Diamondiferous quartzofeldspathic rocks from the Kokchetav Massif, Kazakhstan, and the Erzgebirge, central Europe, are rare witnesses of melting of sedimentary material at great depths. To better understand the melting process, two powdered samples from these localities were objects of experiments conducted in a piston-cylinder apparatus at pressures of 3–5 GPa. In addition, thermodynamic calculations in the system Na2O-CaO-K2O-FeO-MgO-Al2O3-SiO2-H2O using a haplogranitic melt model yielded isochemical phase equilibria diagrams (i.e., pseudosections) at ultrahigh pressure. The solidus for both rocks was located close to 1000 °C and 1100 °C at 3 GPa and 5 GPa, respectively, in the experiments. Initial potassic to ultrapotassic melts form by phengite breakdown. At ∼200 °C above the solidus, clinopyroxene disappears from the restite assemblage, and coexisting melts are granitic. The restite consists of garnet + coesite (±kyanite) up to temperatures close to the liquidus, which occurs at a temperature ∼350 °C above the solidus. The results of the thermodynamic calculations approximate the pressure-temperature conditions and phase relations around the solidus but increasingly deviate from the experimental results with rising temperature. According to the experiments, the melt that crystallized to diamondiferous saidenbachite from the Erzgebirge should have been as hot as 1400 °C, whereas the ultrahigh-pressure rocks from the Kokchetav Massif experienced temperatures of at least 1200 °C. These high melting temperatures are derived for rocks with no free water, which is the most likely scenario for continental crust taken to mantle depths where diamond forms.


Subduction of continental crustal material during plate collision is widely accepted by the geoscientific community (e.g., Chopin, 2003; Liou et al., 2004; Ernst, 2005, 2006). The fate of this material during deep subduction is not well understood, particularly during melting. Two well-studied occurrences of diamondiferous rocks provide direct evidence of melting at ultrahigh pressure (UHP). These rocks occur in the Paleozoic crystalline complexes of the Kokchetav Massif, northern Kazakhstan, and the Erzgebirge, central Europe, and were first described by Sobolev and Shatsky (1990) and Massonne (1999), respectively. The dominant rock types in the UHP areas of these complexes were formed from sedimentary protoliths. Metamorphic and anatectic processes produced the diamondiferous quartzofeldspathic rocks that are the subject of this paper. Experimental studies of the melting and crystallization of hot metasedimentary rocks at UHP are reviewed below. The previous experiments do not define precise phase relations for supersolidus conditions at UHP, in part because compositional information on the melts is broadly lacking. Previous experiments were carried out at variable and relatively high water contents. Water content has a significant influence on the temperature (T) of initial melt (or fluid) formation in experiments; therefore, we performed experiments using powders of selected rocks in which a constant H2O content was bound to the micas.

Our experimental results are relevant for melting of larger volumes of sedimentary material at UHP in nature. In fact, a free water-bearing fluid phase should also be taken into account to enhance melt formation in corresponding rocks, but this process would probably be limited to channelized pathways, if it occurs in nature in deep-seated and hotter and, thus, already broadly dehydrated portions of subduction zones or related pressure-temperature (P-T) environments at all. In addition to the experimental study, we applied thermodynamic data to model melting of sedimentary material in the system Na2O-CaO-K2O-FeO-MgO-Al2O3-SiO2-H2O at UHP. In spite of the preliminary character of the thermodynamic data for relevant silicate melt (see following), we can demonstrate that the thermodynamic modeling, using the computer software PERPLE_X (Connolly, 1990), supports our experimental results for initial melting. This modeling method has the advantage of the simultaneous calculation of physical parameters (e.g., density) for the bulk (anatectic) rock and chemical parameters (chemical and modal composition) for minerals and melt. Moreover, different rock compositions can be modeled in less time than an adequate experimental study.

On the basis of our new experiments and thermodynamic modeling, we present solidus temperatures for metasediments that are significantly higher than previously reported. We also readdress the peak temperature experienced by diamondiferous rocks from the Kokchetav Massif and the Erzgebirge during their melting at UHP. Finally, we hypothesize that melts generated from metasedimentary rocks at UHP are more common than previously thought, but they are very difficult to identify as such after they have reached upper-crustal levels.


Partially Molten Diamondiferous Rocks of Crustal Origin

Saidenbachite, which is a diamondiferous quartzofeldspathic rock with psammopelitic composition (see Behn et al., 2011), forms several elongated, homogeneous bodies in the central Erzgebirge (especially close to the Saidenbach reservoir), Germany, that are several hundred meters in length. The garnet- and white mica–rich rocks of these bodies were originally called gneisses, but the relatively coarse-grained micas are not oriented, and garnets of similar size (mean diameter ∼2 mm) are homogeneously distributed throughout the rock. Thus, the saidenbachite has been interpreted as a massive magmatic rock (Massonne, 2003) enveloped by garnet-bearing, high-pressure gneisses (Willner et al., 1997). Additional important indicators of the magmatic nature of the saidenbachite are multiphase inclusions in garnet, which are 20–30 μm in diameter and mainly composed of various micas, an SiO2 phase, and microdiamond (e.g., Stöckhert et al., 2001). These polyphase inclusions were interpreted as former melt or C-H–bearing fluid trapped at UHP (Hwang et al., 2001; Stöckhert et al., 2001, 2009; Massonne, 2003; Hwang et al., 2006). Further hints at a magmatic origin are (1) strongly corroded zircon grains that have regrown to enclose microdiamonds (Massonne and Nasdala, 2003; Massonne et al., 2007a) and (2) the clusters of small, idiomorphic, and equigranular garnets enclosed in the core of millimeter-sized kyanite crystals, which also contain microdiamonds in their mantle (Massonne, 2003). On the basis of the multiphase inclusions, the lensoid outcrop pattern, and the igneous texture, it was concluded that the diamondiferous saidenbachite originated by melting of metasediments (Massonne, 2003). Massonne (2003) inferred from the appearance of microdiamonds systematically enclosed in outer growth zones of garnet that garnet cores had survived the melting process (as did some zircon and kyanite). Thus, the metasediments were incompletely molten at UHP. Massonne (2003) deduced a pressure of ∼8 GPa for this event based on the detection of a nanocrystal of TiO2 with α-PbO2 structure (Hwang et al., 2000). However, if this nanocrystal were formed as a metastable phase, the peak pressure could have been lower. The melting temperature of the precursor rock of saidenbachite was estimated to have been ∼1200 °C (Massonne, 2003). On the contrary, Hwang et al. (2006) suggested that the fluid inclusions in garnet were caused only by penetrating fluids at UHP; however, they did not explain, for instance, where these fluids were generated or how the massive saidenbachite bodies were formed. Finally, Stöckhert et al. (2009) presented evidence for the fast ascent (>100 m/yr) of saidenbachite after garnet had enclosed the aforementioned fluid inclusions. This rate is actually only compatible with the magmatic nature of saidenbachite.

Except for diamond, no other typical UHP phase has survived the late-crustal stage of saidenbachite from the Erzgebirge. Massonne and Nasdala (2003) reported large SiO2 inclusions in garnet that could have been former coesite. Oval-shaped K-feldspar–rich multiphase inclusions in the diamond-bearing zone of garnet from a saidenbachite were interpreted by these authors to be pseudomorphs after K-cymrite (see also Massonne and O’Brien, 2003). Thus, this phase must be regarded as a potential constituent of metapsammopelitic mineral assemblages at UHP conditions. Even jadeite-rich clinopyroxene was completely altered to plagioclase-rich pseudomorphs (Massonne, 2003), probably during the latest crystallization event or by a high-pressure metamorphic overprint at P ≤ 1.8 GPa.

Multiphase inclusions (e.g., Massonne, 2003; Korsakov et al., 2010) and a garnet rim with microdiamond inclusions (Massonne, 2003) were also reported from compositionally similar rocks of the Kokchetav Massif. This led Massonne (2003) to conclude that these rocks were also partially molten but to a lesser extent than the saidenbachite from the Erzgebirge. The maximum temperature was estimated for various diamondiferous rocks, including carbonate-rich rocks, from the UHP area of the Kokchetav Massif to have been 1100 °C (Massonne, 2003), at a pressure of 7 GPa (Ota et al., 2000; Massonne, 2003). Other researchers (e.g., Korsakov and Hermann, 2006) suggested a lower peak temperature around 1000 °C for these rocks. A recent study of a diamondiferous carbonate-rich rock from the UHP area of the Kokchetav Massif yielded P-T data (Massonne, 2011) compatible with the previous estimate of 1100 °C at 7 GPa.

The UHP rocks of the Kokchetav Massif are more varied in composition (see Shatsky and Sobolev, 2003) compared to those of the Erzgebirge. At one locality near Lake Kumdy-Kol, the UHP rocks can be as massive as the saidenbachite from the Erzgebirge. Such a massive rock is a kumdykolite used for our experiments. This rock is rich in garnet and quartz and probably originated from a calcareous clastic sediment that may have contained abundant quartz and chlorite and less illite (Table 1). However, some Kokchetav UHP rocks, especially original mica-rich saidenbachites, show a clear orientation of potassic white mica and biotite, and are true gneisses. Massonne (2003) suggested that the gneissosity formed during exhumation at P <1.5 GPa.

The bulk-rock composition of saidenbachite from the Erzgebirge is somewhat variable (Behn et al., 2011) and resembles an ordinary psammopelite (see Table 1). The depletion of Th in saidenbachite compared to unmelted psammopelites (Behn et al., 2011) may indicate that a batch of early melt escaped from the rock. This early melt of the saidenbachite protolith would have been rich in H2O and K and have relatively low viscosity, and it would have disappeared from the rock. However, the amount of early melt should have been very small, since the K2O content of saidenbachite from the Erzgebirge is always above 2–3 wt% (Table 1). The observation that hydrous, K-rich granitic melts have interacted with diamondiferous marbles at Lake Kumdy-Kol (Korsakov and Hermann, 2006) points to a similar phenomenon for the UHP rocks from the Kokchetav Massif. Perhaps this explains the relatively low K content of the kumdykolite used in this study (Table 1), although a chemically unchanged sedimentary precursor rock cannot be ruled out.

Experimental Studies and Thermodynamic Modeling

Several experimental studies at P ≥ 3 GPa have addressed the melting of clastic sediments (Stern and Wyllie, 1973; Huang and Wyllie, 1981; Irifune et al., 1994; Patiño Douce and McCarthy, 1998; Hermann and Green, 2001; Auzanneau et al., 2006). A few investigations present information on the P-T position of the solidus. For example, Wu et al. (2009) put the solidus at 900–1100 °C, whereas Ono (1998) estimated ∼1300 °C at 6 GPa. Significantly lower solidus temperatures of 800–900 °C were reported by Auzanneau et al. (2006) for pressures between 3 and 5 GPa. The cited works do not yield a consistent picture of the melting of metasedimentary rocks, but they do demonstrate that initial melts in metapsammopelites are generated by phengite-dehydration melting at UHP. In addition, the solidus temperature increases with rising pressure even to 24 GPa (Wu et al., 2009).

A serious experimental problem at P ≥ 3 GPa and especially at P ≥ 6 GPa, as inferred for the areas above UHP (Ota et al., 2000; Massonne, 2003), is the fact that corresponding silicate melts and hydrous fluids can become supercritical with rising pressures at relatively low temperatures. For instance, Ferrando et al. (2005) estimated pressures between 3.5 GPa and 6.0 GPa for the beginning of this state in various silicate-H2O systems. This problem is outlined in Figure 1, which shows melt-fluid relations in an H2O-rock (e.g., metasediment with mica as the only hydrous phase) versus temperature diagram. An experimentalist studying the melting of a rock at P < 2 GPa should not be worried about the exact H2O content of the starting material (see the two broken lines in Fig. 1) to determine the solidus T. High H2O contents would result in an easily recognizable quantity of melt immediately after reaching the solidus T with increasing temperature. The liquidus T for a dry rock is, however, only reasonably approached with a low-H2O starting material. For the supercritical system at UHP, the presence of a free H2O-rich fluid phase can cause diagnostic problems already at relatively low T. At these temperatures, the experimentalist using a relatively H2O-rich starting material should observe the dissolution of silicates, probably a significant amount of mica in a metasedimentary starting material, which could be misinterpreted as overstepping the solidus T. With rising T, the amount of these silicates (mica) would continuously decrease. At the same P-T conditions, a rock saturated with H2O bound to minerals (micas), but without additional water, would undergo no change unless the solidus T, as defined here (see right-hand side of Fig. 1), is reached. Exceeding this T causes a significant change of the phase relations, as melt (the silicate-rich domain of supercritical fluids is termed here silicate melt) appears and a solid phase (mica) disappears. The experimentalist using more H2O in the starting material is not able to recognize this solidus T.

Since thermodynamic data for relevant silicate melt in the system Na2O-CaO-K2O-FeO-MgO-Al2O3-SiO2-H2O have been recently derived (Ghiorso and Sack, 1995; Holland and Powell, 2001), it is possible to calculate phase relations of metasediments for supersolidus temperatures at high P. This kind of calculation was already utilized in two-dimensional thermomechanical numerical models related to the subduction of crustal material, for instance, by Gerya and Yuen (2003), Gerya et al. (2006), and Gorczyk et al. (2007). Unfortunately, the results published by these authors do not allow us to understand the details of the melting process at UHP. For this reason, our experimental study was accompanied by thermodynamic calculations in order to elucidate the phase relations and compositions of the melts. We used the melt model of Holland and Powell (2001; see also White et al., 2001), since this model had been successfully applied to calculations of P-T pseudosections for migmatites (White et al., 2003; Diener et al., 2008; Štípská et al., 2008). It yielded reasonable results at pressures as high as 2 GPa, although Holland and Powell (2001) pointed to the applicability of their melt model only at low P. The results of calculations of phase relations with silicate melt between 0.5 GPa and 4.5 GPa (Massonne, 2009) suggest a possible overestimation of the solidus T at the highest pressures of this study.


High-Pressure Experiments

The investigated rock compositions (Table 1) are those of a saidenbachite from the Erzgebirge (KD37, which is virtually identical to samples E99-2a, St6105, and others reported by Massonne, 2003; Massonne and Nasdala, 2003; Stöckhert et al., 2001, 2009 [from nearly the same locality]; see also Behn et al., 2011) and a kumdykolite from the Kokchetav Massif (sample 22756). In order to do the experiments with a low constant H2O content, we dried powders of the two samples; micas constitute the main hydrous minerals. These rock powders were obtained by crushing and grinding, with final processing in an agate mortar.

The finely ground rock powder was put into capsules, 2 mm in diameter and several millimeters long, consisting of Au, Au75Pd25, or Pt (see Table 2). The filled noble-metal containers were heated to 130–150 °C for 1 h, to drive off absorbed water, and welded. The capsules were placed inside a pressure cell (half-inch diameter [1.27 cm]) together with a thermocouple (Ni/NiCr type K, Pt/Pt[90]Rh, or W/WRe type D; see Table 2). A 630 ton press with tungsten carbide pressure vessels was used as end-loaded piston-cylinder apparatus for all experiments up to 5 GPa. The experimental conditions are listed in Table 2. Two experiments at 9 GPa were conducted with a Walker-type inset in this press calibrated against the stishovite-coesite transition. As no correction of the pressure influence on the electromotoric force of the thermocouple was made, we estimate a temperature error in the range of ±10 °C (1σ) at 1000 °C, but a significantly higher error (±20 °C) at 1400 °C. The estimated pressure error (1σ) is controlled by the type of pressure cell. Experiments below 1100 °C were conducted with cells mainly composed of rock salt (see, e.g., Massonne and Schreyer, 1986), whereas at higher T, talc-Pyrex-glass cells with a graphite heater (cf. Akella, 1979) were used. A pressure error of ±3% (Pmax) was estimated for the talc-Pyrex-glass cells, which is significantly higher than that for salt cells.

The noble-metal containers, recovered after the experiment, were embedded in Araldite® epoxy resin and polished to expose central portions of the reacted contents. Minerals and quenched melt (glass) were carefully investigated with a CAMECA SX100 electron microprobe (EMP) with five wavelength-dispersive spectrometers and a backscattered electron (BSE) detector. Run products were first studied with this BSE detector or in better resolution with a LEO 1430 scanning electron microscope (SEM). A focused beam was applied for spot analyses on the EMP, except for the glass, to determine the concentrations of Na, Mg, Al, Si, K, Ca, Ti, Mn, Fe, and Ba in the various phases. Counting times were 20 s at the peak and on the background. Synthetic and natural minerals, glasses (e.g., Ba glass for the BaLα1-peak), and pure oxides were used as standards. The acceleration voltage and electric current were 15 kV and 10 nA, respectively. The PaP correction procedure provided by CAMECA was applied. The SiO2 phases, which could not be unequivocally identified with the EMP, were checked with a HORIBA XploRA confocal Raman microscope collecting spectra by using a laser wavelength of 532 nm.

Thermodynamic Calculations

P-T pseudosections were calculated with the PERPLE_X computer software package (see Connolly, 2005; version downloaded from http://www.perplex.ethz.ch, 17 August 2006) in the system Na2O-CaO-K2O-FeO-MgO-Al2O3-SiO2-H2O for the P-T range of 3–7 GPa and 900–2000 °C using the thermodynamic data file hp02ver.dat. This file mainly contains the data set of Holland and Powell (1998 , updated 2002) for numerous mineral end members. In addition, we used the thermodynamic data of K-cymrite, which are compatible with this data set (Massonne, 2009).

The following data for solid-solution series were selected (stored in file newest_format_solut.dat): clinopyroxene [Omph(HP)], garnet [Gt(HP)], and phengite [Phe(HP)]. For plagioclase and alkali feldspar, the model (“feldspar” in file newest_format_solut.dat) by Fuhrman and Lindsley (1988) was employed. The compositional limits for plagioclase and alkali feldspar in model “feldspar” were set to 10 mol% K-feldspar and 10 mol% anorthite + 80 mol% albite, respectively, in order to distinguish these phases (models “feldspar1” and “feldspar2”) in the graphical presentations produced with PERPLE_X. For the fluid phase, the silicate melt model [melt(HP)] by Holland and Powell (2001; see also White et al., 2001) and pure H2O (CORK: Holland and Powell, 1998) were used. Tests with the combination of the melt model (pmelts) by Ghiorso and Sack (1995) and the aforementioned solid-solution models led to unlikely results due to the incompatibility of these models. This combination was successfully applied in two-dimensional (2-D) numerical modeling experiments due to separation of melt-bearing and purely solid assemblages (Gerya, 2009, personal commun.), but it was avoided here in favor of the solid-solution models previously cited plus the silicate melt model by Holland and Powell (2001).

The bulk-rock compositions (Table 1) were simplified to the system SiO2-Al2O3-FeO-MgO-CaO-Na2O-K2O-H2O in order to apply the silicate melt model. Two different water contents (Table 1) were used in the calculation of P-T pseudosections for KD37 and 22756: (1) The lower H2O content (compositions labeled A in Table 1) results in phase relations without hydrous fluid but small contents of K-feldspar (i.e., phengite and/or K-cymrite contents were relatively high at melt-absent conditions). (2) The higher H2O content (compositions labeled B in Table 1) yielded some hydrous fluid but no K-feldspar at subsolidus conditions.

The P-T pseudosections obtained with PERPLE_X were manually redrawn to smooth the reaction curves (see Connolly, 2005). Specific data were taken from the resulting data files to produce density, modal, and compositional diagrams as a function of T at constant P. The corresponding curves were manually smoothed as well using the phase relations of the P-T pseudosections as a guide.


Experimental Results

BSE images of the run products are shown in Figure 2. The main phases (micas, kyanite, garnet, clinopyroxene, melt, and SiO2 phases) are listed in Table 2 for each experiment and are displayed in Figure 3 as a function of P and T. In addition, critical chemical parameters in terms of oxide contents are indicated in this table as average values of at least eight EMP analyses per mineral (or glass) and experimental run. These parameters are also graphically displayed in Figure 4 as a function of T at a constant P. We selected representative mineral analyses that were close to the average in the corresponding run product, and these are listed in Tables 3–6. Table 7 shows average melt compositions that were recalculated to modal mineralogy according to the norm by Cross, Iddings, Pirsson, and Washington (CIPW norm).

The solidus for both rock compositions was located at 3 and 5 GPa, somewhat below 1000 and 1100 °C, respectively (Fig. 3). The solidus curve has a clear positive dP/dT slope; this is supported by experiment 136, which yielded no melt at 9 GPa and 1200 °C. The liquidus occurs at a temperature that is ∼300–350 °C higher than that of the solidus (Fig. 3). Within the studied P range, the crystallization sequence of pure melt with decreasing T is garnet + kyanite, coesite (quartz), Na-rich clinopyroxene for the psammopelitic composition KD37 and coesite (quartz), garnet, Na-poor clinopyroxene for the composition of the calcareous psammite 22756. Clinopyroxene appears at a temperature ∼200 °C above the solidus (Fig. 3). The main subsolidus assemblage between 3 and 5 GPa is kyanite, Na-rich clinopyroxene, garnet, coesite, and phengite in KD37 and Na-poor clinopyroxene, garnet, coesite, and probably some phengite in 22756. These assemblages are broadly realized in the natural rocks with plagioclase instead of clinopyroxene in KD37 and biotite instead of phengite in 22756 due to a metamorphic overprint at P < 1.8 GPa and 1.5 GPa, respectively. It is evident that the preexisting minerals changed composition during the experiments. For instance, the analyzed Si content of phengite in the subsolidus assemblages of KD37 increased with rising P (see Table 3: e.g., 7.18 Si per formula unit [pfu] at 9 GPa, 1200 °C), and the Ca content of garnet decreased with rising T at constant P for both rock compositions (see Tables 2 and 5; Figs. 4A and 4B). Nevertheless, relicts of minerals from the starting material, such as potassic white mica with Si close to 6.4 pfu, were still present in products of experiments at temperatures below and somewhat above the solidus. This is also obvious in BSE images of run products in sample 22756 (Fig. 2), where a contrast in gray tone is discernible between garnet cores with compositions of garnet from the starting material and newly grown, equilibrated garnet rims. As these rims can be as wide as several microns, we analyzed garnet rims to obtain compositions coexisting with melt. A consequence of this behavior is that the reacting rock composition is actually that of the starting material minus such relict phases. Nevertheless, three experiments (103, 104, and 106; see Table 2) with less well-ground rock powder, thus containing relicts of garnet that were significantly larger than those in the other experiments, yielded virtually the same result as experiments with finely ground rock powder at the same P-T conditions. Only the Fe content was observed to be lower in melt and clinopyroxene of run products with large garnet relicts because Fe is mainly stored in the almandine-rich garnet (see Tables 2 and 5). A similar feature resulted from the loss of Fe in the starting materials due to its introduction into the Pt capsule wall at the applied higher temperatures and longer run durations (e.g., run VB9 at 1400 °C, 95 h in Table 2). To counteract this problem, we changed to an Au75Pd25 capsule and shorter runs, but even at 1400 °C and 1 h, a significant loss of Fe was discernible (see Table 2: run 146 and corresponding Fe/Mg ratios in garnet and melt). The change of the melt composition in Na, K, and Ca with rising T does not seem to be significantly influenced by the Fe loss. We noted the formation of potassic to ultrapotassic melts by incipient melting due to the breakdown of phengite, probably within a narrow temperature interval. Such initial melts are granitic or, at 5 GPa, (alkali)granitic-syenitic (Table 7). Due to the continuous dissolution of clinopyroxene with rising T, these melts become richer in Na and Ca (Figs. 4C and 4D) but maintain their granitic character even close to the liquidus.

An additional characteristic of the experimentally produced initial melts, especially observed at 5 GPa, is their relatively high Ti content at low Mg and Fe contents (Table 7). The coexisting clinopyroxene in runs of sample KD37 also contained considerable amounts of Ti (Table 6). Thus, it was not surprising that rutile was not detected in experimental runs at temperatures above 1100 °C, because all Ti of the starting material was consumed by melt, clinopyroxene, and, to a minor extent, garnet.

Results of Thermodynamic Calculations

The results of the thermodynamic calculations are summarized in Figures 5–7, which show P-T pseudosections mainly for the dry case (composition A) of each of the two selected rocks. Isobaric sections at 4 and 6 GPa (Figs. 6 and 7) demonstrate the change of mode, density, and composition of melt and garnet with T. The calculated results are similar to the experimental results at P-T conditions around the solidus. The solidus curve for KD37 (Fig. 5) is located at 3 GPa at ∼50 °C higher temperatures than those derived from the experiments (Fig. 3). In addition, the experimentally derived dP/dT slope of the solidus is greater than calculated. The calculated UHP mineral assemblage at subsolidus conditions is that observed in the experiments (phengite + clinopyroxene + garnet + coesite ± kyanite). Clinopyroxene is diopsidic in 22756 (low Na content of the rock; see Table 1) and jadeitic in KD37 (see also Massonne et al., 2007b). The calculated Si content in phengite (1000 °C, 4 GPa: Si = 7.40 pfu in 22756, 6.80 pfu in KD37; 1100 °C, 6 GPa: Si = 7.62 pfu in 22756, 6.90 pfu in KD37) is broadly identical to that observed in experiments with KD37 (see Table 3). At temperatures above the solidus, the calculations predict a narrow T interval (≤35 °C at P between 3 and 5 GPa) in which phengite is completely consumed. The experiments presented here do not contradict this result. It is quite possible that the calculated appearance of K-cymrite above 5 and 6 GPa for 22756 and KD37, respectively, at temperatures above the solidus can be confirmed by future experiments.

Initial melts are invariably ultrapotassic according to the calculations (Figs. 6 and 7). The calculated reaction, which characterizes phengite-dehydration melting at UHP for composition KD37, is phengite + Na-rich clinopyroxene + coesite = melt + garnet + kyanite (Fig. 6), as also deduced from the experimental study (Fig. 8), but the observed melts show a lower K2O/Na2O (in wt%) ratio, although they are still ultrapotassic (especially at 5 GPa) to potassic (Table 7; Fig. 4C), compared to the calculated melt composition. Thus, more clinopyroxene than calculated is actually dissolved in the silicate melt after overstepping the solidus by some tens of degrees. A major difference between experimental and calculated results is the T difference between solidus and liquidus. The sequence of solid phases that disappear with rising T during melting is the same as calculated (cf. Figs. 3 and 5), but this difference is 300–350 °C in the experiments and up to more than 900 °C (at least 550 °C for KD37 at 3 GPa) in the calculations. Another significant difference is the approach of the silicate melt to a basaltic composition for sample 22756 at 6 GPa (Fig. 7), which could not be confirmed in the experiments (see Fig. 4D; Table 7). On the contrary, the Mg/Fe ratio in garnet of both rock compositions increases with rising T in the supersolidus range (Figs. 6 and 7), as also observed in the experiments (Figs. 4A and 4B). Similarly, a slight to significant decrease of the CaO content in garnet coexisting with melt already after incipient melting was noted in both experiments and calculations with rising T.


Equilibrium in the Experiments and Comparison with Solidus Relations in Previous Experiments

Our experiments demonstrate systematic changes of the assemblages and chemical and modal compositions of minerals and melt with increasing P and T (Table 2; Fig. 3). The analyses of clinopyroxene, garnet, and melt in the run products, obtained in various areas of the capsule, point to a fairly homogeneous chemical composition of these phases in the entire capsule. Such results can be expected when equilibrium is reached in an experimental run just before quenching. However, relicts of muscovite surrounded by newly formed phengite (at subsolidus conditions in KD37) and nonequilibrated cores of large garnet grains (Fig. 2H) are still present at temperatures as high as 1300 °C. Both features are the result of unavoidable larger grains in the starting material despite intense grinding of the rock samples. Garnet relicts cause the reacting bulk compositions to be somewhat different than those given in Table 1. In particular, a shift toward lower Fe/Mg ratios in coexisting phases was noted in the three experiments with coarser powders because these garnet relicts are rich in almandine component. A similar effect was observed due to Fe incorporation in Pt and Au75Pd25 capsules at the highest temperatures applied here. Despite the shift of the effective bulk-rock composition to lower Fe contents, we have no indication that the derived phase relations are metastable. One could consider whether or not it was appropriate to use fine-grained rock powder instead of a gel, glass, or oxide mixture with the composition of a natural rock. We think that there are two advantages to using powdered rocks as a starting material compared to these alternatives: (1) The rock powder maintains a low and constant H2O content in the experiments. In the rock powder, H2O is bound to a fixed quantity of hydrous minerals only (adsorbed water was driven off before the experiment), whereas it is more difficult to add a small and well-defined amount of H2O to one of the alternative starting materials, which, additionally, have to contain a well-known water content (preferably zero) before this addition. (2) Adding H2O to a glass can result in metastable phases. For example, Hermann and Green (2001) obtained metastable phase relations in their high-pressure melting experiments using glasses in the system K2O-CaO-MgO-Al2O3-SiO2-H2O at 850 °C and 900 °C. These authors deduced from their experiments a garnet-in reaction at 3.5 GPa and 875 °C, i.e., more than 100 °C above the stability curve of pyrope + coesite in the system MgO-Al2O3-SiO2-H2O (e.g., Massonne, 1995), which should be shifted toward lower temperatures by the introduction of CaO. As garnet is already in our starting material, metastable phases seen in the experiments without garnet seeds were avoided in our experiments.

Another important aspect is the P-T position of the solidus, because different solidus curves for sedimentary material at UHP are reported in the literature (Stern and Wyllie, 1973; Huang and Wyllie, 1981; Irifune et al., 1994; Patiño Douce and McCarthy, 1998; Hermann and Green, 2001; Auzanneau et al., 2006). Our P-T position of the solidus (Fig. 3), somewhat below 1000 °C and 1100 °C at 3 and 5 GPa, respectively, is, extrapolated to 6 GPa, between the rough quotations by Ono (1998) and Wu et al. (2009). Our results differ significantly from the lower solidus temperatures reported by Auzanneau et al. (2006), who found the coexistence of melt with phengite over the entire investigated temperature range of 800–900 °C at 2.5 GPa. Within this interval, the contents of phengite decreased and those of melt increased with rising temperature. Our experiments and calculations point to a relatively narrow temperature interval for the coexistence of phengite + melt and suggest that the results by Auzanneau et al. (2006) can only be explained by having more water in their starting material. Therefore, Auzanneau et al. (2006) underestimated the true solidus temperature as outlined in the section on “Experimental Studies and Thermodynamic Modeling” and on the right-hand side of Figure 1 (see gray dashed line). In particular, for supercritical conditions at UHP, the presence of a free H2O-rich fluid phase can cause diagnostic problems inasmuch as a relatively H2O-rich starting material results in significant dissolution of silicates, which could be misinterpreted as overstepping the solidus T. In contrast to the free H2O-rich fluid phase in the experiments by Auzanneau, no such fluid phase existed in our experiments, and the solidus is exclusively defined by phengite-dehydration melting (Fig. 1). We emphasize that phengite dehydration melting is most relevant to nature because free H2O is not expected in deeply buried continental crust that is dehydrated during earlier metamorphic stages.

Our results point to a temperature interval between solidus and liquidus of <400 °C (Fig. 3). In contrast, Stern and Wyllie (1973) and Huang and Wyllie (1981) reported a higher temperature interval. For instance, Huang and Wyllie (1981) observed a T difference of ∼500 °C between liquidus and solidus for a muscovite granite (without free water according to the authors) at 3.5 GPa. This difference amounts to 550 °C for a pelagic clay at 3 GPa (Stern and Wyllie, 1973); however, significant amounts of free H2O were present in the UHP subsolidus assemblage formed from this clay. These experiments should also have led to an underestimation of the solidus T, as in the study by Auzanneau et al. (2006). In accordance with our observations, the restite phases are close to liquidus temperatures. Kyanite + quartz + metastable(?) corundum in muscovite granite and garnet + kyanite in pelagic clay were reported by Huang and Wyllie (1981) and Stern and Wyllie (1973), respectively.

Application to the Studied Natural Rocks

The results presented here can help to better understand the extreme metamorphic conditions at which the studied rocks were subject to melting, and subsequently crystallized during fast ascent to crustal levels. This fast ascent of the UHP rocks from the Saidenbach reservoir (Massonne et al., 2007a) and Lake Kumdy-Kol (Hermann et al., 2001) is supported by U-Pb geochronology, the nitrogen aggregation state of diamond (compare Cartigny et al. [2001] and Massonne [2003]), and the recent consideration of plastic and/or brittle behavior of garnet around internally overpressured fluid inclusions (Stöckhert et al., 2009). The latter two arguments refer only to the stage of the ascent after diamond was formed when the character of the melt was that of a melt mush containing garnet and kyanite.

Incipient melting in psammopelitic rocks at UHP is accompanied by slightly rising garnet content during phengite-dehydration melting (Fig. 8) and falling garnet content during K-cymrite dehydration melting (Fig. 7). A considerable increase in T after passing the solidus is required to consume a significant portion of garnet (Fig. 8) in order to allow large amounts of garnet to crystallize from the melt at UHP. This important conclusion can be applied to saidenbachites from the Saxonian Erzgebirge, where a large amount of garnet (on the order of half of the total garnet volume or more) contains microdiamond and melt inclusions (Stöckhert et al., 2001, 2009; Massonne, 2003). This quantity of garnet must have crystallized from a hot melt because the melt contained little garnet, kyanite, and zircon as restite. The solidus might have been overstepped by ∼250 °C in case of the saidenbachite sample KD37 (Figs. 8 and 9). This estimate is valid only if the metapsammopelitic rock was as dry as our starting material. Even if there was some H2O-rich fluid present in pore spaces before reaching the solidus T, a small batch of initial melt had formed, which escaped from the rock, taking away some trace elements, such as Th (Behn et al., 2011), and the H2O-rich fluid that would have resided in the pore spaces.

In contrast to saidenbachite KD37, the abundant garnet grains in kumdykolite 22756 show thin rims with microdiamond inclusions (Massonne, 2003). Nevertheless, an overstepping of the solidus by 100 °C or even somewhat more is conceivable and is also compatible with the hydrous, K-rich granitic melts (see Table 7; Fig. 7) that have invaded diamondiferous marbles at Lake Kumdy-Kol (Korsakov and Hermann, 2006). As kumdykolite and deformed saidenbachite (diamondiferous gneiss) occur in the vicinity of these marbles, the UHP metasedimentary rocks are assumed to be the source of the melts that have interacted with the marbles.

The process responsible for the deep burial of the precursor rocks of saidenbachite and kumdykolite is not well known. Some possibilities include: lithospheric delamination with continental crust involved in a mantle drip (Massonne, 2005), diapirism into the hot mantle after sediments were very deeply subducted (Behn et al., 2011), or simple subduction without diapirism. A tentative P-T path for the magmatic evolution of the studied rocks is presented in Figure 9, although the maximum pressures experienced by the precursor rocks of saidenbachite and kumdykolite are not well known. Estimates of P ∼8 GPa for the saidenbachite from the Erzgebirge are tentatively based on a nanocrystal of TiO2 with α-PbO2 structure (Hwang et al., 2000). The pressure for the saidenbachite precursor may have been closer to 7 GPa (see Fig. 9), which is in the range of the maximum P for UHP rocks from Lake Kumdy-Kol (Ota et al., 2000; Massonne, 2003). At 7 GPa and temperatures close to the solidus, calculated densities for the studied rocks are 3.3–3.4 g/cm3 (Figs. 6 and 7), approaching the density of the upper mantle. In fact, after incipient melting at 1150–1200 °C (10–25 vol% of melt in the rock), only a moderate reduction of the density of the anatectic rocks occurs, possibly contributing to the prevention of further burial. The heating of the partially molten sediments at depths of ∼200 km is caused by the hot mantle (considering the modern mantle geotherm: ∼1450 °C at 7 GPa; Komiya, 2007) and results in higher quantities of melt (estimated maximum quantities: 80–90 vol% for KD37, 20–40 vol% for 22756) and lower densities of the melt mush (for KD37 ∼2.8 g/cm3; see Fig. 6). These characteristics would make the partially molten rock volumes capable of ascending as melt mush. During this ascent, the melts should have cooled along a nearly adiabatic geotherm, but this cooling cannot result in crystallization because the dP/dT slopes of such a geotherm and of curves for the solidus and the appearance of specific minerals by crystallization, as given in Figure 3, are very similar. Crystallization of garnet (+ kyanite in KD37), coesite, and probably also clinopyroxene could have only occurred along cooling paths with low dP/dT segments at UHP similar to those shown in Figure 9. We speculate that the reason for this unusual behavior is the ponding of melt in the transition asthenosphere and cooler mantle lithosphere. The proposed delamination process responsible for the deep burial of the metasedimentary rocks would have also resulted in a severe disturbance of the thermal structure of the upper mantle. After partial crystallization of the melts, including the formation of microdiamond in the transition region, the melt mush would have risen with rates of >100 m/yr, in the case of the saidenbachite from the Erzgebirge (Stöckhert et al., 2009).

Garnet coexisting with melt in saidenbachite KD37 at UHP is poorer in Ca than garnet at subsolidus conditions as shown in experiment (Table 2; Fig. 4A) and by calculation (Fig. 6). Indeed, the garnet mantle with microdiamond and melt inclusions shows the lowest XCa in garnet from the saidenbachite (Massonne, 2003). One problem with using the observed Ca contents of the grossular-poor garnet for geothermobarometry arises because cation diffusion is fast at the invoked high temperatures. Calculations with diffusion coefficients for Ca, Mg, and Fe in garnet (Perchuk et al., 2009) result in an almost complete homogenization of a Ca-rich core (1 mm diameter, Alm55Pyr25Gr20; see garnet profile in Massonne, 2003) and a Ca-poor, 0.2-mm-thick rim (here: Alm55Pyr45Gr0) within 5000 and 50,000 yr at 1500 and 1300 °C, respectively, and 5 GPa. Two facts lead to the conclusion that the rise of this crustal material during the late ascent phase was rapid: (1) the garnet in KD37 is not chemically homogeneous, and (2) a diffuse boundary still occurs between restitic garnet cores and garnet mantles with diamond-bearing fluid inclusions in kumdykolite 22756.

The estimated maximum temperatures (Fig. 9), 1400 °C for KD37 and 1250 °C for 22756, are higher than those obtained with the Ti-in-garnet thermometer (Massonne, 2003), which is based on equilibria such as pyrope (component in garnet) + rutile = Ti-pyrope (theoretical end-member component in garnet) + quartz/coesite. However, rutile (occurring in garnet cores and the matrix of the saidenbachites) was absent at peak T due to significant melting. This is demonstrated by our experiments where relatively Ti-rich potassic melts formed by incipient melting (Table 7). Thus, the T estimations with the Ti-in-garnet thermometer yielded minimum values for saidenbachite and kumdykolite. The rough peak T estimate of 1050 °C at 5 GPa for a marble of the Kumdy-Kol area (Massonne, 2011) is ∼100 °C below the P-T path of kumdykolite in Figure 9, but both T estimates are fairly uncertain.

General Application to UHP Melting of Metasedimentary Rocks

The low dP/dT segment of the cooling path (Fig. 9) leads to another interesting problem: Was this cooling an unusual process, considering that saidenbachites and kumdykolites are extremely rare at Earth’s surface? Melts formed from deeply buried metasediments should directly ascend to the crust along a more or less adiabatic cooling path. Crystallization should not take place along an adiabatic path through the mantle, and remaining UHP restite phases should instead melt until higher crustal levels are reached. Such hot melts can probably penetrate the overlying continental crust to form granitoid bodies in shallow crustal levels or, more likely, rhyolites at the surface. Such granitic igneous rocks could lack evidence for being generated at great mantle depths for two reasons: (1) A perceptible interaction with Earth’s mantle has yet to be observed for the saidenbachites and kumdykolites, and (2) crystallization starts at relatively low pressures in contrast to the diamondiferous rocks. Thus, it can be very difficult to recognize SiO2-rich igneous rocks that originated either by melting at UHP or by anatexis of lower to medium continental crust.

Incipient UHP melts of metasediments may be more easily recognizable because these melts are ultrapotassic to potassic in composition (see Fig. 4). For example, lamproites are known to form deep in the mantle, possibly triggered by melting of downgoing crustal material. The Miocene ultrapotassic igneous rocks in the southern Pamir are noteworthy as they contain xenoliths of metapelitic rocks formed between 2.5 and 2.8 GPa at 1000–1100 °C (Hacker et al., 2005). These P-T conditions are appropriate to form ultrapotassic melts in metapelites according to the results presented here at 3 GPa.

The sites, where granitic and related melts are generated, are illustrated in Figure 10 and related to diverse geotectonic environments. The traditional environments and corresponding granitic and related melts are: (1) Andean-type magmas formed during the collision of oceanic and continental plates and subsequent assimilation and differentiation processes at upper-crustal levels and (2) S-type granite bodies, often post-tectonic, the melts of which formed in middle- to lower-crustal levels after continent-continent collision. In addition, sedimentary material could be molten after diapirism, as recently proposed by Behn et al. (2011), to form granitic melts as demonstrated in this work. Such melts could form discrete bodies or contribute to the Andean-type magmas in the overlying crust (not necessarily a continental one as shown in Fig. 10) during subduction of oceanic crust. Sedimentary material could also be molten in deep mantle regions after delamination of the continental crust and mantle lithosphere after continent-continent collision. The melts either ascend through the crust to extrude as rhyolites or form small plutons in the upper crust or the lower crust (e.g., saidenbachite lenses in the Erzgebirge). The melts might also interact with the mantle, especially when they are rich in H2O and K, to form various kinds of lamprophyres. Not addressed, so far, are magmas that formed below stable cratonic crust (see Fig. 10 at the bottom). Delamination of eclogitized lower cratonic crust could also result in the formation of granitic magmas at mantle depths, as invoked for the formation of saidenbachite and related igneous rocks. Eventually, ultrapotassic or A-type granitic magmas are compositional equivalents of magmas crystallizing to lamprophyres and saidenbachites that formed below this cratonic environment instead of during the waning stages of a continent-continent collision.


Solidus temperatures derived from experiments and calculations for natural rocks without a free hydrous fluid phase at UHP are higher than those recorded in previous experimental studies (e.g., Auzanneau et al., 2006). Our results from a saidenbachite from the Erzgebirge and a kumdykolite from the Kokchetav Massif give temperatures of ∼1000 °C and 1100 °C at 3 and 5 GPa, respectively (Fig. 3). Incipient melting is characterized by phengite-dehydration melting that leads to potassic to ultrapotassic melts at UHP. More ordinary granitic melt compositions were observed in the experiments with increasing temperature. Liquidus temperatures are less than 400 °C above the solidus temperatures according to the experiments, but are >>500 °C in the calculated phase diagrams. Restite minerals close to the liquidus are coesite, garnet, and, depending on the bulk-rock composition, kyanite.

We conclude that the maximum temperatures were 1400 °C for the saidenbachite from the Erzgebirge and >1200 °C for the kumdykolite from the Kokchetav Massif. The melt mush produced at these temperatures and pressures, approximating 7 GPa, was able to ascend rapidly. During the ascent, minerals such as garnet crystallized and enclosed microdiamonds.

The experimental work was partially financially supported by Deutsche Forschungsgemeinschaft (MA1160/26, FO306/1). Thomas Theye (Stuttgart) and Rolf Neuser (Bochum) supported work with the electron microprobe and the scanning electron microscope, respectively. Alex Perchuk provided us with a garnet zonation calculator including diffusion coefficients of major elements in garnet. The manuscript benefited from the comments of two anonymous reviewers. We also thank Jane Gilotti (Iowa City) for improving the English and John Goodge for editorial handling.