Carbonatite-melilitite-phosphate immiscible melts from the aragonite stability field entrained from the mantle by a Pliocene basalt

A plagioclase megacryst containing composite crystal-glass inclusions was ejected in a Pliocene basaltic diatreme in the Carpathian back-arc basin (Pannonian Basin). The megacryst grew from phonolitic melt, relics of which are preserved in the crystal-glass inclusions. Most of the pristine melt has undergone substantial compositional resetting by interaction with several batches of a low-viscosity carbonated, P-rich melilitite melt, which infiltrated and largely replaced the original inclusion content. The melilitite melt also caused partial resorption of the host megacryst and crystallisation of new calcic plagioclase forming stringers and palisades. A P-rich calcic carbonatite melt exsolved from the melilitite and later crystallised to aragonite at ~ 800 °C and 1.9 GPa. The phosphate melt fraction exsolved from the carbonatite and solidified as CO32−-rich A-B type apatite. At a very late evolutionary stage, K- and Si-rich fluids caused potassic and silicic alteration of the solidified melilitite glass along cracks and interfaces between calcic carbonate globules and glass at temperatures below 680 °C. The oxygen isotope composition of the plagioclase megacryst (6.2 ‰ V-SMOW; Vienna Standard Mean Ocean Water) and the 87Sr/86Sr isotope ratio of carbonates in the inclusions (0.7034) are consistent with a mantle-derived melt. 87Sr/86Sr isotope ratios (0.7047–0.7051) in interstitial carbonates from associated syenite and carbonatite xenoliths indicate a metasomatised mantle source contaminated with radiogenic crustal material or altered marine carbonate. The O-isotope ratios in the carbonates, 22.7 ± 0.6 ‰ V-SMOW in calcite and 23.6 ± 0.7 ‰ V-SMOW in aragonite, are also consistent with a sedimentary precursor. Contrasting δ13C values in the calcite, -12.7 ± 0.5 ‰ V-PDB (Vienna PeeDee Belemnite), and the aragonite (-4.6 ± 0.5 ‰ V-PDB) indicate low-temperature modification of calcite assisted by δ\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}13C-depleted CO2 and preservation of primary magmatic δ\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}13C values in aragonite. The microstructural and geochemical evidence points towards heterogeneous silicate-carbonate melt fractions generated during the metasomatism and partial melting of a supra-subduction mantle wedge.

This communication addresses unusual carbonate-silicatephosphate melt inclusions preserved in an oligoclase megacryst ejected by Pliocene alkali basalt in the northern part of the Pannonian Basin. Feldspar megacrysts in alkali basalts are generally interpreted as xenocrysts that are genetically unrelated to their basaltic hosts. Together with associated clinopyroxene, phlogopite, magnetite, apatite, zircon, and corundum xenocrysts, the feldspars are interpreted as the products of low-pressure crystal fractionation of evolved silicate melts (Irving 1974;Chapman and Powell 1976;Bahat 1979;Upton et al. 1983;Irving and Frey 1984;Aspen et al. 1990;Hinton and Upton 1991;Sutherland et al. 2016). Alternatively, they may represent fragments of syenitic rocks (Aspen et al. 1990;Gernon et al. 2016) crystallised from benmoreite or trachyandesite derivates of fractionating alkali basalt (Chapman and Powell 1976;Guo et al. 1992) or other partial melts generated in metasomatised amphibole-or biotite-bearing upper mantle (Upton et al. 1999(Upton et al. , 2009.
Apart from identifying the melt parental to the plagioclase megacryst studied, we present microstructural evidence for the megacryst interaction with immiscible melilitite-carbonatitephosphate liquids. Based on the occurrence of magmatic aragonite, unusual A-B type apatite, and high phosphorus content in the carbonatite melt, we infer that the plagioclase megacryst originated deeper than other similar megacrysts in alkali basalts. We also provide C-O-Sr isotope data to elucidate the source and origin of the carbonate-bearing melts causing metasomatism in the mantle.

Geological setting
The South-Slovakian Volcanic Field (SSVF) is located within the Alpine-Carpathian-Pannonian (ALCAPA) megaunit (Fig. 1a), which comprises the Carpathian orogenic belt and the northern part of the intra-Carpathian back-arc basin (Pannonian Basin). The ALCAPA is the Adriatic block which was attached to the European plate during the northward Cretaceous-Eocene closure of the Alpine Tethys (Csontos 1995). During the Neogene, the ALCAPA was subjected to an extension related to lateral extrusion from the Alpine collision zone, subduction of the oceanic/suboceanic lithosphere underlying former flysch basins, and the ongoing subduction in the Eastern Carpathians (Nemcok et al. 1998;Konečný et al. 2002). The extension in the Pannonian basin was accommodated by the Carpathian slab roll-back (Csontos 1995;Balázs et al. 2016;Kováč et al. 2018).
Episodes of volcanic activity in the Pannonian Basin occurred during the major extensional phases. The first episode fell into the Upper Eocene-Early Oligocene when volcanism occurred along the mid-Hungarian zone -the continuation of the peri-Adriatic fault system (Kovács et al. 2007;Danišík et al. 2015). The second episode of intermediate to felsic volcanism (20-7 Ma) was linked to the Miocene subduction in the Krosno-Moldavian flysch zone (Konečný et al. 1995a(Konečný et al. , 2002Nemcok et al. 1998). The final episode of Upper Miocene-Pleistocene alkali basalt volcanism was coeval with the diapiric updoming of the asthenosphere and partial melting of the overlying mantle (Konečný et al. 2002;Pécskay et al. 2006). An 8-10 Ma time lag between the onset of subduction and the arc-type volcanism implies an average subduction rate of 1.5-2.5 cm/year (Dérerová et al. 2006), which resulted in tearing and detachment of the subducting slab of oceanic crust along the Carpathian arc.
The SSVF covers an area of about 500 km 2 along the Hungarian-Slovakian border in the northern part of the Pannonian Basin (Fig. 1b). The area features a 26-28 km thick crust and a 70-90 km thick lithospheric mantle (Dérerová et al. 2006;Tašárová et al. 2009). The SSVF consists of numerous volcanoes, which penetrated Late Oligocene-Early Miocene siliciclastic sediments deposited onto the Variscan/Alpine, low-to medium-grade metamorphic basement units (Vass and Elečko 1992). A lowconductive, presumably Cadomian crystalline basement is supposed to be present beneath the Paleogene-Neogene basin in the southern part of the SSVF (Bezák et al. 2020). Fig. 1 a) Schematic sketch of the Carpathian arc and the Pannonian back-arc basin (modified after Pécskay et al. 2006;Kováč 2000): 1 -European platform, 2 -foredeep, 3 -Tertiary accretionary wedge, 4 -Klippen belt, 5 -back-arc basin, 6 -Variscan crystalline basement with pre-Tertiary sedimentary cover and Mesozoic nappe units (Cretaceous accretionary wedge), 7 -Neogene andesite, rhyolite, 8 -Upper Miocene-Pleistocene alkali basalt volcanism, 9 -major normal, strike-slip and thrust faults. BBH denotes the Bakony-Balaton Highland. The rectangle marks the study area. b) Enlarged view of the study area with the position of Hajnáčka diatreme and volcanic phase ages from whole-rock K/Ar, U/Pb zircon, and (U-Th)/He apatite dating 1 3 The SSVF volcanic rocks range in composition from basanite through basaltic trachyandesite to phonotephrite. Their trace element patterns reflect various degrees of partial melting within the garnet-spinel transition zone. Limited olivine and pyroxene fractionation, and modification of the ascending basalts by interaction with the lithospheric mantle, which had previously been affected by subductionrelated metasomatism, were inferred from trace element patterns and radiogenic isotope signatures (Embey-Isztin and Dobosi 1995;Harangi et al. 2015).
The SSVF volcanoes erupted during several successive phases within the time interval from ~ 7 Ma to ~ 200 ka (Konečný and Lexa 2007). The earliest, Late Miocene phase was identified in the NW part of the SSVF (Balogh et al. 1981). The superimposed Pliocene-Quaternary basalts of the 2 nd to 5 th volcanic phases erupted in the SE part (Vass and Kraus 1985;Konečný et al. 1995a). Phreatic and phreatomagmatic eruptions gave rise to maar depressions, tuff rings, scoria-and spatter cones associated with lava flows. Vertical tectonic movements along NW-SE and NE-SW trending faults exposed diatremes -feeder conduits of destroyed maars (Konečný et al. 1999).
Feldspar megacrysts were found in several maars and lava flows of the SSVF (Huraiová et al. 1996(Huraiová et al. , 2018Hurai et al. 2010). The megacryst addressed in this study was found in a Late Pliocene diatreme exposed in the centre of Hajnáčka village (48°13′5.25''N, 19°57′18.59''E) in the SE part of the SSVF. The diatreme, towering about 75 m above the Quaternary river terrace, is a feeder conduit of a basaltic maar that had been removed by erosion. The feldspar megacrysts recovered from the Hajnáčka diatreme [ Figure S1 in the electronic supplementary material (ESM)] are accompanied by kaersutite megacrysts and by xenoliths of carbonatite, carbonated syenite, and syenite (Hurai et al. 2013a;Huraiová et al. 2018). An Upper Pliocene (Romanian) age (2.2-2.3 Ma) of the diatreme was obtained from K/Ar whole-rock dating of an intersecting basalt dyke (Konečný et al. 1995b(Konečný et al. , 1999 and (U-Th-Sm)/He dating of apatite extracted from a tectonically sunk block of stratified lapilli tuffs from maar walls (Hurai et al. 2013b).

Analytical methods
The internal 3D structure of the composite crystal-glass inclusions in the oligoclase megacryst was visualised using X-ray computed micro-tomography using a μCT Nanotom 180 (GE Phoenix). X-ray projections were acquired under the following conditions: accelerating voltage and beam current of X-ray tube 150 kV and 50 μA, respectively, tungsten target, beam focussing mode M0, detector accumulation time 500 ms, number of projections 1800 and geometric magnification 5. The spectral characteristics of the X-ray beam were adjusted at the output of the X-ray tube by a 0.2 mm thick copper plate. The 3D volume was reconstructed from measured projections using the Datos|x-rec 1.5 software with 10 μm voxel size. Volume data were processed and visualised using the VGStudio MAX 2.1 software (Volume Graphics). Image processing included adaptive Gauss filtering (smoothing 1.5, edge threshold 0.1), iso-surface rendering, and image segmentation.
An attempt was made to quantify the carbon content in phosphate. Carbon is difficult to analyze with the conventional PC2 crystal due to very low intensity of the main Kα line. Only about 30 counts per second per nanoAmpére are obtained from the TiC standard, thus resulting in a very poor detection limit (~ 40 wt% C), when soft analytical conditions (15 keV, 10 nA, 10 s acquisition time) are applied, which are necessary to avoid beam damage of the phosphate. The detection limit can be improved to several wt% by applying longer counting times or increased sample currents but then the phosphate becames severely damaged. Although the Kα line intensity is insufficient for the conventional WDS analysis, the carbon content can roughly be evaluated using the method of spectral scanning, allowing the spectrometer to move between the 0.42258 and 0.52258 sin Θ positions which span the region of the C Kα line. Phosphate embedded in silicate glass matrix was selected for analysis to avoid possible influence of C-rich minerals, such as carbonates. Plagioclase and carbonate with known CO 2 compositions, zero and 43.96 wt% CO 2 , respectively, were used as internal standards, which were measured at a distance of no more than 200 μm from the phosphate to minimise the effect of uneven thickness of the carbon coating. Plagioclase, phosphate and carbonate scans were acquired with a 0.00001 sin Θ step size, and 0.4 s dwell time using the PC2 crystal under the same conditions (15 kV accelerating voltage, 15 nA sample current, electron beam defocused to 20 μm). Raw scans [Tables S1-S3 in the electronic supplementary material (ESM)] were smoothed, background-corrected and integrated using Labspec5 software from Horiba Ltd., Japan (Fig. S2 in the ESM). A calibration curve was constructed assuming a linear dependence of the peak areas on CO 2 concentrations in the plagioclase and the carbonate. The calculated carbon content must be regarded as a semiquantitative estimate, because no matrix effects have been taken into account.
Electron back-scatter diffraction (EBSD) analysis was performed using an FEI Quanta 3D Field Emission Gun (FEG) scanning electron microscope. The instrument is equipped with a Schottky-type field emission electron source and an EDAX Digiview4 EBSD camera. We used the software OIM DC v6.2.1 for EBSD data collection, and OIM A v8.0 for data processing. During data collection, the electron beam settings were at 15 kV accelerating voltage and approximately 4 nA probe current using the analytic electron beam mode. The chemo-mechanically polished and carboncoated thin section surface was tilted to 70° (beam incidence angle 20°) at a working distance of 14 mm. A 4 × 4 binning of the EBSD camera was used at 62 ms exposure duration for single frame EBSD pattern collection. Both static and dynamic background corrections and a histogram intensity normalization filter were applied to optimise EBSD pattern contrast. Hough settings for automated band detection were at 140 pixels binned pattern size, 1° theta step size, and 85% rho-fraction, using a convolution mask of 9×9 pixels to index 3-16 reflectors with a minimum peak distance of 11 pixels. Band centre positions were used for EBSD pattern indexing at an interplanar tolerance angle of 2° disregarding the bandwidths. Crystal orientation data were collected from three 50×50 μm-sized scan areas in hexagonal grid mode at a step size of 100 nm. In order to reduce the fraction of misindexed points, the Grain Confidence Index Standardization procedure was employed, where grains are defined as consisting of more than 4 data points at a maximum misorientation angle of 3°. For two scans, the Neighbour Orientation Correlation cleanup procedure was applied, changing the orientation of a point according to the orientation of 5 equally oriented neighbour points with a misorientation tolerance angle of 3°, considering only points with minimum Confidence Index 0.1. This second cleanup step caused data orientation modifications for 0.6-0.9% of the data points. The cleanup procedures increased the number of indexed points with Confidence Index > 0.1 from 93.3-96.1 to 99.6-99.9% of all data points of each scan. The reference crystal structures used for indexing are given in Table S4 in the ESM.
An Xplora Raman spectrometer from Horiba Ltd. (Japan) was used for measurements of vibration modes of melt inclusion phases. The system includes a flat-field 20 cm spectrograph, a 532 nm, 25 mW solid-state Nd-YAG laser, a 638 nm, 24 mW and a 785 nm, 90 mW laser diode, and a colour camera optically coupled to an Olympus BX-51 microscope. Raman spectra were collected with a 100× longworking-distance objective lens (NA = 0.8, WD = 3.4 mm). A set of band-pass and edge filters were used to collect scattered light. Spectra were collected at 20 • C with a Peltiercooled (-70 • C) multichannel CCD detector (1024 × 256 pixels). Wavelength calibration was accomplished with a neon light source and with the vibration band of silicon at 520.7 cm −1 . Using an entrance slit of 100 μm, and holographic gratings with 1200 and 1800 grooves/mm, a spectral resolution of 0.7 and 1.2 cm −1 , respectively, was obtained using the 532 nm laser wavelength. Spectral acquisition, treatment and deconvolution were done using the LabSpec5 software from Horiba Ltd. (Japan).
Raman mapping was conducted using a LabRam HR800 spectrometer from Horiba Ltd. (Japan). The system consists of a flat-field, 80 cm spectrograph, a 532 nm, 300 mW solidstate Nd-YAG laser, and a colour camera optically coupled to an Olympus BX-51 microscope. Mapping parameters are as follows: single-window mode, 600 grooves/mm grating, 2 μm step, 2×3 s acquisition time. Spectral data array treatment was conducted using the DCLS algorithm included in the Labspec5 software from Horiba Ltd. (Japan).
Oxygen isotope ratios were determined in megacryst fragments devoid of impurities and inclusions, 0.05-1 mg in weight, thermally fluorinated with F 2 using a CO 2 -laser. The liberated oxygen was cryogenically purified and analysed for 18 O, 17 O, and 16 O isotopes using a continuous-flow Thermo-Scientific MAT 253 isotope ratio mass spectrometer. Analysed isotope ratios were expressed using the conventional δ-notation as ‰ deviation from the international V-SMOW standard. The accuracy is within ± 0.5 ‰ as determined from replicate measurements of the NBS-28 quartz with the accepted δ 18 O = 9.66±0.05 ‰.
Strontium isotopes were analysed in samples dissolved in 2 ml 0.1 N CH 3 COOH. Element separation followed conventional procedures using AG® 50 W-X8 (200-400 mesh, Bio-Rad) resin and HCl as eluation medium. Sr blanks were < 1 ng. The Sr fractions were loaded as chlorides and evaporated from a rhenium double filament using a ThermoFinnigan® Triton TI TIMS. An 87 Sr/ 86 Sr ratio of 0.710259 ± 0.000002 (n = 5) was determined for the NBS-987 (Sr) international standard during the measurement campaign. Within-run mass fractionation was corrected for 1 3 86 Sr/ 88 Sr = 0.1194. The uncertainties on the Sr isotope ratios are quoted as 2 .

Feldspar megacrysts in the Hajnáčka diatreme: Compositions and structural parameters
The feldspar megacrysts from Hajnáčka fall into three distinct compositional clusters (Fig. 2). The most abundant Group 1 comprises the range from oligoclase to anorthoclase (Ab 47-9 An 2-25 Or 5-46 ) and shows elevated SrO and BaO contents of up to 1.2 and 2.1 wt%, respectively. Group 3 corresponds to Na-sanidine (Or 47-59 Ab 39-47 ) and contains up to 2.0 wt% BaO and up to 0.25 wt% SrO. Group 2 has an intermediate composition  Or 20-23 An 8-9 Cn 1-2 ) plotting within the anorthoclase field between Groups 1 and 3. The BaO and SrO contents of the Group 2 feldspars are only slightly elevated, ranging from 0.5 to 1 wt%, and from 0.1 to 0.3 wt%, respectively. The Group 2 anorthoclase megacrysts contain randomly distributed inclusions of Group 1 plagioclase.
About 50% of the Group 1 megacrysts contain one or two generations of lamellar precipitates ("exsolutions") with shape-preferred orientations. In terms of microstructure, the precipitates do not correspond to characteristic exsolution lamellae, which are typically formed during slow cooling of a chemically closed feldspar grain but rather indicate open system, thus metasomatic, melt-or fluid-induced precipitation. The precipitates form two distinct compositional clusters between Group 1 and Group 3 feldspars, one plotting in the sanidine field and the other overlapping with the Group 2 megacrysts. The lamellar precipitates contain 0.6-3.6 wt% BaO and 0.7-1.3 wt% SrO. The K-rich lamellar precipitates within Group 1 feldspars and the inclusions of Group 1 feldspar in Group 2 feldspar define a general evolutionary trend from early Na-Ca to late Na-K feldspars. The temporal relationship of Group 3 feldspar with the other groups remains unclear. Regardless of their variable composition, all feldspar megacrysts from Hajnáčka are strongly disordered members of the high albite-high sanidine series, with t 1 (o) + t 1 (m) values ranging between 0.55 and 0.6 (Huraiová et al. 2018).
The megacryst HA-1 of the Group 1 is special among all other megacrysts found in the SSVF for the presence of peculiar crystal-glass inclusions. The megacryst is about 3 cm in size, and its composition (Ab 68 An 24 Or 5 Sw 3 Cn 1 on average, Table 1) is the most Ca-rich among the Group 1 feldspars. Structural parameters of the HA-1 feldspar calculated from powder diffraction data are given in Table S5 in the ESM.

Petrography and microstructures (optical microscopy, BSE and EBSD)
The composite crystal-glass inclusions in the HA-1 megacryst are up to 4 mm in size and contain globules of carbonate and phosphate phases embedded in a vesicular silicate glass matrix (Figs. 3 and S3 in the ESM). The majority of the silicate glass appears relatively homogeneous and shows only minute variability in BSE contrast. Dark grey glass domains with diffuse transitions to the dominant light grey matrix have been discerned in two large inclusions.
The globular phases occupy 60-80% of the volume of the inclusions and typically have diameters ranging from several tens to 500 µm. The smaller globules are nearly circular in cross-sections, whereas the larger ones have usually elongated or amoeboid shapes. Based on their internal microstructure in transmitted light and BSE, several types of globules can be discerned: 1. Globules showing fibres arranged in radial, fan-like fashion throughout the globule (optical microscope), otherwise homogeneous in BSE contrast ( Fig. 4a, b). 2. Globules containing rhombohedral crystals (Fig. 4c). 3. Globules with coarse blocky microstructure (Fig. 4d).  4. Globules with dispersed bright elongated crystals (BSE), showing granular microstructure in the optical microscope ( Fig. 4e). 5. Fibrous globules with irregular bright domains in BSE contrast (Fig. 4f). 6. Globules with blocky microstructure affected by metasomatism from the margins (Fig. 4d). 7. Rare small homogeneous spherules, which appear brownish in the optical microscope and have bright grey shades in BSE contrast (Fig. 3).
Crystal orientation analysis of fibrous globules based on EBSD ( Fig. S4 in the ESM) revealed a gradational change of calcite orientation across each fibre and abrupt orientation jumps at the fibre boundaries that are in the opposite sense to the gradational orientation variation so that overall the crystal orientation is within a range of 15° misorientation over the entire fibrous calcite fan. The orientation jumps at the fibre boundaries are less than about 10° so that the interfaces between neighbouring fibres constitute low-angle grain boundaries.
About 20-100 µm thick palisade bands with cauliflower structure comprised of plagioclase, silicate glass, carbonate and phosphate occur along the contacts of the crystal-glass Note also a carbonate bleb bulging out of the crack, which represents the feeding channel of the immiscible silicate and carbonate phases seen in the thin oblique crack inclusions to the oligoclase host ( Fig. 5d, e). The plagioclase grains in the palisade bands show shape-preferred orientation approximately perpendicular to the contact with the host oligoclase. The inner domains of the palisade bands, which are in contact with the silicate glass matrix, are somewhat more coarse-grained than the outer domains, which grew into the oligoclase host. The inner domains feature faceted plagioclase crystals with skeletal morphology and without any apparent shape-preferred orientation. The interfaces between the outer domains of the palisade bands and the plagioclase host display a saw-tooth morphology, which is due to the presence of 2-4 µm wide, close to regularly spaced protrusions of glass into the oligoclase host, where the glass-feldspar interfaces appear to follow specific crystallographic planes of the oligoclase host. The same directions are also observed in the shape-preferred orientation of the plagioclase laths in the palisade bands. On a larger scale, the palisade-like microstructures form lobes bulging into the oligoclase host, suggesting the replacement of the oligoclase host by the palisade band assemblage growing outwards from the original inclusion-host contact.

Raman spectroscopy and imaging
Raman spectra of fibrous globules and idiomorphic carbonates growing from the outer borders of fibrous and blocky globules inwards show bands at 156-158, 282-285, 715 and 1089 cm −1 , which are diagnostic of the common lowpressure calcite I. Compared to the idiomorphic calcite, the Raman spectra of the fibrous calcites show significantly reduced band intensities and extended bandwidths ( Raman mapping of globule with blocky microstructure (Fig. 6) also revealed typical aragonite bands and sector zoning with B 1 g, B 1-2 g , and B 1-3 g modes combined with the 707 and 1087 cm −1 A g components.
Bright domains in fibrous calcite globules (Fig. 4f), phosphates in palisade bands (   Partly cross-polarised, transmitted light image of aragonite replaced by calcite growing from the margin. b) False colour-coded map of the area indicated by the white rectangle in the previous image generated using the direct classical least square fitting of the Raman multispec-tral data array. Blue, green, and red colours denote blocks dominated by B 1 g, B 1-2 g, and B 1-3 g symmetry components, respectively. Violet colours correspond to calcite. c-f) Raman spectra of individual sectors shown in b). Raman band assignment after De La Pierre et al. (2014) occasionally containing an irregularly shaped aragonite core depleted in Mg and enriched in Sr. The P-distribution map documents preferential accumulation of phosphorus in the high-Mg globules and within the palisade band along the inclusion-host interface on the left-hand side. The P distribution map also reveals the presence of phosphate in cracks emanating from the upper part of the inclusion. Element distribution maps of the small crystal-glass inclusion shown in Fig. 3 document high phosphorus concentrations along the globule perimeters and in bright domains within the globules (Fig. 7b-c). The Ca-distribution map (Fig. 7d) shows inward growth of the inner domain of a thick palisade band into the silicate melt inclusion indicated by the deformation of carbonate globules by neoformed Fig. 7 a) Series of X-ray maps of Ca, Mg, P and Sr distribution in crystal-glass inclusion documenting low Mg globules depleted in P in bottom part and high Mg carbonate globules enriched in P in the upper part. White arrows in all maps designate the position of one aragonite globule (invisible in Mg and P maps) and a calcite globule with aragonite core. Note also the increased Sr concentration in aragonite, and the presence of phosphate in cracks and within the palisade band at the left boundary of the inclusion. b) BSE image of the inclusion shown in Fig. 3. c-e) X-ray maps of P, Ca, and K distribution in the same inclusion, showing irregular P-rich patches within calcite globules, thin discontinuous P-rich rims along their perimeters, and an isolated phosphate spherule in silicate matrix marked by white arrow. Note also the generally elevated P contents in the homogeneous parts of the calcite globules. In the calcium distribution map in d), the deformation of calcite globules by dendritic palisade band of calcic plagioclase growing from the original edge of the inclusion marked by the long-dashed curve to the boundary marked by the short-dashed curve can be seen. The K-distribution map highlights the alteration of silicate glass matrix along cracks and along its interfaces to the carbonate globules plagioclase. Melt protrusions into the plagioclase host are discernible on the opposite side of the palisade band.
BSE images revealed 1-2 μm thick continuous rims, which consistently separate the carbonate globules from the silicate glass matrix and also occur along cracks within the silicate glass matrix (e.g. Fig. 4c). Interestingly, the rims do not encompass the resorbed scapolite grain and neoformed plagioclase laths (Fig. 5f). The X-ray distribution map in Fig. 7e documents enrichment of the rims in potassium, which causes their dark appearance in BSE contrast. K-enrichment along cracks in the oligoclase host is also seen in this image. Furthermore, the K-and Ca-distribution maps show significant Ca and K enrichment of neoformed plagioclases in the palisade bands compared to the oligoclase host.

Chemical compositions of glass and minerals
The composition of the silicate glass matrix (Table 2) is characterised by a low SiO 2 (43-47 wt%), a high CaO content (25-30 wt%), and by a positive correlation of Na 2 O and K 2 O with SiO 2 , a weak positive correlation of Al 2 O 3 with SiO 2 , and a negative correlation between CaO and SiO 2 (Fig. 8). The Ca-glass exhibits 21-24% of normative nepheline, up to 18% of normative larnite, and a strong predominance of Na 2 O over K 2 O. Recalculation of the most basic glass following the procedure proposed for classifying alkaline rocks (Le Bas 1973) yields the normative mineral contents of 42 wt% wollastonite, 21 wt% gehlenite, 13 wt% albite, 10 wt% carnegieite, 8 wt% nepheline, and 2 wt% diopside + olivine + thernardite + perovskite + apatite. Owing to the presence of normative gehlenite -the Al-rich member of the melilite solid solution series-and wollastonite, the Ca-silicate glass can be classified as a melilitite according to the classification schemes of Wooley et al. (1996) and Le Maitre et al. (2002).
The comparatively small and subordinate dark silicate glass domains within the Ca-silicate groundmass (Fig. 3) show SiO 2 contents of up to 56 wt% accompanied by up to 11.3 wt% Na 2 O and up to 3.7 wt% K 2 O (Table 2). These domains are also rich in Al 2 O 3 (20-25 wt%) and, at the same time, they show low CaO contents (2.3-3.7 wt%). These compositions fall on the overall mixing trend of the glass matrix with positive correlations between Na 2 O, K 2 O, and Al 2 O 3 and SiO 2 , and a negative correlation between CaO and SiO 2 (Fig. 8). According to the total alkali-silica (TAS) classification scheme (Le Maitre et al. 2002), the Na-rich silicate glass domains can be classified as a metaluminous phonolite to tephri-phonolite.
The composition of the glass in the palisade bands also falls on the overall mixing trend of the glass matrix (Fig. 8).
The carbonate globules can be subdivided into three distinct compositional groups: Mg-calcite devoid of Fe and Mn, Mg-calcite with elevated Fe, and Mg-calcite with elevated Mn (Table S6 and Fig. S7 in the ESM). The fibrous carbonate globules are comprised of Mg-rich calcite with up to ~ 3.8 wt% MgO (9 mol % MgCO 3 ), up to 0.55 wt% P 2 O 5 , up to 1.7 wt% MnO and up to 0.6 wt% FeO.
The aragonite globules show comparatively low concentrations of P and Mg, but are enriched in Sr with SrO concentrations of up to 2.2 wt% ( Table S7 in the ESM). The idiomorphic rhomb-shaped carbonates growing within the globular aragonite and calcite are nearly pure calcite with locally elevated phosphorus and silica contents of up to 1.2 wt% P 2 O 5 and 2.2 wt% SiO 2 ( Table S8 in the ESM). The elevated Si concentrations are most likely analytical artefacts due to subjacent K,Si-rich silicate rims. The randomly increased P concentrations are also most likely contaminants from sub-microscopic P-rich phases concentrated at the interface between the crystals and the surrounding carbonate.
The phosphate phases forming isolated spherules, or domains within calcite globules and cracks in the plagioclase host have the same apatite-like composition (Table 3), albeit with low analytical totals (86-92 wt%) and with an excess of positively charged ions recalculated using the ideal apatite formula, Ca 10 (PO 4 ) 6 (OH,F,Cl) 2 , normalised to 10 A-site cations. The recalculation of the crystallochemical formula by using the total cationic charge of 47 and 24 oxygens yields around 9 cations in the A-site, thus indicating that an anionic group with lower valence substitutes for PO 4 3− in the T-site. Recalculation with 6 T-site cations including C, and the X-site with a total of 1 apfu Cl, F, and HCO 3 − yields analytical totals and CaO to P 2 O 5 proportions consistent with our EPMA data. We therefore assume that all phosphates in our megacryst correspond to hydrocarbonate-carbonate apatite with ~ 1 apfu CO 3 2− in the T-site and the X-site dominated by HCO 3 − , yielding a maximum CO 2 content approaching the theoretical limit of 9.2 wt%. The high carbon content in one phosphate globule was also confirmed by the EPMA scanning of the C Kα line, which returned 7.0 wt% CO 2 (Fig. S2 in the ESM).
The scapolite grain shown in Fig. 5f is characterised by elevated SrO and negligible Cl and SO 3 concentrations ( Table 4). The scapolite is a S-poor meionite (Me 77 ) with 65.5 mol% of equivalent anorthite content, which is close to the hypothetical Na-bearing, Cl-deficient "mizzonite " NaCa 3 (Al 5 Si 7 O 24 )(CO 3 ) with two-thirds of the meionite molecule and 66.6 mol% of the equivalent anorthite content  (Shaw 1960a, b;Papike and Stephenson 1966;Evans et al. 1969). The neoformed plagioclases within the palisade bands, stringers and isolated crystals dispersed in the silicate glass matrix are more basic than the megacryst host (Table 1, Fig. S8 in the ESM). The palisade plagioclases (An 79-90 ) are compositionally similar to those in stringers (An 82-83 ) growing from the Ca-silicate glass matrix. Isolated plagioclase crystals located in phonolitic domains have Ca-rich cores (An 83 ) surrounded by more alkalic rims (An 67 ). The stringers in contact with phonolite glass domains grade from An 84 to even more alkalic rims approaching Ab 45-47 . Some calcic rims of palisade plagioclases exhibit exceptionally high Sr concentrations of up to 5100 ppm equivalent to ~ 14 mol% of slawsonite (SrAl 2 Si 2 O 6 ) component, which is the highest concentration ever recorded in magmatic plagioclase. For comparison, Heier (1962) reported a maximum concentration of 1800 ppm Sr in magmatic An 70-90 plagioclase. Plagioclases with ~ 1000 ppm Sr are known from fractionated shoshonitic intrusions (Eklund et al. 1998) and phonolites (Melluso et al. 1996;Hogarth and Robin 2007).

Strontium, carbon, and oxygen isotopes
The 87 Sr/ 86 Sr isotope ratio of 0.703358 of the carbonate contained in the crystal-glass inclusions (Table 5) in fact refers to a mixture of dominant calcite and subordinate aragonite globules, because the phases could not be separated.  Somewhat higher 87 Sr/ 86 Sr isotope ratios were obtained from calcite (0.704724 and 0.704955) and aragonite (0.705064) occurring in carbonated syenite and carbonatite xenoliths associated with the plagioclase megacryst. The oxygen and carbon isotope compositions of carbonatite and aragonite were reported elsewhere (Hurai et al. 2013a), but they are listed in Table 5 for reference. The plagioclase megacryst has an oxygen isotope composition of 6.2 ‰ relative to V-SMOW, which is quite similar to that of an accompanying kaersutite megacryst (6.4 ‰). The oxygen isotope compositions of two associated sanidine megacrysts are 4.8 and 5.5 ‰ relative to V-SMOW.

Magmatic liquid immiscibility
The shapes and the soft-matter deformation features of the calcite globules in the composite crystal-glass inclusions suggest that they originated from carbonatite liquid rather than representing resorbed calcite crystals. Large carbonate globules indented by smaller ones document higher surface tension of the smaller globules, which is in line with what is expected during the mechanical interaction of droplets in an emulsion. Furthermore, the calcite globules are indented by neoformed calcic plagioclase crystals and deformed around the edges of the resorbed scapolite grain (Fig. 5f). Calcite globules embedded in silicate melt also occur in cracks in the oligoclase host, from where they bulge out into the crystalglass inclusions (Fig. 3).
The existence of calcic carbonatite liquid is further corroborated by the elevated Mg and P contents and comparatively low P-and Mg-contents in the globular aragonite with elevated Sr contents. An identical behaviour was observed in carbonatite-syenite xenoliths from the Hajnáčka diatreme (Hurai et al. 2013a) and anorthite-pyroxene xenoliths in a basaltic lava flow from Čamovce (Reato et al. 2022). Srrich, Mg-poor aragonite crystals have also been identified in the Mg-rich and Sr-poor calcite matrix of garnet peridotite xenoliths ejected in the Udachnaya pipe kimberlites of Russia (Golovin and Sharygin 2019).
By contrast, the globular aragonite with blocky microstructure is reminiscent of resorbed crystalline aggregates, which were partially replaced by Mg-calcite growing from the spherule perimeter inwards (Fig. 4d). The directional metasomatic replacement is diagnostic of crystal-melt rather than melt-melt interaction. Hence, the globular aragonite and the rounded scapolite grain are interpreted as solids that were partially resorbed by the infiltrating carbonatite-melilitite melt. In the light of the low viscosity of melilitite, from 0.1 (Montana and Brealey 1989) to 1.5 Pa.s (Hammouda and Keshav 2015), and the ultra-low viscosity of carbonatite (Kono et al. 2014), such a process appears to be feasible. The crystallisation of aragonite in some calcite globules (Fig. 4e) indicates partial recrystallisation of the carbonate globules within the aragonite stability field.
One might be tempted to interpret the scapolite grain and the aragonite globules as resorbed crystals and aggregates accidentally trapped during plagioclase growth. This interpretation does, however, not apply to the aragonite globules, because in contrast to scapolite; the aragonite globules occur in about half of the observed crystal-glass inclusions and occupy significant volumes of some inclusions. For instance, a total of 24 aragonite globules occupying about ~ 30 vol% can be discerned in false colourcoded BSE image of the inclusion shown in Figs. 3 and 9, and additional two small calcite globules with dispersed aragonite crystals occur within the phonolite domain. Some aragonite spherules are almost completely replaced by rhombohedral calcite from the margins. Some globules show recrystallisation from the margins combined with the directional replacement by fibrous calcite along curvilinear boundaries. Moreover, the spherical shapes of the Table 4 Representative EPMA data on scapolite bdl below detection limit * stoichiometry based on 24 O atoms and 12 T-site cations per formula unit ** derived from mineral formula as C = 1-S-Cl and back-calculated to CO 3 wt% ¤ anorthite equivalent calculated as 100(Al + Fe-3)/3 after Evans et al. (1969), marialite and meionite components calculated as 100(Na/ [Na + Ca + K]) and 100(Ca/[Na + Ca + K]), respectively (Shaw 1960a, b)  1 3 aragonite globules do not match with the elongated morphology of aragonite crystals and twins in recrystallised globules (Fig. 4e), carbonatite-syenite xenoliths (Hurai et al. 2013a), and anorthite-pyroxene xenoliths (Reato et al. 2022). Last but not least, some aragonite globules show signs of deformation when occurring between two larger globules (Fig. 9). Hence, crystallisation from a carbonatite melt at conditions within the aragonite stability field is the preferred explanation for the frequent occurrence and microstructural features of the aragonite globules. The blocky internal microstructure is ascribed to slow crystallisation. In contrast, the fibrous calcite globules indicate rapid quenching within the calcite stability field. This hypothesis implies multiple rather than one-shot injection of the carbonatite melts into the megacryst interior, which is also corroborated by the fact that Mg-calcite globules form three distinct compositional clusters (Fig. S7 in the ESM). The inclusion documented in Figs. 3 and 9 indicates at least two stages of the melt infiltration recorded in P-rich Mg-calcite globules with phosphate domains concentrated in the bottom part, coexisting with (gravitationally separated?) aragonite globules in the upper part, and the latest portion of homogeneous, P-poor, Mg,Mn-calcic carbonatite melt bulging out from the feeding crack and replacing the aragonite globules. Spherical shapes (Fig. 3), degraded Raman bands (Fig. S6 in the ESM), occurrence in cracks together with carbonatite and silicate glass (Fig. 7a), and deformations by neoformed plagioclase in palisade bands (Fig. 5e) indicate that the phosphate globules are solidified liquids rather than resorbed mineral grains. The compositionally identical phosphate domains in the globular calcite and segregated phosphate globules in the melilitic glass suggest separation of a Ca-phosphate liquid from P-rich carbonatite similar to those described from metasomatised mantle xenoliths (Rosenbaum et al. 1997;Zhang et al. 2007), alkaline rocks and carbonatites (Naumov et al. 1988;Panina and Motorina 2008), impact structures (Ernstson et al. 2002), an alvikite dyke (Sommerauer and Katz-Lehnert 1985), limburgite (Shao et al. 2000), and highly fractionated peraluminous granitic pegmatite (Thomas et al. 1999). Phosphate-carbonate-silicate immiscibility was also reported from the Mushugai Khuduk alkaline igneous complex of Mongolia, where cogenetic silicate-salt and phosphatecarbonate melt inclusions occur in apatite from theralite associated with carbonatites (Andreeva et al. 1999).

P-T constraints from plagioclase-melt equilibria
The palisade bands indicate a lack of equilibrium between the megacryst host and the melt in the inclusion. They formed by reaction between silicate melt and megacryst host which can schematically be expressed as where Pl 1 and Pl 2 represent the Na-rich oligoclase host and the Ca-rich plagioclase of the palisades, respectively. Melt 1 corresponds to the Ca-silicate glass matrix, and melt 2 to the glass fraction of the palisade bands, which is more sodic than the Ca-silicate glass matrix.
The plagioclase megacryst must have crystallised close to equilibrium with a silicate melt that was substantially more sodic than the Ca-silicate glass matrix which now is the (1) Pl 1 + melt 1 = Pl 2 + melt 2 dominant glass in the inclusions. Thus, the neoformed calcic plagioclase grew from a more calcic melt corresponding to the Ca-rich silicate glass of the crystal-glass inclusions. The dendritic and skeletal morphologies of the neoformed calcic plagioclase indicate substantial supersaturation of the parental melt with respect to calcic plagioclase. This situation implies undercooling, where the term undercooling sensu largo does not necessarily imply cooling below the liquidus but merely refers to a situation, where the system is driven away from plagioclase-melt equilibrium towards supersaturation of the melt with respect to Ca-rich plagioclase. At the same time, the Na-rich plagioclase host becomes metastable and is consumed by disequilibrium incongruent melting, which produces the new Ca-rich plagioclase within the palisade bands. This effect may be triggered by overheating (e.g., Anderson and Eklund 1994;Hammouda et al. 1996;Hammouda and Pichavant 2000) or decompression (e.g., Arebäck and Stigh 1997), which both destabilise the sodic plagioclase host at the contact with the inclusion melt, thus leading to the consumption of the host plagioclase accompanied by the formation of more calcic plagioclase in the residual melt.
However, overheating or decompression cannot explain the relict sodic domains within the calcic silicate groundmass. The most likely scenario explaining the observed phase content and microstructures is the supplement of batches of external hot, Ca-silicate melt mingled and mixed with the earlier sodic phonolite melt, which had been in equilibrium with the plagioclase megacryst. The Na-rich silicate domains in the Ca-rich silicate glass matrix probably represent relics of such an early melt, which must have been cooler and substantially more viscous than the incoming batches of carbonatite-melilitite melt. During mingling and mixing, heat transfer between the two melt fractions resulted in the undercooling of the carbonate-silicate melt and enhanced the segregation of carbonate and phosphate liquids accompanied by the rapid skeletal growth of new calcic plagioclase. Thus, the phonolite melt domains probably represent the original melt parental to the plagioclase host, while the neoformed calcic plagioclase grew from the hotter Ca-rich carbonatite-melilitite melt interacting with the oligocalse megacryst host.
Whether the observed melt and feldspar compositions confer to equilibrium phase relations can be decided based on the K D(Ab-An) parameter defined as (Putirka 2005(Putirka , 2008: where Al, Ca, Si, and Na refer to the respective cation fractions in the silicate liquid (L), while Ab and An denote the albite and anorthite mole fractions in the feldspar. In the case of feldspar-melt equilibrium, the K D value falls within (2) K D = Ab × Al L × Ca L An × Si L × Na L narrow bounds, 0.05-0.15 for temperatures below 1050 °C, and 0.16-0.38 for temperatures above 1050 °C. The K D values of 0.22-0.35 calculated from the composition of the plagioclase megacryst and the Na-rich phonolite glass indicate possible equilibrium at temperatures above 1050 °C. Temperatures and pressures calculated from EPMA data in Tables 1 and 2 and the model of Putirka (2005Putirka ( , 2008 are 1009 ± 26 °C and 3.0 ± 0.2 GPa. By contrast, a K D value of ~ 4.5 calculated from the compositions of the oligoclase megacryst and the Ca-rich melilitite glass is far outside the feasible range of equilibrium K D values confirming the disequilibrium between the two pahses in accordance with the evidence provided by microstructural observations. The Ca-rich plagioclases from the palisade bands and the Ca-rich cores of the stringers seem to have been in equilibrium with the Ca-rich melilititic melt at temperatures above 1050 °C (K D = 0.31-0.32). The Na-rich plagioclase overgrowths from the stringers that are in contact with phonolitic domains return K D values between 0.08 and 0.20. In spite of this, the plagioclase-melt equilibrium model did not converge at any P-T conditions and reasonable water contents and this is particularly intriguing for the Na-plagioclase and phonolite pairs, which provide the most favourable K D values.

P-T constraints from plagioclase megacryst compositions and structure
The compositions of all feldspar megacrysts from the Hajnáčka diatreme plot between the 700° and 800 °C solvus curves calculated for 0.1 MPa and 1.5 GPa (Fig. 10). The average compositions of the two main groups of feldspar megacrysts, the K-rich Group 3 and the Na-Ca-rich Group 1, do not appear to correspond to a conode at any combination of P-T conditions indicating that they never coexisted in equilibrium. Hence, we infer that the two feldspar megacryst groups crystallised from compositionally different parental melts. The 700 °C solvus at 0.1 MPa is the lowermost limit of the crystallisation temperature for the plagioclase megacryst with crystal-glass inclusions, as the solvus moves towards the albite apex with increasing pressures. Therefore, we assume that the suite of feldspar megacrysts from the Hajnačka diatreme probably crystallised at 700-800 °C from successively more K-, Ba-, and Sr-enriched silicate melts. This view is also supported by the metasomatic K-rich lamellar precipitates in the Group 1 megacrysts, which record interaction with K-rich melts or fluids.
The feldspar crystallisation temperatures can also be estimated from the structural properties of feldspars calculated from unit cell dimensions. The Δ2Θ(131) parameter of 1.78 inferred from X-ray diffraction data (Table S5 in the ESM) projects to a temperature of ~ 700 °C at 0.1 MPa pressure, or ~ 770 °C at 1.8 GPa (Kroll 2020), in very good agreement with the inference made from the distribution of feldspar compositions between the 700° and 800 °C solvus curves.

P-T constraints from phosphate composition and phosphorus solubility in carbonatite melts
X-ray maps show a rather heterogeneous distribution of phosphate in the glass-crystal inclusions studied. For example, the carbonate-phosphate fraction in the inclusion documented in Fig. 7b-f contains 11.3 vol% phosphate. Combined with an average content of 0.4 wt% P 2 O 5 in the carbonate and 34-36 wt% P 2 O 5 in the phosphate, the total P 2 O 5 content in the carbonate-phosphate fraction corresponds to 4.3-4.6 wt% ( Table S9 in the ESM). In contrast, another neighbouring inclusion illustrated in Fig. 4a-c contains only one small phosphate droplet, 10 μm in diameter, accompanied by fibrous calcite globules devoid of phosphate-rich domains, but still containing 0.34-0.51 wt% P 2 O 5 . Hence, the total phosphate contents in individual inclusions fluctuate by one order of magnitude and indicate the infiltration of the plagioclase megacryst by several batches of carbonatite-silicate melts with different P 2 O 5 contents. In homogeneous calcite Fig. 10 Feldspar solvus curves for 600, 700, 800 °C and 0.1 MPa (dashed curves) or 1.5 GPa (solid curves) calculated from the ternary feldspar mixing model (Benisek et al. 2010) and the SolvCalc software (Wen and Nekvasil 1994). Circles are projection points of compositions of feldspar megacryst from the Hajnáčka diatreme globules that are devoid of phosphate domains, the P 2 O 5 concentrations are 0.2-0.5 wt%, which is one order of magnitude lower than in experimental carbonatite melts saturated with apatite at > 1050 °C and > 2 GPa (Hammouda et al. 2010;Kiseeva et al. 2012). Hammouda et al. (2010) showed that apatite solubility in carbonatite melts increases with increasing Mg content, because the supressed Ca activity enables more P to be dissolved in the melt before apatite saturation is reached. A linear correlation between natural logarithm of P 2 O 5 mole fraction in the melt and 1000/T enables extrapolation of experimental data obtained at 1200-1380 °C to lower temperatures relevant to our crystal-glass inclusions (Fig. 11). The negatively correlated CaO and P 2 O 5 contents in calcite globules (r 2 = 0.67) agree with the trend typical of experiments conducted at > 2 GPa. The overal temperature range derived from P 2 O 5 contents in homogeneous calcite globules spans the interval from 700 °C to 800 °C. We assume the upper temperature limit to be close to the P 2 O 5 saturation temperature, because the carbonate globules with phosphaterich domains also contain about 0.6 wt% P 2 O 5 .
The low fluorine content in the phosphates from the oligoclase megacryst studied is incompatible with experimental (Baker and Wyllie 1992;Klemme and Dalpé 2003) and natural carbonatites typically containing Cl-poor fluorapatite or F-rich hydroxylapatite (Chakhmouradian et al. 2017). The aragonite-bearing carbonatite-syenite xenolith HA-8 from the Hajnáčka diatreme also contains F-rich hydroxylapatite (Table 3). In contrast, the Ca-phosphates in the oligoclase megacryst correspond to CO 3 -rich fluorapatite with up to 6.8 wt% CO 2 crystallised from experimental carbonatite melts (Hammouda et al. 2010). Such high CO 2 contents can only be accomodated in the A-B type apatite, Ca 10-x (PO 4 ) 6-x (CO 3 ) x (OH) 2-2y (CO 3 ) y (x≈y → 1), hitherto unknown from natural rocks but synthesised in high CO 2 -pressure experiments (Fleet and Liu 2007). In A-B type carbonate apatite, the CO 3 2− group substitutes for PO 4 3− in the tetrahedral site and concomitantly HCO 3 − /CO 3 2− groups substitute for OH − , halogens and other possible substituents (O 2− , H 2 O, vacancies) in the c-axis channel of the apatite structure (Pan and Fleet 2002).
Interestingly, the CaO and P 2 O 5 contents in the interstitial carbonates from the carbonatite-syenite xenolith HA-8 from the Hajnáčka diatreme are opposite to the trend observed in the associated oligoclase megacryst (Fig. 11), despite the comparable Mg-and P-enrichment in both. This enigmatic behaviour is inconsistent with high-pressure experiments and might be related to formation at different P-T conditions and changes in the carbonatite melt structure between 1-2 GPa (Dobson et al. 1996).

Reappraisal of P-T indicators
The inferred P-T conditions are summarised in Fig. 12. The formation conditions of the feldspar megacryst are constrained by the iso-Δ2Θ (131) line, which intersects the boundary separating the aragonite and calcite stability fields at 1.87 GPa and 775 °C. There is, however, no unequivocal evidence for crystallisation of the oligoclase megacryst within the aragonite stability field. The temperatures and pressures around 1000 °C and 3 GPa calculated from the plagioclase-melt equilibrium model are unrealistic due to the significant off-set from the iso-Δ2Θ(131) line.
Temperatures around 800 °C are constrained by the Psaturation limit of calcic carbonatite with 51-51.5 wt% CaO dissolving ~ 0.6 wt% P 2 O 5 . Intersection of the 800 °C isotherm with the calcite-aragonite boundary yields a pressure of 1.9 GPa, which corresponds to a depth of ~ 63 km assuming lithostatic load with an average rock density of 3 g.cm −3 . This pressure must be regarded as a minimum estimate and refers to the incipient solidification of the carbonatite melt fraction exsolved from the carbonated melilitite.
The upper phosphorus saturation limit of interstitial calcic carbonate with euhedral aragonite crystals in the carbonatitesyenite xenolith HA-8 corresponds to ~ 720 °C, which returns the minimum pressure of 1.65 GPa equivalent to a 57 km depth. It should be noted that the interstitial carbonate is not saturated with respect to apatite as indicated by resorbed apatite crystals and by the absence of phosphate exsolution microstructures (Hurai et al. 2021). Hence, the actual crystallisation temperatures and pressures could be higher. At any rate, the estimated minimum P-T conditions are rather contradictory  Hammouda et al. (2010) and projection points of Mg-calcite globules in plagioclase megacryst HA-1 and carbonatitesyenite xenolith HA-8 from the Hajnáčka diatreme with the low-pressure assemblage of accessory minerals, involving Zr-rich titanite, baryte, REE-zirconolite, diopsidehypersthene, baddeleyite, zircon and ilmenite replaced by rutile.
The puzzle of low-pressure syenite xenoliths containing euhedral magmatic aragonite in a calcic carbonatite matrix was tentatively explained as a record of the aragonite crystallisation in steady-state conditions under supralithostatic advective pressure caused by expanding gas bubbles in a quasi-incompressible silicate melt. In contrast, the fibrous carbonate matrix was thought to record rapid quenching of the xenoliths (and megacrysts) within the calcite stability field concomitant with the instantaneous pressure release during transport to the surface (Hurai et al. 2013a).

Isotopic and compositional constraints on the origin of calcic carbonatite
Almost pure calcic carbonatite magmas occur in rift-related and intra-oceanic tectonic settings (e.g., Katz and Keller 1981;Keller 1981;Cooper and Reid 1991;Bailey 1993;Hoernle et al. 2002;Woolley and Church 2005;Keller and Zaitsev 2006;Mourão et al. 2010;Bailey and Kearns 2012). Besides, quenched calcic carbonate melts have been recorded in Pleistocene extrusive and intrusive rocks closely associated with melilitites (Stoppa and Principe 1998;Stoppa et al. 2005Stoppa et al. , 2019 and mantle xenoliths ejected by phreatomagmatic eruptions (Rosatelli et al. 2007). Calcic carbonatite globules with low contents of alkali elements frequently occur in metasomatised mantle xenoliths (e.g., Pyle and Haggerty 1994;Kogarko et al. 1995;Seifert and Thomas 1995;Yaxley et al. 1998;Demény et al. 2004Demény et al. , 2010. The high Ca/(Ca + Mg) values of > 0.9 of the carbonates enclosed in the plagioclase megacryst from Hajnáčka and the other occurrences are apparently controversial with experiments, according to which the fractionation or partial melting in the Earth's mantle can only produce dolomitic carbonatite liquids with Ca/(Ca + Mg) < 0.8, alkalic carbonatites, or carbonatites with an elevated aluminosilicate content (Koster van Groos 1975;Baker and Wyllie 1990;Hamilton and Kjarsgaard 1993;Willey 1997, 1998;Dasgupta et al. 2004Dasgupta et al. , 2006Brooker and Kjarsgaard 2011;Kiseeva et al. 2012Kiseeva et al. , 2013Poli 2015Poli , 2016Weidendorfer et al. 2020). However, carbonatites produced by high-pressure melting of carbonated eclogite exhibit Ca/(Ca + Mg) ~ 0.8, in striking contrast with those produced by the melting of carbonated peridotite (Hammouda 2003). Moreover, interaction of pristine Mg-rich carbonatite with mantle pyroxenes may produce almost pure calcic carbonatite according to the reaction: where the carbonates represent components in the melt (Hammouda et al. 2014). By this mechanism, an originally Mg-rich carbonatite melt may acquire a more calcic composition of up to Ca/(Ca + Mg) = 0.95. It is thus reasonable to assume that the calcic carbonatite is produced in the course of mantle metasomatism and may be trapped as veinlets or inclusions in newly formed minerals, which may finally be brought to the surface as xenoliths and megacrysts in ascending basalts (Frezzotti and Touret 2014 Kogarko et al. 1995], and the Kerguelen Archipelago [Ca/(Ca + Mg) up to 0.96; Moine et al. 2004]. Furthermore, quenched calcic carbonate melts in mantle xenoliths and calcic carbonatite intrusive complexes in oceanic hotspots (e.g., Demény et al. 1998;Hoernle et al. 2002;Mourão et al. 2010;De Ignacio et al. 2012;Schmidt and Weidendorfer 2018) also document transport of mantle carbonatites to shallow depths during adiabatic decompression of hot sub-oceanic mantle.
During slow, warm subduction, sedimentary carbonates dragged down in the subduction channel completely dissolve at sub-arc depths by the interaction with aqueous fluids liberated from OH-bearing silicates (Molina and Poli 2000;Yaxley and Brey 2004;Gorman et al. 2006;Manning et al. 2013). In contrast, as much as 80% of the sedimentary (3) 3 MgCO 3 + CaMgSi 2 O 6 → CaCO 3 + 2 MgSiO 4 + 2 CO 2 Fig. 12 P-T isoline constrained by the lower stability limit of disordered oligoclase with Δ2Θ(131) = 1.78 (Table 2, Kroll 2020), aragonite-calcite stability boundaries in the experimental CaO-H 2 O-CO 2 system (Wyllie and Boettcher 1969, short-dashed curve) and calculated from the thermodynamic data (Holland and Powell 1998, long-dashed curve). Thin dotted lines separate the stability fields of α-quartz, β-quartz, and coesite (Wenk and Bulakh 2016). The mantlecrust boundary of the northern Pannonian lithosphere is derived from geophysical data in Dérerová et al. (2006) carbonates may bypass sub-arc depths during rapid, cold subduction (Kerrick and Connolly 2001), giving rise to carbonate liquids and eventually high-pressure orogenic carbonatites (Ravna et al. 2017), which originate from volatile fluxing of fused limestones and dolomites. Volatile fluxing of sedimentary carbonates may also occur around mafic and intermediate magma chambers in the crust (Lentz 1999;Wenzel et al. 2002;Whitley et al. 2019).
Ocelli and globules of fibrous calcites similar to those described from the Hajnáčka diatreme have also been found in metasomatised mantle xenoliths (peridotites) ejected in basalts of the Bakony-Balaton Highland (BBH) in the central part of the Pannonian Basin (Demény et al. 2004). Trace element contents of the BBH carbonates (Demény et al. 2010) and similar fibrous carbonates embedded in phosphate-rich intergranular glass in mantle xenoliths from the French Massif Central (Rosenbaum et al. 1997) are akin to marine carbonates and phosphates. The peridotite-hosted globular carbonates from the BBH with 87 Sr/ 86 Sr ratio of 0.70533, high δ 18 O (+ 22 ‰), and low δ 13 C (down to -15 ‰) are broadly similar to the fibrous calcite of carbonated syenite and carbonatite xenoliths from the Hajnáčka diatreme (Table 5, Fig. 13).
The 87 Sr/ 86 Sr isotope ratio of 0.7034 in the carbonates encapsulated in composite crystal-melt inclusions in the oligoclase megacryst from Hajnáčka is similar to the range of 0.7031-0.7033 in various minerals of pyroxenites and hornblendites ejected in the Pinciná maar in the NE part of SSVF (Huraiová et al. 2005), and interpreted as mafic cumulates from alkali basalt. Slightly higher 87 Sr/ 86 Sr isotope ratios of 0.7047-0.7051 in calcites and aragonites from syenite and carbonatite xenoliths from the Hajnáčka diatreme are also within the range of 87 Sr/ 86 Sr isotope ratios in silicates, and interstitial glass (0.7035-0.7061) from syenite xenoliths from the Pinciná maar, which are interpreted as flotation cumulates derived from the fractionating alkali basalt. The mantle-like Sr-isotope signatures of carbonates from the Hajnáčka diatreme indicate either a mantle carbonatite Fig. 13 Carbon and oxygen isotope composition of aragonite crystallised from devolatilised calcic carbonatite melt calculated using the general Rayleigh fractionation equation describing continuous removal of extract from source (Hurai et al. 2015). Shaded area highlights the stability of aragonite at 2 GPa and T < 810 °C derived from thermodynamic data (Holland and Powell 1998). Shown are also fields of C-O isotope compositions of Phanerozoic marine limestones (Veizer et al. 1999), metamorphosed subduction marbles (Tao et al. 2018), extrusive calcic carbonatites (Stoppa et al. 2021), intergranu-lar fibrous calcites from metasomatised peridotite xenoliths in the Bakony-Balaton Highland (BBH) basalts (Demény et al. 2004(Demény et al. , 2010, aragonite (white circles) and calcite (black circles) from carbonatite and carbonatite-syenite xenoliths in the Hajnáčka diatreme (Hurai et al. 2013a, Table 5). Isotope fractionation equations used for the calculation of the Rayleigh degassing are as follows: CO 2 -calcite and aragonite-calcite (Chacko and Deines 2008) for oxygen, CO 2 -calcite (Scheele and Hoefs 1992) and aragonite-calcite (Golyshev et al. 1981) for carbon parent that was slightly contaminated by radiogenic crustal material or a marine carbonate modified by fluids from juvenile magma or a carbonatite derived from the dissolution (volatile fluxing) of older, less radiogenic minerals (Shields and Veizer 2002).
The stable carbon and oxygen isotope ratios of carbonates from carbonatite and carbonatite-syenite xenoliths from the Hajnáčka diatreme overlap those in Sr-poor calcites in metasomatised mantle peridotites of the BBH (Fig. 13). On the other hand, the Hajnáčka aragonites are isotopically similar to Sr-rich carbonate veins in the metasomatised peridotites of the BBH, showing mantle-like C-isotope (around -5 ‰) and sediment-like O-isotope (around 21 ‰) signatures. The magmatic δ 13 C signature of the BBH carbonates combined with high Sr contents (up to 3.4 wt% SrO) resemble mantle carbonatites, whereas the high δ 18 O value and low Ce/Pb and Nd/Pb ratios point towards a sedimentary precursor (Demény et al. 2004(Demény et al. , 2010. The excursions from typical Phanerozoic marine carbonates (δ 13 C = -2 to 2 ‰, δ 18 O = 29-35 ‰, e.g., Veizer et al. 1999) have been explained by massive CO 2 devolatilization accompanied by the infiltration of isotopically light upper crustal fluids (Demény et al. 2004(Demény et al. , 2010. The ambiguous C-, O-, and Sr-isotope records from Hajnáčka are rather difficult to interpret using canonical isotope fractionation models. An extensive sub-solidus deuteric or hydrothermal alteration by a pervasive fluid is unlikely, because the aluminosilicate glass associated with carbonates encapsulated in the plagioclase megacryst is fresh and chemically modified only along cracks. The euhedral shapes of aragonite crystals growing from the calcite matrix of carbonate-syenite xenoliths (Hurai et al. 2013a) and the almost identical 87 Sr/ 86 Sr isotope ratios of the coexisting calcite and aragonite indicate equilibrium crystallisation from the same batch of carbonatesilicate melt. In contrast, the only moderate enrichment in 18 O ( Δ arg-cal = 0.9 ± 0.8 ‰) accompanied by the enormous enrichment in 13 C ( Δ arg-cal = 8.2 ± 0.5 ‰) in the aragonite indicates isotopic disequilibrium or a low-temperature isotope fractionation assisted by hydrothermal fluids. While the observed small positive enrichment factor ( Δ arg-cal ) for 18 O is acceptable for magmatic temperatures (Chacko and Deines 2008), the one for 13 C is out of all limits provided by the degassing and fractionation models (Giuliani et al. 2014).
The apparent contradiction between magmatic C-and sedimentary O-isotope signatures in aragonites from the Hajnáčka diatreme can partly be explained in terms of the devolatilization of carbonatite melt during the remelting of hydrothermally modified marine carbonate with δ 13 C and δ 18 O values of -2.7 and + 27.7 ‰, respectively (Fig. 13). According to this model, the carbonatite melt lost 60-65 mol% CO 2 during devolatilization at 500-810 °C and 2 GPa, i.e. within the stability field of aragonite. The proposed simple devolatilization model can further be modified by simultaneous gravitational separation of aragonite crystals and pressure variations. The separation of aragonite cumulate has little effect on the composition of the residual carbonatite at reasonable fractionation rates corresponding to less than 50% of cumulate removed from the melt (Hurai et al. 2015).
The contrasting δ 13 C values of aragonite and fibrous calcite matrix cannot be explained by fractionation or devolatilization at magmatic temperatures (> 500 °C), or by the alteration by magmatic fluids at subsolidus temperatures down to 300 °C. Therefore, one must accept the modification of the stable isotope composition of fibrous calcite by a low-temperature hydrothermal fluid and preservation of the pristine magmatic isotope signature in the associated aragonite crystals.

Sources of melilitite and phonolite
Transient, highly electrically conductive carbonated melilitite and melilitite-carbonatite melts formed by low-degree partial melting of mantle peridotite are probably the most abundant melts in the asthenosphere (e.g., Wallace and Green 1988;Presnall and Gudfinnsson 2005;Sifré et al. 2014). The silicate liquid produced by the partial melting of peridotite evolves from kimberlite through melilitite to basalt with decreasing pressure and increasing temperature, CO 2 and H 2 O contents (Wyllie and Huang 1976;Brey and Green 1976;Moussalam et al. 2015Moussalam et al. , 2016Hammouda and Keshav 2015). The source region of melilitite magma is located between 80 and 120 km (Mitchell 2005), corresponding to lithostatic pressures of 2.7-4.0 GPa. Phonolite melt may be generated by the carbonatite melt-assisted metasomatic alteration of lherzolite to wehrlite (Loges et al. 2019).
Melts derived from subducted oceanic crust have lower silica contents than those generated by the partial melting of peridotite. Carbonated metapelites produce peraluminous K-rich rhyolite at 2.5-3.7 GPa or slightly peralkaline phonolite at 5 GPa at the fluid-absent solidus (Thomsen and Schmidt 2008). At 3.5-5 GPa, carbonated eclogites yield K-rich silicate melts at the lowest degree of melting, transient carbonate-silicate immiscible melts at intermediate, and silicate melts at the highest degrees of melting (Kiseeva et al. 2012). Thus, silicate melts (melilitite at low, melilite nephelinite at higher melt fractions) generated from carbonated eclogite may be credible sources of the melilitites and melilitic nephelinites occurring in oceanic provinces. Conjugate carbonatites are, however, always dolomitic, with Ca/(Ca + Mg) = 0.55-0.6 at 3 GPa (Dasgupta et al. 2006). Continental melilitites are compositionally similar, showing trace element and radiogenic isotope signatures diagnostic of the HIMU (high μ, = high 238 U/ 204 Pb) mantle component with recycled crustal material. Melilitites from central Italy associated with alkalic carbonatites are correlated with a specific metasomatised mantle enriched in incompatible elements (Isakova et al. 2017).
The early phonolite and late melilitite melt fractions preserved in the crystal-glass inclusions in the plagioclase megacryst from Hajnáčka are reminiscent of phonolite melts produced during the carbonatite-assisted metasomatic alteration of the mantle overprinted by deep-seated transient carbonatite-melilitite melts derived from low-degree partial melting of carbonated eclogite. All silicate and carbonatite melt constituents preserved in the plagioclase megacryst are depleted in Fe, Ti, and Mg, indicating either the fractionation and separation of primary magmatic oxides (Ti-rich magnetite, hematite, ilmenite-geikielite) and silicates (olivine, orthopyroxene), or the extraction of Fe, Mg and Ti by neoformed olivine and Fe-Ti oxides (ilmenite) crystallising during the interaction of dolomitic carbonatite with mantle pyroxenes as it has been documented for anorthite-diopside xenoliths in alkali basalts from the Čamovce quarry, 6 km northeast from the Hajnáčka diatreme (Reato et al. 2022).

Conclusions
The microstructural relationships within crystal-glass inclusions in an oligoclase megacryst from the Hajnáčka diatreme in the northern part of the Pannonian Basin reflect a multistage evolution, starting with megacryst growth in a phonolitic melt at temperatures higher than 700 °C. The megacryst was later infiltrated by several batches of a lowviscosity carbonate-melilitite ± phosphate immiscible melt, which largely replaced the original inclusion content and caused partial resorption of scapolite accidentally trapped during the megacryst growth. The infiltration of the carbonated melilitite, partial crystallisation of the immiscible carbonatite liquid fraction, and the interaction with the early phonolite took place at 800 °C and 1.9 GPa within the aragonite stability field. The infiltrating carbonated melilitite was out of equilibrium with the plagioclase megacryst, and reaction between the two gave rise to crystallisation of calcic plagioclases forming palisade bands, stringers and isolated crystals. During this stage, both the melilitite and the carbonate melt fractions were still liquid. At a later stage, the influx of an external K-and Si-rich fluid along cracks resulted in the metasomatic alteration of already solidified melilitic glass at temperatures below 680 °C.
Isotope data on carbonates reveal the mixing of melts derived from the mantle and from subducted oceanic crust (carbonated eclogite). The silicate melt fraction of the crystal-glass inclusions documents the mingling of phonolite derived from metasomatised (carbonated) mantle with transient carbonatite-melilitite-phosphate melts, which probably had been generated by the partial melting of carbonated eclogite intercalated with marine carbonates. The low Mg content of the carbonate melt fraction suggests that the original, supposedly dolomitic carbonatite, was depleted in Mg by the interaction with mantle clinopyroxene and concomitant crystallisation of secondary olivine. Considering the close genetic relationship with the metasomatised mantle and the anatectic melting of carbonated eclogite, the carbonatite melts preserved in the oligoclase megacryst should not be considered true carbonatites. It is possible that the P-rich carbonated melilitite was generated by volatile fluxing of limestone dragged down in the subduction channel.
Our observations indicate that bringing calcic carbonate melts from the mantle to the surface is possible in an intracontinental setting. Apart from the presence of magmatic aragonite, several lines of indirect evidence (type A-B apatite with CO 3 2− substituting for PO 4 3− and HCO 3 − for OH − , negatively correlated Ca and P contents in the carbonatite melt) document that plagioclase megacrysts and syenite xenoliths in alkali basalts may contain relics of high-pressure melts from the deep mantle.