Formation Conditions of Extremely Reduced High-Temperature Mineral Assemblages in Rocks of Combustion Metamorphic Complexes (Mongolia)

1. Introduction

The East Mongolian coal province in central and eastern Mongolia accommodates the Nyalga and Khamaryn–Khyral–Khiid combustion metamorphic (CM) complexes consisting of thermally metamorphosed and partly melted rocks produced by multi-stage Quaternary and modern wild fires of coal stored in Early Cretaceous sediments. We discovered the two complexes in 2016–2018 and have been studying them since then. The history of discoveries and an inventory of the available data on local geology, lithology, and mineralogy of main rock types were reported in our previous publications [1,2,3,4,5,6,7,8].

The Nyalga complex consists of more or less strongly thermally altered pelitic sediments and products of their partial melting (paralavas and vitreous clinker) exposed in three ≤ 40 m high erosion outliers [1,2]. Melilite–nepheline paralava containing minerals of the rhönite–kuratite series occurs as scattered rock fragments or single blocks on the outliers or on eroded surfaces nearby [1]. Clinker and paralavas in the area often enclose xenoliths of burnt sedimentary rocks. Signatures of CM alteration in sediments were also found in the Tugrug coal quarry located 4–5 km southeast of the outliers [2].

The territory of the Khamaryn–Khyral–Khiid complex has a rugged topography typical of coal fire areas, with long ridges, numerous gullies, fissures, and caves produced by weathering and collapse of overburden above combusted coal seams [3]. No active coal combustion has been observed lately though a focus of underground smoldering remains in the zone of the last coal fire of 1932–1947, where erosion stripped warm CM rocks.

The CM rocks are remarkable by high diversity of mineral assemblages due to the lithology of the sedimentary protoliths and local conditions of multi-stage thermal alteration and partly melting. The two complexes consist of predominant CM pelitic sediments and vitreous clinker of various compositions, as well as mineralogically exceptional melilite–nepheline paralava. The paralava mineralogy includes basic plagioclase, clinopyroxene, and melilite in phenocrysts, with interstitial nepheline and high-silica potassic glass that encloses olivine group minerals (Ca-bearing fayalite, kirschsteinite), spinel, K-Ba feldspars (celsian, hyalophane), and other phases, including very rare ones [1,2,3,4,5]. The CM complexes also contain much less abundant paralavas of plagioclase–pyroxene ± indialite (Naylga complex), sekaninaite–tridymite, and cristobalite–fayalite (Khamaryn–Khyral–Khiid complex) main mineral compositions.

The Khamaryn–Khyral–Khiid CM rocks host remnant marly limestone in which calcite and silicate minerals melted successively at elevated CO₂ partial pressure and T > 1250 °C, with formation of carbonate and Ca-rich silicate mafic melts crystallized then into melilite–nepheline paralava [6]. The Khamaryn–Khyral–Khiid cristobalite clinker and tridymite–sekaninaite and cristobalite–fayalite paralava compositions were formed by partial melting of pelites in a large temperature interval [4]. Mullite crystallized in abundance at T > 850°С after dehydration-dehydroxylation and incongruent melting of amorphous pelitic material. The sedimentary overburden above coal fire foci heated up to T > 1050°С (as inferred from crystallization temperatures of cordierite-group minerals), or locally even to T > 1470°С required for melting of detrital quartz and possibly corresponding to the stability field of β-cristobalite [5].

The studies of CM rocks in the area continue. We report new data concerning unusual high-temperature reduced mineral assemblages found sporadically in the Naylga and Khamaryn–Khyral–Khiid complexes. Some melilite–nepheline paralava and clinker samples bear troilite (FeS), metallic iron (Fe⁰), kamacite (Ni-bearing Fe⁰), taenite (Ni-rich Fe⁰), iron phosphides (barringerite or allobogdanite Fe₂P, schreibersite Fe₃P, and steadite Fe₄P = eutectic α-Fe + Fe₃P), as well as wüstite (FeO), and cohenite (Fe₃C). Raman spectroscopy of the constituent minerals revealed schreibersite, cohenite, wüstite, magnesiowüstite–ferropericlase, magnetite, minerals of pyroxene, olivine, cordierite, and spinel groups, rhönite–kuratite series, melilite, nepheline, cristobalite, troilite, pyrrhotite, marcasite, goethite, ferropseudobrookite-pseudobrookite, and rutile.

The crystallization conditions of magnesiowüstite–ferropericlase, Fe⁰, iron monosulfide with Fe⁰ inclusions, wüstite + magnetite, iron phosphides (Fe₂P, Fe₃P, and Fe₄P) + Fe₃C were reconstructed using the Fe–S, MgO–FeO, MgO–FeO–Al₂O₃, Fe–P, Fe–P–C, and Fe–S–P phase diagrams. The obtained temperature estimates were used to infer the redox conditions with respect to the CCO, IW (Iron–Wüstite), IM (Iron–Magnetite), and MW (Magnetite–Wüstite) buffer equilibria during melting of marly limestone and crystallization of melilite–nepheline paralava melts. The origin of the analyzed СМ rocks was explained in a model implying thermal alteration in locally reducing conditions of low-permeable sedimentary overburden above foci of wild subsurface coal fires.

2. Materials and Methods

The collected samples were analyzed at the Center for Isotope-Geochemical Studies of the Vinogradov Institute of Geochemistry, Siberian Branch of the Russian Academy of Sciences (IGC, Irkutsk). Mineralogy and phase compositions were investigated by scanning electron microscopy coupled with energy-dispersive spectrometry (SEM EDS) on a Tescan Mira 3 LMU microscope with an ULTIM MAX 40 microanalysis system (Oxford Instruments). The analyses were performed in a scanning mode on carbon-coated polished cut surfaces of samples, at 1 to and over > 10 μm² spots, 20 kV accelerating voltage, 1 nA beam current, and 30 s acquisition time (excluding dead time). Carbon in cohenite was identified in some gold-sputtered samples. The data quality was checked against analyses of reference samples, including quartz, albite, orthoclase, wollastonite, MgO, Al₂O₃, CaF₂, metals, and synthetic compounds from sets 6316 and 7682 of the reference standards by Microanalysis Consultants Limited (Oxford Instruments Ltd., UK). All the calculations and matrix corrections were conducted automatically in AZtec (software of Oxford Instruments Nanoanalysis, UK). The detection limits of major elements were 0.2–0.3 wt%. Аverage random errors varied from 0.9 tp 13 rel% depending on major-element contents: ± 0.9 rel% at > 10 wt%, ± 3.0 rel% at 1–10 wt%, and ± 13 rel% at 0.3–1 wt% [9]. The crystal chemical formulas were calculated using the Crystal software, ver. 3.0 [10].

The Raman spectra were collected on a WITec confocal Alpha 300R (WITec GmbH) spectrometer with 532 nm Nd:YAG laser (power 0.5–17 mW), at the Analytical Center for Geodynamics and Geochronology in the Institute of the Earth’s Crust (Irkutsk). The 1–2 μm laser beam was focused using a 100x Carl Zeiss lens. The range of the used Raman shift regions was from 100–3600 cm⁻¹. The data was calibrated against a silicon peak of 520.7 cm⁻¹ at an optical resolution of 1800 grooves/mm and 600 grooves/mm groove densities. Spectral acquisition time was from 1 to 10 s. Total spectra with 20–100 (up to 5400) accumulations were collected for each measurement. The spectra were processed without baseline correction in the ArDI system of mineral vibrational spectroscopy data processing and analysis [11]. Mineral phases were identified with reference to the RRUFF database (http://rruff.info/) [12] (last accessed on 15 October 2024) and to published data.

3. Results

3.1. Identification of Minerals

Minerals in > 2 µm inclusions were identified from Raman spectra and SEM-EDS data. The previously obtained Raman spectra for a number of rare phases from CM rocks in Mongolia revealed minerals of several groups, including rhönite–kuratite [1], cordierite (orthorhombic cordierite and sekaninaite, hexagonal indialite and ferroindialite), silica polymorphs (cristobalite and tridymite), and spinel-group minerals (including silicate spinel), as well as mullite and Fe-rich mullite [3,4,5]. Raman spectra were also used to identify iron sulfides, iron oxides, and several silicate minerals (see some examples in Figure 1).

It was problematic to discriminate between pyrrhotite Fe_1-x S _{(x= 0-0.17)} and troilite FeS iron monosulfides which have minor difference in iron and sulfur contents (62.33 against 63.53 wt% Fe and 37.67 against 36.47 wt% S, respectively). The Raman spectra of iron monosulfides were acquired at different laser power values. The monosulfide surface was heated at laser power increasing from 1 to 17 mW, with successive oxidation FeS + O₂ → Fe₃O₄ + SO₂; FeO + O₂ → Fe₂O₃ + SO₂ in the laser-exposed zone [13]. As a result, the spectra showed peaks of magnetite and hematite from an oxide film formed on the monosulfide surface (Figure 2).

Similar shapes and trends in the Raman spectra changing as a function of laser power and temperature are typical of many iron-rich sulfide and oxide phases, such as pyrite, magnetite, goethite, etc. [13], as well as schreibersite (our data). Therefore, the Raman spectroscopy was performed at the lowest laser power and prolonged acquisition time from 10 minutes to 3 hours. However, the Raman spectra of troilite and pyrrhotite were poorly discriminable even at those operation conditions because of weak peaks in the characteristic region of 100 to 400 cm^-1. The peaks of troilite are known to be small in the 148–331 cm^-1 interval [13] while those of pyrrhotite are relatively high at 314 and 372 cm^-1 (RRUFF, [12]), but the spectra we collected lack these bands (Figure 2b). Thus, the iron monosulfide grains in the samples studied by Raman spectroscopy must be troilite.

Iron sulfides can be successfully identified using the Fe–S phase diagram. The Fe–S system at ambient pressure has been largely studied, but the available phase diagrams differ significantly, especially in the T < 400 °C region [14,15] and references therein]. Best consistency was achieved in the diagram based on thermodynamic modeling and experimental evidence (Figure 3). In addition to the FeS and FeS₂ phases, the Fe–S system produce several non-stoichiometric phases with the common formula Fe_1–хS. At P = 1 bar and T > 1000 °C, the phase diagram displays fields of an Fe–S melt, face-centered cubic (fcc) iron (γ-Fe), body-centered-cubic (bcc) iron (α-Fe), and hexagonal pyrrhotite. The latter is a hexagonal monosulfide having an NiAs structure and non-stoichiometric excess of S (Fe_1−xS), which melts congruently at ~1190 °C. Pyrite (FeS₂), with its highest S enrichment in the Fe–S system, undergoes incongruent melting at 743 °C to produce pyrrhotite Fe_1–хS and an almost pure S melt. The compositions between stoichiometric troilite (FeS) and monoclinic pyrrhotite (Fe₇S₈) were constrained only qualitatively in terms of phase relations at T < 320 °C. Troilite is a FeS polymorph stable at T < 140 °C and rarely found in terrestrial rocks. The iron monosulfide known from mineral assemblages with Fe⁰ in stony-iron and iron meteorites is also commonly referred as troilite. The troilite converts into hexagonal pyrrhotite at temperatures above 140 °C.

The mineral assemblages of melilite–nepheline paralava in the sampled CM complexes of Mongolia typically include a phase with its composition close or corresponding to FeS [2,3]. Proceeding from the Fe–S phase diagram (Figure 3) and the Raman spectra (Figure 2b), troilite in this study refers to iron monosulfide grains, globules, and spherules associated with Fe⁰, while other monosulfide phase are identified as pyrrhotite.

Raman spectroscopy data on iron phosphides, including schreibersite (Fe₃P), barringerite or allobogdanite (Fe₂P), and steadite (Fe₄P = eutectic α-Fe + Fe₃P), are absent or poorly consistent [16]. Only schreibersite can be identified from its Raman spectrum similar to that in the RRUFF database. At a small laser power, the spectra of Fe₃P contain several peaks in the 100–200 cm^-1 region and a typical trough between 400 and 700 cm^-1 (Figure 4a). The phase identification of other phosphides (Fe₂P and Fe₄P) was based on the Fe/P ratio from SEM-EDS data.

Cohenite Fe₃C was identified in several 2–4 µm inclusions by Raman spectroscopy (Figure 4b) and SEM-EDS analysis of gold-sputtered samples (see below). The Raman spectrum of Fe/Fe₃C shows D and G bands at 1330–1560 cm⁻¹ which correspond to structural defects/disorder and are associated with first-order scattering of carbon domains characteristic of graphite-like materials. The presence of Fe₃C is also detectable from several peaks between 100 and 300 cm⁻¹ usually attributed to stretching vibration modes between metal and carbon atoms [17].

Kamacite (Ni-bearing α-Fe), taenite (Ni-rich α-Fe), wüstite, and rare sulfides, such as ninengerite (Mn,Fe)S, cubanite CuFe₂S₃, nukundamite Сu₂Fe₂S₄, and bornite Cu₅FeS₄, were provisionally identified according to their compositions and related stoichiometric formulas.

3.2. Reduced Mineral Assemblages in Rocks of the Nyalga CM Complex

The mineralogy of melilite–nepheline and plagioclase–pyroxene ± indialite paralavas in the Nyalga complex was described in our previous publications [1,2]. The matrix of the Nyalga paralavas often encloses iron monosulfide, but small spherules and round grains of troilite with Fe⁰ inclusions were found only in three out of twenty five samples. Some samples of melilite–nepheline paralava also host coarse (up to 3×5 mm) grains of troilite with Fe⁰ inclusions and veinlets filled with wüstite, magnetite, and goethite. Troilite in those samples was identified according to Raman spectra, as well as proceeding from the association with Fe⁰ and with replacing wüstite and magnetite.

Melilite–nepheline paralava in sample MN-1287 bears a coarse troilite grain cut by numerous symplectitic zones with inclusions of wüstite replaced by pure stoichiometric magnetite (Figure 5). The Raman spectra of wüstite differ from those of magnetite in smaller peaks at ~550 and ~670 cm^-1 in the 500–700 cm^-1 range (Figure 6a). In zones near cracks, wüstite is almost fully replaced by magnetite while the latter, in turn, becomes replaced by goethite (see its Raman spectrum in Figure 1e). Troilite near cracks in rims is often replaced by FeS₂ sulfide which has its Raman spectrum similar to marcasite (Figure 1d, Figure 5b,c,e). Symplectitic zones sometimes enclose goethite and (Mg,Mn)-bearing siderite (Figure 5f).

The exposed surface of a large troilite grain in melilite–nepheline paralava sample MN-1133 displays sporadic inclusions of Fe⁰ rimmed with stoichiometric magnetite. Numerous cracks in the grain are filled with wüstite which is replaced by magnetite along undeformed zones (Figure 7a-d). In cracks, magnetite is partly or fully replaced by goethite, while troilite is partly replaced by marcasite (Figure 7a); iron sulfate (calculated per formula Fe₂SO₄) occurs near some cracks.

The matrix of sample MN-1133 is locally composed of a mineral assemblage unusual for melilite–nepheline paralava, with grains of spinel (Mg,Fe)Al₂O₄ and forsterite, as well as Fe-Mg oxide, up to 15–20 µm in the large dimension, and with disseminated magnesioferrite (Mg,Fe²⁺)(Fe³⁺)₂O₄ at grain boundaries or in the grain interior (Figure 7e-g). The Fe-Mg oxide grains contain 42–74 wt% FeO and 59–25% МgO; the total reaches 100 wt%, and the FeO/(FeO + MgO) ratio is 0.42–0.74, or 0.3–0.6 apfu Fe²⁺ as calculated per solid solution (Fe²⁺,Mg)O of wüstite–periclase or magnesiowüstite–ferropericlase (Figure 7e-g, Table 1, an. 1, 2). Fe-Mg oxide of this composition does not fit the formula of magnesioferrite; it appears as a crystalline phase in its Raman spectrum (Figure 6b). The spectra show strong peaks in the 100–200 cm^-1 range and two gentle peaks at ~600 and ~1200 cm^-1. Note that the characteristic peaks in the spectra of wüstite and magnetite from inclusions in symplectitic zones hosted by troilite in sample MN-1287 likewise fall within the 500–700 cm^-1 range (Figure 6).

3.3. Reduced Mineral Assemblages in Rocks of the Khamaryn–Khural–Khiid Complex

The analyzed samples from the Khamaryn–Khural–Khiid complex were melilite–nepheline and ferroan (tridymite–sekaninaite and cristobalite–fayalite) paralavas, as well as crystobalite clinker and vitreous rocks produced by partial melting of pelitic rocks [3,4,5].

Sample MN-1417 is mostly glass that hosts numerous xenoliths of vitreous clinker and molten pelite (possibly, mudstone), abundant partially dissolved or fused grains of detrital quartz, and rare grains of zircon and ferropseudobrookite–pseudobrookite with rutile inclusions (Figure 8). Quartz grains are densely cut by veinlets filled with cristobalite. Other inclusions in glass are numerous clusters of submicrometer magnetite, small acicular mullite and Fe-rich mullite (< 2–3 µm in the long dimension), and round globules of metallic iron or troilite with Fe⁰ inclusions, as well as zoned spherules, up to 100–150 µm in diameter (Figure 8 and Figure 9).

The spherules have a troilitic rim with sporadic inclusions of Fe⁰ and Fe₃P phases, a middle zone of schreibersite with inclusions of Fe⁰, cohenite, and steadite, and a core with a round Fe⁰ globule enclosing cohenite and schreibersite (Figure 9, Table 1). Cohenite either occurs as isolated round inclusions or is enclosed together with schreibersite or Fe⁰, the phases being separated by a distinct boundary. The elemental maps and EDS spectra of cohenite (Figure 10) obtained for several zoned spherules confirmed a high concentration of carbon up to 8 wt% (e.g., 6.7 wt% C in Fe₃C: Table 1, an. 9). Meanwhile, minor amounts of carbon (~2–3 wt%) and oxygen (up to 1 wt%) were found in all phases of the spherules. Schreibersite also contains up to 2 wt% Ni and 0.6–2.6 wt% Co; Fe⁰ inclusions bear ≤ 2–3 wt% P. The middle zone in some spherules is composed of Ni-rich (10–20 wt% Ni) schreibersite with inclusions of kamacite (α-Fe, 5–7 wt% Ni) and taenite (α-Fe, 10–11 wt% Ni) with ≤ 1 wt% Cu, as well as Cu sulfide compositionally similar to nukundamite Сu₂Fe₂S₄.

Sample MN-1420 of melilite–nepheline paralava bears disseminated Fe⁰, fine round inclusions consisting of schreibersite, Fe⁰, and troilite (Figure 11) in the core of elongate basic plagioclase grains, as well as inclusions of pyrrhotite (or troilite) and perovskite intergrown with a Fe-Mn sulfide corresponding to a Fe-dominant analogue of ninengerite (Mg,Fe)S (Figure 11, Table 1, an. 5). The paralava also hosts pelitic xenoliths (mudstone baked to vitreous clinker) composed of K-rich aluminosilicate glass which encloses < 2 µm grains of cristobalite, indialite, mullite, and relatively coarse (20–40 µm) isometric cordierites. The latter sometimes have euhedral contours and, in turn, host submicrometer cristobalite and mullite inclusions. Besides these phases, the glass bears spherules of troilite, Fe⁰, and < 1 to 5 µm round two-phase inclusions of Fe⁰ + Fe₃P or FeS + Fe₃P.

Another sample MN-1412 of melilite–nepheline paralava contains up to 100 µm spherules of Fe⁰, intergrown Fe⁰ and troilite, and Fe⁰ inclusions in troilite (Figure 12a-d). Many phenocrysts of plagioclase enclose ≤ 5 µm troilite ± Fe₃P, troilite + taenite (Ni-rich α-Fe, 18–21 wt% Ni, Table 1, an. 8), and sporadic Fe₂P inclusions in the core (Figure 12e-g). Some coarse plagioclase phenocrysts locally contain spinel microlites coexisting with the Fe₂P phase (Figure 12h). Iron phosphides in some inclusions bear 0.4–1.1 wt% Ni and ≤ 0.8 wt% Co. Interstitial troilite sometimes enclose Cu sulfide inclusions and veinlets correspond to the cubanite CuFe₂S₃ (Figure 12i, Table 1, an. 4).

Sample MN-1406 of plagioclase–orthopyroxene paralava encloses xenoliths of vitreous clinker derived from mudstone, with numerous acicular microlites of mullite, orthorhombic pyroxene (enstatite–ferrosilite), and euhedral cordierite-sekaninaite microlites (Figure 13). Glass in the xenoliths hosts small spherules of troilite, one with a round zone of schreibersite and Fe⁰ and some with Fe⁰ inclusions and a discontinuous chalcopyrite rim (Figure 13d-f). The spherules also enclose submicrometer particles of intergrown non-identified sulfides Cu along the spherule-glass boundaries. The sulfides have large ranges of Cu (50–70 wt%), Fe, and S, and some approach bornite compositions.

4. Discussion

4.1. Formation of Melilite-Nepheline Paralava: Conditions and Processes

Earlier we [3,6] suggested a qualitative petrological model explaining the formation of the Khamaryn–Khyral–Khiid melilite–nepheline paralava and carbonate melt by melting of marly limestone. The model implied crystallization of a gehlenite + Al-clinopyroxene assemblage in marly limestone layers during high-temperature low-pressure combustion metamorphism of sediments as a result of wild underground coal fires. The thermal alteration occurred at relatively high partial pressure of CO₂ in unique P-T conditions without decomposition of calcite into CaO and CO₂ [6].

Melting of K-Na feldspar in the silicate (pelitic) component of marly limestone occurred at T > 1100 °С and P_CO2 > 40 bar, while calcite, gehlenite, Al-clinopyroxene, and other refractory minerals melted incongruently at T > 1250 °С, P_CO2 > 40 bar. The melts comprised immiscible carbonate (non-stoichiometric CaCO₃ + CaO or stoichiometric calcitic) and several compositionally diverse silicate melts [6]. Coalescence of melts derived from silicate minerals led to the formation of a liquid undersaturated with respect to Si and enriched in Ca which became the parent melt of melilite–nepheline paralava. The paralava matrix locally preserves gehlenite grains with rims of newly formed ferroan åkermanite–alumoåkermanite and remnant marly limestone xenoliths composed of gehlenite, monticellite–kirschsteinite, forsterite, perovskite, calcite, sulfates, and other rare phases, in different relative percentages, these assemblages being exotic for melilite–nepheline paralava [2,3,6].

As indicated by petrography and SEM-EDS data, the sequence of crystallization from the parent melts of melilite–nepheline paralava was as follows: (1) spinel-group microlites with large ranges of Fe²⁺, Fe³⁺, Al, and Mg, as well as phosphide Fe₂P in some paralava varieties [3]; (2) a phenocryst assemblage of bytownite–labradorite, ferroan åkermanite–alumoåkermanite, and Al-clinopyroxene (diopside with a high percentage of the kushiroite end-member); (3) interstitial phases of olivine (Ca-bearing fayalite, kirschsteinite), K-Ba feldspar (celsian, hyalophane), rhönite–kuratite series, fluorapatite, spinel, pyrrhotite or troilite, Na-K nepheline (Nyalga complex) or Na-Ca nepheline (davidsmithite, Khamaryn–Khyral–Khiid complex), kalsilite, barite, and other minerals, including very rare ones [2,3,5].

The data on mineral assemblages in the melilite–nepheline paralava of the two CM complexes in Mongolia allowed estimating the crystallization temperatures of the rhönite–kuratite series (~1100 °C) and the temperature range for exsolution of Fe-Ca olivine into Ca-bearing fayalite and kirschsteinite (1000–800 °C) at oxygen fugacity varying between the IW and FMQ (Fayalite–Magnetite–Quartz) buffers [2,3].

The samples of melilite–nepheline paralava (MN-1412 and MN-1420) enclose iron phosphides: Fe₂P + Fe₃P in the early growth zone of anorthite–bytownite phenocrysts and schreibersite in aluminosilicate glass produced by vitrification of melt during the melting of clinker xenoliths in the paralava of samples MN-1420 and MN-1412 (Figure 11 and Figure 12). The clinker xenoliths in sample MN-1420 contain mullite, indialite, and cristobalite due to a high melt temperature which possibly reached the stability field of β-cristobalite at T > 1470 °C [5].

The findings of reduced mineral assemblages with metallic iron, iron phosphides, magnesiowüstite–ferropericlase, and wüstite provide direct evidence of high-temperature extreme redox conditions at a certain crystallization stage of silica-undersaturated Ca-rich mafic melts (parent to melilite–nepheline paralavas) and vitrification of felsic aluminosilicate melts in clinker. Below we discuss possible processes and conditions responsible for the presence of reduced assemblages in the studied CM rocks.

4.2. (Mg,Fe)O, MgO–FeO Solid Solution

The ferropericlase–magnesiowüstite solid solution is an essential component of the Earth's lower mantle [18] where phase change of ringwoodite (Mg,Fe)₂SiO₄ (silicate spinel) at high pressure leads to the formation of perovskite (Mg,Fe)SiO₃ and ferropericlase (Mg,Fe)O. The unusually large range of iron content in ferropericlase (Mg# 12–93) from diamond inclusions is due to its origin at mantle depths of 410 to 660 km [19]. Considering that periclase with < 5 wt% FeO is of rare occurrence in the Earth’s crust and proceeding from analysis of crystal structure, the solid solution between MgO and FeO is expected to be unstable in the low-pressure crustal conditions [20].

The Mg-Fe oxide with variable FeO and MgO contents appears in the Raman spectra as a crystalline compound (Figure 6b) within the joint between wüstite and periclase. The presence of the exotic magnesiowüstite–ferropericlase + spinel (Mg,Fe)Al₂O₄ + forsterite assemblage (Figure 7e-g) in melilite–nepheline paralava hardly can be attributed to high-pressure formation conditions of mantle rocks but rather must result from melting of carbonate-silicate sedimentary protoliths. According to our qualitative petrological model [6], melilite–nepheline paralava was produced by incongruent melting of marly limestone. This prediction agrees with the CaO–CaCO₃ phase diagram in the low-pressure region implying melting of calcite [21], as well as with the abundance of paralava-hosted calcite melt inclusions in mineral phenocrysts and sedimentary xenoliths composed of spinel, forsterite, monticellite, perovskite, calcite, and other phases [2,3]. The Nyalga melilite–nepheline paralava likewise bears remnant sedimentary xenoliths consisting of magnesioferrite, larnite, baritocelestine, sulfates, gehlenite, perovskite, fayalite, and periclase with 10–15 wt% FeO. The parts of paralava matrix with magnesiowüstite–ferropericlase grains in association with spinel and forsterite (Figure 7e-g) are likewise remnants of sedimentary xenolith or ferroan sinter produced by melting of ash after coal combustion.

Published evidence from metallurgical production experience shows that Fe, Mg, and Al oxides enter many sinter and refractory materials used for lining up steel-smelting furnaces, and variations of their contents can affect notably the physical properties (melting point, viscosity, fluidity, etc.) of the lining materials. The FeO–MgO phase diagram has been well constrained experimentally and correlated with thermodynamic calculations [22,23,24]. Both liquid and solid oxides in the FeO–MgO system have unlimited solubility (Figure 14). In reducing conditions at atmospheric pressure, the FeO–MgO–Al₂O₃ phase diagram includes regions of equilibrium between an oxide melt and the magnesiowüstite–ferropericlase and spinel (Mg,Fe)Al₂O₄ solid solutions, while the solubility in MgO is unlimited for FeO and limited for Al₂O₃ (Figure 14).

The FeO–MgO and FeO–MgO–Al₂O₃ phase diagrams show that magnesiowüstite–ferropericlase and spinel in melilite–nepheline paralava sample MN-1133 may have melted (or crystallized) in a reducing environment at 1800–1600 °C. The temperature of underground coal fires not necessarily was that high though, because the conditions and processes during low-pressure melting of sediments have multiple controls, such as compositions of coal and sedimentary protoliths, fluid regime, combustion kinetics, etc. Meanwhile, the mineral assemblage of magnesiowüstite–ferropericlase + spinel + forsterite was formed at low oxygen fugacity and at temperatures higher than for any other known paralava assemblage, possibly, at the fire focus. Note also that magnesiowüstite–ferropericlase grains became replaced by magnesioferrite while the paralava was cooling down or weathered in an oxidizing environment (Figure 7e-g).

4.3. Troilite with Inclusion of Metallic Iron and Pyrrhotite

Melilite–nepheline paralava and vitreous clinker fragments often contain globules or spherules of iron monosulfide (pyrrhotite or less often troilite) with inclusions of metallic iron. High-temperature iron monosulfide crystallized from droplets of Fe–S melt occurring in the silicate melt of paralava.

As follows from the Fe–S phase diagram (Figure 3), pyrrhotite grains (free from Fe⁰ inclusions) may have crystallized from an Fe–S melt with mole fraction 0.54–0.50 S which existed at temperatures above 1100 °C. A melt with a lower S fraction (0.50–0.43) can precipitate pyrrhotite upon cooling below 1190 °C while γ-Fe forms at 988 °C at the eutectic point. That was probably the case for a large pyrrhotite grain from sample MN-1287 which acquired symplectitic zones with inclusions of wüstite and magnetite (Figure 5). The two minerals probably resulted from oxidation of metallic Fe originally present in the inclusions, first to wüstite and then partly to magnetite at increasing oxygen fugacity (Figure 5c,e). Upon cooling to 140 °C, pyrrhotite converted into the stable phase of troilite which became partly replaced by marcasite along cracks during weathering (Figure 3 and Figure 5).

Judging by the texture of the paralava and the phase diagram (Figure 3), the troilite globules with Fe⁰ inclusions crystallized from a Fe–S melt with a low mole fraction of S (< 0.43). The crystallization of iron was followed by that of high-temperature pyrrhotite at 988 °C, which then converted to troilite upon cooling at 140 °C.

4.4. Mineral Assemblages with Iron Phosphides

Iron phosphides (± Ni-bearing) are especially common to mineral assemblages of meteorites but are of rare terrestrial occurrence restricted to CM rocks and some volcanics, e.g., in Disko Island, Greenland [25,26]. Recently structural features of iron phosphides (barringerite, schreibersite, murashkoite FeP, zuktamrurite FeP₂) and Ni–Fe–Mo–Cr phosphides (nickelphosphide (Ni,P)₃P, allobogdanite (Fe,Ni)₂P, orishchinite Ni₂P, transjordanite Ni₂P, negevite NiP₂, halamishite Ni₅P₄, nazarovite Ni₁₂P₅, andreyivanovite FeCrP, nickolayite FeMoP, polekhovskyte MoNiP₂) have been discovered in the Hatrurim Formation (Israel and Ohman) and documented in more or less detail [27,28,29,30,31,32,33,34,35,36,37]. However, the Hatrurim phosphide-bearing CM rocks remain very poorly studied in terms of petrology.

In this respect, it is especially interesting to reconstruct the crystallization conditions of iron phosphides (schreibersite and Fe₂P) found in melilite–nepheline paralava samples MN-1420 and MN-1412 (Figure 11 and Figure 12), as well as in vitreous zones of clinker xenoliths from paralava samples MN-1406 and MN-1417 (Figure 8, Figure 9 and Figure 10 and 13). Note that allobogdanite was previously considered as a high-pressure orthorhombic polymorph (Fe,Ni)₂P [28,38] before it was found out that Ni-free Fe₂P also crystallized in a orthorhombic structure at ambient pressure [39] and the allobogdanite to hexagonal barringerite conversion occurred at ~600 °С [40].

The possible conditions and sequence of crystallization of iron phosphides (Fe₃P and Fe₂P) from inclusions in the core (early growth zone) of many basic plagioclase phenocrysts from melilite-nepheline paralava samples MN-1420 and MN-1412 can be estimated using the Fe–P phase diagram which has been well studied topologically (Figure 15). Plagioclase phenocrysts mostly bear schreibersite, as isolated inclusions or intergrown with troilite (crystallized drops) or Fe⁰, while Fe₂P occurs in a few inclusions (Figure 11c and Figure 12e-h).

According to the Fe–P phase diagram at ambient pressure, the Fe–P melt containing 14–22 wt% P first precipitates the Fe₂P phase at T < 1370 °C, then Fe₃P forms by the peritectic reaction Fe–P melt + Fe₂P at 1165 °C and exists till the equilibrium eutectic phase of Fe₄P (steadite = α-Fe + Fe₃P) crystallizes at 1048 °C (Figure 15).

Thus, the cooling parent melt of paralava apparently captured different inclusions at decreasing temperatures (Figure 15, Paths I and II): Fe₂P at 1370–1165 °C, sometimes together with Al-spinel microlites occurring in the core of plagioclase phenocrysts from paralava samples MN-1420 (Figure 11c) and MN-1412 (Figure 12e-h), and Fe₃P at 1165 °C to 1048 °C. Sporadic inclusions of Fe₄P (steadite) are apparently a eutectic mixture of submicrometer Fe⁰ and Fe₃P phases stable below 1048 °C, which are undetectable by SEM-EDS because of small size.

Phase relationships are more complex in zoned spherules consisting of troilite, Fe⁰ or kamacite, taneite, schreibersite ± (steadite and cohenite) found in melt rock sample MN-1417 (Figure 9 and Figure 10). The crystallization conditions and sequence can be inferred from their texture using the Fe–S, Fe–P–C, and Fe–P–S phase diagrams.

The possible formation conditions of troilite phenocrysts and spherules with inclusions of Fe⁰ in the vitreous zones of sample MN-1417 were estimated using the Fe–P phase diagram (Figure 15). Troilite in other coarser zoned spherules commonly rims the phases of Fe⁰, schreibersite ± (cohenite and steadite) present in different volume percentages (Figure 9).

Different phases in the zoned spherules may have followed two possible models of crystallization from a silicate melt which contained either (1) droplets of a homogeneous Fe–S–P–C melt decomposed into immiscible Fe–S and Fe–P–C liquids upon cooling or (2) droplets of Fe–P–S–C, Fe–S, Fe–P–C, and Fe–P–S–C liquids coalesced partly into larger homogeneous drops with different ratios of S, Fe, P and C or into drops consisting of two immiscible melts. Model 2 appears more realistic because inclusions, globules and spherules show large variations of volume phase relations among troilite, Fe⁰, Fe₃P, and Fe₃C.

The Fe–S–P phase diagram was studied previously [42,43,44] at oxygen fugacity corresponding to the IW and QFI (Quartz–Fayalite–Iron) buffers equilibria. It comprised fields of a homogeneous Fe–S–P melt and a region of immiscibility between a Fe–S (sulfide) and Fe–P (phosphide) melts (Figure 16).

As follows from the phase diagram and published data [42,43,44], the homogeneous Fe–S–P melt in the spherules could exist at ~1350 °С and decomposed into immiscible Fe–P and Fe–S melts in the ~1350–1250°С range. In our case, the Fe–S–P melt contained minor amounts of carbon which complicated the relationships of crystallized phases and could shift the phase change temperatures.

After the homogeneous melt had decomposed upon cooling, the spherules accommodated two immiscible liquids: a dense Fe–P–C melt in the core and a less dense Fe–S melt in the rim. The phases crystallized from the two melts were, respectively, high-temperature pyrrhotitic (4.61 g/cm³ on average) and several higher-density phases: Fe⁰ (7.87 g/cm³), Fe₃P (6.74 g/cm³) ± (Fe₄P + Fe₃C). In the course of further cooling, the melts evolved in accordance with the Fe–S and Fe–P–C phase relations. Metallic iron was the first to crystallize, judging by the texture of spherules (Figure 9), which is possible in a iron-saturated Fe–P–C melt with different relative contents of P and С above the eutectic point at T > 952 °C, as one may infer from the place of the liquidus surface in the Fe–P–C system (Figure 17). Paths 1 and 2 in Figure 17 show composition changes in the Fe–P–C melt from the α-Fe (bbc) or γ-Fe (fcc) fields, with the following crystallization sequence: C-bearing α-Fe or γ-Fe → cotectic α-Fe or γ-Fe + Fe₃P → eutectic (E_T) γ-Fe + Fe₃P + Fe₃C.

In some spherules, the schreibersite zone with inclusions of Fe⁰ ± Fe₃C breaks through the troilite rim (Figure 4 and Figure 9d,g). Therefore, the Fe–P–С parent melt for this zone was liquid (mobile) after the crystallization of high-temperature pyrrhotite in the spherule rim (at T < 988 °C according to the Fe–S phase diagram, Figure 3). The texture features of the spherules are consistent with the crystallization sequence along Path 3 in the Fe–P–C phase diagram: C-bearing γ-Fe → cotectic γ-Fe + Fe₃C → eutectic (E_T) at 952 °C (Figure 17). Thus, the final crystallization stage in the zoned spherules may have occurred in the 988–952 °C range while the large drops of homogeneous Fe–P–S–C melt and immiscible Fe–S and Fe–P melts the could coexist at a temperature as high as 1350 °C (Figure 15, Figure 16 and Figure 17).

4.5. Formation Conditions of Reduced Mineral Assemblages

The findings of enclosed iron phosphides (Fe₂P, Fe₃P) and troilite ± Fe⁰ in the early growth zone of plagioclase phenocrysts from melilite–nepheline paralava samples MN-1412 and MN-1420, as well as in glass from clinker xenoliths of paralava samples MN-1420 and MN-1406, together with zoned spherules composed of troilite, Fe⁰, Fe₃P, and Fe₃C in sample MN-1417, indicate the presence of Fe–S, Fe–P, and Fe–P–S–C melt drops in the paralava. This melt drops were captured by growing plagioclase phenocrysts early during the formation of the mineral assemblage in the paralava matrix.

Iron phosphides and Fe⁰ can co-crystallize only at oxygen fugacity below the IW buffer and, for this reason, are found together in many stony-iron and iron meteorites. In this respect, the presence of extremely reduced mineral assemblages in some rocks of the Nyalga and Khamaryn–Khural–Khiid complexes is hard to explain as they result from multi-stage thermal metamorphism and partial melting of sediments during coal combustion which requires oxygen access.

Hypothetically, some low-permeable overburden domains above underground fire foci may have undergone thermal alteration with very limited to zero air inputs, by conductive heat transferred from adjacent burning parts of coal seams. Similar processes of thermal metamorphism in extremely reducing conditions were inferred for rocks deep inside large burnt waste dumps (so-called black blocks) in the Chelyabinsk coal basin [47]. The fluid (gas) regime in these zones, at impeded access of atmospheric oxygen, excess of atomic carbon, and lack of water, was controlled by the CCO buffer.

The probable redox evolution during melting of marly limestone, crystallization and subsequent cooling of melilite–nepheline paralava is presented in the T–LgP_O2 diagram (Figure 18) based on estimated crystallization temperatures of reduced mineral assemblages.

The highest-temperature mineral assemblage of magnesiowüstite–ferropericlase, Mg-Fe spinel, and forsterite in remnant sinter from the paralava matrix of sample MN-1133 (Figure 7 e-g) was formed at oxygen fugacity between the WM and IW buffers, while the T > 1500–1600 °С melting temperatures of the sinter based on the FeO–MgO and FeO–MgO–Al₂O₃ phase diagrams (Figure 14) can be overestimated (see above).

At ~1350 °С, large zoned spherules may have accommodated a homogeneous Fe–P–S–C melt (Figure 9, Figure 10 and Figure 16) which produced zones of troilite, schreibersite and Fe⁰ ± (Fe₃C + Fe₄P). High-temperature phosphide Fe₂P in plagioclase phenocrysts may have crystallized at temperatures between 1370 °С and 1165 °С (Figure 15) and at oxygen fugacity far below the IW buffer (Δlog fO₂ (IW) – 6 according to [25,26]), in extremely reducing conditions probably controlled by the CCO buffer equilibrium.

Schreibersite from plagioclase phenocrysts, as well as the Fe₃P + Fe⁰ + Fe₃C assemblage in zoned spherules, crystallized between 1165 °С and 952 °С, at fO₂ about the CCO buffer (Figure 15, Figure 17 and Figure 18). The matrix of melilite–nepheline paralava samples MN-1133, MN-1287, MN-1420, MN-1412, and MN-1406 had been completely formed before the system cooled down to 850–800 °С, or the minimum crystallization temperature for the Fe-Ca olivine solid solution [2]. At the same time, mineral assemblages free from Fe⁰ and iron phosphides crystallized at a higher oxygen fugacity about the FMQ buffer [1,2].

The crystallization temperature of iron monosulfide (high-temperature pyrrhotite) in large grains and in eutectic symplectites with wüstite–magnetite inclusions (Figure 5 and Figure 7), which crystallized from Fe⁰, varied from < 1190 °С to 988 °С (Figure 3). As the crystallized paralava was cooling down, metallic iron in monosulfide grains self-oxidized to wüstite and magnetite (Figure 5 and Figure 7) because the affinity of iron to oxygen increases faster than that of carbon. The equilibria overlap at ~720 °С (CCO–IW and CCO–IM) and at ~550 °С (IW–MW and IW–IM) (Figure 18). Meanwhile, the self-oxidation of Fe⁰ to wüstite (sample MN-1133, Figure 7) and wüstite to magnetite (samples MN-1133 and MN-1287, Figure 5 and Figure 7) in iron monosulfide grains remained locally incomplete as paralava cooled down relatively rapidly. Furthermore, Fe⁰, wüstite, and magnetite coexist at < 550 °C in a small LgP_O2 region confined by the WM, IW, and IM buffers equilibria. Upon further cooling to < 140 °С, iron monosulfide underwent solid-state transition to troilite (Figure 3), while magnetite and troilite became partly replaced by goethite and marcasite, respectively, at lower temperatures and/or during weathering (Figure 5 and Figure 7).

5. Conclusions

Reducing mineral assemblages of FeS + Fe⁰ + Fe₂P + Fe₃P and FeS + Fe⁰ + Fe₃P + Fe₃C + Fe₄P were studied comprehensively in several samples of melilite–nepheline paralava and other thermally altered and partly melted rocks from the Naylga and Khamaryn–Khyral–Khiid combustion metamorphic complexes in Mongolia.

The paralava samples include matrix fragment composed of magnesiowüstite–ferropericlase, spinel (Mg,Fe)Al₂O₄, and forsterite. As inferred from the MgO–FeO and MgO–FeO–Al₂O₃ phase diagrams, this mineral assemblage has the highest crystallization temperatures and may represent Fe-rich sinter produced by melting of coal combustion ash.

The crystallization temperature were also inferred from the Fe–S phase diagram for paralava-hosted iron monosulfide (high-temperature pyrrhotite) with Fe⁰ inclusions which converted to the stable phase of troilite by solid-state transformation at 140 °C.

The Fe–P phase diagram was used to reconstruct the crystallization conditions of iron phosphides Fe₂P, Fe₃P, and Fe₄P in inclusions from the early growth zone of plagioclase phenocrysts in the melilite–nepheline paralava matrix: < 1370–1165 °C, 1165–1048 °C, and < 1048 °C, respectively.

The analyzed samples enclose large zoned spherules consisting of troilite, Fe⁰ or kamacite and taneite, Fe₃P ± (Fe₃P + Fe₃C) in complex phase relationships. At ~1350 °С, the spherules may have contained a homogeneous Fe–P–S–C melt which separated into two immiscible melts in the 1350–1250 °С interval: a dense Fe–P–C in the core and a less dense Fe–S melts in the rim. The two melts evolved upon further cooling following the paths for the Fe–S and Fe–P–C systems. After the crystallization of pyrrhotitic rim, the Fe–P–С melt parent to the schreibersite zone preserved yet its fluidity; Fe₃P and Fe₃C crystallized between 988 and 952 °C.

The evolution of redox conditions during crystallization of melilite–nepheline paralava melts and their subsequent cooling was reconstructed with reference to the CCO, IW, IM, and MW buffer equilibria from the inferred crystallization temperatures of the reduced mineral assemblages.

The formation of high-temperature reducing mineral assemblages during multi-stage underground coal fires may have been as follows. Thermal alteration of some low-permeable overburden domains above the fire foci was possible with very little to zero air oxygen access, by conductive heat transfer from adjacent burning parts of the coal seams. The gas flow in those zones, with the lack of atmospheric oxygen and water but excess of atomic carbon, was apparently controlled by the CCO buffer. The mineral assemblages with troilite, metallic iron, iron phosphides, and cohenite crystallized from droplets of immiscible Fe–S, Fe–P, and Fe–P–S–C liquids in high-temperature and extremely reducing conditions.