Nitrogen under Super-Reducing Conditions: Ti Oxynitride Melts in Xenolithic Corundum Aggregates from Mt Carmel (N. Israel)

Titanium oxynitrides (Ti(N,O,C)) are abundant in xenolithic corundum aggregates in pyroclastic ejecta of Cretaceous volcanoes on Mount Carmel, northern Israel. Petrographic observations indicate that most of these nitrides existed as melts, immiscible with coexisting silicate and Fe-Ti-C silicide melts; some nitrides may also have crystallized directly from the silicide melts. The TiN phase shows a wide range of solid solution, taking up 0–10 wt% carbon and 1.7–17 wt% oxygen; these have crystallized in the halite (fcc) structure common to synthetic and natural TiN. Nitrides coexisting with silicide melts have higher C/O than those coexisting with silicate melts. Analyses with no carbon fall along the TiN–TiO join in the Ti–N–O phase space, implying that their Ti is a mixture of Ti3+ and Ti2+, while those with 1–3 at.% C appear to be solid solutions between TiN and Ti0.75O. Analyses with >10 at% C have higher Ti2+/Ti3+, reflecting a decrease in fO2. Oxygen fugacity was 6 to 8 log units below the iron–wüstite buffer, at or below the Ti2O3–TiO buffer. These relationships and coexisting silicide phases indicate temperatures of 1400–1100 °C. Ti oxynitrides are probably locally abundant in the upper mantle, especially in the presence of CH4–H2 fluids derived from the deeper metal-saturated mantle.


Introduction
Nitrogen is a minor element in Earth's mantle; it is usually regarded as a fugitive element, ultimately carried to the surface in magmas or C-O-H fluids, and diamonds. Many studies have treated its behavior under relatively oxidizing conditions, in which such fluids are dominated by CO 2 + H 2 O. However, several analyses (e.g., [1][2][3]) suggest that the sublithospheric mantle is metal-saturated, in which case the oxygen fugacity is controlled by the iron-wüstite buffer (IW), and C-O-H fluids are dominated by CH 4 + H 2 . Under such reducing conditions, N may display different lithophile/siderophile behavior, as witnessed by the occurrence of osbornite (TiN) in reduced enstatite chondrites, together with phases such as cohenite (Fe 3 C) and oldhamite (CaS) ( [4], references therein).
Osbornite was first reported as a terrestrial material in the ejecta of kimberlitic eruptions in the Azov area on the northern edge of the Ukrainian Shield [5], where it is accompanied by other highly reduced phases including moissanite (SiC), as well as abundant corundum. It also occurs as inclusions in chromitites from the Luobusa ophiolite of the Yarlung-Zhangbo suture in SE Tibet [6]; the associated phases are Ti-rich corundum, moissanite and a variety of Fe-Ti silicides and carbides [7][8][9].
The Cretaceous pyroclastic rocks of Mount Carmel (Northern Israel) erupted from six to eight volcanic centers from 100-85 Ma, and over an area of at least 150 km 2 ( Figure S1). Xenoliths and xenocrysts enclosed in these tuffs and in related alluvial deposits appear to represent snapshots of similar magma-fluid systems at different stages of their evolution. They record a wide range of oxidation states, including some of the most reducing conditions yet reported in terrestrial rocks [10,11]. Melts trapped in xenolithic corundum aggregates show melt evolution and fractional crystallization under strongly reducing conditions (f O 2 = ∆IW − 6 to − 9 [10,11]), consistent with fluxing of a magmatic system by CH 4 + H 2 fluids derived from a metal-saturated mantle [12,13]. Inclusions in corundum comprise a range of Ti-rich phases in which Ti is either trivalent (tistarite, Ti 2 O 3 ; carmeltazite, ZrAl 2 Ti 4 O 11 ; khamrabaevite, TiC [12,14]) or divalent (TiB 2 [15]); these are associated with a range of immiscible Fe-Ti-Zr-C silicide melts and their crystallization products [16] moissanite (SiC [17]), hibonite [10] and a series of Ti nitrides and oxynitrides, the subject of this paper.
The geological setting and provenance of the samples have been discussed in detail elsewhere [18] to demonstrate that assertions of their anthropogenic origin [19] are untenable. In brief: (1) all of the mineral assemblages described in this and previous papers are found in situ in the pyroclastic rocks of Cretaceous volcanoes on the Mt Carmel plateau; (2) the alluvial deposits that also contain these xenoliths are paleoplacers that rest on bedrock at the base of Pliocene-Pleistocene terraces well above the current drainage of the Kishon River, and are overlain by 4-10 m of undisturbed sediments; (3) much of the heavy mineral assemblage, including abundant moissanite, is also found in lithified Miocene beach placers, deposited during a marine incursion that redistributed surface materials across the Yizre'el valley, now drained by the Kishon River. All of these deposits predate human occupation in the area.
Here, we present petrographic images and elemental maps to establish the relationships between nitrides, borides, carbides and silicide and silicate melts; we then present mineral chemistry and discuss the origins and evolution of these melts and the associated nitrides.

Methods
The sampling methods, which are designed to avoid anthropogenic contamination, have been described in detail elsewhere [11,18].

SEM, FE-SEM, EMP
Samples were mounted in epoxy blocks, polished and coated with carbon; all mounts used for the analyses presented here were cleaned and carbon-coated together. A Zeiss EVO MA15 scanning electron microscope (SEM) at the Geochemical Analysis Unit (GAU), Macquarie University, Sydney, Australia was used to capture backscattered electron (BSE) images and energy dispersive X-ray spectrometry (EDS) was used to map the elemental composition of the samples, and to analyze phases of interest. An accelerating voltage of 15 keV and a beam current of 1 nA were used. Images and qualitative EDX data were also collected at Macquarie University on an FEI Teneo field emission scanning electron microscope equipped with secondary and backscattered-electron detectors and a pair of integrated Bruker energy dispersive X-ray spectroscopy analyzers (Bruker XFlash Series 6).
Major and minor elements were determined by an electron microprobe (EMP) using a CAMECA SX100 equipped with five wavelength-dispersive spectrometers at the Macquarie University GeoAnalytical facility (MQGA; formerly GAU), Macquarie University, Sydney, Australia. Analyses were performed using a focused beam (1-2 µm) with an accelerating voltage of 15 keV and a beam current of 30 nA. Calibration standards were a suite of natural and synthetic minerals, including synthetic TiB 2 , TiC, TiO and TiN. B, C, N and O were measured using a PC2 crystal (on Spectro 4). Peak counting was 30 s, and background was counted for 15 s on either side of the peak. Oxygen and carbon were corrected for blank effects (carbon coating) by analysis of the synthetic Ti phases and Fe-Ti silicide alloys present in the analyzed samples; matrix corrections were carried out by ZAF software.

Transmission Electron Microscopy (TEM)
Focused ion beam (FIB) foils from the region of interest were prepared for TEM using two dual-beam FIB systems, an FEI Nova NanoLab 200 at UNSW and an FEI Helios G3 CX at the Centre for Microscopy Characterisation and Analysis (CMCA), the University of Western Australia, Perth, Australia. High-angle annular dark field-scanning transmission electron microscopy (HAADF-STEM) imaging and element mapping were carried out using an FEI Titan G2 80-200 TEM/STEM with ChemiSTEM Technology operating at 200 kV at the Centre for Microscopy Characterisation and Analysis (CMCA), the University of Western Australia, Perth, Australia. The element maps were obtained by EDS using the Super-X detector on the Titan with a beam size~1 nm and a probe current of~0.25 nA. Total acquisition times of 20-30 min were used to obtain good signal-to-noise ratios. Electron diffraction was carried out using a field-limiting aperture that selected an area approximately 400 nm in diameter. Electron energy-loss spectroscopy (EELS) was carried out using a Gatan Enfinium spectrometer at the same beam size and beam current as the EDS, and at a dispersion of 0.1 eV/pixel.

Background
As in other xenolith studies, an idealized history of magma/fluid evolution can be constructed by piecing together information expressed in a suite of xenoliths and xenocrysts. Specifically, it is possible to arrange the samples from Mt Carmel in a sequence of progressively lower oxidation states (decreasing f O 2 ) that appears to record the evolution of magmas fluxed by CH 4 + H 2 at high fluid/rock ratios [9,10,12,20]. Grains of vesicular wüstite show f O 2 near the QFM buffer (f O 2 = ∆IW + 3 to + 4). Large euhedral to subhedral crystals of magnesian calcite have high-Sr cores with 87 Sr/ 86 Sr = 0.7033, suggesting a magmatic origin, and may have crystallized near the enstatite-magnesite-olivine-diamond (EMOD) buffer (∆IW + 1.5). Spheres of quenched Fe oxide-silicate and Ti oxide-silicate melts, many with cores of native iron (Fe 0 ), suggest metal/metal oxide melt immiscibility at f O 2 near IW.
Among the most common xenoliths are aggregates of skeletal/hopper crystals of corundum ( Figure 1), a morphology that implies rapid crystallization from a melt supersaturated in Al 2 O 3 . Numerical modeling of the zoning patterns defined by Ti 3+ concentrations [21] indicates a progression from skeletal crystallization in open-system conditions, to hopper forms in a partially closed system and finally to growth of euhedral corundum into closed melt pockets. Analysis of crystal growth rates and diffusion profiles suggests that the aggregates crystallized on very short time scales, from days to years. This rapid growth trapped pockets of Ca-Mg-Al-Si-O melts, and continuing fractional crystallization of the host corundum led to residual melts with high contents of Ti, Zr and REE. The earliest phenocrysts in these pockets are Mg-Al-Ti spinel, followed by tistarite (Ti 2 O 3 ) and then carmeltazite (ZrAl 2 Ti 4 O 11 ) [12,14]. More evolved melts precipitated REE-rich hibonite (CaAl 12 O 19 ), moissanite (SiC) and several Zr-REE phases. The melt pockets also contain a variety of immiscible carbon-rich Fe-Ti-Zr silicide/phosphide melts, mostly in the form of spherical droplets; these crystallized TiC, SiC, TiB 2 and Ti oxynitrides and a range of silicide phases [16]. Crystals of moissanite (up to 4 mm) contain inclusions of Si 0 melts that crystallized Fe, Ca and Al silicides [17]. The crystallization of SiC and the reduction of Ti 4+ to Ti 3+ require fO2 = ΔIW −6 to −7 [12,17]. The most evolved silicate melts produced coarse-grained hibonite+grossite+spinel aggregates with inclusions of V 0 , requiring fO2 ≤ ΔIW−9 [11]. Even lower fO2 may be indicated by the presence of Ti 2+ -bearing phases such as TiB2 [15]. The most reduced conditions imply a hydrogen-dominated atmosphere, which has been confirmed by the discovery of VH2 [22,23] and the presence of abundant H2 in the hibonite crystals that carry inclusions of V 0 [10,11].
Paragenetic studies [9,10,20] suggest that these xenocrysts formed at ca. 1 GPa, close to the crust-mantle boundary, and crystallized over a T range of approximately 1450 °C to 1150 °C. Dating of zircon xenocrysts from the pyroclastic rocks [13] suggests that the crust was underplated by mafic magmas from ca. 285-100 Ma and that these melts differentiated far enough to crystallize large zircons and the typical "basaltic" sapphire found in the tuffs and alluvial deposits. We have suggested previously that volumes of these syenitic (s.l.) residual melts in the underplate were reduced by interaction with mantlederived CH4 + H2 fluids [20], prior to entrainment in the Cretaceous host basalts. The detailed origin and evolution of the reduced melts will be discussed elsewhere.

Petrography
The Ti nitrides (hereafter TiN) are abundant within the corundum aggregates, and appear in a remarkable array of morphologies: small isolated globular to cruciform inclusions in corundum ( Figure 2); irregular blobs, drop-shaped lenses and cruciform inclusions in silicate melts (Figures 3 and 4); long arrays or lattice patterns, apparently controlled by the orthogonal voids in the skeletal structure of the corundum crystals (Figures 4b  and 5); and irregular bodies interstitial to corundum crystals with faces that may be either smooth or dentate ( Figure 6). Some nitrides occur as apparent crystals on the rims of balls of Fe-Ti silicide melts, often together with TiB2 [15] and/or TiC [16]; others clearly are immiscible with the Fe-Ti-silicide melts (Figures 6 and 7). The crystallization of SiC and the reduction of Ti 4+ to Ti 3+ require f O 2 = ∆IW − 6 to −7 [12,17]. The most evolved silicate melts produced coarse-grained hibonite+grossite+spinel aggregates with inclusions of V 0 , requiring f O 2 ≤ ∆IW − 9 [11]. Even lower f O 2 may be indicated by the presence of Ti 2+ -bearing phases such as TiB 2 [15]. The most reduced conditions imply a hydrogen-dominated atmosphere, which has been confirmed by the discovery of VH 2 [22,23] and the presence of abundant H 2 in the hibonite crystals that carry inclusions of V 0 [10,11].
Paragenetic studies [9,10,20] suggest that these xenocrysts formed at ca. 1 GPa, close to the crust-mantle boundary, and crystallized over a T range of approximately 1450 • C to 1150 • C. Dating of zircon xenocrysts from the pyroclastic rocks [13] suggests that the crust was underplated by mafic magmas from ca. 285-100 Ma and that these melts differentiated far enough to crystallize large zircons and the typical "basaltic" sapphire found in the tuffs and alluvial deposits. We have suggested previously that volumes of these syenitic (s.l.) residual melts in the underplate were reduced by interaction with mantle-derived CH 4 + H 2 fluids [20], prior to entrainment in the Cretaceous host basalts. The detailed origin and evolution of the reduced melts will be discussed elsewhere.

Petrography
The Ti nitrides (hereafter TiN) are abundant within the corundum aggregates, and appear in a remarkable array of morphologies: small isolated globular to cruciform inclusions in corundum ( Figure 2); irregular blobs, drop-shaped lenses and cruciform inclusions in silicate melts (Figures 3 and 4); long arrays or lattice patterns, apparently controlled by the orthogonal voids in the skeletal structure of the corundum crystals (Figures 4b and 5); and irregular bodies interstitial to corundum crystals with faces that may be either smooth or dentate ( Figure 6). Some nitrides occur as apparent crystals on the rims of balls of Fe-Ti silicide melts, often together with TiB 2 [15] and/or TiC [16]; others clearly are immiscible with the Fe-Ti-silicide melts (Figures 6 and 7).            The chains of evenly spaced blobs and crosses of Ti nitrides within the silicate melt pockets, and in some long arrays in the skeletal cavities (Figure 4), suggest the necking down of originally elongated or lens-shaped inclusions of a trapped fluid [24]. Nitrides in contact with silicate glass are commonly rimmed by a zone of tadpole-shaped blebs of TiN mixed with the silicate glass, or included in late-crystallizing phases (Figures 3c,d, 5, 8 and 9). Oxides (tistarite, carmeltazite, spinel) crystallizing from the silicate melt may be in direct contact with the TiN, or separated from it by a similar rim (Figures 3, 5, 8 and 9). TiN in contact with corundum seldom shows such rims (Figures 3 and 5). The chains of evenly spaced blobs and crosses of Ti nitrides within the silicate melt pockets, and in some long arrays in the skeletal cavities (Figure 4), suggest the necking down of originally elongated or lens-shaped inclusions of a trapped fluid [24]. Nitrides in contact with silicate glass are commonly rimmed by a zone of tadpole-shaped blebs of TiN mixed with the silicate glass, or included in late-crystallizing phases (Figures 3c,d, 5, 8 and 9). Oxides (tistarite, carmeltazite, spinel) crystallizing from the silicate melt may be in direct contact with the TiN, or separated from it by a similar rim (Figures 3, 5, 8 and 9). TiN in contact with corundum seldom shows such rims (Figures 3 and 5).        (Table 1).

Mineral Chemistry and Structure
EMP analyses of Ti-N phases (Table S1) are summarized in Table 1 Figure 10); the Ti-N phase is best described as a carbon-bearing Ti oxynitride Tix(N,O,C)2-x where X varies from 0.74-1.04 (mean and median = 0.90, stdev = 0.08). The range of TiN-TiO substitution is only slightly larger than that observed by [25] in thin films.   (Table 1).

Mineral Chemistry and Structure
EMP analyses of Ti-N phases (Table S1) are summarized in Table 1 and Figures 9 and 10. The Ti-N phase shows a wide range of solid solution, taking up 0-10 wt% carbon and 1.7-17 wt% oxygen. Most of the analyses have 20-35 at.% N and 8-20 at.% O and more than half have >10 at% C. There is a broad negative correlation between N and C over a range of 20 at.% O ( Figure 10); the Ti-N phase is best described as a carbon-bearing Ti oxynitride Ti x (N,O,C) 2-x where X varies from 0.74-1.04 (mean and median = 0.90, stdev = 0.08). The range of TiN-TiO substitution is only slightly larger than that observed by [25] in thin films. There is a broad correlation between paragenesis and composition (Table 1) hough the paragenesis of some nitrides is uncertain because of the effects of 2D sectio of 3D structures. Nitrides occurring as blebs or crosses in silicate glasses nearly all <5 at.% C (0-7.4%, mean 2.3%) and <15 at.% O (3.3 to 7.8 %, mean 16.2 wt%), and t true also of most large interstitial nitrides (mean C = 2.0 at.%; mean O = 17.5 at.% contrast, nitrides associated with silicides ± TiB2 have much higher contents of carbo 23.1 at.%, mean 13.9 at.%) and lower contents of oxygen (7-24 at.%, mean 12.9 at.% trides filling skeletal cavities in the corundum show a distinct bimodal distribution (Tab corresponding to the two silicate-related and silicide-related populations. Small iso inclusions also show a broadly bimodal distribution of compositions, falling equally low-C/high-O and high-C/low-O populations (mean C = 2.3 and 13.3 at.%; mean O = and 9.6 at.%, respectively; Table 1). The proportions of analyses in the two popula are not necessarily representative of their relative abundances in the system as a w because larger inclusions were preferentially analyzed. The EMP data (Table S1) als clude a number of analyses with Ti between ca. 75 wt% and 67 wt%, suggesting the ysis of fine-grained mixtures of Ti(N,O,C) with another phase. One end member o distribution is represented by a group of analyses with a mean of 69 wt% Ti and an pirical formula of Ti2(N0.43O0.43C0.14)3 (Table 1).
TEM-EDS analysis of a nitride phase coexisting with TiB2 shows it t Ti(N0.57O0.17C0.09) ( Table 1). The Ti-N phase shown in Figure 9, which appears to have a melt immiscible with the host silicate melt, has the empirical formula Ti(N0.50O0.45 (TEM-EDS analysis). Both examples have electron diffraction patterns ( Figure S2) sistent with the halite structure (fcc) common to TiN, with a0 = 4.3-4.4 Å (cf synthetic a0 = 4.242 Å). There is a broad correlation between paragenesis and composition (Table 1), although the paragenesis of some nitrides is uncertain because of the effects of 2D sectioning of 3D structures. Nitrides occurring as blebs or crosses in silicate glasses nearly all have <5 at.% C (0-7.4%, mean 2.3%) and <15 at.% O (3.3 to 7.8 %, mean 16.2 wt%), and this is true also of most large interstitial nitrides (mean C = 2.0 at.%; mean O = 17.5 at.%). In contrast, nitrides associated with silicides ± TiB 2 have much higher contents of carbon (5-23.1 at.%, mean 13.9 at.%) and lower contents of oxygen (7-24 at.%, mean 12.9 at.%). Nitrides filling skeletal cavities in the corundum show a distinct bimodal distribution (Table 1) Table 1). The proportions of analyses in the two populations are not necessarily representative of their relative abundances in the system as a whole, because larger inclusions were preferentially analyzed. The EMP data (Table S1) also include a number of analyses with Ti between ca. 75 wt% and 67 wt%, suggesting the analysis of fine-grained mixtures of Ti(N,O,C) with another phase. One end member of this distribution is represented by a group of analyses with a mean of 69 wt% Ti and an empirical formula of Ti 2 (N 0.43 O 0.43 C 0.14 ) 3 (Table 1).

Discussion
TEM-EDS analysis of a nitride phase coexisting with TiB 2 shows it to be Ti(N 0.57 O 0.17 C 0.09 ) ( Table 1). The Ti-N phase shown in Figure 9, which appears to have been a melt immiscible with the host silicate melt, has the empirical formula Ti(N 0.50 O 0.45 C 0.05 ) (TEM-EDS analysis). Both examples have electron diffraction patterns ( Figure S2) consistent with the halite structure (fcc) common to TiN, with a 0 = 4.3-4.4 Å (cf synthetic TiN a 0 = 4.242 Å).

Valence State of Ti in Ti(N,O,C)
The phase space of the relevant part of the Ti-N-O system [25] is shown in Figure 10. The ternary space is dominated by a field in which all compositions can crystallize as solid solutions (Ti x N y O z ) with the rock-salt structure. The defect chemistry of this structure allows this wide range of compositions and accommodates a range in the oxidation state of Ti, from Ti 3+ in TiN to Ti 2+ in TiO. The data reported here define a coherent range within that space.
Analyses with no detectable carbon fall along the TiN-TiO join, implying that their Ti is a mixture of Ti 3+ and Ti 2+ , while those with 1-3 at.% C define a trend parallel to, but slightly above, the Ti 3+ boundary, consistent with solid solution between TiN and Ti 0.75 O. This suggests that f O 2 was in the range ∆IW − 6 to − 8, with the lower value corresponding to the Ti 2 O 3 -TiO buffer. A major cluster of points with ca. 3-10 at.% C (mean 7 at.%) scatters more toward the TiO-TiN join. Analyses with >10 at% C generally correspond to nitrides that coexist with Fe-Ti silicides, and scatter more widely between the Ti 3+ and Ti 2+ boundaries. The apparent higher abundance of Ti 2+ in the higher-carbon nitrides may reflect a decrease in f O 2 during the exsolution of nitride melts from the silicide melts. If so, this would be consistent with the overall trend of f O 2 during the evolution of the Mt Carmel magmatic "system", which ends with the precipitation of the hibonite+grossite+V 0 assemblage (f O 2 = ∆IW ≤ −9 [10,11,22]).
The suggested "end member" phase Ti 2 (N,O,C) 3 described above does not appear in the analysis by [25] (Figure 11); however, Ti 2 N 3 is a common synthetic product used in thin-film surface coatings. The observed example may represent a subsolidus breakdown product of the dominant Ti(N,O,C) phase.
solutions (TixNyOz) with the rock-salt structure. The defect chemistry of this structure allows this wide range of compositions and accommodates a range in the oxidation state of Ti, from Ti 3+ in TiN to Ti 2+ in TiO. The data reported here define a coherent range within that space.
Analyses with no detectable carbon fall along the TiN-TiO join, implying that their Ti is a mixture of Ti 3+ and Ti 2+ , while those with 1-3 at.% C define a trend parallel to, but slightly above, the Ti 3+ boundary, consistent with solid solution between TiN and Ti0.75O. This suggests that fO2 was in the range ΔIW −6 to −8, with the lower value corresponding to the Ti2O3-TiO buffer. A major cluster of points with ca. 3-10 at.% C (mean 7 at.%) scatters more toward the TiO-TiN join. Analyses with >10 at% C generally correspond to nitrides that coexist with Fe-Ti silicides, and scatter more widely between the Ti 3+ and Ti 2+ boundaries. The apparent higher abundance of Ti 2+ in the higher-carbon nitrides may reflect a decrease in fO2 during the exsolution of nitride melts from the silicide melts. If so, this would be consistent with the overall trend of fO2 during the evolution of the Mt Carmel magmatic "system", which ends with the precipitation of the hibonite+grossite+V 0 assemblage (fO2 = ∆IW ≤ −9 [10,11,22]).
The suggested "end member" phase Ti2(N,O,C)3 described above does not appear in the analysis by [25] (Figure 11); however, Ti2N3 is a common synthetic product used in thin-film surface coatings. The observed example may represent a subsolidus breakdown product of the dominant Ti(N,O,C) phase.

Evolution of the Ti(N,O,C) Melts
The petrographic observations (Figures 2-8) indicate that the rapidly growing corundum aggregates trapped three types of mutually immiscible melts: Fe-Ti-C silicide, Ca-Al-Mg silicate, and Ti(N,O,C). The bimodal distribution of C-O contents in the oxynitride melts (Table 1) suggests that some were in at least partial equilibrium with silicide melts, and others with silicate melts. This implies local equilibrium in chemically restricted environments, probably corresponding to melt pockets dominated by either silicide melts or silicate melts. Such local environments would be created by the rapid growth of the surrounding corundum crystals, which resulted in the necking off of larger melt pockets into smaller ones, with the silicide melts being in many cases physically isolated from the silicate melts.
Petrograph observations (Figures 3 and 9 ) also indicate that exsolution of nitride melts from the silicate melts occurred before or simultaneously with the nucleation of the first oxide phases, tistarite (Ti2O3) and Mg-Al spinel. The degree of substitution of Al for Mg in these spinels [12] is consistent with T of 1450-1400 °C, while the crystallization of

Evolution of the Ti(N,O,C) Melts
The petrographic observations (Figures 2-8) indicate that the rapidly growing corundum aggregates trapped three types of mutually immiscible melts: Fe-Ti-C silicide, Ca-Al-Mg silicate, and Ti(N,O,C). The bimodal distribution of C-O contents in the oxynitride melts (Table 1) suggests that some were in at least partial equilibrium with silicide melts, and others with silicate melts. This implies local equilibrium in chemically restricted environments, probably corresponding to melt pockets dominated by either silicide melts or silicate melts. Such local environments would be created by the rapid growth of the surrounding corundum crystals, which resulted in the necking off of larger melt pockets into smaller ones, with the silicide melts being in many cases physically isolated from the silicate melts.
Petrograph observations (Figures 3 and 9) also indicate that exsolution of nitride melts from the silicate melts occurred before or simultaneously with the nucleation of the first oxide phases, tistarite (Ti 2 O 3 ) and Mg-Al spinel. The degree of substitution of Al for Mg in these spinels [12] is consistent with T of 1450-1400 • C, while the crystallization of tistarite requires f O 2 = ca. ∆IW − 7. The conditions for exsolution of nitride melts and crystallization of nitrides from Fe-Ti silicides are less well constrained, but probably lie in the same range [16].
The solubility of N in silicate melts is strongly correlated with pressure up to ca. 3 GPa, but increasing temperature leads to a drop in N solubility [29]. The main factors controlling N solubility in silicate melts at lower pressures are f O 2 and melt composition [30], which affect N speciation. Under oxidizing conditions, N is physically dissolved in silicate melts as N 2 , even in the presence of water. Near the MW buffer, it is present as NH 2+ and NH 3 complexes if water is present, but as N 2 in the absence of water. At and below the f O 2 of the IW buffer, COH fluids are dominated by CH 4 and H 2 , leading to the presence of NH 2groups [30,31]. The solubility of N in silicate melts increases by five orders of magnitude as f O 2 drops from IW to ∆IW − 8 and N becomes chemically bonded in the silicate melt network as N 3− species [32].
The solubility of N in the Fe-Ti-Si-C metallic melts is less well understood, because most experimental work is limited to Fe or Fe-N ± C systems, relevant either to the formation of Earth's core or to steelmaking. Increasing T leads to decreased solubility of N in Fe-rich melts [33]. However, several experimental studies have observed the high-T (ca 1500 • C) exsolution of a "N-rich fluid" from melts in the Fe-N and Fe-C-N systems, and shown that the exsolved melts re-dissolved when T was increased [29,33,34]. The implied increase in N solubility with increasing T may reflect the presence of high levels of carbon in the experiments; higher C or Si in the melt lowers N solubility, as both elements are more siderophile at low f O 2 and in effect "eject" the N as an immiscible N-rich melt.
In these experiments, the solubility of N in the metallic melts was on the order of 1-2 wt% at 1 GPa and increased strongly with increasing P. In contrast to similar experiments on silicate melts, N solubility in the Fe-rich melts decreased markedly as f O 2 was lowered to ∆IW − 3.7. Estimates of D N metal melt/silicate melt at pressures <4 GPa suggest a drop from 10-30 at ∆IW − 1 [29,33] to 2.9 at ∆IW − 2.4 and to 0.5 at ∆IW − 3.3 [35], i.e., N becomes significantly more lithophile as f O 2 decreases below the IW buffer. However, the sparse observations of TiN apparently crystallizing on the rims of Fe-Si melt spherules (Figure 7) suggest the presence of some N in the Fe-Si melts. As cooling at constant f O 2 would only increase the solubility of N in the metallic melt, the expulsion of TiN from the silicide melts can probably be ascribed to the effect of decreasing f O 2 .
These experimental data suggest that when the corundum aggregates at Mt Carmel began to crystallize at f O 2 = ∆IW − 6 to − 7, T ca. 1450 • C and P ca. 1 GPa [12], partition coefficients D N metal melt/silicate melt would have been < < 1, and N would be largely dissolved in the silicate melts rather than in the coexisting metallic melts. In this case, the cause of the immiscibility of the Ti(N,O,C) melts is not obvious, because both declining T and decreasing f O 2 would only lead to higher N solubility in the silicate melts; another trigger appears to be required.
Our modeling of the zoning patterns in the skeletal corundum crystals (Figure 1; [21]) shows a progressive change from rapid crystal growth in an open system, with a moving (i.e., constant composition) melt/fluid, to more sedate growth as the system became less porous, and finally to exponential reduction of melt volumes as individual pockets were isolated from the circulating melt. The corundum aggregates, at least in the early stages, formed a dynamic open network in which the size and shape of void spaces were continually changing as the crystals grew. If nitrogen was being supplied (along with Al 2 O 3 , C and H) by the moving melt/fluid, then the melt pockets may have been accumulating N even as melt volumes shrank, eventually leading to saturation in N and the expulsion of an immiscible N-rich fluid [34]. Martinez and Javoy [36] have pointed out that TiN is the most easily formed nitride, and that N appears to have an affinity for Ti at any mantle T-P-f O 2 conditions. In the Mt Carmel corundum aggregates, Ti phases are abundant, and the trapped melts typically contain several wt% Ti [12,21]. It is therefore expected that saturation of N in the silicate melts could trigger the release of N and the formation of immiscible Ti-N melts. Conversely, the triggering factor may have been saturation of the silicate melts in Ti, leading Ti 3+ to couple with N 3− complexes present in the silicate melt to form Ti-N melts [31].
The emulsions of Ti nitride "tadpoles" (Figures 3, 5 and 8) appear to represent an interaction between nitride and silicate melts. Where these emulsions are in contact with silicate glasses, they could also reflect a late, and probably disequilibrium, process. As noted above, the solubility of N in silicate melts decreases rapidly with decreasing P, especially below 1-2 GPa. This suggests that the formation of the emulsions could reflect the expulsion of N from the silicate melts during the eruption of the host magma. The distribution of the "tadpoles" along the margins of larger Ti nitride inclusions in the silicate glass may indicate that N contents in the silicate melts were higher in the vicinity of the exsolved Ti nitride melts at the beginning of eruption, thus localizing the exsolution of the remaining dissolved N. With modeled time scales on the order of days to years for the evolution of the corundum aggregates [21], equilibrium was probably difficult to attain. Explosive eruption and quenching of the melts would have been very rapid, preventing microstructural equilibration.
The phase diagram for the Ti-N binary ( Figure 11; [26-28]) shows Ti x N 1-x as a liquidus phase at T > 2350 • C (1 atm). This is ca. 1000 • C above the temperatures registered by the metal alloy melts in the corundum aggregates (ca. 1500-1200 • C [16]). We have not found information on the melting points of Ti oxynitrides in the literature. Neumann et al. [37] reported complete miscibility between TiC, TiN and TiO in the solid state from 1100 • C to 1500 • C, but the presence of abundant H in the system at these low-oxygen fugacities can be expected to lower liquidus temperatures substantially. For example, small proportions of H can lower the melting point of Ti by up to 600 • C [38]. Vapor-deposited films of Ti(N,O) are stable under a hydrogen plasma, but show advanced etching at temperatures near 1100 • C [39], which may suggest that the melting point is not a great deal higher.
In the binary system, the Ti 2 N phase found in one TEM-FIB foil (Table S1) can be produced by breakdown of the subsolidus Ti x N y phase below 1100 • C ( [27], Figure 11), and this single observation may suggest the temperatures reached in some parts of the corundum aggregate-melt system before eruption of the host volcanic rocks.

Conclusions
The Ti(N,C,O) phases in the corundum-aggregate xenoliths from the Mt Carmel volcanoes represent immiscible melts that coexisted with mutually immiscible silicate and silicide melts under highly reducing conditions (f O 2 = ∆IW − 6 to − 7). Nitrides with high C/O apparently exsolved from, and/or crystallized from, silicide melts while those with lower C/O coexisted with silicate melts. Exsolution of Ti(N,O,C) from silicide melts was probably driven by falling f O 2 , as N became more lithophile. Immiscibility between nitride and silicate melts may have been driven by scavenging of N from coexisting C-H-O fluids, or by saturation of Ti 3+ as fractional crystallization of corundum reduced the volumes of trapped silicate melts. The P-T-f O 2 conditions reflected in the Mt Carmel xenolith assemblages appear to reflect interaction between small volumes of differentiated mafic melts and CH 4 -H 2 fluids derived from a metal-saturated mantle. Similar assemblages have been recorded from other examples of explosive volcanism [9,20] and provide insights into the behavior of nitrogen under reducing conditions in the deeper mantle.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/min11070780/s1, Figure S1: Maps showing the regional setting of the Mt Carmel area (a) and the simplified geology with locations of Cretaceous primary volcanic vents; Figure S2: Selected area electron diffraction patterns from two zone axes obtained by tilting the same grain of the titanium nitride phase; Table S1: EMP analyses of Ti oxynitrides; Video S1: #D-µCT image of a slice through a corundum aggregate.