Introduction

Both the original melt composition of primary kimberlitic melt and the ascent mechanism for kimberlites have been contentious issues in understanding their formation. Kimberlite melts do not quench to glass and are extensively modified by mantle and crustal contamination, exsolution of volatiles, and alteration. Therefore, the original composition of these melts is poorly constrained. Studies have shown that these melts ascend rapidly to the surface, traveling from depths greater than 200 km to the surface within hours (Peslier et al. 2008 and references therein; Nishi et al. 2010), entraining large amounts of mantle material en route. Exsolution of volatiles such as CO2 has been suggested to cause this rapid ascent by increasing the buoyancy of the ascending magma or by assisting with dyke propagation (Wilson and Head 2007).

A recent model (Russell et al. 2012) provides both a composition for primary kimberlite melt and a mechanism for CO2 exsolution to drive ascent. Orthopyroxene is the second most abundant mineral in peridotitic mantle, but is nearly absent from the suite of mantle xenocrysts found in kimberlites. Indeed, the few grains that are present show dissolution textures, implying that orthopyroxene is highly unstable in the kimberlite melt (Mitchell 1973, 2008; Russell et al. 2012). Russell et al. (2012) suggest the reason for this instability is because primary kimberlitic melt is carbonatitic, a more SiO2-undersaturated and volatile-rich melt than kimberlite. Because of the low activity of silica of this melt, orthopyroxene (a SiO2 rich mineral) is unstable (Mitchell 1973, 2008; Luth 2009). Assimilation of orthopyroxene into the carbonatitic melt causes an increase in the concentration of SiO2 in the melt, which decreases the solubility of CO2 (Brey et al. 1991; Brey and Ryabchikov 1994; Brooker et al. 2011; Moussallam et al. 2015, 2016). Exsolution of CO2 would then drive ascent of the magma (Russell et al. 2012). In their model, assimilation of orthopyroxene and exsolution of CO2 commences once the melt ascends from its source region, and continues as long as orthopyroxene is not in equilibrium with the melt. It should be noted that the relationship between carbonatitic and kimberlitic melts has a long history (see review in Kamenetsky et al. 2016), so the proposal that the parental melt was carbonatitic was not new.

Russell et al. (2012) tested the viability of their proposed mechanism in experiments at 1000–1200 °C and atmospheric pressure using Na2CO3 as the carbonatitic melt. They explicitly state that their sodium carbonate melts are “…an analog for a natural carbonatitic melt that could be parental (i.e., pre-assimilation) to kimberlites” and further emphasize that “the experiments (e.g., conditions, melt compositions, and scales) are not intended to mimic the natural processes exactly, but to illustrate what is possible and how the process may operate during magma transport.” In these experiments, orthopyroxene reacted with the melt, causing exsolution of CO2 and enrichment of the melt in SiO2. They did not do experiments at higher pressures.

Subsequently, Kamenetsky and Yaxley (2015) studied orthopyroxene interactions with Na2CO3 melt at 2.0–5.0 GPa and 1000–1200 °C. They found that orthopyroxene assimilation did occur, but it did not result in the exsolution of CO2. Instead, at high pressure (5 GPa) it created a silicate–carbonate melt that unmixes at lower pressure to coexisting carbonate and carbonated silicate melts.

Intrigued by the Russell et al. proposal that orthopyroxene assimilation into carbonatitic melts produces CO2-saturated kimberlitic melts capable of rapid ascent, we investigated assimilation of orthopyroxene at high pressure and temperature, to explore possible reactions during ascent. Rather than use Na2CO3 as an analog for the carbonatitic melt, we initially thought to take advantage of previous studies in the model system CaO–MgO–Al2O3–SiO2–CO2 that had determined the composition of melts at the solidus of carbonated garnet lherzolite at different pressures (e.g., Dalton and Presnall 1998a; Gudfinnsson and Presnall 2005). The advantage of this system is that the melt at the solidus, coexisting with a carbonated garnet lherzolite mineral assemblage, is isobarically invariant (see, e.g., discussion in Dalton and Presnall 1998a) and therefore, the bulk composition of the starting material can be tailored to produce large volumes of melt, which allows for more reliable analysis of the melt composition than is possible with small-volume interstitial melts, especially in systems where quench modification of the melt is an issue. The model system does have the disadvantage, however, of not allowing evaluation of the effects of other constituents, such as FeO, Na2O, K2O, or H2O. In addition, solidi of carbonated lherzolite in this system are ~200 °C higher than in the natural system (Gudfinnsson and Presnall 2005).

Dalton and Presnall (1998a) showed that the melts at the solidus of carbonated lherzolite in this system were carbonatitic, with SiO2 between 5.17 and 6.08 wt%, and their (molar) Mg/(Mg + Ca) increased systematically from 0.41 at 3 GPa to 0.56 at 7 GPa. We selected their melt composition from 6 GPa for this study. To explore the possible effects of changing melt composition on the reactivity of orthopyroxene, we also chose a 10 GPa near-solidus melt composition from the study of Keshav and Gudfinnsson (2014), which was more Mg-rich (Mg/(Mg + Ca) = 0.69) and higher in SiO2 (10.22 wt%). To further vary the melt compositions, we took a near-solidus melt from 16.5 GPa from Ghosh et al. (2014) and a hydrous near-solidus melt from 6 GPa from Girnis et al. (2011). These two melt compositions were simplified to CaO–MgO–Al2O3–SiO2–CO2(–H2O), yielding SiO2 contents and Mg/(Mg + Ca) of 15.26 wt% and 0.89 and 13.57 wt% and 0.56, respectively. The compositions of each melt, and the pressure and temperature conditions they were previously determined to be in equilibrium with a lherzolitic assemblage, are listed in Table 1 (Dalton and Presnall 1998b; Girnis et al. 2011; Ghosh et al. 2014; Keshav and Gudfinnsson 2014).

Table 1 Compositions of carbonatitic melt starting materials
Full size table

Most of our experiments were run at lower pressures than those at which these melts had been previously determined to be in equilibrium with a lherzolitic mineral assemblage. This was done to simulate conditions during ascent of such a melt, which is when orthopyroxene should react with the melt according to the model of Russell et al. (2012), as well as the previous theoretical analysis by Luth (2009). This latter analysis predicted that any melt formed in equilibrium with olivine and orthopyroxene would, upon ascent, become out of equilibrium with orthopyroxene, providing a driving force for orthopyroxene to react with the melt.

Methods

Starting materials

The compositions of all the carbonatite melts used for these experiments are in the CaO–MgO–Al2O3–SiO2–CO2 ± H2O system. For the CO2 bearing melts, the starting materials were prepared by mixing and grinding SiO2 (99.95 %), α-Al2O3 (99.99 %), CaCO3 (99.95 %), MgCO3, and MgO (99.95 %) in an agate mortar in ethanol. Before mixing, the silicon, aluminum, and magnesium oxides were fired in a tube furnace for 25 h at 1000 °C (Edgar 1973). The calcium carbonate was dried for 5 h at 300 °C. The magnesite used was a natural magnesite from Mt. Brussilhof, British Columbia, Canada (Enggist et al. 2012). The CO2- and H2O-bearing starting material was prepared using the same set of oxides, and except instead of using MgCO3 and MgO as sources for MgO and/or CO2, magnesium carbonate hydroxide pentahydrate ((MgCO3)4 · Mg(OH)2 · 5H2O), was used as a source of MgO, CO2, and H2O.

Because we wanted to maintain the CMAS nature of the experiments, we decided to synthesize enstatite for use in these experiments, rather than use natural mantle-derived orthopyroxene like Russell et al. (2012) did. The orthopyroxene was prepared by mixing equimolar amounts of SiO2 and MgO under ethanol in an agate mortar for 10 min. The mixture was dried in an oven for 10 min at 120 °C and fired in a tube furnace at 1450 °C for 3 days. Afterward, the powdered mixture was cooled, re-ground, re-dried, and then fired again at 1450 °C for 4 more days. The run product was analyzed on a Rigaku Geigerflex Powder Diffractometer, which confirmed enstatite (both clinoenstatite and protoenstatite) formed. Very minor cristobalite and forsterite peaks were also present.

Experimental setup

For experiments at 2.5 GPa, ~10 mg of starting material was packed into a Pt capsule ~5 mm long with an OD of 3 mm; for experiments >2.5 GPa, capsules were prepared from 5-mm-long by 1.5-mm OD Pt tubing. The capsule was weighed before and after welding in order to detect any sample loss during welding. Minor weight loss of 0.05 mg was routinely observed and was attributed to Pt volatilization during welding. Samples that showed >5 % mass loss were discarded.

The experiments at 2.5 GPa used 12.7-mm talc–Pyrex assemblies in a solid-media, piston-cylinder apparatus (Boyd and England 1960) in the C. M. Scarfe Laboratory for Experimental Petrology, University of Alberta. A W95Re5–W74Re26 thermocouple was used to monitor temperature with no correction for pressure effects on emf. The experiment was first pressurized to approximately 80 % of the run pressure before heating the sample. The sample was heated at a rate of 120 °C min−1 at a constant pressure, and then, the pressure was brought up to the desired level.

All experiments at >2.5 GPa were run in the USSA-2000 multi-anvil apparatus at the University of Alberta. The sample assemblies used were an 18/11 M high-temperature assembly with ZrO2 sleeves (7 mm OD, 3 mm ID) encased in a semi-sintered MgO–5 %Cr2O3 octahedron as a pressure cell (Walter et al. 1995). A stepped graphite furnace was used, with the sample centered within the assembly using MgO spacers. An MgO sleeve protected the Pt capsule from direct contact with the graphite furnace. The W95Re5–W74Re26 thermocouple was inserted to the top of the Pt capsule axially by an Al2O3 four-bore thermocouple ceramic, and no emf correction was used.

Assemblies were inserted into the cavity in the center of eight 32.5-mm edge-length WC cubes with 11-mm triangular truncations. The assemblies were pressurized at room temperature and then heated at a rate of ~30 or ~60 °C/min per hour. The samples were quenched by cutting the power to the assembly, followed by the decompression at ~30 kbar oil pressure/h. Table 2 lists the run conditions used for all experiments.

Table 2 Experimental conditions and results
Full size table

Analytical methods

Experimental charges were mounted, polished, and analyzed for phase compositions using a JEOL 8900R electron microprobe (EPMA) at the University of Alberta. Silicates were analyzed using a 15-kV accelerating voltage, a 15-nA current, and a focused beam with 30-s counting time on the peak and 15-s counting time on the background. Carbonates were analyzed using a 10-µm beam, although in some cases the beam was focused to 5 µm because of the small grain size. An accelerating voltage of 15 kV and a current of 15 or 10 nA were used for the carbonates. Frank Smith pyrope, Fo93 olivine, diopside, dolomite, spinel, anorthite, and MgO were used as standards. Raw count data were reduced to wt% compositions using the Phi-Rho-Z program supplied by JEOL. The resulting data are tabulated in Supplementary Tables 1–3. Backscatter and secondary electron images were taken of the experimental run products to document the textural relationships present. All images were taken at 15 kV voltage, 15 nA current, and a rastering focused beam.

Results

The backscatter electron images in Fig. 1a, b show a pair of experiments run at 2.5 GPa and 1450 °C. The experiment in Fig. 1a used the DP carbonatite starting composition alone, which produced mostly melt at these experimental  conditions. This melt did not quench to glass but to a mass of mostly carbonate crystals with dendritic textures. A few small grains of olivine (Ol) and spinel (Sp) were also present at the edges of the capsule, which are interpreted to be equilibrium phases. The void space present in this experiment is partially a result of plucking, but some may be the result of CO2 exsolution, possibly during the quench.

Fig. 1
figure 1

Backscatter electron images of a experiment PC-DP-RS-001 (2.5 GPa, 1450 °C) using the DP carbonatitic melt (see Table 1), b experiment DPO-RS-002 (2.5 GPa, 1450 °C) using the DP carbonatitic melt plus 31 wt% orthopyroxene, c experiment C-RS-001 (6 GPa, 1400 °C) using the C carbonatitic melt, and d experiment OPX-RS-002 (4 GPa, 1400 °C) using the C carbonatitic melt and 42 wt% orthopyroxene. The irregular voids in c are a result of plucking the friable quench material and do not represent the presence of fluid during the experiment (in contrast to b). Mag magnesite, coes coesite, q-Melt dolomitic carbonatitic melt, opx orthopyroxene, ol olivine, cpx clinopyroxene, sp spinel

Full size image

Adding 31 wt% orthopyroxene to the carbonatite bulk composition at the same P, T conditions produces carbonatitic melt (q-Melt), olivine (Ol), and clinopyroxene (Cpx) as shown in Fig. 1b. The formation of large vesicles and the bulging of the capsule are interpreted to mean that CO2 exsolved from the melt during the experiment. None of the initial orthopyroxene remained. Comparing the electron microprobe analyses of the quenched melt from the two experiments, the assimilation of orthopyroxene increased the SiO2 concentration by ~15 wt% and decreased the CO2 content by ~13 wt% (Fig. 2 and Supplementary Table 1).

Fig. 2
figure 2

Ternary plots showing the composition of the DP carbonatite (Table 1), melts compositions from the orthopyroxene-bearing experiments at 2.5 GPa (DPO-RS-002) and 4.5 GPa (DPO-RS-004), and the composition of the DP carbonatite after the assimilation of orthopyroxene and exsolution of CO2. The green circles represent a range of kimberlite compositions from previous studies (Price et al. 2000; Le Roex et al. 2003; Harris et al. 2004; Kjarsgaard et al. 2009; Kopylova et al. 2009; Russell et al. 2012) (Supplementary Table 4). The dashed line shows the effect of orthopyroxene assimilation on the composition of the DP carbonatite in the 2.5 GPa experiment, and the dotted line shows the effect clinopyroxene assimilation would have at lower pressure

Full size image

At 3.5 GPa and 1400 °C, and at 4.5 GPa and 1500 °C, orthopyroxene showed no sign of assimilating into this melt composition, and there was no evidence of exsolution of CO2 (Table 2, Supplementary Table 1). Electron microprobe analyses confirmed that the melts did not increase in SiO2 nor decrease in CO2 content in the presence of orthopyroxene (Fig. 2, Supplementary Table 1). Indeed, the calculated amounts of orthopyroxene in the experiments were greater than the amount of orthopyroxene added to the starting material, suggesting that orthopyroxene actually precipitated during the experiment. These higher-pressure experiments were repeated with other carbonatitic melts to ensure that this result was not specific to this particular melt composition (Table 2). In these experiments, we did not observe dissolution of orthopyroxene nor evidence of exsolution of CO2. One of these sets of paired experiments used a carbonatitic melt with 10 wt% H2O added is shown in Fig. 1c, d. Although H2O has been shown to destabilize orthopyroxene (Kushiro 1970), even in this case orthopyroxene continues to be stable in the carbonatitic melt.

Discussion

Formation of clinopyroxene at low pressure

In the 2.5 GPa experiment, the production of clinopyroxene in addition to olivine and CO2 is consistent with a reaction of the form:

$$2\;{\text{Mg}}_{2} {\text{Si}}_{2} {\text{O}}_{6} \left( {\text{opx}} \right) + {\text{ CaMg}}\left( {{\text{CO}}_{3} } \right)_{2} \left( {\text{melt}} \right) \to 2\;{\text{Mg}}_{2} {\text{SiO}}_{4} \left( {\text{ol}} \right) + {\text{CaMgSi}}_{2} {\text{O}}_{6} \left( {\text{cpx}} \right) + \, 2\;{\text{CO}}_{2}$$
(1)

taking place. This reaction is the same reaction that forms the “ledge” at ~2.8 GPa in the model lherzolite system CaO–MgO–SiO2–CO2 (Eggler 1975; Wyllie and Huang 1975a, 1976; Eggler 1978; Moore and Wood 1998; Lee et al. 2000), and the idea that exsolution of CO2 from ascending carbonatitic and kimberlitic melts upon intersecting this reaction would take place dates back to the 1970s as well (e.g., Wyllie and Huang 1975a, b). The analogous reaction in the CaO–MgO–Al2O3–SiO2–CO2 is of the form:

$${\text{opx}} + {\text{melt }} = {\text{ol }} + {\text{ cpx }} + {\text{ gt }} + {\text{ CO}}_{2}$$
(2)

(Dalton and Presnall 1998a; Novella et al. 2014). This reaction lies at lower pressures than does (1), and our run conditions were such that our experiment was above this reaction. Combined with the absence of garnet in our experiment, we believe that reaction (1) better describes our experimental results.

In our 2.5 GPa experiment, the 31 wt% of orthopyroxene originally added was completely consumed, which suggests that the efficacy of the orthopyroxene-consuming reaction is rather high. If the stoichiometry of reaction (1) can be used as a guide, ~68 wt% enstatite would react with a dolomitic carbonatitic melt. The original melt composition in this experiment has molar Mg/(Ca + Mg) = 0.509 and low SiO2 and Al2O3 contents, such that it is ~92 wt% CaMg(CO3)2. Such a melt could consume ~67 wt% orthopyroxene; in other words, the melt can assimilate twice its original mass in orthopyroxene.

Attributing orthopyroxene assimilation to the reaction (1) creates an issue because of the accompanying formation of clinopyroxene. Although common in kimberlite xenocryst suites, clinopyroxene is not a phenocryst phase in kimberlites (Mitchell 1986). If this reaction is the mechanism by which orthopyroxene assimilates in kimberlitic melts, then some subsequent process at lower pressure must eliminate clinopyroxene to explain the absence of phenocrystic clinopyroxene in kimberlite samples. Sub-solidus decarbonation reactions can provide some guidance here. As Eggler (1989) noted, there is a lower-pressure decarbonation reaction studied by Käse and Metz (1980) that parallels the sub-solidus analog to reaction (1) and involves clinopyroxene as a reactant:

$$\begin{aligned} & {\text{CaMgSi}}_{2} {\text{O}}_{6} \left( {\text{cpx}} \right) \, + \, 3\;{\text{CaMg}}\left( {{\text{CO}}_{3} } \right)_{2} \left( {\text{dol}} \right) \\ & \quad = 2\;{\text{Mg}}_{2} {\text{SiO}}_{4} \left( {\text{ol}} \right) + \, 4\;{\text{CaCO}}_{3} \left( {\text{cc}} \right) \, + \, 2\;{\text{CO}}_{2} \\ \end{aligned}$$
(3)

(see also Wyllie and Huang 1976; Wyllie et al. 1983). This suggested that there might be an analogous reaction involving carbonatitic liquid rather than crystalline dolomite and that determining the univariant reactions surrounding the invariant points (ol, opx, cpx, dol, L, V) and (ol, cpx, cc, dol, L, V) in CMS-CO2 by Schreinemaker’s analysis would be a useful exercise. The form of the reactions and their relative positions around each of the invariant points are sensitive to the composition of the liquids involved, but a possible result is shown in Fig. 3a.

Fig. 3
figure 3

a Schreinemakers’ analysis showing the positions of the orthopyroxene (blue) and clinopyroxene (green) reactions in P/T space. Each reaction is labeled with the phase not participating in the reaction, b a schematic showing the positions of these reactions with respect to depth and pressure. The solid black line represents the pressure (~3 GPa at 1450 °C) that Wyllie and Huang (1976) predicted that orthopyroxene assimilation reaction would occur. In the diagrams showing in the occurring reactions, blue represents olivine, light green is clinopyroxene, and two different shades of dark green represent olivine growth due to the two different reactions

Full size image

The blue line in Fig. 3a is the reaction (1) by which orthopyroxene reacts with the melt. At lower pressure, the reaction:

$$\begin{aligned} & {\text{CaMgSi}}_{2} {\text{O}}_{6} \left( {\text{cpx}} \right) + 3\;{\text{CaMg}}\left( {{\text{CO}}_{3} } \right)_{2} \left( {\text{melt}} \right) \\ & \quad = 2\;{\text{Mg}}_{2} {\text{SiO}}_{4} \left( {\text{ol}} \right) + 4\;{\text{CaCO}}_{3} \left( {\text{cc}} \right) + 2\;{\text{CO}}_{2} \\ \end{aligned}$$
(4)

is predicted to occur. The production of CO2 in this reaction would provide a second burst of volatile exsolution to further accelerate the ascent of the kimberlite. The formation of calcite at the expense of clinopyroxene would also explain the crystallization of primary calcite in kimberlite (Mitchell 2008). A reaction similar to (4) but involving the CaCO3 component of carbonate melt was used by Harmer and Gittins (1997) to explain dolomitic to calcitic carbonatite evolution, but no experimental constraints exist on either variant of this reaction. A recent study looking at clinopyroxene megacrysts (>1 cm in size) in kimberlites found large reaction rims attesting to the instability of clinopyroxene during the ascent of the kimberlite, and crystallized melt inclusions trapped in these megacrysts consist largely of olivine and calcite showing possible evidence of this reaction occurring in nature (Bussweiler et al. 2016).

Melt evolution

Would assimilation of orthopyroxene and clinopyroxene change a carbonatitic melt to a kimberlitic one? Assimilation of orthopyroxene causes the primary carbonatitic melt to increase in SiO2 content and decrease in CO2, approaching kimberlitic compositions, but not reaching them with 31 wt% orthopyroxene added (Fig. 2). Assimilation of clinopyroxene would cause more CO2 exsolution, moving the melt composition along the dotted line (Fig. 2a), into the field of kimberlite compositions. Therefore, it is possible that primary kimberlites are indeed carbonatitic melts as suggested by previous authors (see historical review in Kamenetsky et al. 2016), or possibly a range of carbonatite to transitional melts, that reach kimberlite compositions based on varying amounts of orthopyroxene assimilation, with xenocrystic clinopyroxene also possibly contributing in a secondary manner, which also explains why there is such a large range of kimberlite compositions observed worldwide.

Conclusions

The summary of our model, which builds off the work of Russell et al. (2012), shows that kimberlite ascent at low pressure (less than ~3.5 GPa) is driven by two reactions, one at a pressure between 2.5 and 3.5 GPa and one at an undetermined pressure below 2.5 GPa (Fig. 3b). The first reaction, which is the one that produces the “ledge” in the carbonated peridotite solidus, causes orthopyroxene to assimilate into the carbonatitic melt leading to the crystallization of olivine and clinopyroxene and the production of CO2, which increases the buoyancy of the melt and drives ascent. Once the melt reaches lower pressures, the clinopyroxene formed in the previous reaction and possibly some xenocrystic clinopyroxene is consumed in the proposed second reaction, forming more olivine and calcite (or possibly a calcite-rich melt). This reaction also produces more CO2, further accelerating ascent. This crystallization of olivine may contribute to the polybaric olivine assemblage observed in kimberlites (Kamenetsky et al. 2008; Mitchell 2008; Brett et al. 2009; Arndt et al. 2010; Bussweiler et al. 2015; Sobolev et al. 2015) and causes the residual melt to evolve to more CaO-rich compositions which results in calcite later crystallizing as primary groundmass mineral. The consumption of orthopyroxene in the first reaction explains its scarcity in kimberlite samples at the surface. Clinopyroxene xenocrysts and megacrysts may be preserved only because they react at a lower pressure and would stop reacting once the melt becomes too SiO2 rich. The assimilation of orthopyroxene in the first reaction, the possible contribution of xenocrystic clinopyroxene in the second, and the resulting exsolution of volatiles drive carbonatite compositions to kimberlite before emplacement.