Fluid Inclusions in Fibrous Diamonds

Most diamonds are metasomatic minerals, which means they grew from fluids or melts that moved through solid mantle rocks and chemically interacted with them. The involvement of fluids was recognized in the early days of diamond research (e.g., Sobolev 1960; Kennedy and Nordie 1968; Meyer and Boyd 1972; Harte et al. 1980) and was mostly attributed to kimberlite melts (either the erupting kimberlite, or a deeper proto-kimberlite composition). Based on Rb–Sr and Sm–Nd systematics of garnet inclusions in diamonds, Richardson et al. (1984) suggested that it involves “introduction and entrapment of asthenosphere-derived alkali, LREE and CO₂ enriched interstitial melt, which remained liquid until the time of diamond crystallization”. However, only sporadic reports on the occurrence of alkali-rich mineral inclusions or fluid- or melt-inclusions in diamonds were reported at that time. In later years, evidence for such fluids and the metasomatic origin of diamonds accumulated, including, the association of diamonds with other metasomatic minerals and with alteration zones in xenoliths (e.g., Schulze et al. 1996; Anand et al. 2004; Zedgenizov et al. 2018); trace element abundance patterns of garnet and clinopyroxene inclusions in diamonds that point to interaction with, or growth from melts that are extremely enriched in incompatible trace elements (Shimizu and Richardson 1987; Stachel and Harris 1997; Stachel et al. 2004 and references therein) and observations of fluids or melts in microinclusions in diamonds, which are the subject of this review.

Before proceeding to the observations of fluids in diamonds, it is necessary to define the terms ‘fluid’ and ‘melt’. The definition of ‘fluid’ in dictionaries is commonly “capable of flowing” (Merriam-Webster Dictionary), or “a substance that flows” (Cambridge and Oxford Learner’s dictionaries). Physics textbooks use a similar definition “a substance (or a phase) that flows under the application of a shearing stress”. This definition includes both liquids and gases, as well as supercritical fluids (Landau and Lifshitz 1987). In inclusion studies, a similar broad definition was suggested by Roedder (1984) and Bodnar (2003). The latter explained that some researchers relate to the state of the inclusion at room temperature and exclude glassy inclusions. However, he suggests including all inclusions that were fluid at the time of entrapment. This definition covers both high- and low-density fluids (Bodnar 2003). High-density fluids (HDFs) include silicate, sulfide, metallic, carbonatitic or saline supersolidus melts and dense supercritical fluids that are above the second critical endpoint (the termination of the solidus in a P–T space; Wyllie and Ryabchikov 2000; Kessel et al. 2005). At lithospheric pressures and temperatures, low-density fluids would have densities of ~1000 kg/m³. The most common are C–O–H-bearing supercritical fluids like H₂O, CO₂, CH₄, H₂, and their mixtures, together with the solute they carry (which can reach up to a few tens of percent). The values for the high-density ones are ~2000 kg/m³ or more, but past the second critical endpoint, the transition between low-and high-density fluids is continuous.

This review concentrates on HDFs that are encapsulated in diamonds. It begins with the historical perspective of the research of fluids in diamonds. We then review the shape and texture of fibrous diamonds which commonly carry HDFs, including the description of the three main forms of such diamonds: ‘fully fibrous’ where the whole diamond consists of fibers (~10 µm in diameter) and grows to develop cuboid {100} faces; ‘cloudy’ where a cuboid fibrous core is surrounded by a non-fibrous, octahedral zone; and ‘coated’ where a clear core is surrounded by fibrous overgrowth. This section also includes a short review of the chemistry of the diamonds themselves, focusing on nitrogen speciation as well as carbon and nitrogen isotope composition. The next section describes the various analytical techniques employed to explore the nature of the trapped fluids. Here the attitude is historic, and in the case of analytical techniques that are not common, or that do not provide compositional information (e.g., X-ray diffraction or transmission electron microscopy), we also present a review of the results. The following section focuses on the geochemistry of the HDFs trapped in fibrous diamonds and their four compositional types: silicic, which are rich in SiO₂, K₂O, Al₂O₃ and water, and form a continuous array with low-Mg carbonatitic compositions, rich in CaO, FeO, MgO and CO₂; High-Mg carbonatitic HDFs are distinguished by their very high MgO (>15 wt%) and high CaO, FeO, K₂O and CO₂; saline HDFs have high K, Na and Cl values, and contain varying proportions of FeO, BaO, CO₂ and water. This section ends by reviewing the direct and indirect evidence for the occurrence of such HDFs and other fluids in non-fibrous diamonds.

The discussion begins with a section on the temporal and geographical distribution of HDFs and a possible connection between HDF type and lithospheric lithology. The following section deals with estimations of pressure and temperature of diamond growth and HDF entrapment, and the current hypotheses on the origin of the four compositional types. We then review the role of HDFs in the formation of diamonds, where the collection of observations to date suggest they are the media from which the majority of lithospheric diamonds grew. The final sections highlight that all HDF types are excellent metasomatic agents, which impact the chemical composition of the continental lithosphere and possibly other mantle environments, emphasize the close relation between high-Mg carbonatitic HDF type and kimberlites, and discuss the possible role of HDFs in alkaline basaltic magmatism.

The earliest mention of fluids in diamonds was, most probably, by Bauer in his book “Edelsteinkunde” (1896, translated into Precious Stones, 1904). He described observations of liquid and gas-bearing microscopic inclusions within diamonds (as quoted by Melton et al. 1972 and Fesq et al. 1975).

Early studies of the three types of microinclusion-bearing diamonds were conducted by Shah and Lang (1963), Kamiya and Lang (1965) and Moore and Lang (1972), who applied X-ray topography to study their internal structure: a) cloudy diamonds with an internal cloud of microinclusions, b) coated diamonds in which a clear octahedral core is surrounded by fibrous growth, and c) cuboid diamonds that consist mostly of fibrous growth, sometimes with a small core (Fig. 1). Orlov (1977) summarized optical studies of diamonds and designated the cuboid diamonds as Variety III and the coated diamonds as Variety IV.

Angress and Smith (1965) used infrared spectroscopy to investigate the coat of a coated diamond but reported mainly on the nitrogen concentration. In a subsequent study, Chrenko et al. (1967) detected characteristic bands of molecular water (~3400 and ~1640 cm⁻¹), carbonate (1430 and 880 cm⁻¹) and nitrate (1360 and 835 cm⁻¹) in the IR spectra of the opaque coat of a diamond, but not in its transparent core (the presence of nitrate was not confirmed by later studies).

Melton and Giardini (1974, 1975) reported the composition of gases released by crushing diamonds under vacuum. Gas signals of transparent or light-yellow diamonds were similar to the system blank levels (0.2–2 × 10⁻⁵ cm³ STP, for an empty crushing chamber at 0 °C and 1 bar). However, the crushing of cubic and translucent diamonds at 200 °C yielded gas volumes that were an order of magnitude larger. The main gases released were H₂O and CO₂. Gas released by graphitization of diamonds contained mostly CO with minor H₂ and the quantities were >10⁻² cm³ STP (Kaiser and Bond 1959; Melton and Giardini 1976), but the H₂ may be attributed to a reaction of water during graphitization (Fesq et al. 1975).

Early attempts to determine the nature of the material trapped in the microinclusions by X-ray diffraction and X-ray fluorescence were not conclusive (Lonsdale and Milledge 1965; Seal 1966, 1970; Harris 1968). Prinz et al. (1975) found “fluffy, filamentous material” in two diamonds from the Democratic Republic of the Congo (DRC), but did not describe the diamonds. The composition obtained by electron probe microanalysis (EPMA) of this material was SiO₂ 67 wt%; TiO₂ 2.5 wt%; Al₂O₃ 14–17 wt%; FeO 2–3 wt%; MgO 1 wt%; Na₂O 0.2–2.5 wt% and K₂O 8.5–10 wt%. As mentioned below, this composition agrees roughly with that of later studies of microinclusion-bearing fibrous diamonds.

Fesq et al. (1975) used instrumental neutron activation analysis (INAA) to investigate the nature of impurities in diamonds. They examined groups of diamonds (~10–20 crystals in each sample) with mineral inclusions as well as diamonds with no visible inclusions. They concluded that in addition to mineral inclusions, the diamonds also carried picritic (high-Mg basaltic) melt associated with an H₂O + CO₂ rich component or phase that carried incompatible trace elements. The presence of a sulfide-rich phase was inferred from the correlation between Fe, Ni, Cu and Co. These impurities were found in diamonds with and without visible inclusions.

INAA of coated diamonds was attempted by Kodochigov and co-workers (according to Orlov 1977) and Glazunov et al. (1967). Bibby (1979) used INAA to determine the trace-element composition of a coated diamond. He found low levels in the core, but the coat carried up to 50 ppm Fe and a few ppm of Na, K, Ba and Ce. Based on the good intercorrelation of K, Na, Ba, and the REE and their poor correlation with Sc, Cr, and Mn, he suggested the presence of carbonate microinclusions.

Fibrous diamonds are imperfect single crystals, consisting of parallel microscopic fibers that are slightly misaligned from the diamond crystallographic axes. They have one set of crystal faces and they produce a single X-ray diffraction pattern (Orlov et al. 1980; Smith et al. 2011). However, they are distinct from the non-fibrous monocrystalline diamonds (i.e., octahedral, cubic and macles) that grew layer by layer with a higher degree of perfection. Although fibrous diamonds are monocrystalline, in view of these differences and the absence of a better distinction, in the present paper we will use the term ‘non-fibrous monocrystalline’ (NFMC) when addressing other monocrystalline varieties as opposed to fibrous ones throughout the paper.

Optical microscopy is a straightforward method for identifying fibrous growth. Microinclusions trapped in the fibrous zones result in a translucent or opaque appearance of the diamond. Under high magnification, the tiny microinclusions (<1 μm) can be observed (but not their internal structure; Fig. 1d). Crossed nicols reveal high birefringence by the diamond due to stresses that may be the result of the fibrous growth or the presence of microinclusions.

Fibrous growth leads to the development of cuboid faces (Lang 1974a), thus ‘fully fibrous’ diamonds have a cubic shape (Fig. 1a). If octahedral, layer by layer growth develops on a fibrous diamond, a ‘cloudy’ diamond forms with a cuboid, fibrous, inclusion-rich, translucent or opaque inner zone and an outer transparent NFMC zone (Fig. 1b). External fibrous overgrowth on an existing NFMC diamond produces a ‘coated diamond’ where the triangular, octahedral {111} faces conserve their original size, while the cubic {100} faces expand with the growing thickness of the coat (Fig. 1c).

X-ray topography can be used to reveal the details of the fibrous texture (e.g., Lang 1974a,b; Yelisseyev et al. 2004; Zedgenizov et al. 2006; Rondeau et al. 2007; Skuzovatov et al. 2011). The fiber diameter varies from 10 µm down to the resolution of the technique (<1 µm). The fibers grow with an average orientation parallel to the <111> direction, but individual fibers may be misoriented by up to 0.5º (Lang 1974b). Branching allows the fibers to fill the full volume of the growing diamond and leads to the formation of cuboid surfaces (parallel to {100}; Moore and Lang 1972; Orlov et al. 1982; Rondeau et al. 2007). The microinclusions (<1 µm) are scattered between the fibers (Moore and Lang 1972).

The fibrous zones have weak cathodoluminescence (CL), but it is still useful to emphasize different stages of growth. Most studies record the total intensity of the luminescence, which is sufficient to produce a map of the diamond growth layers (similar to the growth rings of a tree). An example is Fig. 1c of Klein-BenDavid et al. (2004) that reveals a transparent core and 5 concentric growth layers. In coated diamonds, the NFMC core is usually brighter than the fibrous coat, whereas, in cloudy diamonds, the fibrous inner zone is darker than the NFMC outer one (e.g., Fig. 3 of Zedgenizov et al. 2006, but see Rondeau et al. 2007 for a counterexample).

In summary, fibrous diamonds or growth zones are defined by the presence of fine fibers, as revealed by X-ray topography. They commonly carry microinclusions in variable numbers, but may also be clear of inclusions. Branching of the fibers leads to space-filling and the development of cuboid external faces.

Nitrogen is the main element that substitutes for carbon in the diamond lattice (see Green et al. 2022). Its concentration can reach levels of a few thousand ppm. Fibrous diamonds typically carry a high concentration of N (~60% with more than 500 ppm and 20% with more than 1000 ppm) that resides mostly in A-centers (two nitrogen atoms replacing two carbon atoms, see Green et al. 2022), but some fibrous diamonds show the yellow tint caused by C-centers (a single nitrogen atom substituting for a single carbon atom) and others carry both A- and B-centers (4 nitrogen atoms around a vacancy) and platelets (interstitial planes carrying carbon and nitrogen atoms; Speich et al. 2018). The concentration is determined from the IR spectrum of the diamond based on the unique absorption of each of the three centers (e.g., the CAXBD97. xls spreadsheet produced by Dr. David Fisher, Diamap of Howell et al. 2012 or Quiddit by Speich and Kohn 2020). When analyzing the spectrum of microinclusion-rich diamonds, care must be taken for the overlap between nitrogen absorption at 900–1400 cm⁻¹ and the major peaks of the silicates and phosphates. Most fibrous diamonds carry only A-centers and allow an easy escape from the problem. However, when B- or C-centers are present or suspected, one cannot simply use the standard fits and more elaborate treatment of the baseline and the deconvolution of the various peaks is needed (e.g., Kopylova et al. 2010; Weiss et al. 2018).

Taylor et al. (1990, 1996) compiled experimental data on the systematics of nitrogen aggregation (Evans and Qi 1982; Clark and Davey 1984) with ages and thermometric data of natural diamonds, and quantified the aggregation kinetics. They showed that it is controlled by the residence temperature and the initial amount of nitrogen (assumed to be equal to the total nitrogen measured). Most fibrous diamonds, with only A-centers, had to reside in the mantle long enough to allow the merging of C-centers into A-centers, but not too long to avoid conversion of the A- into B-centers (Fig. 2). For example, according to Taylor’s calibrations, a diamond that carries 1000 ppm nitrogen with 10% in C-centers could not have spent more than 10 Myr at 950 °C or 0.1 Myr at 1050 °C. Formation of a few percent of B-centers calls for longer times, ~100 Myr at 1150 °C and billions of years at ~1000 °C (Fig. 2). Figure 2 demonstrates that the estimation of absolute times is very imprecise (as the time axis is logarithmic), but the aggregation data allow for the separation of distinct growth events. In Fig. 2, an individual diamond would appear as a diagonal line representing the interdependence of time and temperature for a given nitrogen concentration and B/(A+B) ratio. A population of diamonds formed in a single event would appear as a group of closely spaced parallel lines (e.g., Kempe et al. 2021a).

NFMC diamonds cover a wide range of carbon isotope composition (δ¹³C) with a distinct peak at −5‰ (relative to the PDB standard). Diamonds carrying eclogitic mineral inclusions peak at −5±1.2‰ (at half maximum), but about one-third of them have lower isotopic compositions (−10 to −42‰) and a few have more positive δ¹³C values (0 to +3‰; Cartigny et al. 2014). Those with peridotitic inclusions have a somewhat wider peak of δ¹³C at −4.5 ± 1.5‰ (Cartigny et al. 2014), but very few diamonds fall outside this range (most of the light grey area in Fig. 3c consists of eclogitic diamonds). Fibrous diamonds have more negative δ¹³C values with a mode at −6.5 ± 1.2‰ (Fig. 3a,c). Early studies of fibrous diamonds found a narrow range between −6 and −8‰ (Galimov 1984; Javoy et al. 1984; Boyd et al. 1987; Cartigny et al. 2001), but later studies (Klein-BenDavid et al. 2004, 2007; Shiryaev et al. 2005; Zedgenizov et al. 2006, 2009, 2011, 2016, 2017; Skuzovatov et al. 2016; Timmerman et al. 2018a; Gubanov et al. 2019) widened the range, mainly towards heavier values (Fig. 3a).

Although more data is needed, the figure also shows that there is no correlation between the type of HDF and a specific range of carbon isotope compositions of its host fibrous diamonds.

The range of nitrogen isotope compositions (δ¹⁵N) of fibrous diamonds is also narrow (−8 to −2‰, relative to air standard) compared with the full range of NFMC diamonds (−24 to +12‰; see Figs. 7 and 8 of Cartigny et al. 2014). Only a handful of fibrous diamonds were analyzed for both the isotopic composition of their nitrogen and the composition of their HDF microinclusions (Fig. 3b), not enough to search for regularities and correlations.

Most fibrous diamonds exhibit uniform carbon isotope compositions, or small variations (< 2‰), but some are zoned. Some present a continuous evolution during growth and appear to evolve in different directions (e.g., the arrows in Fig. 3c). A few others exhibit sharp changes that are accompanied by a change in the chemistry of the trapped HDF (Klein-BenDavid et al. 2004; Shiryaev et al. 2005; Zedgenizov et al. 2011). As discussed in ‘Homogeneity and zoning’ below, this coupled variation is a strong indication that the diamond carbon is supplied by the HDF.

Optical microscopy of the microinclusions. High magnification optical microscopy allows observations of the individual microinclusions but is insufficient to resolve their internal structure (Fig. 1d). In some cases, it is possible to see the microinclusions arranged in linear arrays (e.g., Fig. 1 of Navon et al. 1988), but in most cases, their distribution appears to be random, except for radial variation in the number-density of the microinclusions (Fig. 1d). The number-density varies widely, from no microinclusions at all (commonly close to the rim) to more than 10⁸ inclusions/mm³ in densely populated bands, corresponding to distances of ~2 μm between neighboring microinclusions (Klein-BenDavid et al. 2006). Light reflected from polished surfaces also reveals the microinclusion-bearing zones. The microinclusions appear as tiny holes that are the origin of fine polishing grooves that extend from each hole.

Except for very rare cases, the fluid inclusions trapped in fibrous diamonds are smaller than 1 µm in size and are designated as microinclusions throughout the paper (where it is clear that we refer to microinclusions we sometimes use ‘inclusions’ for short). Most mineral inclusions in NFMC diamonds are larger than 1 µm and are referred to as mineral inclusions. However, some fibrous diamonds carry mineral inclusions that are smaller than 1 µm. We refer to these as micro-mineral inclusions. Larger mineral inclusions are extremely rare in the fibrous zones, while micro-mineral inclusions are not common in NFMC diamonds.

Transmission electron microscopy (TEM) and high-resolution scanning electron microscopy (SEM). The first glimpse into the microinclusions was achieved using TEM (Lang and Walmsley 1983). In these early days, a 0.5 mm thick diamond wafer was sawn and thinned first by conventional polishing (down to ~50 µm) and then by ion beam down to electron transparency. More recent works by Wirth and coworkers used a focused ion beam (FIB) of gallium ions to cut ultra-thin foils (15 × 10 × 0.1–0.2 µm) that can be mounted directly for examination (Wirth 2004). Lang and Walmsley (1983) found numerous, mostly empty holes, 100–1000 nm in diameter, but in a few cases some material remained, and they succeeded to identify one of the phases as apatite. Energy dispersive spectroscopy (EDS) revealed clear peaks of P and Ca and convergent beam electron diffraction (CBED) revealed hexagonal symmetry with a = 9.55 Å. Using similar methods, Walmsley and Lang (1992a,b) found biotite (with excess silica) and ferroan dolomite in other fibrous diamonds. Guthrie et al. (1991), Klein-BenDavid et al. (2006) and Logvinova et al. (2008a, 2011, 2015, 2018, 2019a) provided more comprehensive descriptions of the microinclusions, studying both open and intact inclusions. Bright-field images reveal microinclusions, a few tens to a few hundred nanometers in size, that were sliced open during the preparation of the 10 nm thick foils. Microinclusions in the fibrous diamonds are dominantly multiphase assemblages that consist of a few mineral phases and a hole that held a low-density fluid before the inclusion was opened (Fig. 4). In some cases, amorphous material is also present. The mineral assemblage is secondary and grew from the trapped HDF during cooling, leaving a residual, low-density hydrous phase.

TEM allows the use of various techniques to identify the phases and determine their composition and structure. As CBED leads, in many cases, to fast destruction of the phase under the focused beam, high-resolution electron microscopy (HREM) is used for lattice fringe images of the various phases, and a Fast Fourier Transform algorithm is used to compute diffraction patterns from these images. Cell symmetry and parameters derived from the patterns are compared with libraries and literature data and allow identification of the phases. EDS spectra can be obtained from very small volumes and yield the composition of phases, down to a few nanometer scales (Fig. 4), but care should be taken as the beam can pass through two, and even more, phases that overlie each other. Electron energy loss spectroscopy (EELS) is used to detect the presence of various elements and to distinguish the chemical environment of various elements, e.g., to tell carbon in carbonates from the diamond carbon. Further details on the FIB and TEM-related techniques are provided in Wirth (2004, 2009) and Jacob et al. (2014).

The inclusions and their internal structure can also be imaged using high-resolution scanning electron microscopy (SEM) and the various crystals can be analyzed. However, the need to coat the surface disturbs either the imaging (carbon coating) or the analyses (gold coating).

Microinclusions with high-Mg carbonatitic compositions carry dolomite and phlogopite as the major phases with magnesite, Ba- and Sr-bearing carbonates, apatite, sylvite, rutile, ferropericlase, magnetite, clinohumite and sulfide, and a fluid phase that was lost during sample preparation (Klein-BenDavid et al. 2006; Logvinova et al. 2011, 2019a). Pyroxene was found in a single microinclusion (Klein-BenDavid et al. 2006), but it may have been trapped as a solid mineral together with the fluid (as described by Izraeli et al. 2001). Inclusions carrying saline HDFs crystallized KCl, ferroan dolomite, Ca–Sr–Ba-carbonate, mica, and apatite and single occurrences of halite and Ba–P phosphate (Klein-BenDavid et al. 2006; Logvinova et al. 2011). In inclusions with low-Mg carbonatitic HDFs the secondary assemblage consists of dolomite and Ca-carbonate (sometimes with Ba and Sr), sheet-silicates, apatite, KCl, ilmenite, magnetite, ferropericlase and sulfides. The sheet silicates were identified as biotite (Walmsley and Lang 1992b) or phlogopite (Logvinova et al. 2008a) and the sulfides include pyrrhotite, Ni-sulfide and K–Fe-sulfide (Logvinova et al. 2008a). When compositions are more silicic, quartz crystallizes and the sheet silicates are phengite and in some cases paragonite (Guthrie et al. 1991; Kvasnytsya and Wirth 2009; Logvinova et al. 2019a). Both are rich in silica and alumina. Klein-BenDavid et al. (2006) obtained compositions that fit high-silica mica with up to 7.6 Si atoms per formula unit (22 oxygen atoms). The above phases are joined by Ca ± Sr ± Ba-carbonate, a single occurrence of sulfide and some unidentified Al-Si rich phases (Logvinova et al. 2019a).

Fourier-transform infrared spectroscopy (FTIR). Infra-red spectroscopy was the analytical technique that first detected the presence of water and carbonates in the microinclusions (Chrenko et al. 1967). Navon et al. (1988) used FTIR spectroscopy to detect water, carbonate, apatite, a sheet silicate and an additional silicate phase (which was later identified as quartz, Navon 1991) in coated and cuboid diamonds from the DRC. They also estimated the carbonate and water concentrations based on absorption coefficients they measured or adapted from the literature. Fig. 5 presents a typical spectrum of a diamond carrying silicic to low-Mg carbonatitic HDF.

In 1991 Navon recognized that the observed absorption peaks at 782–786 and 809–813 cm⁻¹ were caused by secondary quartz in microinclusions. These peaks were shifted from their 1 bar position (779 and 798 cm⁻¹) because of elevated internal pressures of (1.5–2.1 GPa) in the inclusions. He used this pressure at room temperature to show that the diamonds originated from the shallow part of the diamond stability field in the lithospheric mantle. A somewhat lower and uniform pressure of 1.5 ± 0.1 GPa was reported by Schrauder and Navon (1994) for diamonds from Botswana. Kagi et al. (2000) used near-IR to provide another indication for high internal pressure within the microinclusions. In addition to liquid water absorption at 5180 cm⁻¹, they identified a shoulder at 5000 cm⁻¹, which they attribute to ice-VI at 1.9 GPa, similar to the pressure indicated by the shift of the quartz absorption.

Zedgenizov et al. (2004) pointed out that Udachnaya diamonds are richer in the carbonatitic component relative to diamonds from DRC and Botswana. They also noted that water absorption in the Udachnaya diamonds extends to lower wavenumbers and, as a result the intensity ratio of the stretching and bending modes of water I₃₄₂₀/I₁₆₅₀ is lower than that of the African diamonds. Weiss et al. (2010) calibrated the absorption of calcite, dolomite, magnesite, apatite, quartz and phlogopite, and combined them with that of water (Thompson. 1965; Venyaminov and Prendergast 1997) to measure the concentration of the various secondary phases in the diamonds. The IR data agree well with modal calculations based on EPMA data (see ‘H₂O and CO₂ in the HDFs: Connecting EPMA and FTIR’ below). Weiss et al. (2013a) found that the width of the water stretching mode and the position of the bending mode vary between HDF types and used it to identify the nature of the trapped HDF (saline, high-Mg carbonatitic, or silicic to low-Mg carbonatitic; see ‘Major element compositions’ below) based on its IR spectrum. This explains the observed differences in I₃₄₂₀/I₁₆₅₀ mentioned above between Siberian and African diamonds (Zedgenizov et al. 2004).

Together all these studies turn FTIR analyses into a powerful tool that yields important information on the type of the trapped HDF, the identity and abundance of most of the major phases: carbonates, silicates, phosphates, and water (but not halides), and in the case of silicic HDFs, even an estimation of the internal pressure within the inclusions. Additional absorption bands at 2340 and 669 cm⁻¹ appear in many spectra but in most cases, they are caused by atmospheric CO₂ in the spectrometer. A similar case can be made for the absorption at 2700–3000 cm⁻¹ that may be the result of surface hydrocarbons, but it is also possible that these bands are contributed by CO₂ and/or organic compounds in the inclusions (Kopylova et al. 2010).

Micro-Raman spectroscopy. Raman spectroscopy was used by Zedgenizov et al. (2004) and allowed the identification of carbonates, silica glass and apatite in a few microinclusions in fibrous diamonds from Udachnaya which were large enough for analysis. They also found an olivine inclusion and graphite. The analyses did not capture the multiphase nature of the inclusions (aside from the detection of apatite and carbonate in a single inclusion). Most of the inclusions are too small and do not yield useful spectra (Kopylova et al. 2010).

X-ray diffraction (XRD). Another method used to identify the secondary assemblage in the microinclusions is X-ray diffraction. Harris (1968) pioneered the application of this technique to diamond studies and it is now used extensively to characterize mineral inclusions in diamonds and to determine their internal pressure. However, its application to fibrous diamonds was always problematic and did not yield useful results. Kopylova et al. (2010) reported lines corresponding to micas and clay minerals, but noted that the technique failed to identify carbonates or apatite, two main phases in the microinclusions. Smith et al. (2011) conducted the most comprehensive XRD study of fibrous diamonds. They also concluded that the technique fails to identify some important phases. Nevertheless, they were able to confirm the presence of sylvite, halite and Ba–Cl hydrate; dolomite, ikaite (CaCO₃·6H₂O), Na–Mg carbonate, two types of Ba–Mg carbonates and an unidentified phase with a 2.63 Å band. The powder-like patterns confirmed the random orientation of the secondary phases within the microinclusions.

Secondary ion mass-spectrometry (SIMS). This technique was the first to yield a complete chemical analysis of the major elements in the microinclusions (Navon et al. 1988). They used the O⁻ beam of the Cameca IMS-3F at Caltech to drill into fibrous diamonds and read the intensities of the secondary positive ions formed by the diamond carbon and the various major elements in the microinclusions. Ion yields were calibrated via EPMA analysis of the volume that was later drilled by SIMS, and by analysis of mixed graphite and glass pellets. The yields were high and the background, measured on gem-quality diamonds, was very low. However, the analysis of a volume with a few hundred microinclusions, some of them larger than others, produced large fluctuations in the intensity of the various ions. As the ions were counted sequentially, this was problematic and very long counting times were needed to reach a stable average counting rate (Navon 1989).

Navon et al. (1988) compared the SIMS results to those of EDS analyses of individual, subsurface microinclusions and instrumental neutron activation analysis (INAA) and reported good agreement between element ratios determined by the three methods. However, the complex sample preparation and the long duration of the SIMS analysis needed to obtain precise counting statistics, compared with the ease and accuracy of EPMA-EDS analysis of the major elements in individual microinclusions (see next section) led to the prevalence of the latter in following studies.

Electron-probe microanalysis (EPMA). Electron-probe microanalysis of individual, subsurface microinclusions is the main technique used to analyze the major element composition of the trapped HDFs (e.g., Schrauder and Navon 1994; Izraeli et al. 2001; Tomlinson et al. 2006; Zedgenizov et al. 2009). Navon (1989) showed that it is more accurate than broad beam analysis of many inclusions. The inclusions are identified by comparing back-scattered electron (BSE) and secondary-electron (SE) images of the same area. In the BSE images, the microinclusions appear as bright fuzzy dots due to their high electron density relative to the diamond. However, they can be mistaken for dust particles. This is solved by examination of the SE image, which presents surface topography. The dust is clear and sharp, but the subsurface microinclusions are not visible. An electron beam with a typical current of 10–15 nA is accelerated to 15 KeV and focused to a ~0.5 μm diameter or less at the center of the inclusion. The X-rays generated from the microinclusion and the surrounding carbon matrix are analyzed by an energy dispersive spectrometer (EDS) and wavelength dispersive spectrometers (WDS). Using a standard correction procedure and mineral standards for each of the elements of interest, the number of counts for each element is translated into element concentrations.

Monte Carlo simulations of electron scattering indicate a penetration depth of ~2 μm into the diamond and a similar range of scattering to the sides (at 15 kV). As a result, most X-rays are emitted by the diamond carbon atoms and only a few percent of the calculated analysis are due to elements in the microinclusions. The correction procedure calculates the sum of chlorine and the major oxides and assumes that the difference to 100% (commonly >95%) is all carbon. Finally, to get the composition of the HDF itself, the total of chlorine and oxides is normalized to 100%, plus accounting for excess calculated oxygen due to the presence of Cl (see ‘Introduction’ in the supplementary compilation of HDF data, Weiss et al. 2022). In reality, the concentration of the oxides in the HDF is lower, as water and CO₂ (in carbonates) must be included as well. However, because they are measured by FTIR in a different volume, we must first normalize each EPMA analysis in order to combine them and calculate the average composition for the diamond (or a zone in the diamond). Only then, the average EPMA composition and the FTIR dataset can be merged to provide the full major element composition including the volatiles (see next section).

The very low background produced by the carbon matrix allows observations of very small peaks, but the low concentrations of the various elements still lead to large relative errors in the composition of individual inclusions. Jablon and Navon (2016, their Fig. A3) estimated the precision and concluded that the error for major elements of the inclusion is a few percent (relative), but it increases to tens of percent for the minor elements. However, except for a few zoned diamonds, the composition of the HDFs in any one diamond is uniform. Thus, the standard error on the average composition of the HDF in each diamond is reduced by a factor of $n−1$⁠, relative to the standard deviation of the individual analyses. Commonly, 30–50 inclusions along a few profiles are analyzed for each diamond. Plotting the profiles allows verification of homogeneity. Most diamonds are homogeneous, and the standard errors are smaller by factors of 5–7 relative to the standard deviation of the individual analyses. When calculating the average composition, it is important to remove analyses of microinclusions that trapped discrete minerals or mixtures of mineral+fluid (Izraeli et al. 2001; Tomlinson et al. 2006; Weiss et al. 2018; Weiss and Goldstein 2018). As the diamonds are cleaned with HF, fluorine concentration is monitored, and F-rich analyses, which are indicative of exposed inclusions, are also removed.

Other methods. Information on the concentration of some major elements can be obtained using INAA and LA-ICP-MS, along with data on many trace elements. The two techniques are discussed in section ‘Trace element analyses’ below.

To fully characterize the HDF major element composition, it is necessary to integrate both EPMA and FTIR data. Izraeli et al. (2001) provided a rough estimate of carbonate ions as the sum of the molar proportions of Mg + Fe + Ca and calculated the water content using the relative proportions of carbonates and water determined from the IR data. Klein-BenDavid et al. (2007) accounted, in a simplified manner, for cations taken by chlorides, silicates and phosphates and Weiss et al. (2010) provided a better framework. Below, we present their method and examine one more case.

We begin with translating the EPMA data into modal proportions, using the various minerals of the secondary assemblage as a guide. Silicates, phosphates and halides are determined, and the excess cations are assumed to be compensated by carbonate ions. First, the molar proportion of halides and apatite is calculated based on the Cl and P₂O₅ concentrations determined by EPMA. If Cl < K + Na, the excess Na is added to the carbonate. If Cl > K + Na, the excess Cl bonds with Ca. The case of the silicates is more complex. IR and TEM data indicate that the silicate phase is mostly phlogopite-biotite solid solution, K(Mg,Fe)₃AlSi₃O₁₀(OH)₂, but that silica-rich mica is found in the silicic HDF microinclusions (Klein-BenDavid et al. 2006). In most cases, the Si/Al ratio of HDFs is larger than 3 (Table 1), and both micas are necessary.

In pure phlogopite-biotite, Fe + Mg = Si = 3Al and K/2 = Si/6. We assume that Ti⁴⁺ resides in the phlogopite, and obtain:

If Al < Si/3, we assume that the mica is: K(Mg,Fe)_3−xAl_1−xSi_3+xO₁₀(OH)₂, which leads to:

The carbonate is then (Weiss et al. 2010):

In the case of silicic HDFs, where Fe + Mg < 3Al < Si, we assume that all the iron and magnesium, as well as the Al, Ti and Si are in the silicate phases (mica and quartz) and possibly in titanates, and estimate the K bound in the silicates as K_(silicates) = Al. In this case, the carbonate is (Weiss et al. 2010):

If Al > Si/3, the excess Al should enter the octahedral sites (muscovite solid solution) and with Mg_(mica) + Fe_(mica) = 3(Si − Al)/2 and K_(mica)/2 = Si/6, the proper equation is:

Once the weight fraction of the CO₂ component is determined from the molar proportion of carbonates, the water is estimated based on the CO₂/(CO₂ + H₂O) ratio determined from the FTIR absorption bands of water and carbonates. The main obstacle here is that there is no available absorption coefficient for the residual hydrous fluid in the microinclusions, and even for pure water, the two values in the literature differ by ~25%. Venyaminov and Prendergast (1997) determined ε_water = 109 (AU/cm)/(mole/liter), compared with the old value of 86 (AU/cm)/(mole/liter) of Thompson (1965). We use the newer value. Using the older, the water proportion will be higher by 15–20%. Now it is possible to determine the composition of the HDF as wt% of all oxides + Cl + CO₂ + H₂O (Table 2); in molal proportions (moles per 1 kg or 55.5 moles of water, Table 3) or as wt% of the major components: silicates (+titanates), carbonates, halides, phosphates and water (Table 4).

As can be seen in Fig. 6, this procedure works well. The comparison is close for high-Mg carbonatitic HDFs, which have relatively simple secondary mineral assemblage—mostly dolomite, calcite or magnesite, and less (K,Na)Cl (not active in IR), phlogopite, apatite and a hydrous fluid with dissolved halides and carbonates. The fit is reasonable in the case of the low-Mg carbonatitic compositions, but poor for the silicic HDFs. The likely reasons are that the mica is no longer phlogopite, but rather, a Si-rich mica (Klein-BenDavid et al. 2006; Logvinova et al. 2019a) with different absorption coefficient, the presence of an unidentified Al-Si-rich phase (Logvinova et al. 2019) or silicic glass (Zedgenizov et al. 2004). Still, it shows that FTIR spectroscopy alone can be used for identifying the type of HDF present and to provide a semiquantitative estimate of the mineralogy and hence the chemistry of all the major constituents of an HDF.

Instrumental neutron activation analyses (INAA). This technique involves irradiation of the sample with neutrons in a nuclear reactor. Nuclear reactions transform (activate) the elements in the sample into short-lived isotopes that emit γ-rays upon their decay. The γ-rays are counted to determine the concentrations of the new isotopes and consequently, the abundance of their parents in the diamond.

Bibby (1979) reported the first INAA study of a coated diamond, but did not have all the major elements needed to obtain the composition of the HDFs. Fragments of coated and fully fibrous diamonds from DRC and Botswana were also analyzed by Palme and Spettel (unpublished, reported in Navon 1989). Schrauder et al. (1996) used INAA to determine a large number of elements in a suite of diamonds from Botswana. Diamonds were fragmented and hand-picked before cleaning in acids in order to expose and avoid cracks. They were then irradiated, cleaned once more, counted 5, ~10 and ~40 days after the irradiation, and analyzed for the concentrations of Na, K, Fe and 28 trace elements in the diamonds. Na/K and Fe/K ratios obtained via INAA compared well with EPMA measurements (Schrauder and Navon 1994) and allowed re-normalization and determination of the trace element concentrations of the HDFs themselves. Br, I and U (along with a few major elements) were also measured following their transformation to noble gases via neutron activation and noble gas mass spectrometry (see ‘Noble gas and halogen analysis’ below).

Proton induced X-ray emission (PIXE).Rege et al. (2005) presented results obtained by Griffin et al. (1993) using PIXE (also known as proton probe, Ryan et al. 1990). Two diamonds previously studied by Schrauder et al. (1996) were re-analyzed and the duplicate analysis of Ca, Ti, Cr, Fe, Ni, Cu, Zn, Sr, Zr and Ba were used to calibrate a standard for LA-ICP-MS.

Early mass-spectrometry—combustion and crushing. Early analyses of trace elements in diamonds used emission spectroscopy to analyze impurities in the ash left after burning the diamond, but no specific analysis of fibrous diamond was reported (see Bibby 1982 for review). Akagi and Masuda (1988) burnt cuboid fibrous diamonds in a platinum foil within a quartz tube under oxygen flow. They collected the residue from the Pt foil and the gas-trap and analyzed the solution using isotope dilution mass spectrometry. Akagi (1999) crushed the diamonds, leached soluble elements first in water and then with hydrochloric acid, and determined the concentration of Rb, Sr, Ba, REE and ⁸⁷Sr/⁸⁶Sr ratios.

Laser-ablation inductively coupled plasma mass-spectrometry (LA-ICP-MS). The development of LA-ICP-MS allows for a much faster and easier measurement of some major and many trace elements in fibrous diamonds. The laser beam is focused into a 20–100 μm spot on the diamond surface ablating the diamond and its microinclusions. Part of the ablated material accumulates around the ablation pit, but most of it is carried into the plasma torch, ionized and analyzed using a quadrupole mass-spectrometer.

The technique was first applied to NFMC diamonds by Watling et al. (1995), who defined their investigation as preliminary. It took 8 more years for Resano et al. (2003) to try again, using an improved laser and mass spectrometer set-up, but here too, the purpose was looking for chemical fingerprint of diamond provenance for identification of conflict diamonds, and no fibrous diamonds were analyzed. Wang et al. (2003) also reported concentrations of 28 elements determined using LA-ICP-MS, but provided no details.

To date, most LA-ICP-MS analyses of fibrous diamonds have been carried out at the GEMOC lab at Macquarie University, Australia. The method was developed by Rege et al. (2005) and utilizes a 266 nm Nd:YAG laser focused into a custom-made ablation cell. Helium is used as the carrier gas and is mixed with argon before its introduction into the plasma torch of an Agilent 7500 ICP-MS. They introduced a cellulose standard that is doped with ~1000 ppm of a variety of elements. Carbon is also measured on the sample and the standard, allowing the translation of count-rate ratios into the concentration of the various elements in the diamond. A comparison of the concentrations determined by LA-ICP-MS with those determined by INAA (Schrauder et al. 1996) and by PIXE (Griffin et al. 1993) indicated good accuracy with most elements falling close to the 1:1 line (Rege et al. 2005). Rege et al. (2010), Araújo et al. (2009), Zedgenizov et al. (2007, 2009, 2011, 2018), Weiss et al. (2009, 2011, 2013b, 2018) and Weiss and Goldstein (2018) all used the Macquarie system and procedures for determining trace element concentration in the diamonds and their HDFs.

Tomlinson et al. (2005) used an ArF excimer laser at 193 nm, a GeoLes illumination system and an Elan 6100 ICP-MS to analyze 5 coated diamonds from DRC. They used a NIST 612 glass standard, but no carbon standard and thus could report only element ratios. In a later paper, Tomlinson et al. (2009) assumed that matrix effects are not too important and used the NIST 612 to derive concentration in the diamond. They used the carbonate content as measured by FTIR and EPMA in the diamond and the HDF, respectively, to get the trace element content of the HDF. Dalpé et al. (2010) used a CRM-S2150 doped oil standard.

As diamonds have a simple matrix, consisting mostly of carbon and minor nitrogen, the number of significant interferences is limited to ion pairs of C, N, O and Ar (see Supplementary Table 2 in Rege et al. 2010). These interferences affect many of the major elements, up to ⁵⁸Fe = ⁴⁰Ar¹⁸O. However, the high abundance of the major elements of the HDFs still allows for reasonable precision in their determination. Weiss et al. (2008) compared element/Fe ratios for the major elements determined by LA-ICP-MS with those measured using EPMA and found good agreement between the two methods. They followed Schrauder et al. (1996) in using potassium that was measured by both EPMA and LA-ICP-MS to translate the concentrations in the diamond to concentrations in the HDF.

We note that, in principle, a carbon standard is not needed. A preliminary study in our lab at the Hebrew University found similar counting-rate ratios for various element pairs comparing analyses of the pure NIST 610 with analyses of pressed pellets made of graphite powder to which 20% of ground NIST 610 were added (Elazar and Weiss 2021). The similarity suggests that it is sufficient to measure the NIST standard. That means that if the major elements are determined by EPMA, and potassium, and/or other elements, are measured by both techniques, the element/K concentration-ratios determined using the NIST standard can be converted to the trace element concentrations in the HDF.

For example, for the case of La:

where d = diamond,

st = standard,

ppm = parts per million,

cps = count per second,

$[K2O]wt%HDF$ = the concentration of K₂O in the HDF as obtained by normalizing the concentrations of all major oxides and Cl measured by EPMA to 100 wt% (or, alternatively, allowing for water and carbonate as well) and the factor at the end translates $[K2O]wt%HDF$ into $[K]ppmHDF$⁠.

Noting the similar concentration of K₂O in silicic and carbonatitic HDFs, we suggest that in the lack of major element data, the trace element concentrations in the HDFs can be estimated using K₂O abundance of 9 wt% for silicic and carbonatitic HDFs and 20 wt% for saline compositions (Table 2). The nature of the fluid can be determined using IR spectroscopy. Although this is an approximation, the added inaccuracy will only lead to a minor uniform vertical shift of all elements on the trace element abundance diagram.

Off-line laser ablation of fibrous diamonds. In an effort to lower the limit of quantification and to open the way for trace element analysis of NFMC diamonds and Sr–Nd–Pb isotopic analysis of fibrous diamonds, Pearson and co-workers developed the off-line approach (McNeill et al. 2009; Klein-BenDavid et al. 2010, 2014). Following cleaning in ultra-pure HF and HNO₃ and weighing, the diamond was placed in a sealed PTFE beaker with an optical window and ablated using a UP-213 New wave laser system. The sample was moved under the laser so that a large area (a few hundred µm in size) was ablated. The ablated material was collected in HCl that was dried and taken in HNO₃ before splitting into two aliquots: 10% for trace element and 90% for isotopic analysis. The carefully pre-cleaned beakers and windows as well as the use of ultrapure acids in an ultraclean lab led to blanks at the picogram levels and allowed trace element and isotopic analyses of ~500 μg of ablated diamond.

The technique yielded very accurate results for fibrous diamonds, confirming the on-line analyses (see Fig. 2 and 4 of Klein-BenDavid et al. 2014). It was used successfully in trace element analyses of a few NFMC gemmy diamonds, where concentrations were very low, but the patterns resembled those of fibrous diamonds (Melton et al. 2012; Krebs et al. 2019), and for radiogenic isotopic analysis of fibrous diamonds (following section). Recently, a new custom-made off-line ablation system equipped with a Spectra-Physics Talon 532-20 laser was installed in Yaakov Weiss’ lab in Jerusalem.

Radiogenic isotopes are an important tool in dating geologic events and as tracers of melt and fluid sources. Their use as tracers is based on the preservation of isotopic ratios of heavy elements during melting, fractionation, and separation of immiscible liquids. Akagi and Masuda (1988) were the first to determine the Sr isotopic composition of HDFs after burning a diamond in a platinum crucible under flowing oxygen. They dissolved the residue in HCl and used half of the material for the analysis of Sr isotopes, using standard wet chemical methods. The other half was used for analyzing the trace element concentrations. Klein-BenDavid et al. (2010, 2014) replaced the burning of the diamond by laser ablation and the quartz tubes by Teflon beakers fitted with an optical window. The details of the ablation and dissolution procedures were described above, in the off-line laser ablation of fibrous diamonds (‘Trace element analysis’ section). Separation chemistry and isotopic analysis of the derived solution were performed using standard techniques. To date, only 42 fibrous diamonds from Canada, Botswana, DRC, Guinea, and Siberia were analyzed for their Sr isotopic compositions and a handful of those for their Nd and Pb isotopic compositions (Klein-BenDavid et al. 2010, 2014; Smith et al. 2012; Weiss et al. 2015).

Fibrous diamonds were first analyzed for their Ar content and isotopic composition using noble gas mass spectrometry by Zashu et al. (1986), without noticing their unique character and the presence of microinclusions. They determined the K, ³⁶Ar, and ⁴⁰Ar content of a group of diamonds from DRC and obtained linear correlations between K and ⁴⁰Ar concentrations and between ⁴⁰K/³⁶Ar and ⁴⁰Ar/³⁶Ar ratios. Interpreting the slope of the correlation between the two ratios as age (an isochron diagram) they obtained 6 ± 0.3 Ga, older than the age of the Earth. It took three more studies to reach the correct explanation - the correlations reflect the presence of microinclusions full of HDFs that contain not only K₂O and other oxides, but also excess argon that came with the fluid and is not the result of potassium decay in the diamond. The very old age reflects the stronger enrichment of the original trapped fluid in Ar relative to K. Measurement of Cl, after its activation into ³⁸Ar, revealed an even tighter ⁴⁰Ar/³⁶Ar–Cl/³⁶Ar correlation, demonstrating that the Ar–K correlation has nothing to do with the diamond age (Turner et al. 1990).

Following studies yielded data on the concentrations and the isotopic composition of He, Ne, Ar, Kr and Xe in fibrous diamonds from DRC, South Africa, Botswana, Guinea, Yakutia and Canada (Ozima and Zashu 1991; Burgess et al. 1998, 2009; Wada and Matsuda 1998; Weiss et al. 2017, 2021,Broadley et al. 2018; Timmerman et al. 2018b, 2019; Johnson et al. 2000). Noble gas analyses of fibrous diamonds were performed by diamond graphitization and gas released at T > 1800 °C (commonly at ~2000 °C) using a low blank, high-temperature resistance furnace (e.g., Wada and Matsuda 1998), a laser heating system (Broadley et al. 2018), or by crushing the diamond in a high-vacuum crushing vessel (e.g., Burgess et al. 1998). While the heating techniques extract all the gas from the diamond, including microinclusions and diamond lattice, the crushing technique preferentially extracts the gases trapped in the microinclusions (Burgess et al. 1998; Sumino et al. 2011). This issue is important because nuclear production of noble gases can add ³He, ⁴He and ²¹Ne in diamonds (Basu et al. 2013 and references therein), almost entirely to the diamond lattice and not in the microinclusions (which constitute only ~0.1% of the diamond by mass, Navon et al. 1988). Crushing the diamonds rather than burning them, reveals the signature of the microinclusions, and allows the distinction, for example, between the He compositions of the trapped HDFs (mantle source) and He confined to the diamond lattice (from secondary sources; Burgess et al. 1998).

Burgess et al. (1998), Weiss et al. (2017, 2021), Timmerman et al. (2018b, 2019) and Broadley et al. (2018) reported results on He, Ne, Ar, Kr, and Xe concentrations and isotopic composition obtained by crushing fibrous diamonds, rather than graphitizing them. If the extraction efficiency by crushing is not high enough, only part of the gas trapped in the microinclusions is released. This may affect the absolute gas concentrations but should not influence their isotopic ratios. Timmerman et al. (2019) graphitized the crushed diamond material and estimated that only 4–20% of the gas in the microinclusions is released by crushing. However, these measurements yielded a significant variation in ³He/⁴He ratios, of up to 10 Ra values, between crushing and heating techniques (where Ra is the atmospheric ratio of 1.39 × 10⁻⁶). In contrast, Broadley et al. (2018) reported ⁴He concentrations during crushing that are higher by 2–3 orders of magnitude compared with final heating results, and similar ³He/⁴He ratios for the two modes of extraction. Thus, gas extraction efficiency by crushing may be much higher (as a result of different crushing methods and initial sample size), and the measured concentrations and isotopic ratios represent the trapped HDF compositions.

Following the activation of diamonds by neutrons, they provide information on additional elements, whose isotopes react to form synthetic noble gas isotopes that can be measured as well. The best-known example is the ‘Ar–Ar dating’ method where ³⁹K reacts to ³⁹Ar. Turner et al. (1990), Wada and Masuda (1998) and Burgess and co-workers (Burgess et al. 1998, 2002, 2004, 2009; Johnson et al. 2000) have used neutron activation to determine K (via ³⁹Ar), Cl (³⁸Ar), Ca (³⁷Ar), Br (⁸⁰Kr), I (¹²⁸Xe) and U (¹³⁴Xe) concentrations and study noble gases and halogens in fibrous diamonds from Africa, Siberia and Canada.

More than 20 years of research lead to the discovery of the four types of HDFs. Navon et al. (1988) first published compositions that span a range between carbonatitic and silicic HDFs. Thirteen years later Izraeli et al. (2001) described a new saline HDF. Last, Klein-BenDavid et al. (2009) distinguished between low-Mg carbonatitic endmember that forms a continuous array towards the silicic one, and high-Mg carbonatitic HDFs that were defined as a new, fourth type. Many other studies confirmed the global distribution of all four compositional types (Figs. 7 and 8, Tables 1–4), but no new major element composition has been discovered since. Trace element analyses revealed two repeating patterns of enrichment in incompatible elements (Weiss et al. 2008b; Rege et al. 2010): highly fractionated patterns with negative anomalies of alkalis, HFSE and in most cases Sr, and less fractionated, smoother patterns. These two types of patterns have been identified in all four types of HDFs (Fig. 9). Radiogenic isotope ratios correlate with the trace elements indicating a relationship between metasomatic enrichment of the source rocks and HDFs signatures (Klein-BenDavid et al. 2014).

Many fibrous diamonds exhibit radial variation in the number density of microinclusions (Fig. 1). However, out of ~300 fibrous diamonds analyzed for the major element composition of their microinclusion, only a handful are chemically zoned and show significant radial (core-to-rim) variation of HDF composition. Even in these diamonds, there is no consistent chemical evolution of fluid composition (e.g., from silicic to carbonatitic) or systematic isotopic trend in the diamond carbon (Klein-BenDavid et al. 2004; Shiryaev et al. 2005; Weiss et al. 2009; Zedgenizov et al. 2011; see arrows in the inset in Fig. 3c). Thus > 95% of known HDF-bearing fibrous diamonds recorded growth from a homogeneous medium. Oxide concentration in individual microinclusions varied randomly with no clear core-to-rim evolution and by less than 20% relative. Hence, a single, major element composition achieved by averaging a few tens of HDF inclusions is representative of the diamond to better than ±15% (Weiss et al. 2008a), reflecting analytical errors and a limited real variation between the individual microinclusions.

LA-ICP-MS analyses determine the average trace element composition of thousands of microinclusions in the ablated volume of the diamond. Multiple ablation analyses of different parts of a single diamond agree to better than 15% (Tomlinson et al. 2009; Weiss et al. 2009, 2013b, 2018; Rege et al. 2010; Klein-BenDavid et al. 2014), thus showing that HDFs in an individual diamond are also homogenous for trace elements.

We note that the available isotopic data on HDFs in fibrous diamonds provide contrasting evidence. In some cases, repeated analyses of the same diamond reveal homogeneity (Weiss unpublished data), in other cases significant variation is documented (Klein-BenDavid et al. 2010). To clarify this issue of uniform vs. variable HDF isotopic compositions in a single diamond, additional data are necessary.

Most zoned diamonds reflect two or more distinct growth events, for example, two Kankan (Guinea) fibrous diamonds exhibit inner zones with silicic HDFs and outer ones with low-Mg carbonatitic compositions (Weiss et al. 2009). Similar variations from silicic to carbonatite HDFs were reported by Shiryaev et al. (2005) and the reversed direction by Zedgenizov et al. (2011). Weiss et al. (2014) reported two distinct microinclusion-rich layers in a clear octahedral NFMC diamond from Finsch (South Africa); the inner one carries a low-Mg carbonatitic HDF and the outer one a saline composition. The contrasting radial change of the composition and the sharp boundaries that are also reflected by CL imaging suggest that zoning is the result of growth from different HDFs, rather than in situ evolution of a single composition.

Such zoning allows constraining whether the HDFs carry the carbon from which their host diamond grows. Klein-BenDavid et al. (2004) reported saline and high-Mg carbonatitic HDFs in the inner and outer zones, respectively, of a diamond from Diavik (Canada). The carbon isotope compositions of the two zones show a clear difference: −8.1 to −8.9‰ in the inner vs. −11.4 to −12.8‰ in the outer. Shiryaev et al. (2005) reported similar relationships in a cuboid fibrous diamonds from Brazil. The inner part was silicic, while the outer part carried low-Mg carbonatitic HDFs and the carbon isotopic composition changed sharply at the boundary. The simultaneous changes of CL, major element compositions of the HDF and the carbon isotope compositions of the host diamonds thus indicate that the carbon and nitrogen for diamond growth are carried and supplied by the HDFs.

The compositional space spanned by HDFs in fibrous diamonds is huge, from highly silicic melts to carbonatitic and saline compositions. The high water and carbonate content distinguishes them from any of the melts we know at the surface such as basalts or granites and even from erupted kimberlites, lamproites, or carbonatites. Yet, fibrous diamonds from mines all over the world carry the same four types of HDFs (see ‘Geographical distribution of HDFs’ below). The major element composition of all diamonds is presented in Figs. 7 and 8, and in the Supplementary compilation of HDF data Excel file that accompanies this paper (Weiss et al. 2022). Tables 1–4 present the composition of the four HDF types. These compositions (shown in Figs. 7 and 8) were compiled by averaging 11–37 diamonds for the various HDF types, ignoring isolated extreme analyses.

The tables present the compositions in four ways. Table 1 provides the major elements that were measured by EPMA normalized to 100 wt% and allowing for excess oxygen due to chlorine (Cl*15.999/2/35.45). This is the standard way to present the composition, avoiding the uncertainty associated with the IR determination of water and CO₂. The silicic HDFs are highly siliceous (SiO₂ = 57 wt%, with some HDFs reaching up to 65 wt%), and rich in K₂O (12.4 wt%, with more scatter and values up to 22 wt%) and Al₂O₃ (10 wt%). Many diamonds form a continuous array between this component and low-Mg carbonatitic HDFs that are richer in CaO, FeO and MgO (up to 15 wt% MgO). High-Mg carbonatitic HDFs form a distinct group, especially through their very high MgO (>15 wt%). CaO (~20 wt%) and FeO (~14 wt%) are also high, while SiO₂ and Al₂O₃ are low (9 and 3 wt% respectively). The K₂O content varies between localities (see ‘Geographical distribution of HDFs’ below; Fig. 8d) with some carrying more than 20 wt%, while many others only contain 5–12 wt% K₂O. The saline HDFs are mostly (K,Na)Cl with a few percent of Fe-, Ca-, Ba- and Mg-carbonates and little SiO₂, Al₂O_3, TiO₂ and P₂O₅. The BaO concentration in Table 1 averages two distinct groups, saline HDFs with high BaO (~13 wt%) in Canadian and Russian diamonds and those with low BaO content (~1 wt%) in South-African diamonds.

In Table 2 we added the calculated CO₂ (based on the EPMA analysis, see section ‘H₂O and CO₂ in the HDFs: Connecting EPMA and FTIR’ above) and the H₂O components (based on the IR data) and renormalized to 100 wt% (again accounting for excess oxygen due to chlorine). These two values are less accurate but are useful for understanding the nature of the HDFs. CO₂ makes up 25 and 32 wt% of the low- and high-MgO carbonatitic HDFs, respectively, but only 11 wt% of the saline HDFs and <5 wt% of the silicic ones. The water content of the HDFs varies between 8 wt% in the high-Mg carbonatitic HDFs to 21 wt% in the saline ones. We note that the water content we calculate depends on the choice of the absorption coefficient, which is not well constrained. The results presented here were calculated using the high value of ε_water = 109 (AU/cm)/(mole/liter). If the lower value is used the water content increases by a few percent, and up to ~26 wt% in the saline endmember.

Table 3 presents the data in atomic proportions. The water content was normalized to 55.5 moles or 1 kg of water, so all elements are in molal concentration (moles/1 kg water). With 1.6 to ~4.6 ions per water molecule, it is clear that the HDFs are not hydrous solutions, but rather melts or supercritical fluids with a high content of carbonate and water. To get a feeling for the nature of these melts, Table 4 groups the various components by their anions: silicates, carbonates (including both metal and carbonate ions), phosphates, halides and water. Together, the HDFs show a wide range of miscibility among the various components. Carbonate and water are always present. Phosphate is always a minor constituent. Silicates and halides are variable but also appear in all HDFs. During analysis, the best way to distinguish an HDF from a micromineral inclusion is that the HDF analysis carries all the major elements.

The sole indication for the presence of sulfide HDFs in diamonds is provided by three fibrous diamonds from Yubileinaya, Siberia (Klein-BenDavid et al. 2003). The diamonds contain carbonatitic HDF with CO₂/(CO₂ + H₂O) ratio between 0.26–0.65, but also carry other microinclusions rich in S, Ni and Fe with Ni / (Ni + Fe) that ranges between 0.3–0.7. The S / (S + Ni + Fe) ratio of individual inclusions varies between 0.35–0.64, suggesting that this is a melt rather than a mineral composition. The presence of small amounts of alkali elements and Mg are indicative of a melt as well. The sulfide melts coexist with carbonatitic HDFs, but the data is insufficient for defining the relation between the two fluids. TEM studies of microinclusions in other diamonds (e.g., Logvinova et al. 2019a) showed that sulfides, if present, make up only a trivial fraction of most inclusions.

Trace element abundance patterns of HDFs (Fig. 9) reveal very high concentrations of incompatible elements relative to primitive mantle (PM) values. The REE are fractionated and variable, mostly negative anomalies exist for the alkali elements (K, Rb, and Cs), high field strength elements (HFSE: Ti, Zr, Nb, Hf, Ta), Sr and Y relative to elements of similar compatibility (e.g., Schrauder et al. 1996; Zedgenizov et al. 2007, 2009, 2011; Tomlinson et al. 2009; Weiss et al. 2009, 2013b; Skuzovatov et al. 2012; Smith et al. 2012; Klein-BenDavid et al. 2014; Gubanov et al. 2019; Timmerman et al. 2019). A closer look reveals two main patterns (see below), which are present in all the four HDF types.

Correlations between the trace element composition and HDF types were only identified within a few studies, and mainly within diamonds of the same locality. Namely, the concentrations of Rb, Cs, Ba, Th, U, LREE and Sr decrease linearly from silicic to low-Mg carbonatitic HDFs in diamonds from Jwaneng, Botswana (Schrauder et al. 1996). Similarly, La/Dy ratios decreasing from silicic to low-Mg carbonatitic compositions were observed in HDFs in diamonds from Yakutia, Siberia, (Zedgenizov et al. 2007, 2009, 2011). These observations suggest that carbonate-rich HDFs are more enriched in incompatible elements than silicic ones. Conversely, Ba, U and Th concentrations and La/Dy ratios all increase from silicic to low-Mg carbonatitic HDFs in diamonds from Kankan, Guinea (Weiss et al. 2009). Compared with the silicic to low-Mg carbonatitic array, the high-Mg carbonatitic compositions exhibit the highest (Nb,Ta,Ba,Th,U)/alkalis, LREE/MREE and LREE/HREE ratios, both in Yakutia and in Kankan (e.g., Zedgenizov et al. 2007; Weiss et al. 2009, 2011; Skuzovatov et al. 2012; Gubanov et al. 2019). In general, high-Mg carbonatitic HDFs in Siberian diamonds are characterized by high levels of Nb and Ta compared to Ba, Th, U, alkalis and LREE contents (i.e., PM normalized (La,Th,Ba,U)/Nb_PM ratios <1; Figs. 10 and 11). But lower Nb and Ta levels were recorded in some other Siberian diamonds (Zedgenizov et al. 2009, 2011), in diamonds from Kankan and Diavik (Weiss et al. 2011, 2015), and are even lower in high-Mg carbonatitic HDFs from Finsch (i.e., (La,Th,Ba,U)/Nb_PM ≫ 1; Fig. 9; Weiss and Goldstein 2018).

The saline HDFs are distinct from the other types. This is seen in their elevated La/Pr and La/Dy ratios compared to all other HDFs (Fig. 11); Th/U and Ba/U ratios are also higher in the saline HDFs. Moreover, a few studies pointed out that the Sr* (Sr / √(Pr × Nd))_PM anomaly is positive for most saline HDFs and negative for all other types (Fig. 11d; Klein-BenDavid et al. 2014; Weiss et al. 2015). Similarly, the Eu* anomaly (expressed here as Eu/Sm_PM, because of the lack of sufficient Gd data) in the saline HDFs is > 1, compared to the majority of all other HDFs, which have either negative anomalies (Eu/Sm_PM < 1) or no anomaly (Fig. 11c; Smith et al. 2012; Weiss et al. 2013b, 2015; Klein-BenDavid et al. 2014).

When all the available HDF trace element data are examined (Supplementary compilation of HDF data, Weiss et al. 2022), the observed relationship between Ba, Th, U and LREE enrichment and Nb, Ta and alkalis depletion is the most persistent. This is expressed as positive correlations between ratios such as La/Nb, Th/Nb, Ba/Rb and U/K, regardless of the HDF major-element compositions or their host diamond provenance (Fig. 10a,b).

Based on these characteristics, Weiss at al. (2008b, 2013b) identified two main trace element patterns in all four HDF types (Fig. 9). One with large ion lithophile element (LILE) and LREE enrichment, and HFSE and alkali depletion, similar to calcalkaline magmas and continental crustal rocks. The other has lower LILE and LREE abundances and ‘smoother’ overall trace-element patterns similar to oceanic basalts. These patterns were designated first as ‘Tables‘ and ‘Benches‘ (Weiss et al. 2008b), and later as ‘fibrous-low’ and ‘fibrous-high‘ (Rege et al. 2010) or ‘Ribbed’ and ‘Planed’ patterns (Weiss et al. 2013b), respectively. The two are best distinguished using co-variation diagrams of (Th,U)/(Nb,Rb) vs. (La,Ce)/(Nb,Rb) (Fig. 10), where (La,Th)/Nb_PM ≈ 100, La/Rb_PM ≈ 25 and Th/Rb_PM ≈ 50 for the fractionated ‘Ribbed’ pattern and (La,Th)/Nb_PM ≈ 0.5, La/Rb_PM ≈ 0.2 and Th/Rb_PM ≈ 0.5 for the unfractionated ‘Planed’ pattern (excluding a few extreme compositions). The two trace element patterns are observed in all four major element HDF types (Fig. 9). Intermediate patterns that vary through the complete range between the two also exist. The consistency of the two patterns in HDFs of differing compositions, from different mantle localities, well separated in space and time, overcomes the small trace element variations between HDF types observed in single localities. This emphasizes the main feature of major-trace element systematics—the trace elements are decoupled from the major element composition of the HDF (Weiss et al. 2013b).

The pioneering work of Akagi and Masuda (1988) provided the first Sr isotope data of HDFs in five diamonds from DRC. The ⁸⁷Sr/⁸⁶Sr ratios ranged between 0.7038 to 0.7052 and showed no correlation with ⁸⁷Rb/⁸⁶Sr values indicative of an isochronous relationship. Similarly, all later studies of HDFs from Botswana, Canada, Siberia, DRC, and Guinea (42 diamonds in total), yielded no Rb–Sr isochron systematics (Klein-BenDavid et al. 2010, 2014; Smith et al. 2012; Weiss et al. 2015).

Saline HDFs in diamonds from the ~2.7 Ga Wawa deposit (Smith et al. 2012) show very low initial ⁸⁷Sr/⁸⁶Sr ratios (⁸⁷Sr/⁸⁶Sr_i = 0.699–0.703, Fig. 10c). In contrast, all four types of HDFs from Phanerozoic kimberlites span a wide range of ⁸⁷Sr/⁸⁶Sr_i ratios, from MORB-like depleted values (i.e., 0.703–0.704) to more radiogenic compositions. Highly radiogenic compositions of ⁸⁷Sr/⁸⁶Sr_i > 0.710 and up to 0.723 are observed solely in silicic to low-Mg carbonatitic HDFs (Fig. 10c,d; Klein-BenDavid et al. 2010, 2014). These HDFs, mainly in DRC and Botswanan diamonds, also exhibit positive correlations between Sr isotopic composition and trace element concentrations. Klein-BenDavid et al. (2010, 2014) suggested that the correlation indicates a relationship between metasomatic enrichment of the source rocks and the elevated ⁸⁷Sr/⁸⁶Sr signature of HDFs derived from these sources. For silicic to low-Mg carbonatitic HDFs, a general and robust relationship is apparent between higher ⁸⁷Sr/⁸⁶Sr isotopic values and elevated La/Nb, Th/Nb and Ba/Rb ratios (Fig. 10c,d). Similar systematics exist for high-Mg carbonatitic compositions. These relationships suggest that a radiogenic Sr isotopic signature is related to a more strongly fractionated (‘Ribbed’) trace element pattern within HDFs of the same type. Saline HDFs from the Northwest Territories (Ekati and Diavik) do not show these relationships; rather they exhibit positive correlations between ⁸⁷Sr/⁸⁶Sr_i and Th/U, La/Pr, and La/Sm, as well as Sr* and Eu* anomalies (see Fig. 3 of Weiss et al. 2015).

Radiogenic isotopic data for Nd and Pb from HDFs are very scarce, only five diamonds (all silicic to low-Mg carbonatitic) were analyzed for their Nd and only three for their Pb isotopic composition (Klein-BenDavid et al. 2010, 2014). The ¹⁴³Nd/¹⁴⁴Nd ratios range between 0.5118 to 0.5105 and correlate negatively with the Sr isotopic composition. The few diamonds analyzed for Pb isotopic compositions display high ²⁰⁷Pb/²⁰⁴Pb ratios (15.62–15.67) compared to depleted MORB mantle (DMM) or kimberlites and modest ²⁰⁶Pb/²⁰⁴Pb ratios (17.87–18.12), placing them at a slightly more radiogenic position than calculated primitive mantle sources (Kamber and Collerson 1999; Murphy et al. 2003; Klein-BenDavid et al. 2010, 2014).

Fibrous diamonds are highly enriched in noble gases. Crushing yields ⁴He concentrations of 10⁻⁶–10⁻⁵ cm³ STP/g and ⁴⁰Ar between 10⁻⁷–10⁻⁶ cm³ STP/g (Burgess et al. 1998; Broadley et al. 2018; Timmerman et al. 2018b, 2019; Weiss et al. 2021). In most cases, similar concentrations were measured by fibrous diamond graphitization (Ozima and Zashu 1991; Wada and Matsuda 1998; Burgess et al. 2009), which suggests that most of the gas is trapped in the microinclusions and released by crushing alone. Equivalent high He concentrations were only found in “popping rock” mid-ocean ridge basalts (Moreira et al. 1998). Burgess et al. (1998) combined these data with ⁴⁰Ar/K and ⁴He/⁴⁰Ar* (*corrected for atmospheric ⁴⁰Ar) and determined that He and Ar are present in the trapped HDFs (rather than HDF+diamond) at concentrations of 10⁻³ cm³ STP/g, corresponding to enrichment by 3 orders of magnitude over present-day upper mantle values.

The ³He/⁴He isotopic ratios of HDFs in most analyzed diamonds from Africa, Siberia, and Canada vary between 2–8 Ra (the atmospheric ratio of 1.39 × 10⁻⁶; Burgess et al. 1998, 2009; Weiss et al. 2017, 2021; Timmerman et al. 2018b), similar to the range of MORB (³He/⁴He = 8 ± 1 Ra; Graham 2002), recycled subducted surface material (Kurz et al. 1982) and subcontinental lithospheric mantle (SCLM, 6.1 ± 2.1 Ra; Day et al. 2015). Only a few HDFs have higher ratios (10–11.2 Ra) or lower, radiogenic values of <1 Ra (Burgess et al. 1998; Broadley et al. 2018; Timmerman et al. 2019; Weiss et al. 2021). Broadley et al. (2018) suggested that the variation between SCLM and the upper range of ³He/⁴He ratios of the HDFs reflects a mixture between lithospheric and deep volatile noble gas components originating from a mantle plume, whereas ³He/⁴He<1 was postulated to result from the accumulation of radiogenic ⁴He atoms produced by α-decay of U and Th (Burgess et al. 1998). Weiss et al. (2017) showed that ³He/⁴He ratios of HDFs correlate negatively with the amount of K₂O and positively with the carbonate component in fibrous diamonds from Africa, which reflects increasing ³He/⁴He from saline to low-Mg carbonatitic to high-Mg carbonatitic HDFs. Exceptional are the South African silicic and carbonatitic HDFs which have much lower ³He/⁴He compared to silicic and carbonatitic compositions from Botswana and Guinea, to the degree that their ³He/⁴He ratios are lower (more radiogenic) compared to diamonds with saline HDFs from the same locations. This could suggest a complex evolution of the source of the fluids or represent different metasomatic events at different times over the history of the local continental mantle. The latter alternative is consistent with the observed negative correlation between ⁴He contents and Ra values and positive correlation with increasing U–Th concentrations in HDFs from South Africa and DRC (Timmerman et al. 2018b, 2019).

It was suggested that by combining He isotope analyses with U–Th-He abundance measurements, the geochronology of HDF-bearing diamonds and C–O–H metasomatism can be constrained (Weiss et al. 2017; Timmerman et al. 2019). Indeed, the slope of ⁴He vs. U–Th variation of HDFs in the few DRC diamonds deviates from the ~70 Ma eruption age of their possibly related kimberlite, indicating that these diamonds formed 10 to several hundred Myr before their host kimberlites (Timmerman et al. 2019). However, that study did not take into account the possibility of diffusive loss of He, which is critical for a meaningful interpretation of (U–Th)/He data and for constraining ages (Farley 2002). The issue was recently addressed by Weiss et al. (2021), who showed that the budget of He in a diamond is affected by diffusion over geological time scales. Taking into account the impact of possible He diffusion on the ³He/⁴He ratios of South African diamonds and considering the thermal and tectonic history of the Kaapvaal craton, they suggested an upper limit of ~1.8 ± 0.2 × 10⁻¹⁹ cm²/s for the diffusivity of He in fibrous diamonds. This diffusion limit provides a means to directly place an upper and lower limit on the timing of diamond crystallization and C–O–H metasomatism of the SCLM. For the analyzed diamonds from South Africa, it constrained the timing of diamond formation during three metasomatic events in the late-Mesozoic, Paleozoic and Proterozoic, each by a different HDF type (saline, silicic and carbonatitic, respectively; Weiss et al. 2021).

The ⁴⁰Ar/³⁶Ar values of fibrous diamonds are high, typically >3000, and reach values above 40,000 (Turner et al. 1990). Wada and Matsuda (1998) showed that in fibrous diamonds from DRC, such isotopic variation form a trend with ³⁶Ar concentrations, between air-like Ar (³⁶Ar > 10⁻⁹ cm³ STP/g and ⁴⁰Ar/³⁶Ar ≈ 3 00; Lee et al. 2006) and a mantle component similar to MORB ‘popping rock’ source (³⁶Ar < 10⁻¹⁰ and ⁴⁰Ar/³⁶Ar >30,000; Moreira et al. 1998). However not all fibrous diamonds fall on such a trend, as evident by those from Canada which exhibit high ³⁶Ar and high ⁴⁰Ar/³⁶Ar (Johnson et al. 2000; Burgess et al. 2009). These isotopic values indicate that the high content of ³⁶Ar in the HDF microinclusions is derived from their mantle source. ²⁰Ne/²²Ne and ²¹Ne/²²Ne ratios reach values up to ~11 and ~0.06, respectively (Wada and Matsuda 1998; Broadley et al. 2018). Within the ²⁰Ne/²²Ne–²¹Ne/²²Ne space, the HDFs Ne isotopic compositions are distinct from those of air and fall between the air–MORB mixing line and the air–solar mixing line. For Xe, the ¹²⁹Xe/¹³⁰Xe vs. ¹³⁶Xe/¹³⁰Xe isotopic ratios of a few fibrous diamonds from DRC correlate and lie on a mixing line defined by air and MORB (Wada and Matsuda 1998). Timmerman et al. (2018b, 2019) are the only available studies thus far that provide information about noble gases heavier than He for fibrous diamonds with known HDF compositions. Using the crushing technique, these authors measured ⁴⁰Ar/³⁶Ar ratios between 480–3600 for fibrous diamonds containing silicic to low-Mg carbonatitic HDFs, ⁴⁰Ar/³⁶Ar ratios between 390–1940 for high-Mg carbonatitic HDFs, and values in the range of 380–30,000 for saline compositions. The ⁴⁰Ar/³⁶Ar ratios obtained by step-heating diamond graphitization of a few of the silicic to low-Mg carbonatitic HDFs yielded values that varied between 670 and 15,400 (Timmerman et al. 2019). ¹²⁹Xe/¹³⁰Xe vs. ¹³⁶Xe/¹³⁰Xe isotopic compositions of all HDF types in 7 diamonds that they analyzed lay on mixing lines between air and MORB or air and crust (Staudacher and Allègre 1982). Timmerman et al. (2018b) noted that the ³⁶Ar/¹³⁰Xe and ⁸⁴Kr/¹³⁰Xe ratios for the analyzed saline, silicic to low-Mg carbonatitic, and high-Mg carbonatitic HDFs all fall in the compositional field for sea-water, sediments, and crustal endmembers and are distinct from atmospheric Ar/Xe and Kr/Xe compositions.

Unfortunately, only four diamonds were measured for both their major element and halogen compositions (Jwaneng diamonds analyzed by Schrauder and Navon 1994 and Johnson et al. 2000). However, K is commonly analyzed together with the halogens and its concentration in the HDFs is overall uniform, allowing for the use of K/Cl to identify the type of the HDF and normalize the results. The average K/Cl of diamonds from the Ekati mine, Canada (Johnson et al. 2000; Burgess et al. 2009) is close to one, which is suggestive of saline HDFs, similar to most major element analyses of diamonds from that mine (Fig. 12).

Cl concentrations in the Ekati HDFs vary between 20–35 wt%, the Br content of most samples is 1000–2000 ppm, with Br/Cl ratios of 0.001–0.003. Some diamonds reach Br levels of up to 3 wt% and extreme molar Br/Cl ratios (up to 0.063, Johnson et al. 2000). Iodine is also enriched: 10–30 ppm and I/Cl of 1–3 × 10⁻⁵ in most samples, and up to 1200 ppm and I/Cl = 1.7 × 10⁻³ in the most fractionated sample. The ratios of most samples fall close to the MORB field and extend towards values typical of altered oceanic crust. However, Br concentrations are highly enriched in comparison to PM (~10⁴ × PM) and even more in comparison to the depleted MORB reservoir. Iodine is enriched to levels of 1500–2500 × PM. Separation of a saline fluid from a silicate melt was proposed as a possible reason for the extreme values (Burgess et al. 2009). Fractionation of halides, which was observed experimentally (Förster et al. 2019) may also play a role in removing Cl, K and Na and enrich the residual fluid in Br and I (Johnson et al. 2000).

Diamonds from DRC, Jwaneng (Botswana) and Aikhal (Siberia) have K/Cl of 2–10 and are silicic to low-Mg carbonatitic. Assuming K₂O content of 13 wt%, (Table 2), we obtain Cl concentration of 1–3 wt%, similar to the EPMA measurements, 30–60 ppm Br, and 1–3 ppm I (with some diamonds reaching up to 17 ppm). Br/Cl ratios vary between 1–2 × 10⁻³, close to the seawater value, somewhat higher than MORB and similar to altered oceanic crust (Burgess et al. 2009; Broadley et al. 2018). I/Cl ratios of 0.5–3 × 10⁻⁵ are higher than seawater and span a broad area with MORB values at the center (see Fig. 5 of Burgess et al. 2009). These ratios represent enrichment of 300–800 over the PM value for Cl, 250–700 × PM for Br and 80–300 for I, similar to levels of other incompatible elements. Halogen concentrations are extremely high in saline HDFs, though they are highly enriched in the other types as well. Such enrichment in halogen concentrations and elevated Br/Cl and I/Cl ratios further strengthen the argument for the involvement of a subducted component in the generation of HDFs.

The formation mechanism for fibrous diamonds from the HDFs they trapped is broadly accepted, whereas the mode of formation of NFMC diamonds is less clear. The differences in texture, nitrogen aggregation and age, as well as the tight carbon isotopic composition of fibrous diamonds relative to the wide range of NFMC and polycrystalline diamonds, left the door open for other mechanisms of growth. Thus, it is interesting to examine whether HDFs, their fingerprints, or other fluids are found in non-fibrous diamonds.

Bulanova et al. (1988) and Novgorodov et al. (1990) reported 10–30 μm inclusions of potassium aluminosilicate glasses associated with inclusions of K-feldspar and other eclogitic minerals in colorless NFMC diamonds from Mir. Shatsky et al. (2019) reported two melt+clinopyroxene inclusions, with a SiO₂- and Al₂O₃-rich bulk composition, in two alluvial NFMC diamonds from Ebelyakh, Siberia. A single microinclusion composed of Fe, Si, P, Mg, K, Ca, Al and carbonate was identified by Jacob et al. (2011) in a FIB foil cut from a polycrystalline framesite from Orapa, but no additional information was provided. Using TEM on foils cut from Juina (Brazil) NFMC diamonds, Kaminsky et al. (2009, 2013) described a suite of HDF-related multiphase inclusions, mostly smaller than 1 μm in size. The suite included carbonates (magnesite, Na₂Mg(CO₃)₂ and Ba- and Sr-carbonates), phosphates, phlogopite, chlorides, fluorides, sulfides and pores (that probably included residual low-density fluids), and was designated: carbonate-halide association.

Weiss et al. (2014) were the first to identify HDFs, similar to those found in fibrous diamonds, within NFMC diamonds. They found and analyzed two microinclusion-rich layers near the edge of an NFMC Finsch diamond and used X-ray topography to show that the diamond is monocrystalline throughout. The inner layer carried carbonatitic HDFs, while the outer layer was saline. They also characterized six saline inclusions in the NFMC core of a coated diamond from Kankan. Recently, Kempe et al. (2021a) identified 3 silicic microinclusions in the NFMC core of a coated diamond from Voorspoed. A systematic search for inclusions in NFMC diamonds was conducted by Jablon and Navon (2016), who examined twinning planes of 30 twinned diamonds (macles) from Venetia and Voorspoed, as well as the diamond adjacent to the planes. Inclusions were absent in the majority of diamonds however, 32 shallow, subsurface microinclusions were identified in 8 of the macles. All were found where the twinning plane formed corners. Twelve inclusions in two diamonds carried submicrometer orthopyroxene and a K–Mg carbonate. The other 20 carried HDFs. Twelve inclusions in four Venetia diamonds carried high-Mg carbonatitic HDF and a silicic inclusion was found in another. In Voorspoed, three saline microinclusions were found in one diamond, and four silicic to carbonatitic ones in another. The level of nitrogen aggregation of the inclusion-bearing diamonds is the same as that of other 40 macles and 23 octahedral diamonds from the same batch. This is true for both localities (Jablon and Navon 2016).

Larger carbonate inclusions were found in a few Siberian diamonds. Sobolev et al. (1997) reported phlogopite associated with calcite (and dolomite, see Sobolev et al. 2009) in three NFMC diamonds from the Sputnik kimberlite. They did not provide the proportions of the two phases, but the combined composition of CaCO₃ and phlogopite brings the analysis close to that of silicic–carbonatitic HDFs. On a TiO₂ vs. Mg# diagram, the phlogopites fall in the eclogitic field that is associated with the silicic–carbonatitic HDFs (Sobolev et al. 2009). Raman spectroscopy of a particularly large (43 × 30 × 23 µm) carbonate inclusion in a NFMC diamond (Logvinova et al. 2019b) revealed that it consists of three distinct carbonates, which according to TEM-EDS analyses have a combined bulk composition of 0.1(Na_0.5K_1.5)CO₃·0.9(Ca_0.57Mg_0.43)CO₃. This is very close to the composition of high-Mg carbonatitic HDFs, but with slightly elevated CaO content and no measurable SiO₂, Al₂O₃, TiO₂, BaO or Cl, which are normally present at the few wt% level in such HDFs.

Aiming their micro-Raman at the interface between mineral inclusions and the host NFMC diamond, Nimis et al. (2016) characterized bands of Si₂O(OH)₆ and Si(OH)₄ at 650 and 800 cm⁻¹ as well as water bands at 1650 and > 3300 cm⁻¹. These bands are caused by a thin film of a hydrous residual fluid with dissolved silicates. The presence of the thin layer (<1.5 μm) of low-density material at the diamond-inclusion interface was confirmed by X-ray tomography. CO₂ could not be identified as its Raman bands are too close to the intense diamond line. Neither CH₄ nor N₂ were detected.

Another line of evidence for the presence of HDFs in NFMC diamonds derives from trace element analysis. Early efforts were reviewed in section ‘Trace elements analyses’ above. A major improvement came with the introduction of the off-line laser ablation method (McNeill et al. 2009; Klein BenDavid et al. 2010). McNeill et al. (2009) studied gem-quality NFMC diamonds from South Africa, Siberia and Venezuela, distinguished between a group enriched in Ba, U and La over Nb and another with smoother, less fractionated patterns, and concluded that the two groups bear resemblance to HDFs in fibrous diamonds (Weiss et al. 2008a,b). Melton et al. (2012) examined fragments of 10 clear NFMC diamonds that carried garnet inclusions. The trace element abundance pattern of the diamond richest in trace elements was similar to that of HDFs in fibrous diamonds. Krebs et al. (2019) conducted a more comprehensive study of gem and near-gem NFMC diamonds from Finsch and Newlands mines, South Africa. They reported that they “see the same spectrum of fluids in both high-purity gem and near-gem diamonds that was previously documented in fibrous diamonds. ‘Planed’ and ‘Ribbed’ trace element patterns characterize not only the HDF inclusions in fibrous diamonds but also in gem diamonds”. Another significant result of this study was the identification of similar patterns in silicate-bearing and sulfide-bearing NFMC diamonds from the Victor mine, Canada. Timmerman et al. (2019b) reached a similar conclusion for silicate- and sulfide-bearing NFMC diamonds from the Koffiefontein, Letlhakane and Orapa mines, South Africa; their trace element signatures are similar to those of fibrous diamonds and they record both ‘Planed’ (unfractionated) and ‘Ribbed’ (significant inter-element fractionation) patterns.

There is little evidence for other fluids trapped in NFMC diamonds. Chepurov et al. (1994), Tomilenko et al. (1997) and Smith et al. (2014, 2015) reported the presence of CO₂, N₂ and hydrocarbons in secondary fluid inclusions, larger than 1 µm, along healed cracks in diamonds from North-East Siberia and Africa. Nanoinclusions carrying solid molecular nitrogen were described in sublithospheric diamonds from Juina, the Urals and South Africa (Kagi et al. 2016, Rudloff-Grund et al. 2016, Navon et al. 2017, Sobolev et al. 2019). Navon et al. (2017) suggested that they formed by exsolution of nitrogen from the diamond matrix and not by direct trapping of nitrogen-rich fluid. Solid CO₂ was reported in diamonds from Colorado (Chinn et al. 1995a,b) and an unknown source (Schrauder and Navon 1993). Mixed-habit diamonds from Zimbabwe contain measurable CH₄ (Smit et al. 2016) and Smith et al. (2016) found thin films of CH₄+H₂ around metallic inclusions in clear sublithospheric diamonds.

In summary, the present dataset of direct measurements of fluids and trace elements in NFMC diamonds clearly shows the presence of HDFs similar to those found in fibrous diamonds. Evidence for other fluids are sporadic and their source is less clear.

A decade ago, the prevailing concept on the formation of fibrous diamonds and their HDFs asserted that they formed episodically in the mantle, close in time to the eruption of their host kimberlite over the last ~500 Ma. This assumption, that fibrous diamonds are young, was based on the short mantle residence time of fibrous diamonds at mantle temperatures, as evident by their low nitrogen aggregation (~100% A-centers, no B-centers and no platelets; Fig. 2), and the Phanerozoic emplacement ages of most kimberlites that carried these diamonds to the surface (e.g., Boyd et al. 1987; Navon 1999; Faure 2010; Gurney et al. 2010). However, in recent years, new data changed this idea: a) even diamonds with only A-centers may survive for many millions of years at 900–1000 °C (Fig. 2); b) HDFs were discovered in fibrous diamonds from Proterozoic and Archean deposits; c) fibrous diamonds with both A- and B-centers and platelets were identified at various mines d) in a few mines, nitrogen aggregation indicates several events of fibrous diamond formation, each with its own characteristic HDF type and e) U–Th/He dating of fibrous diamonds yielded ages older than 750 Myr.

The 2.7 Ga metaconglomerates near Wawa (Smith et al. 2012), as well as the 1.2 Ga Premier kimberlite (Weiss unpublished data), host fibrous diamonds that contain HDFs (Fig. 12). This is direct evidence for the antiquity of HDFs and their presence in the mantle at least since late Archean times. The fibrous diamonds, in both localities, have nitrogen in A-centers only, indicative of a short mantle residence time. The HDFs in both cases are solely saline, and more specifically, of the “high-Ba” group (Fig. 8f).

Reports of fibrous diamonds that carry nitrogen in both A- and B-centers came first from the Siberian craton. Nitrogen aggregation ratios varying between 10–70 %B (where %B = 100N_B/(N_A+N_B), and N is the nitrogen content in A- and B-centers) were found in fibrous diamonds from the Aikhal, Internationalnaya, Mir, Nyurbinskaya and Udachnaya kimberlites (Zedgenizov et al. 2006, 2009, 2020; Skuzovatov et al. 2011). These diamonds carried silicic to low-Mg carbonatitic and high-Mg carbonatitic HDFs. Later, fibrous diamonds with B-centers were documented at Finsch, De Beers Pool and Voorspoed on the Kaapvaal craton (Weiss et al. 2018; Weiss and Goldstein 2018; Kempe et al. 2021a). The aggregation ratios varied between 10–35 %B, and the diamonds carried all four HDF types (saline, silicic, low- and high-Mg carbonatitic compositions). In these localities, a few generations of fibrous diamonds were identified, each with its distinct HDF composition. Lastly, ratios of 10–40 %B were recently detected in fibrous diamonds from Pulandian, North China craton (a dozen of those carried high-Mg carbonatitic HDFs and one contained saline compositions; Mershon et al. 2021). HDFs were also found in a NFMC diamond from Finsch with 17 %B (low-Mg carbonatitic HDFs; Weiss et al. 2014), and in clear NFMC diamonds from Voorspoed and Venetia with 40–65 %B (saline, silicic to low-Mg carbonatitic and high-Mg carbonatitic HDFs; Jablon and Navon 2016).

Together, all these diamonds indicate much longer mantle residence times compared with the majority of fibrous diamonds that carry only A-centers at similar nitrogen concentrations. They show that fibrous diamond and the HDF they carry are not necessarily connected to kimberlite eruptions and formed episodically over extended geological periods. This is true in various lithospheric provinces and for all four HDF types. The above ideas are supported by the recent U–Th/He geochronology of HDF-bearing fibrous diamonds from the Congo and Kaapvaal cratons, indicating penetration of HDFs and formation and survival of fibrous diamonds in the SCLM over a time span of >1 billion years (Timmerman et al. 2019b; Weiss et al. 2021).

Fibrous diamonds are documented in many localities from all Archean cratons (Fig. 12). This indicates the global distribution, the long history, and the ubiquity of the four HDF types in the deep continental lithosphere. However, some differences exist when comparing the compositions of similar HDF types from various localities.

Silicic to low-Mg carbonatitic HDFs. Silicic to low-Mg carbonatitic compositions were reported from the Kaapvaal craton (Schrauder and Navon 1994; Weiss et al. 2018; Kempe et al. 2021a; Weiss unpublished data), the Congo craton (Navon et al. 1988; Kopylova et al. 2010; Kosman et al. 2016; Timmerman et al. 2019), the West African craton (Weiss et al. 2009), the Amazonian craton (Shiryaev et al. 2005), the Slave craton (Klein-BenDavid et al. 2007; Logvinova et al. 2008b; Weiss et al. 2015), the North Atlantic craton (Elazar and Weiss 2021), India (Schrauder et al. 1994), and the Siberian craton (Klein-BenDavid et al. 2009; Zedgenizov et al. 2009, 2011; Skuzovatov et al. 2016; Gubanov et al. 2019). The Siberian HDFs have the most extreme compositions on both ends of the silicic to low-Mg carbonatitic array, with SiO₂ both >65 and <15 wt% (on a carbonate- and water-free basis; Figs. 7a and 8a). As noted by Elazar et al. (2021) the HDFs from Siberia also form a distinct high-Al trend compared to similar HDF compositions from all other localities/cratons (Fig. 8c). CaO enrichment, while more restricted, is observed for HDFs in Ebelyakh fibrous diamonds (Fig. 8b) and elevated FeO values were recorded for Internationalnaya (not shown). These enrichments are compensated by lower K₂O contents (typically <10 wt%, Fig. 8d), as well as lower P₂O₅ (not shown) and Cl values (Fig. 7b). Another distinction of the Siberian silicic to low-Mg carbonatitic HDFs is their lower (Th,La,Ba)/Nb ratios (Figs. 10a and 11).

High-Mg carbonatitic HDFs. The high-Mg carbonatitic HDFs in Siberian fibrous diamonds (e.g., Klein-BenDavid et al. 2009; Zedgenizov et al. 2009, 2011; Skuzovatov et al. 2016; Gubanov et al. 2019) also show some unique compositional characteristics compared with those from the Slave craton (Klein-BenDavid et al. 2007; Logvinova et al. 2008b), the Kalahari craton (Timmerman et al. 2018a; Weiss and Goldstein 2018; Kempe et al. 2021a,b; Weiss unpublished data), the West African craton (Weiss et al. 2009) and the North China craton (Mershon et al. 2021). At Internationalnaya, Al₂O₃ and FeO contents are higher than at any other locality worldwide (Fig. 8c). All Siberian high-Mg carbonatitic HDFs have lower BaO contents resulting in lower Ba/Rb and Ba/K ratios compared to high-Mg carbonatitic HDFs from other localities/cratons (Figs. 10b and 11). In addition, the general enrichment of incompatible elements in such HDFs from Siberia is moderate, characterized by La/Sm_PM < 20, compared with values between 20–80 in other localities; (La,Th)/(Nb,Rb) and Sr/Rb show similar behavior, i.e., lower ratios in Siberia compared to other provinces.

Saline HDFs. Saline HDFs were identified in fibrous diamonds from the Kaapvaal craton (Izraeli et al. 2001; Timmerman et al. 2018a,b; Weiss and Goldstein 2018; Weiss et al. 2018; Kempe et al. 2021a; Weiss unpublished data), the West African craton (Weiss et al. 2009), the Slave craton (Klein-BenDavid et al. 2007; Tomlinson et al. 2006; Weiss et al. 2015), the Superior craton (Smith et al. 2012), the North China craton (Mershon et al. 2021), and the Siberian craton (Zedgenizov et al. 2018). No diamonds with saline HDFs were found in the Congo craton. Saline HDFs from Udachnaya have higher TiO₂ concentrations (~5 wt%), the ones from Koffiefontein are higher in FeO (10–20 wt%), and those from De Beers Pool have elevated Na₂O (mostly >15 wt%), compared to all other localities (Fig. 8e).

A striking dichotomy is observed for BaO (Fig. 8f), which is very high in saline HDFs from the Slave craton (10–18 wt%) and Udachnaya (13–15 wt%), but much lower in those from the Kaapvaal craton (BaO < 8 wt%). HDFs from the 1.2 Ga Premier kimberlite, on the Kaapvaal craton, have the highest BaO concentration (8 wt%, Weiss unpublished data) and those from the > 2.7 Ga Wawa deposit (14–17 wt%) are as enriched as other Canadian HDFs, indicating that high Ba levels in saline HDFs existed for billions of years. The Ba enrichment in the Slave craton HDFs is reflected in higher Ba/Rb and Ba/Nb ratios, which coincide with well-developed fractionated ‘Ribbed’ trace element patterns of the Slave saline HDFs (i.e., high (La,Th)/(Nb,Rb) and LREE/(MREE,HREE) ratios; Figs. 10a,b and 11).

The petrology of the source rocks of the HDFs, as well as the wall-rocks they interact with during migration, must be important in determining their major and trace element composition (e.g., Tomlinson et al. 2009; Zedgenizov et al. 2009; Weiss et al. 2011, 2013b; Gubanov et al. 2019). Figure 12 presents the abundances of the various HDF types globally and emphasizes that it varies between localities and cratons. For example, only silicic to low-Mg carbonatitic HDFs were reported in fibrous diamonds from Jwaneng, Botswana (Schrauder and Navon 1994), DRC (mostly from the Mbuji-Mayi area, Navon et al. 1988; Kopylova et al. 2010) and the CH-7 kimberlite at Chidliak (Baffin Island, Canada, Elazar and Weiss 2021). In Jwaneng and Chidliak, an eclogite-dominated source for diamonds is indicated by mineral inclusions in diamonds and kimberlite indicator minerals (Gurney et al. 1995; Pell et al. 2013; Gress et al. 2018; Xia 2018). Similarly, eclogitic minerals dominate the assemblage in Mbuji Mayi diamond inclusions (Mvuemba-Ntanda et al. 1982) and ~90% of the xenoliths are eclogitic (El-Fadili et al. 1999). The association between HDF type and the dominant diamond-host lithology in these three deposits agrees with the strong connection that has been established between silicic to low-Mg carbonatitic HDF compositions and hydrous/carbonated eclogites (see ‘Formation of silicic to low-Mg carbonatitic HDFs’ below for details). Similar relations can be identified at other localities, although the correlations are not as defined as in the examples above.

In Siberia, most of the 100 analyzed fibrous diamonds are from the Udachnaya and Internationalnaya kimberlites (25 and 39 diamonds, respectively). In both kimberlites, fibrous diamonds carry all the HDF types (Figs. 7 and 12), but the representation of the various types is not uniform. The proportions of {saline}:{silicic to low-Mg carbonatitic}:{high-Mg carbonatitic} HDFs is 25:36:39 in Udachnaya, and 3:62:35 in Internationalnaya. Peridotite is the predominant diamond host lithology in Yakutian kimberlites, as evident by diamond inclusions and xenoliths (Sobolev 1977; Kharkiv et al. 1997, as cited by Ionov et al. 2010; Sobolev et al. 2004). This explains the predominance of high-Mg carbonatitic HDFs in the two localities, as this HDF type has been linked to a carbonated peridotite source (see ‘Formation of high-Mg carbonatitic HDFs’ below for details). However, it cannot account for the significant abundance of silicic to low-Mg carbonatitic HDFs. We also note that saline HDFs are more abundant in fibrous diamonds from the Udachnaya kimberlite, where Cl was detected in melt inclusions in groundmass olivine (Kamenetsky et al. 2009). In comparison, at the Ebelyakh placers the majority of the 19 analyzed fibrous diamonds carry silicic to low-Mg carbonatitic HDFs (90%), with a minor representation of high-Mg carbonatitic (10%) and no saline compositions, quite distinct from the observed proportions in Udachnaya and Internationalnaya (Fig. 12). Within the framework of the apparent relationship between HDF type and diamond-host lithology, these differences follow the known distribution of mineral inclusions in diamonds from the Siberian placers, which are dominated by an eclogitic suite (>70%; Shatsky et al. 2015), compared to largely peridotitic inclusions in diamond at Udachnaya and Internationalnaya (Sobolev et al. 2004).

High-density fluid types in the southwestern Kaapvaal and the central Slave craton are dominated by saline compositions (>80%; Fig. 12). Many of the fibrous diamonds with saline HDFs carry micro-mineral inclusions or mixed mineral-HDF microinclusions and can therefore be assigned to either an eclogitic or a peridotitic paragenesis. Most of those samples are associated with peridotitic hosts (Izraeli et al. 2001; Tomlinson et al. 2006; Klein-BenDavid et al. 2007; Weiss et al. 2015, 2018). This is consistent with the predominance of peridotite in both lithosphere provinces based on mineral inclusions in diamonds and xenoliths abundances (e.g., Gurney et al. 1979; Lawless et al. 1979; Skinner 1989; Stachel et al. 2003; Menzies et al. 2004; Tappert et al. 2005). In the southwestern Kaapvaal, the minority (25%) of saline HDFs associated with micro-mineral inclusions are of eclogitic affinity (most of which are from Koffiefontein; Izraeli et al. 2001; Timmerman et al. 2018a,b; Weiss and Goldstein 2018; Weiss et al. 2018; Weiss unpublished data), while the remaining 75% are associated with peridotitic inclusions. In contrast, in the central Slave eclogitic minerals are only associated with silicic HDFs and only peridotitic minerals are found in diamonds with saline HDFs (Tomlinson et al. 2006; Weiss et al. 2015). High-Mg carbonatitic HDFs are solely linked to peridotitic micro-mineral inclusions in both cratons (including the Witwatersrand East Block of the Kaapvaal; Klein-BenDavid et al. 2004; Weiss et al. 2018; Kempe et al. 2021a; Weiss unpublished data).

In addition, a temporal gap was identified between saline HDFs and other compositional types in the Kaapvaal (Weiss and Goldstein 2018; Weiss et al. 2018, 2021; Kempe et al. 2021a). Such a gap, based on nitrogen aggregation of the host diamonds, was not observed in the central Slave (Tomlinson et al. 2006; Klein-BenDavid et al. 2007; Weiss et al. 2015). In both cases, saline HDFs invaded the lithosphere and interacted with different lithologies, as evident by major and trace elements and isotopic variations of the HDFs, and the HDF-associated micro-mineral inclusions (Izraeli et al. 2001, 2004; Tomlinson et al. 2009; Weiss et al. 2015, Weiss and Goldstein 2018). However, while in the Kaapvaal this fluid-rock interaction did not lead to melting of silicates and formation of non-saline HDFs (Weiss et al. 2018), in the Slave both silicic and carbonatitic HDFs formed by in situ melting as the saline HDFs traversed mixed peridotite-eclogite lithosphere (Weiss et al. 2015). This difference in the impact of saline HDF percolation through both lithospheres is attributed to the extremely refractory nature of diamond-bearing harzburgite/dunite in the Kaapvaal, compared to the significantly less depleted Slave lithosphere (as well as the Siberian one; Stachel et al. 2003; Stachel and Harris 2008).

The lithospheric origin of fibrous diamonds is manifested in a few independent lines of evidence:

• a)

  The majority of fibrous diamonds carry high nitrogen concentrations that reside only in A-centers, and most of those with both A- and B-centers have low B/(B+A) ratios (see ‘Temporal distribution of HDFs’ above). At the top of the asthenosphere, at ~1400 °C, B-centers would develop in a very short time, a few thousand years or less (Fig. 2, Taylor et al. 1996; Leahy and Taylor 1997), indicating that the fibrous diamonds grew in the lithosphere at cooler temperatures.

• b)

  Based on the shift of the IR peaks of secondary quartz in silicic HDF-bearing diamonds, Navon (1991) estimated that fibrous diamond originated at lithospheric pressures of 4–7 GPa.

• c)

  Geothermobarometry on micro-mineral inclusions that are found together with the HDF microinclusions yielded low temperatures of 1000–1185 ± 200 °C for Koffiefontein fibrous diamonds (Izraeli et al. 2001) and 1000 ± 100 °C for De Beers Pool ones (Weiss et al. 2018). Tomlinson et al. (2006) obtained a range of 930–1010 °C from larger mineral inclusions they found in fibrous diamonds from Panda. These are all lithospheric temperatures that correspond to pressures of 4–6 GPa along cratonic geotherms (e.g., Hasterok and Chapman 2011), and fall in the lower part of the temperature range obtained from mineral inclusions in NFMC lithospheric diamonds (e.g., Stachel and Harris 2008; Nimis et al. 2020; Nimis 2022).

The major and trace element chemistry of the HDFs indicates that they are low-volume melts, e.g., low-degree partial melts, percolating melts that interacted with a large volume of rock, or the very last residue following crystallization of a volumetrically more abundant parental melt. The high levels of enrichment in many incompatible elements can only be reached if the HDFs represent small fractions. Water comprises ~8–21 wt% of the HDFs (Table 2) compared with 100–500 ppm in the primitive mantle (McDonough and Sun 1995), corresponding to enrichment by a factor of 300–1500. Carbon, K and Cl yield similar factors of ~450, 500 and 1,000–10,000, respectively. Even if the source was enriched relative to primitive mantle, the melt/rock ratio must be small. Trace elements reveal similar enrichment in incompatible elements (Fig. 9) as do estimates based on the enrichment in He and Ar (~10³–10⁴, Turner et al. 1990, Burgess et al. 1998). Also, experimental studies of melting of peridotite or eclogite in the presence of CO₂ and/or H₂O have shown that melts of similar major element composition only form close to the solidus (see below).

Trace element abundance patterns of HDFs are best explained by equilibration with mantle peridotite without accessory phases for the ‘Planed’, and with the presence of mica and titanates for ‘Ribbed’ patterns (Weiss et al. 2013b). Such equilibration may happen during melting, but also during percolation or crystallization. The higher abundances of Ba, Th, U and LREE in the ‘Ribbed’ patterns, suggest an even lower fraction of partial melting for such HDFs. This allows for the survival of mica, titanates, and zircon in the source (or their growth during percolation) and leads to the formation of the typical negative anomalies in alkalies, Ti, Nb, Ta, Zr and Hf.

The positive correlation of Sr isotopic ratios with La/Nb ratios (low in ‘Planed’ and high in ‘Ribbed’ HDFs, Klein-BenDavid et al. 2014) suggests that the different patterns originate at the source or during percolation, as it is difficult to explain the isotope-trace-element correlation by fractionation. The origin of the ‘Planed’ patterns with low ⁸⁷Sr/⁸⁶Sr ratios can be linked to a source in the convective asthenosphere (similar to that of Group I kimberlites and ocean-island basalts, OIBs, e.g., Hofmann 1997; Paton et al. 2007), while the high ⁸⁷Sr/⁸⁶Sr ratios and the ‘Ribbed’ patterns reflect old enrichment of the source region that lies, most probably, in the lithosphere (e.g., Smith 1983; Giuliani et al. 2015).

Schrauder and Navon (1994) considered three processes for the formation of the silicic to low-Mg carbonatitic HDFs: melting, crystallization, and immiscibility. The immiscibility scenario gained popularity with the discovery of the saline HDFs (Izraeli et al. 2001; Klein-BenDavid et al. 2004, 2007). Klein-BenDavid et al. (2007) pointed out the absence of intermediate compositions between the silicic and the saline endmembers (Fig. 7) and suggested that crystallization of carbonates drives carbonatitic melts into an immiscibility gap where silicic and saline melts segregate. Safonov et al. (2007, 2009) conducted experiments in the CaMgSi₂O₆–(CaCO₃, Na₂CO₃)–KCl system and found such an immiscibility gap. However, in their water-free system, the evolution was from the immiscible saline and silicic melts at high temperatures towards one carbonatitic composition at lower temperatures. Contrasting fluid evolution recorded in zoned diamonds (Shiryaev et al. 2005; Weiss et al. 2009) and the variability of ⁸⁷Sr/⁸⁶Sr led to the rejection of the immiscibility model (Weiss et al. 2009).

Zedgenizov et al. (2007, 2009, 2011) reviewed a few possible models for the formation of high- and low-Mg carbonatitic HDFs. They recognized the similarity of the high-Mg compositions to kimberlites, but also pointed out differences between these HDFs and kimberlitic magmas. Following the distinction between low-Mg and high-Mg carbonatitic HDFs (Klein-BenDavid et al. 2009), Weiss et al. (2009, 2011) established a strong connection between high-Mg carbonatitic HDFs and a carbonated peridotite source, based on the results of near-solidus melting experiments (Dalton and Presnall 1998; Dasgupta and Hirschmann 2007; Brey et al. 2008). This is supported by the association of high-Mg carbonatitic HDFs solely with peridotitic micro-mineral inclusions (e.g., Klein-BenDavid et al. 2004; Weiss et al. 2018; Gubanov et al. 2019).

In view of the possible connection between high-Mg carbonatitic and saline HDFs (Klein-BenDavid et al. 2004), Weiss et al. (2009) examined potential options for the formation of such an array. Their preferred scenario was the penetration of saline HDF into a carbonated peridotite. The saline component provides a source for K, Cl and Ba, and may provide many of the trace elements of the carbonatitic HDF. Later, Weiss et al. (2011) distinguished between Udachnaya HDFs with high-K₂O and ‘Planed’ patterns and those from Kankan with low-K₂O and more ‘Ribbed’ pattern. They emphasized the close similarity of the trace element abundance patterns to those of Group I and Group II kimberlites and suggested that both high-Mg carbonatitic HDFs and kimberlites form by melting of carbonated peridotite, with or without phlogopite. The HDFs represent a lower degree of melting (~0.5% for Udachnaya and ~0.2% for Kankan) compared with Group I and II kimberlites (2% and 0.8%, respectively; see Fig. 6 and 7 of Weiss et al. 2011). Klein-BenDavid et al. (2014) reported the Sr isotope compositions of Udachnaya and Kankan HDFs. The values measured for Udachnaya (0.7043–0.7053) fall in the range of Group I kimberlites, and those measured in a Kankan diamond (0.7079) overlap with Group II kimberlites.

More evidence for the similarities between kimberlites and high-Mg carbonatitic HDFs comes from melting experiments of carbonated peridotite, where the first melt at the solidus is similar in composition to the HDFs (e.g., Dasgupta and Hirschmann 2007) and approaches that of kimberlite at higher melt fractions (e.g., Dalton and Presnall 1998; Brey et al. 2009). In summary, melting of a carbonated peridotite, either due to heating or due to penetration of saline fluids is the preferred scenario for the formation of high-Mg carbonatitic HDFs.

The high silica content of the silicic HDFs is not at equilibrium with peridotite and calls for another source rock. Experiments in the eclogite-H₂O system at 4–6 GPa produce highly silicic, water-rich fluids that resemble HDFs (Kessel et al. 2005). Kiseeva et al. (2012) demonstrated that even in the eclogite–CO₂ system, the first melt is silicic, rich in SiO₂, Al₂O₃ and K₂O, which is joined by an immiscible carbonatitic melt at higher temperatures. The carbonatitic melt they produced is low in MgO and its composition resembles that of the low-Mg carbonatitic HDFs, but with low K₂O, reflecting the high CO₂/K₂O ratio of 11 of the starting material (much higher than the mantle value of ~1.5; McDonough and Sun 1995). Other experimental melts in the eclogite–CO₂ system are also close to the low-Mg carbonatitic HDFs (e.g., Hammouda 2003; Yaxley and Brey 2004). The connection between silicic to low-Mg carbonatitic HDFs and eclogites is also supported by the association of such compositions with omphacite and eclogitic garnet micromineral inclusions (e.g., Klein-BenDavid et al. 2009; Weiss et al. 2009, 2015; Gubanov et al. 2019). Subducted sediments are another possible source for producing some silicic to low-Mg carbonatitic HDFs (e.g., Bulatov et al. 2014).

Weiss et al. (2009) suggested that the continuous array from the hydrous-silicic to low-Mg carbonatitic HDFs is the result of dissolution of carbonates in H₂O- and K₂O-rich silicic melts with increasing temperature. Litasov et al. (2010) favored evolution in the opposite direction and argue that in an eclogite–CO₂–Cl system, more silicates are dissolved into an initial carbonatitic melt with increasing temperature and degree of melting.

Recently Elazar et al. (2019) have shown that for eclogites containing both water and CO₂ the initial melt is not silicic or carbonatitic, but rather of intermediate composition. They suggested that it is the CO₂/H₂O ratio of the source rather than temperature that controls the position of the near-solidus melt along the silicic to low-Mg carbonatitic array. Dvir et al. (2017) provided evidence that near solidus fluids and melts in the lherzolite–H₂O–CO₂ system also fall close to the central part of the silicic to low-Mg carbonatitic array. In their experiments with CO₂/H₂O ratios of ~0.1 and 0.5 (by weight), the melt of the CO₂-rich composition has higher MgO and lower SiO₂ content compared with the melt of the H₂O-rich one, supporting the role of the source CO₂/H₂O ratio in controlling the silicic–carbonatitic character of the HDF. Further evidence for the control exerted by the source CO₂/H₂O ratio rather than by temperature and the degree of melting is provided by the composition of the HDFs. Silicic and low-Mg carbonatitic compositions have similar levels of K₂O, TiO₂ and Cl (Table 2). Temperature control should lead to increasing melt fraction and dilution of these incompatible elements, but such a trend is not observed in the analytical data. The similar (U,Th,La)/K of the silicic and low-Mg carbonatitic HDF types (Fig. 10b) further strengthens this suggestion.

In summary, the array represents the melting of eclogites that contain water and carbonates in variable amounts. Figures 7 and 8 show that most of the silicic to low-Mg carbonatitic HDFs fall in the middle of the array, indicating that the majority originate from rocks that carry both water and carbonates. Sediments are less probable, but possible sources of some of the HDFs.

The saline HDFs are the most intriguing as they do not resemble any mantle melts that erupt on Earth’s surface. Melting of peridotite or eclogite source-rocks is unlikely to form K-rich saline HDFs. K and Cl have similar incompatibility, but while the average molar K/Cl ratio of the HDFs is 0.68 ± 0.14 1σ (Supplementary compilation of HDF data, Weiss et al. 2022), the mantle ratio, as defined by MORB and OIB glasses, falls in the range of 6–45 (e.g., Kendrick et al. 2013, 2017). Fractionation of mantle melts is also unlikely. Most carry low concentrations of both K and Cl and require extensive fractionation that is not expected to produce a uniform residual K- and Cl-rich saline HDF.

Klein-BenDavid et al. (2007) noted a consistent negative correlation of MgO and CaO with Cl content in saline HDFs. They examined fractionation of Mg-rich carbonates from a carbonatitic HDF, concluded that extreme fractionation is needed, and noted the difficulty to remove potassium (relative to Cl). Carbonates, phlogopite and high-Si mica micro-mineral inclusions were found in association with saline HDFs (Izraeli et al. 2004, Weiss et al. 2018), but mica removal is limited by the low abundance of silica and alumina, and crystallization of K-bearing carbonates is required. This is not impossible (e.g., Shatskiy et al. 2016, Arefiev et al. 2019), but such phases are rare (Logvinova et al. 2019a). Last, HDFs of intermediate compositions between carbonatitic and saline compositions are not common in fibrous diamonds (Figs. 7 and 8, but see Klein-BenDavid et al. 2004). Similar difficulties exist for a possible fractionation from a silicic composition to form saline HDFs.

The high Cl content of the saline HDFs and the association with water and carbonates strongly suggests a subducted source (Izraeli et al. 2001; Tomlinson et al. 2006; Weiss et al. 2015, 2018; Timmerman et al. 2018a). Indeed, the K/Cl ratio of the saline HDFs is within the range of altered oceanic crust (K/Cl = 0.3–2; Ito et al. 1983; Sano et al. 2008). This alternative is also supported by the pronounced positive Eu and Sr anomalies of saline HDFs in fibrous diamonds from the Slave craton (Weiss et al. 2015), which reflect the involvement of plagioclase during low-pressure crustal processes of protolith formation. Low ³He/⁴He isotopic ratios of South African saline HDF (2.7–4.4 Ra), further strengthen a connection with recycled subducted surface material (Weiss et al. 2017, 2021a; Timmerman et al. 2018b).

The chemical evolution of saline HDFs is not yet clear. Weiss et al. (2015) suggested that they originated as pore fluids trapped in the crust during low-pressure hydrothermal alteration. The high K/Na ratios evolved through the consumption of H₂O and Na during spilitization (hydration of basalt) and/or due to preferential incorporation of sodium into eclogitic minerals (principally omphacite) at high pressure (Izraeli et al. 2001). Dehydration at 150–200 km depth can generate highly saline fluids. Marine sediments are another possible source for the saline HDFs, based on high-pressure–temperature reaction experiments with peridotite (Förster et al. 2019). The equilibrium fluid in these experiments was not analyzed, but the crystallization of euhedral alkali chloride with K/Na = 0.3–6 and Cl/(Na+K) = 1.03 ± 0.44 molar ratios suggest growth from a fluid that is comparable to the saline HDFs.

Once formed, saline HDFs flow through different mantle lithologies. This is indicated by their association with micro-mineral inclusions. The most common are olivine, pyroxene and garnet with compositions typical for ‘normal’ mantle peridotites, websterites or eclogites (Izraeli et al. 2001, 2004; Klein-BenDavid et al. 2004; Tomlinson et al. 2006; Weiss et al. 2015, 2018; Timmerman et al. 2018a,b). These compositions and the association of similar HDFs with different mantle lithologies suggest that the minerals represent a preexisting mantle through which the saline HDF percolates rather than the lithology of their source. Fluid–rock interaction is also reflected in the trace element and isotopic composition of the saline HDFs.

Weiss et al. (2009, 2015) further suggested that the introduction of saline HDFs may induce in situ melting of carbonated-peridotite or eclogite, leading to the formation of high-Mg carbonatitic or silicic to low-Mg carbonatitic HDFs, respectively. This model connects all the HDF types to a common parental saline composition.

High-density fluids are excellent media for the nucleation and growth of diamonds. The lowest conditions for diamond synthesis, 1150 °C, 5.7 GPa, were achieved in the system (K,Na)₂CO₃ + C₂H₂O₄ (oxalic acid) + C (graphite) (Pal’yanov et al. 1999), which is a simple analog of carbonatitic HDFs. Similar temperatures were calculated for minerals included in fibrous diamonds (Izraeli et al. 2001; Weiss et al. 2018). Still, while the growth of fibrous diamonds from the HDFs they trap is widely accepted, the formation mode of NFMC diamonds is debated (e.g., Cartigny et al. 2014; Stachel et al. 2017).

It is agreed that mantle diamonds are metasomatic minerals and that their carbon is generally supplied by the metasomatic fluids (Haggerty 1986; Stachel and Luth 2015). However, two issues are still debated: a) the state of the fluids—subsolidus low-density fluids such as water or methane, or high-density fluids such as the diamond HDFs, kimberlites and similar melts. b) the oxidation state of the fluid—evidence was collected supporting the full wide range of carbon redox states, from carbonates and CO₂ (with C⁺⁴) to CH₄ (C⁻⁴).

Stachel and Luth (2015) used a solidus curve sketched by Wyllie (1987) to argue that the solidus of harzburgites+H₂O+CO₂ lies at ~1270 °C (at P >6 GPa), some 170 °C above that for lherzolite. However, Brey et al. (2011) have shown that at 1100 °C and 6 GPa K-rich carbonatitic melts coexist with harzburgites. The solidus of lherzolite + K₂O + CO₂ overlaps that of the harzburgite system. Adding water may lower the solidi to even lower temperatures.

Low-density hydrous fluids with 60–80 wt% water also exist with peridotite under such conditions (5–6 GPa, 1000 °C, Kessel et al. 2015). But if carbonate is also present, the water fraction decreases to 35–45 wt% (Dvir et al. 2017), not much higher than in the HDFs. At 1100 °C, above the solidus it falls to 28 wt%. It may be even lower if the K₂O/H₂O in the experiment (0.03–0.08) would be closer to the mantle ratio of ~1 (Kessel et al. 2015). We conclude that hydrous fluids at 5–6 GPa and 1100 °C would dissolve incompatible components such as carbonates, K₂O, Cl and incompatible trace elements and would resemble the HDFs documented in diamonds. Presumably, the water fraction in equilibrium with harzburgites is somewhat higher, but we expect that the difference between such assumed hydrous fluid and HDFs is rather small.

The carbonate-bearing HDFs present the best choice for the oxidation state of carbon in diamond-forming fluids, not only for fibrous diamonds, but for most NFMC diamonds as well. HDFs were identified in NFMC diamonds (Weiss et al. 2014; Jablon and Navon 2016; Kempe et al. 2021) and their trace element signature was also found in gem and semi-gem diamonds (Krebs et al. 2019). We note that they may also carry a small proportion of reduced carbon compounds (Kopylova et al. 2010).

A common argument against an extensive role for HDFs as the source of most NFMC diamonds is the oxidation state of the lithosphere that is below the EMOD buffer (MgSiO₃ + MgCO₃ = Mg₂SiO₄ + C + O₂). However, as shown by Stagno et al. (2013), pure carbonate melt is stable at more reducing conditions than EMOD, and if the carbonate fraction is diluted, its stability in the melt extends to cover even more reducing conditions (by up to ~2 log units). As pointed out by Stachel and Luth (2015) the buffering capacity of the lithosphere is small, and an oxidized melt can migrate through a reduced lithosphere, react with it and deposit diamonds over a long distance.

Can HDFs carry other species of carbon besides carbonates? This interesting possibility was presented by Stachel et al. (2017) for a C–O–H low-density fluid. Based on thermodynamic modeling, they argued that CO₂ (or carbonates) and CH₄ can coexist and react with each other to form diamonds. The idea is also supported by thermodynamic modeling of water-peridotite and water-eclogite systems that predict the coexistence of diverse oxidized and reduced carbon species (Sverjensky and Huang 2015; Huang and Sverjensky 2020). Can this be applied to HDFs? Kopylova et al. (2010) reported absorption at 2800–3000 cm⁻¹ in IR spectra of fibrous diamonds from DRC and related it to CH₃ and CH₂ groups. We note that this region is prone to contamination on the diamond surface, but future studies should examine carefully the IR spectra of fibrous diamonds and look for the existence of organic reduced and oxidized species predicted by the thermodynamic models. If found, this will increase the diamond forming capacity of HDFs.

The collection of observations to date identifies HDFs as the best candidates for the formation of most lithospheric diamonds, however they are not the only ones. Observations of CH₄ in diamonds are rare (Smit et al. 2016) and growth from CH₄, or CH₄-H₂O low-density fluids was suggested based on modeling of isotopic fractionation (e.g., Thomassot et al. 2007; Smit et al. 2019). Experimental studies found that “in oxidized fluids (CO₂; CO₂–H₂O; H₂O), the intensity of spontaneous nucleation, the rate of diamond growth on seed crystals, and the degree of graphite-diamond transformation are considerably higher than those in reduced fluids (H₂O–CH₄; CH₄–H₂)” (Palyanov et al. 2015), but a recent study does suggest nucleation and growth of diamond in the harzburgite+CH₄ system (Matjuschkin et al. 2020). Smith et al. (2016) reported the finding of iron–nickel–carbon–sulfur inclusion in lower mantle diamonds, but metallic iron inclusions are known from shallower diamonds as well (Bulanova et al. 1998), and this mode of growth may be important and efficient where metallic melts exist (e.g., below 250 km, Frost and McCammon 2008).

Experimental studies showed that hydrous, saline, carbonate and carbonate–silicate melts with compositions that are similar to all HDF types can efficiently crystallize diamonds (e.g., Akaishi et al. 1990; Pal’yanov et al. 1999, 2002; Litvin 2009; Bureau et al. 2016). Carbon isotopes confirm the growth of diamonds from carbon supplied by the HDFs (Klein-BenDavid et al. 2004: Shiryaev et al. 2005). Growth from a fluid is also consistent with the morphology and symmetric growth patterns of diamonds (Sunagawa 1984; Bulanova 1995; Tappert and Tappert 2005). Their close association with veins and alteration zones in xenoliths (e.g., Schulze et al. 1996; Howarth et al. 2014; Logvinova et al. 2015), and metasomatic minerals (e.g., Meyer 1987; Leost et al. 2003; Bulanova et al. 2004; Zedgenizov et al. 2018), indicate that diamond formation is a by-product of HDF-rock metasomatic interaction.

Elevated concentrations of incompatible elements in garnet and clinopyroxene, which would otherwise be refractory and depleted in composition, are evidence for metasomatism (Shimizu 1975; Hoal et al. 1994; Shimizu and Richardson 1987). Among these, sub-calcic garnets globally exhibit sinusoidal chondrite-normalized (CN) REE pattern, which is, perhaps, the most identified geochemical feature attributed to the formation of diamonds (Stachel et al. 1998, 2004; Grütter et al. 2004). The varying REE enrichment patterns in metasomatized garnets, both in xenoliths and as mineral inclusions in diamonds, are commonly explained as the result of a two-stage evolution of the continental lithospheric mantle, i.e., intensive chemical depletion by melting, followed by re-fertilization via enriched fluids or melts (e.g., Stachel and Harris 2008; Klein-BenDavid and Pearson 2009; Lazarov et al. 2012; Aulbach et al. 2013; Shu and Brey 2015; Shchukina et al. 2019). Stachel et al. (2004) attributed the sinusoidal REE patterns to the interaction of a C–O–H fluid highly enriched in incompatible elements with a depleted ‘pre-metasomatic’ garnet. Alternatively, Pearson et al. (1995) and Malkovets et al. (2007) suggested that the sinusoidal pattern is a primary feature and was formed upon direct garnet crystallization during a reaction between a silica-bearing CH₄-rich fluid and chromite-bearing harzburgite, which represent a pre-metasomatized Archean lithospheric mantle (Griffin et al. 2008). In either case, the fluids responsible for the sinusoidal REE pattern in mantle garnets are also associated with diamond formation.

Methane-rich fluid inclusions were identified in olivine phenocrysts from orogenic harzburgite (Song et al. 2009) and ophiolitic dunite (Liu and Fei 2006), but trace element data of such fluids are not available. Thus, the formation of sinusoidal patterns in lherzolite and harzburgite garnets and garnet inclusions in diamonds by interaction with methane-rich fluids cannot be assessed.

Ziberna et al. (2013) and Shu and Brey (2015) modeled the sinusoidal patterns in peridotitic garnets and showed that they can result through interaction between a representative kimberlite or a hypothetical enriched carbonatitic melt with a depleted garnet. The choice of kimberlites (or any other highly alkaline melt) as a metasomatic agent is reasonable, as they satisfy the need for an LREE/HREE enriched fluid to interact with the ‘pre-metasomatic’ garnet (Stachel et al. 2004). However, their composition at depth is still debatable. HDFs have equal, and some even higher, LREE/HREE ratios and are well characterized, pristine samples of deep fluids capable of metasomatizing the continental lithosphere.

Weiss et al. (2009, 2014) used a simple mass balance expression, ϕC_m,i + (1−ϕ)C_g,i = ϕC_m,f + (1−ϕ)C_g,f, to model the evolution of REE patterns in a garnet during interaction with a HDF. Where ϕ is the melt mass fraction, which in an open system metasomatic process may be larger than the actual porosity, and C_i and C_f are the initial and final concentrations of an element in the melt (m) or the garnet (g). This simplified expression ignores the other phases in the system, but provides a good approximation for harzburgite where garnet is the main depository of many trace elements. Using C_m,i = HDF, C_g,i = ’pre-metasomatic’ garnet of Stachel et al. (2004), and assuming equilibrium at the final stage, C_m,f = C_g,f /D_(garnet/HDF), Weiss et al. (2009, 2014) showed that the reacted garnet (C_g,f) can develop sinusoidal and ‘normal’ REE patterns that are close to those of harzburgite and lherzolite garnets (Fig. 13). Similarly, ‘U’ or ‘V’ shaped REE patterns in ultradepleted harzburgite garnets can be formed due to interaction with a very small amount of HDFs (ϕ ≪ 1%; Gibson et al. 2013). The important terms controlling the amplitude and shape of the garnet pattern are the REE pattern of the initial HDF and ϕ. For example, a more pronounced sinusoidal pattern is developed in garnet interacting with HDFs that have La/Dy_CN ≈ 100 compared to those having La/Dy_CN ≈ 10 (Fig. 13a), and the ‘Dip’ around Dy gets shallower (Lu/Dy decreases) with increasing ϕ (Fig. 13b). Based on this, Weiss et al. (2014) suggested that the progression of peridotite garnets from strongly sinusoidal (mostly in harzburgite) to normal REE patterns (common in lherzolite) can be explained by interaction with increasing amounts of the same HDF, rather than attributing the harzburgite patterns to interaction with low-density fluids and the lherzolite patterns to interaction with melts (Griffin et al. 1999; Stachel and Harris 2008; Banas et al. 2009). Here, we demonstrate that introducing HDFs to an eclogitic ‘depleted’ garnet produces an REE pattern with flat and elevated HREE–MREE and decreasing LREE values, which closely resemble the characteristic ‘normal’ REE pattern of eclogite garnets in mantle xenoliths and diamond inclusions (Fig. 13c; e.g., Stachel et al. 2004; Stachel and Harris 2008; Jacob 2004). Thus, HDFs can account for metasomatism in both peridotite and eclogite, two lithologies that are in close association with them, as evident from the micro-mineral inclusions recorded in fibrous diamonds (e.g., Tomlinson et al. 2006; Weiss et al. 2018).

High-density fluids provide a unique record of the compositions and nature of deep mantle fluids. Their high volatile content (Table 2) and the strong enrichment in incompatible elements of all HDF types (Fig. 9), make them effective metasomatic agents. The presence of HDFs in eclogitic garnets (Elazar et al. 2021) demonstrates their ability to migrate and react with lithospheric rocks over a range of depths (McKenzie 1989). Indeed, the mineralogy and chemical composition of xenoliths/xenocrysts and some alkaline magmas suggest a prevalent involvement of HDFs, in addition to other melts, in mantle metasomatism through time (e.g., Richardson et al. 1984; Erlank et al. 1987; Menzies 1990; O’Reilly and Griffin 2013).

Diamonds are xenocrysts in kimberlites, as indicated by their occurrence in mantle xenoliths (Sobolev and Kuznetsova 1966; Viljoen et al. 1992; Smart et al. 2009; Aulbach et al. 2011), by the high pressures of inclusions in diamond (e.g., Meyer 1987; Liu et al. 1990; Izraeli et al. 2001; Stachel and Harris 2008; Nimis 2020), and by the old ages of most diamonds, much older than that of their host kimberlite (e.g., Richardson et al. 1984; Gurney et al. 2010; Koornneef et al. 2017, and references therein). Still, Akagi and Masuda (1988) and Schrauder et al. (1996) proposed a genetic relation between diamond-forming HDFs and kimberlites at depth, based on their similar trace element patterns. This similarity has been broadly confirmed for HDFs and kimberlites from different lithospheric provinces (e.g., Tomlinson et al. 2009; Zedgenizov et al. 2007, 2017; Rege et al. 2010; Weiss et al. 2011; Skuzovatov et al. 2012). Among the HDFs, the closest resemblance to kimberlites is demonstrated by the high-Mg carbonatitic type. This is based on their common high Nb,Ta/La ratio and their depletion in K, Rb and Cs compared with other highly incompatible elements like Ba, U, Th and La (Fig. 14a; Weiss et al. 2011). Notably, this trace element pattern characterizes the unique HIMU alkali ocean island basalts (OIB; having a high U/Pb ratio, μ) from the Cook-Austral islands (Chauvel et al. 1997; Willbold and Stracke 2006), St. Helena (Kawabata et al. 2011), and Bermuda (Mazza et al. 2019), as well as lavas from the Grande Comore island (Class et al. 1998) and oceanic and continental basanites (e.g., Weinstein et al. 2006; Pilet et al. 2008, 2011).

The trace element patterns of HDFs, kimberlites and HIMU-basalts are subparallel (Fig. 14a), with their main difference being the abundances of highly incompatible elements and varying depletion in Ti, Zr and Hf (and Sr in the case of some basanites) compared to rare earth elements of similar incompatibility. These differences may reflect source heterogeneity, the degree of partial melting, the retention of accessory phases during melting, or different element partitioning in carbonatitic versus silicate melts. An important issue is the close link between the trace-element composition of HDFs trapped at depth and surface kimberlites and basalts, which indicates that the incompatible element content of the latter does not change much on their way to the surface. Moreover, it suggests similar enrichment and depletion processes experienced by their mantle source over geologic time, because batch melting of either a depleted source or a primitive mantle composition cannot reproduce the characteristic negative anomalies of the alkalis together with high Nb and Ta levels of these enriched melts.

Both carbonates (as minerals or melts) and peridotite are required for the generation of high-Mg carbonatitic HDFs and kimberlites (see section ‘Formation of high-Mg carbonatitic HDFs’ above). The same constituents were suggested for the source of HIMU OIBs, based on their high CaO and Ca/Al ratios (Dasgupta et al. 2007; Jackson and Dasgupta 2008; Castillo 2015) and the high Mn/Fe and Ca/Al ratios of their olivine phenocrysts compared to other oceanic basalts (Weiss et al. 2016; Mazza et al. 2019). Similarly, a metasomatized peridotite source was proposed for basanites based on experimentally produced key major- and trace-element features (Pilet et al. 2008, 2011). Within this framework, carbonate can be introduced into peridotite through metasomatic interaction with HDFs and/or kimberlite melts, which form almost identical trace elements enriched sources at different mixing ratios (Weiss et al. 2011). Such interaction, for example, the addition of 0.5% of high-Mg carbonatitic HDF or 2.5% of kimberlitic melt into a depleted peridotite, mimics the trace-element patterns of some Archean metasomatized mantle xenoliths (e.g., Schmidberger and Francis 2001; Grégoire et al. 2003; Simon et al. 2007; Wittig et al. 2008) and the average post-Archean SCLM (McDonough 1990). These mantle-derived samples are a natural manifestation that illustrates the impact of carbonatite/kimberlite on depleted peridotite within the lithosphere. A source with similar trace element compositions is expected to form in the mantle transition-zone and deep upper mantle, as a result of similar interaction between carbonatitic melts which originate from subducting carbonated oceanic crust and the ambient mantle (Walter et al. 2008; Litasov et al. 2013; Regier et al. 2020).

To test the possible petrogenetic link between high-Mg carbonatitic HDFs and kimberlites, Weiss et al. (2011) used the trace element composition of available Archean xenoliths and the average post-Archean SCLM as natural representatives for a metasomatized carbonated peridotite and examined the products of melting. Figure 14b shows that simple batch melting of such a source closely reproduces the pattern of kimberlites. The same source can account for the pattern of HIMU basalts and basanites at higher melt fraction; however, the more compatible elements are too low in the produced melt. Weiss et al. (2016) found that batch melting of a mixture of ‘normal’ (trace-element-depleted) and metasomatized peridotite closely reproduces the complete trace element pattern of such basalts (Fig. 14c). They assumed a simple mixing of 80:20, respectively, between the two lithologies to produce a mixed peridotite source. The more enriched Bermuda lavas lie above the calculated melt, but with parallel pattern (Fig. 14c, Mazza et al. 2019). Such a pattern is reproduced at a lower melt fraction of a source with higher proportions of metasomatized peridotite.

The unique trace element patterns of kimberlites and HIMU-flavored alkali basalts require a distinctive enriched source that carries the signature of carbonatitic metasomatism (Hauri et al. 1993; Dalton and Presnall 1998; Saal et al. 1998; Greenwood et al. 1999; Le Roex et al. 2003; Dasgupta et al. 2007; Brey et al. 2008; Jackson and Dasgupta 2008). The impact of high-Mg carbonatitic HDFs on mantle peridotite can produce such a source. Other HDF types have different compositions and trace element patterns, but similar metasomatic potential upon interaction with mantle lithologies. The passage of such HDFs through and the interaction with different mantle lithologies impart chemical ‘stains’ that result in enriched mantle sources. In turn, the melting of such sources conveys these unique chemical and isotope compositional signatures to mantle-derived magmas that erupt at the Earth’s surface. However, this is not a “chicken & egg” situation. HDFs with ‘Planed’ pattern (Fig. 9) may be derived directly from melting of depleted or primitive mantle and evolve into the other patterns as they percolate through the mantle (Weiss et al. 2013b).

Lithospheric diamonds form at depths > 140 km beneath the Earth’s surface from circulating carbon- and water-rich (C–O–H) fluids. High-density fluids (HDFs) encapsulated in microinclusions in fibrous diamonds are the best-preserved and most common samples of such fluids. Fibrous diamonds and the HDFs they carry are documented in localities of diverse ages and geographical provenance, mostly from Africa, Canada and Siberia, but also from Brazil, India and China, indicating their global distribution and long history in the subcontinental lithospheric mantle.

About 300 fibrous diamonds have been analyzed to date for their HDF major element compositions, roughly 200 of those have also been analyzed for trace element content, and some for their radiogenic isotope composition (mostly Sr isotope ratios). Four types of major element compositions, i.e., saline, silicic to low-Mg carbonatitic and high-Mg carbonatitic HDFs, are ubiquitous worldwide. Trace element compositions of HDFs also have consistent patterns globally, regardless of their host diamond provenance. Two repeating patterns are revealed: a highly fractionated one and a smoother one. The first is characterized by LILE and LREE enrichment, and HFSE and alkali depletion, similar to calcalkaline magmas and continental crustal rocks. The second has lower LILE and LREE abundances and ‘smoother’ overall trace-element pattern similar to oceanic basalts. These two patterns, as well as intermediate compositions, have been identified in all four types of HDFs, indicating the decoupling between major and trace elements of HDFs. The available Sr isotopic signatures suggest a relationship to the characteristics of their mantle sources, as also reflected in the trace element patterns.

A strong connection has been established between silicic to low-Mg carbonatitic HDF compositions and hydrous/carbonated eclogites, based on the association of these HDFs with eclogitic mineral inclusions, as well as their similarity to experimental near-solidus melts in the H₂O-CO₂-eclogite system. Similar arguments have been used to establish a connection between high-Mg carbonatitic HDFs and a carbonated peridotite source. Saline HDFs are more enigmatic and their chemical evolution and source rock are not as clear. At present, the most robust hypothesis is that saline HDFs originate from a subducted source, which is supported by their high Cl content, low K/Cl ratio, positive Eu and Sr anomalies and low ³He/⁴He isotopic ratios. Once formed, saline HDFs can percolate through different mantle lithologies, and induce in situ melting of carbonated-peridotite or eclogite, leading to compositional transition and the formation of silicic and carbonatitic HDF types.

The varying abundance of HDF types in different cratons is fascinating, and likely related to the petrological and physical characteristics of the local lithospheres. For example, the abundance of silicic to low-Mg carbonatitic HDFs in DRC and Jwaneng, two lithospheres with dominance of eclogitic diamonds, is prominent. Other observations, like the high abundance of saline HDFs in the Slave and southwestern Kaapvaal, are clear, but not yet reconciled. In addition, the distinct high-Al trend of silicic to low-Mg carbonatitic HDFs in Siberia compared to other lithosphere provinces, or the striking dichotomy between high-Ba saline HDFs from the Slave and low-Ba ones in the Kaapvaal, are unresolved. Integrated studies on HDF compositions and the local host lithosphere, as well as additional HDF isotopic analyses which are still scarce, are likely to shed light on such issues.

High-density fluids are excellent media for the nucleation and growth of diamonds. Here we summarized recent observations that indicate the involvement of HDFs in the formation of most (but not all) diamonds, fibrous and non-fibrous alike. The most widely recognized geochemical feature attributed to the formation of lithospheric diamonds are metasomatized garnets, which globally exhibit enriched REE patterns. Such patterns can form by the interaction of any HDF type with a depleted garnet in both peridotite and eclogite lithologies, thus strengthening the connection between diamond formation and HDFs. The prevalence of similar metasomatized garnets in mantle-derived xenoliths, together with the similarity in trace elements between HDFs (in particular high-Mg carbonatitic ones), kimberlites, HIMU-flavored alkali OIB and oceanic and continental basanites, reflect the impact of HDFs on the deep lithospheric mantle and their role in deep Earth processes.

High-density fluids provide a unique record of the compositions and nature of deep mantle C–O–H-bearing fluids. Their high volatile content, strong enrichment in incompatible elements and possible relation to subduction processes, make HDFs a key player in the global circulation of volatiles. However, to date, there are no robust age constraints for HDFs, nor for their host fibrous diamonds. Future progress in analytical techniques and/or sampling methods for analyzing HDFs, especially for determining their radiogenic isotope composition, are likely to offer a remedy and provide comprehensive chemical characteristics and meaningful geological ages of HDFs in diamonds. It will potentially elucidate the origin of all HDF types, the relationship between them, and the ages of their host diamonds and the metasomatic processes that led to their formation. Such new data will provide the opportunity to understand the complex history of alteration events in different lithospheric provinces, the type of fluids involved, and relate these deep mantle events to regional tectonics and magmatic history. Since HDFs trapped in diamonds are the most pristine samples of an abundant C–O–H-bearing fluids in Earth’s mantle, they are likely to provide important contributions to our understanding of the deep carbon and water cycle throughout Earth’s history.

We thank Ofra Klein-BenDavid, Oded Elazar, Yael Kempe, and Reed Mershon for sharing their unpublished data and Dan Howell for a modified version of the DiaMap software. We thank Thomas Stachel and Karen Smit, the editors, for handling the paper and Dimitry Zedgenizov, Thomas Stachel, and an anonymous reviewer for constructive reviews that greatly improved the paper. This research was supported by the Israel Science Foundation grant No. 2015/18 to YW and grant No. 734/17 to ON.