Melting phase equilibria of model carbonated peridotite from 8 to 12 GPa in the system CaO-MgO-Al₂O₃-SiO₂-CO₂ and kimberlitic liquids in the Earth’s upper mantle^†

Kimberlites are carbon dioxide (CO₂)- and water (H₂O)-rich, ultrabasic, silica-undersaturated liquids (magmas/melts; Mitchell 1986, 1995; Ulmer and Sweeney 2002; Sparks 2013). Kimberlites perhaps form part of a sequence of silica-undersaturated rocks that can vary quite widely in their composition, and might include rocks as diverse as melilitites, nephelinites, and lamproites. Kimberlites that contain diamonds must have originated from depths of at least 140–150 km in the Earth (Shirey et al. 2013; Sparks 2013).

On the basis of mineralogy, geochemistry, and isotopic compositions, two broad groups of kimberlites have been identified: Group I (basaltic) kimberlites and Group II kimberlites (micaceous; sometimes also called “orangeites”) in previous studies (Smith et al. 1985; Mitchell 1986, 1995; Becker and Le Roex 2006). Because the erupted kimberlites have been extensively contaminated by “foreign” material, and as there is not a single glassy (sensu stricto) erupted kimberlite, such a classification is fraught with its own set of difficulties (Mitchell 2008; Kjarsgaard et al. 2009). Yet, from rare hypabyssal and aphanitic eruptions, estimates of “primary” kimberlite magma compositions have appeared in the published literature (e.g., Kopylova et al. 2007; Kjarsgaard et al. 2009; Patterson et al. 2009 and references therein; Sparks et al. 2009).

A causal link between kimberlites and carbonatites, from direct melting of a carbonated peridotite has been suggested to exist (Wyllie and Huang 1975; Canil and Scarfe 1990; Bailey 1993; Dalton and Presnall 1998; Gudfinnsson and Presnall 2005). That is, in experimental petrologic studies, carbonatites, which are low-degree melting products, give way to kimberlites, as the degree of melting of carbonated peridotite progresses. Among the studies mentioned, the ones of Dalton and Presnall (1998) and Gudfinnsson and Presnall (2005), undertaken in the system CaO-MgO-Al₂O₃-SiO₂-CO₂ (CMAS-CO₂), demonstrate experimentally the continuum between carbonatites and kimberlites. Wyllie (1980, 1987, 1995) and Green et al. (1987) preferred a petrological scenario wherein volatiles are transferred from somewhere in the continental asthenospheric mantle to shallower depths (e.g., asthenosphere-lithosphere “boundary”), and this transfer of volatiles generates kimberlites and carbonatites. Kimberlites are not generated only from high-pressure melting of carbonated peridotite, however. For instance, Kawamoto and Holloway (1997) in their study on melting phase relations of multicomponent, hydrous (water-saturated) peridotite also noticed kimberlite-like liquid compositions at the solidus at about 10–11 GPa; at lower pressures, the liquids resemble basalts and komatiites. Hence, there is experimental evidence that water-rich kimberlites could have originated from direct melting of a hydrous mantle peridotite in the deeper portions of the Earth’s upper mantle.

The other approach in addressing the petrogenesis of kimberlites lies in what is called the “multiple saturation” technique. This approach forms the basis of a petrological philosophy that equilibrium should exist between a primary kimberlite and the source. That is, a primary kimberlite should saturate with a multi-phase mantle assemblage at its conditions of origin (Eggler and Wendlandt 1979; Edgar et al. 1988; Ringwood et al. 1992; Edgar and Charbonneau 1993; Girnis et al. 1995; Yamashita et al. 1995; Ulmer and Sweeney 2002; Mitchell 2004; Girnis et al. 2011; Sokol et al. 2013).

Here, we report melting phase relations of model carbonated peridotite in the system CMAS-CO₂ from 8 to 12 GPa. The objective is twofold: (1) does the earlier-proposed continuum between model carbonatites and kimberlites over 3–8 GPa (Gudfinnsson and Presnall 2005), cease to exist at and beyond 8 GPa? If so, could kimberlites be directly produced at the solidus of model carbonated peridotite at and beyond 8 GPa. This first part is relevant to consider, for, recently a model for the eruption dynamics of kimberlites was developed wherein kimberlites in natural settings were purported to start out as carbonatites; these carbonatites at various depths in the mantle assimilate mantle material (opx) upon their ascent, and become kimberlites (Russell et al. 2012). The second (2) question is could the naturally occurring kimberlites as magnesian as Group IA, and perhaps even more magnesian than this Group, be produced simply from melting of carbonated peridotite, at pressures yet greater than 10–12 GPa (Gudfinnsson and Presnall 2005)?

The intent of the present experimental work is illustrated in Figure 1 (modified after Dalton and Presnall 1998; Gudfinnsson and Presnall 2005). The present experimental study is toward the higher pressure side of the vertical, dashed line on the P-T divariant surface (Fig. 1). The present experimental study builds upon melting phase relations in model carbonated peridotite reported earlier (Dalton and Presnall 1998; Gudfinnsson and Presnall 2005). Besides presenting more complete data set at 8 GPa than was done in Gudfinnsson and Presnall (2005), we further explore the topology of the P-T divariant surface up to 12 GPa. The influence of alkalies (Na₂O and K₂O), water (H₂O), iron oxide (FeO), and titanium oxide (TiO₂), components that might have significant influence on the melting phase relations is unknown at the moment and is not considered here.

The experiments were conducted by using starting mixtures listed in Table 1. These were prepared so as to have a large amount of liquid that is in equilibrium with a peridotite crystalline phase assemblage. This philosophy was adopted to facilitate analysis of liquids using the electron microprobe. Starting mixtures labeled JADSCM-3, JADSCM-7, JADSCM-8, and JADSCM-14, were identical to those used by Gudfinnsson and Presnall (2005), and were taken from the batches prepared by Gudfinnsson and Presnall (2005). Other starting mixtures were prepared afresh (for details, the reader is referred to the studies of Dalton and Presnall 1998 and Gudfinnsson and Presnall 2005). Briefly, all the starting compositions were prepared using a combination of shelf carbonate and oxides: CaCO₃ (Alfa Aesar, 99.998%), MgO (Alfa Aesar, 99.998%), Al₂O₃ (Alfa Aesar, 99.998%), and SiO₂ (Aldrich, 99.995%). Magnesite from Oberdorf (Austria; generously supplied by Peter Ulmer, ETH Zürich) provided all CO₂ and some MgO. This particular magnesite was used in the experimental work on the pseudobinary CaCO₃-MgCO₃ by Buob et al. (2006). The silicate portion of the starting mixtures was prepared first as a glass. All the components were fired in platinum (Pt) crucibles for at least 18–20 h in air: MgO, Al₂O₃, and SiO₂ at 1250 °C, and CaCO₃ at 400 °C. Pt crucibles were used that had not been previously utilized to dry either iron metal or iron oxides. To prepare the starting mixture, fired oxides and CaCO₃ were immediately mixed in appropriate proportions, and ground for at least 1 h under ethanol in an agate mortar. After drying under an infrared (IR) heat lamp, this mixture, contained in a Pt-crucible, was decarbonated in air in a box furnace. To decarbonate, temperature of the furnace was slowly ramped to 1000 °C over 10 h, and maintained at this temperature for 6–7 h. After this step, the Pt-crucible was taken out of the furnace, and then transferred to another high-temperature furnace to prepare glass by melting in air at 1650 °C for about 4 h. To quench this liquid to glass, the Pt-crucible was partially immersed in an ice-water bath, and the resulting glass was optically pure and transparent. This glass was finely ground under ethanol for about 30–40 min, and using the same procedure, glassing was repeated once more. To ensure homogeneity, the silicate glass was re-ground for 1 h under ethanol in an agate mortar, and then dried under an IR heat lamp for about 1 h. Finely ground magnesite was fired at 200–250 °C in air in a Pt-crucible for over 17 h. Silicate glass and magnesite were mixed and ground in an agate mortar under ethanol for another 1 h, and the final contents poured in a glass vial and stored in a desiccator.

All the experiments from 8 to 12 GPa were performed in a traditional, uniaxial, split-sphere, multi-anvil apparatus (MA6/8 module; Sumitomo 1200 press; Rubie 1999; Keppler and Frost 2005) installed at Bayerisches Geoinstitut (BGI). The experiments were performed using Cr₂O₃-doped MgO pressure cells with an edge length of 14 mm. Second-stage tungsten carbide anvils (32 mm edge length) with 8 mm (Toshiba F grade) truncated edge lengths were used in the experiments. Cells in the multi-anvil press were brought to target pressures over 10–12 h, and then temperature was directly increased to the target values (Table 2). On the basis of pressure calibrations reported earlier (Keppler and Frost 2005), the pressure uncertainty in the present set of experiments is estimated to be on the order of 0.2 GPa. On the basis of two-pyroxene thermometry, the temperature gradient in the 14/8 type pressure cells used in the present study, is on the order of ±50 °C. For details regarding the rest of the experimental procedures, the reader is referred to Keshav et al. (2011). After each run, the capsule was recovered from the pressure cell, mounted longitudinally in Petropoxy-154 resin, and ground and polished for optical and electron microprobe examination. Charges were polished under water-absent conditions, using kerosene-based oil (Buehler) on SiC grit (240–1000) paper. Charges were vacuum-impregnated multiple (3–4) times with resin (Petropoxy-154). After this step, regrinding of the charges was done on SiC grit papers under kerosene-based oil (Buehler), until a satisfactory surface for oil-based diamond polishing (3–0.25 μm) was obtained.

The compositions of crystalline phases and quenched liquid were determined by wavelength-dispersive electron microprobe (5-spectrometer JEOL-JXA 8200 Superprobe at BGI) with an accelerating voltage of 15 kV and 15 nA probe current (at the Faraday cup). Quenched liquid was analyzed using a beam diameter of 3–10 μm, and the beam diameter was 1 μm for crystalline phases. The analyses were performed in a fixed spot mode, and were reduced using the ZAF correction scheme. The amount of CO₂ in liquid was calculated by difference, and the four oxides (CMAS) were measured using a combination of diopside, forsterite, pyrope, and enstatite standards.

Experiments reported here were not reversed, and hence attainment of chemical equilibrium is not proven. Lack of zoning in the crystalline phases and internally consistent liquid compositions in the experiments are consistent with near-equilibrium conditions. Liquid did not quench to glass in the present experiments. Instead, it is quenched to a fine mixture of silicate and carbonate fractions, resembling the textures reported previously (Dalton and Presnall 1998; Gudfinnsson and Presnall 2005; Keshav et al. 2011). (The presence of both silicate and carbonate fractions in the quenched materials does not imply that there are two, separate liquids, silicate and carbonate in composition.) For this reason, at least 30–45 analyses were acquired. Quenched liquid is, in some cases, segregated from rest of the crystalline mass. This effect is most likely due to temperature gradient that originates in the pressure cells employed here (also discussed in Gudfinnsson and Presnall 2005). The experimental conditions and composition of phases are shown in Table 2.

All the experiments reported here contain assemblages consisting of forsterite + opx + cpx + garnet + liquid. There is no evidence of liquid immiscibility, and hence the experiments in the present study are consistent with studies at lower pressure (Dalton and Presnall 1998; Gudfinnsson and Presnall 2005). From 8 to 12 GPa in the present experimental study, there is a gradual change in the liquid compositions, from magnesiocarbonatitic at low temperatures to kimberlitic at high temperatures with about 30–40 and 7–18 wt% dissolved CO₂, respectively (Table 2). Changes in liquid compositions, in the form of isopleths for the five oxides, are shown in Figures 2 to 6. These isopleths were sketched on the basis of liquid compositions (Table 2), and represent visual fits to the data; also shown for reference are the isopleths from Gudfinnsson and Presnall (2005). The results of the present experimental study from 8 to 12 GPa confirm the continuum between carbonatites at low temperatures and model kimberlites at high temperatures, as was documented at lower pressures (Dalton and Presnall 1998; Gudfinnsson and Presnall 2005). With increasing pressure from 8 to 12 GPa, liquids on the low-temperature side of the P-T divariant surface become increasingly magnesian with increasing temperature, although they do not really become kimberlite in composition (“Kimberlites” in Fig. 7). We note that with increasing pressure from 8 to 12 GPa, the liquid compositions on the studied P-T divariant surface do not change much (Fig. 7), compared with liquid compositions at lower pressures (Gudfinnsson and Presnall 2005). Along an isobar, however, and with increasing temperature, liquids gradually become more magnesian, and also acquire more silica (SiO₂) and alumina (Al₂O₃), and get gradually depleted in CO₂, a feature similar to what was reported at lower pressure, 3–7 GPa (Dalton and Presnall 1998; Gudfinnsson and Presnall 2005).

Figure 7 shows changes in the compositions of experimentally produced quenched liquid as a function of MgO/CaO vs. SiO₂/Al₂O₃. Experimental liquids near the carbonate-bearing peridotite solidus (Fig. 7) do not become kimberlitic in composition, and, if the phase assemblage at the solidus remains that of a carbonated peridotite, liquids are also unlikely to become kimberlitic at higher pressures (Keshav et al. 2011).

Melt compositions away from the low-temperature side of the P-T divariant surface from 8 to 12 GPa partly extend into the field of (reconstructed) kimberlite compositions (in the area labeled as “Kimberlites;” Fig. 7) as reported in the literature (see Mitchell 1995). Provided that the components not included in this study (iron, alkalis, titanium) do not affect the melting phase relations much, the average composition of kimberlites from Kimberley, South Africa (Clement 1982; shown as “Ki” in Fig. 7) is consistent with an origin at about 8 GPa. Equally, kimberlites of Group IB (Smith et al. 1985; shown as “1B” in Fig. 7), those of Majuagaa, Greenland (Nielsen and Jensen 2005; shown as “G” in Fig. 7), and the high-titanium kimberlites of Lac de Gras, Canada (Kjarsgaard et al. 2009; shown as “C-2” in Fig. 7) may have originated at pressures of about 6–7 GPa. Kimberlites at Jericho, Canada (Kopylova et al. 2007; shown as “J” in Fig. 7), those at Udachnaya-East (Kamenetsky et al. 2007; shown as “U” in Fig. 7), and the low-titanium kimberlites of Lac de Gras, Canada (Kjarsgaard et al. 2009; shown as “C-1” in Fig. 7) appear to have come from the highest pressures of about 10–12 GPa. If, however, the assertion of Kjarsgaard et al. (2009), that Kamenetsky et al. (2007) in arriving at a “primary” kimberlite magma composition at Udachnaya-East have incorrectly estimated the concentrations of water and carbon dioxide from the erupted kimberlites, is taken at its face value, then there are two implications: (1) it would make understanding the origin of Udachnaya-East kimberlites in the light of melting phase relations presented either here or elsewhere in the published literature extremely difficult, and (2) the low-titanium kimberlites of Lac de Gras, Canada (Kjarsgaard et al. 2009; shown as “C-1” in Fig. 7) will become the only ones to have come from the highest pressure of about 12 GPa. The kimberlites of Group IA (Smith et al. 1985; shown as “IA” in Fig. 7) are very far from the highest pressure datum investigated here (12 GPa), and from what we know in the model system CMAS-CO₂, it is unlikely that they could have their origin simply from melting of carbonated mantle peridotite at pressures less than that of about 18–20 GPa (see below). On the other hand, the reconstructed kimberlite at Wesselton, South Africa (Sparks et al. 2009; shown as “WR” in Fig. 7) indicates pressure of origin at about 4 GPa, making it the shallowest kimberlite among those considered here.

There are other kimberlitic compositions, which are even more magnesian than those of Group IA, and some of these, the so-called “contamination-free kimberlites” (Mitchell 1986), occupy roughly half the field of some “Kimberlites” shown in Figure 7 (also see Fig. 5 of Gudfinnsson and Presnall 2005). On the basis of data from 8 to 12 GPa we argue that the influence of increasing pressure beyond 12 GPa, on the liquid compositions on the P-T divariant surface, would be minimal. This is because the liquid compositions on the P-T divariant surface show the largest compositional range over the pressure range of 3–8 GPa, as seen in the widely spaced isobars (and oxide-ratio isopleths) that traverse this surface (Fig. 7). Given these observations, Group 1A and kimberlites more magnesian than those of Group 1A at pressures beyond 12 GPa could not have simply come from melting of carbonated peridotite. Therefore, melting phase relations in the system CMAS-CO₂, especially the closely stacked nature of oxide-ratio isopleths from 8 to 12 GPa (Fig. 7) limits the generation of kimberlites most likely in the Earth’s upper mantle. If the genesis of kimberlites is indeed restricted to upper mantle depths in Earth, then how does one account for the transport of “deep” Transition Zone and even deeper “lower mantle” assemblages in diamonds to the surface of the Earth? Such assemblages have been described in diamonds from localities in Brazil, Canada, Guinea, South Africa, Venezuela, and perhaps from localities in Zimbabwe and Tanzania (Hayman et al. 2005 and references therein; Kaminsky 2012 and references therein). This transport of “deep” diamonds could occur via mantle flow from greater to shallower depths; from such depths later kimberlites could carry such diamonds and their enclosed inclusions, and bring them to the surface of the Earth.

In summary, natural kimberlites (hydrous and carbonated), could likely have their origin in one of the following ways: (1) carbonated and hydrous kimberlites in general, and those belonging to Groups IA and IB, begin as dry carbonatites, and gain water somewhere during their ascent in the Earth’s upper mantle or at very shallow depths in the Earth’s crust; (2) kimberlites begin as wet (water-saturated?) carbonatites, and these carbonatites somehow “donate” water to liquids that eventually are called kimberlites, or (3) kimberlites are simply the low-degree melting products of some sort of a hydrated mantle at high pressures (Kawamoto and Holloway 1997), and eventually become carbonated as they rise through the mantle column. (4) The initial melts at the solidus of carbonated peridotite are carbonatite-like; but one possibility might be that the mantle rises and follows an adiabat (or close), so the (still very low-degree) melt eventually becomes kimberlitic, before it segregates and erupts.

One of the aspects to come from the present experimental investigation is the finding that isopleths along individual isobars in oxide-ratio diagram are closer together to each other than at lower pressures in the same system, CaO-MgO-Al₂O₃-SiO₂-CO₂ (CMAS-CO₂). This observation (Fig. 7) suggests that from 8 to 12 GPa, liquid compositions do not change much, than is the case at lower pressures (Gudfinnsson and Presnall 2005). On the basis of the nature of isopleths (Fig. 7), it would seem that to achieve even a further peripheral change in liquid compositions (and so long as liquids are held in equilibrium with the four-phase assemblage of forsterite, orthopyroxene, clinopyroxene, garnet), a fairly large increase in pressure could be required. Yet, at pressures of 14–16 GPa, orthopyroxene would disappear from the carbonated mantle peridotite composition space (Keshav et al. 2011), and hence the liquid will only be in equilibrium with forsterite (or one of its polymorphs, wadsleyite/ringwoodite), clinopyroxene, and garnet. This loss of orthopyroxene from the studied composition space would increase the variance by 1; the compositions of such liquids in equilibrium with rest of the peridotite crystalline phase assemblage are not known at the moment, but should be a subject of interest because of their potential significance for primary kimberlite-like liquids at pressures corresponding to those of Earth’s Transition Zone (14–22 GPa).

Additionally, in this study we have focused on the systematic evaluation of melting phase relations only of carbonated peridotite, which encompasses a fairly limited composition space in the system CMAS-CO₂. Systematic investigation of fusion relations on pressure-temperature divariant surfaces in other composition spaces, for instance, in broadly carbonated basalt or mixed lithologies (carbonated basalt-carbonated peridotite) would also be important. Likewise, within the model carbonated peridotite composition space, a methodical study of the influence of other components, especially of water, iron oxide, and perhaps alkalies, would be worthwhile. Such studies have the potential to clarify further the causal relationship between carbonatites and kimberlites in the Earth’s mantle.

For discussions on the petrogenesis of carbonatites and kimberlites, the authors thank Hans Keppler, Davide Novella, and Dean Presnall. On different occasions Steven Foley, Roger Mitchell, Michael Walter, a referee who wished to remain anonymous, Journal Associate Editor Bjørn Mysen, and Journal Editor Keith Putirka officially reviewed this manuscript, and are thanked for their comments. Roger Mitchell, Bjørn Mysen, Keith Putirka, and Michael Walter are acknowledged for their tough reviews that considerably improved the contents of this manuscript. We thank our technical and administrative colleagues in France, Germany, and Iceland for their assistance. Bayerisches Geoinstitut, Germany and partial support from the 7th framework CIG Marie-Curie EU grant ( 303301) to the first author supported the present research.