Mineral Inclusions in Lithospheric Diamonds

Initially, it may seem extremely daunting to be sitting in front of several thousand carats of small diamonds, knowing that the likelihood of finding a diamond with a meaningful inclusion is about 1% of that parcel. Like other events in life, however, the joy of finding such material soon outweighs the perceived hardship. Mineral inclusions in diamond provide unique and pristine information about inner Earth, from depths of about 120 to just over 800 km and over 3.5 to 0.1 billion years of Earth history. Such is the scope and importance of inclusion research.

Once the inclusion-bearing diamonds have been collected, hopefully from a known source, an intricate set of procedures falls into place. A microscopic description of the physical features of the diamond and its valuable cargo are needed, a preliminary paragenesis is assigned based on inclusion color, and both diamond and inclusion are photographed in detail. Then comes the release of the µm-sized inclusion by crushing of the diamond host, the hope being that the mineral is recovered in one piece, an infrequent event. The next stage is mounting in epoxy and polishing, minimizing rounding of the inclusion surface en route to quarter μm perfection. When this stage is complete, the researcher can afford to relax as modern-day analytical techniques are non-destructive (electron microprobe) or dig only small holes in the perfect inclusion surface (ion microprobe and to a lesser degree, laser ablation-based micro-sampling).

In this chapter, we present results of the above procedures for minerals included in diamonds that originate in the lithospheric mantle. A comprehensive review of the mineral chemistry of inclusions, based on an already largely saturated database of almost 5000 inclusion analyses, was previously presented by Stachel and Harris (2008) and is not repeated here. Instead, the two main sections of the review focus on detailed comparisons of the major and trace element compositions of peridotitic and eclogitic suite inclusions with the equivalent minerals in cratonic xenoliths.

When John Evelyn was advised by his parents in 1642 that an extended European tour would be better than fighting on the Royalist’s side in the English Civil War (1642–1651), a connection between this sojourn and the probable first description of a mineral inclusion in diamond might be considered distinctly tenuous. Nevertheless, on Friday, September 29^th, 1645 his diary records that he was in Venice and he wrote, “I went to see the collection of a noble Venetian Signor Rugini: he has a stately Palace richly furnished, …. but above all was a diamond which had growing in it a faire ruby” (de Beer 1955). It is not surprising that he identified a ruby, a mineral known from ancient times. However, since ruby is exceptionally rare as an inclusion in diamond, the inclusion was most likely a peridotitic garnet.

Another 17^th century traveler, Jean Baptist Tavernier, noted black material in some Indian diamonds he bought, referring to them as “l’herbe pourrie”, translated as either “rotten weed” or “decomposed vegetable matter” (Tavernier 1676). The burning of a diamond by Lavoisier in 1772 and similar quantitative experiments with charcoal convinced the French scientist that these two substances were distinct forms of carbon. When graphite was further shown to be a third form of carbon, the origin and formation of diamond was linked to processes involving biogenic matter and gases, a concept that lasted for most of the 19^th century. Göppert (1864) outlines the beliefs of over twenty scientists published between 1816 and 1860. The majority view was that diamond formation was linked to biogenic matter, although water, topaz and gold were surmised as other inclusions. One author had a different view, suggesting that diamond formation was the result of compression of carbonic acid in Earth, which caused carbon to crystallize as diamond. This rather perceptive view of diamond formation was not appreciated by the monograph’s author, a botanist and paleontologist. Brewster (1861) examined fifty Indian diamonds and remarked that some had small cavities having “shapes that resembled those of insects and lobsters” (see below).

If one steps back in time for a moment and assumes a scientific knowledge base of the mid-19^th century, Figures 1 to 4 illustrate what the early scientists may have been looking at. Figure 1 shows graphite clusters in a diamond from Sierra Leone exhibiting a ‘plant-like’ impression. The realistic, encapsulated flying insect inclusion in Figure 2 is from the Karowe Mine in Botswana, the head and feelers being the result of some minute inclusion, which also caused two disproportionately large graphite-filled cleavage fractures to create the wings. The very rounded outline of the clear inclusion in Figure 3 might easily confirm the presence of water or gas within that diamond. Finally, Figure 4 would reinforce some of these initial observations as vegetable matter (graphite flakes) occurring on the interfacial surfaces between the water or gas inclusions and the diamond.

With the discovery of the first primary diamond deposits in South Africa in 1870, inclusion identification was still by observation only, and the new igneous environment did not have an immediate impact on research thinking. Boutan (1886) divided inclusions into crystalline and non-crystalline, and post-1870 research proposed hematite (iron glance), ilmenite and pyrite as inclusions. Lewis and Bonney (1897) considered that compressed gas was responsible for the common yellow color of diamonds from Kimberley. Bauer (1904) believed he had recognized fluid enclosures in diamond and, from their behavior when heated, considered the liquid to consist of either carbon dioxide and water or a saline solution. They also mention diamond as an inclusion.

With a new century came the understanding that the minerals found in diamond were linked to the mineralogy of the kimberlite rock and its xenoliths. In perhaps the shortest letter ever to be published in the periodical Nature (14 column lines long), Sutton (1907) reasonably argued the first identification of garnet (color unspecified) inside a South African diamond. Additionally, refractive index work on inclusions by Wagner (1914) identified pyrope and recorded the presence of (chrome) diopside in a diamond. Sutton (1921) listed garnet, ilmenite, pyrite, and mentions olivine for the first time as definite inclusions and notes zircon, mica and cognate graphite as probable inclusions. Sutton also studied epigenetic material in diamond fractures proposing calcite, an apophyllite mineral and pectolite. Magnetite was identified as an inclusion by Spencer (1924). The book on diamonds by Williams (1932) listed the known inclusion minerals up to that time and provided clear evidence of diamond as an inclusion: a mauve stone embedded in a colorless one. Williams concludes, as might be expected from the evidence at hand, that diamond was a phenocryst in kimberlite.

The first color micro-photographs of a suite of inclusions, mostly in polished diamonds, were described by Gübelin (1952). The inclusion in the sixth diamond in the color set likely resolves why zircons were commonly noted by previous authors as an inclusion in diamond, as the color of that mineral looks very much like the typical pale orange of an eclogitic garnet. The connection to zircon may also have been influenced by the typical honey color of zircons recovered from kimberlite heavy media concentrate. Two colorless inclusions were visually identified as quartz, a rare mineral in diamond, though they were likely the much more commonly found colorless mineral, olivine. Apart from crystalline inclusions, this work also captured photographic detail of fractures within diamonds containing dark/black material. For example, what the author called hematite ‘iron-roses’ look very much like typical sulfide inclusions within their rosette fractures. Graphite was identified by experiment, but much of the other dark material was considered as original magma. There were also observations and descriptions about diamond and cloud-like masses as inclusions.

Diamonds containing clouds were specifically examined by Giardini and Melton (1975). Similar to Gübelin, these authors considered such inclusions as gaseous. This is because the four, six or twelve sided cavities of 1 to 5 µm size contained no solid material when broken or polished through. Subsequent studies have shown this not to be the case (Weiss et al. 2022). Clouds range in color from white to mid-grey with an overall shape that can be irregular but is more often distinctly cuboid or star-shaped and visually aligned with the shape of the diamond. In the first case, the sides of the cube are aligned at right-angles to the fourfold direction of the diamond and in the second orientation, the six arms of the star cloud end at the apices of the octahedron, if that is the enclosing diamond shape.

The identification of an olivine inclusion by Mitchell and Giardini (1953) using X-ray diffraction techniques quickly developed into the principal way of defining inclusions in diamond. Orlov (1959) and Futergendler (1960) used this procedure to document the newly discovered (in 1954) Siberian diamonds, and the latter author reviewed her inclusion work over the second half of the 1950s. Futergendler first identified enstatite and chrome spinel, and from both authors other documented inclusions were olivine, chrome diopside, garnet and diamond. With Orlov, some cell parameters and refractive indices are given and inclusions are divided into syngenetic and epigenetic, the latter being serpentine, as a pseudomorph after olivine, red quartz and red iron hydroxide both filling diamond fractures. Much of Orlov’s inclusion work can be found in his translated book (Orlov 1977). Also, orientation studies between diamond and some of the principal inclusions were developed by Futergendler and Frank-Kamenetsky (1961), a subject fully considered in Angel et al. (2022).

Eppler (1961) studied the specific gravity and refractive indexes of some inclusions. From the associated description to his Figure 4, the previously common descriptions of ilmenite as an inclusion are most likely chrome spinel. Also identified were a high Mg# (molar 100Mg/(Mg+Fe)) olivine, a garnet of a composition approaching pyrope, the aforementioned ilmenite (likely chrome spinel) and apatite. He also remarked upon gaseous cloud-like inclusions, black expansion fractures around minerals, and considered that graphite and quartz were not likely inclusions.

From large batches of diamond samples from all the South African mines then operated by De Beers (and with one stone from the DRC), Sharp (1966) selected 89 diamonds containing black inclusions. After diamond crushing, brassy-colored non-diamond material was collected and individually examined by X-ray powder diffraction. Apart from a single cohenite (from a Jagersfontein diamond), pyrrhotite was identified as the dominant phase. Cell parameters indicated that two inclusions were nearly stoichiometric (i.e., showed no Fe deficiency) and therefore troilite. Most of the identified specimens were intergrown with pentlandite and to a lesser extent pyrite. Although there was some uncertainty about whether pyrite was primary, Sharp (1966) had found most of the minerals in the typical sub-solidus re-equilibration sequence of sulfide inclusions in diamond.

Harris (1968a,b) outlined criteria for recognizing whether inclusions in diamond were syngenetic or epigenetic. These works were based principally on the nature of the inclusions and their response to X-rays. Of the syngenetic inclusions, rutile and coesite were new identifications along with a dispersed dark cloud of microcrystalline olivine. Epigenetic inclusions were grouped into pseudomorphs after primary inclusions (goethite, quartz and possibly mica), fractures surrounding inclusions (graphite, pyrrhotite and pentlandite), and material in surface fractures (calcite, hematite, kaolinite, sellaite and xenotime).

Although commercial electron microprobe analyzers (EMPA) were available from the late 1950s, their use in determining the chemistry of inclusions in diamond only began in the late 1960s. This development essentially eliminated the need for X-ray techniques in inclusion recognition and came at a time when there was a growing awareness of the mineralogical and chemical links between inclusions in diamond and the equivalent minerals in peridotitic and eclogitic xenoliths found in kimberlites. The earliest EMPA study was completed by Meyer (1968), on a suite of inclusion garnets. Subsequent work by Meyer and Boyd (1972) separated garnets into peridotitic and eclogitic types, with inclusion assemblages further emphasizing the two principal growth environments for diamond. At this time, EMPA did not find many new inclusion minerals, although kyanite, ilmenite and zircon are examples, but instead rather precisely defined the chemistry of the existing mineralogies.

During the 1970s two major inclusion research centres emerged, one in South Africa through John Gurney at the University of Cape Town, the other through Nick Sobolev at the Siberian Branch of the Russian Academy of Sciences in Novosibirsk. Part of the inclusion research at Cape Town was linked to a prospecting program established by De Beers Consolidated Mines (organized by Barry Hawthorne and managed by Jeff Harris), who, over a ten-year period starting in 1970/71, provided representative inclusions for analysis from each of the De Beers mines in South Africa and Botswana. Similar independent inclusion studies were conducted in Novosibirsk on diamonds from the developing Siberian mines and linked to xenolith and xenocryst studies, providing important new insights into the crystal chemistry of inclusions. Much of the work on southern African localities, but also studies from other localities around the world, can be found in the 1^st to 4^th IKC Proceedings volumes (1975–1989 and references therein), where, for example, such rare inclusions as kyanite, sanidine, albite, phlogopite, zircon and moissanite are identified by EMPA and linked to a diamond paragenesis. Much of the early results of the Novosibirsk group were reviewed in a seminal book by Sobolev (1977; translated from the Russian edition of 1974).

With the southern African work, additional projects involving inclusions were set in place. For example, one was the determination of inclusion abundance tables and the relationship between inclusions and specific diamond shapes for some of the mines (Harris and Gurney 1979). A second project led to the first successful determination of genesis ages for diamond. Although by the mid-to-late 70s the eruption ages of several diamondiferous kimberlites were known (Davis 1977), the question still remained as to whether diamond was a phenocryst or xenocryst in the rock that brought it to the surface. The purity of diamond excluded any kind of dating analysis but the visual characteristics of the inclusions indicated a probably syngenetic growth with the diamond, and some of those minerals contained radioactive isotopes. Whilst the first studies used U–Pb in sulfides (Welke et al. 1974; Kramers 1979), the most successful work was started by Richardson et al. (1984), by determining the Sm–Nd isotopes in aliquoted harzburgitic garnet inclusions from the De Beers Pool and Finsch mines in South Africa. This work showed beyond doubt that diamonds were xenocrysts in kimberlite and easily justified the collection of 200 inclusion-bearing diamonds from both sources for the analyses. The Sm–Nd chronometer was then applied to date separate aliquots of eclogitic garnet and clinopyroxene inclusions (Richardson 1986) and, over the subsequent 20 years, to determine diamond ages in southern Africa, Russia and Australia (reviewed in Smit et al. 2022). With the considerable improvement in analytical precision, the Sm–Nd dating technique can now obtain an age from an array formed by single garnet or clinopyroxene inclusions (Koornneef et al. 2017; Gress et al. 2021).

Other dating techniques followed; the principal one being Re–Os isotopes from sulfide inclusions, first developed by Pearson et al. (1998) in diamonds from the Koffiefontein mine in South Africa specifically to date single diamonds. Results confirmed a xenocrystal origin for diamond but also, where comparison was possible, gave similar ages to those obtained from the silicates, indicating that the common dates were not coincidences (see Table 9 in Stachel and Harris 2008). The use of Ar–Ar isotopes from peridotitic or eclogitic clinopyroxene inclusions, although initially considered to be a genesis dating technique (see Burgess et al. 1989), became more useful in determining diamond eruption ages, with particular importance in diamond prospecting (see, for example, Phillips et al. 2018).

From the mid-1980s, therefore, not only had many different inclusion minerals been assigned to the peridotitic and eclogitic growth environments, or to the minor websteritic suite (Gurney et al. 1984), but also, inclusions were providing important information about mantle evolution. A significant consequence of the first two diamond dating programs, for example, was that the results generally correlated with tectonothermal events which had affected the lithospheric mantle where the diamonds occurred; a subject more fully considered by Smit et al. (2022). Also, the discovery and subsequent study of eclogitic inclusions in lamproite-hosted diamonds at Argyle, in Australia (Sobolev 1985, Jaques et al. 1989) led to the first proposal that the mantle source for a predominantly eclogitic diamond deposit was generated during subduction (in this case along the Kimberley Craton margin). Additionally, inclusion chemistry linked to HPHT experimental work enabled the temperature and pressure of inclusion formation to be determined (see Boyd and Gurney 1986; Nimis 2022), and broader inclusion studies began to explore the relationship between mineralogy and particularly the chemistry of inclusions and the carbon and nitrogen isotopes and nitrogen abundances of the host diamonds (see Stachel et al. 2022).

Returning specifically to the inclusions, a significant review of the two principal parageneses was provided by Meyer (1987) and for all three parageneses by Stachel and Harris (2008). Examples of further studies that have added to our inclusion knowledge are the finding of rare single crystals of primary graphite in diamond from the Panda kimberlite in Canada (Glinnemann et al. 2003) and a thorough study of rare phlogophite inclusions, identified in peridotitic and eclogitic diamonds from Siberia, Arkhangelsk and Venezuela (Sobolev et al. 2009). Thus, inclusions in diamonds seemed to firmly place diamond genesis in the subcratonic lithospheric mantle.

However, after four crystalline isotropic reddish-brown inclusions of magnesio-wüstite (now more commonly called ferropericlase) were recovered from four diamonds from the Koffiefontein mine in South Africa in the early 1970s (later studied by Cardoso 1980), this belief began to be queried. It was the recovery of a lower mantle inclusion assemblage of low-nickel enstatite (retrogressed bridgmanite) and a ferropericlase in a single diamond from the Orroroo kimberlite dykes in South Australia (Scott Smith et al. 1984) that established a whole new dimension to studies of inclusions in diamond. Thus were born sublithospheric and lower mantle diamond studies, a subject covered by Walter et al. (2022).

In conclusion to this historical overview of cratonic diamonds the reader is referred to Table 1, which updates the inclusion compilation of Stachel (2014) with subsequent discoveries.

The research on the geochemistry of inclusions in diamond captured in the database underpinning this review generally relied on sampling of run-of-mine diamonds between 1.8 and 3.5 mm in size (corresponding to round aperture Diamond Trading Company (DTC) sieve sizes +5 to −11), although in some rare instances inclusions were recovered from diamonds measuring less than 1.1 mm (DTC sieve size −1; e.g., Stachel et al. 2006). Based on the examination of over 1 million diamonds ~2 mm in size mined from nine kimberlites in South Africa and Botswana, the abundance of diamonds with visually identifiable mineral inclusions is 1.0% (see Table 1 of Stachel and Harris 2008). The size of studied mineral inclusions commonly is about 100 µm in longest dimension, although there are exceptions both to smaller and larger sizes. Our observations suggest an only minor increase in inclusion size towards larger diamonds and for the same inclusion mineral, the chemical composition does not significantly change with diamond size.

Disequilibrium among non-touching inclusions is occasionally observed, reflecting incorporation during diamond growth in chemically evolving environments (e.g., Griffin et al. 1988; Bulanova 1995; Wang 1998). Based on multiple lines of evidence (reviewed, e.g., in Gurney 1989 and Stachel and Luth 2015), disequilibrium between non-touching inclusions is nevertheless considered the exception rather than the rule.

Inclusions exhibit an often cubo-octahedral shape, irrespective of mineral symmetry or paragenesis (Harris et al. 2022). For many years, this characteristic was considered to reflect syngenetic growth with that of the host diamond (e.g., Harris 1968a; Bulanova 1995). This hypothesis was recently queried by Nestola et al. (2014), who observed an absence of an overall preferred orientation of inclusions in diamond indicating a protogenetic origin. Occurrence of both syngenetic and protogenetic relationships was subsequently documented through a combination of cathodoluminescence imaging and crystallographic orientation (EBSD) data (Davies et al. 2018). Based on small inclusion sizes (see above) and fast diffusion speeds in hot mantle environments, pre-existing host-rock minerals are nevertheless expected to generally equilibrate during fluid-mediated diamond growth events and thus the term “synchronous” inclusion was introduced (Nestola et al. 2017, 2019a; Pamato et al. 2021). That a lack of syngeneity does not necessarily preclude inclusion-based diamond dating is established in detail in Smit et al. (2022).

A first order division of mantle-derived diamonds is based on their origin in either the lithospheric mantle (our topic) or the sublithospheric mantle, the latter including the convecting upper mantle, the transition zone and the lower mantle (reviewed in Walter et al. 2022). Correcting for an extreme overrepresentation of sublithospheric diamonds in the literature, Stachel and Harris (2008) estimated that about 99% of diamonds produced worldwide ultimately derive from the lithospheric mantle and only about 1% from sublithospheric sources. The subsequent recognition that the CLIPPIR subpopulation of large, irregular, nitrogen-free (Type II) diamonds largely or entirely derives from transition zone to lower mantle environments (Smith et al. 2016) suggests that the estimate for sublithospheric diamonds needs to be increased to about 2% and the proportion of lithospheric diamonds decreased to 98%. Based on their mineral inclusions, lithospheric diamonds are conventionally subdivided into two main suites, peridotitic and eclogitic, plus a minor websteritic or pyroxenitic (websterites are a subclass of pyroxenites) suite (Meyer 1987; Gurney 1989; Stachel and Harris 2008). Based on 2844 diamonds with analyzed inclusions, Stachel and Harris (2008) derived relative abundances of 65% for the peridotitic, 33% for the eclogitic and 2% for the websteritic suite.

Peridotitic suite. Based on a primitive mantle Mg# of 89–90 (see McDonough and Sun 1995; their Table 4), olivine, orthopyroxene and clinopyroxene inclusions from depleted cratonic peridotite substrates¹ have Mg# ≥ 90, with infrequent Mg# values between 90–85 indicating increasingly melt-modified peridotitic substrates, which become transitional to websterites (Viljoen et al. 2018). In addition to their more magnesian character, peridotitic inclusions are separated from the eclogitic inclusion suite by Cr₂O₃ in garnet > 1 wt% (Schulze 2003) and molar Cr# (100Cr/(Cr+Al)) in clinopyroxene ≳7–10 (Stachel and Harris 2008). The peridotitic inclusion suite may be further subdivided into harzburgitic (to dunitic), lherzolitic, and wehrlitic parageneses, based principally on the groundbreaking work on garnet classification using Cr and Ca abundances by Sobolev et al. (1969, 1973). In this review, we employ the updated garnet classification scheme of Grütter et al. (2004) for this purpose. It should be noted that classification of peridotites based on garnet compositions is distinct from classification based on petrographic modes (Streckeisen 1976): garnet chemistry tracks equilibration with (1) orthopyroxene only (harzburgitic paragenesis), (2) clinopyroxene only (wehrlitic paragenesis), or (3) both pyroxenes (lherzolitic paragenesis), whilst petrographically defined boundaries are based on 5 vol% of either clinopyroxene (for the harzburgite/lherzolite distinction) or orthopyroxene (for the lherzolite/wehrlite distinction). Consequently, using garnet chemistry, highly depleted rocks may classify as lherzolites in spite of very minor (< 1 vol%) clinopyroxene contents. However, in contrast to xenolith studies, classification based on mineral modes is not feasible in inclusion research.

The minerals encountered as peridotitic suite inclusions in diamond are given in Table 1 and here we will exclusively focus on the common inclusion types (garnet, olivine, orthopyroxene, clinopyroxene and spinel), with the exception of sulfides that are discussed elsewhere in this volume (Smit et al. 2022). A table with inclusion abundance counts for lithospheric diamonds is given by Stachel and Harris (2008; their Table 2).

Eclogitic and websteritic/pyroxenitic suites. Eclogites are metamorphic rocks with a broadly basaltic composition, composed predominantly of jadeite-rich clinopyroxene and grossular-almandine-rich garnet (cf. Figs. 31, 32 in section Eclogitic–Pyroxenitic Substrates). In one more formal definition of eclogite, the clinopyroxene has Al/(Al + Fe³⁺) > 0.5 on the M1 site and 0.2 ≤ Na/(Na + Ca) < 0.8 on the M2 site (Clark and Papike 1968). Conversely, garnet–clinopyroxene rocks with jadeite-poor clinopyroxene are classified as pyroxenite, which, sensuStreckeisen (1976), encompasses all ultramafic rocks with <40 vol% olivine. There are gradations between garnet clinopyroxenites, which can be mineralogically and texturally similar to mantle eclogite, and websterites containing additional orthopyroxene ± olivine.

The assignment of silicate inclusions in diamond to the eclogitic suite is relatively straightforward and based on low Cr₂O₃ (< 1 wt%; see above) at elevated CaO contents in garnet, and low Cr₂O₃ and elevated Al₂O₃ contents in clinopyroxene (Gurney 1984; Grütter et al. 2004; Stachel and Harris 2008). In contrast, the websteritic inclusion suite, with compositions intermediate between the peridotitic and eclogitic suites, is less well defined, and relies on elevated Cr# and Mg#, and low Ca# in garnet, and/or the presence of orthopyroxene (Deines et al. 1993; Stachel and Harris 2008). Eclogitic and pyroxenitic garnet occurring as inclusions, as xenocrysts or in xenoliths can be grouped, based on Cr and Ca contents and Mg#, into classes G3 (eclogitic), G4 (eclogitic and pyroxenitic) and G5 (pyroxenitic) (Grütter et al. 2004). More recent work aimed at distinguishing Cr-poor garnet sourced from granulite and orogenic (crustal) eclogite vs. mantle eclogite recognized Ti/Si and Mg/Fe as useful discriminants (Hardman et al. 2018). The classification used here follows Aulbach and Jacob (2016) by employing a simplified version of the definition of Clark and Papike (1968), whereby cpx with 0.2 ≤ Na/(Na+Ca) < 0.8 classifies as eclogitic and that with lower ratios as pyroxenitic. This translates into distinctive bulk-rock compositions and is therefore not controlled by a pressure-dependent change in mineral chemical composition.

In addition, using a combination of major and trace elements, gabbroic eclogites with positive Eu anomalies indicative of high normative plagioclase modes (Eu/Eu* = chondrite-normalized Eu/(Sm × Gd)^0.5 > 1.05; Rudnick and Fountain 1995) are distinguished from basaltic eclogites with lower Eu/Eu*. The latter are classified into high-Ca, high-Mg and low-Mg eclogites. High-Ca eclogites have garnet with Ca# (molar 100Ca / (Ca + Fe^total + Mg + Mn)) > 20, high-Mg eclogites garnet with Ca# ≤ 20 and Mg# > 60, and low-Mg eclogites garnet with Ca# ≤ 20 and Mg# ≤ 60 (cf. Fig. 32a in section Eclogitic–Pyroxenitic Substrates). This classification contains petrogenetic information, reviewed below, but relies on the availability of both major and trace element compositions for both garnet and clinopyroxene. Also, these groupings do not typically have sharp distinctions, especially when samples from various localities with their own endemic compositional ranges are considered.

Inclusion minerals associated with the eclogitic and websteritic suites are given in Table 1 and again, our focus is on the common minerals (garnet and clinopyroxene) with the exception of sulfides, which are discussed by Smit et al. (2022).

As previously discussed, diamonds are principally associated with peridotitic and eclogitic substrates occurring in subcratonic lithospheric mantle. Through dating of inclusions following pioneering developments by Richardson et al. (1984) and Pearson et al. (1998) (for a review and full references see Smit et al. 2022) it emerged that formation of diamonds in harzburgitic substrates principally occurred in the Paleoarchean (Gurney et al. 2010; Shirey and Richardson 2011). Formation of lherzolitic diamonds preferentially occurred in the Paleoproterozoic but extends through the Mesoproterozoic and, with diamonds from the Victor Mine (Aulbach et al. 2018), even into the Neoproterozoic. For individual diamonds with presumably lherzolitic sulfides from Koffiefontein and Jagersfontein, even Mesozoic formation ages are observed (Pearson et al. 1998; Aulbach et al. 2009). Probably as a consequence of younger formation ages, lherzolitic diamonds are not exclusively restricted to undisturbed Archean lithospheric mantle but also occur in areas impacted by younger tectonothermal events (Shirey et al. 2002; Richardson and Shirey 2008; Smit et al. 2014a; Stachel et al. 2018b) or underpinned by entirely post-Archean lithospheric mantle (e.g., Banas et al. 2007; Hunt et al. 2009; Czas et al. 2020; Liu et al. 2021). Episodic addition of eclogitic diamonds to the subcratonic lithospheric mantle was associated with subduction along craton margins immediately following the presumed onset of modern style Wilson cycles at about 3 Ga (Helmstaedt et al. 2010; Shirey and Richardson 2011) and likely continues to the present, with the youngest eclogitic diamonds dated at 0.14 (± 0.09) Ga (Timmerman et al. 2017).

The formation of the most important diamond substrate, cratonic peridotites, is associated with intense primary melt depletion (Jordan 1978; Boyd 1989; Herzberg 2004) that reached up to the point of orthopyroxene exhaustion from the melting residues, i.e., requiring up to about 40% melt extraction (Walter 1999; Bernstein et al. 2007). These large degrees of melt extraction are presumed to relate to 100–300 °C higher mantle potential temperatures in the Archean (Labrosse and Jaupart 2007; Korenaga 2008; Aulbach and Arndt 2019a). The very low FeO of some cratonic peridotites (filtered for orthopyroxene addition and refertilization), which requires melt extraction at pressures > 5 GPa, has been used among other arguments to infer melting associated with ascending mantle plumes (e.g., Boyd 1989; Griffin et al. 1999b; Aulbach 2012). Based, for example, on Cr/Al ratios (Stachel et al. 1998) and concentrations of mildly incompatible trace element contents including HREE (Canil 2004; Pearson and Wittig 2008), a principally low-pressure origin in the Archean equivalent of oceanic ridges and/or arc mantle is favoured by others (reviewed in Brey and Shu 2018 and Pearson et al. 2021), which implies a metamorphic origin for cratonic garnet peridotites from Cr-spinel peridotite precursors. Suggested modes of emplacement into the subcratonic lithosphere then involve imbrication associated with subduction (Helmstaedt and Schulze 1989; Kesson and Ringwood 1989), or lateral compression and thickening (Jordan 1978; Wang et al. 2018; Pearson et al. 2021) or vertical attaching (Herzberg and Rudnick 2012; Perchuk et al. 2020).

Irrespective of the exact mode of origin, in several cratons (e.g., Kaapvaal, Yakutia, Tanzania) the composition of cratonic peridotites points to enrichment in silica (i.e., enrichment in orthopyroxene; Boyd 1989; Kesson and Ringwood 1989; Kelemen et al. 1998). Major element relationships (Al/Si, Mg/Si) of cratonic peridotites indicate involvement of high-temperature aluminous pyroxene in this process, inconsistent with Si enrichment through infiltration of siliceous melts in cold supra-subduction zones (Aulbach 2012). Also, the absence of detectable perturbations in δ¹⁸O values of orthopyroxene-enriched cratonic peridotites clearly precludes slab-derived melts as the principal agent of Si enrichment (Regier et al. 2018) and points to melt–rock reactions involving mantle-derived melts instead (e.g., Kelemen et al. 1992; Koornneef et al. 2009; Tomlinson and Kamber 2021). For the Kimberley area of the Kaapvaal Craton, orthopyroxene enrichment in the lithospheric mantle has been linked to the 2.7 Ga Ventersdorp large igneous province (Branchetti et al. 2021), which agrees well with the model of Si enrichment with komatiite infiltration (Tomlinson and Kamber 2021). More subtle metasomatic enrichment of principally trace elements (cryptic metasomatism; Dawson 1984) in Kaapvaal cratonic peridotites was attributed to percolation of small-volume melts of the carbonatite-kimberlite spectrum, peaking between 3.3 and 2.7 Ga and then tapering off in the Mesoproterozoic (Shu and Brey 2015). As a final important stage of metasomatic overprint of the Kaapvaal Craton lithospheric mantle, Simon et al. (2007) identified localized clinopyroxene ± garnet re-enrichment, likely associated with kimberlite magmatism < 200 Ma.

Mantle eclogites and associated inclusions in diamond are widely interpreted as representing various parts of subducted oceanic crust, which inherited major element, trace element and isotopic signatures from low-pressure accumulation-differentiation processes of igneous protoliths, seawater alteration and subsequent dehydration and partial melting (MacGregor and Manton 1986; Beard et al. 1996; Jacob 2004; Stachel et al. 2004). Strong indices for a subduction origin of inclusions and xenoliths comprise non-mantle oxygen isotope compositions, implying low temperature crustal fractionation, and Eu anomalies, requiring the participation of plagioclase during protolith formation (Schulze et al. 2003; Jacob 2004; Stachel et al. 2018a). Moreover, sulfide inclusions have yielded a range of mass-dependently fractionated S isotope compositions, with δ³⁴S from −13 to +14‰, similar to sediments and altered oceanic crust (Chaussidon et al. 1987; Eldridge et al. 1991). Finally, mass-independently fractionated S isotope compositions reflect processes unique to the O₂-poor Archean atmosphere, which were incorporated first into sediments and, after subduction, into sulfides trapped by diamond (Farquhar et al. 2002; Thomassot et al. 2009; Smit et al. 2019a). Apart from such signatures of a low-pressure origin and the stable isotope evidence, direct formation from mafic melts in the mantle is precluded because major and trace element relationships are inconsistent with the participation of garnet, which is on the liquidus of mafic melts at depths from which eclogites are entrained (Aulbach and Arndt 2019b). The occurrence of eclogite with low-pressure signatures in the cratonic lithosphere, combined with age constraints (Smit et al. 2022), indicates emplacement during two principal processes: (1) Archean amalgamation of cratonic nuclei into larger cratonic masses, with closure of intervening ocean basins and (2) subsequent accretion/collision at craton margins, best exemplified by the Slave craton.

For the present review, we compiled two databases for inclusions in diamond of all suites from the literature, one focusing on major element compositions (8412 inclusion analyses), the other on trace element compositions (949 inclusion analyses). For the presented data, a blanket detection limit of 0.02 wt% was assumed for major element oxides. For trace elements, collated analytical tables in the literature typically only report data above the specific limits of detection. These databases are available for download at https://doi.org/10.7939/DVN/EJUE1G (major element database) and https://doi.org/10.7939/DVN/581P99 (trace element database).

Peridotitic inclusions and xenolith minerals. For comparison of inclusion compositions with the same mineral phases in cratonic peridotites, we assembled a further database of minerals in well-equilibrated mantle xenoliths, including 967 samples from on-craton kimberlite localities (used in plots presented here), 909 being garnet peridotites. This database is not yet in a publication ready state but can be obtained from us on request. Mineral analyses of olivine, orthopyroxene, clinopyroxene, garnet or spinel were filtered for analytical totals between 98.5 and 101.5 (98.5–102.0 for garnet and 97.5–102.0 for spinel) and for cation per formula unit ranges that match accepted bounds for these mineral species. To assess differences and similarities in the trace element composition of inclusion and xenolith/xenocryst garnets, we compiled an additional database with trace element compositions of 152 harzburgitic and 281 lherzolitic garnets from xenoliths (or in a few instances occurring as xenocrysts) in on-craton kimberlites. For both xenolith databases, a paragenesis was assigned based on garnet major element chemistry (Grütter et al. 2004), which for the major element database in a few instances was revised based on assignments in the original publications (where garnets with “lherzolitic” compositions sometimes come from pyroxenite xenoliths). Of the 843 peridotitic garnets that pass the analytical filters described above, 679 (80.5%) derive from lherzolites (classified as G1 = megacrysts (n = 64), G9 = lherzolitic (n = 460), and G11 = high-TiO₂ peridotitic (n = 155)), 154 (18.0%) derive from harzburgites (with 105 of the G10 garnets classifying as “diamond facies” (G10D)), and 12 (1.4%) derive from wehrlites (G12). Only samples with an assigned harzburgitic or lherzolitic paragenesis were used in this review. Textural information (allowing to separate, e.g., strongly deformed varieties) is generally not available for these samples.

For the inclusion database, orthopyroxenes and clinopyroxenes with Mg# < 90 recovered from diamonds from the Voorspoed Mine were excluded from the peridotitic data set because of their compositionally unique character related to intense melt metasomatism (Viljoen et al. 2018).

Eclogitic inclusions and xenolith minerals. For comparison with their xenolithic counterparts, only garnet and/or clinopyroxene-bearing diamonds are considered. The database comprises reported eclogitic and websteritic diamonds with 23 touching clinopyroxene–garnet pairs and 249 non-touching pairs, 567 unpaired clinopyroxenes, and 906 unpaired garnet inclusions. Multiple non-touching inclusions per mineral and diamond are not counted individually but were averaged instead. However, touching inclusions were not averaged with non-touching ones as they may have re-equilibrated in response to changing pressure–temperature conditions, contrary to non-touching inclusions.

The xenolith database encompasses on-craton samples for which both major and trace element compositions are reported for both garnet and clinopyroxene, and excluding those where trace element compositions were affected by kimberlite contamination evident in anomalous Ba and Nb concentrations (cf., Aulbach and Jacob 2016). Cut-off values for contamination are estimated using mineral–melt distribution coefficients and concentrations in the contaminant, typically kimberlite, which determine how much Ba and Nb can be incorporated in garnet and clinopyroxene under equilibrium conditions. Garnet/kimberlite distribution coefficients for Ba of 0.0002 and for Nb of 0.0008, and clinopyroxene/kimberlite distribution coefficients for Ba of 0.001 and for Nb of 0.0002 are taken from Girnis et al. (2013). Combined with Ba and Nb concentrations in primitive kimberlites of 1250 and 260 ppm, respectively (Becker and Le Roex 2006), this yields maximum values of 0.25 ppm for Ba and 0.052 ppm for Nb in garnet, and of 1.25 ppm for Ba and 0.208 ppm for Nb in clinopyroxene at equilibrium. Allowing for metasomatism by more incompatible trace element enriched melts, we apply cut-offs for Ba in garnet and clinopyroxene of 0.5 and 1.5 ppm, respectively, and for Nb of 0.5 and 1 ppm, respectively. For a comparison of mineralogy, only xenoliths for which at least clinopyroxene and garnet major element data are reported are considered, which is the case for 1907 eclogite and 544 pyroxenite xenoliths. For a comparison of compositions, diamond-absent xenoliths are only considered when REE data are available, allowing full classification (695 samples), whereas diamond-bearing xenoliths (n = 251) are retained regardless of trace-element data availability (cf. Table 3 in section Eclogitic–Pyroxenitic Substrates). Cratons with corresponding data coverage encompass Kaapvaal, Zimbabwe, West Africa, India, Siberia, Slave, Sask, Superior, the North Atlantic craton in West Greenland, and the East European Platform (references in Appendix Table 1).

Values below the detection limit are often not given and can therefore not be counted for calculation of average or median compositions, which implies that these are maxima. Since this applies only to a small proportion of the data for elements considered here, the resultant bias is considered minor.

Of the 1468 analysed peridotitic olivine inclusions in diamond, the majority (79%) cannot be assigned to a specific paragenesis due to an absence of coexisting garnet or clinopyroxene, 16% are harzburgitic, 5% are lherzolitic and a single inclusion (equivalent to 0.07%) is of wehrlitic paragenesis. The olivine inclusions are high in forsterite content, with a median Mg# of 92.8 (mean is 92.7 ± 1.3; 1 sigma). Olivine from cratonic garnet peridotites is lower in Mg# (median value of 91.8, mean of 91.7 ± 1.3). Comparing the Mg# of harzburgitic (median and mean of 93.1) and lherzolitic (median and average of 91.8) olivine inclusions, harzburgitic inclusions are more magnesian (Mg# difference is 1.3; Fig. 5). The same distinction is seen for harzburgitic (median is 92.8, average is 92.9) and lherzolitic (median is 91.6, average is 91.5) olivines in cratonic garnet peridotites (average difference is 1.3; Fig. 5). For both parageneses, equality of means in Mg# of olivines included in diamond and contained in xenoliths cannot be rejected, based on a two tailed Student’s t-test for two independent samples, assuming unequal variances and an alpha value of 5%. Olivine Mg# is widely interpreted to gauge the degree of depletion of the protoliths of subcratonic lithospheric mantle (Boyd 1989; Herzberg 2004; Bernstein et al. 2007; Pearson and Wittig 2014) with typically only minor subsequent modification. The observation of equal means then suggests that, on the paragenesis level, diamond substrates and diamond-free garnet peridotites experienced effectively the same depletion history. Equality of means on the paragenesis level is reconciled with a higher Mg# of inclusion olivines relative to xenolith olivines based on very different relative proportions of the harzburgitic and lherzolitic parageneses, with 75% of inclusions with an assigned paragenesis being harzburgitic, as opposed to 15% of olivines from garnet harzburgite xenoliths.

A long-recognized difference between olivine included in diamond and from garnet peridotite xenoliths is a higher Cr content of the former (Meyer and Boyd 1972; Meyer 1987). Peridotitic olivine inclusions have a median Cr₂O₃ content of 0.05 wt% (mean is 0.06 ± 0.08 wt%), with an interquartile range of 0.03–0.14 wt% and a maximum value of 1.36 wt%. Olivine in cratonic garnet peridotites has a median Cr₂O₃ content of 0.03 wt% (mean is 0.03 ± 0.02 wt%), with an interquartile range from below the limit of detection (< 0.02 wt%) to 0.05 wt%, with a maximum value of 0.12 wt%. Higher Cr in inclusion olivines has been related to incorporation of Cr²⁺ (Meyer and Boyd 1972) during crystallization at reducing conditions and at high temperature (Li et al. 1995). High Cr, however, is not a universal characteristic of olivine inclusions and only 10% have Cr₂O₃ > 0.10 wt%.

As with olivine inclusions, the majority of the 417 analysed peridotitic orthopyroxene inclusions (57%) are of unspecified paragenesis, 32% are harzburgitic and 11% are of lherzolitic paragenesis. Inclusion orthopyroxene is almost pure endmember enstatite, with a median Mg# of 94.1 (mean is 94.0 ± 1.9)². These Mg# values are 1.3 higher than the median and mean for olivine inclusions in diamond (see above). Orthopyroxene in cratonic garnet peridotites is less magnesian than inclusions (median Mg# of 92.9, mean of 92.8 ± 1.1). On the paragenesis level, harzburgitic orthopyroxene inclusions (median value of 94.9, mean of 94.8 ± 1.2) are more magnesian than lherzolitic inclusions (median Mg# of 92.8, mean of 92.4 ± 1.9; Fig. 6). The harzburgitic inclusion dataset is strongly biased by a large number (n = 59) of highly magnesian orthopyroxenes from De Beers Pool (the combined production of the Kimberley Mines), and after removing this sample set, the median Mg# drops to 94.1 (mean of 94.2 ± 1.1; Fig. 6). These paragenesis level values are very similar to harzburgitic (median Mg# of 93.8, mean of 93.9 ± 0.7) and lherzolitic (median Mg# of 92.7, mean of 92.6 ± 1.0; Fig. 6) orthopyroxenes from cratonic garnet peridotites. A two tailed Student’s t-test for two independent samples, assuming unequal variances and an alpha value of 5%, however, rejects the hypothesis of equal means for both the harzburgitic and lherzolitic inclusion and xenolith orthopyroxenes. Based on the nevertheless high similarity of inclusion and xenolith orthopyroxene on the paragenesis level, the lower Mg# of xenolith orthopyroxenes for the combined peridotitic data sets largely reflects a much lower proportion of the harzburgitic paragenesis, which makes up only 16% (as opposed to 75% of inclusions). This matches the observation made for olivine inclusions.

Among minor elements, NiO in orthopyroxene is higher in inclusions in diamond (median value is 0.13 wt%, mean is 0.13 ± 0.03 wt%) compared to cratonic garnet peridotites (median value is 0.09 wt%, mean is 0.07 ± 0.05 wt%). There is no variation in NiO with paragenesis: for inclusions of known lherzolitic and harzburgitic paragenesis the median NiO content is 0.13 wt% and the mean is 0.14 ± 0.03 wt%; for xenoliths the two parageneses have the same median and mean as the combined peridotitic orthopyroxene dataset. Given the similarity in Mg# of orthopyroxene inclusions and in xenoliths on the paragenesis level and the similarity of NiO in olivine inclusions (median NiO is 0.36 wt%, mean is 0.36 ± 0.05 wt%) and in xenoliths (median NiO is 0.38 wt%, mean is 0.36 ± 0.08 wt%), this observation cannot be interpreted as a bulk rock signal but suggests a temperature control on Ni in orthopyroxene instead, similar to Ni in garnet (Griffin et al. 1989a). Indeed, a blurred positive correlation (r² = 0.2, not illustrated) between temperature of last equilibration (enstatite in clinopyroxene-based estimates; Brey and Köhler 1990) and NiO content is observed for our database of 189 orthopyroxene analyses from cratonic garnet lherzolite xenoliths.

Due to their incompatible behavior during mantle melting, Na and Ti were efficiently removed from the Archean melting residues that now form subcratonic lithospheric mantle (O’Hara et al. 1975; McDonough and Rudnick 1998). This is well reflected in harzburgitic orthopyroxene inclusions, where over 50% of analyses for Na₂O and TiO₂ are below the limit of detection (≤ 0.02 wt%). For inclusions of the lherzolitic paragenesis (and presumably lherzolitic inclusions contained in the “unspecified peridotitic” group), metasomatic reenrichment of the diamond substrates is documented in maximum Na₂O contents of 0.57 wt% and maximum TiO₂ contents of 0.29 wt%. For orthopyroxene from cratonic garnet peridotites, the metasomatic re-enrichment of Na and Ti can be much more intense (Fig. 7). From the xenolith data set, a positive correlation between Na and Ti is apparent, which on a cation basis has a slope of about 2, consistent with a coupled substitution of the type 2Na¹⁺-^VITi⁴⁺ (Cameron and Papike 1981). Using the 95^th percentile values for Na₂O (≤ 0.16 wt%) and TiO₂ (≤ 0.06 wt%) of orthopyroxene inclusions as upper cutoffs, which in combination encompass 93% of the inclusion data set, permits (empirical) definition of a diamond inclusion field for orthopyroxenes. Forty two percent of orthopyroxenes from cratonic garnet peridotites fall outside this inclusion field, indicating that substantive metasomatic re-enrichment of Na and Ti is much more widespread in xenolith samples.

Lherzolitic clinopyroxene inclusions (n = 148) are Cr₂O₃-rich, with contents typically near 1 wt%. Their median Mg# (92.8) is the same as that of lherzolitic orthopyroxene inclusions and mildly elevated relative to clinopyroxene from cratonic garnet lherzolites (92.3; Table 2). Compositional differences between inclusion and xenolith clinopyroxenes are evident for Al, Ti, Na, but largely overlapping compositional ranges preclude the definition of compositional cut-offs between the two groups (Table 2). Al₂O₃ in inclusions shows a strongly right (towards higher Al) skewed distribution with a mode at 0.8 wt% (median value 1.42 wt%) whereas xenolith-derived clinopyroxene displays a near-normal distribution with a mode at 1.8 wt% (median value 1.96 wt%; see Fig. 8). Inclusion clinopyroxenes also have lower median and 25^th percentile (Q1) values in TiO₂ and Na₂O contents (Table 2). The strongest compositional difference exists for K₂O, where the inclusion distribution shows a tail up to 1.68 wt%, whereas xenolith clinopyroxenes do not exceed 0.26 wt% (Fig. 9). The occurrence of strong to extreme metasomatic enrichment in potassium is a unique characteristic of inclusions, which also have a higher median value and interquartile range (Q1 and Q3 values Table 2), but nevertheless, the bulk of clinopyroxenes from diamonds and cratonic garnet lherzolites overlap in K₂O content (Fig. 9).

The “classical” Cr₂O₃ versus CaO diagram (Sobolev et al. 1973; Gurney and Switzer 1973; Gurney 1984; Grütter et al. 2004; Fig. 10), with fields separating the three principal peridotitic parageneses, harzburgitic (-dunitic), lherzolitic and wehrlitic, is widely used in assessing the compositions of mantle-derived garnets. For our database (n = 1175), 81% of inclusion garnets are of harzburgitic paragenesis, 18% are lherzolitic and 1% are wehrlitic. Ultra-depleted harzburgitic and dunitic diamond substrates are represented by garnets with CaO contents < 1.8 wt% (Grütter et al. 1999; Fig. 10), which applies to 38% of peridotitic inclusions (or 47% of garnets classified as harzburgitic). In evaluating these relative proportions, it should, however, be noted that the published garnet inclusion data set is strongly biased towards a few well studied locations, with about 40% of analyses in our database being derived from only four occurrences (De Beers Pool, Finsch and Premier (Cullinan Mine) on the Kaapvaal Craton and Victor on the Superior Craton), with a clear under-representation of inclusion garnets from Yakutian occurrences (9%). For that reason, we consider the original proportion of 85% harzburgitic (-dunitic) or “G10” garnet inclusions established by Gurney (1984), using a much smaller data set, to be still equally valid. Garnets from cratonic peridotite xenoliths (n = 843) show an exactly inverse relationship in the proportions of the harzburgitic (18% in our database) and lherzolitic (81%) parageneses, with again 1% of garnets with wehrlitic compositions (Fig. 10). Ultra-depleted (< 1.8 wt% CaO) xenolith garnets are rare (2% of all peridotitic and 13% of harzburgitic garnets).

Calcium intercept values (calculated after Grütter et al. 2004) are a one-dimensional measure of how far a garnet is removed in Ca content from the harzburgite–lherzolite division, which intersects the x-axis in a Cr₂O₃–CaO plot (Fig. 10) at 3.375 wt% CaO. In garnet compositional space, the transition to Ca intercept values below 3.375 is considered petrologically equivalent to the exhaustion of clinopyroxene from the residues during primary melt depletion and corresponds to about 20% melt removal (Jaques and Green 1980; Kushiro 1994). Further decreasing Ca intercept values of garnet document increasing degrees of primary melt depletion, progressing to the exhaustion of orthopyroxene in the residues at about 40% melt extraction (Jaques and Green 1980; Bernstein et al. 2007 and references therein). For harzburgitic garnets, Ca intercept values, therefore, are a measure of the primary chemical depletion of their substrates (see Grütter et al. 1999 for a detailed discussion). Correspondingly, Mg# and Ca intercept values are negatively correlated for both inclusion and xenolith garnets (not shown). Garnet Mg# is, however, controlled by temperature-dependent Mg–Fe partitioning with olivine, orthopyroxene and clinopyroxene and hence only provides a distorted image of bulk rock Mg#. Pairing garnet Ca intercept values with the Mg# of coexisting olivines (Fig. 11; see also Eaton et al. 2009, their Fig. 4) sidesteps this distortion problem, with the drawback of greatly reducing the data set for inclusions (from 1175 individual garnets to 198 garnet–olivine pairs). Despite some scatter, regression of the inclusion data set yields a clear linear trend (r² = 0.51; Fig. 11), with olivine Mg# increasing from 92.0 at the transition from lherzolitic to harzburgitic garnets (Ca intercept value = 3.375) to 94.2 for Ca-free garnets. Note that for the lherzolitic paragenesis some deviation from this linear trend is expected, as (1) varying degrees of primary melt depletion prior to the point of exhaustion of clinopyroxene in the protoliths would not be reflected in the Ca intercept values of garnet and (2) metasomatic addition of clinopyroxene to original harzburgitic melting residues may have occurred. In spite of that, including only harzburgitic garnet–olivine pairs results in an indistinguishable regression line (y = 94.2 − 0.62x versus y = 94.2 − 0.65x for the entire data set) but with a lower r² (0.31). For the xenolith data set, the same trend of decreasing Mg# with increasing Ca intercept values is visible but with much more scatter (Fig. 11). For the lherzolitic paragenesis, the variation in olivine Mg# at a given garnet Ca intercept value visually appears more pronounced than for the inclusion data set. Given the equality of mean Mg# of lherzolitic olivines inside and outside of diamond (see above), this apparently increased variability likely only reflects a much larger number of analyzed lherzolitic xenoliths compared to inclusion pairs. Highly similar median and mean Mg# values of lherzolitic garnets inside and outside diamond (inclusions: median = 83.0 and mean = 83.3 ± 2.8; xenolith garnets: median = 83.4, mean = 83.1 ± 2.2) support this interpretation. The overlapping trends in olivine Mg# versus garnet Ca intercept value document that peridotitic diamond substrates and cratonic garnet peridotites followed the same melt depletion path, with the key difference that carbon-bearing fluids preferentially react with the cpx-absent, depleted peridotitic substrates during mineralization events that produce diamond.

As a consequence of intense primary melt depletion, Na and Ti contents in peridotitic diamond substrates and cratonic peridotites are expected to be very low (see above). For harzburgitic garnets included in diamond and from cratonic peridotite xenoliths, median Na₂O and TiO₂ contents are indeed at or below the limit of detection (≤ 0.02 wt%; averages are not reported because of strongly non-normal distributions), with concentrations above the limit of detection being considered to indicate secondary metasomatic re-enrichment (Figs. 12 and 13). For the lherzolitic paragenesis, median contents are above the limit of detection for TiO₂ (0.13 wt% for inclusions and 0.24 wt% for xenolith garnets) and for xenolith garnets also for Na₂O (median of 0.04 wt%). In this work, we repurpose Na₂O ≤ 0.02 wt% (below the limit of detection) and TiO₂ ≤ 0.04 wt% from our previous work (Stachel and Harris 2008) to represent upper-limit thresholds for garnets included in diamond and from xenoliths, above which cratonic lherzolites are considered enriched in Na and Ti by secondary metasomatism.

On that basis, Figures 12 and 13 document widespread metasomatic modification of the harzburgitic and lherzolitic substrates of both inclusion and xenolith garnets. Maximum Na₂O concentrations in garnet are 0.31 wt% (a harzburgitic inclusion outside the range of Figure 12) and 0.19 wt% (xenoliths) and extreme TiO₂ values reach 1.08 wt% (inclusions) and 1.91 wt% (xenoliths). For xenolith garnets there is a clear positive correlation of Na with Ti (r² = 0.67; Figure not shown), which is barely discernible for inclusions (r² = 0.15). A slope of Na versus Ti of about 0.2 (on a cation basis) indicates that Na likely is fully accommodated by the substitution mechanism ^VIIIM²⁺ + ^VIAl³⁺=^VIIINa¹⁺ + ^VITi⁴⁺ (Bishop et al. 1976). The strong excess of Ti over Na reflects additional substitution mechanisms of the type 2^VIAl³⁺=^VITi⁴⁺ + ^VIM²⁺ or ^VIAl³⁺ + ^IVSi⁴⁺=^VITi⁴⁺ + ^IVAl³⁺ (Ackerson et al. 2017).

Intense metasomatic addition of Ti is particularly common for lherzolitic xenolith garnets, with 25% of samples containing > 0.5 wt% TiO₂ (Q3 value), as opposed to only 2.4% of inclusions. A large set of pressure–temperature estimates for well-equilibrated cratonic garnet lherzolites (n = 372) allows the relationship of Ti-metasomatism and temperature of last equilibration to be assessed. This relationship indicates that TiO₂ contents > 0.5 wt% (the above quoted Q3 value for lherzolitic xenolith garnets), with one exception, are restricted to temperatures > 1130 °C (Fig. 14). A comparable data set does not currently exist for inclusions in diamond but the observed paucity of inclusion garnets with TiO₂ > 0.5 wt% documents that the style of high-temperature, Ti-rich (“melt”) metasomatism (Griffin and Ryan 1995) documented in one quarter of lherzolite xenoliths either did not significantly impact diamond substrates or only occurred in a different, much milder form (note footnote 1 in section Introduction here). Given that typical temperatures of diamond formation in peridotitic substrates are estimated at 1130 ± 120 °C (Stachel and Luth 2015; see also Nimis 2022), this near absence of Ti metasomatism of diamond substrates is not a temperature effect but either indicates an increasing abundance of Ti-rich melt metasomatism in the more recent geological past (Mesoproterozoic to Phanerozoic) and/or a diamond unfriendly effect of such metasomatism that renders affected potential substrates infertile.

Low Al contents and consequent high bulk rock Cr/Al ratios cause spinel to persist to diamond-stable pressures in cratonic peridotites and, in particular, in peridotitic diamond substrates (Doroshev et al. 1997; Klemme 2004; Ziberna et al. 2013). Of the 1001 peridotitic spinel inclusions in diamond in our database, most (97%) are associated with neither garnet nor clinopyroxene and hence are of unspecified peridotitic paragenesis. Based on the observation that only 9% of the Mg-chromite inclusions with an assigned paragenesis are lherzolitic and 91% are harzburgitic, a mostly harzburgitic paragenesis may also be assumed for the inclusions with unspecified paragenesis. Based on high pressure experiments, spinel, in equilibrium with garnet and orthopyroxene, at the pressure–temperature conditions of the graphite–diamond transition along a typical cratonic geotherm, should have a molar Cr# (100Cr/(Cr+Al)) of ≥ 80 (Doroshev et al. 1997; Girnis and Brey 1999; Klemme 2004). In good agreement with these experimental outcomes, the median Cr# of spinel inclusions in diamond is 86.8 (mean = 86.6 ± 3.5) and the few samples of unspecified peridotitic paragenesis falling towards lower Cr# (Fig. 15) likely represent orthopyroxene-free (i.e., dunitic) assemblages, for which spinel + olivine remains stable for all lithospheric pressures.

For comparison of spinels from cratonic garnet peridotite xenoliths with inclusions in diamond, we excluded all xenolith spinels with Cr# < 60 because of a likely shallow garnet facies (< 100 km, see Klemme 2004) or retrograde origin. Despite this cut-off, the Cr# of spinel from garnet peridotites (median 80.5, mean 78.8 ± 6.8) is lower than that of inclusions. The Mg# of all but two (0.02%) spinel inclusions is > 50 (median = 69.1, mean = 68.9 ± 5.2) and compared to xenolith spinels (median Mg# = 62.6, mean = 61.6 ± 6.0) higher by about 6–7. Besides bulk rock composition, the Mg# of spinel depends on the temperature dependent exchange of Mg–Fe with olivine (O’Neill and Wall 1987); elevated Mg# for inclusion spinels may thus, at least partially, represent a temperature (depth) effect. Combined with high Cr#, spinel Mg# generally >50 allows classification as Mg-chromite. The ferric iron number (Fe³⁺# = 100Fe³⁺ / (Fe²⁺ + Fe³⁺); calculated from stoichiometry after Droop 1987) of spinel inclusions (median Fe³⁺# = 15.9, mean = 15.7 ± 7.4; Fig. 16) is lower than that of xenolith spinels (median = 18.8, mean = 19.6 ± 7.6).

Overall low Fe³⁺# in both inclusions and xenolith spinels is consistent with the moderately reducing conditions observed in the deep lithospheric mantle (Frost and McCammon 2008; Stagno et al. 2013; Luth et al. 2022). In view of the almost exclusively harzburgitic association of inclusion spinel, focusing on garnet harzburgites only, the Cr# (median = 83.1, mean = 82.8 ± 4.2) and Mg# (median = 63.9, mean = 64.2 ± 4.3) in xenolith spinel increases whilst the Fe³⁺# (median = 18.1, mean = 17.5 ± 6.9) decreases slightly, but still not to the levels observed for inclusions (Figs. 15 and 16). TiO₂ contents are generally similarly low in both inclusions (median = 0.09 wt%; no mean reported for strongly non-normal distribution) and xenolith spinels (median = 0.10 wt%) but elevated 75^th percentile values (Q3 = 0.22 wt% for inclusions and 0.33 wt% for xenolith spinels) indicate that moderate metasomatic reenrichment of Ti is relatively widespread.

The principal distinction between diamond substrates and the cratonic mantle roots as sampled by garnet peridotite xenoliths is the well-established, much higher proportion of harzburgite (-dunite) to lherzolite (-wehrlite) in the former (~85:15%; Gurney 1984) compared to the latter (18:82% in our xenolith database). Potentially dunitic mineral assemblages, indicated by CaO contents in garnet < 1.8 wt%, are common diamond substrates (~38%) but rarely documented in xenoliths (~2%). This latter discrepancy may be somewhat enhanced by sampling bias during xenolith collection, as garnet- and/or clinopyroxene-rich peridotites are more likely to be picked. Using mineral Mg# as an indicator of source depletion through melt extraction again documents the overall more depleted character of diamond substrates relative to the deep (garnet-facies) cratonic lithospheric mantle represented in the xenolith record. Olivine, because of its very high modal abundance in cratonic peridotites (72 vol% as mean and median value; Pearson et al. 2004), is the most robust proxy for this purpose, as its Mg# effectively becomes independent of temperature and instead equivalent to bulk rock Mg#. The observation that on the paragenesis level olivine inside and outside of diamond has statistically indistinguishable means in Mg#, therefore, has fundamental implications: (1) that the major element composition of inclusions is imposed by the substrate and not by the diamond forming medium and (2) that widespread Fe-rich metasomatism of the lithospheric mantle did not occur subsequent to diamond formation, which is focused in the Paleoarchean for harzburgitic substrates and the Paleoproterozoic–Mesoproterozoic for lherzolitic substrates. The latter conclusion precludes neither metasomatism by small melt fractions/fluids subsequent to diamond formation, as such events have limited impact on bulk rock Mg#, nor localized metasomatic shifts in Mg#.

A distinguishing feature between inclusions and xenolith minerals lies in the higher Cr content and Cr/Al ratio of garnet and Mg-chromite in diamond. The extensive melt depletion events that characterize the protoliths of subcratonic lithospheric mantle did not affect the Cr content of the melting residues but efficiently removed Al (from primitive mantle to average cratonic harzburgite, Cr contents remain constant but about 90% of Al is removed; e.g., McDonough and Rudnick 1998, their Table 1). Higher Cr contents for inclusions are not limited to garnets with very low Ca intercept values and high Mg# (in fact, the highest Cr contents are associated with harzburgitic garnets with relatively low Mg# of 84–86) but occur equally for mildly subcalcic and even for lherzolitic garnets. This suggests that the almost exclusive restriction of Cr₂O₃ contents greater than 13 wt% to inclusion garnets is not a consequence of higher degrees of primary melt depletion being restricted to diamond substrates.

Nevertheless, the systematic offset to higher Cr contents, as best seen for spinel inclusions, may, at least partially, represent a source signal. Cr is not usually considered as a mobile element during mantle metasomatism but high pressure (4–6 GPa) experiments indicate an order of magnitude increase in the solubility of Cr in hydrous fluids associated with increased salinity (addition of 1.3–3 wt% of KCl; Klein-BenDavid et al. 2011). These authors consider that during percolation of such saline high-pressure fluids, Cr could dissolve from the peridotitic wall rocks without significantly affecting their high Cr concentrations. The ability of such saline fluids to impose their high-Cr characteristics on garnet inclusions in diamond, however, is likely limited, given that their mean Cr# (100Cr/(Cr+Al)) is only 7.5 (high-density fluid compilation of Weiss et al. 2022) compared to Cr# of 38–65 for the unusually Cr₂O₃-rich (> 13 wt%) garnet inclusions discussed here. In contrast, enrichment of Al in cratonic peridotites is frequently observed and possibly relates to fluids derived from recycled sediments (e.g., Rapp et al. 2017). Consequently, widespread but relatively mild reenrichment in Al may have decreased Cr/Al ratios and increased modal garnet in cratonic peridotites sampled as xenoliths, with this effect being strongest for rocks with extreme primary Al depletion (very high Cr/Al ratios). This type of metasomatism may be responsible for both the lower average and maximum Cr contents in xenolith-hosted garnet and spinel; it may have affected diamond substrates as well, just to a lesser degree.

An alternate explanation relates to the pressure and temperature dependence of the distribution of Cr between garnet and spinel (Doroshev et al. 1997; Grütter et al. 2006). Experiments demonstrated inclusion-like high Cr/Al ratios for coexisting Cr-pyrope and Crspinel in harzburgite at high pressures and temperatures (> 5 GPa and > 1200 °C; Malinovsky and Doroshev 1975; Girnis and Brey 1999). High Cr/Al inclusion compositions are accordingly interpreted as indicative of some diamond growth occurring over a wide range of temperatures, elevated above a cratonic geotherm during high-temperature thermal perturbations (Girnis et al. 1999; Grütter 2006).

The most sensitive indicators of metasomatism on the major-minor element level are Na, Ti and, for clinopyroxene, K. In orthopyroxene, clinopyroxene and garnet, metasomatic enrichment of Na and Ti in both inclusions and xenoliths is most prominent in the lherzolitic paragenesis and very intense Ti-rich metasomatism is almost entirely restricted to lherzolite xenoliths. Intense Ti metasomatism (leading to > 0.5 wt% TiO₂ in garnet) is restricted to temperatures of last equilibration in excess of 1130 °C, i.e., occurs above the hydrous solidus of lherzolite along typical cratonic geotherms (e.g., Wyllie and Ryabchikov 2000). Since equilibration temperatures of > 1130 °C are frequently observed for inclusions, the almost complete absence of evidence of intense Ti-metasomatism in inclusions (mild Ti-metasomatism is more frequent) likely relates to either a diamond unfriendly character of such metasomatism or an increase in Ti-metasomatic intensity or frequency after the formation of most peridotitic diamonds in the Archean-Mesoproterozoic. The opposite is true for potassic metasomatism documented in clinopyroxene, where only fairly moderate levels of enrichment are seen in xenoliths (K₂O ≤ 0.26 wt%) whilst clinopyroxene inclusions show a tail in their frequency distribution of up to 1.68 wt% K₂O. This type of metasomatism may relate to potassic brines that are frequently observed as high-density fluid inclusions in fibrous diamonds (Izraeli et al. 2001; Weiss et al. 2022). That evidence for such metasomatism is best preserved inside diamond may relate to reactivation of potassic mantle metasomes and migration to shallower mantle in the course of transient heating events (e.g., Waters and Erlank 1988; Tappe et al. 2008). The overall low abundance of strongly potassic clinopyroxene inclusions suggests that potassic brines may be one medium of diamond formation in peridotitic substrates, but not the dominant one.

To our knowledge, the first REE pattern for peridotitic garnet was obtained by Philpotts et al. (1972), via isotope dilution thermal ionization mass spectrometry on separated minerals from a phlogopite garnet lherzolite xenolith (GSFC #187) from Roberts Victor (Kaapvaal Craton). From today’s perspective, the reported humped REE_N pattern with a prominent peak at Dy would be considered anomalous.

The two common types of REE_N patterns of garnet from cratonic peridotites, normal and sinuous or sinusoidal, were first reported by Shimizu (1975): (1) garnets from three sheared (mosaic-porphyroclastic texture) lherzolite xenoliths from kimberlites in Lesotho showed steep positive slopes in the LREE_N and fairly flat MREE–HREE at ≥10 times chondritic abundance, similar to garnets from eclogite xenoliths and megacrysts (hence the term “normal” patterns), (2) garnets from three granular (coarse texture) lherzolite xenoliths from Lesotho and the Kimberley area had sinusoidal patterns, with a prominent peak at Sm_N–Eu_N and an upward kink in slope near Er_N. For inclusions in diamond, similar sinusoidal REE_N pattern were subsequently reported by Shimizu and Richardson (1987) for harzburgitic garnets in diamonds from Kimberley and Finsch, while normal REE_N patterns were first observed for lherzolitic garnet inclusions from Akwatia, Ghana (Stachel and Harris 1997).

REE_N patterns of garnet inclusions of harzburgitic and lherzolitic paragenesis. Since the pioneering work of Shimizu and Richardson (1987), the number of REE analyses of garnet inclusions has risen considerably as ion microprobes enjoyed wider use, with 223 harzburgitic and 72 lherzolitic garnets in our database (see Fig. 17). These analyses clearly establish sinusoidal REE_N patterns as the hallmark of both harzburgitic and lherzolitic inclusion garnets. On average, the degree of sinuosity, however, is more pronounced for harzburgitic garnets (see median, 10^th and 90^th quartile patterns in Fig. 17).

In detail, the range of REE_N patterns displayed by peridotitic garnet inclusions is much more complex than apparent from Figure 17. Based on overall shape and the presence and position of a peak in the LREE_N–MREE_N, we identified 10 types of REE_N patterns, which due to uncertainties in the classification of garnet analyses that do not include the LREE praseodymium (Pr) were assigned into 12 classes. The REE_N patterns for these classes are shown in Appendix Figure 1 and the frequency of their occurrence for garnets of the harzburgitic and lherzolitic paragenesis is shown in Figure 18. Patterns with peaks at Ce, Pr, Nd, Sm and Eu are all sinusoidal. Non-sinusoidal patterns are either V-shaped with a peak at La, show steep positive or nearly flat slopes from LREE_N to HREE_N, or classify as normal (rise in LREE_N, flat and superchondritic MREE–HREE). Abnormal patterns fit none of these categories and typically are “noisy” with multiple peaks. For harzburgitic garnets, sinusoidal patterns peaking at Nd are clearly dominant (Fig. 18). Including patterns with a peak at Nd with Pr not analyzed, 49% of harzburgitic garnets peak at Nd. By contrast, 33% of lherzolitic garnets peak at Sm and only 21% at Nd (27% including the peaks at Nd with Pr missing) and 15% of lherzolitic garnets show normal patterns, as opposed to 1% of harzburgitic garnets (Fig. 18). The transition between the two most important REE_N patterns, peak at Nd (including samples with Pr missing) to peak at Sm, is associated with a decrease in the degree of sinuosity, measured as the Nd_N/Y_N ratio (MREE_N slope) in Figure 19.

Other trace elements. Other commonly and reliably analyzed trace elements in garnet inclusions are Sr, Nb, Ti, Hf, Zr and Y (in sequence of increasing compatibility; Fig. 20). Whilst Sr concentrations are higher in harzburgitic garnets (median value of 2.4 ppm, compared to 0.67 ppm for lherzolitic garnets; likely controlled by the absence/presence of clinopyroxene, see Stachel et al. 2004), the patterns for the remaining elements are nearly parallel but with higher concentrations in lherzolitic garnets (Fig. 20). Both parageneses show distinct negative anomalies for Sr and Ti (subchondritic median values), flat and typically superchondritic Nb, Hf and Zr, and slightly depressed Y_N (with a subchondritic median for harzburgitic garnets).

Correlations with major elements.Griffin et al. (1999a) highlighted the usefulness of garnet Cr# as a measure for the degree of chemical depletion of cratonic peridotites. In addition to possible metasomatic modification (e.g., Burgess and Harte 1999; Grütter and Menzies 2017), the interpretation of garnet Cr# is, however, complicated by the pressure-controlled exchange of Cr and Al between spinel and garnet (in spinel–garnet peridotites, garnet Cr# is a geobarometer; Grütter et al. 2006). In garnet, increasing Cr# is accompanied by broad trends of enrichment in strictly incompatible trace elements such as Sr and the LREE (Fig. 21).

A more unambiguous indicator in garnet of the degree of primary melt extraction from its source protolith is its position relative to the lherzolitic trend in a Cr/Ca plot, which can be expressed as the Ca intercept value (Grütter et al. 2004; see above) or more simply but less accurately, as the Ca# (molar 100Ca/(Ca+Mg+Fe)). In addition to primary melt depletion, garnet Ca intercept values (or Ca#) also track secondary metasomatic modification of the diamond substrates. Figure 22 shows a positive correlation between the Yb content and Ca# of garnet. To interpret this correlation as a straightforward geochemical trend may, however, be misleading as crystal chemical controls have to be considered as well. For eclogites, increasing partitioning of REE into garnet with increasing Ca content was recognized by Harte and Kirkley (1997). In a study of trace element partitioning between mantle minerals and silico-carbonate melts, Girnis et al. (2013) confirmed these findings and observed a strong increase in garnet–melt partition coefficients for REE with Ca#. This effect renders HREE like Yb and Lu incompatible in strongly subcalcic garnets (Girnis et al. 2013). Figure 22 shows a predicted trend for Yb in garnet, based purely on the crystal chemical effect imposed by varying the Ca# number from 15 (approximately equivalent to the lherzolite/harzburgite boundary) to 0. The predicted effect is over an order of magnitude stronger than the observed correlation in the inclusion data set, suggesting that either the crystal chemical effect was overestimated by Girnis et al. (2013) or that other effects (e.g., bulk rock chemistry) partially compensate the crystal chemically predicted strong HREE depletion in low-Ca garnets.

Comparison with garnet from cratonic peridotites. In cratonic peridotites, harzburgitic garnets inside and outside of diamond share overall similar, sinusoidal REE_N patterns (Fig. 17). On average, xenolith garnets, however, have higher MREE and HREE, resulting in a shift of the peak from Nd (for inclusion garnets) to Sm for the median REE_N pattern. Sr contents in xenolith garnets tend to be lower and Zr and Y contents higher than in inclusions garnets (Fig. 20). Distinct Nd and Sr isotope composition of harzburgitic garnets included in diamond and from concentrate were first reported by Richardson et al. (1984) and are discussed in detail by Smit et al. (2022).

For lherzolitic garnets, differences are even more pronounced: xenolith garnets have significantly higher contents in the MREE–HREE from Gd onwards, resulting in normal REE_N patterns for median and 25^th and 75^th quartile compositions, as opposed to sinusoidal patterns for the median and interquartile range of lherzolitic garnet inclusions in diamond (Fig. 17). Lherzolitic xenolith garnets generally also lack the pronounced negative Ti_N anomalies characteristic for inclusion garnets and on average have higher Y contents (Fig. 20).

Shimizu (1975) not only defined the two principal types of REE_N patterns of peridotitic garnet but also characterized the REE_N patterns of associated clinopyroxene: (1) clinopyroxene occurring together with garnet with normal REE_N in the three studied sheared lherzolite xenoliths showed mildly fractionated REE_N patterns with about 10× chondritic LREE and chondritic to mildly subchondritic HREE, similar to clinopyroxene in eclogite xenoliths and megacrysts. (2) Clinopyroxene in equilibrium with garnet with sinusoidal REE_N in three coarse lherzolites was found to have much more fractionated LREE–HREE, falling from ~20–30 chondritic LREE to about 0.4× chondritic HREE.

Types of REE_N patterns. The two groupings introduced by Shimizu (1975) largely hold for peridotitic clinopyroxene inclusions in diamond. For inclusions, the mildly fractionated group is characterized by humped patterns peaking at Nd_N (Fig. 23), with La_N/Sm_N between 0.3 and 2.5 with a median value of 0.6. Two clinopyroxenes associated with the humped group show modified patterns, which by comparison with the humped patterns show enrichment of La and Ce in one case and enriched LREE and a trough in MREE_N in the other. In slight contrast to the xenolith data of Shimizu (1975), the key distinguishing criterion of the second group are steep negative LREE_N slopes (steep-LREE_N) and not necessarily a high LREE–HREE ratio (Fig. 23). La_N/Sm_N ranges between 4.6 and 87 with a median value of 22.

Clinopyroxene inclusions with humped REE_N patterns coexist with garnets with normal patterns (n = 2 for inclusion database), characterized by steep positive slopes in the LREE_N. This coexistence matches observations made for peridotite xenoliths beyond the limited sample set of Shimizu (1975), with, e.g., the occurrence of clinopyroxene with humped patterns together with garnet with normal patterns being characteristic for the Type 1 garnet lherzolites of Grégoire et al. (2003). Clinopyroxene inclusions with steep LREE_N coexist with garnets with sinusoidal or positively sloped REE_N (n = 4), the latter characterized by a less positive slope in the LREE compared to garnets with normal REE_N patterns. Clinopyroxene with steep LREE_N (but with a peak at Ce) and garnets with sinusoidal or positively sloped REE_N patterns also characterize the Type 2 garnet lherzolites of Grégoire et al. (2003).

In major/minor element composition, clinopyroxenes with humped REE_N patterns have lower Mg# (87.9–93.1, median of 90.6) and higher TiO₂ contents (0.04–0.23 wt%, median of 0.10 wt%) than clinopyroxenes with steep-LREE_N (Mg# 91.5–94.8, median of 92.8; TiO₂ < 0.02–0.05 wt%, median of 0.03).

Olivine is the last of the three peridotitic inclusion minerals that have received attention for trace element studies. Following up on an initial study (De Hoog et al. 2010) on trace elements in olivine from peridotite xenoliths and their application to thermobarometry, De Hoog et al. (2019) analyzed 10 trace elements in 36 olivine inclusions in diamonds from Akwatia, Ghana with an ion microprobe. They found that the contents of Ca and Na and their ratios with Al can be used in combination to constrain the paragenesis (harzburgitic versus lherzolitic) of peridotitic olivine inclusions, while Ti in olivine may be employed as a tracer of metasomatic re-enrichment of diamond substrates. Based on new reference materials and recent advances in the analysis of trace elements in olivine via LA-ICPMS (Bussweiler et al. 2019), we expect a large increase in trace element data for olivine inclusions over the coming years. Only such an expanded data set will allow us to conclusively assess whether the paragenetic fields defined by De Hoog et al. (2019) on the basis of Akwatia inclusions can be applied universally.

Normalization to primitive (J4) garnet and cpx. To derive the trace element characteristics of the rock a mineral derives from, normalization to CI-chondrite or primitive mantle is of limited use, as the derived patterns are strongly determined by the degree of compatibility of the studied trace elements in the available lattice sites. To overcome this problem, Stachel et al. (1998) instead suggested normalization to minerals that are well equilibrated in a bulk rock with a primitive mantle-like composition and chose high-temperature sheared garnet-lherzolite xenolith J4 from Jagersfontein (Jagoutz and Spettel, unpublished data; Wolff-Boenisch 1994). Since then, other samples with near-primitive compositions have been recognized (e.g., Vitim xenolith 313-105 of Ionov et al. 2005) but despite offering a more complete range of analyzed trace elements, their bulk rock REE patterns are neither a perfect match for primitive nor for depleted MORB-source mantle. For that reason, we carefully re-analyzed mineral separates from J4 using laser ablation sector field ICPMS (Appendix Table 2) and continue the use of J4 garnet and clinopyroxene for normalization purposes.

For simplicity, for J4-normalization we only use the median patterns of the principal REE_N pattern types (see Fig. 18 and Appendix Fig. 1): harzburgitic garnet inclusions peaking at Nd, lherzolitic garnet inclusions peaking at Sm, and lherzolitic garnet inclusions with normal REE_N patterns. Normalization to primitive garnet reveals that harzburgitic garnets with strongly sinusoidal patterns reflect sources with REE contents steadily decreasing from mildly enriched LREE through the MREE to a trough in the HREE at Er, with increasingly less-depleted HREE from Er to Lu, overall resulting in a V-shaped REE_J4 pattern (Fig. 24). On the other end of the spectrum, normal REE_N patterns of lherzolitic garnet inclusions, which are dominant among garnets from lherzolite xenoliths (Fig. 17), represent sources with near-primitive trace element compositions similar to J4 (Fig. 24). Lherzolitic garnet inclusions with REE_N patterns peaking at Sm fall in between these two endmembers, with near primitive LREE concentrations, a less prominent drop towards Er_J4 than for harzburgitic garnets, and a shallower increase towards Lu_J4. Given the higher distribution coefficients for LREE in clinopyroxene relative to garnet (e.g., Johnson 1998), minor relative depletion of LREE in lherzolitic relative to harzburgitic garnets may partially relate to metamorphic equilibration with clinopyroxene for the lherzolitic paragenesis.

For lherzolitic substrates, the median values of the two principal pattern types in clinopyroxenes, humped and steep-LREE_N, may be used to assess the validity of the garnet-based findings. HREE concentrations in clinopyroxene inclusions in diamond generally are low (in part below detection), resulting in noisy median HREE patterns. In addition, Yb analyses in clinopyroxene via SIMS appear to frequently suffer from an analytical problem (possibly an unrecognized interference) and hence are not reported in a number of studies. This drawback relative to garnet inclusions is compensated by higher and therefore more accurately determined LREE concentrations (e.g., La is approximately 1–100 × chondritic in abundance, compared to 0.03–15 × chondritic for garnet inclusions). Normalized to J4-clinopyroxene, inclusions with humped REE_N patterns also reveal a near-primitive composition (Fig. 24). This agrees well with the observation that such clinopyroxenes occur together with garnets with normal REE_N patterns. Clinopyroxene with steep LREE_N patterns shows enrichment in LREE and steep negative slopes towards the MREE, when normalized to J4 (Fig. 24). Overall, this fits with the observation made for lherzolitic garnets with sinusoidal patterns peaking at Sm but suggests a more continuous negative slope from La–Sm in the diamond substrates than indicated by these garnets. Due to the problems outlined above, the HREE patterns of clinopyroxenes are considered less informative than those of garnet and not further discussed here.

Considering the well-established origin of cratonic peridotites as residues of intense melt extraction events in the Archean (see above), the LREE-enriched character of peridotitic diamond substrates appears counterintuitive. Based on REE patterns of unmodified MORB residues (e.g., Sharma and Wasserburg 1996), the positive slope within the HREE_J4 from Tm to Lu, observed for garnets with chondrite-normalized patterns peaking at Nd and Sm (Fig. 24), fits the steep positive LREE_N–HREE_N slopes in such residues. The common negative slopes from La_J4–Er_J4 then are a consequence of metasomatic re-enrichment in incompatible trace elements, first recognized as a universal process in lithospheric peridotites by Frey and Green (1974). From harzburgitic garnets, through lherzolitic garnets with sinusoidal REE_N and associated clinopyroxenes with steep-LREE_N patterns, to lherzolitic garnets with normal REE_N and associated clinopyroxenes with humped REE_N patterns, the signature of primary depletion becomes increasingly obliterated and eventually completely erased in the J4-normalized mineral trace elements patterns (and hence the bulk rocks). The decrease in Mg# and increase in TiO₂ for the transition from clinopyroxenes with steep-LREE_N to humped patterns implies that this metasomatic progression is not limited to trace elements but falls in the categories of “enrichment in major–minor–trace elements without modal change” or “modal metasomatism” of Harte (1983, 1987).

Fluid versus melt metasomatism. Having established (1) that garnets with sinusoidal REE_N patterns, when normalized to a primitive garnet composition, document preferential reenrichment in LREE and (2) that the progression from sinusoidal towards normal REE_N patterns (or from V-shaped to primitive J4-normalized patterns) increasingly also involves addition of moderately incompatible trace elements and major/minor elements, raises the question what metasomatic agent(s) drive this re-enrichment process. Trace element signatures, in particular ratios of strongly to mildly incompatible elements (LREE/HREE, Zr/Y, Sm/Hf), have long been used to make a first order division between fluid- and melt-driven metasomatism (Stosch and Lugmair 1986; Shimizu and Richardson 1987; Griffin and Ryan 1995; Stachel et al. 1998; Viljoen et al. 2014). Griffin and Ryan (1995) empirically developed a widely used diagram of Y versus Zr (Fig. 25), with vectors for low-temperature phlogopite (fluid) metasomatism (high Zr/Y) and high-temperature melt metasomatism (intermediate Zr/Y). These two trends emerge from a field for “depleted” garnets near the origin, where a distinction is no longer possible (approximately at Y < 5 ppm and Zr < 20 ppm).

For inclusion garnets exceeding this threshold, harzburgitic garnets mostly fall onto or close to the fluid metasomatic trend whilst lherzolitic garnets generally have lower Zr/Y, compatible with melt metasomatic re-enrichment of their substrates. A notable exception is the large population of lherzolitic garnet inclusions from the Victor Mine in Canada, which mainly displays Zr/Y ratios indicative of fluid metasomatism or intermediate between fluid and melt metasomatism (Fig. 25). Excluding Victor garnets, this overall distribution was linked (Stachel and Harris 1997; Stachel and Luth 2015) to harzburgitic diamond substrates mostly residing below their hydrous solidus (thus permitting percolation of fluids only), while lherzolitic diamonds form generally above the hydrous (± carbonated) solidus (allowing for the percolation of melt). Geothermobarometric data for lherzolitic inclusions in diamonds from Victor indicate their derivation from just below the hydrous lherzolite solidus, consistent with their comparatively high Zr/Y.

For the study of garnets included in diamond, however, Y and Zr generally are not sensitive enough as tracers of metasomatism and 70% of garnets fall below the threshold of Y < 5 and Zr < 20 ppm. This is in stark contrast to cratonic peridotite xenoliths, where only 29% of garnets fall below the Y and Zr threshold. This much higher proportion among inclusions of garnets below the threshold again highlights the overall much more depleted nature of peridotitic diamond substrates compared to the xenolith and xenocryst sample. Despite their low Y and Zr, garnets below the threshold nevertheless are almost invariably affected by metasomatic re-enrichment. Garnets reflecting the intense primary melt depletion characteristic for cratonic peridotites should have trace element patterns with LREE_N/MREE_N ≪ 1 (see above), a signature that is virtually absent among inclusion garnets even below the Y and Zr threshold. Of the inclusion garnets below the threshold, only 7% have Nd_N/Y_N < 1, and one sample (0.5%) has Nd_N/Y_N < 0.1. Xenolith garnets below the threshold show a very similar picture, with only 11% having Nd_N/Y_N < 1 and 0.7% < 0.1. With over 2/3 of garnet inclusions being depleted below the threshold in Y–Zr space, it is evident that diamond formation usually is associated with mild metasomatic events, reflecting low fluid–rock or melt–rock ratios. Note that low fluid/melt–rock ratios do not imply diamond formation under fluid-limited conditions: protracted diamond growth involving multiple metasomatic pulses evidently involves sufficiently high ratios of fluid/melt–diamond to preclude widespread Rayleigh fractionation of stable isotopes and nitrogen in diamond (Howell et al. 2020; Stachel et al. 2022).

On the level of mineral inclusions, the effects of multiple metasomatic events cannot be separated using only trace elements but require radiogenic isotope approaches. For example, based on unsupported (low Rb) highly radiogenic Sr in harzburgitic garnet inclusions, Richardson et al. (1984) proposed that a melt metasomatic event affected the harzburgitic diamond substrates beneath the Kimberley area of the Western Kaapvaal Craton within 300 m.y. before diamond formation (at about 3.3–3.2 Ga).

Fluids as metasomatic agents. The concept of fluid-driven metasomatism of lithospheric mantle was widely popularized by Bailey (1982) and two key arguments support the involvement of supercritical CHO fluids in mantle metasomatism, particularly of harzburgitic substrates: (1) As discussed above, extremely high ratios of strongly to moderately incompatible trace elements and a negligible metasomatic effect on both moderately incompatible trace elements (Ti, Y, HREE) and major elements (e.g., Mg#, Ca, Al and Cr) clearly are not compatible with melt involvement. (2) The average cratonic model geotherm (40 mW/m²; Hasterok and Chapman 2011) intersects the carbonated, hydrous solidus of harzburgite (Wyllie 1987) at ~1300 °C. This implies that with the exception of the deepest regions (≳180 km) located near the lithosphere–asthenosphere boundary, harzburgite in the lithospheric mantle is impermeable to melt percolation, as migration along grain boundaries causes both thermal and compositional equilibration, which inevitably leads to freezing of melts in subsolidus substrates.

The range in composition of possible metasomatic fluids can be considerably narrowed by examining a few basic constraints:

1. The fluid must be stable in equilibrium with peridotite. This constraint is not met for CO₂-rich fluids, which may occur only on the oxidizing side of the EMOD/G (enstatite + magnesite = olivine + diamond/graphite; Eggler and Baker 1982) buffer. Along cratonic geotherms throughout the entire peridotitic lithospheric mantle, CO₂ is buffered through carbonation reactions involving olivine ± pyroxene (Wyllie and Huang 1976; Brey et al. 1983). Depending on the redox conditions, H₂O-rich and CH₄-rich fluids can both be stable in equilibrium with peridotite. Excluding a subgroup of unusually oxidized samples from the Slave craton, the xenolith record for the depth range 140–180 km has a main range in ΔlogfO₂ (FMQ), i.e., the oxygen fugacity in log unit difference relative to the FMQ buffer, of ~ −3 to −1.5 (see, e.g., Luth and Stachel 2014). For the average conditions of diamond formation (~1140 °C/5 GPa, based on garnet–olivine thermometry; Stachel and Luth 2015), this fO₂ range implies that CHO fluids are at the water maximum, with water contents ranging between 85 and 98 mol% (calculated using GFluids of Zhang and Duan 2010; see Fig. 1 of Stachel et al. 2017). Matjuschkin et al. (2019) showed that on the reducing side of the water maximum (lower end of the main range considered here), GFluids may significantly underestimate CH₄ contents but this does not affect the principal conclusion that the xenolith record over the 140–180 km depth range is centered at fO₂ conditions corresponding approximately to the water maximum.

2. The fluid must be able to percolate through peridotite at low fluid:rock ratios. High dihedral angles for both CO₂–olivine and CH₄–olivine (Watson et al. 1990; Wyllie 1992; Ferrando et al. 2017) preclude efficient percolation of such fluids through the lithospheric mantle. The dihedral angle for aqueous fluid–olivine decreases with increasing pressure (Watson et al. 1990), permitting efficient percolation at pressures ≥ 3 GPa (Mibe et al. 1998). Added salinity decreases dihedral angles of aqueous fluids with olivine even further (Huang et al. 2019).

3. The fluid must dissolve and transport the chemical components associated with mantle metasomatism. Low solubility of trace elements such as REE is well documented for CO₂-rich fluids under mantle conditions (e.g., Meen et al. 1989). An experimental study of trace element solubility in CH₄-rich fluids is still outstanding but given the non-polar character of the symmetrical CH₄ tetrahedron, CH₄ fluids are expected to be poor solvents. In contrast, the solubility of solids in aqueous fluids increases strongly with pressure and is on the wt% level at mantle conditions (e.g., Ayers et al. 1997). At diamond stable conditions (≥ 4 GPa), trace element partitioning between peridotite minerals and aqueous fluids is similar to partitioning with silicate melts, with strong LREE/HREE fractionation (Fig. 26), and about 36 wt% (15 mol%) dissolved solids in the fluid at subsolidus conditions (1000 °C, 5–6 GPa; Kessel et al. 2015).

In conclusion, (1) the redox state recorded in mantle xenoliths from the main portion of the lithospheric diamond stability field falls at or near the water maximum, (2) H₂O is the only CHO fluid species with a sufficiently low dihedral angle with olivine to readily percolate along grain boundaries and (3) aqueous fluids also are the only CHO species with the ability to dissolve incompatible trace elements in the required quantities. The high load of dissolved solids carried by aqueous fluids at the pressure–temperature conditions inside the diamond stability field justifies using the term “high-density fluid” (HDF; Klein-BenDavid et al. 2007; Weiss et al. 2022) instead. Research on primary fluid inclusions in fibrous, clouded and coated diamonds indicates that for peridotitic substrates, hydrous HDFs generally range from potassium-rich brines to Mg-carbonatitic, while in eclogitic substrates an evolution from saline through low-Mg carbonatitic to silicic HDFs is observed (Weiss et al. 2015, 2022). The carbonatitic and silicic inclusions are transitional between hydrous HDFs and hydrous melts. For example, inclusions representing an original carbonate melt with molar 100H₂O / (H₂O + CO₂) of only 5–20, were recognized in cuboid diamonds of unknown paragenesis from Udachnaya (Zedgenizov et al. 2004). Fluid inclusion-rich diamonds are often interpreted as having grown at high levels of carbon supersaturation in the HDF, to be genetically linked to kimberlite or proto-kimberlite magmatism, and to be typically of similar age as the transporting host kimberlite (e.g., Boyd et al. 1987; Gurney et al. 2010; Weiss et al. 2011). On that basis, the HDFs observed in clouded and fibrous diamonds may represent a growth medium that is distinct from that for generally much older (Gurney et al. 2010; Smit et al. 2022) smooth-faced monocrystalline diamonds. The discovery of rare but compositionally highly similar, saline to carbonatitic hydrous HDFs as inclusions in gem-quality monocrystalline diamonds from Kankan, Finsch, Voorspoed and Venetia (Weiss et al. 2014; Jablon and Navon 2016), however, clearly established a much wider role of HDFs in diamond formation, both with respect to diamond types and time. We emphasize, however, that the presence of HDFs as inclusions in some smooth-faced monocrystalline diamonds does not automatically imply that all diamonds formed from HDFs or that the types of HDFs observed in diamond are the principal agents of mantle metasomatism. These conjectures will have to be validated or disproven through future research and will be discussed further below.

Melts as metasomatic agents. Hydrous silicate, silico-carbonate and carbonate melt compositions have all been suggested as agents of mantle metasomatism, based on studies of very distinct aspects of subcratonic lithospheric mantle:

1. A group of peridotite xenoliths with fertile compositions—often displaying “sheared” textures, equilibration temperatures that exceed the local conductive geotherm, and evidence for rim-ward enrichment in Fe, Ti, Zr, Y and Ga in constituent minerals—document re-fertilization in the course of (likely repeated) infiltration of melts. Implicated melt compositions range from proto-kimberlitic (megacryst magma) through primitive OIB-like asthenospheric melts and alkali picrites to komatiitic (Gurney and Harte 1980; Nixon et al. 1981; Hervig et al. 1986; Smith 1988; Griffin et al. 1989b; Harte et al. 1993; Mahotkin et al. 2000).

2. A subset of peridotite xenoliths that are enriched in orthopyroxene from, e.g., the Kaapvaal and Siberian cratons (Boyd 1989; Boyd et al. 1997) is interpreted as evidence for Si-enrichment through melts (Rudnick et al. 1994; Kelemen et al. 1998). Based on the narrow, mantle-like δ¹⁸O record of olivine from cratonic peridotites, including those with an opx-enriched signature, Si-enrichment through dacitic (trondhjemitic–tonalitic–granodioritic) slab melts (Rapp and Watson 1995) can be excluded, as it would inevitably lead to ¹⁸O enriched signatures (Regier et al. 2018). Instead, Si-enrichment in cratonic peridotites may have occurred by melts generated through hydrous melting of peridotite (Mitchell and Grove 2015); reaction with basic melts, which initially dissolve clinopyroxene and, migrating further into the wall rock, precipitate orthopyroxene (Kelemen et al. 1992); or reaction with komatiite at high pressure (> 4 GPa; Tomlinson and Kamber 2021).

3. A comparative study of diamond suites from areas with normal (Finsch Mine) and anomalously low (Premier kimberlite at Cullinan Mine) seismic P-wave velocities in the diamond-stable lithospheric mantle beneath the Kaapvaal Craton allowed the identification of contrasting styles of metasomatic re-enrichment (Viljoen et al. 2014; see also Shirey et al. 2002). While harzburgitic diamond substrates at Finsch document low-temperature fluid style metasomatism, the peridotitic diamond sample at Premier shows a high proportion of the lherzolitic paragenesis among peridotitic diamonds, with evidence of high temperature melt metasomatism introducing Ca, Fe, Ti, Zr, Y and REE (Viljoen et al. 2014). The refertilization in major and trace elements of the peridotitic diamond substrates beneath Premier and the P-wave anomaly linked to the associated changes in mineralogy (conversion of harzburgite to lherzolite) is attributed to the formation of the 2 Ga Bushveld large igneous province (Shirey et al. 2002; Viljoen et al. 2014) and thus an example of extensive silicate melt metasomatism associated with a mantle plume. Another example of intense modification of diamond substrates associated with plume magmatism is the 2.7 Ga Ventersdorp large igneous province and the inclusions in diamonds from the proximal Voorspoed Mine, which document the partial to complete conversion of depleted peridotite to websterite (Viljoen et al. 2018).

4. Over a wide range of pressures in the lithospheric and sublithospheric mantle, incipient melting of carbonated peridotites yields magnesio-carbonatites (e.g., Dalton and Presnall 1998). Their low density, high mobility, and high contents in incompatible trace elements make carbonate melts highly efficient agents of mantle metasomatism. An important feature of carbonate metasomatism is a minimal impact on the major element composition of affected substrates, typically with the exception of Ca only (Rudnick et al. 1993). Characteristic geochemical signatures of carbonatite metasomatism are high LREE/HREE, Nb/La, Ca/Al and Zr/Hf ratios and low Ti/Eu ratios (Rudnick et al. 1993; Yaxley 1993). Metasomatism through sodic magnesiocarbonatite, derived from melting of carbonated pargasite peridotite, was identified, e.g., in spinel peridotites from Mt Leura, Victoria, and may lead to conversion of spinel lherzolite to wehrlite at pressure < 2.1 GPa (Green and Wallace 1988). Sodic magnesio-carbonate metasomatism of peridotites is, however, limited to a narrow pressure range (2.1–3.1 GPa) bracketed by the breakdown of carbonate at lower pressures and the breakdown of pargasite at higher pressures.

In diamond stable lithospheric mantle, fO₂ conditions generally are on the reducing side of the EMOD buffer (see Stagno et al. 2013; Luth and Stachel 2014), limiting the in situ generation of carbonated melts to localized domains with elevated redox conditions, as observed for some samples from the Central Slave Craton (Creighton et al. 2010). In the absence of such carbonated domains, carbonate metasomatism of the deep lithospheric mantle must be caused by melts derived from below, as inferred, e.g., for the metasomatic modification of highly depleted peridotitic lithosphere beneath the Gregory Rift Valley in northern Tanzania (Rudnick et al. 1993). Asthenosphere-derived carbonatites may originate from upwelling metasomatized mantle that experiences redox melting via oxidation of diamond and reduction of Fe³⁺ when reaching depths < 250 km (Rohrbach and Schmidt 2011). In present day Earth, subducting oceanic crust only exceeds the anhydrous solidus of carbonated eclogite at depth > 400 km (i.e., ~ transition zone depth; Dasgupta et al. 2004; Thomson et al. 2016) but would have melted at increasingly shallower depth with the mantle potential temperature increasing towards the Archean. At some intermediate time, this may have allowed for carbonatite metasomatism of the lithospheric mantle directly from melting of carbonated eclogites and possibly sediments. Depending on the exact estimate for the temporal evolution of mantle potential temperature, Archean and Paleoproterozoic subduction zones may have been sufficiently hot to cause shallow carbonate melting and direct recycling back to the surface (Dasgupta and Hirschmann 2010), precluding subduction-driven mantle metasomatism through carbonated melts. Observational evidence to the contrary was, however, produced by Shu and Brey (2015), who employed subcalcic garnets in peridotite xenoliths and included in diamonds to study Archean metasomatism in the Kaapvaal lithospheric mantle. Based principally on Ti/Eu and Zr/Hf characteristics, they identified carbonate and carbonate–silicate (kimberlite) melts as the principal agents of metasomatism and associated diamond growth in harzburgitic substrates.

Ti/Eu and Zr/Hf systematics of metasomatic agents. The ratios of Ti/Eu and Zr/Hf of whole rocks, garnet and clinopyroxene are used to trace metasomatic interactions and to constrain the nature of metasomatic agents (Rudnick et al. 1993; Yaxley 1993; Shu and Brey 2015; Uenver-Thiele et al. 2017). Directly comparing the Ti/Eu and Zr/Hf systematics of garnets with those of potential metasomatic agents requires that garnet does not strongly fractionate these ratios during metasomatic equilibration. Experimentally determined partitioning data between garnet and dry basaltic melt (Johnson 1998), strongly hydrous basanite (18–27 wt% H₂O; Green et al. 2000) and hydrous fluid (Stalder et al. 1998) show that garnet faithfully reflects the Zr/Hf ratio of the melt/fluid. The same applies to strongly subcalcic garnet (Ca# = 100Ca / (Ca + Mg + Fe) ≤ 5, approximately equivalent to CaO ≤ 2 wt%) in equilibrium with silico-carbonate (“kimberlite”) melts, while more calcic lherzolitic garnets (average Ca# ~15) have D_Zr/D_Hf of ~ 0.6, i.e., Zr/Hf ratios 40% lower than the coexisting melt (Girnis et al. 2013). Garnet in equilibrium with carbonate melt at high pressure (6.6–8.6 GPa; Dasgupta et al. 2009) exhibits an offset to 25% lower ratios (D_Zr/D_Hf of ~ 0.75). With the exception of lherzolitic garnets in equilibrium with kimberlite melt, garnet Zr/Hf ratios thus are a good match for the fluid/melt they last equilibrated with. More limited partitioning data for Ti and Eu indicate that garnet in equilibrium with dry and hydrous (4.8 wt% H₂O) basalt will have approximately 30–40% lower Ti/Eu (Johnson 1998; Gaetani et al. 2003), while garnet in equilibrium with carbonatite has a 7 times higher ratio than the melt (Dasgupta et al. 2009). However, the extent of the offset in Ti/Eu of garnet relative to basaltic melts is not relevant on the very large scale of observed variations in Ti/Eu (Fig. 27). Also, considering the extremely low Ti/Eu of magnesio-carbonatites (average of 52; Chakhmouradian 2006; compared to, e.g., 7451 for N-MORB; Sun and McDonough 1989), the actual effect of preferential partitioning of Ti over Eu into garnet in equilibrium with carbonatites is very limited.

Strongly subchondritic Ti/Eu and superchondritic Zr/Hf (Rudnick et al. 1993; Yaxley 1993) of most of the studied Kaapvaal subcalcic garnets was used by Shu and Brey (2015) to establish mantle metasomatism through carbonate and silico-carbonate melts. Our database of xenolith-derived peridotitic garnets overall agrees with this finding, but suggests that HDFs should also be considered as a potential metasomatic agent for garnets with near-chondritic Zr/Hf (Fig. 27). For garnet inclusions in diamond with super-chondritic Zr/Hf (> 37), a number of samples have very low Ti/Eu (< 500) and closely match average compositions of magnesio-carbonatite while samples with more elevated Ti/Eu fall close to the primitive Group I and II kimberlite compositions (Fig. 27). Nearly half of the inclusions (55% of harzburgitic and 47% of lherzolitic garnets), however, have subchondritic Zr/Hf. Overall, HDFs in fibrous diamonds and the melts in equilibrium with MARID (mica–amphibole–rutile–ilmenite–diopside) and PIC (phlogopite–ilmenite–diopside) xenoliths from the Kaapvaal Craton (see caption of Fig. 27 for references), provide a better fit for the metasomatic agent required for the majority of inclusions, with some harzburgitic inclusions indicating even lower Zr/Hf in the metasomatic agent. Considering the principal types of REE_N patterns observed for peridotitic garnet inclusions, the median composition for harzburgitic garnets peaking at Nd (Zr/Hf = 30; Ti/Eu = 1297) closely matches the field for MARID and PIC melts, lherzolitic garnets peaking at Sm_N (median Zr/Hf = 53; Ti/Eu = 292) nearly overlap average magnesio-carbonatites, and lherzolitic garnets with normal REE_N (median Zr/Hf = 40; Ti/Eu = 5734) evolve towards primitive mantle (Zr/Hf = 37; Ti/Eu = 7825). These observations indicate that the carbonatitic metasomatism invoked for xenolith garnets (Shu and Brey 2015) likely affected some inclusion garnets, but is not the dominant metasomatic process reflected by inclusions. This interpretation is consistent with the observation that carbonate inclusions in smooth-faced monocrystalline diamonds are very rare; for the peridotitic suite, only one well established example is documented, consisting of a magnesite inclusion in a lherzolitic diamond from Namibia. As an alternative, equilibration with MARID-PIC melts (for inclusions with low Zr/Hf), HDFs or kimberlites (carbonate–silicate melts) provides better matches for a large number of inclusion garnets.

Considering the evolution of the entire garnet data set shown in Figure 27, the starting point are the harzburgite residues left behind by major Archean melt extraction events (see above). After exhaustion of clinopyroxene (D_Zr/D_Hf of clinopyroxene–melt is ~0.5–0.6; Johnson 1998; Norman et al. 2005), orthopyroxene controlled the continued decrease in the Zr/Hf ratio of the melting residues (D_Zr/D_Hf of orthopyroxene–melt is ~0.2–0.5; Ulmer 1989; Green et al. 2000). After elimination of cpx from the residues (D_Ti/D_Eu of clinopyroxene–melt is ~ 1; Johnson 1998; Norman et al. 2005), an increase in Ti/Eu ratio of the residues is controlled by orthopyroxene (D_Ti/D_Eu of orthopyroxene–melt is ~3; Ulmer 1989; Green et al. 2000) and spinel (compatibility of Ti, at strict incompatibility of Eu, in Cr- and Fe³⁺-rich spinels; Wijbrans et al. 2015). The approximate evolution of residues of large-scale Archean melt depletion is indicated as a solid arrow in Figure 27. The resulting high Ti/Eu and low Zr/Hf of the depleted protoliths of subcratonic lithospheric mantle were associated with very low concentrations of Ti, Eu, Zr and Hf, making these rocks highly susceptible to metasomatic overprint. This metasomatism involved a range of fluids/melts (see above) with variable Zr/Hf. High Eu (reflecting high LREE/HREE) and low Ti in the metasomatic agent(s) moved the affected diamond substrates to very low Ti/Eu.

Focusing only on inclusion garnets with Ti/Eu ≳2500 and subchondritic Zr/Hf: if the observed spread from high Ti/Eu, akin to melt-depleted compositions, to ~2500 was due to metasomatic enrichment of Eu (trend of decreasing Ti/Eu), then a negative correlation of Ti/Eu with Eu content would be expected, but this is not the case (Fig. 28). Instead, there is a crude positive correlation of Ti/Eu with Ti content (Fig. 28), suggesting that the spread in Ti/Eu is largely a consequence of preferential re-enrichment in Ti that occurred as a second metasomatic event (dashed trend of increasing Ti/Eu in Fig. 27). Compared to the original melt-depleted harzburgites, which also have high Ti/Eu (melt depletion trend in Fig. 27), the inclusion garnets with Ti/Eu > 2500 have higher Eu (metasomatic event 1, affecting all garnets inclusions) and higher Ti contents (metasomatic event 2, affecting only some garnets). Very high Ti/Eu ratios are documented as a consequence of melt metasomatism in garnet rim compositions from hot sheared peridotites, with Ti/Eu ratios of, e.g., 9800 from Lesotho (Smith et al. 1993) and 23,000 from Arkahangelsk (Kargin et al. 2017). They also characterize garnets precipitated from the megacrysts magma, with Ti/Eu of 4600–44,000 (Yakutia: Kostrovitsky et al. 2013; Arkahangelsk: Kargin et al. 2016, 2017). Thus, a possibly modified, generally mild form (cf. Figs. 7 and 13) of the Fe–Ti metasomatism documented in xenoliths (see above) and tentatively linked to the proto-kimberlite/megacryst magma (e.g., Kargin et al. 2016; Ashchepkov et al. 2017a; Shchukina et al. 2017) may have affected some diamond substrates. The inferred second stage re-enrichment in Ti appears to have had little effect on Zr/Hf ratios, implying moderate to low Zr/Hf and/or low Zr and Hf concentrations in the metasomatic melt (consistent with an absence of positive correlations of either Zr or Hf with Ti/Eu). With D_Zr/D_Hf of garnet–melt being in the range 0.6–1 (see above), the near-chondritic Zr/Hf ratio of megacrysts (34–44; Kargin et al. 2017) is approximately representative of their parental melt. Combined with its very high Ti/Eu ratio, the megacryst magma would consequently be consistent with the evolution seen in Figure 27 (dashed pink arrow labelled Fe-Ti metasomatism, pointing towards megacryst garnets).

Calculated melts in equilibrium with garnet inclusions. A different approach to constrain the nature of metasomatic agents is to calculate the composition of melts or fluids in equilibrium with garnet or clinopyroxene inclusions and to compare these calculated compositions to observed mantle melts or fluids (e.g., Shimizu 1975). For this approach to be valid in the context of diamond studies, the inclusion mineral and the metasomatic agent must have reached equilibrium. For syngenetic inclusions, which precipitate together with their diamond hosts from a metasomatic agent, this precondition is implicitly met. For preexisting minerals, equilibration is a diffusion-controlled process and for centimeter-sized crystals in coarse peridotites may take over a billion years at the temperatures of diamond stable lithospheric mantle. For protogenetic inclusions in diamond (see above), which are typically recovered in the 0.1–0.2 mm range, diffusive equilibration during a fluid or melt infiltration event will, however, be rapid, e.g., ~20,000 years for REE in a 0.1 mm defect-free garnet residing at 1150 °C (Nestola et al. 2019b). Faster diffusion along dislocations could decrease the required time significantly. On that basis, we consider equilibrium between inclusion minerals and diamond-forming metasomatic fluids/melts highly likely even in the case of protogenetic inclusions.

A significant recent breakthrough for modelling the composition of metasomatic agents came from the experimental work of Kessel et al. (2015), who provided a large set (including the REE) of partition coefficients for both garnet and clinopyroxene with subsolidus HDFs (aqueous fluids with up to 36 wt% dissolved solids). For the calculation of model melts, we use the three principal garnet groups derived in Figure 18: harzburgitic garnets with REE_N peaking at Nd, lherzolitic garnets with sinusoidal patterns peaking at Sm, and lherzolitic garnets with normal REE_N patterns. For each of these garnet groups, we chose the 25^th percentile, the median and the 75^th percentile patterns to calculate the equilibrium compositions of HDF (6 GPa, 1000 °C; Kessel et al. 2015), hydrous melt (6 GPa, 1100 °C, melt with 39 wt% H₂O; Kessel et al. 2015) and “dry” melt (Johnson 1998). The suffix “dry” relates to the nominally dry Kilauea olivine tholeiite used in the experiments by Johnson (1998), but does not imply that the calculated melt compositions are only applicable to volatile-free systems. Because of the similarity in Zr/Hf and Ti/Eu of the median composition of lherzolitic garnets peaking at Sm and average magnesio-carbonatite (Fig. 27), for this garnet group we also use the partition coefficients for garnet-carbonatite of Dasgupta et al. (2009). The calculated equilibrium melts (Fig. 29) are then compared to (1) the spectrum of HDFs observed in diamonds, consisting of the typical “planed” and “ribbed” patterns of Weiss et al. (2013) and a potentially parental, saline HDF observed in a diamond from the Fox kimberlite (Weiss et al. 2015), (2) the primitive Group I and II kimberlite compositions of Becker and Le Roex (2006) and (3) in the case of the lherzolitic garnets peaking at Sm, the average magnesio-carbonatite composition of Chakhmouradian et al. (2009). The REE_N patterns for average MARID melt overlap the primary kimberlite compositions of Becker and Le Roex (2006), while the pattern for PIC melt shows the same slope but at about 1/3 of the REE concentrations (see caption of Fig. 27 for details on the calculation of MARID and PIC melt compositions). On the scale of Figure 29, the resulting offset, however, is so small that the kimberlite patterns may be viewed as representative for MARID and PIC melts as well.

For harzburgitic garnets peaking at Nd, all calculated HDF and melt REE_N patterns have very steep negative slopes (Fig. 29, top). The upward kink in the HREE_N is interpreted to be imposed by the substrates rather than as a characteristic of the primary metasomatic medium (see above). As harzburgitic substrates generally reside at subsolidus conditions, only the calculated HDF pattern is relevant for comparison in this case (Fig. 29, top, right panel). The absolute trace element concentrations in HDFs observed in fibrous diamonds are mainly a reflection of inclusion density and hence were normalized to match the Dy content of the calculated HDF. The comparison reveals a modest fit, which is closest for the primitive saline pattern (“saline Fox”; Fig. 29, top). The LREE_N–HREE_N fractionation in the calculated HDF covers almost three orders of magnitude, “ribbed” patterns span two orders of magnitude and “planed” patterns one order. Extrapolating the HREE slope (below the limit of quantification; Weiss et al. 2015) for the primitive saline pattern results in more similar overall LREE_N–HREE_N fractionation, but differences to the calculated HDF are still prominent in the much steeper slope from Nd_N to Eu_N for the latter. This misfit in REE patterns between calculated HDF and primitive saline fluid is exacerbated when considering the experimental study of Rustioni et al. (2021), which showed that with increasing salinity the LREE/HREE ratio of garnet in equilibrium with fluid of a given REE content decreases strongly. For saline fluids, HDF compositions (Fig. 29) calculated using the partition coefficients of Kessel et al. (2015), therefore, underestimate the LREE_N–HREE_N fractionation.

For lherzolitic garnets peaking at Sm, the calculated HDF pattern shows a kink at Sm but otherwise is again most similar to the primitive saline HDF pattern (Fig. 29, middle), with the above caveat that the LREE_N–HREE_N fractionation of the calculated HDF would increase if the effect of salinity was considered. The patterns for calculated silicate and carbonate melt compositions are even steeper than for the calculated HDF and hence do not match observed kimberlite and magnesio–carbonatite compositions. In comparison, the calculated HDF and melt compositions for lherzolitic garnets with normal REE_N patterns show much lower levels of LREE_N–HREE_N fractionation. The calculated HDF and hydrous melt compositions (the latter shown only in the left panel of Fig. 29, bottom) are intermediate in slope between the observed “planed” and “ribbed” patterns, while the “dry” melt composition parallels primitive Group I and II kimberlites. The formation of lherzolitic diamonds generally above the wet lherzolite solidus makes the match with kimberlites particularly relevant.

An inherent problem of visual comparisons of calculated and observed fluid/melt trace element patterns is that REE concentrations in the various media cover almost five orders of magnitude. Consequently, as long as the sign of the slope is the same, very distinct patterns appear visually similar. To examine the differences in slope in the REE_N patterns of calculated and observed fluid and melt compositions in more detail, we use the chondrite-normalized ratio of Nd/Dy (Fig. 30). In addition, focusing exclusively on REE blurs the distinction between HDFs with “planed” and “ribbed” trace element patterns, for which a key distinguishing criterion is the presence (“ribbed”) or absence (“planed”) of a strong negative Nb anomaly (Weiss et al. 2013), expressed as a strongly elevated chondrite-normalized Ce/Nb ratio in Figure 30. Calculated HDF, hydrous melt and “dry” melt compositions all show similar Nd_N/Dy_N, decreasing strongly from equilibrium with garnets with sinusoidal REE_N peaking and Nd to those with normal patterns. Only for the melts in equilibrium with garnets with normal patterns are Nd_N/Dy_N ratios observed similar to HDFs and low volume mantle melts. Using partition coefficients for garnet-carbonatite (Dasgupta et al. 2009) leads to even more extreme LREE–MREE fractionation, with Nd_N/Dy_N ratios much higher than natural magnesio-carbonatites for all three garnet groups utilized for melt calculation. In terms of Ce_N/Nb_N ratio, only calculated carbonate melts in equilibrium with garnets with sinusoidal REE_N patterns peaking at Nd would show negative Nb anomalies (Ce_N/Nb_N of about 5). This excludes HDFs with “ribbed” trace element patterns as a principal diamond-forming metasomatic agent and instead highlights a similarity in the occurrence of mildly positive Nb anomalies (Ce_N/Nb_N < 1) for calculated HDF/melts with Group I kimberlites and with natural HDFs with “planed” trace element patterns.

Garnet, and for the lherzolitic paragenesis also clinopyroxene, are the principal carriers of incompatible trace elements among peridotitic inclusions in diamond. The REE_N patterns of garnet inclusions are typically sinusoidal, with harzburgitic garnet most commonly peaking at Nd and lherzolitic garnets at Sm. As another end-member type of REE_N patterns, some lherzolitic and very few harzburgitic garnets have normal patterns, characterized by flat MREE_N–HREE_N. This evolution from sinusoidal to normal garnet REE_N patterns is mirrored for clinopyroxene inclusions by a transition from patterns with a characteristic steep negative slope in the LREE_N to humped patterns peaking at Nd_N.

Normalization to “primitive mantle” garnet/clinopyroxene documents a two-stage evolution of the protoliths (cf. Frey and Green 1974), starting with intense melt depletion causing overall REE depletion and steep positive LREE_N/HREE_N slopes in the bulk rocks (preserved only in the HREE from Er to Lu). This depletion is typically followed by metasomatic re-enrichment through a medium with very high LREE and LREE/HREE and distinctly subchondritic HREE. Garnet with normal REE_N patterns and coexisting clinopyroxene with humped patterns indicate primitive mantle-like bulk rock REE concentrations and are a consequence of metasomatism through media with only moderately fractionated trace element contents. Whilst such near-primitive REE_N patterns are rare among inclusions – about 5% of all garnet inclusions have normal REE_N—they strongly dominate among peridotite xenolith and xenocryst garnets. This is not only a function of reversed proportions of the harzburgitic and lherzolitic parageneses for inclusions (high ratio) and xenoliths (low ratio) but is also observed when looking at lherzolitic garnets only. Whether this observation reflects the well-established affinity of diamond to the most depleted peridotitic substrates (e.g., Gurney 1984), or an increasing metasomatic modification through time that inclusions were shielded from by their diamond hosts, or probably a combination of both, is not well constrained by currently existing data.

The nature of the metasomatic agents that re-enriched the diamond substrates and presumably also caused repeated episodes of diamond formation cannot be fully resolved. The HDF inclusions observed in coated and fibrous diamonds and sometimes detected in clear, smooth-faced monocrystalline diamonds are too “melt-like” in their trace element patterns to create the extreme LREE/HREE fractionation seen in the vast majority of garnet and clinopyroxene inclusions. Whether that reflects a temporal evolution, with Archean diamond-forming HDFs showing overall stronger trace element fractionation, or if the HDFs precipitating either fibrous diamonds or diamond coats are more primitive members of a HDF fractionation series (e.g., through chromatographic column type fractionation; Navon and Stolper 1987; Bodinier et al. 1990) will have to be established through future research. Near-primitive trace element patterns are consistent with melt-like metasomatic agents (including the HDFs seen in fibrous diamonds) and, given the relationship between cratonic geotherms and the hydrous solidus of lherzolite, likely reflect infiltration of low volume hydrous melts, similar to primitive kimberlites. Conversely, the high temperatures required for hydrous melting of strongly depleted peridotites (harzburgites-dunites) shield such rocks against pervasive melt metasomatism, as such melts must freeze upon equilibration with their wall rocks.

Occurrence. As broadly metabasaltic rocks sometimes containing coesite or diamond, eclogite occurs in dynamic settings where (ultra-)high-pressure metamorphism takes place. This encompasses deeply subducted and exhumed portions of oceanic crust consisting of pillow lavas, sheeted dikes and gabbros (e.g., White and Klein 2014), which are now sampled as lenses in orogens (e.g., Liou et al. 2004) and as xenoliths in kimberlites and related rocks (e.g., Jacob 2004). Diamond populations dominated by eclogitic inclusions are often associated with craton margins or cratonic lithosphere affected by later tectono-magmatism, suggestive of a subduction origin of the source rock (Sobolev 1985; Shirey et al. 2002; Stachel and Harris 2008). Some mantle xenolith suites are dominated by eclogite (e.g., at the Roberts Victor and Orapa mines in the Kaapvaal craton, and at Koidu in Sierra Leone in the West African craton; MacGregor and Carter 1970; Hills and Haggerty 1989). As the lithospheric mantle contains only a few percent (Schulze 1989), at most several 10s of percent (Garber et al. 2018) eclogite, this local dominance highlights preferential preservation of eclogite versus peridotite and/or non-representative sampling by kimberlite and lamproite magmas (e.g., Moss et al. 2018).

A proportion of garnet–clinopyroxene rocks in xenolith suites with jadeite-poor clinopyroxene classify as pyroxenite, although these are sometimes referred to as high-Mg or Group A eclogite (Coleman et al. 1965; Taylor and Neal 1989). While bimineralic pyroxenites grading into high-Mg eclogites form part of most cratonic xenolith suites (Aulbach et al. 2020a), orthopyroxene-bearing pyroxenites are more common in craton-margin and off-craton localities (Jacob 2004), where they record generally lower equilibration pressures than, and have an origin different from, eclogite (Schmädicke et al. 2011).

Mineralogy.Stachel and Harris (2008) found that garnet and clinopyroxene, the main rock-forming minerals in eclogite, comprised 56% and 39% of the total of inclusions in 1311 eclogitic diamonds; other common silicate and oxide inclusions being rutile (2.7%) and coesite/quartz (1.8%). In contrast, garnet, clinopyroxene, orthopyroxene, coesite and olivine constitute 38%, 37%, 21%, 3% and 1%, respectively of inclusions in 136 websteritic diamonds (Stachel and Harris 2008). Despite their accessory mineral status in common lithospheric mantle sources, sulfides are the most common inclusion in diamond (Harris and Gurney 1979; Haggerty 1986). However, they will not be further considered here, as their occurrence as an inclusion in diamond is considered as part of the diamond age-dating work and is reviewed in Smit et al. (2022).

Of 2451 xenoliths considered for comparison of mineralogy, commonly reported primary accessory and minor silicate and oxide minerals include diamond (n = 251), rutile (n = 83), kyanite (n = 67), graphite (n = 59; of these 10 also contain diamond), corundum (n = 25) and coesite/quartz (n = 11). Sulfides are common, though not necessarily reported as part of the assemblage (pers. observation). Garnet clinopyroxenites (n = 334), which grade into websterite and olivine-poor lherzolite, more commonly contain orthopyroxene (n = 30 vs. 8 in eclogite xenoliths), but only 15 of 544 pyroxenite xenoliths (3%) are diamondiferous, compared to 12% of eclogite xenoliths. As the description of websteritic inclusion mineralogy provided in the previous paragraph is based on a different classification scheme than that applied to pyroxenite xenoliths (see Inclusion suites and parageneses), these findings are not entirely comparable.

The proportion of diamondiferous eclogite (12%) and pyroxenite samples (3%) may be high because such rocks are frequently specifically targeted for study. On the other hand, accessory minerals may be more common than they appear, because they need not be exposed at the section scale or may not have been reported, depending on the aims of the study. Whilst other minerals in the xenoliths, such as amphibole, plagioclase and phlogopite are likely to be of secondary, metasomatic origin, zircon, apatite and ilmenite could be primary, although more likely to be metasomatically introduced (Aulbach et al. 2020a, and references therein). In addition, Jacob (2004) notes the typical purity of minerals in mantle eclogite, which rarely contain inclusions, as opposed to their orogenic counterparts. Eclogites rich in CaO and Na₂O are particularly affected by texturally late alteration, resulting in the partial or complete replacement of primary jadeite-rich clinopyroxene by “spongy” diopside-rich clinopyroxene plus K-rich glass and other minerals (e.g., Snyder et al. 1997; Spetsius and Taylor 2002). This texturally late assemblage reflects the greater instability of jadeite-rich clinopyroxene in contact with an ultramafic carbonated agent such as kimberlite (e.g., Misra et al. 2004). Also, the presence of amphibole and phlogopite in the kelyphite rims of garnet and of K-rich glass in the spongy clinopyroxene suggests partial melting and kelyphite formation catalyzed by kimberlite-related fluids (Snyder et al. 1997; Spetsius and Taylor 2002).

Classification. Of the aforementioned 2451 xenoliths considered for mineralogy, only 946 have reported major element abundances for both garnet and clinopyroxene, allowing distinction of eclogites and pyroxenites. This yields 776 mantle eclogite samples (236 diamondiferous) and 170 pyroxenite samples (15 diamondiferous). Only a small proportion of diamondiferous samples have reported REE concentrations that further allow assignment to gabbroic, with positive Eu anomalies, vs. non-gabbroic sources (Table 3).

When inclusions are recast in the scheme applied to mantle xenoliths, clinopyroxene–garnet inclusion pairs comprise 235 eclogitic and 37 pyroxenitic diamonds. Of unpaired clinopyroxene, 481 are eclogitic (15 of 26 with REE data have a gabbroic signature) and 86 pyroxenitic (2 of 3 with REE data are gabbroic). Unpaired garnet can be assigned to high-Ca, high-Mg, low-Mg and gabbroic groups, bearing in mind that some high-Mg samples may actually derive from pyroxenite sources (~70% of pyroxenite xenoliths have high-Mg garnet) and some high-Ca samples without REE data may actually derive from gabbroic sources (67% of gabbroic xenoliths contain high-Ca garnet) (Table 3).

Obtaining accurate and precise formation ages from mantle eclogite and pyroxenite is challenging because bulk rocks are affected by kimberlite contamination, and because two-mineral isochrons yield kimberlite eruption or cooling ages. Also, zircon is rare in these rocks and can be of metasomatic origin (Jacob 2004; Aulbach et al. 2019a). Nevertheless, ages and the methods by which they were derived are summarized in Appendix Table 1. Worldwide, the oldest age arrays obtained from Pb–Pb, Sm–Nd and Re–Os geochronology on bulk rocks or mineral separates are often Mesoarchean to Neoarchean, (3.2 to 2.5 Ga) reflecting the suturing of continental nuclei and haphazard incorporation of intervening oceanic lithosphere to form larger cratonic entities. Exclusively Paleoproterozoic ages (2.5 to 1.6 Ga) have been obtained from eclogite and pyroxenite xenoliths in the northern and central Slave craton, which coincide with collision/accretion at the western craton margin, preceded by subduction of an intervening ocean floor.

In the following, inclusions in diamond are compared with similar minerals in xenoliths, the latter distinguished according to whether they are reported as diamondiferous or barren. Median and average concentrations are typically very similar, especially for large numbers (100s) of samples in a given group. Differences are more apparent for pyroxenites, for which smaller numbers are available and which are a heterogeneous group with respect to their compositions and origins (Stachel and Harris 2008). Median compositions are used for comparison to safeguard against outliers, and median absolute deviations are reported as the corresponding uncertainty. Nevertheless, the relatively small number of inclusions (especially pyroxenitic) with trace-element or δ¹⁸O data make comparisons and unambiguous interpretation tenuous. Also, minor elements, such Cr, Ni and K, are not consistently analysed and reported. Appendix Table 3 shows the statistical analysis for four eclogite and two pyroxenite classes. Minerals in diamond and similar ones in diamondiferous xenoliths are distinguished according to whether they are eclogitic or pyroxenitic and whether they are touching or non-touching, as the latter relationship affects calculated temperatures, and also some distribution coefficients. To simplify the following compositional comparison, touching and non-touching inclusions are considered together, and unpaired clinopyroxene, which allows classification into eclogite and pyroxenite, is subsumed with paired clinopyroxene. Finally, the gabbroic pyroxenite class shares the cumulate signature with gabbroic eclogites and the enriched signature with nongabbroic pyroxenite xenoliths. For the sake of clarity, the gabbroic pyroxenite class is not described below or shown in most figures.

Mineral major and minor element compositions. Clinopyroxene in eclogitic diamond has a median Mg# of 76.5, a diopside content of 0.46 and a jadeite (Jd) component of 0.32 (Appendix Table 3). For clinopyroxene in diamondiferous eclogite these values are a higher Mg# (82.4) and a diopside content of (0.41) but similar Jd (0.35). In contrast, clinopyroxene in all classes of barren eclogite (gabbroic, high-Ca, high-Mg, low-Mg) have Mg# of 78.3–85.3, consistently higher diopside content (0.54–0.65) and mostly lower Jd, varying between 0.33 in gabbroic and 0.20 in high-Mg eclogites. Median K₂O contents of clinopyroxene in diamond (0.20 wt%) are more than twice as high as in diamondiferous eclogite (0.09 wt%) and 3–4 times as high as in barren eclogites (≤ 0.06 wt%). In a diagram of Jd vs. Mg#, (Fig. 31), clinopyroxene inclusions in eclogitic diamond span a large range, with diamondiferous eclogites shifted to higher Mg#, and barren eclogites to both higher Mg# and lower Jd. In pyroxenitic diamond, clinopyroxene has a median Mg# of 83.5 and a diopside content of 0.72, which is much lower than the Mg# (≥ 90.1) or diopside content (≥0.77) in clinopyroxene from diamondiferous or barren pyroxenite, but still higher than in eclogitic clinopyroxene. By definition, jadeite contents are lower in pyroxenitic than in eclogitic clinopyroxene, but Cr₂O₃ contents are higher (≥ 0.13 wt%), though overlapped by clinopyroxene in high-Mg eclogite with 0.14 wt%.

Garnet in eclogitic diamond has distinctly higher TiO₂ (0.61 wt%) and Na₂O contents (0.20 wt%) than garnet in diamondiferous eclogite (0.36 and 0.13 wt%, respectively) and various barren eclogite classes (≤ 0.27 and 0.09 wt%, respectively) (Appendix Table 3). Median Mg# of included garnet is 52.7. Garnet in diamondiferous eclogite has Mg# 61.7, which is very similar to that in gabbroic barren eclogite (62.2). All the three garnet groups referred to above have a Ca# (100Ca / (Ca + Mg + Fe^total + Mn)) of 24. Garnet in high-Mg barren eclogite has the highest Cr₂O₃ contents (0.13 wt%) and Mg# (68.6) and the lowest Ca# (11.2) of all eclogite classes, the nearest comparison being the unpaired high-Mg garnet in diamond (Cr₂O₃ 0.17 wt%, Mg# 66.1, Ca# 11.1). Compared to paired inclusions, unpaired garnet specifically extends into the field of diamondiferous eclogites with high Mg#–Ca# typical of some gabbroic eclogites (Fig. 32a). Garnet inclusions in pyroxenitic diamond have again lower Mg# (63.2) and higher Ca# (14.3) than in diamondiferous pyroxenites (77.7 and 10.2, respectively). Median Cr₂O₃ contents in pyroxenitic garnet (≥ 0.29 wt% for all classes) are higher than in eclogitic garnet (≤ 0.13 wt% for all classes), and many fall on a “lherzolite” trend of CaO vs. Cr₂O₃ that indicates buffering of CaO contents by orthopyroxene (Fig. 32b).

Ca# in garnet shows a broad positive correlation with Jd, where inclusions range to higher Jd at a given Ca# and are only weakly represented at Ca# ≳37 compared to diamondiferous eclogites (Fig. 33a). Combined, Na₂O in garnet and K₂O in clinopyroxene show a strongly diminished range both in diamondiferous and in barren eclogites/pyroxenites compared to inclusions (Fig. 33b).

Mineral trace element compositions. Trace elements for which at least tens of analyses for inclusions in diamond are available include Ni, Sr, Y, Zr, Nb, the REE and Hf (Appendix Table 3). For clinopyroxene, median Ni contents are lowest in inclusions (160 ppm, vs. 251 ppm in clinopyroxene from diamondiferous eclogite and 196–312 ppm from barren eclogite). Median Sr contents in inclusions (136 ppm) and in clinopyroxene from diamondiferous eclogite (142 ppm) are similar and intermediate between those from high-Ca eclogite (118 ppm) and gabbroic eclogite (162 ppm), whereas that in high-Mg and low-Mg eclogite is higher (186 and 193 ppm, respectively). Clinopyroxene in diamondiferous eclogite has low Y content (0.9 ppm), between that in gabbroic eclogite (0.6 ppm) and high-Ca eclogite (0.9 ppm), but inclusions have higher abundances (2.4 ppm). Median Zr contents of clinopyroxene in diamond (14.1 ppm) and diamondiferous xenoliths (10.5 ppm) are again similar to those in gabbroic and high-Ca eclogite (10.3 ppm and 16.4 ppm, respectively), being higher still in high-Mg and low-Mg eclogites. Conversely, by far the highest median Nb contents (0.25 ppm) are recorded for clinopyroxene in diamondiferous eclogites, followed by inclusions (0.10 ppm), whereas those in barren eclogites are lowest (≤ 0.08 ppm). The comparison for pyroxenite classes is hampered by the low number of trace-element analyses both for inclusions and for diamondiferous xenoliths (Appendix Table 3), and by the heterogeneous nature of their petrogenesis (Stachel and Harris 2008; Aulbach and Jacob 2016). Median concentrations of trace elements in clinopyroxene from pyroxenitic diamond are very different from those of their xenolithic counterpart, which are, in turn, more similar to those of high-Mg eclogite xenoliths (Appendix Table 3).

Chondrite-normalized median REE patterns of clinopyroxene (Fig. 34a) show systematically decreasing abundances between Gd and Lu for all classes, except clinopyroxene in diamondiferous eclogite, which shows an upturn in Yb-Lu. The pattern of clinopyroxene in eclogitic diamond shows a hump in Nd_N, most similar to that in high-Ca eclogite, but differs from xenolithic clinopyroxene in having a less steep slope in the MREE_N–HREE_N and higher concentrations of these elements. REE abundances in clinopyroxene from high-Mg and low-Mg eclogites are higher than for inclusions. Clinopyroxene in gabbroic eclogite shows a positive Eu anomaly (Eu/Eu* = 1.22), shared to a lesser extent by that in diamondiferous eclogite (1.19) and that in diamond (1.07). Individual REE_N patterns of clinopyroxene in pyroxenitic diamond are extremely heterogeneous, not only with respect to abundances, which vary by several orders of magnitude, but also to shape (not shown), which precludes calculation of a meaningful median composition. Individual inclusions have Eu/Eu* from 0.61 to 2.4, with a median of 1.07, i.e., slightly gabbroic. Clinopyroxene in the single diamondiferous pyroxenite with REE data, together with its barren counterpart, has the highest LREE abundances of all classes, but shows a stronger negative slope in the MREE_N–HREE_N, similar to that in gabbroic pyroxenite (Fig. 34a).

For garnet, median Ni and Sr contents are highest for inclusions in diamond (62 and 1.76 ppm, respectively), followed by that in diamondiferous eclogite (49 and 0.92 ppm, respectively) and barren eclogite classes (41–14 ppm and 0.62–0.26 ppm, respectively). Median Y contents of garnet in diamond (30 ppm), high-Ca eclogite (34 ppm) and low-Mg eclogite (36 ppm) are highest, whereas those of garnet in diamondiferous eclogite (19 ppm) are more similar to those of garnet in gabbroic eclogite (17 ppm). Median Zr contents are identical for garnet in diamond and in diamondiferous xenoliths (19 ppm), whereas those for garnet in high-Ca and low-Mg eclogite (11 ppm) and in gabbroic eclogite (9 ppm) are much lower. Median Nb contents vary widely, and are lowest for garnet in gabbroic eclogite (0.03 ppm), followed by inclusions (0.06 ppm) and highest for garnet in diamondiferous eclogite (0.20 ppm), as applies to coexisting clinopyroxene.

Chondrite-normalized median REE patterns for garnet in the various eclogite classes are very similar at first glance, with low LREE, more or less strongly increasing abundances to Dy_N and then small or no increases to Lu_N (Fig. 34b). Garnet in diamond and in diamondiferous eclogite have the highest La and Ce abundances, but, in contrast to inclusions, garnet in diamondiferous eclogite shows a flatter HREE_N pattern, similar to garnet in gabbroic eclogite. Contrary to clinopyroxene, REE_N patterns of individual garnet inclusions in pyroxenitic diamond are relatively coherent (not shown). The median REE_N pattern for garnet in pyroxenitic diamond shows a strong increase between Pr and Lu, and is very different from that of garnet in diamondiferous pyroxenite, which increases strongly between La and Sm and then flattens rather abruptly. This in turn differs from the REE_N pattern of garnet in barren pyroxenites, which does not attain as high LREE abundances.

Garnet oxygen isotopes For six paired eclogitic garnet inclusions the median δ¹⁸O is +7.4‰, whereas in diamondiferous eclogite it is +6.1‰ and in barren xenoliths it is lower still, varying between +6.0‰ for garnet in low-Mg eclogite (n = 9) and +5.4‰ for that in high-Ca and gabbroic eclogite (both +5.4‰, n = 23 and 81, respectively). Unpaired high-Ca and low-Mg garnet inclusions, which are likely eclogitic, both have median δ¹⁸O of +6.8‰ (n = 32 and 3, respectively), whereas for high-Mg garnet δ¹⁸O is +6.6‰ (n = 9). The variability of δ¹⁸O is high (> 6‰ for garnets in eclogite xenoliths with Mg# < 50 but diminishes to < 3‰ at Mg# > 75, especially in barren pyroxenite xenoliths (Fig. 35). Inclusions show a crude negative correlation of δ¹⁸O vs. Mg# (r² = 0.2, n = 50) and are barely represented in the field of sub-mantle δ¹⁸O at low Mg#, where a group of garnets from barren high-Ca and gabbroic eclogite xenoliths plots (Fig. 35). The latter are exclusively from the Roberts Victor eclogite. No δ¹⁸O data are available for clinopyroxene inclusions in the database, or for garnet in pyroxenitic diamond or diamondiferous pyroxenite. Garnet in barren pyroxenite has δ¹⁸O of +5.5‰.

Temperature estimates. For all samples, temperatures are calculated with the Mg–Fe exchange thermometer of Krogh (1988; T_KR88), either at a pre-set pressure of 5 GPa using total Fe contents, in keeping with Stachel and Luth (2015), or solved iteratively with regional model conductive geotherms from Hasterok and Chapman (2011; hereafter HC11). The geotherm for non-touching inclusions corresponds to a surface heat flow of 38 mW/m², chosen to yield median pressures identical to those obtained for touching inclusions (see below). Samples yielding temperatures < 800 °C are not displayed or counted because temperatures may be underestimated at T < 900 °C (Krogh 1988) and because there is overlap with lower crustal xenoliths, e.g., in the Kaapvaal craton (Pearson et al. 1995). Temperatures > 1500 °C are inferred to reflect disequilibrium, although the cut-off may actually be lower depending on lithosphere thickness.

Non-touching and touching eclogitic inclusions record median temperatures of 1225 and 1132 °C, respectively, reduced to 1207 and 1099 °C, respectively when the inclusions in Argyle diamonds, which are derived from an atypical high-temperature environment, are not considered (cf. Stachel and Harris 2008). These values compare with 1134 °C for diamondiferous eclogites (Appendix Table 3). Barren eclogites yield lower median temperatures, between 1087 °C for high-Mg eclogites and 962 °C for low-Mg eclogites. Non-touching and touching pyroxenitic inclusions give 1198 and 1065 °C, respectively, whereas temperature estimates for diamondiferous and barren pyroxenites are much lower, 920 and 974 °C, respectively. Temperature distributions are displayed in Figure 36a.

Trace element distribution. The distribution of trace elements between garnet and clinopyroxene in mantle eclogite and pyroxenite xenoliths has been shown to variably depend on garnet Ca#, jadeite (Jd) in clinopyroxene, and temperature and pressure (O’Reilly and Griffin 1995; Harte and Kirkley 1997). Mantle metasomatism of eclogite tends to lower Ca# and jadeite content, with preferred trace element partitioning into clinopyroxene, which can lead to the paradoxical observation that garnet from metasomatized eclogite has abundances of some incompatible elements similar to or even lower than garnet in unmetasomatized specimens (Aulbach et al. 2020a).

Garnet Ca# and jadeite content in clinopyroxene from diamond and diamondiferous eclogite show a broad positive covariation (r² = 0.25, n = 515). This is also recognizable in barren eclogite (r² = 0.34, n = 694), albeit at a lower jadeite intercept value, reflecting a higher proportion of pyroxenite (Fig. 33a). Thus, Ca# in garnet and Jd in clinopyroxene, which do not co-vary significantly with temperature, are coupled. With the exception of one outlier, inclusions, diamondiferous xenoliths and barren xenoliths show a negative correlation of log D_Sr (the distribution coefficient D of Sr between clinopyroxene and garnet) with garnet Ca# (r² = 0.64, n = 24; r² = 0.29, n = 78 and r² = 0.38, n = 635, respectively). These trends are offset to increasingly higher log D_Sr at a given garnet Ca# for diamondiferous and barren xenoliths (Fig. 37a). This variation is also true for the more incompatible trace element Ce (excluding four outliers), where inclusions have r² = 0.35 (n = 31), diamondiferous xenoliths have r² = 0.44 (n = 71) and barren xenoliths have r² = 0.59 (n = 674) (Fig. 37b). This trace-element relationship similarly applies to Zr and Nb (not shown), except that there is greater overlap between inclusions in diamond and barren eclogites. The distribution of Y, which is compatible in bulk eclogite (see, e.g., distribution coefficients for garnet and clinopyroxene in equilibrium with tonalite melt in Barth et al. 2002), is strongly negatively correlated with Ca# (r² = 0.64, n = 783), but there are no distinct trends for diamondiferous and barren samples (not shown). Both log D_Sr and log D_Ce show marked negative correlation with TKR88(5 GPa) (r² = 0.43, n = 747 and r² = 0.31, n = 752, respectively), whereas log D_Y shows no correlation. As trace element data are available for only one touching inclusion, the effect of cooling and re-equilibration on trace element distribution cannot be assessed.

Reconstructed whole rocks. Following Aulbach and Jacob (2016), whole rocks are reconstructed from 55 wt% garnet and 45 wt% clinopyroxene, bearing in mind that HFSE abundances are minima when rutile is not considered. The presence of minor kyanite and/or orthopyroxene can further adversely affect accuracy of whole rock estimates, which is the case for only < 3% kyanite-bearing and < 2% orthopyroxene-bearing xenoliths, coesite being reported for even fewer xenoliths in this study. The above mineral modes correspond to the median modes determined for large eclogite xenoliths reported in Agashev et al. (2018), with a standard deviation of 5%. While bulk major-element compositional variations are largely accommodated by both garnet and clinopyroxene compositions (Aulbach and Jacob 2016), a single mode is unlikely to be representative of the true variation, and, in particular, pyroxenites may have higher clinopyroxene/garnet ratios. Typical uncertainties in major-element contents of reconstructed bulk rocks for a 10% modal uncertainty are displayed in Figure 38.

From the compilations in Figure 38, median whole-rock FeO contents for inclusions (11.8 wt%) are higher than for diamondiferous eclogites (9.8 wt%), and most similar to barren high-Ca eclogites (12.4 wt%), whereas those in high-Mg and gabbroic eclogites are much lower (9.7 and 9.4 wt%) (Appendix Table 3). Conversely, whole-rock MgO contents for inclusions (10.1 wt%) are lower than those for diamondiferous eclogites (11.5 wt%), and are lowest for barren high-Ca eclogites (9.6 wt%) and, by definition, highest for high-Mg eclogites (14.8 wt%). Also by definition, CaO contents are highest in high-Ca eclogites (12.5 wt%), whereas diamondiferous eclogites have 11.6 wt% and inclusions have 10.8 wt%. For other oxides not shown in Figure 38, the whole-rock Al₂O₃ contents of inclusions (16.3 wt%) and diamondiferous eclogites (16.5 wt%) are most similar to gabbroic eclogites (16.3 wt%), and higher than in other barren eclogites, the lowest values being observed in high-Mg eclogites (14.7 wt%). Similar variations are mirrored by Na₂O contents: 2.5 wt% for inclusions and diamondiferous eclogites, 2.3 wt% for gabbroic and high-Ca eclogites, and only 1.6 wt% for high-Mg eclogites. Whole-rock Cr₂O₃ contents are lowest for high-Ca eclogites (0.05 wt%) and inclusions (0.06 wt%), and highest for high-Mg eclogites (0.14 wt%).

Whole rock reconstruction for pyroxenitic inclusions shows FeO contents (11.1 wt%) that are higher than for diamondiferous pyroxenites (7.0 wt%) or barren pyroxenites (7.3 wt%), as applies to eclogites. Conversely, MgO contents are lowest for inclusions (15.9 wt%) and highest for diamondiferous pyroxenites (18.6 wt%), whereas CaO (11.6–11.7 wt%) and Cr₂O₃ contents (0.28–0.30 wt%) are indistinguishable and Al₂O₃ contents similar (13.4 wt% for inclusions, 13.9 wt% for diamondiferous and barren pyroxenites). Moreover, Na₂O contents are higher for inclusions (0.90 wt%) than for diamondiferous pyroxenites (0.65 wt%), but similar to barren pyroxenites (0.92 wt%). The major-element relationships of Figure 38 illustrate the offset of xenolith chemistry towards overall lower FeO and higher MgO, Cr₂O₃ and CaO contents compared to inclusions.

Chondrite-normalized median REE patterns for bulk rocks reconstructed from inclusions (Fig. 39) show strong LREE depletion, similar to reconstructed diamondiferous, high-Ca and gabbroic eclogite. The REE pattern flattens out for gabbroic eclogite and, at twice higher abundances, high-Ca eclogite. In contrast, both diamond and diamondiferous eclogite show a continuous increase towards the HREE. Low-Mg and high-Mg eclogites are characterized by much higher LREE abundances and slightly increasing abundances in the HREE. Positive Eu anomalies are common to whole rocks reconstructed from inclusions (1.05), for diamondiferous eclogites (1.16) and, by definition, for gabbroic eclogites (1.21). The other eclogite classes have lower median Eu/Eu* (0.96–0.98). Reconstruction of pyroxenites from inclusions is hampered by the small number of samples and their wide compositional variation, as described for individual minerals. The single diamondiferous pyroxenite has a REE_N pattern similar to its barren non-gabbroic counterpart (Fig. 39). Barren gabbroic pyroxenite shares positive Eu/Eu* and low HREE with barren eclogite, but is LREE-enriched.

Some thoughts on inclusion thermobarometry. Accurate barometry for eclogitic and pyroxenitic garnet–clinopyroxene pairs remains elusive (cf., Nimis 2022), although some formulations have been proposed (e.g., Beyer et al. 2015; Ashchepkov et al. 2017b). In most instances, the thermal gradient of the regional lithospheric mantle has been independently determined from peridotitic minerals or assemblages, so that pressures can be estimated by solving T_KR88 iteratively with the appropriate gradient, assuming equilibration to the conductive geotherm (e.g., Aulbach et al. 2020a). Because in xenoliths, minerals used for geothermobarometry remain in contact and equilibrate to ambient mantle temperatures until entrainment, they reflect the thermal state of the lithosphere at the time of kimberlite emplacement. The same should apply to touching garnet–clinopyroxene pairs included in diamond. Thus, for xenoliths and touching inclusions, T_KR88 is solved iteratively with geothermal gradients parameterized from Hasterok and Chapman (2011; HC11), corresponding to surface heat flows of 38–35 mW/m². These model geotherms either correspond to those directly determined based on peridotitic assemblages or clinopyroxene relative to HC11 geotherms, as is the case for Koidu, West Africa. In other instances, they were roughly translated from the older Pollack and Chapman (1977) family of geotherms as quoted in older literature (Appendix Table 1). Application of iterative solutions moves a fraction of xenoliths to temperatures < 800 °C, and results in pressures > 8 GPa (> 250 km) for high-temperature samples equilibrated to cool geotherms (35 mW/m²). Keeping all solutions, this results in average pressures of 4.9 ± 1.5 GPa and temperatures of 950 ± 110 °C for the small number of touching eclogitic inclusions (n = 15). These values compare relatively well with 5.2 ± 1.2 and 970 ± 130 °C for 228 diamondiferous eclogites (37 mW/m²). With touching pyroxenite inclusions (n = 8) average pressures and temperatures vary between 3.8 ± 0.5 GPa and 860 ± 80 °C (38 mW/m²), whilst 14 diamondiferous pyroxenites yield 4.1 ± 0.7 and 740 ± 100 °C (36 mW/m²). Of the latter, ten are from the northern Slave craton where the lithosphere is characterized by a cold geotherm (Grütter 2009).

The appropriate choice of a geotherm for non-touching inclusions, which last equilibrated at the time of encapsulation, is less certain. As outlined above, eclogitic diamond formation commenced around 3 Ga, when lithospheric conductive gradients would have been higher. One could assume that the lithosphere experienced only cooling, and that touching and non-touching inclusions formed over the same pressure range. The conductive geotherm for iterative calculations is then chosen such that the median pressure for non-touching inclusion pairs equals that of touching inclusions. According to Hasterok and Chapman (2011), this situation occurs when the geotherm corresponds to a surface heat flow of 38 mW/m², and 220 non-touching eclogitic inclusions give values of 4.9 ± 0.8 GPa and 1040 ± 130 °C. The resultant cooling by only some 100 °C is consistent with stabilization of cool continental lithosphere by the Mesoarchean, before substantial eclogitic diamond formation. In addition, pressure–temperature conditions for non-touching inclusions can be underestimated if they experienced decompression due to extension, as suggested for the Kimberley area in the Kaapvaal craton (Phillips et al. 2004; Stachel and Luth 2015) and the western Zimbabwe craton margin (Aulbach et al. 2017). A comparison of Figure 36a and 36b shows a marked difference in the determined temperatures using pre-set and iteratively determined estimates.

Age constraints indicate that at least some eclogitic diamond suites (Slave, Kaapvaal) formed in close temporal relationship to major subduction events (for details on suggested diamond forming processes during subduction, see further below). If so, they could have followed ancient subduction geotherms, different from those in the coeval continental lithosphere or modern subduction zones. These thermobaric gradients must have been lower than those obtained for exhumed ≥ 2 Ga metamorphic rocks (> 375 °C/GPa; Brown and Johnson 2018), in order to allow a pressure–temperature path consistent with diamond stability. A possible Archean slab gradient is roughly reconstructed in Figure 40a, based on following observations: Orogenic eclogite with an oceanic crustal protolith from the Archean Belomorian Belt indicates peak conditions of 1.85 GPa and 710 °C (Li et al. 2015). Elastic geobarometry on quartz inclusions in garnet from mantle eclogite at Mir (Siberian craton) yielded a pressure of 3 GPa at 850 °C, and was interpreted as reflecting entrapment during prograde metamorphism corresponding to a surface heat flow of 40–50 mW/m² (Alvaro et al. 2020). This may be linked to Archean subduction, as eclogite xenoliths from the Siberian craton have yielded Mesoarchean ages (Appendix Table 1). Combined with the pressure–temperature estimates for non-touching inclusions, these provide a predictive pressure–temperature path for subducting Archean oceanic crust (Fig. 40a).

In contrast to the low temperatures obtained from iterative solutions on non-touching inclusions (average 1040 ± 120 °C), constraints from the aggregation of N impurities in eclogitic diamonds, which is a function of N concentration, temperature and age, indicate average mantle residence temperatures of 1140 ± 50 °C for a 3 Ga eclogitic diamond age (Stachel and Harris 2008). As N-based temperatures reflect time-integrated thermal states, they may include transient heating events associated with melt migration. Moreover, not all eclogitic diamonds formed during relatively cold subduction, but may instead be associated with (renewed) growth during later melt metasomatism (Stachel and Luth 2015), as supported by multiple Proterozoic eclogitic diamond formation events based on inclusion dating (Wiggers de Vries et al. 2013; Gress et al. 2021), and by ca. 1.6 Ga eclogitic diamonds from Argyle recording atypically high temperatures (Stachel and Harris 2008).

Inhibited minor element redistribution in touching inclusions. As described above, the distribution of some minor and trace elements between clinopyroxene and garnet is temperature-dependent. Thus, touching inclusions and their xenolithic counterparts should show similar systematics, whereas non-touching inclusions might differ due to higher equilibration temperatures. Specifically, the distribution coefficient of both Na₂O and TiO₂ between clinopyroxene and garnet decreases with increasing temperature, as higher concentrations of these elements are accommodated in garnet (Aulbach 2020). If this should be the case, then D_Na2O,TiO2 for clinopyroxene–garnet would be expected to be higher for touching inclusions than for non-touching inclusions. However, when D_Na2O is considered as a function of T_KR88 calculated at 5GPa, it becomes evident that several touching inclusions in eclogitic and pyroxenitic diamonds have low D values at a given temperature (Fig. 41a). These low D values are more similar to those of (higher-temperature) non-touching inclusions than to (similar-temperature) diamondiferous xenoliths, which suggests that at least minor element redistribution for inclusions is inefficient. Figure 41b shows that a similar situation applies for TiO₂. The decreasing TiO₂ solubility in both garnet and clinopyroxene with decreasing temperature leads to exsolution of rutile, with the implication that this mineral is more likely to be observed in lower-temperature specimens, and that TiO₂ concentrations in bulk rocks are increasingly underestimated when only garnet and clinopyroxene are considered (Aulbach 2020). Minimum (rutile-free) bulk-rock TiO₂ contents reconstructed from non-touching inclusions as a function of T_KR88 loosely follow the xenolith trend, but with higher average contents consistent with higher entrapment temperatures and rutile solubility (Fig. 41c). However, there is also an off-set to higher TiO₂ at a given temperature compared to xenoliths, which may reflect the inability of included garnet to purge the excess TiO₂ from its lattice, first via nucleation of lamellae and then by recrystallization and formation of discrete rutile grains. One may speculate that this is due to the dryness of the inclusions where structural hydrogen, once lost to the inclusion–diamond interface (Nimis et al. 2016), is not replenished in contrast to matrix minerals. Conversely, the similar temperature estimates for touching inclusions and diamondiferous eclogites suggest that the redistribution of MgO and FeO, on which T_KR88 is based, is more efficient.

Source rocks of eclogitic diamond. Inclusions in diamonds are expected to reflect the mineralogy and composition of their various substrates (Boyd and Gurney 1986; Meyer 1987), in this case mantle eclogite. With respect to major elements, eclogitic inclusions cover nearly the entire range displayed by xenoliths, but are more common at bulk-rock MgO < 13 wt%. They seemingly overlap mostly with high-Ca and gabbroic eclogites (Fig. 38). Mantle eclogites with positive Eu anomalies (gabbroic class) and low total REE abundances (ΣREE) derive from sources that experienced plagioclase–olivine accumulation, partially followed by reaction with melt, which explains the wide range of compositions in this group (Aulbach and Jacob 2016). The fractionated melts develop negative Eu anomalies. Broadly, high-Ca and some high-Mg eclogites have more- and less-differentiated basaltic protoliths, respectively, whereas low-Mg eclogites may require Fe-rich protoliths. Inclusions follow the expected systematics of low-pressure cumulates vs. differentiated melts with respect to Eu/Eu* and ΣHREE (Fig. 42a; only HREE are summed to exclude effects of LREE loss via melt depletion and gain via metasomatism). Also, 65% of unpaired garnet and 58% of unpaired clinopyroxene inclusions have Eu/Eu* > 1.05. The proportion of gabbroic xenoliths (i.e., showing positive Eu anomalies) amongst diamondiferous xenoliths with reported REE contents is much higher at 80%. These numbers agree well with estimates that some 70% of modern oceanic crust is constituted by a gabbroic layer, the remainder being pillows, flows and sheeted dike complexes where magma freezes in layers and conduits (White and Klein 2014). Conversely, as noted for cratonic eclogite (Aulbach and Arndt 2019a), a mantle cumulate origin is precluded because eclogitic inclusions do not show the predicted positive correlation of Y with MgO, but weak negative correlation instead (r² = 0.25, n = 21; Fig. 42b).

Following the low-pressure igneous stage at spreading ridges, low- and high-temperature seawater alteration produces minerals with higher and lower δ¹⁸O, respectively, compared to the mantle (Muehlenbachs and Clayton 1972a,b). These signatures are still recognizable in many mantle eclogites (Jacob 2004), and in garnet inclusions (Fig. 35). Due to the crossover of alteration signatures and the oxygen isotope mantle range at intermediate temperature conditions of crustal alteration, or just a lack of alteration, mantle-like δ¹⁸O is not proof against low-pressure protoliths (Smart et al. 2012). Subsequent subduction and metamorphism of oceanic crust causes dehydration and loss of a LREE-enriched silicic partial melt in some, though not all eclogites (Schulze et al. 2003), especially in warm slabs. Eclogitic diamond sources now residing at temperatures ≳ 800 °C would have crossed the solidus of wet metabasalt (Schmidt and Poli 2014), which is supported by the pressure–temperature paths of exhumed ancient orogenic eclogites (Fig. 40a). Experimental phase relationships further indicate that this entails loss of SiO₂ and Na₂O, accompanied by an increase in MgO, FeO and CaO, while the presence of substantial amounts of garnet in the residue leads to compatible behaviour of the HREE and Y, and incompatible behaviour of the LREE during melting (Aulbach and Jacob 2016). Indeed, high MgO, low SiO₂ and Na₂O as well as low LREE/HREE in both inclusions and mantle eclogites have been interpreted as signatures of melt loss from eclogite (e.g., Ireland et al. 1994; Jacob 2004; Stachel et al. 2004), although low SiO₂ and LREE/HREE can also be a primary signature of oceanic crustal cumulates (Aulbach and Jacob 2016). The strong LREE-depletion in high-Ca and gabbroic eclogite xenoliths (Fig. 39) may causally reflect their higher original Na₂O content, because the jadeite component in clinopyroxene, together with quartz/coesite and H₂O, participate in melt production (Laurie and Stevens 2012). A higher initial Na level will ultimately cause higher levels of depletion after melt loss. Again, the majority of bulk rocks reconstructed from inclusions have LREE-depleted REE_N patterns most similar to gabbroic and high-Ca eclogites, but with some important differences (Fig. 39) that provide insights into diamond formation processes discussed below.

Depending on the fertility after initial dehydration melting at low pressure and on the redox state of the eclogite and infiltrating fluid, oceanic crust may experience second-stage partial melting during fluid ingress (Fig. 40a). Diamondiferous eclogites and inclusions do not extend to the extremely melt-depleted compositions (very low Ce/Yb_NMORB) observed in some barren eclogites, perhaps due to the failure of diamond to crystallize from fluids in such depleted rocks. Otherwise, as noted above, the proportion of inclusions (~60%) and diamondiferous eclogites (80%) with gabbroic protoliths is similar to that in modern oceanic crust, indicating that diamond formation occurs in all portions of the oceanic crust, with a preference for intrusive sections indicated by diamondiferous eclogites. Given mechanisms for eclogitic diamond formation discussed below, this need not, however, reflect the extent to which C is originally distributed in these rocks. Curiously, few inclusions have strongly gabbroic signatures (low whole-rock Y or HREE abundances, strong positive Eu anomalies, Fig. 42), possibly indicating that deep crustal cumulate sections are less propitious substrates for the formation of inclusion-bearing diamonds.

Source rocks of pyroxenitic/websteritic diamond. For inclusions in diamond, the websteritic paragenesis is compositionally less well defined than the eclogitic one, which probably reflects multiple formation mechanisms (Stachel and Harris 2008). Similarly, pyroxenite xenoliths, which include rocks referred to as “high-Mg eclogites” in the literature, have multiple suggested origins (Aulbach and Jacob 2016). Bimineralic pyroxenite xenoliths grading into orthopyroxene-bearing websterites and, for inclusions in diamond, the websteritic inclusion suite from Venetia have been interpreted as the products of the interaction of slab-derived silicic melt with peridotite (Aulbach et al. 2002, 2007; Smit et al. 2014b), based on the similarity to experiments where peridotite was hybridized with eclogite-derived melts (Yaxley and Green 1998). On the other hand, a large proportion of bimineralic pyroxenite grading into high-Mg eclogites (20–40% of most eclogite-bearing xenolith suites) shows evidence for an origin via metasomatic interaction of mantle eclogite with a small-volume carbonated ultramafic melt similar to kimberlite (Aulbach et al. 2020a). Some pyroxenitic inclusions share major element features with the metasomatized group. This includes high MgO and low FeO (Fig. 38a), high CaO (Fig. 38b) and high Cr₂O₃ at high MgO (Fig. 38c), all of which have been interpreted as reflecting addition of a high-temperature pyroxene from the metasomatic melt (Aulbach et al. 2020a). A single diamondiferous pyroxenite with trace element data (from the Siberian craton) has a pattern similar to median barren pyroxenite (Fig. 39). Diamond in pyroxenite xenoliths from the northern Slave craton, where kimberlite-like metasomatism has been recognized, may have a similar origin (Smart et al. 2009). However, the few inclusions in pyroxenitic diamonds with REE data have spoon-shaped patterns with steep positive slopes in the MREE_N–HREE_N that cross-cut those of pyroxenite xenoliths (Fig. 39). Similarly high resultant La_N/Pr_N and Lu_N/Gd_N are seen in only a few pyroxenite xenoliths. Finally, some pyroxenitic inclusions have lower MgO contents combined with relatively high FeO (Fig. 38a), and low ΣHREE contents (Fig. 42b), perhaps reflecting a deep crustal cumulate protolith, with insufficient Na₂O to metamorphose into eclogite with Jd-rich clinopyroxene.

The dearth of trace-element data makes it difficult to confidently distinguish between different scenarios for the origin of pyroxenitic diamond source rocks. As applies to high-Mg eclogite, an origin of pyroxenitic inclusions as direct mantle cumulates is precluded by the lack of the predicted element relationships proposed for precipitation of substantial amounts of garnet (e.g., correlated Y-MgO; Fig. 42b). The small proportion of pyroxenitic diamond (15%, of all pyroxenitic–eclogitic inclusions, based on paired and isolated clinopyroxene) suggests that this rock type is a subordinate diamond source. Note that Stachel and Harris (2008), based on different classification criteria (elevated Mg# and Cr₂O₃, low Ca content in garnet, presence of orthopyroxene), obtain a relative abundance for websteritic inclusions (amongst lithospheric diamonds) of only 2.3%, as opposed to 5.3% in this study.

Diamond formation from saline fluids. A group of eclogitic inclusions has very high Na₂O in garnet and K₂O in clinopyroxene, occupying a field where diamondiferous or barren eclogite does not occur (Fig. 33b). Garnets with Na₂O content > 0.4 wt% are off-set from trends of increasing Na concentrations with increasing temperature (inset in Fig. 33b), and therefore reflect a bulk compositional effect, as previously noted by Grütter and Quadling (1999). The same interpretation applies for K-rich clinopyroxenes, and therefore, these inclusion compositions reflect Na–K-rich bulk rocks. They are not associated with any major or trace element signature except low Cr₂O₃ and high Sr contents, although not enough trace-element data are available to further establish the nature of the metasomatic agent. This may point to diamond precipitation from saline high-density fluids, which are observed in (though not limited to) eclogitic diamond (Izraeli et al. 2001, 2004), and may be ultimately linked to subducted slabs based on observational data (Weiss et al. 2014, 2015, 2022), modelling (Sverjensky and Huang 2015) and experiments (Shatskiy et al. 2019). Omphacite occurring with saline high-density fluid in the same eclogitic diamond has elevated Na₂O (6.7 wt%) and K₂O (0.6 wt%; Weiss and Goldstein 2018). Any medium involved in the formation of diamond cannot be dissociated from the substrate in which the diamond forms (i.e., mineral inclusions and eclogite have the same composition at the locus of encapsulation in diamond). Therefore, the exclusive occurrence of high Na₂O in garnet and high K₂O in clinopyroxene inclusions in diamonds (Fig. 33b), whereas eclogite xenoliths lack these signatures in their garnet or clinopyroxene, indicates that such compositions do not survive during subsequent mantle eclogite evolution. This suggests efficient removal of this signature from the substrate during subsequent interaction with Na₂O- and K₂O undersaturated fluids or melts. If this interpretation is correct, 14% of eclogitic diamond (containing garnets with Na₂O content > 0.4 wt%) may have formed from saline high-density fluids.

Diamond formation from small-volume melts or COH fluids. Mantle eclogites are dominantly LREE-depleted, indicative of partial melt loss without subsequent re-enrichment. This is gauged by normalization to a putative low-pressure protolith with flat REE_N patterns, such as NMORB, and equally applicable to oceanic crust formed from a higher melt fraction, such as picrite. A sizable proportion of bulk high-Mg eclogites and pyroxenites (37%) show LREE-enrichment (Nd/YbNMORB or La/SmNMORB > 1), accompanied by a lowering of Jd (Fig. 43) and also of Ca# in garnet, which are both hallmarks of metasomatism by kimberlite-like melt (Aulbach et al. 2020a). As the lowering of garnet Ca# is accompanied by decreased partitioning of many incompatible elements into garnet, this leads to weak signatures of metasomatism based on trace elements in garnet alone (Aulbach et al. 2020a). From an fO₂ perspective, carbonated silicate melts, such as kimberlite, are stable to lower fO₂ than pure carbonatite (Fig. 40b), allowing for coexistence of melts with dilute carbonate content and diamond over a large fO₂ range (Stagno et al. 2015; Mikhailenko et al. 2020). The ability of kimberlite-like melt to be stable with eclogite to lower fO₂ than with peridotite implies carbon mobility over a wider range of conditions, and may thus help explain the overabundance of eclogitic relative to peridotitic diamonds (Stagno et al. 2015). In addition, small-volume melts (like kimberlite) or fluids imparting LREE-enrichment are important drivers of diamond formation in peridotite source rocks (Stachel et al. 2004; Stachel and Luth 2015; this work). However, less than a quarter of eclogitic-pyroxenitic bulk rocks reconstructed from inclusions (23%) show LREE-enrichment and associated high LREE_N/HREE_N (Fig. 43), the hallmark of mantle metasomatism by diamond-forming CHO-fluids and small-volume melts in peridotite. Thus, although the trace-element database for eclogitic/pyroxenitic inclusions in diamond is limited, it appears that diamond formation in the eclogite system is not dominantly linked to a typical mantle metasomatic signature.

Diamond formation during subduction. Age distributions (Smit et al. 2022) suggest formation of the first eclogitic diamond populations in or associated with Archean (Kalahari Craton) and Paleoproterozoic (Slave craton) subduction zones, perhaps linked to the onset of plate tectonics at ~ 3 Ga (Shirey and Richardson 2011). This initial formation was followed by renewed growth or formation of eclogitic diamond during Proterozoic tectonothermal events, up to at least the Mesozoic (Timmerman et al. 2017). The coincidence of eclogite xenolith ages with eclogitic diamond ages and with craton amalgamation in the Kaapvaal craton (ca. 3.0–2.9 Ga) and with collision at the craton margin in the Slave craton (ca. 1.9 Ga) is most readily explained by interaction of oceanic crust with C-bearing fluids liberated in subduction zones. The median composition of diamondiferous eclogites is most similar to that of barren gabbroic eclogites with respect to strong positive Eu anomalies and LREE depletion (Fig. 39). The median composition of whole rocks reconstructed from eclogitic inclusions does not have as strong a gabbroic character and is more similar to barren high-Ca eclogite. The MREE_N–HREE_N patterns of bulk rocks reconstructed from both eclogitic and pyroxenitic inclusions show marked positive slopes that cut across the flatter pattern of high-Ca eclogite (Fig. 39), resulting in high NMORB-normalized Yb/Gd. This slope is unlike that of potential protoliths and can also not be achieved by extraction of partial melt (modelled compositions in Fig. 39b). Similar patterns have been ascribed to disequilibrium processes involving large fluid volumes, such as those encountered in subduction zones (Aulbach et al. 2020b). In particular, serpentinite has been implicated as a major fluid source in subducting slabs (Rüpke et al. 2004), and disequilibrium REE_N patterns with LREE–MREE depletion and HREE enrichment have been observed in orogenic eclogites affected by fluids derived from deserpentization (Beinlich et al. 2010). In this scenario, the low LREE contents of diamondiferous eclogites and inclusions need not be (solely) due to melt extraction, but can result from auto-metasomatism (i.e., fluid indigenous to the slab), whereby the compatible HREE partition into garnet, and the incompatible LREE are removed with the fluid (Aulbach et al. 2020b).

The viability of a scenario of diamond formation via interaction of slab-derived fluids with oceanic crust is supported by thermal modelling, which indicates that carbon and volatiles in subducting slabs can escape shallow decarbonation and dehydration (Shirey et al. 2021). It can be assessed by considering phase relations of oceanic slabs and the stability of diamond vs. other carbon minerals as a function of pressure, temperature and oxygen fugacity. Oceanic crust dehydrates, accompanied by loss of a small amount of melt, at forearc depth (Spandler and Pirard 2013), where diamond is not stable (Fig. 40a). The small melt fraction explains the persistence of coesite (which participates in the melting reaction) in some mantle eclogites (Schulze et al. 2003). In contrast, pulsed dehydration of deep crust and oceanic mantle, which follows cooler subduction geotherms than slab tops (John et al. 2012; Laurie and Stevens 2012; Spandler and Pirard 2013), could flush oceanic crust within the diamond stability field. If ancient oceanic crust followed a cool subduction path, aided by the proximity of early-cooled cratonic lithosphere (Eaton and Perry 2013), it could enter the diamond stability field at pressures < 4 GPa (Fig. 40a). Whether the high-pressure interaction of eclogite with fluids causes partial melting, as suggested by Stachel and Luth (2015), depends on the refractoriness of the oceanic crust after initial dehydration melting and the redox state of the fluid, which determine the eclogite solidus.

Carbonate, disordered carbonaceous matter, and CaO, are efficiently mobilized in crustal COH fluids (Frezzotti et al. 2011; Menzel et al. 2020; Vitale Brovarone et al. 2020). These materials may be one factor contributing to a preponderance of Ca-rich compositions amongst inclusions and diamondiferous eclogites. As graphite and dolomite are observed in eclogite-hosted veins (Li et al. 2020; Zhu et al. 2020), and microdiamond occurs in fluid inclusions in oceanic subduction zones (Frezzotti et al. 2014), clearly subduction-related fluids can mobilize and supersaturate in carbon or carbonate. Almost all mantle eclogites have fO₂ where simple CHO fluids are near the H₂O-maximum or on the reduced side, increasingly rich in CH₄ (Fig. 40b). However, fluids at high pressures are more complex, containing dissolved solids (Kessel et al. 2005) and possibly a multitude of additional hydrocarbon species (Huang et al. 2017). Since massive fluid flow is invoked in the subduction model, diamondiferous eclogites may represent a fluid-buffered system, in which reducing fluids would impose their fO₂ on eclogite. This lends support to models of diamond formation via oxygen-conserving reactions in CHO fluids (Stachel and Luth 2015), occurring during interaction of deep deserpentinization fluids with overlying oceanic crust. At the relatively low, subsolidus temperatures in a subducting slab, oxygen-conserving diamond formation from fluids may also be achieved due to a drop in pH (Sverjensky and Huang 2015).

The small number of eclogite xenoliths reflecting fO₂ conditions above the H₂O-maximum (Fig. 40b) suggests that the stability of diamond in eclogite to 1 log unit higher fO₂ values compared to peridotite (Luth 1993) plays a subordinate role in explaining the relative overabundance of eclogitic diamonds relative to the small volume of eclogite in the lithospheric mantle. Deserpentization fluids have been inferred to be both oxidising (Debret and Sverjensky 2017) and reducing (Piccoli et al. 2019). Thus, percolation of more oxidized, CO₂-rich fluid into originally reduced eclogite would allow for diamond formation via reduction of CO₂. But the mixing of such fluids with different redox character alone would be a highly effective process. At the water maximum, only about 2% of other species (CO₂, CH₄) are permitted, and if, as an example, fluid 1 contains 4% CO₂ and fluid 2 contains 4% CH₄, they will precipitate 50% of their carbon as diamond upon mixing in an oxygen preserving reaction (Deines 1980; Luth and Stachel 2014). A CH₄-dominated fluid at lower fO₂ interacting with initially more oxidized eclogite at mantle depth might precipitate diamond coupled to reduction of Fe³⁺ in eclogite. Both interpretations are consistent with an evaluation of diamond formation mechanisms in eclogite calling for isochemical diamond precipitation from CH₄-dominated fluids or oxidation of methane on infiltration of eclogite (Smit et al. 2019b). In summary, the preponderance of diamonds forming in eclogitic over peridotitic substrates may reflect that serpentinized harzburgite can carry far more water than the crust, and, as it dewaters at greater depth than the oceanic crust, those fluids leave to form diamond in the overlying crustal portion of the slab now sampled as eclogites.

A composite present-day mantle eclogite/pyroxenite reservoir. Inclusions in eclogitic/pyroxenitic diamonds represent the most pristine record of the mantle eclogite reservoir at the time of encapsulation, with the caveat that the diamond-forming process itself may have introduced some modification (e.g., Ca–HREE enrichment during disequilibrium interaction with dehydration fluids in subduction zones). Inclusions do not reflect the kimberlite-related late (pre-eruption) metasomatism typically affecting the mantle lithosphere (and xenoliths during transport), unless their diamond host only formed during this process, as discussed for fibrous diamonds (Boyd et al. 1987; Gurney et al. 2010). Conversely, mantle xenoliths reflect the state of the eclogite/pyroxenite source at the time of typically much later kimberlite magmatism. In order to facilitate a comparison between inclusions, diamondiferous eclogites/pyroxenites and their barren counterparts, we calculate the composition of a composite present-day mantle eclogite/pyroxenite source, taking median compositions for each of the six classes weighted by their proportion. Pyroxenite xenoliths are included in the composite as many have geochemical signatures indicative of interaction of carbonated ultramafic melt with mantle eclogite (Aulbach et al. 2020a), and they therefore constitute an integral part of the evolution of the mantle eclogite source. Median compositions of the composite eclogite, diamondiferous eclogite and bulk rocks reconstructed from eclogitic inclusions are given in Table 4.

Mantle eclogite evolution. The mantle eclogite reservoir is affected by mantle metasomatism, as also recognized in its peridotitic counterpart (e.g., Dawson 1984). It encompasses stealth effects (addition of minerals forming part of the primary mineral assemblage), modal effects (addition of clearly metasomatic minerals such as phlogopite) and cryptic effects (changes to the trace element composition such as LREE-enrichment) (Harte 1983; O’Reilly and Griffin 2013). Metasomatic effects on mantle eclogite include a reduction of the jadeite component in clinopyroxene and grossular component in garnet, a decrease in bulk-rock FeO and Al₂O₃ contents and increase in MgO, Cr₂O₃, ± SiO₂, as well as LREE-enrichment accompanied by higher Sr, Pb, Th, U, ± Zr, ± Nb, and lower Li, Cu, ± Zn (Aulbach et al. 2020a). Strong metasomatism was suggested to be accompanied by a muting of typically crustal signatures, such as non-mantle δ¹⁸O and the size of Eu anomalies. Rare studies on eclogite xenoliths with inclusion-bearing diamonds have shown that the inclusions retain evidence for partial melting, while the host eclogite has been subsequently enriched in MgO, Cr₂O₃ and incompatible elements, and depleted in Na₂O and CaO by a kimberlite-like metasomatic agent (Ireland et al. 1994; Smart et al. 2012). A comparison of composite mantle eclogite with bulk rocks reconstructed from eclogitic inclusions shows that the mantle eclogite reservoir has gained 0.52 wt% CaO (+5% relative), 3.0 wt% MgO (+29%) and 0.06 wt% Cr₂O₃ (+87%), and lost 0.91 wt% Al₂O₃ (−6%), 2.2 wt% FeO (−18%), 0.58 wt% Na₂O (−24%), and 0.04 wt% K₂O (−56%). These changes are ascribed to the effects of mantle metasomatism over time. Lower minimum TiO₂ content (−0.35 wt%, −63%) in the mantle eclogite reservoir is linked to lower average temperatures (1050 °C at 5 GPa vs. 1220 °C for inclusions), and corresponding lower TiO₂ solubility in garnet, clinopyroxene and consequently reconstructed bulk rock, and higher inferred rutile modes (Aulbach 2020). With respect to trace elements, evolution of the mantle eclogite reservoir comprises an increase in Sr (+43%) and LREE abundances (e.g., +140% for La), and a dilution of Y (−22%) and HREE abundances (e.g., −42% for Lu), whereby Sm is least affected (−2%), and Zr and Hf also change little (−2%) (Table 4). These systematics are consistent with interaction with a small-volume melt that is LREE-enriched and HREE-depleted relative to bulk eclogite, such as kimberlite. The similar relative abundance of garnet with positive Eu anomalies (Eu/Eu* > 1.05) amongst inclusions (65%) and barren eclogites and pyroxenites (66%) suggests that the metasomatic evolution does not lead to significant muting of this crustal signature in garnet, whereas it is muted in clinopyroxene. Like TiO₂, lower Nb abundances in the xenoliths (−46%) are not (solely) related to metasomatism, but reflect the increased presence of accessory rutile, which strongly partitions Nb and Ta, at lower average temperatures (Aulbach 2020).

Diamond-destructive effect of mantle metasomatism. Diamondiferous xenoliths are expected to have experienced the metasomatic evolution recorded by the barren eclogite reservoir, with the important exception that they do not capture mantle processes that were diamond-destructive. A direct comparison with inclusions must consider that diamondiferous eclogites have a much stronger gabbroic signature (bulk-rock Eu/Eu*=1.16) than eclogitic inclusions (1.05). This difference may at least in part explain 1.7% higher bulk Al₂O₃, 7% higher CaO contents, nearly 20% lower FeO, and lower REE abundances (Table 4; Figs. 38a,b, 39b) in diamondiferous xenoliths. The increase in bulk MgO and Cr₂O₃ is not as marked as for barren xenoliths (+14% vs. +29% for Mg, and +10% vs. +87% for Cr), and there is no dilution of bulk Na₂O content in contrast to barren xenoliths (Table 4). Similarly, in diamondiferous xenoliths, Sr is less enriched (+12%) than in barren xenoliths (+43%), and the strong positive slope in the REE is retained (i.e., high Yb/GdNMORB). These systematics are taken to indicate that strong mantle metasomatism, with greater enrichment in MgO, Cr₂O₃ and some incompatible elements leads to diamond destruction, via dissolution into C-undersaturated melts. The intermediate TiO₂ content in diamondiferous eclogite (0.36 wt%; a minimum estimate as rutile is not considered) is consistent with its intermediate median temperature (1130 °C at 5 GPa). In contrast, for unclear reasons, median Nb abundances are highest in diamondiferous eclogites (0.24 ppm).

The higher median temperatures (pressures along conductive geotherms) recorded in diamondiferous compared to barren eclogites suggest that the low-pressure population was thinned out by diamond-destructive mantle metasomatism that particularly affects the midlithosphere, close to the lower stability limit of diamond (Aulbach et al. 2020a). The low Na₂O contents in garnet from barren eclogite and pyroxenite xenoliths therefore reflect the superposition of two effects: Firstly, the effect of lower average temperatures, and pressures along conductive geotherms, where diamond is not stable, and secondly, the effect of diamond-destructive mantle metasomatism, during which Na₂O is diluted (Aulbach et al. 2020a). Diamond destruction during pyroxenitization, with a marked lowering of bulk Na₂O and Jd component associated with strong metasomatism by kimberlite-like melt, may also explain why diamondiferous pyroxenite xenoliths make up only 6% of all diamondiferous xenoliths. Indeed, the diamond-destructive effect of kimberlite melt at mantle depth has been experimentally demonstrated (Fedortchouk et al. 2017; 2019). Thus, the enrichment in LREE over MREE (La/Sm) in almost a quarter of bulk eclogites and pyroxenites reconstructed from inclusions is probably not linked to such mantle metasomatism, except in a few localities (e.g., northern Slave craton; Smart et al. 2009), suggesting a delicate balance between diamond saturation vs. resorption in kimberlite-like melts, which is also documented by the inferred precipitation of fibrous diamond from kimberlites (Boyd et al. 1987; Gurney et al. 2010).

Compositional indicators for diamondiferous vs. barren sources. When both clinopyroxene and garnet are available for study, log D_Ce for clinopyroxene–garnet at a given garnet Ca# may be useful to predict the presence of diamond in eclogite, whereby the combination of low garnet Ca# and high log D_Ce reflects the small-volume melt metasomatism dominating in barren eclogites and pyroxenites (Fig. 37b).

The following relationship can be formulated for the recognition of diamondiferous eclogite substrates:

This equation predicts that 50% of diamond-free eclogite and pyroxenite may actually contain diamond. It accurately determines the presence of diamond in 97% of diamondiferous xenoliths and in 94% of the eclogitic inclusions. Based on garnet only, the density of inclusions and diamondiferous xenoliths is smaller at garnet MgO content > 16 wt%, as only 11% of diamond-associated garnets have this feature, compared to 31% of barren xenoliths, typically high-Mg eclogites and pyroxenites (Fig. 44a). Finally, not only elevated Na₂O (here > 0.04 wt%) in garnet can be indicative of a diamondiferous eclogite source (Robinson et al. 1984; McCandless and Gurney 1989), but also elevated TiO₂ contents (> 0.1 wt%). Despite unquestionable bulk compositional effects (Grütter and Quadling 1999), both of these minor elements increase in concentration in garnet with increasing temperature (pressure along conductive geotherms), therefore representing graphite- vs. diamond-stable conditions. More than 66% of diamond-free eclogite and pyroxenite xenoliths and 98% of inclusions plus diamondiferous xenoliths are correctly assigned by this combined criterion (Fig. 44b). The number of inclusions and diamondiferous xenoliths increases at values of Na₂O ≥ 0.07 wt% in garnet (Gurney 1984) and TiO₂ ≥ 0.2 wt% in garnet, matched by 37% of barren xenoliths, and 89% of inclusions plus diamondiferous xenoliths.

Eclogitic inclusions in lithospheric diamond have major and trace element systematics broadly similar to those of mantle eclogites, which originated as variably differentiated oceanic crust. Direct formation of eclogite from magma at mantle pressures is not supported. Non-mantle δ¹⁸O compositions in some inclusions and xenoliths point to a variably seawater-altered oceanic crust (AOC) protolith, which lost a small volume partial melt during subsequent dehydration melting in the course of subduction, causing some LREE-depletion. Known oxygen fugacity values of mantle eclogite are mostly below the FMQ buffer. Thus, the stability of diamond in eclogite to higher fO₂ than in peridotite plays a subordinate role in explaining the overabundance of eclogitic diamonds relative to the abundance of eclogite in the mantle lithosphere. Less than a quarter of inclusions (23%) and even fewer diamondiferous xenoliths (11%) show the LREE-enrichment that ensues from kimberlite–eclogite or CHO–fluid–eclogite interaction. Thus, eclogitic diamond formation is not dominantly linked to a typical mantle metasomatic signature, which in eclogite appears to be responsible for diamond destruction rather than growth. Therefore, the stability of diamond together with carbonated silicate melt at fO₂ conditions much lower than is the case for pure carbonatite, also is not important in explaining eclogitic diamond formation. Inclusions and diamondiferous xenoliths are depleted in MREE_N relative to HREE_N, which is not seen in barren xenoliths and possible crustal protoliths. This depletion is interpreted as a disequilibrium feature caused by interaction with large fluid volumes during subduction-related dehydration of underlying oceanic mantle. If so, oxygen fugacity values indicate that diamond may have formed either by reduction of CO₂ in relatively oxidized fluids, or during mixing of CHO fluids falling on either side of the water maximum, a feature observed during deserpentinization. The subduction environment, with the availability of large redox contrasts and large volumes of mobile fluids, may be favourable for the disproportionately higher amount of eclogitic than peridotitic diamond growth. Whether the carbon involved in this growth is derived directly from the dehydration of serpentinized peridotite or picked up by the dehydration fluid as it migrates through the oceanic crust may be resolved using multiple stable isotopes (e.g., C–N isotopes of diamond and O–S isotopes of inclusions compared to various portions of oceanic lithosphere). While elevated MgO and Cr₂O₃ contents in garnet and high D_Ce between clinopyroxene–garnet at a given Ca# are associated with barren eclogites, high combined TiO₂ and Na₂O in garnet are indicative of high-temperature (pressure along conductive geotherm) sources that were little affected by mantle metasomatism and have high diamond potential.

Kelsey Graversen (University of Alberta) is thanked for her tremendous help in adding full references (DOIs) and in error checking the inclusion-in-diamond database. Amber Yu (University of Alberta) upgraded the xenolith database into a truly useful tool and assembled an additional database for trace elements in xenolith minerals. Fanus Viljoen (University of Johannesburg), Steve Richardson (University of Cape Town), as formal reviewers, and Steve Shirey (Carnegie Institution), as our editor, significantly improved the manuscript. Herman Grütter (SRK Consulting) is thanked for valuable comments on an early version of the manuscript. Saebyul Choe of The American Museum of Natural History kindly shipped crush and a fragment of the primitive J4 xenolith to us. Andrew Locock and Yan Luo (both University of Alberta) are thanked for their help in analysing J4 mineral separates with EPMA and sector field ICPMS, respectively. T.S. acknowledges long-term support for his diamond research program through the Discovery and the Collaborative Research Development (CRD) grant programs of the Natural Sciences and Engineering Research Council of Canada (NSERC), and additional grants from the Canada Research Chairs program and the Canada Foundation for Innovation (CFI). S.A. thanks the German Research Foundation for their continued support, currently via grant DFG AU356/11. Diamond research would not happen without the continued generous support of major (principally De Beers and Rio Tinto) and many junior companies active in diamond exploration and mining. We sincerely thank all the enthusiastic geologists, engineers and diamond sorters in these companies who helped us with samples, gave of their expertise and occasionally organized some financial support.