- © 2016 Mineralogical Society of America
Cratonic lithospheric mantle is composed of predominantly refractory materials that formed at higher mantle potential temperatures (TP) than recorded in non-cratonic peridotites. It also shows stronger depletion and fractionation of Pd and Pt from Ru, Os and Ir than oceanic, supra-subduction zone or off-cratonic lithospheric mantle, as well as some of the lowest Se and Te contents. The varied response of the highly siderophile elements (HSE: Os, Ir, Ru, Rh, Pt, Pd, Re, Au), and their embedded radioactive decay systems, to changes in oxygen fugacity (fO2), sulfur fugacity (fS2) and pressure (P)—in particular through the impact of these parameters on the stability of the main HSE-bearing sulfide and alloy phases makes them potentially powerful tracers of their melting environment. Therefore, investigation of the HSE systematics of cratonic mantle peridotites, in combination with information from Re–Os isotopes on time-integrated enrichment or depletion, can help us to understand processes leading to mantle differentiation and continental lithosphere formation in the Archean, which are controversial subjects despite decades of research.
The longevity of the cratonic lithosphere implies that there was ample opportunity for secondary overprint, obscuring our view of earlier processes. For example, destabilization of platinum-group element (PGE: Os, Ir, Ru, Rh, Pt, Pd) alloy leading to depletions in the compatible PGE, and perhaps Pt, in some cratonic mantle samples may occur in an oxidizing mantle wedge or through interaction with oxidizing small-volume, volatile-rich melts that typically invade cratonic roots. Such melts may eventually deposit S, Pd, Pt and Re and also capture remaining PGE alloys, consistent with the anomalous S-rich character of many kimberlite-borne xenoliths. Their basalt-borne counterparts show additional late effects of subaerial degassing that can deplete volatile elements (S, Re, Os). Basaltic melts can also scavenge PGE alloys at depth, while still sulfide-undersaturated. Such melts, may, on ascent, add sulfides when they become sulfur-saturated and, during the process, refertilize the mantle and modify major-element and modal compositions. The investigation of minor lithologies in the cratonic lithosphere, such as eclogites and pyroxenites, which are expressions of tectonothermal events ranging from subduction to melt infiltration, can enhance our understanding of the effects of these processes on HSE redistribution.
Thus, three major topics will be discussed, using HSE systematics in cratonic mantle samples: (1) How did the HSE behave during the (in part) extreme degrees of partial melt extraction experienced by cratonic lithospheric mantle; (2) What were the effects of the secular metasomatic overprint of the cratonic mantle; (3) What was the composition of the Archean convecting mantle, for which cratonic mantle samples may afford better insight than modern samples, provided, of course, that we have an accurate grasp of how HSE are redistributed during partial melting and metasomatism. Models based on experiments done under controlled pressure (P), temperature (T), fO2 and fS2 conditions can help place the data in context and to distinguish between melt- and metasomatism-related processes. Disentangling the various primary and secondary effects is only possible when HSE are studied in combination with lithophile elements, with due attention to petrography and mineralogy. This adds many layers of complexity, but ultimately allows a more complete understanding of the variegated processes that have shaped the cratonic lithosphere through time.
In this review, we commence by discussing the peculiarities and complexities of continental lithospheric mantle origin, evolution and current state. We then introduce the database used in this contribution, followed by a brief review of the mineral hosts of HSE in peridotite and of the diverse approaches to isolate the HSE for measurement. We examine the behavior of the HSE during the formation of cratonic lithospheric mantle under non-uniformitarian conditions, where the application of the Re–Os isotope system has afforded particularly useful information on the timing of initial melt depletion and the stabilization of cratonic roots. We then turn to the effects of mantle metasomatism, both during intra-plate and craton-margin processes (see also Gannoun et al. 2016, this volume), on HSE systematics in cratonic mantle. We also discuss the data in the context of melt extraction modelling that shed light on the primary versus secondary HSE signatures in cratonic mantle rocks. Finally, we evaluate the possibility that the HSE in cratonic mantle retain a memory of core formation and subsequent accretionary processes.
THE CRATONIC MANTLE SAMPLE: PECULIARITIES, OPPORTUNITIES AND PITFALLS
Cratons are usually taken to represent the ancient (≥ 2.5 Ga) cores of stable continental regions that are commonly, but not always, underlain by broadly coeval and possibly cogenetic lithospheric roots, and where voluminous tectonomagmatic activity is confined to the margin (Bleeker 2003). Their areal extent has been mostly defined based on outcropping Archean basement (Fig. 1). Some cratonic entities were themselves formed from continental nuclei that were amalgamated and “stabilized” predominantly during the NeoArchean, and welded together to form larger landmasses (composite cratons) during Proterozoic collisions (Bleeker 2003). The resultant mobile belts provided protection against later tectonothermal overprint and helped ensure craton longevity (Lenardic et al. 2000).
Cratonic lithosphere generally extends to a maximum of 200–250 km depth (Artemieva and Mooney 2001; Priestley and McKenzie 2006). The mechanism(s) by which this depth was attained—in collisional settings or during plume tectonics or both—remains a controversial issue (Arndt et al. 2002, 2009; Griffin et al. 2003a, 2009; Lee 2006; Aulbach et al. 2007; Pearson and Wittig 2008, 2014; Lee et al. 2011; Aulbach 2012; Herzberg and Rudnick 2012). Peculiar properties of cratonic lithosphere include not only its greater thickness compared to average younger lithosphere, but also its higher average degree of melt depletion, viscosity and seismic velocity and cooler geothermal gradients (e.g., O’Reilly et al. 2001) that can be linked to the secular evolution of TP. While the details of this evolution are debated, higher TP in the Archean is generally advocated (e.g., Korenaga 2008; Davies 2009). The exceptional depletion of incompatible FeO and H2O that were possible at higher Archean TP and the resultant increased buoyancy and viscosity are thought to be pivotal in the long-term stability of cratonic lithosphere. The relative importance of these two physical parameters for craton stability has been modeled by various authors and remains debated (Jordan 1988; Abbott et al. 1997; Lenardic et al. 2003; Afonso and Schutt 2012; Wang et al. 2014), but it is clear that metasomatic modification appears to be a pre-requisite for weakening and generating negative buoyancy in the craton root (Wang et al. 2015) that can lead to its destruction in some instances.
The thick and cool cratonic lithospheric lid exceeds the depth where dry fertile convecting mantle crosses its solidus and reaches not much shallower depths than that of mantle with moderate volatile contents (e.g., Dasgupta et al. 2013), preventing large degrees of partial melting in upwelling mantle at current ambient TP (1310–1460 °C; e.g., Dalton et al. 2014). Consequently, intact cratonic mantle is typically sampled by deep-seated (> 250 km) low-volume carbonate-rich melts, such as kimberlites (e.g., Nixon 1987; Tappe et al. 2006). These melts entrain fragments (xenoliths) or their disaggregation products (xenocrysts including diamond and its inclusions) from the conduit wall consisting predominantly of variably depleted and metasomatized peridotite, the petrology of which reflects the sum of processes that occurred during cratonic lithosphere formation and evolution. Cratonic mantle that has lost a portion of its deep root due to erosion or delamination is occasionally sampled by higher-volume basaltic melts, as is the case for the North China craton (Gao et al. 2002) and the Tanzanian craton (Burton et al. 2000).
Besides peridotites, pyroxenites and eclogites form a generally subordinate, though in some cases locally dominant constituent of mantle xenolith and diamond inclusion suites (Schulze 1989). In particular cratonic mantle eclogites having elemental and isotopic compositions consistent with a crustal origin document the emplacement of crustal components into the cratonic lithosphere (Macgregor and Manton 1986; Jacob 2004). The presence of such lithologies has been argued to attest to the relevance of accretionary processes in the construction or evolution of cratons (e.g., Helmstaedt and Schulze 1989; Pearson et al. 1995a, 1998; Shirey et al. 2001; Shirey and Richardson 2011). Regardless of interpretation, it is important to bear in mind that magma-entrained material depicts the state of the cratonic lithosphere not at the present day, but at the time of magma emplacement, which encompasses Cenozoic to rare Proterozoic and Neo-Archean kimberlites, lamproites and ultramafic lamprophyres (summarized in Gurney et al. 2010).
Several potential pitfalls are associated with the accidental mode of lithospheric mantle sampling by xenoliths, which have to be taken into account in an evaluation of constraints that can be obtained from these samples: (1) It is likely that the mantle material entrained in kimberlites is not representative of cratonic lithospheric mantle as a whole because kimberlites are thought to erupt through reactivated lithospheric weak zones that were altered by the passage of fluids and melts (e.g., Poudjom-Djomani et al. 2005; Afonso et al. 2008) and because the entire lithosphere column is rarely equally sampled by the rising kimberlite magma (e.g., Finnerty and Boyd 1987; Lee et al. 2011). (2) Kimberlite infiltration into entrained xenoliths is ubiquitous (e.g., Schmidberger and Francis 2001; Jacob 2004) and its effects range from the deposition of host magma-derived material at grain boundaries and in grain fractures, to the formation of melt pools (e.g., Barth et al. 2001; Dawson 2002). Chemical data from the measurement of whole rocks are therefore mixtures of the pre-entrainment compositions and fractionated kimberlite magma (e.g., Pearson and Wittig 2014). (3) Although occasionally very large xenoliths are entrained (> 25 cm; e.g., Bell et al. 2005), the sample size is limited by the loading capacity of the host magma and dike width (e.g., Sparks et al. 2006), and by access to often small specimens (≤ 3 cm) that have passed the processing stage from modern-day diamond exploration and mining companies. This makes accurate estimates of representative bulk composition and relationships between different lithologies difficult, and the definition of structural relationships impossible. On the other hand, xenoliths are brought to the surface quickly, in a matter of hours or days (e.g., Canil and Fedortchouk 1999; Sparks et al. 2006) and, as such, there are no complications from slow exhumation, retrogression and re-equilibration that plague slowly exhumed orogenic mantle slices (e.g., Ernst and Liou 2008).
The above illustrates that the longevity of cratons simultaneously offers great benefits and drawbacks. Cratons provide the only rock archive of the Earth’s earliest lithosphere and its interaction with the emerging hydrosphere, atmosphere and biosphere, and of mantle dynamics that eventually resulted in the face of the planet we inhabit today (e.g., Santosh 2013). This archive may hold clues to the earliest planetary differentiation processes, such as core formation, mixing of late accreted material into the mantle and magmatic differentiation, which are particularly sensitively monitored by HSE (e.g., Shirey and Walker 1998; Lorand et al. 2013; Day et al. 2016, this volume). Furthermore, cratons are unique hosts to specific deposits, such as world-class magmatic PGE–Ni ores and gem-quality diamonds (de Wit and Thiart 2005; Gurney et al. 2010). Because cratonic mantle is often sampled during multiple magma emplacement episodes at different times, as is the case for the Kaapvaal, North China, Siberia, Wyoming and Slave cratons (Smith 1983; Menzies et al. 1993; Mitchell 1995; Heaman et al. 2004; Pernet-Fisher et al. 2015; Sarkar et al. 2015), several “snapshots” of the state of the mantle lithosphere at different times are sometimes available. This temporally punctuated view can be augmented by information from inclusions in diamond (typically 10–300 μm), which are shielded from post-entrapment chemical modification by their inert and often ancient hosts (reviewed in Gurney et al. 2010). On the other hand, this antiquity ensures that over the course of eons, cratonic roots were multiply affected by passing melts and fluids (Fig. 2), and hence unmodified depleted material may be all but non-existent (e.g., Pearson and Wittig 2014). Cooling and subsolidus re-equilibration subsequent to craton formation leads to element redistribution and modal changes, further adding to the complexity of present-day compositions (Cox et al. 1987; Canil 1991).
Despite the caveats outlined above, cratonic mantle xenolith and xenocryst samples are an invaluable resource with which we can study the formation and evolution of cratonic mantle roots, and interrogate the early state of the convecting mantle. A significant contribution to our understanding of early mantle geodynamics, the nascence of continents and the subsequent evolution of the lithospheric mantle has come from the investigation of the HSE systematics of cratonic mantle-derived rocks (Walker et al. 1989; Reisberg et al. 1991; Carlson and Irving 1994; Pearson et al. 1995b,c, 2004; Pearson 1999; Carlson et al. 2008; Shirey and Richardson 2011; Maier et al. 2012). In particular, the Re–Os isotope system has become the tool of choice to date lithosphere stabilization, to identify the presence of cratonic mantle and to investigate the relationship between cratonic mantle and overlying crust (e.g., Pearson et al. 2007; Griffin et al. 2014; Pearson and Wittig 2014), using peridotite and eclogite whole rocks, their mineral separates, single sub-mm-sized sulfide (e.g., Griffin et al. 2002, 2004; Aulbach et al. 2004; Luguet et al. 2008), and even sulfide inclusions in diamond (e.g., Pearson et al. 1998, 1999; Richardson et al. 2001, 2004; Aulbach et al. 2009a, 2011).
This contribution builds on published data for mantle xenoliths from ten cratons (i.e., ≥ 2.5 Ga: Kalahari comprising Kaapvaal and Zimbabwe, Tanzania, Karelia, Siberia, Slave, Rae, North American with Greeland, Wyoming, and North China; Fig. 1), mostly entrained in kimberlites and related rocks of various ages. The data cover not only the “classic” HSE composed of PGE plus Re and Au, but also the chalcogens, S, Se, and Te, which partition into sulfides and tellurides (Hattori et al. 2002; Lorand and Luguet 2016, this volume). Selenium and Te display highly siderophile element character at conditions relevant to core formation (Rose-Weston et al. 2009) and the database on cratonic peridotites is expanding through high-precision analyses obtained by isotope dilution techniques (reviewed in Lorand and Luguet 2016, this volume). Lithologically, the featured samples encompass peridotites ranging from chemically depleted to enriched varieties, and minor lithologies where only limited data are available. In addition, there is a sizable database for Re and Os concentrations in sulfides, where grains occurring in peridotite and eclogite xenoliths are distinguished from those included in diamonds. With regard to sample digestion and analytical strategies, the data include whole rocks and mineral separates, as well as single grains of amenable minerals, as further reviewed below.
Within the cratonic database, we distinguish a sub-group of “disturbed” cratonic root samples, where the xenoliths are not hosted in kimberlites, but are erupted by less deeply derived magmas such as lamprophyres and alkali basalts. Such samples are typified by certain localities in the Wyoming, Tanzanian and North China cratons where the xenoliths have been erupted following variably severe thermo-tectonic processes that have led to thinning and removal of a portion of lithospheric mantle roots. We further consider regions where cratonic mantle was identified beneath Proterozoic crust (e.g., Western Australia: Luguet et al. 2009; Western Greenland: Wittig et al. 2010a) and circum-cratonic Paleoproterozoic areas, some of which are underlain by mantle showing geochemical, microstructural, thermal or geophysical affinities with their Archean neigbors (Table 1), such as the Rehoboth terrane in Namibia (west of the Kaapvaal craton; Luchs et al. 2013, and references therein). To illustrate the secular evolution of the lithospheric mantle and the effects of various processes on PGE systematics, we compare data from cratonic peridotites to those from younger circum- and off-cratonic areas, typically sampled by basalts, to orogenic peridotites, which have been affected by melt percolation and strong refertilization processes, and to Phanerozoic supra-subduction zone (SSZ) peridotites and ophiolites (see also Luguet and Reisberg 2016, this volume and Becker and Dale 2016, this volume), which have interacted with slab-related melts and fluids, the latter also affected by PGE mineralization, for example associated with chromitites.
The minimum, maximum, mean and median concentrations and ratios of some PGE in the various peridotite groups and in eclogites are given in Table 1, along with their standard deviation. In general, median concentrations will be presented as a more robust measure of central tendency, due to the often non-normal nature of the data and due to their robustness to the biasing affects of outliers, of which there are many (Fig. 3). Compared to orogenic peridotites (median Ir = 3.5 ng g−1, Os = 3.8 ppb, Pt = 6.9 ppb, and Pd = 5.7 ppb), which represent the modern convecting upper mantle, kimberlite-borne xenoliths have low median concentrations of all PGE (Ir = 3.1 ppb, Os = 3.1 ppb, Pt = 2.8 ppb, and Pd = 1.2 ppb, respectively; Fig. 3 and Table 1). This is true also for circum-cratonic peridotites (Ir = 2.9 ppb, Os = 3.3 ppb, Pt = 4.6 ppb, Pd = 1.2 ppb). Basalt-borne cratonic xenoliths show even lower concentrations of the Ir-group PGE (Ru, Os, Ir: IPGE) (Medians: Ir = 2.4 ppb, Os = 2.2 ppb, Pt = 3.8 ppb, Pd = 1.5 ppb), similar to younger off-cratonic xenoliths, which are also basalt-borne (2.5 ppb, 1.9 ppb, 4.2 ppb, and 2.2 ppb). By comparison, supra-subduction zone peridotites have strikingly higher Pt abundances (Ir = 3.4 ppb, Os = 1.8 ppb, Pt = 9.1 ppb, Pd = 3.0 ppb). Ophiolites, the majority of which are subduction-related (Dilek and Furnes 2011), have concentrations similar to orogenic peridotites (Ir = 3.5 ppb, Os = 3.9 ppb, Pt = 6.9 ppb, Pd = 4.6 ppb). The processes leading to these observed differences in HSE systematics of peridotites of different ages and from different tectonic settings will be explored below.
As mentioned above, with the exception of Re and Os, few data are reported for minor lithologies occurring in cratonic lithosphere and for sulfides. As meta-basaltic or -gabbroic rocks, cratonic eclogites are characterized by a significantly lower median Os concentration (0.30 ppb, n = 73) compared to peridotite which is, however, significantly higher than that observed for orogenic eclogites (0.011 ppb, n = 35). Median S, Se, and Te contents in 27 eclogite xenoliths from Roberts Victor (Kaapvaal) are 260 ppm, 73 ppb, and 9 ppb, respectively (Gréau et al. 2013), compared with 360 ppm S, 115 ppb Se, and 0.8 ppb Te in cratonic peridotite xenoliths. Exotic lithologies, such as Mica–Amphibole–Rutile–Ilmenite–Diopside rocks (MARID) and Phlogopite–Ilmenite–Clinopyroxene (PIC) occur mainly in the Kaapvaal craton where they are associated with different types of kimberlite magmatism, as is true for the megacryst suite (Grégoire et al. 2002). The PGE systematics of six MARID samples were investigated by Maier et al. (2012), in addition to a few pyroxenites, wehrlites and megacrysts of metasomatic origin. MARID xenoliths have low HSE concentrations (average Ir = 0.3 ppb, Ru = 0.92 ppb, Pt = 2.0 ppb, Pd = 2.1 ppb, Os not available), which is true also for the megacrysts (average Os = 0.5 ppb, Ir = 0.1 ppb, Ru = 1.1 ppb, Pt = 1.0 ppb, Pd not available). The reason for the paucity of data for these minor metasomatic lithologies may be that most efforts of using Re–Os isotope systematics in combination with HSE abundances are directed towards obtaining lithosphere stabilization ages, for which the least visibly metasomatized samples are targeted. Further analyses of both bulk rock and in-situ mineral HSE systematics in these rocks would be a welcome addition to furthering our understanding of the metasomatic evolution of the cratonic mantle.
Sulfides, which concentrate the HSE, have higher concentrations than their corresponding whole rocks (Table 1), which is true for sulfides in xenoliths (median peridotitic sulfide Os = 17000 ppb, n = 280; eclogitic sulfide Os = 40 ppb, n = 62), and for corresponding sulfide populations occurring as inclusions in diamond (median peridotitic sulfide Os = 42000 ppb, n = 54; eclogitic sulfide Os = 34 ppb, n = 82). Few additional data are available for these minerals, which have been targeted mostly for single-mineral Re–Os dating (e.g., Pearson et al. 1998, 1999; Richardson et al. 2001, 2004). Nickel-rich peridotitic monosulfides tend to exsolve to a low-temperature assemblage upon cooling, accompanied by fractionation of HSE between the low-temperature minerals (pyrrhotite, pendlandite, chalcopyrite), which complicates acquisition of representative trace-element concentrations (Richardson et al. 2001; Dale et al. 2009; van Acken et al. 2010). Nickel-poor sulfides in eclogitic rocks are little affected by exsolutions; sulfide inclusions in eclogitic diamonds (n = 41) have median concentrations of 41 ppb Ir, 140 ppb Ru, 315 ppb Pt and 340 ppb Pd (Aulbach et al. 2012).
MINERALOGY AND HSE HOSTS IN CRATONIC MANTLE PERIDOTITES
Depending on degree of depletion and metasomatic re-enrichment, mantle peridotites are composed of olivine, orthopyroxene, clinopyroxene, an Al-phase (plagioclase, spinel or garnet, with increasing pressure) plus accessory sulfides (base metal sulfides, BMS: Fe–Ni–Cu sulfides) or trace platinum-group minerals (PGM) (Lorand and Luguet, 2016 this volume). Here we define PGM as any phase in which PGE are major or stoichiometric components; these can be sulfides (e.g., the erlichmanite–laurite series; Bockrath et al. 2004a), arsenides, sulfarsenides, bismutho-tellurides (Lorand et al. 2010), alloys or intermetallic compounds, depending on sulfur fugacity and temperature (Fonseca et al. 2012; also see O’Driscoll and González-Jiménez 2016, this volume). Typically metasomatic minerals, such as amphibole, phlogopite, ilmenite and diamond are sometimes present, but the primary minerals can be also metasomatically added (and removed) and are then sometimes difficult to distinguish from the original assemblage (O’Reilly and Griffin 2013; Pearson and Wittig 2014).
Olivine and pyroxene separates have yielded variably elevated HSE contents (Pearson et al. 1995c; Burton et al. 2000, 2002; Luguet et al. 2009; Wittig et al. 2010a; Aulbach et al. 2014; Smit et al. 2014a; Pernet-Fisher et al. 2015). With the exception of Re, which has been suggested to be compatible in silicate minerals including garnet, especially under oxidizing conditions after sulfide break-down (Righter and Hauri 1998; Burton et al. 2002; Mallmann and O’Neill 2007; Righter et al. 2008), such enrichments may in part be due to inclusions of PGE alloys and/or sulfides that were trapped during formation or recrystallization (e.g., Walker et al. 1991; Ballhaus and Sylvester 2000; Mungall and Brenan 2014). This is also evident from the considerably higher Os concentrations in visibly inclusion-bearing mineral separates compared to inclusions-free separates from the same sample (Burton et al. 2000). Nonetheless, the compatibility of the IPGE and Rh in olivine has also been inferred from inter-element relationships in komatiites (e.g., Brügmann et al. 1987) and confirmed in several experimental studies where care was taken to avoid inclusions of PGE-rich trace phases during the analysis of the run-products (Brenan et al. 2003, 2005).
Pearson et al. (1995c) found that olivine in a dunite from Udachnaya (Siberian craton) contains ~ 400 ppt Os and less radiogenic 187Os/188Os than the whole rock. In contrast, a fertile lherzolite (whole rock) from the same locality yielded only 2 ppt Os. Work on mantle xenoliths from the Tanzanian craton has shown that olivine, inclusion-bearing or inclusion-free, has higher 187Os/188Os than sulfide or the bulk rock, which is thought to be dominated by accessory HSE hosts (Burton et al. 2000). In xenoliths from western Greenland, lower 187Os/188Os, lower IPGE and Pt concentrations but similar chondrite-normalized HSE patterns have been determined in olivine separates compared with whole rocks (Wittig et al. 2010a). Olivine separated from peridotite xenoliths from the Superior craton have HSE patterns similar to sulfide, but at lower concentrations, due to a dilution effect. Such systematics are consistent with a signature of sulfide inclusions diluted by olivine, whereas other olivine separates from different peridotites have patterns mimicking those of highly depleted, S-poor whole-rock peridotites, suggestive of PGEs being hosted within olivine by alloy inclusions (Smit et al. 2014a).
Chromite is a typical spinel present within the deep cratonic lithosphere and has been discussed as a potential host for PGE based on the observation that chromite-rich cumulates in basalts and komatiites, chromite xenocrysts and chromite in some cratonic mantle peridotites have elevated contents of the IPGE (e.g., Mitchell and Keays 1981; Barnes and Picard 1993; Hart and Ravizza 1996; Chesley et al. 1999; Graham et al. 1999, 2004; Walker et al. 2002; Luguet et al. 2009). Both the compatibility of the IPGE in the chromite crystal lattice (Capobianco and Drake 1990; Righter et al. 2004; Brenan et al. 2012) and the presence of minute PGE alloy inclusions (Ballhaus and Sylvester 2000; Sattari et al. 2002; Finnigan et al. 2008; Godel et al. 2010) have been discussed as reasons for elevated PGE in chromite. Under reducing conditions that may be applicable to the cratonic mantle, Ru most strongly partitions into chromite, whereas Ir and Rh are mildly incompatible to compatible and Re, Au, Pt, and Pd are strongly incompatible (Righter et al. 2004; Brenan et al. 2012; Park et al. 2013). Several chromite generations in peridotite xenoliths from the Argyle lamproites (just off the Kimberley craton margin, Western Australia) were identified, separated and measured for Re–Os isotopes by Luguet et al (2009). These authors found that a primary chromite generation preserved extremely unradiogenic 187Os/188Os giving Archean ages, but likely experienced (late) Re gain, whereas a late, symplectite-derived generation (from the break-down of garnet) gave Mesoproterozoic TRD similar to the age of the host lamproite (Luguet et al. 2009).
Given the siderophile and chalcophile nature of the HSE (Peach et al. 1990; Fleet et al. 1999; Bockrath et al. 2004b), their concentration in PGM (Luguet et al. 2007; Lorand et al. 2010, 2016, this volume; O’Driscoll and Gonzáles Jiménez 2016, this volume) and in BMS is orders of magnitude higher than in whole rocks dominated by silicates (e.g., Pearson et al. 1998, 1999; Lorand et al. 2008, 2010, 2013). This has also been demonstrated experimentally (Brenan et al. 2016, this volume). Consequently, these minerals dominate the bulk-rock budget of mantle rocks, despite having ≪ 1% modal abundances in primitive or depleted mantle, and they are amenable to being analysed by single-grain techniques. The presence of accessory sulfide or alloy in cratonic mantle peridotites has long been postulated based on direct observations on mantle samples (e.g., Frick 1973), the observation of low PGE contents of constituent silicates (Mitchell and Keays 1981) and because the poor reproducibility of PGE and Re in some rocks requires the so-called “nugget-effect”, i.e., the irregular distribution of HSE-rich accessory minerals in the sample powder (e.g., Ravizza and Pyle 1997; Pearson and Woodland 2000; Reisberg and Meisel 2002).
In the absence of PGM, HSE are expected to strongly prefer BMS over silicate minerals or melt in mantle rocks (Mitchell and Keays 1981; Lorand 1989; Fleet et al. 1996; Hart and Ravizza 1996; Burton et al. 1999; Alard et al. 2000; Sattari et al. 2002). In particular, within cratonic peridotites, sulfides can contain 1000s of ppm or even % levels of the HSE, either due to the effect of extreme concentration these elements into remaining sulfide during partial melt extraction (Jgaoutz et al. 1979; Alard et al. 2000 Griffin et al. 2002), or because they derive from PGE alloys in a highly refractory residue that was captured by BMS, in which case discrete Pt-rich microphases can be present in the sulfides (Griffin et al. 2002, 2004). For simplicity, all such secondary PGE-rich sulfide assemblages will be referred to here as “resulfidized PGM”. In addition, HSE-rich monosulfides may yield abundant discrete PGM by subsolidus cooling (Lorand and Luguet 2016, this volume).
Pre-metasomatic cratonic mantle sulfides are typically monosulfide solid solutions (mss) that exsolve to low-temperature assemblages of pyrrhotite plus pentlandite and chalcopyrite in the mantle and during transport to the surface (Mitchell and Keays 1981; Lorand and Conquéré 1983). Lorand and Luguet (2016, this volume) review a wide range of exsolved PGM (laurite, Ru–Os–Ir sulfarsenides, Pt–Fe alloys) in cratonic mantle samples. The presence of Pt–Ir–Os alloys embedded in BMS has been inferred by Griffin et al. (2002) due to the irregularities in signal intensity during in-situ laser analysis. The occurrence of such minerals in residual mantle, possibly as micro- or nano-inclusions, has also been inferred from the PGE systematics of basalts and komatiites (Fiorentini et al. 2011; Fonseca et al. 2011) and is expected for peridotites that have been depleted to S exhaustion, based on analytical (e.g., Pearson et al. 2004) and experimental evidence (e.g., Fonseca et al. 2012). The modelling of experimentally derived alloy solubilities and stability at a range of P, T and oxygen and sulfur fugacities (fO2, fS2) by Mungall & Brenan (2014) indicates that there is no field of mantle stability for Pd alloys in nature but that Pt alloy stability is expected at ambient TP whereas Ir alloy stability is broader still, suggesting the presence of Os, Ir, and Ru alloys even at very high (50%) levels of melt depletion and excess TP. This, combined with the possibility that some cratonic mantle roots reach depths where iron-nickel metal saturation is expected, due to falling fO2 with increasing P (Ballhaus 1995; Rohrbach et al. 2007; Rohrbach and Schmidt 2011; Stagno et al. 2013), suggests that HSE could be hosted in PGE alloys or in iron-nickel metal in strongly depleted, sulfide-free shallow lithosphere and in the deeper reaches of cratonic mantle, respectively. These minerals may be easy to overlook and even if actively searched for, may be too small to be observed with conventional analytical techniques. Their presence as micro- or nanoinclusions may be responsible for PGE patterns with low Pd/Ir in olivine separates that strongly resemble those of highly depleted, S-poor cratonic xenoliths (Smit et al. 2014a), as outlined above.
ANALYTICAL TECHNIQUES FOR CRATONIC MANTLE PERIDOTITES
Analytical techniques for the accurate and precise determination of HSE concentrations and Re–Os isotopes are reviewed in Meisel and Horan (2016, this volume). The spectrum of possible HSE hosts in peridotites has given rise to various sample digestion and analytical strategies that each has their advantages and drawbacks. Clearly, the best approach is to use a combination of techniques in order to obtain maximum constraints on the origin and evolution of the cratonic lithosphere from a HSE perspective. Here, we focus on those that are particularly usefully applied to cratonic mantle samples. Actual data from various sample types and their implications will be discussed in later sections dealing with signatures of melt depletion and metasomatism.
Whole rocks and mineral separates
Cratonic peridotite xenoliths contain sufficient Re and PGE abundances—mostly concentrated in BMS and PGM—to allow determination of Re–Os isotopes and PGE abundances by routine isotope dilution mass spectrometry techniques (Creaser et al. 1991; Walker et al. 1989). The presence of irregularly distributed accessory HSE-rich minerals poses several challenges to obtaining precise and accurate results from whole rock powders of mantle xenoliths. Firstly, a technique has to be chosen that ensures that refractory minerals are reliably digested. This may be achieved by—and requires—high-temperature and -pressure techniques, such as Carius tube or high-pressure asher digestions at 300 ºC. A drawback of such methods is that in general, sample sizes are limited to 5 g or less (Shirey and Walker 1995; Meisel et al. 2001; Becker et al. 2006), leading to potential problems from sampling errors and the “nugget effect” (Pearson and Woodland 2000). Alternatively, a Ni–S fire assay can be used (e.g., Pattou et al. 1996; Maier et al. 2012), which allows larger aliquots to be analyzed (therefore not practicable for mineral separates), potentially lessening the impact of the “nugget effect” (e.g., Lorand et al. 2013). This method suffers from larger blank contributions and, unless great care is taken, possible incomplete digestion of refractory phases, producing data with relatively large scatter (Pearson and Woodland 2000; Becker et al. 2006). It is clear amongst practitioners that a Te co-precipitation step is required to ensure best recovery of PGE during Ni–S fire assay (e.g., Gros et al. 2002; Maier et al. 2012). Perhaps the most significant draw-back of this technique, especially for cratonic peridotites, is that obtaining high-quality Os isotope and concentration data is difficult and it is not possible to obtain quantitative Re data, such that combined Re–Os plus PGE studies on the same sample dissolution (Pearson and Woodland 2000) are precluded. Also, while it is desirable for any of the above methods that the sample size be sufficiently large to diminish sampling errors due to the “nugget effect”, this is often not achievable for kimberlite-entrained cratonic mantle xenoliths, which—by nature or because of their comminution in diamond processing plants—are limited to hand specimens rarely exceeding 5–10 cm in the longest dimension, especially outside of the historic mines of South Africa, where access and size is controlled by exploration companies and smaller concentrate samples are produced by modern kimberlite processing procedures.
Several instrumental approaches are available for PGE abundance and isotope ratio measurement. Measurement of Os isotopes by N-TIMS remains the most widely used method (Creaser et al. 1991; Völkening et al. 1991). The sparging technique, where Os is volatilized directly from the sample solution and measured online using an ICPMS instrument (Schoenberg et al. 2000; Hassler et al. 2000), suffers from incompletely understood problems (Liu et al. 2010). In contrast, measurement of PGE concentrations by isotope dilution sector-field ICPMS and of Re–Os isotopes by multicollector ICPMS, where the sample solution is aspirated into the torch with a nebulizer/spray chamber (Pearson and Woodland 2000; Day et al. 2003; Nowell et al. 2008), has yielded reproducible and accurate results of sufficient precision to be usefully applied to mantle xenoliths (Sen et al. 2011; Aulbach et al. 2014). In this case, care has to be taken to exclude the severe memory problems that can affect the sample introduction system (Pearson and Woodland 2000).
The potential presence of multiple generations of BMS and/or PGM in a whole-rock sample, described above, implies that a mixed signal is obtained (Alard et al. 2002; Griffin et al. 2002, 2004). In general, for cratonic mantle samples, sulfide-rich rocks and sulfides having Ni/Fe ratios that are inappropriate for residual phases remaining after melting extents exceeding 30% (Lorand and Grégoire 2006) should be avoided in the whole rock approach. Rather, S-poor, refractory samples, in which all the Os resides in laurite/erlichmanite series minerals or alloys that were produced during the melting event that formed the rock should be targeted (Pearson and Wittig 2014). Single sulfide grains measured in some cratonic samples have yielded significantly lower 187Os/188Os, and therefore older minimum ages, than their corresponding whole rocks (e.g., Griffin et al. 2004; Bragagni et al. 2014).
Several studies report HSE and/or Re–Os isotopes in peridotitic mineral separates that were measured with the aim to identify mineral hosts of the HSE, to obtain information uncorrupted by ubiquitous kimberlite addition during xenolith entrainment or to access a component in xenoliths that is dominated by included and presumably therefore least modified HSE carriers (e.g., Chesley et al. 1999; Burton et al. 2000; Hanghøj et al. 2001; Luguet et al. 2009; Wittig et al. 2010a; Aulbach et al. 2014; Pernet-Fisher et al. 2015). For example, in the case of Smit et al (2014a) the small kimberlite-derived xenoliths were so extensively invaded by kimberlite melt that digestion of a whole-rock sample was not useful; olivine separates were the least compromised part of the rocks. The sample preparation and analytical requirements for peridotitic mineral separates are identical with those of the whole rocks reviewed above, since complete digestion also of refractory phases is required for meaningful results. Ruthenium analyses by ICPMS in chromites must be evaluated with caution due to the detrimental effects of Cr-based polyatomic interference.
Advancements in analytical techniques over the ~15 years have allowed Re–Os isotope measurement of single sulfide grains (see review by Harvey et al. 2016, this volume). Two approaches have been applied. One measures Re and Os isotopes using Negative Thermal Ionization Mass Spectrometry (NTIMS) after digestion of entire, or close to entire, single grains isolated from mantle xenoliths or diamonds (e.g., Pearson et al. 1998, 1999; Richardson et al. 2001, 2004; Luguet et al. 2008; Aulbach et al. 2009a,b,c, 2011; Harvey et al. 2011). The other uses an in situ approach, where the sulfide is analysed using laser ablation microprobe inductively-coupled plasma mass spectrometry (LA-ICPMS); if applied to sectioned rocks, valuable microstructural context is preserved (Alard et al. 2000, 2002; Pearson et al. 2002a; Aulbach et al. 2004; Griffin et al. 2004; Zheng et al. 2007). A hybrid method consists of extracting sulfide remains in polished thick sections, followed by solution and measurement by NTIMS (Bragagni et al. 2014; Luguet et al. 2015). The concentration of the analyte in the sulfide grains dictates the size of the grains that must be processed to achieve satisfactory counting statistics during measurement and useful precision on the concentration and isotope ratio, if measured.
Typically, unless a diamond inclusion study is being performed, sulfide grains < 30–50 μm are not analyzed, meaning that if such a fraction is present in the whole rock, its HSE and isotopic contribution cannot be directly accounted for. These issues limit how representative the analysis of single or multiple sulfide grains per sample is of the whole rock and mantle source region, especially in highly depleted peridotites where melt extraction has proceeded to the level of sulfide exhaustion and PGE such as Os, Ir and Ru are likely hosted in micron to sub-micron phases such as laurite or alloys (Luguet et al. 2007; Lorand et al. 2008, 2010; Kogiso et al. 2008) in addition to olivine (Mungall and Brenan 2014). The in situ method additionally requires the correction of 187Os from the isobaric interference of 187Re, which in practice means that accurate Re–Os isotope data can only be confidently obtained for sulfides with low Re/Os (N.J. Pearson et al. 2002b; Nowell et al. 2008), leading to exclusion of unsuitably high-Re/Os grains present in some samples. Given the mostly young emplacement ages of kimberlite, which likely reflect the limited preservation of older kimberlites (e.g., Tappe et al. 2014), the differential distribution of Re and Os associated with sub-solidus exsolution of mss has a limited effect on Os isotopes, but will contribute to uncertainties in model ages and produce scatter on Re–Os isochron diagrams.
Thus, where possible, intact or recombined whole sulfide grains should be analyzed to avoid these complications (Richardson et al. 2001; Rudnick and Walker 2009; Pearson and Wittig 2014). Both solution and in situ techniques have shortcomings in this regard because the former do not always succeed in recovering the entire sulfide (Richardson et al. 2001; A. Wainwright pers. comm. 2014) and the latter may not sample representative portions of grains exposed in sections even if total ablation is attempted (see discussion in Pearson and Wittig 2014). Nevertheless, the single grain technique has led to the recognition that cratonic peridotites—and even single diamonds—can harbor multiple generations of sulfide (Pearson et al. 1998; Alard et al. 2000; Griffin et al. 2002, 2004) and is the only way of demonstrating that some diamonds grow in pulses that may be separated by 1 Ga (Wiggers de Vries et al. 2013). Although PGM lend themselves to in situ or single mineral techniques similar to those described for BMS, large enough grains suitable for in-situ analyses in cratonic mantle samples have not been reported so far.
UTILIZATION OF THE Re–Os ISOTOPE SYSTEM IN CRATONIC MANTLE STUDIES
The development of Re–Os isotope geochemistry has played a pivotal role in dating lithosphere stabilization and exploring links between continental crust and mantle formation (Walker et al. 1989; Carlson and Irving 1994; Pearson et al. 1995b,c, 2007; Pearson 1999; Griffin et al. 2002, 2004; Carlson et al. 2005, 2008). Rudnick and Walker (2009) give a comprehensive general review of the principles and issues of interpreting Re–Os isotope ages, and an assessment specifically with regard to dating cratonic mantle samples is provided by Pearson and Wittig (2014). The salient characteristics and complexities in the application of the Re–Os isotope system are summarized here.
Because Re is mildly incompatible in the absence of sulfide melt (with a distribution coefficient broadly similar to Al, Yb, or Ti), whereas Os is compatible during peridotite partial melting and generation of basalt (e.g., Righter and Hauri 1998; Brenan et al. 2003; Righter et al. 2004; Fonseca et al. 2007; Mallmann and O’Neill 2007), isolation of lithospheric from convecting mantle by melt extraction leads to strong fractionation of the daughter from the parent element, in particular if the degree of melting is great enough to exhaust sulfide from the mantle residue, which occurs after ~ 25% of partial melting (Keays 1995), but may occur at very low degrees of melting in oxidized mantle (Mungall et al. 2006; Botcharnikov et al. 2013). This results in Os isotope compositions being virtually frozen in at the time of the strong melt depletion inferred for cratonic mantle (Walker et al. 1989; Shirey and Walker 1998). In addition, the high Os concentrations in residual mantle compared to low concentrations in metasomatic melts and fluids confer a high degree of robustness to post-formation overprint (Reisberg and Lorand 1995; Rudnick and Walker 2009). However, no Re–Os isotope study of cratonic peridotites has so far resulted in a meaningful isochron relationship that defines a clear partial melting event, with the possible exception of peridotites from Hannuoba (Gao et al. 2002), which have isochron relationships dating the formation of new lithosphere at ~1.9 Ga. Much of the scatter in Re–Os geochronology results from direct modification of Os isotope compositions and/or from changes Re addition and time-integrated ingrowth of radiogenic Os, e.g., following sulfide breakdown or addition (Rudnick and Walker 2009).
Two types of model age have been formulated for the application of the Re–Os isotope system to mantle samples (Walker et al. 1989; Shirey and Walker 1998): (1) An age for the isolation from the model convecting mantle reservoir (TMA) assuming that all Re is primary—this is analogous to model ages calculated using Sm–Nd (McCulloch and Wasserburg 1978); (2) A Re-depletion age (TRD) assuming that all Re has been extracted during melt depletion and that any Re present was added later—this is a minimum age whereby the measured 187Os/188Os is projected to the mantle evolution line assuming Re/Os is zero. A variation on this age is the TRD eruption age (Pearson et al. 1995b) calculated for mantle xenoliths, assuming that all measured Re was added during xenolith transport in the host magma, which is manifested by sub-horizontal arrays in the Re–Os isochron diagram (e.g., Carlson et al. 2005). In this case, the measured 187Os/188Os is corrected for the ingrowth of 187Os from Re since kimberlite emplacement and a TRD is calculated for the resultant value (Pearson et al. 1995b; Shirey and Walker 1998; see their Fig. 1 for an illustration of the various model ages).
An initially depleted, but refertilized mantle may have PGE and Re–Os isotope systematics that are difficult to distinguish from initially less depleted mantle. Unless Re addition has occurred quasi-contemporaneously with melt depletion or with emplacement of the host magma, metasomatic Re addition to the cratonic mantle during its long isolation from the convecting mantle will lead to calculated TMA that are too old and TRD that severely underestimate the true age—assuming preservation of original 187Os/188Os (Rudnick and Walker 2009). Thus, TRD should not be calculated for moderately to minimally depleted mantle samples that did not suffer quantitative Re extraction, and TMA may be meaningless for those bulk rocks which likely had a multi-stage Re–Os isotope evolution as inferred from PGE systematics (Pearson and Wittig 2014). Consequently, application of the Re–Os isotope system to whole-rock peridotites requires careful selection of samples that show the highest levels of melt depletion (usually the least diopside- and garnet rich), minimal modal metasomatism and minimal addition of metasomatic sulfides. The use of combined PGE and Re–Os isotopes in whole-rock studies reveals that often the oldest Re depletion ages are obtained from the samples with the most Platinum-group PGE (Pt and Pd: PPGE) depleted normalized PGE patterns (D.G. Pearson et al. 2002), irrespective of their Re enrichment.
Recently, König et al. (2014) and Luguet et al. (2015) have shown that combined S and Se/Te measurements are highly useful in deciphering whole-rock Re–Os and PGE systematics. They find that, within a suite of peridotites from Letlakhane, Botswana, the most depleted rocks that are sulfide-free contain high but variable Se/Te ratios (15–40) and very Pd- and Pt-depleted normalised patterns. Such characteristics are indicative of melting residues. Indeed, these samples yield the oldest, most robust whole-rock Os isotope ages. In the same suite of rocks, only the most metasomatized BMS-rich peridotites experienced enrichment of both 187Os/188Os and Pd/Ir at constant Se/Te, such that the Os isotopes yielded unrealistically young TRD eruption ages for mantle root stabilization in this region.
In an effort to isolate the oldest Os isotopic component in mantle xenoliths, sulfide inclusions, or mineral separates, where the Os signature is presumably dominated by primary inclusions, have been targeted. Sulfide inclusions have in some instances been shown to preserve older age information and more residual HSE systematics compared to sulfides occurring in the interstices (Burton et al. 1999; Alard et al. 2002, 2011; Harvey et al. 2011). This has led to the expectation that silicate-hosted sulfide grains, being protected from post-formation modification by their host, will preserve older age information than interstitial grains. However, mantle metasomatism induces recrystallization (Drury and van Roermund 1989), during which metasomatic minerals including sulfides can be trapped in primary minerals. Therefore, inclusions do not necessarily yield Archean or even the most ancient ages in cratonic—or any—mantle xenoliths (e.g., Burton et al. 1999). Detailed recent studies of whole rocks and associated sulfide grains with particular attention to various textural contexts (included versus interstitial or in melt pools) have revealed an almost bewildering lack of correspondence between textural position and Os isotope systematics, and interstitial sulfide grains in strongly metasomatized cratonic and other peridotites can give Archean model ages (Bragagni et al. 2014; González-Jiménez et al. 2014; Wainwright et al. 2015). Nevertheless, because metasomatic sulfur addition sometimes appears to be accompanied by only minor HSE input, in particular the compatible IPGE, previously hosted in residual alloys, can retain their original, unradiogenic 187Os/188Os for eons (Marchesi et al. 2010; Fonseca et al. 2012) even if they are now present as sulfides (Griffin et al. 2002, 2004; Lorand et al. 2013). An example of the “resulfidation” of older residual alloys, imparting the ancient signatures of these grains on more recent metasomatic sulfide phases has been documented by Wainwright et al. (2015).
Though rare, meaningful ages have been been obtained from Re–Os isochrons for several peridotitic sulfide suites from the central Slave craton, yielding radiogenic initial 187Os/188Os (Aulbach et al. 2004, 2009a, 2011; Westerlund et al. 2006). Consequently, model ages for samples with unradiogenic Os are systematically underestimated when projected back to a primitive mantle evolution curve. This highlights the effect of uncertainty regarding source composition for model age calculations. The uncertain evolution and pronounced heterogeneity of present-day convecting mantle (Brandon et al. 2000; Meibom et al. 2002; Walker et al. 2002; Pearson et al. 2007) further diminish the resolution of Re–Os ages. Despite the smaller isotopic diversity of the early convecting mantle (Rudnick and Walker 2009), an uncertainty of ca. 200 Ma may still be indicated for model ages calculated for Archean samples, and cratonic mantle lithosphere can only be unambiguously identified if a sufficiently large number of samples yielding Archean ages are analyzed (see also discussion of data presentation in Pearson and Wittig 2014).
Despite the many complications, the Re–Os isotope system has confirmed the presence of Archean mantle beneath many cratonic regions and has provided remarkably detailed information on coupled cratonic crust and mantle evolution. One of the clearest examples of the success of both sulfide and whole rock analytical approaches in defining cratonic mantle lithosphere is given by the multitude of studies performed on and adjacent to the Kaapvaal craton, summarized in Janney et al. (2010) and Pearson and Wittig (2014). The mode in TRD ages given by xenolith suites from all kimberlites erupted within the recognized crustal bounds of the craton is Archean, with the exception of those from the Premier mine, which samples mantle with a major Mesoproterozoic overprint (Shirey et al. 2002). In contrast, while there are signs of Archean heritage in some xenoliths from circum-cratonic kimberlites, their dominant age signature is Mesoproterozoic (Janney et al. 2010; Aulbach et al. 2014). A link between large-scale mantle melt depletion events, gauged by Os isotopes in a variety of mantle materials (BMS, PGM, peridotites), and spurts of continental crust formation are observed no earlier than 3.3 Ga ago, and again at 2.7, 1.9, and 1.2 Ga (Pearson et al. 2007). Accordingly, juvenile crust formation pulses at 4.2 and 3.8 Ga—evidenced by zircon U-Pb ages combined with Hf and O isotopes—were apparently not accompanied by the formation of cratonic lithosphere (at least none that survived to be sampled today). The oldest preserved vestiges of cratonic mantle probably formed ca. 3.5 Ga ago (Griffin et al. 2014). In some instances, cratonic mantle has been discovered beneath younger mobile belts: This applies to the ca. 1.8 Ga Nagssugtoqidian mobile belt in northwestern Greenland (adjacent to Greenlandic part of the North Atlantic craton; Wittig et al. 2010a) and the ca 1.9–1.8 Ga Halls Creek Orogen in northwestern Australia (at the southeastern margin of the Kimberley craton; Luguet et al. 2009). In other instances, Re–Os isotope systematics reveal that deep cratonic keels were lost (by an uncertain mechanism) and replaced by younger mantle (of uncertain origin), the most prominent case being that of the North China craton (Gao et al. 2002), which will be discussed further below. Finally, it is worth noting that crust-mantle links exist not only for the timing of cratonic lithosphere formation, but also for its subsequent tectonothermal evolution, as evidenced by multiple peaks in age spectra for sulfide or peridotite suites from a single mantle section, which may reflect new lithosphere formation or overprint during collisions or rifting subsequent to cratonization (e.g., Pearson et al. 1995c; Pearson 1999; Griffin et al. 2002, 2004; Wittig et al. 2010a).
EFFECT OF MELT DEPLETION DURING CRATONIC LITHOSPHERE STABILIZATION ON SULFUR AND HSE SYSTEMATICS
Partial melting of convecting mantle generates a lithosphere that should be enriched in compatible elements, such as the IPGE and depleted in incompatible elements, such as the PPGE and Re, depending on extent of melting (e.g., Pearson et al. 2004). Because of the extreme compatibility of HSE with sulfide phases, the main effect of partial melt extraction will be through consumption of sulfide; after sulfide has been exhausted, the HSE are dominated by IPGE compatibility in olivine and by possible saturation in alloy (e.g., Mungall and Brenan 2014).
Sulfur and the persistence of sulfides
Komatiites are high-degree melts that have been suggested to be complementary melts to cratonic lithosphere (e.g., Herzberg and O’Hara 1998). Since some komatiite types are sulfide-undersaturated at source (Fiorentini et al. 2011), an absence of primary residual sulfides in highly depleted cratonic mantle could be implied, with the potential exception of sulfide trapped in silicates during the partial melting. This is true, for example, for the Kaapvaal craton, where petrographic evidence and young Re–Os isotope ages for peridotitic sulfide inclusions in diamond suggest secondary introduction of sulfide into an initially (post-melting) sulfide-barren lithosphere (Lorand and Grégoire 2006; Aulbach et al. 2009a).
Despite the general expectation that the primary cratonic mantle residue is sulfide-free, the thickness of cratonic mantle roots combined with their formation during polybaric melting (e.g., Walter 1998, 2005; Herzberg 2004) warrants the consideration that primary residual sulfide could be present in the deeper, less depleted cratonic mantle regions, although the least depleted, deepest portions may have been lost (Arndt et al. 2009). The presence—or not—of sulfide at the base of the cratonic mantle depends on depth of initiation of melting (Pi) and the depth of any pre-existing lithospheric lid limiting further decompression melting (e.g., Sobolev et al. 2007), and also depends on the fO2 of the convecting mantle, as will be discussed in more detail in a later section. If all PGE are quantitatively retained and little fractionated in the residue, as long as sulfide is stable (Mungall and Brenan 2014), one might expect that the deepest and hence least depleted cratonic mantle residues have relatively high PGE concentrations at near-chondritic ratios of PPGE to IPGE (bearing in mind that during cratonic lithosphere formation all sulfide is molten and hence there is no incongruent melting, during which PGE would be fractionated).
The relatively low PPGE/IPGE, compared to other melts, in Al-depleted komatiites that formed at great depths and relatively small degrees of partial melting (> 200 km, F = 0.2; Arndt et al. 2008) were suggested to indicate the presence of residual sulfide in their source. Figure 4 shows that there are few cratonic samples with high PGE concentrations and high Al2O3 contents consistent with extraction of small amounts of partial melt in the presence of sulfide, whereas such systematics are common in orogenic peridotites. Indeed, none of the cratonic peridotites in the dataset should contain primary residual sulfide, which would be exhausted before Al2O3 decreases to < 1.5 wt% (Pearson et al. 2004), yet most have S concentrations that are vastly too high—even in excess of primitive mantle values—for the degree of depletion gauged from Al2O3 concentrations at the conditions considered here (Fig. 5). In contrast, most orogenic and some off-cratonic peridotites, which formed at lower initial pressures, where Al2O3 decreases more rapidly compared to S, may have retained primary sulfide.
At decreasing sulfide modes prior to final exhaustion, concentrations of the PGE within the sulfide liquid will increase dramatically and may reach wt% concentrations, possibly leading to the saturation of PGM immediately prior to final exhaustion of sulfide liquid (Mungall and Brenan 2014). This could explain why some sulfides from cratonic peridotites have much higher HSE concentrations than those from younger samples that were generally never depleted to the point of sulfide exhaustion, such as abyssal and orogenic peridotites (Fig. 6).
The expectation of PGE alloy saturation during generation of cratonic mantle combined with the presence of BMS, some containing discrete PGM grains, indicates that resulfidation of the mantle may have occurred (Griffin et al. 2002, 2004). The associated capture of residual PGE alloys by BMS melt droplets should lead to formation of composite alloy–BMS grains (Lorand and Luguet 2016, this volume). Alternatively, in the absence of residual IPGE alloy, addition of sulfur to previously melt-depleted mantle may extract IPGE from olivine to form mixtures of BMS and PGE sulfides.
Primary residual sulfides should be distinguishable from resulfidized PGM, which are expected to have low Pd/Os and Re/Os since Pd and Re do not form alloys and are incompatible in olivine, in addition to low Pt/Os for Pi >3 GPa (Mungall and Brenan 2014). There are few complete PGE datasets for sulfide in cratonic mantle samples. Peridotitic sulfide inclusions in diamond, and sulfide grains in peridotite xenoliths from the Kaapvaal and Siberian craton, have Pt/Os from ~0.002 to 20 and show a rough negative correlation with Os concentration (Fig. 6). In time-resolved laser ablation spectra these sulfides display signal intensity spikes indicative of the presence of PGE alloy inclusions, which, combined with their low Pt/Os, is consistent with them being mixtures of BMS and PGM; however, the timing of exsolution of PGM from the BMS is impossible to constrain (Griffin et al. 2002, 2004). Given that most cratonic peridotites apparently gained considerable amounts of secondary S, pervasive resulfidation of the cratonic mantle may help explain why primary alloys, though expected and suspected in many though not all restites, have to date not been directly observed. Nevertheless, little-fractionated Pt/Os and Re/Os at elevated concentrations are observed for sulfide inclusions in diamond and for some sulfide grains in cratonic peridotites (Fig. 6), perhaps consistent with a primary residual sulfide population that is not depleted enough to evolve to unradiogenic Os giving Archean TRD.
A single common PGE host—whether primary sulfide, BMS–PGM mixture or olivine—is indicated for sample suites showing tight correlations of PGE concentrations (Lorand et al. 2013). Such correlations are observed for Os with Ir in all cratonic peridotite suites (Fig. 7A), which may indicate saturation of Os–Ir alloys at high concentrations. In contrast, the less common very low Os and Ir concentrations may indicate failure of alloy to saturate, in which case both elements would be hosted by olivine, or secondary loss of (resulfidized) alloy. Some superposition of secondary effects is particularly evident in the basalt-borne peridotite xenolith variety (Fig. 7A). Ruthenium and Ir are also positively correlated in cratonic peridotites (Fig. 7B), probably due to compatibility of both of these IPGE in olivine, but show more scatter to high Ru at a given Ir concentration for the Kalahari and Greenland sections. This could point to a nugget effect from inhomogeneous distribution of chromite, to inhomogeneous distribution of alloys containing Ru or to analytical artefacts due to uncorrected polyatomic interferences involving Cr.
Large scatter is also observed for Pt versus Ir concentrations in some cratonic mantle sections (Kalahari, Greenland; Fig. 7C). Samples with coherent Pt and Ir at the high concentration end of the array probably indicate the presence of variable amounts of primary sulfide, but coherent behavior at lower concentrations may represent mantle that experienced co-saturation of Pt and Ir alloy at low Pi and fO2 during melt extraction to shallow depths. Offsets to higher Pt at a given Ir concentration may be explained by metasomatic Pt addition, whereas lower Pt may point to the failure of Pt to saturate in the melt at higher Pi and/or fO2 and consequent incompatible behavior during partial melting (Mungall and Brenan 2014). In the absence of primary residual sulfide, roughly equal concentrations of Pt and Pd (Fig. 7D) may be indicative of metasomatic introduction with sulfide, whereas depletions of Pd relative to Pt would be consistent with the stronger incompatibility of Pd during partial melting, as inferred from high-volume melts (e.g., Barnes et al. 1985; Puchtel and Humayun 2000), and/or stabilization of Pt in the residue under reducing conditions and at low pressures.
The concentrations of both Se and Te in selected cratonic mantle harzburgites are among the lowest of any measured from the Earth’s mantle (König et al. 2014) and this is empirically consistent with their melt-depleted character. The relative partitioning behavior of the chalcogens Se and Te has been inferred from studies of various peridotites and sulfides (Hattori et al. 2002; Lorand and Alard 2010; König et al. 2012, 2014, 2015; Wang et al. 2013; Wang and Becker 2013). In peridotites that have suffered sulfide exhaustion, Te has been shown to behave incompatibly—more so than Se—leading to superchondritic Se/Te (> 20; König et al. 2014, 2015). In agreement with this, the abundances of these elements in cratonic harzburgites are distinctly lower than in orogenic peridotites, but at highly variable absolute concentrations (Fig. 5). This observation may, in part, be related to a nugget effect caused by accessory sulfide or by PGM that inherited high Se/Te from residual sulfide (Fig. 6), or to secondary overprint (König et al. 2014, 2015). Metasomatic enrichment by high-Te, low-Se metasomatic phases, such as Bi–Pt tellurides, will lower Se/Te ratios at constant Al2O3 re-creating “chondritic” Se/Te ratios in samples varying from lherzolite to harzburgite (König et al. 2014). Basalt-borne xenoliths show conspicuous offsets toward lower S concentrations at a given Se or Te, consistent with S loss and Se–Te stability during degassing in subaerially emplaced host basalt (Lorand et al. 2013). While Se and Te are more robust than S to loss during degassing in peridotites, because of metasomatic disturbances, none of these elements is likely to be a faithful recorder of melting conditions in cratonic peridotites.
HSE PROCESSING DURING MANTLE METASOMATISM
Modification of the lithospheric mantle composition during interaction with fluids or melts is known as “mantle metasomatism”. This process may be “cryptic”, resulting in changes to the trace-element budget usually manifest as strong enrichment of LREE over HREE and addition of LILE introduced by invading volatile-rich fluids or small-volume melts. Alternatively “patent” or “modal” metasomatism is accompanied by major-element enrichment or depletion and modification of the modal mineralogy, including the addition or removal of minerals (e.g., Frey and Green 1974; Harte 1983; Dawson 1984; Menzies and Hawkesworth 1987; Erlank et al. 1987). The large scatter in the PPGE and Re data suggests that such secondary processes have obscured a significant part of the original signatures. The diversity of fluids and melts—reduced or oxidized, poor or rich in volatiles, small or large volumes—that may percolate and interact with the mantle (O’Reilly and Griffin 2013), combined with the antiquity of cratonic lithosphere, gives rise to a wide spectrum of conditions under which HSE can be added, removed and redistributed over the long course of cratonic histories.
The style and intensity of mantle metasomatism have consequences for the long-term survival of cratons, which depends critically on their viscosity and density as a function of H2O contents and temperature (Jordan 1988; Abbott et al. 1997; Lenardic et al. 2003; Afonso and Schutt 2012; Wang et al. 2015). In extreme cases, intra-plate and subduction processes may lead to tectonic erosion, thermochemical erosion or whole-sale delamination and replacement by convecting mantle. This may be evidenced, inter alia, by cratonic mantle xenoliths suites entrained in basalts rather than kimberlites. Because of their varied geochemical behavior, HSE and Re–Os isotope systematics are suited to address the extent of craton survival and disruption beneath thinned cratonic areas. Below, we discuss the effects of various metasomatic styles on HSE abundances and Re–Os isotopes of cratonic mantle, presenting case studies that document various stages of cratonic lithosphere destruction through metasomatism.
Modification during Intraplate Mantle Metasomatism
Silica-undersaturated melts and fluids
During incipient partial melting of a carbonated peridotite source, low-volume, silica-undersaturated melts are generated that form a continuum between carbonatite and carbonated silicate melts, such as kimberlites (Dalton and Presnall 1998). Low-volume, residual melts may also form through reactions during percolation of a silicate melt through the lithosphere (e.g., Vernières et al. 1997) or by immiscibility (e.g., Kogarko et al. 1995). The signatures of the passage of such melts through the cratonic mantle are ubiquitous (Pearson et al. 2005; O’Reilly and Griffin 2013). The contents of HSE in carbonatite and kimberlite melts at depth are presumably initially very low since low-volume melts should leave a sulfide-bearing residue, which effectively retains the HSE (Mungall and Brenan 2014), assuming that the convecting mantle is not highly oxidized. Relatively high PGE and Re contents (> 1 ppb) have been measured in some kimberlites and related rocks (e.g., McDonald et al. 1995; Graham et al. 1999, 2004; Chalapathi Rao et al. 2014), but were ascribed to contamination of the kimberlite magma by a high load of lithospheric mantle-derived xenoliths and xenocrysts (e.g., Kopylova et al. 2007). Sub-ppb levels of PGE and Re with fractionated HSE patterns have been determined for carbonatites, with higher concentrations also presumably due to contamination by mantle materials (Widom et al. 1999; Graham et al. 2004; Xu et al. 2008). Indeed, cryptic metasomatism by low-volume melt has been shown to have little effect on the HSE systematics of mantle xenoliths having highly unradiogenic (enriched) Nd and which retained Os with an ancient melt depletion signature (Carlson and Irving 1994; Pearson et al. 1995c; Olive et al. 1997; Aulbach et al. 2014).
Consistent with a predominantly cryptic effect, metasomatic S addition—required to explain the elevated S contents in refractory cratonic peridotites—is not linked to the introduction of a major element, such as Al2O3 (Fig. 5; Pearson et al. 2004). Similarly, Se and Te concentrations in harzburgite xenoliths from northern Lesotho (SE margin of the Kaapvaal craton) do not correlate with Al2O3 (e.g., Lorand et al. 2003a) (Fig. 5). Metasomatic introduction of sub-micrometric tellurides containing little Os and Re has also been documented (König et al. 2014), but the, on average, distinctly low Se and Te concentrations in cratonic harzburgites indicate that very little of this type of secondary material has been added. Of course, the re-introduction of sulfides (where S is likely sited) or tellurides constitutes a form of modal metasomatism but is rarely treated as such if not also accompanied by changes to silicate mineralogy.
A feature of both cratonic residues and residues from other tectonic settings, is the wide variation of Ir and Os concentrations, far in excess of levels that can be generated by melt depletion for reasonable redox conditions during partial melting of convecting mantle (ΔFMQ ≤ 0) (Figs. 4A,B). Metasomatic effects may play a role here, such as (resulfidized) PGE alloy destabilization during post-formation open-system interaction with low-volume oxidizing melts. Indeed, the general assumption of low-volume melt extraction leaving behind sulfide-saturated sources is challenged by the presence of high and unfractionated relative abundances of the HSE in some low-degree melts, and preferential digestion of more fusible components has been suggested before (Day, 2013). Such characteristics attest to efficient removal of sulfide (and PGE alloy) from their source at low F, under relatively oxidizing conditions (Mungall et al. 2006; Botcharnikov et al. 2013).
Kimberlites and carbonatites may form through redox reactions involving oxidizing fluids and deep, initially reducing and diamondiferous cratonic roots (Foley 2011). The imprint of minimally-fractionated, oxidizing silico-carbonatite and its destructive effect on diamond in the deeper lithosphere has been well-documented (e.g., Fedortchouk and Canil 2004; Creighton et al. 2008; Aulbach et al. 2013). As garnet-bearing cratonic mantle has a low oxygen buffering capacity, it is likely to record fO2 related to the last metasomatic overprint prior to entrainment (Luth and Stachel 2014). The fluid-dominated imposition of fO2 on most cratonic mantle roots appears to be reflected by the demonstration that many cratonic mantle sections are more oxidized than expected for depleted or primitive mantle at similar depths, due to melt metasomatism (Foley 2011). Thus, interaction with oxidizing melts may be a plausible mechanism to explain the strong IPGE depletions observed for many cratonic mantle samples, especially because the sampled xenoliths are, by default, the wall-rocks to melt flow channels. Alloy destabilization during elevated fO2 conditions should produce coherent depletions in the IPGE, as observed for most cratonic and circum-cratonic mantle samples (Fig. 7A,B). This also implies that elevated HSE concentrations in carbonatites and kimberlites may be due to redox-driven assimilation of HSE in or near the mantle source region, and cannot be unambiguously ascribed to bulk digestion of entrained mantle fragments.
Given the likely exhaustion of sulfide during lithosphere formation, the very high S contents of cratonic mantle xenoliths compared to orogenic and off-cratonic peridotites at similar Al2O3 contents, is striking (Fig. 5). This observation invites consideration of the specifically cratonic mantle setting, where the presence of a thick lithospheric lid enables interaction with volatile-rich, silica-undersaturated melts and fluids leading to a cryptic overprint. Evolving oxidized melts that scavenged PGE alloys and sulfides at deeper levels could be efficient metasomatic agents, eventually depositing S and HSE after continued interaction with reducing lithosphere. Deposition of S and HSE may also occur at shallower levels when these melts exsolve volatiles while entraining their wall rocks as xenoliths. Metasomatic precipitation of BMS has been documented in peridotite xenoliths from the Kaapvaal craton that have interacted with low-volume melts prior to and perhaps related to kimberlite magmatism (Lorand and Grégoire 2006; Giuliani et al. 2013) and are manifest as part of the ilmenite–rutile–phlogopite–sulfide (IRPS) group of metasomatic rocks (see review of rock groups in Pearson et al. 2005). Enrichments in the more volatile chalcophile elements (S, Pd, Re, Os, Au) in mantle peridotites, with superchondritic S/Se, Pd/Ir, Pt/Ir and Os/Ir, subchondritic Pt/Pd and variable 187Os/188Os, were suggested to fingerprint this type of metasomatism (Alard et a. 2011; Lorand et al. 2013). A comparison of S with HSE concentrations in cratonic samples shows no clear co-variations, but some high Re, Pt and Pd concentrations appear to be tied to high S concentrations, especially for the very S-rich group of Kalahari xenoliths entrained in Group II kimberlites (Fig. 8), implying co-enrichment from the same metasomatic agent. High IPGE concentrations at high S contents may point to “resulfidation” of pre-existing PGE alloys, but they may also have been introduced from initially oxidizing, alloy-scavenging small-volume melts. If so, this implies that ancient, unradiogenic Os can be transferred in the cratonic lithosphere along with PGE alloy and deposited with S from the volatile-rich melt. While this may decouple Os isotope signatures from their host rocks, this process would represent a re-distribution of ancient ages within ancient cratonic mantle and would still yield meaningful minimum Re-depletion ages.
Despite elevated Re/Os (Fig. 9A), depleted (unradiogenic) Os isotopic signatures are documented in most cratonic peridotite samples, probably because the Os budget in cratonic mantle is controlled by a residual PGE alloy population or their “resulfidized” equivalents. This is true even for very S-rich samples from Group II kimberlites (Fig. 9B), but requires that the Re addition frequently observed in these samples be a recent event (e.g., Carlson et al. 2005). The notion of late-stage Re enrichment, if not due to contamination with kimberlite in measured whole rocks, is consistent with a frequently recognized metasomatic event that is precursory to kimberlite magmatism, based on trace-element zoning (Smith and Boyd 1992; O’Reilly and Griffin 2013).
In summary, it is plausible that interaction with unfractionated, oxidizing low-volume fluids that typically infiltrate intact cratonic roots, caused loss of part of the HSE budget. Conversely, reaction with evolved low-volume melts may lead to S addition accompanied by increases in Pt, Pd, and Re, but without significant disturbance of the Archean Os isotope signatures.
Effects of mafic (silicate) melt metasomatism
Interaction with melts of basaltic composition is only possible in regions underlain by relatively thin lithospheric roots because such melts typically cannot form beneath the thick (~ 200 km) lithospheric roots present beneath most cratons. Thinned lithosphere can support a long enough melting column to generate relatively large-degree mafic rather than small-volume volatile-rich melts even at moderate TP. This scenario applies to cratonic lithosphere where the deep root has been asthenospherized (e.g., the eastern North China craton), where xenoliths are typically entrained in alkali basalts rather than kimberlites. In addition to lithospheric thinning, unusually high TP and/or fertile sources can enable the generation of high melt volumes despite the presence of a deep lithospheric lid, as proposed for the 2.05 Ga Bushveld intrusion, which has strongly affected the cratonic mantle—and Re–Os isotope systematics—beneath parts of the Kalahari craton (e.g., Shirey et al. 2002). Most mafic melts penetrating the deep lithosphere represent relatively small melt fractions (≪ 20%) leaving a sulfide-saturated source, where the PGE plus Re will partition into the residual sulfide that is molten at the pressures and temperatures of basalt generation (Mungall and Brenan 2014). Thus, low concentrations of these elements are expected in the silicate melts, unless they contain physically entrained sulfide liquid (Bockrath et al. 2004b; Ballhaus et al. 2006), a process argued to be highly unlikely by Chung and Mungall (2009) on the grounds that capillary forces would prevent entrainment of immiscible liquids during melt extraction from partially molten peridotite.
Due to the chalcophile behavior of the HSE, the main effects of interaction with mafic melts may be related to the introduction or removal of dissolved sulfur (sulfide). Considering the stronger dependence of sulfur content at sulfide saturation (SCSS) on pressure than on temperature (Wendlandt 1982; Mavrogenes and O’Neill 1999; Holzheid and Grove 2002), two end-member depth-dependent effects on the interaction of such melts with the mantle can be envisaged. First, at high pressure, while the melt is sulfide-saturated, S (and other chalcogens) may be added from the melt to depleted peridotite, and pre-existing PGM may be “resulfidized” (e.g., Griffin et al. 2002, 2004; Lorand and Grégoire 2006). Second, during ascent, as the melt becomes sulfide-undersaturated with decreasing pressure and increasing SCSS, peridotitic sulfide might be dissolved into the melt and stripped from the mantle (e.g., Hart and Ravizza 1996).
However, sulfide can also saturate from a sulfate- or sulfide-bearing silicate melt during ascent if S solubility limits are exceeded due to cooling, fractional crystallization of metasomatic minerals from the percolating liquid (decreasing melt volume), reduction of sulfate-rich magma (Thakurta et al. 2008), or assimilation of pre-existing sulfide in the mantle wall-rock (e.g., Lorand and Alard 2001; Wittig et al. 2010b). Hence, a simple gradient of sulfide/PGE alloy persistence in the deep root and scavenging in the shallow root may not develop, especially if superposed on the effects of interaction with oxidizing silica-undersaturated melts as discussed above. Nevertheless, it is possible that interaction with sulfide-undersaturated silicate melts resulted in the strong IPGE depletions exhibited by basalt-borne cratonic peridotite (e.g., Cenozoic North China craton xenoliths) and off-craton peridotite xenoliths. In particular, percolation at high melt-rock ratios should lead to IPGE depletion, imposition of melt-like 187Os/188Os (Lorand et al. 2013; Figs. 4A–C, 8B) and re-fertilization in magmaphile major elements such as Al and Ca. This type of interaction means that such apparently fertile peridotites should not be used in the estimation of Primitive Mantle HSE abundances (e.g., Lorand et al. 2013; König et al. 2014). Indeed, co-saturation of a mafic melt in cpx and sulfide can explain the observed correlation of S or 187Os/188Os with Al2O3 or cpx mode in orogenic peridotites and some mantle xenolith suites that experienced metasomatism (Lorand 1989; Pearson et al. 2004; Wittig et al. 2010b).
In other off-cratonic suites, such systematics are interpreted as melt extraction trends that did not proceed beyond sulfide and cpx exhaustion (Ionov et al. 2006; Janney et al. 2010). A distinction between these two possibilities is best achieved when considering a highly incompatible element, such as TiO2, which shows different behavior during melt extraction vs refertilization (Niu 1997). However, xenoliths suites from most cratonic mantle sections show a distinct lack of Al2O3–187Os/188Os correlations (Pearson et al. 2004; Maier et al. 2012; Figs. 4, 9C), possibly because sulfide saturation or undersaturation is not coincident with pyroxene saturation. In addition, invasion of the bulk xenolith by late-stage S and Ti-rich kimberlite melt during xenolith eruption and low-T contamination during emplacement are superimposed complications (Frick 1973; Lorand and Grégoire 2006).
Effects of incongruent sulfide melting
The conductive geotherm of the continental mantle suggests that once metasomatic fluids infiltrate the lithosphere, HSE redistribution during sulfide addition and removal is governed by mss-sulfide melt partitioning, where sulfide partial melts concentrate Pd, Pt, and Au (in addition to Cu and, under some conditions, Ni), whereas mss partitions Ru, Rh, Ir, Re, and Os (Li et al. 1996; Barnes et al. 1997; Ballhaus et al. 2001; Bockrath et al. 2004b; Mungall et al. 2005; Brenan 2008; Fonseca et al. 2011). This translates to marked HSE fractionation, which is not achieved if sulfide is completely molten.
In a seminal study, Alard et al. (2000) showed that in cratonic and other peridotites, primary sulfides enclosed in silicate minerals have high IPGE concentrations and low Pd/Ir, whereas interstitial metasomatic sulfides have complementary signatures. Later work has demonstrated that often there is no link between textural position and PGE or Re–Os isotopes systematics (Alard et al. 2011; Bragagni et al. 2014; González-Jiménez et al. 2014; Wainwright et al. 2015). Furthermore, Lorand and Grégoire (2006) point out that most mss enclosed in refractory silicate grains have Ni/Fe ratios that are too low to be consistent with equilibration with olivine at magmatic temperatures. Thus, although (re)-introduced sulfides would intuitively occur primarily as interstitial grains, metasomatic sulfides may become occluded in silicate or oxide minerals during recrystallization and therefore, sulfide inclusions cannot be taken as prima-facie evidence of their “primary” residual nature.
The introduction of sulfide melt into peridotite by sulfide-saturated silicate melt should lead to enrichments in Pd, Pt, Cu ± Ni, because they would be stripped from the silicate melt. There are few data from cratonic peridotites, but a trend of Ni enrichment with a weak change to Os concentrations observed for interstitial sulfide in eclogite xenoliths compared to sulfide included in silicates or diamond (Aulbach et al. 2009b; Fig. 6) may attest to metasomatic addition of an IPGE-poor sulfide liquid. High Pt and Pd concentrations at low Pt/Pd are observed in massif peridotites. This signature could be consistent with pentlandite addition, a metasomatic sulfide mineral, which forms from sulfide melts and partitions Pd over Pt, showing strong negative Pt anomalies in chondrite-normalized PGE patterns (Lorand et al. 2013). Thus, addition of a sulfide liquid, as an independently moving melt or precipitating from sulfate-bearing silicates through redox-reactions, should lead to co-enrichment of Pd and Pt, maintaining low Pt/Pd, whereas metasomatism by fluids could be accompanied by enrichment in Pd without enrichment in Pt. Several cratonic samples have both high Pt and Pd concentrations at depleted Al2O3 suggestive of sulfide melt addition (Figs. 4D,E). However, for many samples, some degree of secondary Pd and Pt addition is probably hidden in their overall depleted PPGE systematics, leading to apparently lower melt extraction pressures when compared against modelled residues. These superposed effects may be unravelled when bulk-rock HSE systematics are used in concert with sulfide mineralogy and in situ analytical techniques, but such data are seldom reported.
Plume impingement and rifting
Some 40 km of the deep Kaapvaal craton root was strongly refertilized in the Cretaceous during interaction with asthenosphere-derived broadly basaltic to hydrous alkaline melts, perhaps related to impingement of a plume at the base of the lithosphere that ultimately also caused group I kimberlite magmatism (Boyd 1987; Smith and Boyd 1987; Griffin et al. 1996; Grégoire et al. 2003; Le Roex et al. 2003; Becker and Le Roex 2006; Kobussen et al. 2008; O’Reilly and Griffin 2010). Despite this, some sheared garnet peridotites, which sample the deep, hot and more fertile cratonic root, retain unradiogenic Os isotope compositions with Archean minimum ages, precluding wholesale loss of cratonic lithosphere (Walker et al. 1989; Pearson et al. 1995b). The original Os budget was possibly not erased because strong depletion even in the deeper surviving cratonic mantle (assuming that yet deeper, more fertile portions have been lost; Arndt et al. 2009) enables retention of IPGE in alloys and/or olivine, as also evidenced by other peridotite suites that were pervasively percolated by basaltic melts (e.g., Marchesi et al. 2010; Harvey et al. 2011). The retention of Archean and Paleoproterozoic ages in the deeper portions of the Kaapvaal craton, and perhaps also the surrounding mobile belts, are examples of incomplete thermochemical erosion of cratonic subcontinental lithospheric mantle that did not, however, result in observable rifting.
Some xenoliths from Obnazhennaya in the central Siberian craton, which have been entrained after the thermal climax of the Siberian plume, display radiogenic Os-isotope ratios (percent deviation of the sample from chondritic mantle at the time of formation = γOs > −1) combined with Pd depletions and high olivine Mg#, consistent with newly formed residues associated with plume impingement (Pernet-Fisher et al. 2015). Some samples retain Archean TRD, indicating that, as in the case of the Kaapvaal craton, plume impingement did not completely destroy ancient lithospheric signatures. The incomplete removal of cratonic mantle is consistent with dynamic modelling of the combined effects of metasomatism and plume impingement (Wang et al. 2015), and attests to the robustness of cratonic lithosphere even during extreme thermo-mechanical erosion.
Conversely, there are several examples of rift-related, thinned cratonic roots that have been strongly affected by melt infiltration followed by sampling as basalt-borne xenoliths. The mantle keel beneath the central Tanzanian craton appears to remain largely intact, but development of the East African Rift, combined with earlier Proterozoic collision, has led to disruption and thinning of the craton margins. Interaction with dry carbonatitic to basaltic melts caused strong enrichments in CaO and FeO, respectively, and oxidation compared to less overprinted samples (Rudnick et al. 1993, 1994; Ritsema et al. 1998; Lee and Rudnick 1999). Peridotite xenoliths derived from the lithosphere to a depth of ~140 km yield Archean TRD, whereas those derived from greater depth have Os that is more radiogenic, approaching convecting-mantle (plume) Os isotope compositions; modally metasomatized peridotites and replacive dunites experienced recent Re addition and introduction of radiogenic Os, respectively (Chesley et al. 1999; Burton et al. 2000). Basalt-borne xenoliths from Tok (Stavonoy-Aldan Block, SE margin of the Siberian Craton) also comprise a high abundance of Fe-rich peridotites and wehrlites, where alumina and 187Os/188Os correlate for the less-metasomatized samples that give an Archean model age, whereas strongly metasomatized xenoliths have more radiogenic Os, in cases associated with elevated Pd and Pt, although the cause of the lithosphere thinning is uncertain (Ionov et al. 2005a,b, 2006). Peridotite xenoliths from the Moroccan Atlas mountains, which may have been underlain by a cratonic root prior to rifting and orogenesis, show evidence for co-precipitation of cpx and sulfide from a melt that led to positive co-variations of Ir with Cu and of 187Os/188Os, Re, and Ir with Al2O3 (Wittig et al. 2010b). Magmatism beneath the Atlas has been ascribed to mantle plume flow, but, alternatively, this region may have been affected by craton-edge driven convection currents (Kaislaniemi and van Hunen 2014).
These examples demonstrate that although the Os isotope signature of lithospheric mantle in rift settings is sometimes impervious to melt percolation (e.g., Massif Central, Central European Volcanic Province; Harvey et al. 2010), the IPGE and Os isotopes are in other cases strongly modified, depending on the severity of the metasomatic overprint. The occasional persistence of strongly unradiogenic Os isotope signatures also in deeper rift-related cratonic xenoliths and/or constituent sulfides appears to indicate the survival of ancient lithosphere and argues against whole-sale loss of cratonic roots. Unfortunately, it is difficult to ascertain whether in any of these examples, xenoliths with mildly unradiogenic to primitive or radiogenic Os are a testament to the replacement of portions of the cratonic lithosphere by isotopically heterogeneous asthenospheric mantle, or whether they represent strongly overprinted, but still intact cratonic lithosphere (Pearson et al. 2004; Rudnick and Walker 2009; Janney et al. 2010; McCoy-West et al. 2013).
Modification during Craton Margin Processes—Subduction
Mantle sources modified by crustal recycling may have yielded boninite-like melts, precipitating mafic–ultramafic cumulates with associated chromitites and in some cases PGE mineralization, as early as the Paleo-Archean (Greenland: Bennett et al. 2002; Polat et al. 2002), Meso-Archean, (Pilbara: van Kranendonk et al. 2010; India: Khatun et al. 2014) and Neo-Archean (Superior: Polat and Kerrich 2001; North China: Polat et al. 2005). Though in some cases contested, the identification of mélange complexes and juxtaposition of contrasting lithotectonic elements in these, and other cratonic, regions have been taken as firm evidence for a transition to Phanerozoic-style plate tectonics (Shervais 2006). Thus, in contrast to the diamond inclusion record, which has been interpreted to reflect the initiation of subduction in the Neo-Archean (Shirey and Richardson 2011), the crustal record suggests that cratonic mantle may have been affected by plate interactions from the Paleo-Archean onwards, during its amalgamation from early continental nuclei and/or during later collisions, leading to larger cratonic landmasses (e.g., Bleeker 2003). Such continental collisions have been suggested to represent the main mode of cratonic mantle thickening (Jordan 1975; Tainton and Mckenzie, 1994; Canil 2004; Lee 2006; Pearson and Wittig 2008, 2014; Lee et al. 2011).
Radiometric dating of eclogite xenoliths or related inclusions in diamond has yielded ages that can be linked to regional tectonothermal events and accretionary processes during craton amalgamation or at the craton margins. This is true for the Kaapvaal, Slave and Siberian cratons, where the presence of radiogenic initial 187Os/188Os in some sample suites (γOs of +45 for Kaapvaal, +14 for Slave and +2.2 for Siberia) is consistent with ingrowth of 187Os in a high-Re/Os crustal precursor prior to eclogitization (Pearson et al. 1995a; Richardson et al. 2001, 2004; Shirey et al. 2001; Barth et al. 2002; Menzies et al. 2003; Aulbach et al. 2009a,b; Wiggers de Vries et al. 2013). Voluminous fluid emanating from subducting oceanic mantle undergoing serpentinite breakdown has been linked to 1.85 Ga eclogites from the Slave craton, and was proposed induce both eclogitization of the overlying oceanic crust and to generate diamonds entrapping sulfides with nearly homogeneous and radiogenic initial 187Os/188Os (Aulbach et al. 2009a). Higher median Os concentrations in ancient mantle eclogite xenoliths (0.3 ppb) compared to young orogenic eclogites (0.011 ppb; Table 1; Fig. 3) are consistent with formation of the former during higher degrees of partial melting, or may in part also be due to metasomatic effects. For example, the secondary deposition of Ni-rich sulfides in mantle eclogite (Aulbach et al. 2009b) is suggested by lower average Os concentrations of sulfides included in diamond (Table 1)
If the mantle wedge is an oxidizing environment, sulfide will be destabilized, giving rise to arc basalts with high Os and Re concentrations compared to melts originating from more reduced mantle due to the lithophile element behavior of Re at high fO2 (Fonseca et al. 2007; Righter et al. 2008). A corollary of this is that lithospheric mantle affected by accretionary processes may be relatively depleted in Re and Os, depending on whether and when mantle wedges became oxidized, as discussed again later. Indeed, lower Os contents coupled with radiogenic Os are observed for peridotite xenoliths from Phanerozoic arc mantle. These peridotites are characterized by high PPGE/IPGE, Pt/Pd and Ru/Ir signatures suggested to be due to interaction with oxidizing, Cl- and 187Os-enriched fluids that scavenged Os and some or all of the sulfide (Brandon et al. 1996, 1999; McInnes et al. 1999; Peslier et al. 2000a,b; Kepezhinskas et al. 2002; Widom et al. 2003; Saha et al. 2005). Such signatures are observed in some cratonic mantle samples (Fig. 10). Interactions between sulfate-rich aqueous subduction fluids and reducing cratonic mantle mineral assemblages might lead to deposition of S as sulfide, with the potential for large increases in S concentration accompanied by increased fO2 up to but not exceeding that of the sulfide-sulfate fence (Mungall 2002; Jugo et al. 2010). Thus, such fluids can also generate bulk rocks with elevated S/Se, as exhibited by cratonic xenoliths (Fig. 5), because of the differential solubility of these elements (Lorand et al. 2003a).
Sulfur concentrations in eclogite xenoliths from the Kaapvaal craton range from 50 to 2000 ppm and although sulfide has been suggested to be of metasomatic origin in the majority of samples (Group I), their average S concentrations are indistinguishable from those in unmetasomatized Group II eclogites (Gréau et al. 2013). The upper limit of S concentrations in eclogite xenoliths is similar to those in basalts and basaltic eclogites with oceanic crust precursors (Dale et al. 2009), whereas the low end of S concentrations may indicate that sulfide that was originally present has been lost, potentially during subduction, along with Se and Te hosted in the sulfides, which control the abundances of these elements in eclogites (Aulbach et al. 2012).
Assuming some form of plate tectonics operated at the time of craton formation and growth, subducting oceanic crust is expected to have more efficiently melted due to higher mantle potential temperatures indicated for the Archean; such melts have been linked to early continental crust formation (e.g., tonalite–trondhjemite–granodiorite complexes; Martin et al. 2005). However, because slab-derived melts are sulfide-saturated, they are not likely to be efficient carriers of a slab signature with regard to the HSE, which will be largely retained in the sulfide-bearing residue (Chesley et al. 2004; Jégo and Dasgupta 2014; Mungall and Brenan 2014). The interaction of siliceous slab-derived fluid with peridotite would lead to the formation of opx at the expense of olivine and the generation of lithologies ranging from opx-rich harzburgites and lherzolites to websterites to pyroxenites (Rudnick et al. 1994; Kelemen et al. 1998; Yaxley and Green 1998; Rapp et al. 2010). The expected effect on the HSE would be dilution and perhaps mildly increased 187Os/188Os. A peculiarity of some cratons (Kaapvaal and Siberia) is the opx-rich nature of the shallow lithosphere (Boyd 1989; Boyd et al. 1997), which has been ascribed by some to introduction siliceous, slab-derived fluids (Kesson and Ringwood 1989; Kelemen et al. 1998; Bell et al. 2005). A weak positive correlation between opx modes and 187Os/188Os for peridotite xenoliths from the Kaapvaal craton has been tentatively linked to introduction of a subduction fluid during 2.9 Ga craton amalgamation (Simon et al. 2007), although the expected dilution of HSE contents is not observed for opx-rich xenoliths from the Kaapvaal craton (data-set of Maier et al. 2012). The formation of pyroxenite xenoliths from the Slave and Superior cratons, with compositions intermediate between eclogites and peridotites and initial 187Os/188Os ranging from unradiogenic to radiogenic, has been ascribed to reaction of peridotites with slab-derived melts that could be linked to craton-margin processes (Aulbach et al. 2009b; Smit et al. 2014b).
Considering that eclogites and pyroxenites constitute high-Re/Os environments that will very rapidly evolve to radiogenic 187Os/188Os (e.g., Shirey and Walker 1998), it is clear that the reaction of these lithologies with cratonic mantle, either by mechanical mingling, melt transfer or isotopic equilibration aided by the percolation of fluids, could lead to mantle regions with elevated 187Os/188Os, disturbed Re–Os isotope systematics and increases in Pt and Pd. For ancient cratonic peridotites, the disentangling of the signatures of these subduction-related processes from those related to the infiltration of convecting mantle-derived decompression melts is challenging and requires precise Re–Os isotope ages that correlate with known regional collisional events.
Subduction-related hydration, lithosphere weakening and delamination
In recent years, increasing attention has been paid to the role of water in the destabilization of cratons and its link to subduction, either through direct fluid-transfer into the mantle wedge or dehydration from deeply subducted slabs that are stalled in the transition zone (Windley et al. 2010; Kusky et al. 2014; Pearson et al. 2014; Wang et al. 2014). The North China craton (NCC) may be the most prominent example of cratonic lithosphere destruction. The dramatic reduction in lithospheric thickness is evident from a comparison of mantle xenoliths entrained in Paleozoic kimberlites, which attest to the presence of a thick and cool diamondiferous cratonic mantle keel giving Archean to Paleoproterozoic TRD ages, with Cenozoic basalt-borne xenoliths sampling a considerably hotter, thinner mantle root that gives Proterozoic and younger TRD ages (Menzies et al. 1993; Griffin et al. 1998; Fan et al. 2000; Gao et al. 2002, 2009). There are competing hypotheses to account for this striking loss of cratonic lithosphere, including rheological weakening and destabilization by injection of water from multiple subduction zones (e.g., Tang et al. 2013). The deeper mantle beneath the central craton may have been replaced during continental collision at ~1.9 Ga (Gao et al. 2002). Particularly strong tectonic erosion in the Mesozoic is recorded in the eastern NCC, and a wealth of 187Os/188Os data now available suggests that the destruction may have been diachronous, occurring earlier in the east and later in the west (Liu et al. 2011). Relatively high Ru/Ir and Pd/Ir, similar to those of orogenic peridotites, are suggestive of the presence of newly accreted, asthenospheric mantle (Zheng et al. 2005; Liu et al. 2010), which was likely affected by melt percolation. This major disruption of the North China Craton means that the compiled Os isotope ratios and resulting TRD ages from this craton show much greater bias towards non-Archean ages than any other craton (Fig. 9B).
The cratonic mantle beneath the western USA, including the Mojavia and Wyoming cratons, which were affected by subduction of the Farallon plate beneath western North America, is unusually thin, fertile and hydrated (as indicated by abundant glimmerite xenoliths), suggesting removal of the deeper lithosphere due to metasomatism (Lee et al. 2001; Lee 2005; Li et al. 2008). The remaining shallow cratonic root persists despite strong refertilization and retains Os isotope evidence for NeoArchean to Paleoproterozoic formation (Lee et al. 2001), as is true also for the North China craton (Sun et al. 2012). Peridotite and websterite xenoliths from the thinned Wyoming craton root also retain some highly unradiogenic Os with Archean TRD, but most samples give younger, overprinted ages (Carlson and Irving 1994).
Thus, as is true for cratons affected by plume impingement leading to melt infiltration due various degrees of stretching (Kaapvaal) up to rifting (Tanzania), interaction with subducted slabs will lead to overprints on Re–Os isotope and HSE systematics that are indistinguishable from those off-cratonic mantle (Figs. 4, 9). The main difference appears to lie in Os isotope signatures that sometimes preserve a memory of Archean lithosphere formation in the disrupted cratonic roots, due to incomplete thermochemical erosion, but a loss of this signature in deep Archean roots that were affected by subduction-related hydration weakening, delamination and replacement by asthenospheric mantle. All these results are in keeping with recent dynamical modelling that indicates the necessity of metasomatism (probably hydrous) in weakening and removing the lower sections of lithospheric roots and yet still preserving some portion of an original ancient keel (Wang et al. 2015).
MODELLING OF PRIMARY VS. SECONDARY HSE SIGNATURES IN CRATONIC MANTLE
Here, a published model is used to constrain the distribution of Ru, Pd, Ir, and Pt between mantle restite phases and primary silicate melts, where HSE concentrations of residues are determined as a function of Pi (TP) and fO2 (Mungall and Brenan 2014), with the aims to (1) determine whether HSE systematics are useful indicators of the pressures and fO2 of melt extraction and hence melting environment, (2) determine at which level of depletion Re will be quantitatively removed from the mantle allowing determination of accurate TRD by using proxies less susceptible to overprint and (3) compare natural samples to modelled melting residues in order to distinguish between those that have possibly retained primary residual characteristics from those that have been affected by secondary processes. The Re and Os concentrations were modelled in accordance with the rationale outlined in Mungall and Brenan (2014), with further details given below. For this exercise to be useful, it is first necessary to review the constraints on the initial HSE concentrations, i.e., those of the Archean convecting mantle from which cratonic lithosphere formed.
HSE Concentrations of the Archean convecting mantle (ACM)
Taking modern mantle, as represented by orogenic peridotites, as the benchmark (median Ir = 3.5 ppb, Os = 3.8 ppb, Pt = 6.9 ppb, and Pd = 5.7 ppb), kimberlite-borne xenoliths have strikingly lower median concentrations of all PGE (Ir = 3.1 ppb, Os = 3.1 ppb, Pt = 2.8 ppb, and Pd = 1.2 ppb, respectively; Fig. 3 and Table 1), an observation that has been made before (Maier et al. 2012). Basalt-borne cratonic xenoliths appear to be affected by strong, secondary IPGE depletions, similar to younger off-cratonic xenoliths, which therefore are unlikely to relate to original mantle PGE concentrations, as discussed earlier. The large variability of HSE contents in usually small samples of cratonic peridotite xenoliths may in part reflect control of IPGE by HSE-rich nanometer size nuggets (e.g., Luguet et al., 2007; Lorand et al. 2013) due to melt extraction well beyond sulfide stability. Effects due to inhomogeneous distribution of PGM or to incomplete digestion may be excluded at least for those cratonic mantle xenoliths where low PGE concentrations were reproduced by multiple dissolutions (e.g., Irvine et al. 2003), although the nugget effect can occur at various scales and need not be detected from multiple digestions from the same aliquot powder. At any rate, the median concentrations calculated for kimberlite-borne cratonic peridotites should be robust against such outliers.
Barring artefacts related to the nugget effect, the depletion of cratonic mantle relative to modern convecting mantle may have several causes: (1) It is possible that very low IPGE concentrations in some kimberlite-borne cratonic peridotites relate to secondary, metasomatic effects; (2) The high HSE concentrations in orogenic peridotites are in part caused by secondary refertilization; (3) It may be the signature of a reservoir that was depleted by core formation and was subsequently incompletely mixed with incoming accreted material (Maier et al. 2009, 2012). If we are to successfully model primary residual HSE concentrations in the newly-formed lithosphere and to confidently distinguish them from secondary effects, initial HSE concentrations (C0) must be better constrained, the rationale for which is laid out in the following.
Knowing the initial concentrations—C0—of HSE of the undifferentiated “Primitive Upper Mantle (PUM)” is one of the most sought after and yet difficult to define goals of mantle geochemistry. Much of the problem stems from the fact that almost all samples of mantle that are available have been subject to various types of secondary modification. Notwithstanding this, all parts of the upper mantle have undergone partial melting episodes at some stage during Earth history (e.g., Allègre et al. 1995), so estimating a “primitive” composition necessitates selecting a suite of rocks from which a minimally melted “original” composition can be estimated.
A currently widely used model is based on the Re–Os isotope systematics and HSE concentrations of variably depleted xenoliths and massif peridotites, with suggested concentrations of 3.5 ppb Ir, 3.9 ppb Os, 7.6 ppb Pt, and 7.1 ppb Pd (Table 2) (Meisel et al. 1996; Becker et al. 2006; Fischer-Gödde et al. 2011; Wang et al. 2013). In recent years, it has been cautioned that this PUM model introduces a bias in that orogenic peridotites, from which these estimates are mostly drawn, are variably depleted, strongly melt-metasomatized and refertilized rocks with harzburgite protoliths (Le Roux et al. 2007; Lorand et al. 2009, 2010, 2013; Alard et al. 2011). Similarly, it has been suggested that near chondritic chalcogen ratios in orogenic and xenolithic peridotites reflect ubiquitous refertilization of variably depleted residual mantle containing mss ± PGE alloys plus metasomatic sulfides and tellurides (König et al. 2014, 2015). The effects of mafic melt infiltration are evident in the presence of texturally late spinel and pyroxene clusters, metasomatic pentlandite and ubiquitous pyroxenite dikes, and in linear correlations for incompatible elements with vastly different distribution coefficients, which is inconsistent with simple melt extraction (Lorand et al. 2013).
The refertilized nature of the samples on which PUM is based not only makes them unsuitable for modelling the concentrations of partial melts and their complementary residues, but also raises doubts over suggestions that their non-chondritic relative abundances relate to late accretion of an unknown meteorite component (Pattou et al. 1996; Becker et al. 2006; Fischer-Gödde et al. 2011). In addition, it is problematic that the PUM estimate results in markedly higher Pd/Ir and Ru/Ir than those of various chondrites classes (Horan et al. 2003), which are considered building blocks of Earth and components of the Late Veneer (e.g., Day et al. 2012, 2016, this volume). The non-chondritic ratios in orogenic peridotites, many of which appear to sample supra-subduction zone mantle, may be the result of metasomatic addition of Cu–Ni-rich sulfide melt that later crystallises pentlandite (Alard et al. 2000; Lorand and Grégoire 2006) and chromite accumulation, given the affinity of Ru for Cr-rich spinel (Righter et al. 2004; Brenan et al. 2012).
Nevertheless, because orogenic lherzolites are by far the “best-behaved” in terms of producing coherent trends of PGE concentrations versus melting indices such as Al2O3 (Reisberg and Lorand 1995; Pearson et al. 2004; Fig. 4), they might be considered the most suitable samples from which to extract convecting mantle compositions. We derived an estimate after careful examination and exclusion (“filtering”) of samples showing evidence for secondary effects related to refertilization discussed above, to incomplete digestion/recovery, to analytical artefacts or having statistically anomalous abundances. A median was taken from the resulting subset of samples that have Al2O3 contents in the interval between 3.5% and 4.06%, i.e., up to the value suggested for PUM (Jagoutz et al. 1979; Hart and Zindler 1986). The results from this exercise (Table 2; Os = 3.6 ppb, Ir = 3.6, ppb, Ru = 7.2 ppb, Rh = 1.1 ppb, Pt = 7.4 ppb, Pd = 6.4 ppb) show that even in a judiciously filtered data-set, Pd and Ru appear to be implausibly high relative to a chondritic starting material and may be compromised by melt refertilization.
Sub-PUM PGE concentrations (Ir = 2.9 ppb, Os = 3.0 ppb, Pt = 6.2 ppb, and Pd = 5.2 ppb) have been presented for the depleted mantle (DM), based on average concentrations in ophiolites, abyssal peridotites and orogenic peridotites (Salters and Stracke 2004), which likely also include samples affected by melt percolation. If the DM composition reflects extraction of 2–3% partial melt from the primitive mantle (Workman and Hart 2005), then the effect on the PGE and Re should be negligible because the presence of residual sulfide at such low degrees of melt depletion would effectively prevent depletion of all HSE except Au (e.g., Keays, 1995; Mungall and Brenan 2014). Despite this, based on relatively constant Re/Yb in MORB, a Re concentration of 0.157 ppb is provided by Salters and Stracke (2004), which generates subchondritic and sub-PUM 187Os/188Os for the present-day DM.
Can C0 be better constrained, at least for the Archean mantle? Assuming that the IPGE concentrations in kimberlite-borne cratonic xenoliths are least affected by secondary enrichment (as evidenced by the retention of depleted Os isotope signatures), and since the concentrations of compatible elements weakly increase with increasing degree of depletion, a maximum concentration for the HSE can be estimated using the median Ir of 3.1 ppb for this peridotite class (Table 1). The concentrations of the other HSE are calculated assuming chondritic relative abundances, resulting in Os = 3.3 ppb, Pd = 3.7 ppb and Pt = 6.9 ppb (Table 2). Iridium was chosen because it appears to be less affected by secondary processes than Os, as evidenced by much lower variance of Ir in the kimberlite-borne peridotite dataset. Modelling presented below shows that Ir remains compatible with combinations of alloy and olivine in the restite throughout the melting interval responsible for generating cratonic mantle. Uncertainties remain regarding the nature of the late accreted material, as different chondrite classes show significant differences in HSE abundances (Horan et al. 2003; Day et al. 2016, this volume). Nonetheless, we feel that this initial estimate of the ACM starting composition may be the one that is the least compromised by metasomatism and secondary alteration. For the modelling that follows, these new values will be combined with an Al2O3 concentration of 4 wt% in keeping with suggested values of 3.97 wt% (Jagoutz et al. 1979) and 4.06 wt% (Hart and Zindler 1986), which are based on the least depleted specimens in a suite of ultramafic xenoliths and comparison of chondritic and mantle peridotitic fractionation trends.
In order to use the HSE to constrain conditions of formation and modification of the cratonic lithosphere, not only must the HSE composition of the Archean mantle source be well constrained, but HSE fractionation(s) related to partial melting must be defined and HSE carriers in such restitic peridotites must be correctly and unambiguously identified. We address the behavior of HSE during partial melting by considering their distribution between sulfide melt, silicate melt, and solid restite phases including silicates and alloys using equilibrium partition coefficients. This approach is complemented by a consideration of alloy solubility, which takes into account that PGE are major elements in alloys, for which the use of a limiting metal solubility is more appropriate than a simple Nernst partition coefficient (Mungall and Brenan 2014). The model assumes a dry, primitive mantle (PM) source with initial HSE concentrations for “primitive” ACM discussed above. This ignores the possibility of mantle source heterogeneities, such as the presence of pyroxenites or early-depleted mantle portions, which have been inferred from the trace-element and isotopic composition of komatiites and cratonic mantle (e.g., Sobolev et al. 2007; Arndt et al. 2008; Aulbach et al. 2011; Fiorentini et al. 2011). We note, however, that the extent and abundance of such heterogeneity in the Archean mantle should be considerably smaller than that evident in today’s mantle following 3 Ga of subduction processing.
The behavior of the HSE during partial melting depends on a range of parameters including bulk composition, oxygen and sulfur fugacity, as well as pressure and temperature (Mungall and Brenan 2014, and references therein). Assuming the shallow portion of the cratonic mantle to be the result of picritic melt extraction at hot ridges (TP 1500–1600 ºC according to the TP evolution of Korenaga 2008), with average melt fractions (F) of ~0.3 and initial and final melting pressures of 3–5 GPa and 1–2 GPa, respectively, a mantle lithosphere of 85–135 km is formed, in addition to the crust (Herzberg et al. 2010). Plume subcretion beneath a pre-existing shallow cratonic nucleus or melting to shallow pressures is assumed to occur at excess TP of 1600–1900 ºC and pressures of up to 10 GPa, whereby the plume would exhibit an internal temperature structure with a hot conduit and cooler periphery (Herzberg and O’Hara 1998; Herzberg et al. 2007, 2010). Komatiites representing ca. 30–50% partial melts (Al-undepleted komatiites) would only be extracted from the shallower portion of the melting column that experienced large melting intervals, leaving a residue of harzburgite to dunite at 100–150 km depth. Basalts and picrites would be extracted either at lower TP (ambient or cool plume), from the cooler periphery of buoyantly upwelling mantle or from the deeper portions of the melting column, leaving a residue that changes from lherzolite to harzburgite to dunite as F increases (Herzberg et al. 2010).
Given the above, it is useful to model HSE concentrations for a range of Pi and corresponding TP. High degrees of melting applicable to cratonic lithosphere formation are not accessible to the lower-pressure melting paths (1–2 GPa) because decompressing mantle reaches the surface after F < 0.2. We will limit our discussion to P ≥ 3 GPa, corresponding to a range from ambient to excess TP according to the model of Korenaga (2008), but to excess TP according to the model of Davies (2009). Pressure during adiabatic decompression is allowed to vary as a function of F, using the parameterization of Iwamori et al. (1995). The solidus temperature for dry melting and the adiabatic melting path of 50 °C/GPa were parameterized from Herzberg and O’Hara (2002). At these conditions, the decompressing mantle intersects the surface at F = 0.6 for Pi = 7 GPa, F = 0.46 for Pi = 5 GPa, and F = 0.27 for Pi = 3 GPa, although we note that the final pressures would be dictated by the thickness of the pre-existing lithosphere and the exhaustion of fusible components, which may limit melt production after melting intervals corresponding to 3–4 GPa (Herzberg et al. 2010).
Due to the siderophile and chalcophile character of the HSE, sulfides, PGE alloys and iron-nickel metals—if present—play dominant roles in determining the bulk distribution coefficient during partial melting. As described before, the sulfur solubility in the extracted silicate melt at sulfide saturation (sulfur concentration at sulfide saturation: SCSS) depends predominantly on pressure and FeO concentration in the silicate melt, and less on temperature (Wendlandt 1982; Mavrogenes and O’Neill 1999; Holzheid and Grove 2002; Liu et al. 2007). Given the dependence on FeO, there is an indirect control of fO2 on SCSS in that it determines the ratio of FeO to Fe2O3. Thus, a term for Fe2O3 has been added to the calculation of SCSS in addition to constants parameterized from Wendlandt (1982) and Mavrogenes and O’Neill (1999). Finally, from the P–T dependence, it follows that the choice of TP and corresponding pressure where the solidus is crossed (Pi) has implications for the S (and Se, Te?) capacity of partial melts extracted (e.g., Mavrogenes and O’Neill 1999).
Partition coefficients and alloy solubilities as a function of fO2, fS2, and Pi were parameterized from published experimental partition coefficients, following the approach outlined in Mungall and Brenan (2014). At the P–T conditions considered here, a parameterization of results in Rhyzenko and Kennedy (1973) and Hart and Gaetani (2006) shows that the sulfide phase is at all times well above its liquidus, which, for a starting composition of 250 ppm S (McDonough and Sun 1995), has a mass fraction ~0.07 wt% (assuming that all S resides in a sulfide with ~36 wt% S). As long as the silicate melt is saturated in sulfide, HSE partitioning into sulfide liquid and residual silicate minerals control residual concentrations. The distribution coefficients for sulfide melt-silicate melt were parameterized as a function of fO2 and fS2 and lie between 3.0 × 105 and 1.9 × 106 for the PGE. DResulfide/silicate varies strongly as a function of fO2 and fS2 (Brenan 2008), ranging from 102 to 104 over the conditions addressed here. As a result, the PGE are little fractionated and effectively retained as long as sulfide melt is stable and Re remains chalcophile and little fractionated from Os under reducing conditions applicable to oceanic mantle melting (Fonseca et al. 2007; Brenan 2008; Mungall and Brenan 2014). After sulfide has been consumed by the melting reactions, bulk Ds reflect the relative compatibilities in silicate minerals as summarized by Mungall and Brenan (2014) and correspond well to those determined indirectly from peridotites having experienced low degrees of melt extraction (≤ 25%), i.e., DOs ~ DIr ~ DRu > DPt > DPd ≥ DS ~ DRe (Pearson et al. 2004; Wang et al. 2013).
Alloy solubility in silicate melts is strongly dependent on fO2 and fS2 (Laurenz et al. 2013, 2014; Mungall and Brenan 2014; this work). Sulfur fugacity in the model is estimated using free energy data from Barin (1995) and Holzheid et al. (1997) to find an equilibrium constant for the mass action expression FeOsilicate melt + 0.5 S2 = FeSsulfide melt + 0.5 O2 (see also Brenan 2008). The pressure-dependence for this reaction was estimated using a ΔrV obtained by comparing the partial molar volume of FeO (Lange and Carmichael 1987) to that of γFeS (Balabin and Urusov 1995), adjusted for ΔmeltingV of γFeS (Moretti and Baker 2008). After exhaustion of sulfide liquid from the restite, fS2 is scaled linearly with S concentration in the silicate melt from its value at sulfide saturation (e.g., Mungall and Brenan 2014). Modern convecting mantle has a log fO2 relative the fayalite–magnetite–quartz oxygen buffer (ΔFMQ) of approximately −2 to 0 at low pressure (Frost and McCammon 2008). There is some uncertainty regarding convecting mantle fO2 in the Archean. It appears to have been constant since ca 3.5 Ga, i.e., since the stabilization of cratonic lithospheric mantle, based on investigations of continental mantle rocks (Canil 1997, 2002; Berry et al. 2008), but more reducing conditions prior to the oxygenation of the oceans are advocated by some authors (Kasting 1993; Foley 2011; Aulbach and Viljoen in press). Since abyssal peridotites form via melting at relatively low pressures (1–2 GPa) and since the mantle becomes more reducing with increasing pressure, at constant concentration (Frost and McCammon 2008), higher Pi during Archean melt generation in the convecting mantle implies correspondingly lower ΔFMQ (~ −4 at 200 km depth).
Though the evidence is controversial (e.g., Lee et al. 2010), the modern arc mantle may be more oxidizing than the convecting mantle due to the transfer of oxidized material from the slab during subduction (Wood et al. 1990; Mungall 2002), with arc xenoliths recording mean ΔFMQ of +0.5 (Foley 2011). If transfer of oxidized material does occur, H2O-fluxed melting in the mantle wedge is expected to entail significantly different HSE behavior compared to decompression melting of convecting mantle (e.g., Brenan et al. 2003), which may be relevant to scenarios of craton formation and/or thickening in Archean subduction zones. In the absence of oxygenated seawater in the Archean, the flux of oxidized material from the wedge was probably small and the transfer of water alone is insufficient to increase mantle fO2 at pressures < 5 GPa (Frost and McCammon 2008), resulting in arc mantle that was likely more reducing than today (Foley 2011). We therefore confine our discussion to situations in which sulfur remains in its reduced state (c.f., Jugo et al. 2010) and do not consider the role of sulfate in the comportment of S.
The total solubility of each HSE in the silicate melt is calculated as the sum of its solubility as metal oxide and as metal sulfide as a function of fS2 and fO2. Metals are apportioned to the residual solid silicates according to mineral modal abundances and distribution coefficients, which were parameterized as a function of fO2 (Mungall and Brenan 2014), with the remainder assumed to reside in PGE alloy phases after saturation. Palladium does not form alloys and is assumed to be perfectly incompatible in olivine. As a monovalent element with a similar ionic radius, Pd may partition into opx with a D similar to Na, taken from Bédard (2007), although its slightly smaller cation size may imply slightly larger D (Dygert et al. 2014).
The model approximates columnar fractional melting by step-wise extraction of 1.1% partial melt, of which 1% is retained in and combined with the residue as the “starting composition” for the next melting increment (Mungall and Brenan 2014). The effect of the modelled melt retention is to retard the exhaustion of incompatible elements, such as S, Pd and Re, and of Pt. Concentrations of major elements and silicate mineral modes in the residue were parameterized from isobaric batch melting experiments of Walter (1998) as a function of F in order to approximate how much pyroxene ± garnet vs olivine remain in the restite, which control the partitioning of HSE between solid and melt. For example, for melting beginning at 70 kbar, the initial mode is 0.453 olivine, 0.323 pyroxenes (opx + cpx combined) and 0.223 garnet, for that beginning at 30 kbar, the initial mode is 0.537 olivine and 0.463 pyroxenes (no garnet). These modes are not very sensitive to the nature of melt extraction (fractional vs. batch) or to changes in the melting reaction.
A comparison not only of the modelled residues to cratonic mantle, but also of the modelled melts to picrites and komatiites (Mungall and Brenan 2014), provides an additional control of the quality of the models and assumptions—in particular with regard to the starting composition (ACM)—as poor choices can produce reasonable matches for the restite while magnifying the mismatch from the melts. Although a discussion of the modelled versus natural melts is beyond the scope of this work, such checks were carried out as part of the quality control process.
On a final note, cratonic mantle formation at higher Archean TP implies deeper onset of partial melting, including at depths where metal saturation in the mantle is potentially reached. This would lead to stabilization of solid Fe–Ni metal that partitions and fractionates the HSE and is progressively diluted by less noble siderophiles, such as Ni and Co (Ballhaus 1995; Woodland and Koch 2003; Rohrbach et al. 2007; Rohrbach and Schmidt 2011; Stagno et al. 2013). Exploration of the effect of iron-nickel metal in the modelling presented here is precluded by the lack of appropriate partitioning and solubility data.
Effects of Partial Melt Extraction on HSE based on Modelling
We describe here the modelled HSE relationships as a function of F and Pi, and then compare natural samples against these modelled residues. The data and modelling form a combined perspective that, though not providing unique constraints, help distinguish primary depletion-related signatures from those that must have been inherited during subsequent processes, provided the model is an accurate reflection of nature. In the model results, Al2O3 is extracted increasingly efficiently with decreasing Pi (<1 wt% Al2O3 at F = 0.20 and 3 GPa, or at F = 0.31 at 7 GPa). The change in residual S concentration is initially small but then asymptotically approaches zero when sulfide is practically exhausted, concurrent with PGE alloy stabilization. Due to increased S solubility, this occurs at lower melt fractions for more oxidizing conditions (F = 0.20 for ΔFMQ = −2, F = 0.16 at ΔFMQ = 0). In contrast, the pressure-sensitivity is small because the increased SCSS at higher TP and FeO in the melt are balanced by decreased SCSS at higher Pi. Trends of modelled HSE concentrations as a function of melt fraction (F) are shown in Figure 11A–F.
The IPGE are compatible in sulfide liquid, and both experiments and natural samples indicate that extraction of picrite and komatiite melts at high pressure lead to saturation in Ru-Os–Ir alloys at some time around sulfide exhaustion, and that neither Rh or Pd are concentrated in alloy minerals (Barnes and Fiorentini 2008; Fonseca et al. 2012; Mungall and Brenan 2014). The concentrations of Ru, Ir, and Os increase continuously throughout the melting interval, but mostly at a slower rate after alloy saturation, in particular under higher fO2 (Fig. 11A,B,F). The highest concentrations are, of course, reached for the longest melting columns (highest melt fraction F), corresponding to the highest Pi. After apportioning the appropriate amount of HSE to olivine, the calculated weight fraction of alloy in the residue is minute and in the ppb range, and yet in some instances it dominates the HSE budget. Because of the combination of very low alloy abundances with alloy–silicate liquidDIPGE on the order of 106, the choice of alloy–silicate liquidDIPGE has only a weak effect on bulk D if conventional modelling approaches are used.
At S concentrations of ~20–30 ppm in the residual mantle, maximum concentrations are reached for Pt and Pd, which are ~20% higher than starting concentrations for ΔFMQ = −2 and ~ 15% higher for ΔFMQ = 0, after which the concentrations of both start to decrease relatively sharply over small melting intervals as they are liberated into the silicate melt from a rapidly diminishing amount of sulfide. Platinum concentrations show a notable exception at ΔFMQ = −2 and 3 GPa, as they decrease much more slowly than at higher fO2 or pressures, owing to saturation of the silicate melt in Pt alloy. As outlined above, Pd does not saturate under any of the modelled conditions and becomes strongly incompatible after sulfide is exhausted. Rhenium also does not form an alloy and its concentration increases by at most a few percent while sulfide is stable. After sulfide exhaustion, Re concentration decreases quickly at ΔFMQ = 0 and more slowly at lower fO2, showing the greatest relative dependence on fO2 of all HSE considered here. At low Pi (3 GPa), Re is retained in the residue more efficiently than at higher pressures (Fig. 11E).
Effects of Partial Melt Extraction on HSE—Comparison of Model and Nature
The effects of partial melting on modelled HSE concentrations are summarized in chondrite-normalized diagrams (Fig. 11G), with the median concentrations of various sample classes from Figure 3 shown for comparison. The modelled residues are useful in highlighting processes that are not consistent with decompression melting of dry primitive mantle under the chosen pressure (7–3 GPa) and redox conditions (ΔFMQ = −2 to 0).
Kimberlite-borne cratonic peridotites have highly variable IPGE concentrations at a given Al2O3 (Figs. 4A–C). In contrast, over the range of fO2 considered, the concentrations in the modelled residues vary by less than a factor 2 and show little sensitivity to pressure of melt extraction, as they are effectively retained first by sulfide and then by PGE alloys together with olivine, in which they are compatible or only mildly incompatible, depending on fO2. Although the modelled range of ΔFMQ has been chosen to be small (−2 to 0) and consistent with conditions expected in convecting mantle during MORB extraction at ~ 3 GPa (Frost and McCammon 2008), the results show that at increasing fO2, alloy solubility is enhanced. For example, at ΔFMQ of +2, which is reached by the most oxidized of modern arc lavas, but not by arc xenoliths (Frost and McCammon 2008; Foley 2011), Ir and Ru become mildly incompatible. Thus, strongly depleted IPGE concentrations—by failure of IPGE alloys to saturate or by secondary dissolution—could only be achieved at very high fO2 and either flux melting or metasomatism by oxidized small-volume melts, such as carbonatites and kimberlites, at high melt-rock ratios (Frost and McCammon 2008; Foley 2011). Alternatively, IPGE alloys may be physically entrained, leading to heterogeneity (Lorand et al. 2013).
Samples from the part of the North American craton sampled in Greenland (G-NAC; Wittig et al. 2010a) show IPGE depletions relative to the modelled residues at high F (low Al2O3; Figs. 4A–C) that mimic the behavior of incompatible HSE, such as Pd and Re (Figs. 4D,F). Barring nugget effects, this begs the question whether these depletions were generated during melt extraction under high-fO2 conditions with significantly enhanced IPGE solubilities. Accretionary processes may have been involved in the NeoArchean amalgamation of the G-NAC (Wittig et al. 2008, 2010a; Tappe et al. 2011), and, if Archean (or modern) subduction zone mantle did form under higher fO2 than convecting mantle, variable flushing with putative oxidizing melts and fluids might explain some of the variability of Os concentrations at the low-Al2O3 end.
In the partial melting model, Os and Ir are mildly fractionated at high melt fractions where Os concentrations increase more slowly than Ir concentrations with increasing F (Fig. 11). The effect of increasing fO2 is to increase PGE solubility in silicate melts, leading to lower concentrations in the residue, but it will not sufficiently fractionate Os from Ir to account for the observed deviation from reasonable starting compositions because of the similar solubilities. Thus, low Os concentrations over a range of Al2O3, such as those observed for other cratonic mantle suites (Fig. 4A), were possibly imposed during secondary processes, such as metasomatic interaction with high-fO2 melts or fluids that would lead to destabilization of (resulfidized) PGE alloys and loss of HSE. The particularly pervasive Os depletions shown by both on- and off-craton basalt-borne xenoliths (Fig. 4A; median concentrations of 2.2 and 1.9 ppb, respectively) relative to kimberlite-borne cratonic and circum-cratonic xenoliths (median concentrations of 3.1 and 3.3 ppb, respectively; Table 1) suggest a relationship to the nature of the host magma and may be related to Os volatility during degassing of the slowly cooling host basalts, with perhaps greater susceptibility to loss occurring when Os is still hosted in sulfides. In contrast, kimberlites, as fast-rising, volatile-rich magmas, are emplaced during explosive volcanism, with little time for devolatilization of mantle-derived xenoliths (e.g., Sparks et al. 2006).
If Ru and Ir behave congruently during partial melting, the observed suprachondritic Ru/Ir in various peridotitic materials could be taken to indicate that the mantle’s HSE systematics do not derive from carbonaceous chondrite, but from a different or unknown meteoritic component (Rehkämper et al. 1997; Snow and Schmidt 1998; Becker et al. 2006; Liu et al. 2009). Alternatively, they could reflect earlier Ru alloy saturation compared to Ir, secondary processes, such as chromite accumulation, or analytical artefacts related to Cr-based polyatomic interferences, as mentioned above.
Platinum and palladium
Once sulfide is exhausted, the concentrations of elements that do not form alloys—Pt under oxidizing conditions or at high P, Pd and Re under any conditions)—are dictated by their low compatibilities in the residual silicate mineral assemblage and their variable sensitivity to fO2. They are efficiently depleted at large melt fractions or high pressures of melt extraction or a combination of both (Figs. 4D–F). One of the most striking differences between cratonic/circum-cratonic and off-cratonic or orogenic peridotites is the strongly Pt- and Pd-depleted character of the former, with median concentrations of 2.8 and 1.2 ppb, respectively in cratonic peridotites (Table 1). The drastic drop in Pt and in Pt/Ir over small ranges of Al2O3 (1–2 wt%; Figs. 4D and 10B) in cratonic peridotites—with variations spanning more than two orders of magnitude at a given Al2O3 concentration—contrasts with the near-constant or slightly decreasing Pt concentrations in orogenic peridotites over the same Al2O3 interval.
It has been suggested that high-temperature PGE alloys are Pt-poor and that during high-pressure melting, Os–Ir–Ru alloy is stabilized, whereas Pt alloy is too soluble in the melt to be stable (Brenan and Andrews 2001; Mungall and Brenan 2014), allowing Pt to behave as an incompatible element. However, at lower pressures and under reducing conditions (ΔFMQ = −2; Fig. 11C), Pt may saturate at a much lower concentration in silicate melt than at higher Pi or fO2. In theory, therefore, Pt can serve as a pressure sensor if the sensitivity of Pt alloy solubility to fO2 and fS2 is sufficiently tightly constrained. For example, the modelled Pt concentrations at 5 and 7 GPa drop to < 0.1 ppb much faster than Pd concentrations, implying DPt < DPd due to incorporation of Pd in pyroxene at high pressure. However, from a consideration of natural high-degree melts at low pressure (picrites, komatiites) and of residual mantle, a stronger incompatibility of Pd is evident (e.g., Barnes et al. 1985; Puchtel and Humayun 2000; Pearson et al. 2004; Fischer-Gödde et al. 2012; Lorand et al. 2013). Hence the stronger depletion of cratonic peridotite samples in Pd than in Pt is consistent with the relative partitioning inferred from partial melts.
Because Pd does not partition into alloys, there are no uncertainties related to alloy solubility, and the concentration of this element is controlled by its incompatibility in the silicate residue and fO2 once sulfide is exhausted. Modelled melt residues are sensitive to Pi (Fig. 4E), with stronger depletions at a given Al2O3 content with increasing pressure. The FeO content of the melt has recently been shown to be an important parameter, as the solubility of Pd increases with increasing FeO in the melt (Laurenz et al. 2010). Because the FeO content of the melt increases with increasing pressure (e.g., Kinzler and Grove 1992), deep Pi enables strong Pd depletion of the residue. Indeed, Pd concentrations and Pd/Ir in both cratonic and orogenic peridotites generally start to decrease rapidly at about 1.2% Al2O3 (Figs. 4E, 10C).
Significant scatter in Pd concentrations and Pd/Ir is evident both for kimberlite- and basalt-borne peridotite xenoliths. This may be related to gains due to metasomatic sulfide or melt addition. In the Kaapvaal craton, the conspicuously S-enriched mantle xenoliths entrained in the older Group II kimberlites have markedly higher median Pd concentrations (1.8 ppb) compared to those from Group I kimberlites (0.53 ppb), suggestive of co-enrichment of S and Pd, and implying that their Pi of 3–5 GPa that could be inferred from Figure 4E is underestimated. Distinguishing between Pi of 5 and 7 GPa is not possible at the low Al2O3 end of the compositional array, but a large number of samples from Kalahari and G-NAC appear to have been depleted at Pi ≥ 5 GPa. In contrast, samples from the Rae craton form a comparatively tight array consistent with Pi of 3–5 GPa, which probably also applies to the Karelian craton, whereas the few samples from China and the Slave craton apparently require Pi > 7 GPa. As mentioned above, it is possible that the incompatibility of Pd in the residue is underestimated and new experiments to constrain this better are needed. Hence, although Pd shows promise as a pressure sensor due to its Pi-sensitivity, which is not complicated by alloy saturation and solubility, it is not currently possible to confidently distinguish between primary and secondary Pd–Al2O3 relationships and to use them as a measure of Pi and, by implication, as an indicator of the tectonic setting where melt extraction occurred (ambient or excess TP).
The Pt/Pd ratio in natural cratonic peridotites does not co-vary with Pt concentrations, but shows a broad negative co-variation with Pd concentration (Fig. 12A.B), falling between modelled melt depletion trends at Pi of 3 and 5 GPa, seemingly indicating control by Pd depletion while Pt alloy is stable. On the other hand, there is a complete lack of the expected covariation of Pt/Pd with Al2O3 (a melt depletion index), and the large scatter of Pt/Pd at a given Al2O3 together with high Pt/Pd spread over a range of Al2O3 may indicate the influence of secondary processes (Fig. 12C). Alternatively, the scatter may result from a nugget effect related to the tendency for Pt and Pd to be hosted by different phases after sulfide has been removed (e.g., Pt alloy, Pd in silicates?) with consequent loss of the correlated behavior observed when both are hosted by the same phase prior to sulfide exhaustion. In the basalt-borne xenoliths, higher than initial Pt/Pd are observed for an even larger range of Al2O3 (Fig. 12C). It is possible that Pt alloy solubility in the model is overestimated, especially given that the fS2 model, on which Pt solubility depends, is currently constrained by an extremely limited experimental database. It is also possible that the assumption of moderate Pd incorporation into pyroxene is not true. Thus, Pt values and Pt/Pd rations in residues cannot yet be used to infer the conditions of melting with confidence, but they are consistent with controls by fO2 and P. Given current uncertainty, the most that can be said is that the appearance of compatible behavior by Pt is highly suggestive of low-pressure melting.
Re/Os and 187Os/188Os
Kimberlite-borne cratonic mantle xenoliths are conspicuously enriched in Re at low Al2O3 contents (Fig. 4F) compared to orogenic peridotites. This is a well-known feature for xenoliths in this type of host rock, which can be explained by infiltration of Re-rich kimberlite melt, and is visualized by sub-horizontal arrays in Re–Os isochron diagrams, at least for Phanerozoic kimberlites (e.g., Carlson et al. 2005). It is also evident in extended chondrite-normalized PGE plots, with normalized values plotted in the order of decreasing compatibility, where Re is often enriched relative to less incompatible HSE, such as Pd and Pt. In contrast, the effects of the infiltration of more viscous basaltic melts are largely limited to the xenolith rinds (e.g., Chesley et al. 1999). Basalt-borne xenoliths generally have Re concentrations similar to or lower than orogenic peridotites, but show large scatter, with a range of Re depletions at a given Al2O3, which suggests secondary volatility-related Re-loss. Thus, the use of Re to constrain conditions of melt extraction is completely compromised by its volatility and mobility in metasomatic agents (e.g., Li 2014). This decoupling of Re from the primary melting process also leads to geologically meaningless TMA model ages being calculated for most cratonic peridotites, as discussed earlier. Meaningful TRD can be obtained if (1) sufficient Re extraction has occurred during melt depletion, such that the initial 187Os/188Os is essentially frozen in and (2) if secondary Re addition (or loss) is so recent that no significant increase of 187Os ingrowth has occurred.
Melt modelling shows that Re/Os decreases with increasing Pi and that residues with < 20 ppt Re—below which decay to 187Os is too small to lead to resolvable change in the isotope ratio, and thus TRD, with the currently attainable precision—can be generated for Pi of 5 and 7 GPa at F ≥0.17 and ΔFMQ of 0. In contrast, higher F (≥ 0.30) is required at ΔFMQ of −2, and for Pi of 3 GPa, such low Re concentrations are not accessible at all (Fig. 11E). A comparison of modelled residual Re/Os–Al2O3 to measured samples shows that a large proportion of cratonic mantle peridotites have Re/Os that is too high for their degree of melt depletion and even higher than that of orogenic peridotites, which formed at lower average Pi and F, implying secondary Re addition and/or Os loss (Fig. 9A). Also, many basalt-borne cratonic peridotites have Re/Os that overlap with orogenic peridotites, even though cratonic peridotites formed by on average larger degrees of melt extraction and at higher pressures.
The disturbance of Re concentrations in peridotites of all types is evident in the large scatter in the Re–Os isochron diagram (Fig. 9B). While orogenic and basalt-borne peridotites show some coherence of 187Os/188Os with Al2O3 (Fig. 9C), as has been recognized early (Reisberg and Lorand 1995) and exemplified by the excellent correlation documented in Vitim peridotites (Pearson et al. 2004), such correlations are not observed in kimberlite-borne cratonic xenoliths (Pearson et al. 2004). The wide range of Al2O3 at 187Os/188Os ≤ 0.11 (i.e., giving ≥ 2.5 Ga TRD ages) could be explained by variable melt extraction ages or degrees of Re depletion, later Re addition and/or Al2O3 addition. The addition of cpx and minor garnet is usually ascribed to interaction with kimberlite (e.g., Simon et al. 2003), whereas opx could be part of the mantle from which the Kaapvaal craton formed (Aulbach et al. 2011) or due to interaction with siliceous fluids (Pearson and Wittig 2008). As neither kimberlite nor fluids will add significant Al2O3, it requires interaction with silicate melt, which is rarely produced beneath thick cratonic lithosphere. Moreover, such interaction should have led to co-introduction of Re. This is not supported by the lack of co-variation of Re/Os with Al2O3. Rhenium addition for samples with 187Os/188Os ≤ 0.11 must have been relatively recent, as otherwise ingrowth of radiogenic Os would have obliterated the Archean signature.
Despite these complications from multi-stage metasomatic overprint, a striking feature of the database is that ancient Os isotope signatures are preserved in many cratonic mantle xenoliths and individual sulfide grains (Figs. 6, 9C). This resilience may be due to high Os abundances in sulfides where residual PGE alloys recording primary melting events were captured by metasomatic BMS. Although Re (and perhaps also Pt and Pd) appears to be too overprinted to be a useful indicator of melt depletion conditions, the persistence of unradiogenic Os—as record-keeper of time-integrated low Re/Os—allows some insights into the required conditions of partial melting. For cratonic mantle formed between 3.5 and 2.9 Ga to have present-day 187Os/188Os ≤ 0.11, time-integrated Re/Os must have been < 0.025 and 0.012, respectively. Reproduction of the present-day 187Os/188Os of these samples using modelled Re/Os and assuming formation from primitive mantle 3.5 or 2.9 Ga ago, shows that in order to reach sufficiently low Re/Os to preserve 187Os/188Os of 0.11—as displayed by a sizable proportion of cratonic peridotites—Pi ≥ 5 GPa appears to be required.
The origin of circum-cratonic mantle is controversial, as it may represent lithosphere formed in the Proterozoic (e.g., Pearson et al. 2004; Janney et al. 2010) or cratonic lithosphere that was reworked during Proterozoic collisions at the craton margins (e.g., Griffin et al. 2009). Although most circum-Kalahari kimberlite-borne mantle xenoliths also appear to have suffered a secondary increase in Re/Os, a subset of samples may have retained primary Re/Os (Fig. 9A). However, unambiguously Archean TRD have not been determined for these samples (Pearson et al. 2004; Janney et al. 2010; Aulbach et al. 2014). This could be taken to indicate that this mantle did form in the Proterozoic, in keeping with their association with exclusively Proterozoic crust. On the other hand, Al2O3 and Re may have been re-introduced during strong reworking of cratonic margins. If Re addition was ancient, ingrowth of 187Os may lead to a correlation of measured 187Os/188Os with Al2O3 and individual sample suites would lie on an apparent Proterozoic partial melting trend. This may also be the case for some on-craton samples that yield Proterozoic TRD model ages. As noted above, it is not easy to pin-point the melt composition existing at the base of the lithospheric mantle that is capable of introducing Al in sufficient quantities. More strongly incompatible, minor elements such as TiO2 are better suited to address this question, as partial melting will cause such elements to lie on curved melting trends with moderately incompatible elements, such as Al2O3 and Re (Niu 1997), compared to straight trends for bulk mixing, but such data are not often reported for cratonic peridotites and are also subject to nugget effects from accessory host minerals, such as TiO2 in chromite.
Post-core formation, sluggish downward mixing of a late veneer
It has long been recognized that the abundances of the HSE in the terrestrial mantle are higher and their ratios more similar to undifferentiated chondrite meteorites than those expected from theoretical modelling of post-core formation mantle, during which the siderophile elements should have been near-quantitatively, but differentially extracted into the metallic core (Ringwood 1966; Jagoutz et al. 1979). A widely held view is that heterogeneous accretion and addition of the HSE from an oxidized “late veneer” remains the most viable explanation for the abundances and ratios of the HSE (Kimura et al. 1974; Chou 1978; Jagoutz et al. 1979; Morgan et al. 1981; Lodders and Palme 1991; Brenan and McDonough 2009; Siebert et al. 2011; Mann et al. 2012; Bennett et al. 2014). This veneer appears to have been spatially widespread and diachronous across the inner Solar System (Dale et al. 2012; Day et al. 2012, 2016, this volume). This hypothesis can be examined from the perspective of ancient high-volume melts, such as komatiites sampling deep mantle sources, and from cratonic peridotites that formed from Archean convecting mantle. Older (3.5–3.2 Ga) komatiites have been suggested to show PGE depletions relative to younger (2.9–2.7 Ga) komatiites, with the implication that this was related to slow downward mixing of a late accreted component into a lower mantle depleted by core formation, after accounting for, or discounting, effects of fractional crystallization, degree and depth of partial melting, sulfide saturation at source, or dilution by an eclogite component (Maier et al. 2009; Fiorentini et al. 2011).
Lower median Os and Ir concentrations in kimberlite-borne xenoliths than in modern mantle represented by orogenic peridotites (Fig. 3) are seemingly permissive of the idea that at least some mantle regions had not “seen” the late veneer (Maier et al. 2012). Although it is difficult to assess whether progressive homogenization with a late veneer occurred during the Archean owing to the inherent challenges to obtain precise and accurate ages from cratonic mantle xenoliths with multi-stage histories, the two radiogenic isotope systems embedded in the HSE—186Os from the decay of 190Pt and 187Os from the decay of 187Re—can provide some constraints on the time-integrated ratios of Pt/Os and Re/Os (e.g., Carlson et al. 2008; Puchtel et al. 2014). Core formation fractionates the HSE even under magma ocean conditions (Righter 2005) and should lead to suprachondritic Re/Os and Pt/Os in the silicate Earth (Walker et al. 1995, 1997; Cottrell and Walker 2006; Brenan and McDonough 2009; Bennett et al. 2014). If a late-added veneer was indeed incompletely mixed with the post-core formation mantle, one would expect to find supra-chondritic 186,187Os isotope compositions of the deep sources of komatiites and cratonic mantle until homogenization was reached. The record of initial 187Os/188Os of komatiites and mantle samples reveals that two types of sources were present since at least 3.8 Ga: One evolving at roughly chondritic parent-daughter ratios—and thus precluding global sluggish downward mixing of a late-accreted component—the other at supra-chondritic Os isotopic ratios and thus possibly indicative of the persistence of some mantle regions that did not equilibrate with the late veneer (Walker et al. 1997; Bennett et al. 2002; Walker and Nisbet 2002). Recent analyses of EoArchean ultramafic rocks, including chromitites, from western Greenland suggest sub-chondritic Pt/Os, likely generated in the Hadaean, at 4.1 Ga (Coggon et al. 2013).
There are a few examples of peridotite-derived age arrays, with geologically meaningful ages corresponding to events expressed in the overlying crust, that have indeed yielded radiogenic initial 187Os/188Os: This is the case for the central Slave craton (Aulbach et al. 2004, 2009a, 2011; Westerlund et al. 2006) and potentially the Pilbara craton and SW Greenland (Graham et al. 1999; Bennett et al. 2002). While such data can be, and have been, explained by addition of crustal components to the mantle source, they are also permissive—from an isotopic perspective—of regions of post-core formation HSE-depleted mantle source that evolved for some time at elevated Re/Os prior to mixing with the late accreted component.
Due to higher Archean TP, partial melting during generation of cratonic lithospheric mantle is characterized by deep melting paths enabling unusually strong depletion observed only rarely in later-formed mantle. We reviewed the HSE and Re–Os isotope systematics of cratonic mantle samples in the context of partial melting and subsequent multi-stage metasomatic overprint. In addition, we use partial melting models to determine whether HSE systematics can be used as barometers for oxygen and sulfur partial pressures (fO2 and fS2) as well as for pressures of onset of partial melting. We quantitatively compare published HSE systematics of natural samples to modelled melting residues in order to distinguish between those with potentially primary signatures and those that must have experienced secondary modification. A review of the literature combined with the modelling presented here lead us to the following general conclusions:
Cratonic peridotites are the most PPGE depleted peridotites sampled from the Earth’s mantle, showing consistently the most extreme fractionations of PPGE over IPGE. They also have the lowest average and median Os isotope ratios, as might be expected from their tectonic setting, and they have amongst the lowest Se and Te concentrations so far measured in mantle peridotites, although some depleted orogenic harzburgites have similar concentrations.
Modelling of HSE behavior during mantle melting shows that the IPGE hold promise as redox sensors, whereas PPGE and Re can be employed as pressure sensors. The IPGE are compatible and at higher degrees of melting they form alloys under all conditions considered (Pi = 3–7 GPa, ΔFMQ = −2 to 0). Their compatibility decreases sharply at higher fO2, which increases the solubilities of their alloys in silicate melts while also diminishing their partition coefficients in olivine. Platinum forms alloy only at the low-P and –fO2 end of the melting environment, whereas Pd, Rh, Re, and Au never form alloys and behave incompatibly in residual silicates once sulfide liquid is exhausted, whereupon they partition into the melt more strongly with increasing P and fO2.
Kimberlite-borne cratonic xenoliths have median Ir concentrations of 3.1 ppb, Os = 3.1 ppb, Pt = 2.8 ppb, and Pd = 1.2 ppb. Higher median Pt but similar Pd contents in circum-cratonic xenoliths may attest to formation at shallower pressures allowing Pt alloy stabilization and/or Pt-dominated overprint during reworking of cratonic margins. The exceptionally strong Pd and Pt depletions in kimberlite-borne samples compared to off-cratonic and orogenic mantle are consistent with their formation at elevated Archean TP and correspondingly higher Pi. Mantle xenoliths from the Greenland, Rae and Karelian cratons have HSE systematics suggestive of formation at lower Pi than those from the Kaapvaal and Slave cratons.
All primary sulfide is expected to be exhausted at the depleted Al2O3 contents exhibited by the vast majority of cratonic mantle xenoliths (initial S = 250 ppm, melt fraction F ≤ 0.20 for ΔFMQ = −2 and F ≤ 0.16 for ΔFMQ = 0), yet significant S concentrations are measured in many of these samples, requiring secondary S addition from low-volume volatile-rich melts. If this process occurs while the peridotite resides in the deep lithosphere, this likely leads to “resulfidation” of residual PGE alloys or extraction into newly formed sulfide liquid of IPGE formerly hosted by residual olivine, and co-introduction of Pd, Pt and Re, leading to the false appearance of lower apparent melt extraction P. The pervasiveness of this signature in cratonic mantle peridotites means that PPGE and Re systematics cannot be reliably used at present to infer depths of partial melting.
The extreme variability of IPGE concentrations in cratonic peridotites cannot be explained by decompression melting of convecting mantle. Instead, they appear to document partial melting under more oxidizing conditions or percolation of oxidizing melts. Concomitant loss of Pt alloy, if present, will lead to high apparent melt extraction P. The melts are consequently HSE enriched and can act as metasomatic agents in the cratonic lithosphere. Alternatively, the more oxidizing conditions required for IPGE and Pt alloy destabilization could be interpreted as being due to flux-melting in the presence of water. However, as with the PPGE, the signature is ambiguous and precludes drawing firm conclusions regarding melting environment.
The lower median IPGE concentrations of kimberlite-borne cratonic mantle xenoliths compared to orogenic mantle is permissive of sluggish downward mixing of late-accreted components post-core formation, but may, alternatively, reflect the inadequacy of the currently used estimates of primitive upper mantle concentrations, which are based on heavily refertilized Phanerozoic samples that experienced enrichment in Pd, Pt, and Ru and possibly some IPGE. We therefore estimate the HSE concentrations of Archean convecting mantle based on median concentrations of Ir, a compatible, non-volatile element, in kimberlite-borne mantle xenoliths erupted through cratons, which, coming from intact lithospheric mantle, are the least-disturbed samples with respect to Ir. Using chondritic relative abundances to calculate other HSEs relative to Ir, yields concentrations of Ir = 3.1 ppb, Os = 3.4 ppb, Pd = 4.3 ppb, Pt = 6.5 ppb, and Re = 0.27 ppb.
Re–Os isotopes combined with HSE systematics remain an unrivalled method to establish lithosphere stabilization ages. In using such ages, it is critical to assess whether the TRD ages provide a meaningful record of the extraction age leading to lithosphere formation, as opposed to minimum ages that can severely underestimate true age due to metasomatic disturbance or incomplete removal of Re during melting. Careful interpretation requires additional constraints, such as the use of major oxide relationships to distinguish melt- vs. metasomatism-related effects, lithophile trace-element fingerprinting to unravel the nature of the metasomatic agents, conventional oxybarometry to quantify consequences for the mantle redox state, and thermobarometry to allow localization in the lithosphere column.
A quantitative application of mantle HSE barometry will be possible once tighter constraints on the fS2-dependent alloy solubilities into silicate melts are established together with fully quantitative data on the partitioning into residual silicates (e.g., orthopyroxene and garnet). In addition, the use of Pi–fO2 as an indicator of melting environment (ridge, plume, arc) must await resolution of the as yet uncertain secular TP and redox evolution of the convecting and arc mantle.
We thank the editors of this special publication—Jason Harvey and James Day—for their invitation to contribute. Jean-Pierre Lorand, John Pernet-Fisher and the editors provided detailed and constructive comments that significantly improved the manuscript. Illuminating discussions with Chris Ballhaus and Raúl Fonseca are greatly appreciated.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution and reproduction in any medium provided that the original work is properly attributed.