Petrogenesis of alkaline basalt-hosted sapphire megacrysts. Petrological and geochemical investigations of in situ sapphire occurrences from the Siebengebirge Volcanic Field, Germany

Megacrystic sapphires are frequently associated with alkaline basalts, most notably in Asia and Australia, although basalt is not generally normative in corundum. Most of these sapphire occurrences are located in alluvial or eluvial deposits, making it difficult to study the enigmatic relationship between the sapphires and their host rocks. Here, we present detailed petrological and geochemical investigations of in situ megacrystic sapphires within alkaline basalts from the Cenozoic Siebengebirge Volcanic Field (SVF) in Germany. Markedly, the sapphires show several micrometer thick spinel coronas at the contact with the host basalt, indicating chemical disequilibrium between the sapphire and the basaltic melt, supporting a xenogenetic relationship. However, in situ U–Pb dating of a Columbite Group inclusion within one Siebengebirge sapphire using laser ablation–inductively coupled plasma–mass spectrometry (LA–ICP–MS) indicates a close genetic relationship between sapphire crystallization and alkaline mafic volcanism in the SVF. The syngenetic mineral inclusion suite including carbonates, members of the Pyrochlore, Betafite and Columbite Groupe minerals, as well as a high abundance of HFSE and of gaseous low-density CO2 inclusions support a parentage of a highly evolved, MgO and FeO deficient carbonatitic melt. We identified CO2 to be the link between alkaline basaltic volcanism and the xenocrystic sapphires. Only alkaline volcanic suites can build up enough CO2 in this magma chamber upon fractionation so that at high degrees of fractionation a carbonatitic melt exsolves which in turn can crystallize sapphires.


Introduction
Sapphire and ruby are gem varieties of corundum (α-Al 2 O 3 ) and belong to the most prominent and precious gemstones, along with diamond and emerald. Deposits of gem-quality corundum are well distributed around the world (e.g., Hughes 1997). A review of the literature on gem corundum shows that corundum occurs in various geological settings and seems to be generated via several different petrogenetic processes, such as metamorphic reactions or through igneous crystallization from aluminarich magmas (see references in this work and references therein). As to the broad variability in the depositional type of corundum, Simonet et al. (2008) established a classification scheme according to their geological occurrence, distinguishing between primary and secondary corundum deposits. Primary deposits are those where corundum is entrained inside the same rock that it crystallized from. In secondary deposits, corundum is an inherited mineral that 1 3 43 Page 2 of 27 formed in a different petrogenetic setting as to where it is deposited.
Of special interest are secondary corundum deposits that are associated with rift-related alkaline basaltic volcanism as these are among the world's largest gem corundum suppliers (e.g., Guo et al. 1996;Limtrakun et al. 2001;Shore and Weldon 2009) and are hence economically of great importance. Such basaltic secondary deposits are located all over the world, most notably in Asia (e.g., Guo et al. 1996, and in eastern Australia, at the western Pacific margin (e.g., Coenraads 1990;Graham et al. 2008;Sutherland 1996;Sutherland et al. 1998aSutherland et al. , 2002Sutherland et al. , 2009bSutherland et al. , 2015Sutherland et al. , 2017Saminpanya 2001;McGee 2005;Abduriyim et al. 2012), but also in Europe (e.g., Malikova 1999;Giuliani et al. 2009;Uher et al. 2012), Africa (Kiefert and Schmetzer 1987;Krzemnicki et al. 1996;Schwarz et al. 2000;Pardieux et al. 2014), in South America , and in North America (e.g., Garland 2002;Berg 2007Berg , 2014Palke et al. 2017). In most basalt fields, corundum is present as blue, green, or yellow sapphires, and is commonly referred to as blue-green-yellow (BGY) sapphire suite (e.g., Sutherland et al. 1998a, b). These BGY sapphires may reach up to 5 cm in diameter and usually occur in alluvial or eluvial placer deposits. In situ occurrences where the sapphires are still inside their host basalts are rarely found (e.g., MacNevin 1972;Stephenson 1976;Vichit 1987). The best known example is a basalt flow at Changle, China (Guo et al. 1992;Zelong et al. 2007;Giuliani et al. 2015), while numerous corundum, including gem crystals, were recorded in extensive flows south of Jos Plateau, Nigeria (Wright 1985).
Although this type of corundum deposit is economically the most valuable amongst corundum deposits, the origin of these corundum is to date still poorly understood (e.g., Izokh et al. 2010;Uher et al. 2012). The problem is that primitive alkali basalts are not corundum-normative when they leave their mantle sources, regardless of the pressure of formation (Liu and Presnall 1990). Hence, unless these sapphires precipitated via some yet unknown process out of the basaltic melts, they must instead have a xenogenetic relation with their host rocks. Indeed, anhedral crystal shapes, rounded edges, and macroscopically visible spinel coronas at the sapphire-basalt interface (e.g., Stephenson 1976;Guo et al. 1996;Sutherland et al. 2002;McGee 2005) indicate disequilibrium between these sapphires and their host rocks, supporting a xenocrystic nature. There seems to be a consensus (see references in this work) that the sapphires originate from a reservoir in the mantle or the crust, and were later transported to the surface by ascending basaltic melts. Yet, the origin of the basalt-associated sapphires remains a puzzle. Moreover, it is still an open question as to why sapphires most often occur as xenocrystals in alkaline basalts from intra-continental volcanic fields, but only rarely in SiO 2 saturated rocks (e.g., Berger and Berg 2006;Berg 2007Berg , 2014Palke et al. 2017). The petrogenesis of alkaline basalt-hosted sapphires was intensively studied from various gem fields in the 1990s and early 2000s. These studies revealed that the sapphires that were brought to the surface by basaltic eruptions may have various pre-depositional histories and may be of metamorphic, magmatic, or of metasomatic origin (e.g., Sutherland et al. 2009a;Giuliani et al. 2014), depending on the δ 18 O values (e.g., Giuliani et al. 2005Giuliani et al. , 2007Yui et al. 2006;Sutherland et al. 2009a, b), the mineral inclusion suite (e.g., Guo et al. 1996), their trace element geochemistry (Sutherland et al. 2009b(Sutherland et al. , 2017, and most notably on the Ga/Mg ratio (Peucat et al. 2007).
Several hypotheses have been proposed to explain the igneous crystallization of sapphires, the most prominent being: (1) Crystallization from highly evolved melts such as syenites (e.g., Irving 1986;Coenraads et al. 1990;Aspen et al. 1990;Giuliani et al. 2009). (2) Crystallization from alkaline Si-and Al-rich melts formed by partial melting of amphibole-bearing lithospheric mantle (Sutherland 1996;Sutherland et al. 1998b). (3) Growth as restitic component during partial melting of Al-rich crustal protoliths such as anorthosites (Palke et al. 2016(Palke et al. , 2017. (4) Hybrid formation where sapphire is stabilized by reaction of a fractionated silicate with a carbonatitic melt at mid-crustal levels (Guo et al. 1996). This hybrid model is based on the occurrence of a very complex mineral inclusion suite in sapphire that must have precipitated from melts of two different compositions, most probably a carbonatitic and a highly evolved silicate melt. Guo et al. (1996) proposed that evolved granite/syenitic pegmatite silicate melts become Al 2 O 3 oversaturated in response to mixing with carbonatitic melts, resulting in local precipitation of corundum in the hybridization zone. (5) Crystallization at shallow crustal levels from an iron-rich syenitic melt with the participation of a carbonate-H 2 O-CO 2 fluid phase (Izokh et al. 2010). This model is similar to the model proposed by Guo et al. (1996) but does not include the participation of a carbonatitic melt. Izokh et al. (2010) rather suggested that the syenitic melt must have had initially high CO 2 contents, as it was formed by differentiation of a CO 2 -rich basaltic melt. Differentiation of these CO 2 rich syenitic magmas led to the formation of alkalicarbonate complexes, so that the residual melt becomes oversaturated in alumina and corundum may precipitate in the most fractionated magma chambers.
In this paper, we present a detailed microchemical investigation of up to 2 cm sized sapphire xenocrysts that occur within primitive alkali basalts and basanites from the Siebengebirge Volcanic Field (SVF) in Germany. The sapphires from the SVF are found in situ within alkali basalt, a relatively rare occurrence with the potential for providing important insights into how this type of sapphire formed. To date, there are only limited investigations of sapphire that are still attached to their host rocks using modern analytical techniques. Hence, the Siebengebirge sapphires provide an exceptional opportunity to study the genetical relationship of the sapphires and their host basalt. We present a unique petrogenetic model for the origin of the Siebengebirge sapphires that may provide important implications for the general origin of basalt-hosted sapphires and for predicting sites of possible further deposits.

Geological setting
The Siebengebirge Volcanic Field (SVF) is part of the Central European Volcanic Province (CEVP). The latter forms a 200-300 km broad Cenozoic intra-continental volcanic belt north of the Alpine Orogen. In Germany, volcanic fields belonging to the CEVP are (from west to east) the Eifel, the Siebengebirge, the Westerwald, the Vogelsberg, the Hessian Depression, the Rhön area, Heldburg, and the Oberpfalz (Fig. 1). Further European regions belonging to the CEVP are the Bohemian Massif in the Czech Republic, the Eger Rift System in the Czech Republic and Poland, the Pannonian Basin in Hungary, and the French Massif Central. Igneous activity in the CEVP lasted throughout the entire Cenozoic (Wilson and Downes 1991).
The SVF is located at the SE margin of the Lower Rhine Graben and is part of the Rhine rift system that formed in Eocene to Oligocene times as a consequence of syn-to post-orogenic extension in the Alpine foreland (Ziegler 1992). The NW-SE-trending Lower Rhine Graben cuts into the Rhenish Shield that is a Variscan tectonic block within the Rhenohercynian Zone and has been uplifted since Eocene times (e.g., Illies et al. 1979;Ziegler 1992;Oncken et al. 1999).
The total volcanic area of the SVF covers an area of about 900 km 2 . The volcanic activity in the Siebengebirge occurred over several eruptional phases, resulting in the eruption of a very broad spectrum of both SiO 2 saturated  pohl et al. 1994). The main sampling localities Petersberg, Oelberg, and Jungfernberg are shown in the boxbox at the bottom of the figure and SiO 2 undersaturated mafic and felsic melts. The most primitive mafic volcanics are basanites to alkali basalts; more felsic volcanics are SiO 2 saturated latites to trachytes and SiO 2 undersaturated phonotephrites to tephriphonolites (Vieten et al. 1988;Kolb et al. 2012;Jung et al. 2012). The period of volcanic activity in the SVF has been dated using the K/Ar and Ar/Ar chronometers and lasted from 30 to 19 Ma (Todt and Lippolt 1980;Linthout et al. 2009;Przybyla 2013). The main magmatic activity is thought to have occurred between 26 and 24 Ma (Przybyla 2013).
Alkaline basalts from the SVF have long been known to host megacrystic sapphires. They were described scientifically in the late nineteenth and early twentieth century (Brauns 1922, and references therein). However, until now, no effort has been made to investigate the sapphires entrained in the alkaline mafic rocks from the SVF with modern techniques and to identify possible genetic links to their host basalts.

Samples
Most of the sapphires studied in this work belong to the mineralogical collection of the Geological and Mineralogical Department of the University of Bonn and were recovered in the early twentieth century when the basalt occurrences (now protected) were still subject to quarrying. Other samples were donated by local collectors. The principal localities in the Siebengebirge from which sapphires were recovered are the Petersberg and Oelberg alkaline mafic rocks. Other known localities are the Jungfernberg in Heisterbacher Rott at the northern part of the Siebengebirge, and Unkel at the south-western extension of the Siebengebirge.
Samples studied included 11 sapphires with an anhedral to subhedral crystal shape and sizes of 5-20 mm. Sample descriptions, localities, and host rock lithologies are summarized in Table 1. Colors vary from a milky pale-blue to a clear dark blue, and two sapphires have large brown to gray domains (Fig. 2), distinguishing them macroscopically from the other samples. All sapphires were still attached to their host rocks and showed evidence for chemical disequilibrium with their host basalt, including resorbed edges, rounded shapes, and macroscopically visible dark green spinel coronas (Fig. 3a, b). The contact between sapphire and spinel is quite sharp, but in their contact with the basaltic matrix the spinel coronas become more ragged and porous, and tend to disintegrate to discrete grains (Fig. 3a). The basalt in direct contact with the spinel coronas is enriched in plagioclase and depleted in modal pyroxene and titanomagnetite (Fig. 3b,   The basaltic matrix in direct contact with the spinel rim comprises almost pure plagioclase. Fe-Mg phases such as pyroxene are missing. The plagioclase-rich basaltic matrix is followed by the 'normal' basaltic matrix with abundant plagioclase, pyroxene, and Ti-phases samples (Fig. 3d), the spinel coronas are virtually free of inclusions.

EPMA
The sapphires, their reaction rims, and mineral inclusions were analyzed for major and minor elements with a JEOL JXA 8900 electron probe microanalyzer equipped with a tungsten cathode at the Steinmann Institute, University of Bonn. All measurements were performed at 15 kV acceleration voltage and a probe current of 15 nA.

XRF
The sapphire crystals were still inside their host rocks, offering the opportunity to classify the basalts that carried them to the surface. After removal of visible alteration rims, around 100 g of basaltic material was crushed and ground in an agate mill. Major elements were analyzed from fused lithium fluoride-based glass disks using a PANalytical Axios wavelength dispersive XRF.

LA-ICP-MS
Trace elements in the sapphires were measured using a Resonetics Resolution M50E 193 nm laser ablation system coupled to a Thermo Scientific X Series II quadrupole inductively coupled plasma mass spectrometer (Q-ICP-MS). The laser was operated at a repetition rate of 15 Hz and a fluence of 7 J/cm 2 . Spot sizes were 100 µm in diameter, and 20-30 spots were quantified depending on grain size and homogeneity of the sapphire. Count rates were normalized using 27 Al as the internal standard, assuming an average of 98.5 wt% Al 2 O 3 as based on the EPMA analyses. The NIST-SRM-610 and NIST-SRM-612 standard reference materials were used for calibration as external standards using the preferred reference values given by Jochum et al. (2011). Isotopes monitored were 9 Be, 23 Na, 24  One Columbite Group mineral inclusion was dated on the basis of the U-Pb decay system, using in situ LA-ICP-MS analysis. This inclusion was analyzed for U, Th, and Pb isotopes by LA-ICP-MS at the Institute of Geosciences, Johann Wolfgang Goethe University Frankfurt, using a Thermo Scientific Element II sector field ICP-MS coupled to a New Wave Research UP-213 ultraviolet laser system. Data were acquired in the analog mode for 238 U and in the counting mode for all other isotopes with 20 s measurement on the background followed by sample ablation of 20 s. Prior to analysis, the sample surface was cleaned with 3 pulses of pre-ablation. Square laser spots with an edge length of 28 µm were used. The laser was operated at a repetition rate of 6 Hz and an energy fluence of 2 J/cm 2 with an energy output of 50%. The signal was tuned to obtain maximum sensitivity for 238 U and 206 Pb, and to keep the oxide production monitored as 254 UO/ 238 U below 3%. Measured masses were 202 Hg,204 Hg,204 Pb,206 Pb,207 Pb,208 Pb, 232 Th, 235 U, and 238 U. Mercury was also measured because 204 Hg can be a minor contaminant of the Ar carrier gas, potentially causing isobaric interferences on 204 Pb.
The primary standard used was the GJ-1 zircon, reported by Jackson et al. (2004) to have a TIMS age of 608 ± 0.4 Ma. Isotopic ratios were corrected for ablationinduced elemental fractionation offset between zircon matrix and columbite offset between zircon matrix and columbite matrix using a columbite-tantalite solid solution (Coltan 139) from Madagascar that is being used as an in-house standard for coltan LA-ICP-MS analyses by the Federal Institute for Geosciences and Natural Resources, Germany (Gäbler et al. 2011). The usability of Coltan 139 as a matrix matching standard for Columbite Group Mineral LA-ICP-MS U-Pb dating was recently demonstrated by Che et al. (2015). Coltan 007 from Kokobin, Ghana (Melcher et al. 2008(Melcher et al. , 2015 was used as a secondary standard to control the accuracy and reproducibility of the measurements. For Coltan 139, Melcher et al. (2015) reported TIMS ages of 505.4 ± 1.0 Ma (BGR, Hannover) and 506.6 ± 2.4 Ma (University of Toronto), and an LA-ICP-MS age of 506.2 ± 5.0 Ma (University of Frankfurt). Coltan 007 has an age of 2079.6 ± 3.1 Ma (Melcher et al. 2008(Melcher et al. , 2015. A total number of 45 unknowns were analyzed, using the standard bracketing method. Acquired data were then processed using an in-house MS Excel© spreadsheet Zeh 2006, 2009). A 208 Pb-based common lead correction was applied, as there is very low Th in the sample.

Raman spectroscopy
Raman spectroscopy was used to further characterize all types of inclusions in the sapphire, and special attention was paid to those that are not exposed on the polished surface, including solid inclusions as well as fluid inclusions and their daughter minerals. As fluid inclusions were not recognizable in polished sapphire mounts, for the Raman analyses, polished thick sections with a thickness of ~500 µm were prepared, so that the sapphires were translucent. Unfortunately, some sapphires were too brittle for thick sectioning. Because of this, we cannot say for certain that all the sapphires depicted here have inclusions, or whether the inclusions shown here are representative of all SVF sapphires.
Raman spectroscopy was also employed to distinguish between calcium carbonate polymorphs in carbonate inclusions. Raman spectra were acquired with a confocal Horiba HR800 equipped with an Olympus BX41 optical microscope operated in 180° backscatter geometry. The excitation source was an Nd-YAG Laser with a wave length of 532.18 nm. The laser beam was focused on the target with a ×50 objective lense. Spectra were accumulated with an exposure time of 30 s on each point within a frequency range of 100-4000 cm −1 . The laser power was 200 mW.

TEM
One sapphire (sample PE25) was investigated with the transmission electron microscope (TEM) at the Geo-ForschungsZentrum (GFZ) in Potsdam, Germany. Next to the typical blue colored areas, this sapphire has a gray to brown colored domain ( Fig. 2) that is characterized by a significant enrichment in the HFSE up to two orders of magnitude higher than the HFSE content in the blue domains (up to 5000 ppm Nb in the brownish domain). Given that some sapphire domains contain up to 5000 ppm HFSE, it would be interesting to determine whether these elements substitute into the corundum crystal lattice, or whether they are present as nano-inclusions. Therefore, a combination of the focused ion beam (FIB) and the TEM techniques was applied to retrieve information about the crystallographic orientation relations of the sapphire and any potential nano-inclusions at an atomic scale, revealing the physicochemical state of the HFSE in the sapphires. For this purpose, three electron-transparent foils (15 × 10 × 0.15 µm)-two from the brownish areas, and one from the blue area-were cut, using the focused ion beam (FIB) technique Ishitani et al. 1990) where thin membranes of an area of interest are cut from a polished sample by sputtering with Ga + ions accelerated to 30 keV. The FIB sections from the brownish area were cut into two different crystallographic orientations, one near parallel to [0001] and another one inclined to [0001] of the sapphire. The section of the blue area was cut near parallel to [0001]. Subsequently, the FIB sections were thinned with a Ga ion beam to a final thickness of 25-30 nm to guarantee electron transparency. A detailed description of the principles of the FIB technique is given by Wirth (2004Wirth ( , 2009). The FIB sections were subsequently analyzed with an FEI Tecnai F20 X-Twin TEM with a Schottky field emitter as electron source. The instrument is equipped with a Fishione high-angle annular darkfield (HAADF) detector allowing for Z-contrast sensitive imaging, an EDAX energy dispersive X-ray spectroscopic system (EDS) for the identification of nano-scale chemical compositions from the excited characteristic X-rays of the sample, as well as a Gatan Tridiem energy filter and an electron energy loss spectrometer (EELS) for spectrum imaging.

Host rock composition and corundum mineral chemistry
Host rock compositions and their CIPW norms are presented in Table 2. All rocks are nepheline normative, though to various extends, ranging from 0.3 to 10 wt% normative nepheline. In the Total Alkali vs. Silica (TAS) diagram (Le Maitre 2002), most host rocks plot in the field for alkali basalts. One sapphire was entrained in a basanite, and three samples lie in the trachybasalt (hawaiite) field.
The Siebengebirge sapphires include up to 2 wt% FeO and up to 0.8 wt% TiO 2 . The major and trace element compositions of the sapphires, as well as the major element composition of the spinel coronas, are listed in Table 3.
Next to Fe and Ti as the main impurities, all sapphires contain detectable amounts of Mg, Mn, Ga, V, Nb, W, and Th. Beryllium, Na, K, Cr, Zn, Zr, Ta, and Hf were detected in few analyses, whereas Sr, Y, and U are usually below detection limit. Remarkably, the highest concentrations of the HFSE (up to 5000 ppm Nb and up to 2900 ppm Ta), as well as of Th (up to 226 ppm) were detected in the brown to gray domains in samples PE21 and PE25 where the concentrations are up to two orders of magnitude higher than in the blue domains of the same sapphire. The other elements are not concentrated at any preferred site within the sapphire. The Ga contents range from 48 to 256 ppm.
In terms of their Ga/Mg ratio and the total Fe concentration, the Siebengebirge sapphires can be divided into two groups: (1) Sapphires with low Ga/Mg (<10) and high Fe tot and (2) sapphires with high Ga/Mg (>10) ratios and lower Fe tot .

Inclusions in sapphire
Inclusions have high potential for providing important information on the origin of sapphires. If the inclusions are primary, i.e., crystallized during growth of the sapphires, then they may represent melts, fluids, or accessory minerals that coexisted with the sapphires during crystal growth. Secondary inclusions may provide information about a potential post-magmatic alteration of the sapphires. Inclusions are interpreted to be primary, if they are completely enclosed within the sapphire and are not located at any visible cracks. An inclusion has to be regarded to be of secondary origin if it is xenomorphic and restricted to cracks in the sapphire.
The Siebengebirge sapphires contain a broad spectrum of mineral inclusions, including silicates, carbonates, and a wide variety of Nb, Ta and Th oxides. While some sapphires are virtually inclusion free, others have a high abundance and compositional variability among the mineral inclusions. Notably, there is no correlation between the mineral inclusion suite and the minor and trace element compositions of the host sapphires. Interestingly, many inclusions in the sapphires are carbonates, which have only been described previously in sapphires associated with alkaline basaltic volcanism from the Bo Phloi gemfield in Kanchanaburi, western Thailand (Khamloet et al. 2014). Furthermore, the U-bearing inclusions thorite, thorianite and Pyrochlore and Columbite Group minerals are of great interest, as they potentially allow to date the age of the sapphire crystallization using the U-Pb chronometer. With  such age information, the sapphires may be temporally related with to their host basalts, as the age of the basaltic volcanism in the SVF is well known (Todt and Lippolt 1980;Linthout et al. 2009;Przybyla 2013).

Silicates
Silicate mineral inclusions are potassic feldspar and labradoritic plagioclase. In one sample (OL25), feldspar was found associated with calcite, either as discrete mineral grains attached to calcite or as a thin rim around the carbonate phase. These associations are interpreted to be primary. In sample 26, the feldspars are located in vesicles along cracks within sapphire and associated with clay minerals and zeolites; hence, these are considered to be secondary in origin. This population of silicate mineral inclusions may not be representative for all investigated sapphires, as only few polished surfaces were studied. Yet, it is striking that apparently only feldspar has been incorporated as a crystalline silicate mineral, but not pyroxene or olivine, which are ubiquitous liquidus phases in the host basalt samples.

Carbonates
Carbonate inclusions are observed in samples 3, 15, 26, OL25, and S52. Compositions range from calcium carbonate to magnesian carbonate. Vibrational spectroscopic studies of the carbonate inclusions in sample 26 using Raman spectroscopy revealed that these carbonates are either dolomite, calcite, or aragonite. Calcite and dolomite inclusions can be regarded as primary as these are completely included in the host sapphire. By contrast, aragonite tends to form vesicular textures within secondary potassic feldspar in the sapphire. Consequently, aragonite is considered a secondary low-temperature alteration phase. Figure 4 shows selected carbonate inclusions inside the Siebengebirge sapphires.

Oxides
Samples 3, 15, 26, and OL25 have various oxide inclusions. All oxide inclusions are completely isolated inside the sapphire, and are not located at any visible cracks (Fig. 5). Hence, they may be regarded as primary inclusions that formed cogenetically with the host sapphires. Figure 5a shows a Columbite Group (Mn)(Fe,Nb) 2 O 6 inclusion that in turn includes a complex aggregate of pyrochlore Ca 2 Nb 2 O 7 , thorite ThO 2 , and thorianite ThSiO 4 (Fig. 5b) Figure 5c shows a complex oxide inclusion where ilmenite-pyrophanite ss showing with Nb-rich rutile and hercynite exsolutions is rimmed by spinel at the contact to corundum, indicating a reaction of the ilmenite-pyrophanite ss with the host corundum at magmatic temperatures accounting for a primary origin. In addition, one Nb-rutile-Betafite Group (Ca,U) 2 (Ti,Nb,Ta) 2 O 6 (OH) composite aggregate was found in sample OL25. The Columbite Group inclusion from sample 26 (Fig. 5a) was used for in situ U-Pb dating using LA-ICP-MS.

Melt inclusions
Sample 26 contains composite silicate glasses of two different compositions (Fig. 6), interpreted to represent silicate melts that were trapped by the growing sapphires. One composition (melt 1) forms vesicular structures within a second glass referred to as melt 2. Melt 1 is quite complex and heterogeneous in composition, and appears to represent a carbonated silicate melt, while melt 2 is close to an aluminous phonolite. The average compositions of the silicate glasses are compiled in

Fluid inclusions
Fluid inclusions were found in samples 15, 26, S52, and KS21. They occur either randomly distributed within the whole sapphire without any preferred crystallographic orientation, or arranged along trails (Fig. 7a). Sizes of the fluid inclusions are highly variable and range from <1 µm to several tens of micrometers. Many fluid inclusions show negative crystal shapes (Fig. 7b) following the crystal symmetry of the host sapphire, indicating them to be of primary origin. Many fluid inclusions seem to be pure CO 2 with two strong vibrational bands at 1285 and at 1388 cm −1 (Rosso and Bodnar 1995). Some fluid inclusions contain daughter phases (Fig. 7c, d) whose compositions are hard to characterize with Raman spectroscopy due to the overlap between the vibrational spectra of multiple phases at low wavenumbers as well as with the spectrum of the host sapphire. One strong band at 992.4 cm −1 repeatedly occurs in the vibrational spectrum of many fluid inclusions. This band is characteristic for minerals such as natroalunite NaAl 3 (SO 4 ) 2 (OH 6 ), thenardite Na 2 (SO 4 ), aphthitalite K 3 Na(SO 4 ) 2 , and blödite Na 2 Mg(SO 4 ) 2 * 4H 2 O (Fig. 8). Further identified daughter minerals are burkeite Na 4 (SO 4 )(CO 3 ), paravauxite Fe 2+ Al 2 (PO 4 ) 2 (OH) 2 *8H 2 O, and dawsonite NaAl(CO 3 ) (OH) 2 . Although an unambiguous identification of the daughter minerals is not possible from the vibrational spectrum, it can be stated with some confidence that the fluid trapped in the inclusions was a complex sulfatephosphate-H 2 O bearing CO 2 fluid with some dissolved cations, notably alumina and alkalis. Similar melt inclusions daughter phases including sulfates and carbonates were reported by Song and Hu (2009) from the Changle sapphires.

Nano-inclusions
Selected TEM images of the brown (Nb-Ta-enriched) and blue sapphire domains of sample PE25 are shown in Fig. 9. The brown sapphire domains exhibit a very high population density of nano-inclusions that are probably responsible for the brown coloration. The inclusions form ~10 to 20 nm long rods with a shape-preferred orientation having the long axes of the needles near parallel to corundum [0001], though the scatter of the orientation of the long axes ( Fig. 9 a, b; TEM bright-field and dark-field images) demonstrates that the nano-inclusions do not show a strict orientation relation relative to the crystal lattice of the host sapphire and thus must be discrete phases. The crystallinity of the nano-inclusions is verified by electron diffraction and dark-field imaging using a reflection from the inclusions ( Fig. 9b; TEM darkfield image), lattice fringes from the inclusions and by the Moirée interference patterns visible in the high-resolution image (HRTEM) in Fig. 9c. The latter are produced by the superposition of the crystal lattices of the nano-inclusions and that of the host sapphire. The observed diffraction contrast around the inclusions is characteristic of the so-called Guinier-Preston Zones which are nano-crystalline platelets in a host material. The compositions of the inclusions were qualitatively estimated by energy dispersive methods. All analyzed inclusions are high in Nb, Ta, Ti, and Fe. Given that Nb-rich rutile (Nb, Ti, Fe)O 2 is a common macroscopic mineral inclusion in the Siebengebirge sapphires, we consider it most likely that Ta-bearing Nb-rich rutile is the phase forming the majority of the nano-inclusions. The nano-inclusions may be either solid-state exsolution products or they may be primary crystals that grew epitaxially along with the sapphire. The lack of a preferred crystallographic orientation relation between the nanoinclusions and the host sapphire lattice supports the latter possibility, as it has also been argued for platinum group mineral nano-inclusions (nPGM) in pentlandite from the Merensky Reef platinum group element ore deposit of the Bushveld Complex in South Africa by Wirth et al. (2013) and Junge et al. (2014). An additional argument opposing the nano-inclusions to be solid-state exsolutions is provided by their large-scale distribution within the sapphire. Figure 9d (TEM bright-field image) shows that the population density of inclusions varies periodically with inclusion-rich zones intermittently alternating with inclusion-poor zones. Strikingly, the boundaries between the zones with different population densities are very sharp and almost perfectly parallel to each other. The orientation of the lamellar zone/ boundaries is near parallel to [0001] of the corundum lattice and hence may be parallel to the (11-20) hexagonal prism of corundum. The distribution of the nano-inclusions within the sapphire differs significantly from the distribution of larger rutile inclusions that have a needlelike shape and show a strong crystallographic orientation relationship with the host sapphire (Fig. 9e, f).

Geochronology
The Columbite Group mineral dated in this study is a 200-500 µm sized crystal with the formula Fe 0.2 Mn 0.5 Nb 2.3 O 6 , suggesting columbite-Mn and was found as an inclusion in sapphire 26 from the Oelberg. As it contained a complex oxide aggregate (Fig. 5a, b), it was repolished until it was free of inclusions which allowed for in situ LA-ICP-MS dating. The reliability of in situ U-Pb dating with LA-ICP-MS has previously been demonstrated by Smith et al. (2004), Dill et al. (2007), Melcher et al. (2008Melcher et al. ( , 2015, and Che et al. The returned weighted mean ages of the columbite-Mn inclusion (Table 5) are 24.74 ± 0.24 Ma (MSWD = 3.4) (Fig. 10). The accuracy and reproducibility were checked by repeated analyses of Coltan 007 (n = 8) which yields an intercept age of 2074.1 ± 7.5 Ma

Discussion
All sapphires investigated in this study are rimmed by tens to several hundred micrometer sized spinel rims, indicating chemical disequilibrium with the host basalt and supporting a xenocrystic nature of the sapphires. This is supported by the phase relations in the CMAS system (Sen and Presnall 1984;Milholland and Presnall 1998;Presnall 1990, 2000), showing that a hypothetical cooling path of a basaltic model melt will never intersect the corundum stability field (Fig. 11). In the presence of MgO (in CMAS) or MgO+FeO (in basalt), stable aluminous phases co-existing with near-liquidus melts are spinel at 1-2 GPa, garnet at ~3 GPa, or anorthite at pressures <1 GPa. Hence, the stability of corundum is not only a function of the alumina content of a melt, but is also sensitive to the presence of MgO and FeO. In a primitive basaltic melt with high MgO and FeO contents, the corundum stability field in the CMAS system will never be intersected upon cooling from magmatic temperatures, so that corundum will never be a stable alumina phase.
Questions addressed in the discussion of this paper are the nature of the parental melt that crystallized the sapphires, as well as the relationship between the sapphires and the host primitive alkaline basalts, as sapphires only occur in such primitive basalts from intra-continental settings, but never in basalts from other geotectonic settings.

Magmatic vs. metamorphic origin
Accepting a xenogenetic relation between the sapphires and the host basalt, the first question to be answered is the geological nature of the Siebengebirge sapphires, i.e., if they formed via igneous or metamorphic processes. The distinction between sapphire crystals of magmatic and of metamorphic origin is mainly based on their Ga/ Mg ratio (Peucat et al. 2007), but also on the mineral inclusion suite (e.g., Sutherland and Schwarz 2001;Graham et al. 2004). According to Peucat et al. (2007), magmatic blue sapphires have rather high Ga/Mg ratios (>10) and metamorphic or metasomatic blue sapphires have rather low Ga/Mg ratios (<10). According to that discrimination, the Siebengebirge sapphires would mostly fall into the "magmatic" field, whereas three samples (26, KS21, OL60) would clearly plot in the "metamorphic" field. This discrimination between magmatic and metamorphic sapphires by Peucat et al. (2007) is based on the fractionation of Ga/Al and of Mg in sapphires during crystallization. In the corundum crystal lattice, Ga substitutes for Al, whereas Mg is located at interstitial lattice sites. Wahlen et al. (1987) found that magmatic corundum generally has higher Ga/Al ratios than metamorphic corundum, explaining this by the extraction of Ga from the metamorphic rocks by circulating F-rich fluids during partial melting of granulites. During fluid circulation, F − would complex with Ga 3+ to form GaF 3− 6 ions that are extracted, leaving the restitic granulite depleted in Ga over Al. If this process implicated by Whalen et al. (1987) controlled the Ga/Mg ratio of a sapphire and hence indicated a metamorphic or magmatic origin, the Ga/Mg ratio should be a function of the Ga content of the sapphire. Such a Ga control of the Ga/Mg ratio cannot be observed for the Siebengebirge sapphires. In contrast, the Ga/Mg ratio is correlated with the bulk Mg content of the sapphire (Fig. 12). Therefore, we consider it unlikely that the Siebengebirge sapphires represent a polymodal suite, i.e., that some samples were formed via metamorphic processes and others are igneous precipitates. Support for a monogenetic, magmatic, origin for all Siebengebirge sapphires is provided by several other mineralogical characteristics which are (1) syngenetic mineral inclusions of Pyrochlore Group, Columbite Group, calcite and feldspar series members that are more typical for a magmatic origin, (2) the presence of glassy inclusions Other bands cannot be unambiguously assigned to a certain mineral. But certainly, the daughter phase must be a sodium sulfate. The reference spectra are from the RRUFF Database. The following reference spectra were used: blödite: R050341; thenardite: R040178; aphthitalite: R050651; corundum: R060020 which were likely melts that were trapped as the corundum crystals grew, and (3) the presence of numerous Nb-rich nano-inclusions in the brownish domain of sapphire PE25 which are most likely magmatic precipitates that co-precipitated with the sapphire host during crystal growth. This lamellar distribution of intermittently alternating inclusionpoor and inclusion-rich zones of the nano-inclusions parallel to the prism faces of the host sapphire supports an epitaxial growth of the inclusions on the sapphire crystal faces. Such oscillatory patterns are ubiquitous in many geochemical systems (L'Heureux 2013) and are either interpreted to be externally controlled and to reflect periodic chemical changes in the crystallizing environment, or to be internally controlled by intrinsic self-organization in a nonequilibrium system with positive feedback (e.g., Ortoleva 1994;Merino and Wang 2001). Perhaps the most compelling evidence for a magmatic origin for the Siebengebirge sapphires comes from the age of their inclusions. If the sapphires and hence their mineral inclusions had been of metamorphic origin, then they should yield Devonian ages (i.e., the age of the metamorphic basement below the SVF). However, our own in situ dating resulted in Cenozoic ages which, when taking into account the uncertainty, are the same age as the host basalt (24.74 ± 0.24 Ma), indicating a clear magmatic origin for these inclusions and hence for their host sapphires.
Based on this discussion, we state that the discrimination between magmatic and metamorphic sapphires based on the Ga/Mg ratio as proposed by Peucat et al. (2007) has to be regarded with caution and may not be applied to any sapphire without additional information provided by the mineral inclusion suite. The Ga/Mg ratio may be used as a discrimination feature for the sapphire samples investigated in the study from Peucat et al. (2007), but at least for the sapphires presented in this study this may not be applicable. Similar discrepancies for the discrimination between "magmatic"  and "metamorphic" sapphires based on their Ga content, and the Ga/Mg ratio as proposed by Peucat et al. (2007) were pointed out by Izokh et al. (2010), Uher et al. (2012), Sutherland et al. (2015), and Palke et al. (2015Palke et al. ( , 2017. The discrepancies in trace element data and the resulting ambiguity about the origin of sapphires from alluvial deposits in Montana, USA, lead Palke et al. (2017) to suggest the trace element classification scheme for sapphires to be revisited and revised. Based on our trace element data on the Siebengebirge sapphires, we support the necessity of the revision of the Peucat et al. (2007) trace element-based discrimination for the identification of the origin of sapphires. Spot size = 19 and 28 µm, respectively; depth of crater ~20 µm. 206 Pb/ 238 U error is the quadratic addition of the within run precision (2 SE) and an estimated repeatability of 2% (2 SD). 207 Pb/ 206 Pb error propagation ( 207 Pb signal dependent) following Gerdes and Zeh (2009 g Accuracy and reproducibility were checked by repeated analyses (n = 8 and 9) of reference columbite 139 and 007; data given as mean with 2 standard deviation uncertainties

Constraints on the parental melt
Important indicators for the composition of the parental melt of the Siebengebirge sapphires are the trace element composition and the composition of syngenetic mineral, fluid, and glass inclusions, as these directly sample the parental melt. One key characteristic of the Siebengebirge sapphires is the abundance of HFSE-rich mineral inclusions and of HFSE-rich nano-inclusions in the brownish domains, indicating that the sapphires' parental melt must have been rich in the HFSE. Such high concentrations in the HFSE are well known from highly evolved silicate melts and from carbonatitic melts. Most authors (e.g., Sutherland et al. 1998b) relate the HFSE enrichment of the sapphires and the presence of Columbite Group, Pyrochlore Group and of Nb-rich rutile mineral inclusions to highly evolved SiO 2 undersaturated silicate melts such as syenites, or nepheline syenites. A further observation leading these authors to favor syenites as parental melts is that repeatedly melt inclusions of syenitic composition were reported from magmatic sapphires (e.g., Pakhomova et al. 2006;Zaw et al. 2006;Izokh et al. 2010). Up until now, carbonatites have not been considered as potential parental melts, although many authors (e.g., Coenraads et al. 1990;Van Long et al. 2004;McGee 2005;Nechaev et al. 2009;Izokh et al. 2010;Palke et al. 2017) reported that CO 2 must have played a significant role in the formation of these sapphires, as CO 2 -rich fluid inclusions are commonly observed in many magmatic sapphires. Palke et al. (2017) interpreted the presence of analcime-calcite inclusions in alluvial magmatic sapphires from Montana, USA as an indication for that the sapphires crystallized from a melt that simultaneously exsolved a carbonatite melt, and fractions of that carbonatite melt were trapped by the growing sapphire and later recrystallized to the analcime-calcite paragenesis. However, in their genetic model, the carbonatite melt was only a "by-product" that resulted from the carbonated nature of the source rock, and they did not state a compelling necessity of the presence of CO 2 or a carbonatite melt for the sapphire formation. Only Guo et al. (1996) and Limatrakun et al. (2001) assign carbonatitic melts a role in sapphire petrogenesis. Yet, the only role that was assigned to carbonatites in these studies was their function as a CO 2 source, but no direct parentage was proposed, mainly due to the lack of primary evidence for the presence of a carbonatitic melt during sapphire crystallization. The Siebengebirge sapphires by contrast do show direct evidence for the participation of a carbonatitic melt in the sapphire petrogenesis which is given by the several primary carbonate inclusions and by the glasses in sample 26. These glasses have two different compositions and are either a carbonated silicate or a phonolitic glass, and probably represent quenched liquids that formed via liquid immiscibility from a carbonated silicate melt, as indicated by the textures shown in Fig. 6. We, therefore, suggest that liquid immiscibility in the carbonatite-silicate melt system played a major role in the genesis of magmatic sapphires, in a similar fashion as stated before by Palke et al. (2017) for magmatic sapphires from Montana, USA.
The liquid miscibility relations in the carbonatitesilicate system have been studied intensively in the past decades (Koster VanGroos andWyllie 1963, 1966;Hamilton and Freestone 1979;Freestone and Hamilton 1980;Kjarsgaard and Hamilton 1988;Brooker and Hamilton  wt. % Crn in c r e a s in g A l-c o n t e n t

Fig. 11
Phase relations in the CMAS system at 2 GPa (Liu und Presnall 1990) pressures. Abbreviations after Whitney and Evans (2010). The star in the forsterite field represents a typical composition of an alkali basalt. Upon cooling, the melt will follow the red path. The final eutectic composition of the melt is given by the small stars at the anorthite-enstatite-quartz triple point There is no visible effect of Ga on the Ga/Mg ratio, whereas the total Mg content in the sapphires is highly correlated with the Ga/Mg ratio, indicating that the strong variability of the Ga/Mg ratio in the Siebengebirge sapphires is controlled by changes in the Mg concentration, and largely unaffected by changes in the Ga content 1990; Wyllie 1997, 1998;Brooker 1998;Brooker and Kjarsgaard 2010). From the phase relations shown in Fig. 13, it becomes obvious that the two carbonated silicate and the phonolitic melts do not represent two melts that exsolved from each other, as two liquids that lie in the field of liquid immiscibility would exsolve along the cotectic lines, so that one composition would lie at the lowermost (SiO 2 and Al 2 O 3 rich) two field boundary, and the co-existing composition would lie at the opposite (Na 2 O and K 2 O/CaO rich) two field boundary. Yet, the glasses show striking evidence for liquid immiscibility, such as the vesicular shape of the innermost glass (melt 1). Consequently, if these glasses do represent quenched products of two melts that formed via liquid immiscibility, as indicated by the vesicular shape, then there must have been an additional, Na 2 O-and K 2 O-rich liquid that simultaneously separated from the co-existing SiO 2 -and Al 2 O 3 -rich liquids, but is now not present in the sapphire. Evidence for the existence of such an Na 2 O-and K 2 O-rich liquid is given by the alkali-rich daughter phases in the fluid inclusions that were observed in some of the sapphires. This Na 2 O-and K 2 O-rich liquid would have a composition that lies somewhere on the 'a-b' two liquid field boundary in Fig. 13. Upon cooling, the melt will evolve from a to b via crystallization of silicate phases, and will then evolve along the silicate-carbonate field boundary 'b-c' with the co-precipitation of silicate and carbonate phases (Lee and Wyllie 1998). Such a melt would be a carbonatite containing ≤5 wt% SiO 2 +Al 2 O 3 . The prerequisite for a liquid that lies along the a-b two liquid field boundary to cross the silicate liquidus surface and reach the silicate-carbonate field boundary to precipitate carbonate minerals is the physical separation of the carbonatitic melt from the co-existing silicate melt (e.g., Kjarsgaard and Peterson 1991;Lee and Wyllie 1998). When the necessity for the presence of carbonatitic melt is combined with the abundance of carbonate mineral inclusions and CO 2 -rich fluid inclusions in the sapphires, it seems most probable that the sapphires crystallized from the carbonatite that separated from the melts now present as glass inclusions within the sapphires. After physical separation of the putative carbonatitic melt from the co-existing silicate melt, the carbonatitic melt will evolve separately. It is likely that the sapphires from the SVF crystallized from these carbonatitic melts that exsolved from a highly evolved melt of phonolitic composition.
Additional support to the idea that sapphires crystallized from a carbonatitic parental melt is given by the extremely low viscosity of carbonatitic melts of ~1.5 × 10 −2 -5 × 10 −3 Pa s (Dobson et al. 1996) which would promote rapid ion transport through the melt that is needed to form megacrystic corundum at low degrees of alumina oversaturation.
A possible origin of the sapphires from carbonatitic melts could be identified employing oxygen isotope measurements on the sapphires, as carbonate-associated corundum has generally higher δ 18 O values than others ). Yet, interpreting oxygen isotope data from sapphires that prospectively precipitated from a carbonatitic melt would be quite difficult as there are no data available about oxygen isotope fractionation between corundum and carbonatitic melt.

Genetic relation between the sapphires and their host rocks
In situ LA-ICP-MS U-Pb dating of a syngenetic columbite-Mn inclusion within a xenocrystic sapphire hosted by alkaline basalt from the Oelberg in the SVF yielded an age of 24.73 ± 0.35 Ma. This age closely matches the age of alkaline mafic volcanism in the area. The alkali basalts from the Oelberg have an 40 Ar-39 Ar isochron age of 24.91 ± 0.45 Ma (Przybyla 2013) and can thus be regarded as cogenetic with the Siebengebirge sapphires. The Oligocene age of the sapphires confirms a magmatic origin, as a sapphire that originated from the basement below the SVF would be expected to yield a Variscan age. The agreement between the age of the Siebengebirge sapphires and the alkaline basaltic volcanism provides crucial evidence for a genetic link between the processes  Fig. 13 Location of the glass compositions (melt 1: green stars; melt 2: red star) in the schematic pseudoternary phase diagram of the Hamilton projection at 1 GPa (Lee and Wyllie 1998). The line 'a-b' represents the two liquid field boundary at the Na 2 O, K 2 O rich side, and line 'b-c' represents the silicate-carbonate field boundary. The composition of both glasses plot in the same field (silicate liquidus surface at the SiO 2 +Al 2 O 3 rich apex), ruling out that they formed solely via liquid immiscibility triggering alkaline basaltic magmatism and the crystallization of sapphire megacrysts. The genetic link between the sapphire crystallization and host alkaline basalt must be via CO 2 . As mentioned in the introduction, the BGY sapphires are restricted in occurrence to a very specific geotectonic setting, i.e., to intra-continental rift-related alkaline basalts. It has been inferred by many previous studies (e.g., Hirose 1997; Green and Falloon 1998;Dasgupta et al. 2007;Zeng et al. 2010) that the mantle source generating continental intra-plate alkaline volcanics must be either carbonate-bearing, or metasomatically enriched by CO 2 bearing fluids. Also, the alkaline mafic melts from the SVF are suggested to be generated from refertilized carbonated spinel peridotite ). This infers that BGY sapphires are only generated in intra-continental settings where the mantle source is enriched in CO 2 , so that primary melts can evolve to phonolites and later expell carbonatites which in turn may crystallize sapphires. The necessity of a CO 2 enriched mantle source to produce melts that are capable of precipitating BGY sapphires explains the lack of sapphires in SiO 2 saturated basalts such as tholeiites.

Barometry
When a magmatic mineral contains CO 2 -bearing fluid inclusions, the density of the inclusions may be used to quantify the pressure during crystallization (Bertrán 1983). The density of CO 2 -bearing fluid inclusions can be calculated from the distance between the CO 2 main peaks at ~1388.0 cm −1 (upper band) and ~1285.5 cm −1 (lower band) of its vibrational spectrum obtained using Raman spectroscopy (e.g., Garrabos et al. 1980;Rosso and Bodnar 1995;Yamamoto and Kagi 2006;Wang et al. 2011). For the Siebengebirge sapphires, we used the equation of Wang et al. (2011) that relates the distance between the CO 2 upper and lower band to the density of a CO 2 -bearing fluid inclusion, and obtained densities in the range of 0.3-0.9 g/cm 3 . A systematic relation between the estimated density of a fluid inclusion and its location within the host sapphire was not observed. The pressure during entrapment of the fluid by the host crystal can be calculated from the volumetric properties of the fluid inclusion, given that composition and trapping temperature are known and given that the system was closed. Pressure and temperature vary along isochores. The P-V-T relations of pure CO 2 were calculated by Yamamoto et al. (2002) from the equation of state for CO 2 by Pitzer and Sterner (1994). Assuming that the Siebengebirge sapphires crystallized from carbonatitic melts and taking average liquidus temperatures for carbonatites of 700-1000 °C (Dobson et al. 1996;Genge et al. 1995;Wolff 1994), the pressure of sapphire crystallization must have been between 0.1 and 0.4 GPa. As we cannot exclude the possibility that during ascent-related decompression of the sapphires some fluid inclusions decrepitated to form inclusions of lower density than the original one, we therefore assume that the highest estimated pressure of 0.4 GPa to be the minimum pressure of fluid trapping and hence of sapphire crystallization. This pressure corresponds to a depth of ~12 km. Based on fluid inclusions, the lithostatic pressure during crystallization has been previously obtained for sapphires from Weldborough and Kings Plains, Australia  (Fig. 14). The temperature estimates for the Nezametnoye (780-820 °C) sapphires are also well in accord with a putative crystallization from carbonatites at low magmatic temperatures. All P-T estimates from the literature, and from this study are well in accord with a crystallization of the magmatic sapphires from carbonatitic melts at mid-crustal levels. The field for the Siebengebirge sapphires is drawn based on assumed average temperatures of 700-1000 °C for the carbonatites that crystallized the sapphires (Dobson et al. 1996;Genge et al. 1995;Wolff 1994

Petrogenetic model for the Siebengebirge sapphires
The formation of the alkaline basalt-hosted sapphires from the SVF is a result of a complex line of magmatic processes that elapsed during a very short time scale. During Miocene times, extensive active volcanism started in the SVF, producing a broad spectrum of SiO 2 saturated and undersaturated primitive and evolved volcanic rocks. The alkaline mafic melts probably formed due to melting of amphibole and phlogopite-bearing peridotite in the spinel peridotite stability zone Schubert et al. 2015), at pressures between 1 and 2.5 GPa and temperatures between 1150 and 1350 °C (Fig. 15a). These P-T conditions correspond to the lithospheric mantle. It has been repeatedly suggested that the alkaline mafic rocks were formed from a carbonatite-metasomatized mantle source Pfänder et al. 2012). As a result, the melts must have been enriched in volatiles including CO 2 . With ensuing fractionation of the primitive melts towards more evolved phonolitic compositions, the partial pressure of CO 2 (pCO 2 ) in the melt increases ( Fig. 15b) until at high degrees of differentiation and high pCO 2 , a carbonatitic melt exsolves from the phonolite (Kogarko 1997). The carbonatitic melt exsolving from a highly evolved phonolite will be CaO-and volatile-rich but is expected to be depleted in SiO 2 , MgO, and FeO (Kogarko 1997). The carbonatite will cool and start to crystallize carbonate phases when it is physically separated from the co-existing silicate melt (Lee and Wyllie 1998). The physical separation of the carbonatitic melt from the silicate melt will probably occur rapidly after exsolution due to the significant viscosity and density difference of both melts. The crystallization of the sapphires from highly fractionated carbonatite occurred at middle-to upper crustal levels ( Fig. 15c) after the physical separation of the carbonatite and the phonolite melt, as indicated by the presence of carbonate inclusions in the sapphires. A fresh pulse of hot ascending alkali basalts later encounters the carbonatite melt pocket in the crust on their way upwards, decomposes the carbonatitic melt and incorporates the sapphires to transport them to the surface (Fig. 15d).

Summary and conclusions
Magmatic sapphires are well known to occur associated with intra-continental alkaline mafic volcanism. Around the world, predominantly in Asia and Australia, many alkaline basalt fields are reported to carry megacrysts of magmatic sapphires. To date, many hypotheses have been proposed to investigate the nature of these so-called BGY sapphires. Still, although these studies comprise detailed investigations on the sapphires including trace element analyses, analysis of mineral inclusions as well as melt and/or fluid inclusion and oxygen isotope measurements, no consensus was reached about the origin of these sapphires. Moreover, these studies cannot explain the intimate association of blue-green-yellow (BGY) sapphires with alkaline basaltic volcanism. This study on the BGY sapphires hosted by alkaline basalts from the Siebengebirge Volcanic Field (SVF) elucidates many yet unknown details about the formation of this special type of gemstone. The main conclusions from this study can be summarized as follows: 1) Sapphire megacrysts are xenogenetic to the host basalts, as indicated by the spinel corona surrounding each sapphire at the contact towards the basalt. 2) The parental melt was probably a carbonatitic melt that exsolved from a highly evolved phonolitic melt, as evidenced by carbonate inclusions, strong enrichment in the HFSE, by the glass inclusions of carbonated silicate in contact with phonolite, and by CO 2 -bearing fluid inclusions. 3) There is a close genetic relation between the sapphires and their host alkaline mafic rocks, as evidenced by the overlapping age of the Siebengebirge sapphires dated in this work and the age of the alkaline mafic volcanism in the area (Przybyla 2013). We identified the genetic link between magmatic sapphires and the host basalt that has been a puzzle to date, to bear upon CO 2 . Only alkaline volcanic suites can build up enough CO 2 in the magma chamber upon fractionation so that at high degrees of fractionation a carbonatitic melt exsolves which in turn can crystallize sapphires at a late stage of evolution. During ascent, fresh alkaline basaltic melts that encounter carbonatite melt pockets will decompose the carbonatite and incorporate the sapphires to carry them to the surface.
Our findings from the Siebengebirge sapphires raise the question about an overall applicability of the genetic model to worldwide occurrences of magmatic sapphires. It is hard to tell if all magmatic sapphires from the world must have precipitated from a carbonatitic melt that exsolved from highly fractionated SiO 2 undersaturated silicate melts. If so, one would expect more sapphires from other locations to show similar features evidencing for a carbonatitic parental melt. Yet, it is important to notice that even those sapphires from this study do not all show the whole range of features pointing towards a carbonatitic parentage, i.e., carbonate and HFSE-rich mineral inclusions + CO 2 fluid inclusions with carbonate daughter phases + silicate/carbonate melt a Alkaline basaltic volcanism is initiated in the SVF. b Differentiation of alkaline mafic magmas leads to the formation of highly evolved phonolites, which is accompanied by a build-up of pCO 2 in the melt. c In the highly evolved phonolite that is now saturated in CO 2 , a carbonatitic melt exsolves and separates from the phonolite. Sapphires precipitate from the carbonatite after physical separation. d Fresh pulses of alkaline mafic melts ascend. When encountering the carbonatite lenses, the alkaline mafic melts decompose the carbonatite, incorporate the sapphires and carry them to the surface inclusions. It is rather the sum of single pieces of mineralogical evidence from the whole sample set that inferred a carbonatite source for the Siebengebirge sapphires. Therefore, it cannot be excluded that other magmatic sapphires that do not bear mineralogical evidence for a carbonatite origin do not have the same petrogenetic history. Hence, an overall applicability of our petrogenetic model for the origin of magmatic sapphires should be considered, but cannot be stated with certainty. However, from the findings of not only our study, but also from referenced herein, we can state with some confidence that in general, sapphire crystallizing melts were produced via melting of CO 2 -rich, probably carbonatite metasomatized, source rocks.