Introduction

Fluids and melts are the main agents responsible for mass transfer and element recycling during subduction, thus playing a major role in the chemical evolution of the Earth (Hermann et al. 2013). While deep fluids are characterized through fluid inclusion studies in high-pressure (HP) and ultrahigh-pressure (UHP) rocks (e.g., Touret 2001; Ferrando et al. 2005; Malaspina et al. 2006, 2015; Frezzotti and Ferrando 2015), partial melts produced during deep subduction have been mainly investigated via melting experiments (e.g., Hermann and Rubatto 2014 and references therein). Preserved portions of natural partial melt have been recently identified in HP felsic granulites of the Orlica-Śnieżnik dome (Bohemian Massif) as polycrystalline inclusions (Ferrero et al. 2015; see also Walczak 2011), also called nanogranites (Cesare et al. 2009). Their characterization after experimental re-homogenization provided the first compositional data, including H2O content, directly measured in situ of melt produced under near-UHP conditions, 875 °C and 2.7 GPa (Ferrero et al. 2015). Through studies of nanogranites, partial melts can be characterized directly in the source rocks, most likely before any modification occurred (Cesare et al. 2011; Ferrero et al. 2012, 2014; Cesare et al. 2015). This approach provides novel and crucial data necessary to better describe and quantify crustal differentiation (e.g., Brown 2013) as well as the influence of melting in the geodynamic evolution of orogens by providing reliable estimates of melt viscosity (Bartoli et al. 2013, 2014).

In the Orlica-Śnieżnik granulites, a unique assemblage including kokchetavite, kumdykolite and cristobalite is present in nanogranites without cracks connecting with the matrix (hereafter referred to as “preserved”). Kumdykolite is an orthorhombic polymorph of albite, with a disordered structure consistent with crystallization at high T (Hwang et al. 2009), while kokchetavite is a hexagonal polymorph of K-feldspar (Hwang et al. 2004). Both phases have been interpreted to crystallize metastably in the stability field of albite and sanidine, respectively (Hwang et al. 2004, 2009). Previously kumdykolite and kokchetavite have been mainly reported in diamond-bearing metamorphic rocks from the Kokchetav Massif (both phases; Hwang et al. 2004, 2009) and the Bohemian Massif (only kumdykolite; Kotková et al. 2014; Perraki and Faryad 2014), while Orlica-Śnieżnik granulites never experienced pressure in excess of 3.0 GPa (Ferrero et al. 2015). In the present study, the crystallization of these polymorphs inside these preserved inclusions—a closed system isolated inside a stiff host such as garnet—is the ideal setting to calculate the pressure–temperature (PT) path followed by the inclusions during cooling with the method of Angel et al. (2014, 2015), thus providing new and more precise constraints to their formation. Neither kumdykolite, kokchetavite nor cristobalite is present instead in inclusions with cracks reaching the garnet–matrix boundary (hereafter referred to as “decrepitated”), despite both preserved and decrepitated inclusions originally contained the same melt and often occur in the same garnet (Ferrero et al. 2015).

This study presents the results of the investigation of the Orlica-Śnieżnik nanogranites by micro-Raman and electron-microprobe means. The systematic difference in phase assemblage between preserved and decrepitated inclusions as well as the significance of kumdykolite and kokchetavite and their petrological implications for the inclusions cooling history is discussed in detail, with the ultimate goal being the provision of novel mineralogical and microstructural criteria necessary for a more rigorous interpretation of these preserved portions of deep partial melt.

Samples and analytical methods

The investigated inclusions occur in HP granulites with a granitic protolith composition sampled from Stary Gierałtów (N50°18,509′, E016°56,032′) in the Orlica-Śnieżnik dome (NE Bohemian Massif), Poland (Anczkiewicz et al. 2007). The granulites, mainly quartzo-feldspathic rocks with minor mafic omphacite granulite lenses, form a narrow, ~15-km-long, SW–NE trending body, hosted in migmatitic orthogneisses (Gierałtów gneisses). Metamorphism occurred during the Variscan orogeny at c. 350–330 Ma (Anczkiewicz et al. 2007).

The samples comprise abundant sub-mm-sized garnet porphyroblasts, up to 30 % in modal amount, scattered in a leucocratic matrix of quartz, plagioclase, and perthitic feldspar. Garnet is generally subhedral to euhedral (Fig. 1a) and commonly contains inclusions of nanogranite along with quartz, mesoperthite, kyanite, and rare clinopyroxene. The garnet is homogeneous in terms of major elements and its internal portion, where the inclusions are located, and exhibits high almandine and grossular contents (Ferrero et al. 2015). PT estimates for the granulites range from 1.8 to 3.0 GPa at 900–1100 °C for the metamorphic peak, which was followed by rapid isothermal exhumation to shallower depths (Kryza et al. 1996; Bröcker and Klemd 1996); no clear evidence for the former presence of coesite has been identified in the samples used in this study. Geothermobarometric calculations and re-homogenization experiments on the inclusions constrain partial melting and garnet formation near or at the metamorphic peak, at 2.7 GPa and 875–900 °C (Ferrero et al. 2015).

Fig. 1
figure 1

Polycrystalline inclusions (nanogranites) in Orlica-Śnieżnik granulites (Bohemian Massif), plane-polarized light photomicrographs. a Garnet porphyroblast showing MI of variable size. The large inclusions have cracks (gray arrow), often reaching the garnet–matrix boundary, whereas the smaller ones (white arrow) are preserved. b Close-up of a preserved inclusion with linear boundaries, almost approaching a negative crystal shape. c Cluster of preserved and decrepitated inclusions with subparallel cracks connecting the inclusion with the matrix surrounding the garnet (black arrow). Mineral abbreviations after Whitney and Evans (2010)

Full size image

For this study, more than 30 inclusions hosted in garnet have been investigated, in polished thin sections and separated garnets mounted in epoxy. After identification of suitable inclusions using a polarized-light optical microscope under transmitted/reflected light, high-resolution microstructural investigation was performed using a field emission gun electron microprobe (FEG-EMP) JEOL Hyperprobe JXA-8500F (Natural History Museum, Berlin). The polymorphs of feldspars and quartz have been identified by micro-Raman spectroscopy, using a HORIBA Jobin–Yvon LabRAM HR 800 (Institute of Earth and Environmental Science, University of Potsdam) equipped with a Peltier cooled multichannel CCD detector and coupled with an Olympus BX41 petrographic microscope. An air-cooled Nd:YAG laser was used for excitation (λ = 532 nm, laser power on the sample was 2–3 mW) with a grating of 300 lines/mm, slit width set to 100 μm, and confocal hole set to 200 μm. The Raman spectra of the crystal phases were acquired using a 100× objective between 100 and 4000 cm−1, integrating three repetitions of 60 s, with spectral resolution of 10 cm−1 and compared with available literature and databases. Raman maps of two different preserved inclusions were acquired using a spot diameter of 1–1.5 µm and consist of a grid of 15 × 15 µm2 with equidistant points separated by 0.75 µm. Each spectrum was acquired using a 100× objective between 100 and 4000 cm−1, integrating three repetitions of 30 s, with spectral resolution of 10 cm−1. Mineral phases were analyzed using a JEOL JXA-8200 microprobe (Institute of Earth and Environmental Science, University of Potsdam) at conditions of 15 kV, 8–15 nA, and variable beam diameter depending on the size of the phases (1 or 3 µm), in order to avoid, or minimize, the contribution of the surrounding phases and host.

Results of microstructural and mineralogical investigation of HP nanogranites

The investigated polycrystalline inclusions have different sizes, from 5 to 150 micron in diameter, and occur as clusters in the inner part of garnet (Fig. 1a). The preserved inclusions are generally the smallest, ≤20 µm in diameter, and have a negative crystal shape, as is common in primary inclusions in garnet (Fig. 1b; Ferrero et al. 2014). Cracks of very limited extension are locally visible around preserved inclusions (Fig. 2b, d). Decrepitated inclusions have diameters of 40–150 µm, with microcracks that cross the whole garnet, reaching the garnet–matrix boundary (Fig. 1c).

Fig. 2
figure 2

FEG-EMP backscattered images of representative preserved inclusions. a Kumdykolite (Kml) in interstitial position, with accessory epidote indenting the MI walls; b Kokchetavite (Kok), kumdykolite, cristobalite (Crs), and glass coexist in the same inclusion. Kokchetavite and cristobalite occurs as large euhedral crystals, whereas kumdykolite forms fine-grained crystals in a glassy groundmass. c Inclusion with microporosity. Calcite is visible as a minor crystallization product. d Inclusion with coexisting K-feldspar and kokchetavite. Gray arrows cracks

Full size image

Preserved inclusions consistently contain the same unusual assemblage consisting of kumdykolite, kokchetavite, cristobalite or quartz (or more rarely tridymite), and glass along with calcite as a minor component. Kumdykolite forms anhedral crystals in fully crystallized inclusions (Fig. 2a, d), whereas it is fine grained and associated with glass in partially crystallized inclusions (Fig. 2b). The Raman spectrum of kumdykolite (Fig. 3a) shows the characteristic peaks at 493, 221, 156, and 409 cm−1 (Hwang et al. 2009; Kotková et al. 2014), and it varies slightly in composition between different inclusions, from Ab83 to Ab97 (Table 1). The glass (Fig. 2b) is residual (see discussion in Ferrero et al. 2015), is granitic in composition, and contains a significant amount of H2O (Fig. 4). Kokchetavite forms subhedral grains on the nanogranites walls (Fig. 2b) and has a chemical composition similar to that of sanidine (Table 1). Its Raman spectrum shows peaks at 395, 110, and 835 cm−1 (Fig. 3b), identical to that of synthetic kokchetavite (Kanzaki et al. 2012). This phase is less common than kumdykolite, and K-feldspar often occurs in the inclusions instead of kokchetavite; both phases do, however, occur together in a few inclusions (e.g., Figure 2d). Cristobalite often occurs with kumdykolite in an interstitial position, in some cases with elongated shape (Fig. 2b, c). It shows characteristic Raman peaks at 421, 230, and 117 cm−1 Fig. 3c) and contains only limited amounts of impurities such as Al, Fe, and Ca (Table 1). Tridymite was observed instead of cristobalite in one single inclusion. Quartz is always present in the investigated inclusions when cristobalite (or tridymite) is absent. Phengitic muscovite, with a minor celadonite component, occurs in some inclusions where it appears as euhedral grains (Fig. 2b, d; Table 1). Biotite (X Mg = 0.46–0.49, TiO2 negligible: Table 1), is generally interstitial (Fig. 2a, d). A carbonate phase, identified in several preserved inclusions (Fig. 2c), shows the characteristic Raman bands of calcite at 1088, 712, 284, and 157 cm−1 (Fig. 3d).

Fig. 3
figure 3

Raman spectra of phases in preserved MI. Only background subtraction has been applied to the reported spectra. a kumdykolite; b kokchetavite; c cristobalite. No H2O-related signal is visible in the region 3400 and 3800 cm−1, confirming that K-cymrite is not present in the investigated inclusions (compare with data from Kanzaki et al. 2012). d Combined spectrum of calcite, white mica, and garnet from the partially crystallized inclusions in Fig. 4

Full size image
Table 1 Representative electron microprobe analyses of the relevant phases
Full size table
Fig. 4
figure 4

Raman spectrum of the glass visible in Fig. 2b

Full size image

Inclusions on the polished surface of the garnet do not allow verification of the absolute absence of cracks, as they may have been present in the portion of inclusion removed during sample polishing. Raman maps were therefore acquired on preserved inclusions located below the surface of the sample, confirming the presence of the same phase assemblage reported in preserved inclusions exposed on the garnet surface (Fig. 5).

Fig. 5
figure 5

Raman map of a preserved MI below the garnet surface. The different maps show the spatial distribution of the following peaks coded by colors: 356 cm−1 for the host garnet (blue), 420 cm−1 for cristobalite (green), 490 cm−1 for kumdykolite (red), 1086 cm−1 for calcite, 3556 cm−1 for OH bands in the glass (green). The identification of kokchetavite in unexposed inclusions is extremely difficult due to the partial superimposition of its main Raman band (395 cm−1) with signals from glass, mica, and garnet

Full size image

Decrepitated inclusions (Fig. 6a, b) contain K-feldspar (sanidine), plagioclase, and quartz instead of kokchetavite, kumdykolite, and cristobalite, respectively, as verified by Raman measurements. Two different generations of plagioclase can be distinguished based on petrographic and compositional features (Fig. 6). Euhedral crystals termed Pl1 show higher An content, whereas more albitic, anhedral grains are termed Pl2 (Table 1; see also elemental maps in Fig. 2 of Ferrero et al. 2015). Grains of Pl2 are intimately associated with elongated and dendritic crystals of K-feldspar (Fig. 6c) which are locally vesicular (Fig. 6c, d). In addition, biotite is always present, whereas white mica is very rare, and glass and calcite are absent. Epidote is a common phase both in preserved and decrepitated melt inclusions (MI), with much larger grains than the other mineral phases (Figs. 2a, 6b), and it often indents the MI walls (Fig. 2a), besides being present as mineral inclusions in garnet. For these reasons, it is interpreted as an accessory phase, formed during crystallization of the plutons which represent the protolith of the Orlica-Śnieżnik granulites, as also suggested by its composition (Table 1) and the presence of an allanite core (Schmidt and Poli 2004) visible in Fig. 6b (see also Ferrero et al. 2015).

Fig. 6
figure 6

Microstructural features of decrepitated inclusions, FEG-EMP backscattered images. Black arrow porosity. White star elongated/dendritic Kfs. a, b Decrepitated inclusions characterized by abundant porosity and decrepitation cracks (white arrows) reaching the garnet boundary; b Detail of elongated Kfs crystals in a groundmass of plagioclase from figure a (white square); c Dendritic crystals of Kfs in decrepitated inclusions

Full size image

Discussion

Nature and behavior on cooling of near-UHP nanogranites

Polycrystalline inclusions from the Orlica-Śnieżnik dome granulites were originally droplets of granitic melt (Ferrero et al. 2015). After entrapment, each inclusion underwent different degrees of crystallization, as indicated by the presence of residual glass in some cases: an occurrence already observed in other case studies of anatectic MI (Ferrero et al. 2012).

Evidence for chemical interaction with the host garnet, such as compositional inhomogeneities in the garnet directly surrounding the inclusions, or embayments in MI walls, is lacking in preserved MI (Ferrero et al. 2015). Although the presence of partially healed cracks around the inclusions may only be excluded by TEM investigation (e.g., Ferrero et al. 2011), the successful re-homogenization of MI at PT conditions consistent with independent calculations suggests that no H2O was lost after MI formation and also that H+ diffusion through the host can be excluded (Bartoli et al. 2014; see also discussion in Ferrero et al. 2015). Preserved inclusions thus behaved as closed systems after entrapment, maintaining their bulk composition.

In the investigated inclusions, kumdykolite and cristobalite are almost ubiquitous, whereas kokchetavite is less common. These phases coexist with phengitic muscovite, biotite, and calcite along with epidote as an accessory mineral. With the exception of epidote, all of these minerals are absent in the host garnet, confirming that these phases are crystallization products of the original melt. As there is no obvious mechanism or reason why metastable kumdykolite, kokchetavite, and cristobalite should have transformed from their thermodynamically stable analogues (albite, K-feldspar and quartz), we can infer that these phases crystallized directly from the melt. This is also supported by the common occurrence of residual glass in preserved MI, coupled with the lack of structures such as granoblastic textures inside the inclusions that would indicate a later general rearrangement of the internal crystal boundaries. The mutual microstructural relationships between the different phases instead suggest a quite well-defined crystallization order (visible in Fig. 2b, d): phyllosilicates → kokchetavite → cristobalite → kumdykolite → calcite.

The formation conditions of the phases within nanogranites are constrained by their behavior as isolated systems enclosed in garnet. During exhumation, the internal pressure conditions of the inclusions do not follow those of the host rock (Angel et al. 2015), as the PT paths of rock and inclusion are decoupled (Fig. 7). Instead, garnet controls the volume variation of the inclusion on cooling because of the large difference in compressibility between the melt and the garnet. Because the garnet is significantly stiffer that the melt, the MI is constrained to have a smaller volume and thus has a higher pressure than if it was a free phase (Angel et al. 2015). The garnet therefore, in some sense, “protects” the inclusion from the external pressure changes. The PT path followed by the inclusion before crystallization has been calculated following the methods of Angel et al. (2014, 2015), and it is shown in Fig. 7 (blue line). For the garnet, the parameters for a thermal pressure equation of state (EoS) were derived by linear interpolation from values for the garnet end members. The melt was treated as a single fluid using the EoS model and parameters for the components (Holland and Powell 2011). Although the crystallization T of the melt in the inclusions is unknown, an undercooling of 50–70 °C can be reasonably assumed as necessary to start the process. After entrapment of the MI at 875 °C and 2.7 GPa, the host rock experienced a near-isothermal decompression to lower crustal levels (~1 GPa, see, e.g., Bröcker and Klemd 1996; Anczkiewicz et al. 2007). Because it was trapped inside the relatively stiff garnet, the internal P of the inclusion lagged behind the external pressure on the garnet and only reached 2.0 GPa at 800 °C (Fig. 7). Thus, at this temperature, the inclusions will have experienced an over-pressure of approximately 1 GPa with respect to the external lithostatic P (Fig. 7). A precise calculation of the evolution of the internal P after crystallization starts is hampered by the lack of knowledge of exact formation T and thermodynamic properties under HP conditions of the polymorphs identified in the preserved inclusions, and thus, the post-crystallization path visible in Fig. 7 (dashed blue line) must be considered as a semiquantitative estimate. Crystallization is, however, known to significantly change melt properties by increasing the relative water content in the remaining melt, and therefore, it likely results in an even faster decrease in internal P during crystallization and cooling (Steele-Macinnis et al. 2011). Because of the near-UHP conditions at which the original melt was trapped, however, we can infer that crystallization probably took place at inclusion pressures in excess of 1 GPa (Fig. 7).

Fig. 7
figure 7

Exhumation/cooling paths of the investigated inclusions and the Orlica-Śnieżnik granulites. The experimental re-homogenization of the inclusions via piston cylinder experiments (Ferrero et al. 2015) indicates as minimum conditions for entrapment ~875 °C and ~2.7 GPa (yellow dot). The calculated PT path before crystallization of the trapped melt in the inclusions on cooling (blue solid line) lies above the PT path of the host granulite. On crystallization the P–T trajectory of the inclusion is deflected to lower pressures (blue dashed line 1) but remains above that of the host rock. If there is decrepitation (star), the inclusion pressure is reduced to that of the host rock (blue dotted line 2). The solidus curve (in red) is derived by Hermann and Spandler (2008), considering an original H2O content in the melt of 6 wt% as reported by Ferrero et al. (2015) for the original melt. Qz-Coe transition is from Bose and Ganguly (1995), Arg–Cal transition as reported by O’Brien and Ziemann (2008). Kfs + H2O = K-cym and K-cym + Coe + H2O = melt are taken from Mikhno et al. (2013). Qz = Jd + Qz reaction from Holland (1980). Tridymite stability field reported from Darling et al. (1997)

Full size image

Constraints on polymorph formation and preservation

Kumdykolite and kokchetavite

As a polymorph of albite, kumdykolite is expected to show a continuous solid solution with the correspondent polymorph of anorthite, svyatoslawite (Chesnokov et al. 1989). Kumdykolite has not yet been synthesized in the laboratory, but despite being reported in diamond-bearing terrains (Kotková et al. 2014 and references therein), it has been interpreted to form at low pressure in meteorites (Németh et al. 2013), whereas svyatoslavite has been observed to crystallize at atmospheric pressure in coal from burning dumps (Chesnokov et al. 1989). The anorthite/svyatoslawite component in the analyzed crystals is variable but generally low, similar to published compositions (Hwang et al. 2009; Kotková et al. 2014).

The present study reports the first occurrence of kokchetavite in rocks that were not equilibrated in the diamond stability field and in rocks outside the Kokchetav Massif, up to now the only location where this phase has been identified and characterized (Hwang et al. 2004). In the Kokchetav Massif, this phase occurs in polycrystalline inclusions from UHP eclogites, where according to Hwang et al. (2004), it crystallized metastably in the stability field of sanidine from an infiltrated melt. It has also been proposed that kokchetavite may be formed as a product of the dehydration of K-cymrite (Mikhno et al. 2013), a hydrated analogue of kokchetavite and K-feldspar stable at UHP conditions (Zhang et al. 2009). However, the Orlica-Śnieżnik granulites lack of evidence for equilibration at P in excess of 4.0 GPa, i.e., in the diamond and coesite field, as required for K-cymrite formation at T of 900–1100 °C (see Mikhno et al. 2013 and references therein), while the PT path of the MI is far removed from the field in which dehydration of K-cymrite occurs in natural rocks (Fig. 7).

Despite the near-UHP conditions of entrapment, the calculated retrograde PT path of the inclusion suggests that the trapped melt crystallizes at much lower P, likely below the reaction line jadeite + quartz = albite (Fig. 7). This is in agreement with the proposal of Hwang et al. (2009) for kumdykolite formation, whereas kokchetavite likely forms at slightly higher P, as its formation predates that of kumdykolite as previously discussed. However, there is no evidence for a thermodynamic stability field for either of these polymorphs, and their low densities (Hwang et al. 2004, 2009) indicate that they would not be stable phases at high pressures. All the phases in the inclusions crystallize at T ≪ 875 °C, much lower temperatures than those proposed for the formation of kumdykolite in meteorites (≥1000 °C, Németh et al. 2013) as well as for formation of svyatoslawite (~1000 °C, Hwang et al. 2009).

The formation and preservation of kumdykolite, because of its metastable nature and disordered structure, probably requires rapid cooling (Hwang et al. 2009). Similar process can be inferred to be influential in the formation and preservation of kokchetavite, although in this case the sluggish nature of the transition kokchetavite–sanidine may further contribute to its preservation (Hwang et al. 2004). The absence of water has been also proposed as an influential factor in the preservation of both phases (Hwang et al. 2004, 2009) but that cannot be the case here as neither kumdykolite nor kokchetavite transforms into their more stable low-T polymorphs on cooling despite the presence of a water-bearing residual melt (Figs. 2b, 5). Both phases instead “survive” until the melt quenches to glass at T of 300–400 °C—the glass transition was calculated according to Giordano et al. (2008), using the original melt composition modified after crystal fractionation.

Cristobalite

The occurrence of cristobalite in the Orlica-Śnieżnik inclusions is another feature in common with the UHP rocks of the Kokchetav Massif (e.g., Korsakov and Hermann 2006). Cristobalite has low density (2.33 g/cm3) and has a stability field restricted to low pressures significantly <1 GPa (e.g., Holland and Powell 2011). Its presence is clearly not consistent with either the PT path of the host rock or the inclusion (Fig. 7; see also Kryza et al. 1996), which favor instead the formation of quartz. Impurities, e.g., Na, Ca, or Al, may fill the cavities of the tectosilicate framework of cristobalite, thus extending the stability field of cristobalite (Hwang et al. 2004). However, the cristobalite in our inclusions has a very low amount of elements other than Si (Table 1), rather weakening this possible explanation. Darling et al. (1997) proposed that diffusive water loss at high T may lower the internal pressure of the inclusions until it approaches the tridymite stability field (Fig. 7), where cristobalite has been observed to crystallize during experiments (see Darling et al. 1997 and references therein). This mechanism is, however, unlikely to have operated here, since there is no evidence of significant water loss from preserved MI (see discussion in Ferrero et al. 2015), and any such loss cannot drop the inclusion pressure to below the external pressure (as required for the inclusions to approach the stability field of trydimite; see Fig. 7). Cristobalite must have crystallized directly from the melt, locally still present as glass (Fig. 2b), kinetically favored by the similarities between the structure of cristobalite and the melt (Huang and Kieffer 2004; Hwang et al. 2004) and consistent with the presence of cristobalite up to 1.8 GPa in experimental charges (Downs and Palmer 1994).

Calcite and CO2 content of the melt

Calcite occurs in preserved inclusions as a minor crystallization product of the partial melt, thus suggesting that the original melt contained a small, yet significant, amount of CO2 besides the H2O identified in re-homogenized MI (Ferrero et al. 2015). The presence of calcite instead of aragonite is consistent with the end of melt crystallization at P ≪ 1.5 GPa (Fig. 7; see also O’Brien and Ziemann 2008). Calcite has been observed to crystallize from granodioritic melts in the presence of CO2–H2O fluids, when degassing is prevented during crystallization, for example during piston cylinder experiments (Swanson 1979). This similarity is a further confirmation that crack-free nanogranites behaved as closed systems, thus fully preserving their original composition also in terms of volatiles other than H2O.

No CO2 has been identified by Raman investigation in natural inclusions, and all CO2 can be inferred to have reacted with the available Ca2+ to form calcite, thus allowing an estimate of the minimum amount of CO2 present in the original melt. Mass balance calculations suggest that the original CO2 content of the melt was ≥0.60 wt% (or 6000 ppm). As the investigated granulites have a magmatic protolith and no C-bearing crystal phase was involved in melting (Ferrero et al. 2015), the CO2 dissolved in the melt is likely externally derived, i.e., from a CO2-rich fluid fluxing during melting.

Decrepitation effects and significance of kumdykolite, kokchetavite, and cristobalite for inclusion interpretation

Most of the investigated inclusions underwent decrepitation as a result of the combined effect of internal over-pressurization and host rock deformation during the retrograde path. Over-pressurization was caused by the diverging retrograde PT path of inclusions and host rocks (Fig. 7), which produces deviatoric stresses in the host garnet (e.g., Angel et al. 2015) and thus promotes cracking. Only small inclusions are crack free (Fig. 2c; see also Fig. 1 in Ferrero et al. 2015), consistent with what is observed in fluid inclusion studies where, for any given host, small inclusions can withstand higher differential P than large ones (Bodnar 2003). The presence of a differential stress field likely to have further promoted and enhanced inclusions decrepitation is supported by the presence of subparallel crack across the whole host garnet (Fig. 1a, c), intersecting the decrepitated inclusions at opposite corners (see, e.g., Fig. 6a). Also the crack lengths around larger decrepitated inclusions exceed their radius (Fig. 1a, b), while cracks due only to internal over-pressurization are generally inferior to the inclusion radius (Tait 1992; see also discussion in Ferrero et al. 2012).

The occurrence of kokchetavite, kumdykolite, and cristobalite is a distinctive feature of the preserved inclusions, despite the occasional presence of minor cracks. Although these microstructures are generally evidence for inclusion decrepitation (Ferrero et al. 2012; Stöckhert et al. 2009), it appears to have caused only limited depressurization, whereas inclusions with cracks reaching the garnet–matrix boundary contain more common granitic phases, i.e., plagioclase, K-feldspar, and quartz. In the latter case, the cracks probably became pathways for the entrance of fluids present in the rock matrix during the retrograde path, catalyzing the transformation of high-T polymorphs into their stable low-T analogues, as well as for CO2 loss, which likely prevented calcite formation. With the exception of CO2, decrepitated inclusions still have a granitic assemblage (Fig. 6) supporting the interpretation that both inclusion types originally contained the same melt (Ferrero et al. 2015).

Melt in larger inclusions is likely to have started to crystallize earlier than in smaller MI (Holness and Sawyer 2008), and thus when decrepitation occurred, large inclusions were already partially crystallized. After decrepitation, the inclusion is open, via the cracks, to the external environment, so its subsequent PT path coincides with that of the host rock, thus at much lower P than those of the preserved inclusions (Fig. 7). When decrepitation occurs at T near the wet solidus, the pressure drop moves the inclusion internal conditions below the solidus (Fig. 7). This promotes fast crystallization of the residual melt, high in silica and thus likely too viscous to leave the inclusions through the newly formed cracks, and rapid growth of dendritic K-feldspar crystals (Fig. 6b, c, d). Similar microstructures have been observed in rapidly grown plagioclase from experiments where granodioritic melt with 6.5 wt% water was rapidly undercooled below the solidus (ΔT = 150 °C, Swanson 1977).

The comparison between preserved and decrepitated inclusions clearly shows how kokchetavite, kumdykolite, and cristobalite are highly susceptible to transformation into their thermodynamically stable polymorphs as result of decrepitation. The presence of these high-temperature polymorphs should therefore be regarded as a direct mineralogical indicator that the polycrystalline inclusions in which they occur were did not undergo chemical changes after entrapment.

Conclusions

Fluid inclusions in deeply subducted rocks often undergo post-entrapment changes as a result of extreme changes in PT conditions during exhumation (see also Frezzotti and Ferrando 2015), mostly decrepitation with loss of material and/or chemical interactions with the external matrix (Touret 2001); similar behavior was recognized also in melt inclusions from HP and UHP rocks (e.g., Stöckhert et al. 2009). The study of preserved nanogranites from Orlica-Śnieżnik granulites reveals the presence of a unique assemblage including unusual phases such as kokchetavite, kumdykolite, and cristobalite, along with micas, calcite, and often residual glass, crystallized from a preserved anatectic melt trapped at depths of around 100 km (Ferrero et al. 2015). We find that neither kumdykolite, kokchetavite, nor cristobalite is present in decrepitated inclusions: The detection of these metastable polymorphs, therefore, provides a clear indication that the inclusions have not been modified by decrepitation. Thus, the inclusion is a pristine chemical record of the original HP melt, which have been kept completely isolated from the external environment during exhumation. Both the composition and volatile content of the melt, retrieved in situ after experimental re-homogenization (Bartoli et al. 2013; Ferrero et al. 2015), can thus be considered as representative of the original composition of the melt present during crustal subduction to mantle depth. The present study proves how the detailed characterization of microstructural and microchemical features of inclusions and their crystallization products has a direct bearing on understanding the complex nature of the melts that were originally trapped, as well as the evolution of these inclusions.

The phases in preserved inclusions likely crystallized at pressure far removed from the entrapment conditions, most probably in the range 1–2 GPa as constrained by the calculated retrograde PT path of the MI (Fig. 7). These conditions of formation are, however, still in strong contrast with kumdykolite and kokchetavite properties, i.e., density, and very different from the formation conditions reported in previous studies (Chesnokov et al. 1989; Kanzaki et al. 2012; Huang and Kieffer 2004). This suggests that pressure cannot, therefore, be considered critical for the formation of these phases, despite them having being so far mainly identified in crustal rocks equilibrated in the diamond stability field (e.g., Hwang et al. 2004; 2009). Thus, neither kumdykolite nor kokchetavite is actually direct indicators of UHP or HP conditions in the host rock. Such an apparent contradiction can, however, be explained by taking in account that the formation and preservation of these metastable phases most likely requires rapid cooling. At mantle depths, crustal rocks are much more buoyant than the surrounding rocks (cf. models for the Bohemian Massif by Massonne and O’Brien 2003). This condition is further enhanced by the presence of partial melting, also necessary for MI formation (Ferrero et al. 2015). As a consequence of melting, these rocks can no longer be dragged deeper into the subduction zone. They are instead rapidly exhumed along a near-isothermal path and emplaced while still at high temperature in the relatively colder crust, where they are likely to undergo a rapid cooling (e.g., 25–50 °C/Ma, O’Brien and Rötzler 2003). Although this cooling rate is clearly not comparable with those attained via quenching during synthesis experiments of metastable phases (e.g., Kanzaki et al. 2012), it is still very high in terms for metamorphic rocks in collisional settings. Hence, deep subduction of crustal rocks indirectly provides the necessary conditions for formation and preservation of these polymorphs inside “natural capsules” such as the nanogranites, as long as the inclusions remained sealed, i.e., did not decrepitate.