Pd and Au contents, resulting from pre-emplacement magma contamination or fractionation Platinum solubility in basalt decreases with falling ƒO 2 (and T) (Borisov and Palme 1997;Mungall and Brenan 2014). In the context of the Bushveld Complex, one could thus suggest that the relatively low Pt/Pd at the base of the intrusion reflects pre-emplacement contamination of the initial magma with reducing (organic) material, e.g., graphitic shale that is locally abundant in the Duitschland Formation of the Transvaal Supergroup (Yudovskaya et al. 2021 and present authors' observations on Flatreef drill cores), leading to precipitation of Pt alloys prior to, or during, final emplacement. Later magma pulses forming the overlying cumulates (including the Merensky Reef) could have undergone less contamination with the floor rocks, perhaps due to armouring of the conduits by the precipitates of the early pulses. The late pulses were thus less Pt depleted. Instead, they could have assimilated some of the Pt-rich cumulates in the staging chambers and feeder conduits. However, there is currently no direct evidence for the model, e.g., in the form of Pt-depleted fine-grained sills or dykes, or floor cumulates with high Pt/Pd.

Summary and conclusions
In the present paper, we compiled Pt, Pd and Au data on Bushveld magmas as represented by fine-grained sills in the floor of the complex and Bushveld cumulates in the western and northern limbs.
Compared to global mantle-derived magmas, the Bushveld parent magmas have unusually high Pt contents and, as a result, high Pt/Pd. After examining a number of models, we conclude that the Bushveld magmas are likely of asthenospheric derivation, with a significant SCLM component being unlikely. Also, there is no clear evidence for an unusually PGE-rich mantle, e.g., containing undissolved remnants of the late veneer. The Bushveld magmas were strongly contaminated during ascent through the continental crust, facilitated by the abnormal size and heat flux of the Bushveld event. This could have led to assimilation of Pt from Pt-rich crustal rocks, potentially containing a meteoritic component. However, no Pt-rich crustal rocks are currently known to occur in the Bushveld floor.
Bushveld cumulates show more significant variation in Pt/Pd and Au/Pt than the parent magmas, suggesting that much of the observed variation in PGE ratios occurred within the Bushveld magma chamber. Likely processes include (1) primary magmatic fractionation of sulfide melts, silicate melts and PGE alloys caused by varying partitioning of PGE, potentially locally enhanced by variation in oxidation state of magma; (2) sulfide melt percolation through semi-consolidated cumulates and associated PGE fractionation, facilitated potentially by partial melting of cumulates in response to flux of volatiles; and (3) mobilisation of some PGE (particularly Pd and potentially Au) by volatiles derived from the sedimentary floor rocks or the basal cumulates.

Conflict of interest The authors declare no competing interests.
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/.