3.1. Preparation of the starting material

The starting material used for our experiments is the sample DR08 that was dredged during the early phases of the 2018–2021 eruption by the MAYOBS-2 oceanographic campaign [Figure 1B; Table 1; Jorry 2019]. This sample has been petrologically characterized by Berthod et al. [2021a]; according to these authors (see their Tables 2–4), it is an evolved glassy popping-rock basanite (Mg ∼ 4.5 wt%; Table 1) with normative olivine content being ⩾10%, moderately vesicular (∼31%) with 5–9% (calculated on a bubble-free basis) of olivine (∼Fo72 ± 2 mol%), and trace amounts of Ti-magnetite and apatite.

Several pieces of this glassy rock were first finely ground in an agate mortar, put in a Pt crucible at 1400 °C and melted twice, with grinding in between, during 5 h in open atmosphere. Electron microprobe analyses (EMPA) of the starting glass show a homogeneous composition with no significant Na or Fe loss compared to data from Berthod et al. [2021a, Table 1]. The resulting dry glass was then ground to obtain the powder that was used as starting material for the phase equilibrium experiments and stored in an oven at 120 °C.

3.2. Experimental equipment and strategy

In total, six experiments were performed (Table 2) at the ISTO experimental laboratory, using the same experimental apparatus and procedure of Andújar et al. [2013], Moussallam et al. [2013], Scaillet et al. [1992], which are briefly summarized below. Experiments were carried out in an Internally Heated Pressure Vessel (IHPV) operating vertically, loaded with Ar–H₂ mixtures at room temperature to achieve the desired fO₂ conditions (see below). Total pressure was recorded by a transducer calibrated against a Heise Bourdon gauge with an uncertainty of ±20 bars. A double-winding kanthal furnace was used as this allows to achieve near-isothermal conditions (gradient <2–3 °C/cm) along a 3 cm long hot spot. Temperature was measured using K-type thermocouples with an accuracy of ±5 °C. A rapid-quench technique was systematically used, imparting isobaric cooling rates of >100 °C/s [e.g., Andújar et al. 2013, 2015, Scaillet et al. 2008]. In all runs reported here, the drop quench was successful as indicated by the rise in total pressure upon the falling of the sample holder into the cold (bottom) part of the vessel.

Experiments were mainly conducted at T–P conditions covering those determined for similar phonolites worldwide [e.g., Andújar et al. 2008, 2010, 2013, Andújar and Scaillet 2012, Giehl et al. 2013, Moussallam et al. 2013, Scaillet et al. 2008]. Temperature was set at 925 and 950 °C whereas explored pressures were 100, 150, 200 and 400 MPa. Based on the iron-rich character of the Mayotte phonolites [Berthod et al. 2021b], the initial fO₂ was set at ⩽ NNO buffer (where NNO refers to the Ni–NiO buffer) as Fe enrichment in magmas is favoured by low fO₂ [e.g., Giehl et al. 2013, Moussallam et al. 2013, Toplis and Carroll 1995].

3.3. Capsule preparation

We used Au gold capsules (1.5 cm long, 2.5 mm inner diameter, 0.2 mm wall thickness) since this metal minimizes the Fe-loss towards the capsule walls under reducing conditions. Distilled H₂O was first loaded, then silver oxalate as the source of CO₂ for H₂O-undersaturated runs, and then the glass powder. Capsules were weighed and then welded using an arc-electric welder. After welding, capsules were re-weighed and if no significant weight loss occurred (considered to occur for a difference >0.0004 g (or ⩽0.1% of the final weight of the capsule)), they were left in an oven for a few hours at 100 °C, to ensure homogeneous volatiles distribution. Both the amount of H₂O + CO₂ and fluid/silicate ratio were maintained constant (3 ± 0.5 mg of H₂O + CO₂, and 30 mg silicate). At a given T–P conditions, various starting H₂O–CO₂ mixtures were explored: XH₂O_in, defined as H₂O∕(H₂O + CO₂) (in moles), varied in the range 1–0.08 (Table 2).

A typical experiment contained five or six capsules, each loaded with a different H₂O∕CO₂ ratio. Run duration varied between 64 and 74 h, depending on temperature. Experiments were terminated by using the drop quench device and then switching off the power supply. After the experiments, capsules were checked for leaks, opened, and half of the run product was embedded in a probe mount with an epoxy resin and polished for optical observation, and carbon coated for subsequent EMPA and scanning electron microscopy (SEM) characterisation.

3.4. Water content, fH₂, fO₂ in the capsules

The variation of the XH₂O_in in the capsules allowed us to explore different water fugacities, and thus different H₂O_melt (Table 2). The amount of dissolved water (XH₂O_in) in the glass of the different H₂O-saturated charges was calculated using the H₂O-solubility model of Jiménez-Mejías et al. [2021] which is suited for basanitic–phonotephritic compositions. The water contents of charges with XH₂O_in < 1 ran at the same temperature and pressure than the H₂O-saturated charge, were calculated by mass balance using the constraint of equilibrium fugacities of H₂O and CO₂ species between melt and fluid, along with the H₂O–CO₂ solubility model of Jiménez-Mejías et al. [2021].

The fH₂ of the experiments was determined using an empirical calibration curve established in previous experimental works, relating the H₂ pressure loaded initially to the autoclave at room temperature to the final fH₂ as determined using NiPd sensors run at similar P–T–fO₂ [Andújar and Scaillet 2012, Andújar et al. 2015, Romano et al. 2018]. Once the prevailing fH₂ is known, the fO₂ can be determined by using the dissociation constant of water [Robie et al. 1979], knowing the fH₂O at the experimental temperature and pressure. fH₂O was calculated as fH₂O = XH₂O_in ∗ fH₂O^o where fH₂O^o is the fugacity of pure water [Burnham et al. 1969]. Afterwards, the fO₂ for each capsule was calculated (Table 2) (see Andújar and Scaillet [2012] for further details). The experiments were conducted at oxygen fugacities around NNO ± 0.3 at H₂O-saturation. As fO₂ decreases with XH₂O_in [e.g., Andújar and Scaillet 2012, Freise et al. 2009, Scaillet et al. 1995, Webster et al. 1987], at fixed T, P, fH₂, for water-undersaturated charges the prevailing oxygen fugacity is necessarily <NNO ± 0.3. As a result, in a single experimental series, the decrease in XH₂O_in from 1 to 0.08 decreased the fO₂ up to about 2.5 log units below the NNO buffer (or ∼1.8 log units below the Fayalite-Magnetite-Quarz buffer, FMQ; Table 2).

3.5. Analytical techniques

Experimental runs were first characterized using a Zeiss Merlin Compact electron microscope equipped with an EDS micro-analysis system (Bruker-Quantax-XFlash6) at ISTO. EDS spectra allowed a first order identification of the minerals present in each charge. Experimental phases were analyzed using a Cameca SX-Five EPMA with an acceleration voltage of 15 kV, a sample current of 6 nA, and a counting time of 10 s. For glasses, a defocused beam of 20 μm was used whereas for minerals a focused beam was employed instead. Sodium and potassium were analyzed first and a Phi-Rho-Z correction procedure was applied. Major elements calibration was done using the following standards: albite (Na, Si), orthoclase (K), andradite (Ca), apatite (P), chromite (Cr), corundum (Al), magnesium oxide (Mg), hematite (Fe) and pyrophanite (Mn, Ti). The relative analytical errors are 1% (SiO₂, Al₂O₃, CaO), 3% (FeO, MgO, TiO₂) and 5% (MnO, Na₂O, K₂O, P₂O₅).

In the case of experimental plagioclase (Pl), the small size of this phase in the charges (usually ⩽3 μm) along with the presence of small (⩽0.5 μm) Fe–Ti oxides-apatite inclusions inside the crystals, made difficult the proper analysis of this phase with EPMA. To circumvent this problem, Pl composition was obtained via using a SEM-EDS calibrated analysis: albite, orthose and andradite EMPA standards were analyzed first with a focused beam (1 μm), a voltage of 15 kV and counting times of 15 s for the different elements (Na, K, Si, Al, Ca, Na, K). SEM-EDS compositions of the standards were found to be within ⩽5% to those obtained by EMPA (see Supplementary Table). Consequently, experimental plagioclases were determined by SEM-EDS in each charge, using the same conditions and the average SEM-EDS calibrated compositions fulfilling the stoichiometric requirements for this phase (see Supplementary Table).

It should be noted that, compared to EMPA analyses, which are characterized by lower standard deviations (see Tables 3–8), SEM-EDS compositions are slightly more variable (see Supplementary Table), with a variation of ∼2–3 mol% on end-member molecules. However, despite this, we can still use these compositions for comparison purposes.

3.6. Attainment of equilibrium

The crystallization experiments performed in this work share similar phase stabilities, textural and compositional characteristics to those found in equivalent studies performed on alkali-rich basalts [e.g., Andújar et al. 2015, Freise et al. 2009, Iacovino et al. 2016] using a similar procedure. As in these works, the following observations show that near-equilibrium conditions were attained in our experiments: the euhedral shape of crystals, the homogeneous distribution of phases within the charges, the smooth variation of phase proportions and compositions with changes in experimental conditions, and the small sum of residuals of mass-balance (generally ⩽1; Table 2). Since each experimental charge is a closed-system, the low-residuals indicate that there was no mass loss or gain during the experiments, and that no major mineral phase has been omitted in our calculations.

4.1. Phase relationships

The main mineral phases crystallizing in our runs are Bt (biotite), Amp (Kaersutite-type), Ol, Mt, Ilm, Ap, Pl and Cpx, with different glass proportions. Variations in temperature, pressure and H₂O_melt directly affect the stability and relationships of the different minerals. These changes are displayed in two isothermal-polybaric sections described below (Figure 3).

At 925 °C, the water-rich part of the diagram (H₂O_melt > 4 wt%) is dominated by the assemblage Bt+Amp+Mt+Ap+melt, followed by Ilm, Pl and Cpx with decreasing H₂O_melt (Figure 3A). Cpx coexists in a relatively narrow H₂O_melt region with Amp, both being involved along with Pl and melt, in a reaction-relationship that results in the disappearance of Amp at lower H₂O_melt (Figure 3A). Bt is stable everywhere in the P–H₂O_melt space explored. However, the decrease of its modal proportion with H₂O_melt (Table 2) and the consequent enrichment of the residual liquid in K₂O point towards its instability in the driest part of the system (Figure 3A). Hence, at 925 °C and for H₂O_melt < 1 wt%, the mineral assemblage that dominates the explored P-H₂O_melt region is Ol+Cpx+Pl+Ap+Mt+Ilm.

At this temperature and for pressures between 100 and 200 MPa, the Amp-out and Cpx, Ol, Ilm and Pl-in curves display steep or even vertical slopes, all occurring in a narrow H₂O_melt interval between 2.5–3 wt% (Figure 3A). Above 200 MPa, Cpx-in and Amp-out curves shrink towards lower H₂O_melt (⩽2 wt%) whereas Ilm stability shifts towards higher H₂O_melt. Pressure does not have any major effect on Ol and Pl saturation curves.

At 950 °C, the general topology is broadly similar to that at 925 °C (Figure 3B), with Pl and Cpx appearing at H₂O_melt < 2–3 wt%, as anticipated, and Ol crystallizing first relative to Pl.

4.2. Phase proportions

Since experiments were performed at relatively low temperatures, liquidus conditions were not reached during the experiments. Charges always contained different amounts of crystals, which vary between 29 and 66 wt% (Figure 4A). In general, at fixed P–T conditions, the crystal proportions increase as H₂O_melt decreases (Figure 4A). Amp is the dominant phase whenever present, with proportions of up to 50–70 wt% of the crystallizing assemblage (Figure 4B); as H₂O_melt decreases, the Amp content decreases below 15 wt%. With decreasing H₂O_melt, Mt, Bt and Ap proportions steadily decrease while Ol, Ilm, Cpx and Pl increase. However, compared to Amp, any of these phases never exceed 15–25 wt% (Table 2).

4.3. Phase compositions

Experimental phase compositions (minerals and glass) are displayed in Tables 3–9, in supplementary and their variation with experimental variables (P–T and H₂O_melt) is discussed in the following sections.

4.3.1. Olivine

Ol crystallizing in our charges varies between Fo₅₁₋₃₃, with relatively constant contents of MnO (1.4 ± 0.2 wt%), depending on H₂O_melt, temperature and pressure conditions (Table 3). At a given P and T, a decrease in H₂O_melt decreases the Fo and increases the CaO content of Ol, similarly to the trends observed in other basaltic compositions [Andújar et al. 2015, 2017, Berndt et al. 2005, Di Carlo et al. 2006]. This is well exemplified at 400 MPa and 950 °C where a decrease in H₂O_melt of 2.5 wt% produces a diminution of 8 mol% of Fo (Figure 5A) and an increase of ∼0.5 wt% in the CaO content of Ol (Figure 5B). A drop of 25 °C results in a relatively modest variation of Fo content, generally <5 mol% (Figure 5A). Series conducted at 925 °C allow to assess the effect of pressure on Ol composition. Considering a fixed H₂O_melt (e.g. 2.5–3 wt%), runs conducted between 100–150 MPa crystallized Ol with a similar composition (e.g. ∼Fo₅₀). Increasing P by 50 MPa (so up to 200 MPa) produces changes in Ol composition that are equivalent to those observed with a T diminution (i.e., decrease of 8 mol% in Fo; Figure 5A). At higher pressures, the effect of this parameter in olivine composition is more pronounced, since at 400 MPa, crystallizing Ol are highly more fayalitic (Fo₃₃) compared to their shallower counterparts (e.g., Fo₄₅₋₅₀ P⩽200 MPa). The average Ol–liquid exchange (KdFe*–Mg; using FeO* as total iron) is 0.29 ± 0.1 (varying between 0.25 and 0.44).

4.3.4. Fe–Ti oxides

Mt has a relatively constant TiO₂ content at ∼22 ± 2.5 wt%, with more variable FeO* (68–74 wt%). When present, Ilm was often too small to obtain reliable compositions by EPMA. Nevertheless, analyzed Ilm in 5 charges have almost constant TiO₂, at 50 wt%, and FeO* contents in the range 42–46 wt% (Tables 6–7). As for other phases, changes in experimental parameters affect Mt and Ilm compositions. The effect of H₂O_melt is significant: for example, in charges annealed at 950 °C–400 MPa, a decrease of 4 wt% H₂O_melt decreases Mg#by ∼4% (Figure 7A), FeO* by 1.5 wt% and increases %Ulvöspinel from 60 to 70% (Figure 7B). The other experimental series display quite similar trends (Figure 7). A decrease in temperature of 25 °C produces similar changes. As for olivine, the effect of pressure on Mt composition is enhanced within the interval 200–400 MPa, low pressure Mt being enriched in Mg#and depleted in FeO* and %Ulvöspinel (compare series run at 950 °C–400 and 200 MPa in Figure 7).

4.3.5. Plagioclase

Pl are mostly andesines with compositions between An₂₇–An₄₂ [Deer et al. 1972]. The most evolved Pl (oligoclase) crystallized instead in charges DR08-132, 155 and 161. As already observed in previous experimental studies performed on hydrous basaltic or basaltic andesite melts, Pl composition is highly sensitive to changes in H₂O_melt [e.g. Andújar et al. 2015, 2017]. The most important compositional variations are observed in charges run at 925 °C and P⩽150 MPa, where the decrease in 2 wt% H₂O_melt results in a reduction of 10% in the mole fraction of An (see Supplementary Table). Similar variations in the CaO content of the experimental plagioclases can be produced when pressure decreases from 200 to 150 or 100 MPa at a given H₂O_melt, an effect that is again in agreement with the results of previous works [e.g. Andújar et al. 2015, 2017].

4.3.6. Experimental glasses

Residual glass compositions (calculated on a 100% anhydrous basis) have benmoreitic–trachytic affinities (according to the classification of Le Bas and Streckeisen [1991] with P.I. (peralkaline indice, =Na+K/Al in moles) at 0.7–1 (Figures 8–9, Table 9). Only charges DR08-143 and DR08-131 produced liquids that straddle the phonolite–trachyte boundary line (Figure 9), having a P.I. = 0.95–1, normative nepheline (6.5–7%) and containing 2–4 wt% H₂O_melt (Table 8). The production of benmoreitic–trachytic–phonolitic liquids reflects the significant degree of crystallization of the charges. In the range of temperatures explored in this work, crystallization is dominated by Amp+Bt along with different proportions of Fe–Ti oxides+Ap for H₂O_melt in the range 2–7 wt% (Figures 3–4, Table 2). The onset of Pl+Ol+Cpx crystallization results in a drastic decrease of Mt, Amp and Bt proportions until the breakdown of the two last phases which occurs at ⩽2 wt% H₂O_melt (Figure 3). The combination of the above factors increases both the SiO₂ and the alkalinity of the residual melt, which reaches the phonolitic field (Figure 9A). In parallel, TiO₂ (not shown), CaO and MgO decrease (Figure 8B,D). In contrast, FeO* may even slightly increase when H₂O_melt decreases (Figure 8C), though by a modest amount (usually <0.5 wt%). This Fe-enrichment is due to the changing proportions of hydrous Fe–Mg phases and Mt (Table 2). In comparison, a decrease of 200 MPa lowers FeO* by ∼1.5–2 wt% for a given H₂O_melt (Figure 8C). Below 3 wt% H₂O_melt, the liquid produced at 400 MPa are even more depleted in FeO* (Figure 8C). This is due to the fact that, at 200 MPa, the precipitation of a relatively Fe-poor Ol (Fo₄₅₋₄₀) enhances the iron-enrichment process. In contrast, at 400 MPa, the crystallization of significant amounts of relatively fayalitic-rich Ol (e.g., Fo⩽38) counteracts the depleting role of Mt at higher H₂O_melt (Figures 3, 8C). Such a difference explains why charges at 200 MPa yield melts with the highest analyzed FeO* contents (7.4 wt%) compared to their high-pressure counterparts, which achieved only a maximum of 6.7 wt% FeO*, and under very high H₂O_melt.

Another striking point concerns the charges annealed at 100–150 MPa since these yield residual melt compositions overlaping those produced at higher P (⩾200 MPa), not only in terms of iron but also in the other major elements (Figure 8). In fact, despite the presence of moderate proportions of olivine with compositions Fo₅₁₋₄₃ (Figure 5A, Table 3), which according to 200 MPa data promotes the Fe-enrichment of the melt, the co-crystallization of ∼8 wt% of Fe-rich magnetite and different amounts of Cpx+Pl±Bt±Amp inhibit the Fe-enrichment process and drive the residual melt towards more Si- and Al-rich and low P.I. compositions [Figures 8–9; Table 8; Botcharnikov et al. 2008, Giehl et al. 2013, Scaillet and MacDonald 2006, Toplis and Carroll 1995]. The effect of P variation is not restricted to FeO*; as pressure increases, the SiO₂, CaO increase with decreasing Al₂O₃ content, MgO and alkalis being somewhat less affected (Figure 8).

It should be also noted that between 7 and 3 wt% H₂O_melt, the precipitation of Amp along with Bt in amounts, representing 60–80% of the total crystal cargo, increases the SiO₂ and Al₂O₃ content but buffers K₂O at ∼3 wt% (Figure 8F). Below this H₂O_melt, the crystallization of Ol+Cpx+Pl, along with a drastic drop of Amp+Bt+Mt proportions, increases the potassium content of the melt by ∼50% relative (Figure 8F). Na₂O follows a similar trend but the observed enrichment is less significant compared to K₂O. The evolution of Al₂O₃ is well illustrated in series run at 200–400 MPa where Al₂O₃ increases when H₂O_melt decreases until the onset of Pl crystallization (Figure 8B). Pl crystallization in relative high proportions (up to ∼30 wt%; Table 2) decreases the Al₂O₃ and accelerates the evolution of the residual melts towards alkali-rich compositions.

Aside from water and pressure, temperature is also an important parameter controlling melt composition. At lower temperatures (i.e. 925 °C) and for a fixed H₂O_melt, residual liquids tend to be enriched in SiO₂ and Na₂O but depleted in TiO₂, MgO and CaO when compared to melts produced at 950 °C. FeO*, K₂O and Al₂O₃ do not show any clear relationship with T in charges run at P⩽200 MPa. Only those at 400 MPa and 950 °C have higher Al₂O₃ and FeO compared to their colder counterparts (Figure 8).

In summary, the fractionation of a relatively hydrous basanite under low oxygen fugacity conditions (fO₂⩽ FMQ; Table 2), at temperatures between 925–950 °C and 200–400 MPa (corresponding to depths of 6–12 km) produces Fe-rich phonolitic liquids containing 2–4 wt% dissolved H₂O. In the next section, we use the composition of the natural phonolites dredged along the Mayotte submarine ridge or collected on-land to shed light on the conditions needed to generate this type of evolved melts at Mayotte.

5.1. Production of phonolitic liquids at the WNW–ESE submarine volcanic ridge of Mayotte

The low porphyritic character of the submarine phonolites dredged in the vicinity of the active Horseshoe structure of Mayotte [Berthod et al. 2021a, b] allows us to consider the bulk-rock composition as representative of liquids. Accordingly, the experimental results detailed above are useful for understanding the conditions for the production of phonolitic melts at this site (Figures 8–9, Tables 1 and 8). Although glasses produced at conditions of 925 °C–200 MPa and 950 °C–400 MPa with 2–7 wt% H₂O_melt broadly reproduce the SiO₂, Al₂O₃, MgO and CaO contents of natural phonolites (Figure 8), only the residual glasses of charges DR08-143 (950 °C–400 MPa; ∼2 wt% H₂O_melt) and DR08-131 (925 °C–200 MPa; ∼4 wt% H₂O_melt) closely match the phonolitic affinities (P.I. ∼ 0.9–1) of the natural products (Figure 9A; Table 1). Both charges have the same mineral assemblage and phase proportions (∼65 wt% of Ol+Pl+Amp+Mt+Ilm+Ap±Bt; Table 2), but differ in the composition of the phases (see above). In detail, only the 200 MPa experiment closely reproduces the FeO* and SiO₂ contents (and other oxides) as well as the FeO*/MgO ratios (Figure 9B). The glass produced at 400 MPa has a FeO* content ∼2 wt% lower and a higher SiO₂ with respect to the dredged phonolites (Figure 9B). Our data therefore show that the crystallization of DR08 basanite at 925 °C–200 MPa (corresponding to depths of 6–8 km) and under relatively reduced conditions (fO₂ ∼ FMQ; Table 2) generates phonolitic melts (with up to 3.8 wt% H₂O_melt) with FeO* contents and FeO*/MgO ratios similar to phonolites sampled offshore. Whereas such T–P–H₂O_melt estimates are in agreement with the conditions set for most of the phonolites and evolved peralkaline magmas worldwide [Andújar et al. 2008, 2010, 2013, Andújar and Scaillet 2012, Di Carlo et al. 2010, Giehl et al. 2013, Moussallam et al. 2013, Romano et al. 2018, Scaillet and MacDonald 2006, Scaillet et al. 2008], they strongly differ from those inferred by Berthod et al. [2021b], in particular with respect to pressure conditions: as stated previously, these authors conclude that the fractionation of a 80 wt% dry mineral assemblage from a hydrous basanite (2.3 wt% H₂O_melt) at mantle depths (18–20 km, 600 MPa) produces the Fe-rich phonolites of the ridge. However, in this model, no role is given to the hydrous phases like amphibole or biotite in the evolutionary process. Assuming that H₂O_melt (2.3 wt%) determined in olivine melt inclusions in DR08 sample by Berthod et al. [2021b] is representative, a fractionating assemblage lacking Amp with such an H₂O_melt requires a maximum pressure of ∼200 MPa (Figure 3). Our results show, however, that basanite crystallization under such wet conditions (i.e., ending with residual liquids with H₂O_melt > 6 wt%, as required by the largely incompatible behaviour of H₂O, for 80 wt% crystallization of a basanite with 2.3 wt% H₂O_melt) produces mostly trachytic melts (Figure 9A), true phonolites requiring drier conditions ( <3–4 wt%). This mismatch is seemingly due to the fact that MELTs does not incorporate amphibole, which is an important fractionating phase in hydrous basanite as our experiments demonstrate.

It should be also stressed here that the differences between MELTs calculations and our results not only concern the role of hydrous phases in the fractionating process, but also the composition of feldspar that is in equilibrium with these Fe-rich phonolitic melts. Whereas the calculations of Berthod et al. [2021b] predict the precipitation of at least 30% of alkali feldspar (anorthoclase) to produce the phonolitic melts, the experimental charges yielding to such type of liquids are instead characterized by the presence of relatively primitive Plagioclase (oligoclase) with composition ∼27 mol% in the fractionating mineral assemblage. The submarine phonolites have a dominant alkali feldspar population at ⩽An₁₅ mol%, but also contain a few more calcic cores (An₂₄ mol%) [Berthod et al. 2021b]. The latter is close to that of experimental Pl (An_27±3 mol%) produced in charge DR08-131 (925 °C–200 MPa), which is as shown above, in equilibrium with a Mayotte-type phonolitic liquid. Hence, the more calcic Pl cores in the natural phonolite can be considered as crystals inherited from the fractionation process involving the crystallization of the mineral assemblage Ol+Pl+Amp+Mt+Ilm+Ap±Bt down to the phonolitic production stage. The less calcic and dominant variety of feldspar microphenocrysts in the phonolite is likely the result of a small increment of crystallization that occurred after the phonolitic liquid was extracted from its parental basanitic magma, perhaps due to an adiabatic cooling effect upon decompression that took place when the phonolitic melt had left the basanitic reservoir.

Hence, we conclude that conditions of 950/925 °C–200 MPa with H₂O_melt⩽3–4 wt% are best suited for the production of Fe-rich phonolites at Mayotte. Deeper conditions produce significantly different compositions, as stated previously (Figure 8).

5.2. The production of the north-central Mayotte Karthala type phonolites

As stated in previous sections, the compositions of the magmas at Mayotte clearly define two distinct magmatic series. Since our starting basanitic composition belongs to the moderate-silica undersaturated series (i.e. Karthala type), our results primarily apply to this specific compositional lineage (Figure 2).

Detailed mineralogical information concerning the phonolitic products of Mayotte is scarce, however: only the works of Pelleter et al. [2014] and Späth et al. [1996] provide bulk-rock information and few mineral data from these evolved compositions (Table 1). For comparison purposes, we have only considered the bulk-rock compositions having MgO contents up to 0.4 wt%, since these are the maximum amounts of our experimental melts (Table 8). The phonolites-trachytes with MgO contents lower than ∼0.25 wt% represent the most evolved products of the Karthala LLD [Pelleter et al. 2014, Späth et al. 1996]. In this case, the liquids produced in our experiments possibly represent the parental compositions from which the Mayotte on-land phonolites derive.

Compared to phonolites from the submarine ridge, those from Mayotte island are characterized by a contrasted mineralogy [Amp+Cpx+Mt+Ilm+ Afs+Ap±Nph) and have slightly lower Fe-contents (between 3.7 to 5.1 wt%; Pelleter et al. 2014]. Rare Ol+Afs bearing phonolites [i.e., equivalent to the dredged products; Berthod et al. 2021b], and aphyric glassy samples [sample M29 in Pelleter et al. 2014; Table 1] can be also found in the northern-central parts of Mayotte.

Apart from sample M29, the Mayotte phonolites are reproduced by the 200–400 MPa series products on TAS, FeO*/MgO, CaO, Al₂O₃ versus SiO₂ Harker diagrams (Figure 9), suggesting that these melts could have been potentially generated at ∼300 MPa (9 km depth). This pressure is about 100 MPa (3 km) higher (deeper) than that determined for the conditions of production of their 2018–2021 submarine equivalents. However, both the lower iron content of the Mayotte island phonolites compared to the submarine ridge products and the presence of Amp in these compositions (Amp stability is enhanced with pressure; Figure 3), support deeper conditions for generation of the on-land phonolites.

The M29 aphyric sample has the highest SiO₂ (61.5 wt%) and the lowest FeO* (3.2 wt%) contents of all phonolites considered in this work (Table 1). According to our results, the production of this phonolitic composition will require ponding conditions of a DR08 type basanite at pressures ⩾400 MPa. In fact, the melt from the charge DR08-143 run at 400 MPa closely matches the total alkalis and SiO₂ contents of this phonolite (Figure 9A). However, when it comes to FeO*/MgO ratio and TiO₂ and CaO contents, the best fits are obtained by charges produced at 925 °C–100 MPa (Figure 9, Table 8). The absence of phenocrysts in the natural sample does not allow us to use the mineralogical information to discriminate between these conditions. However, we cannot exclude that this aphyric phonolite was produced by the fractionation of a basanitic magma at low-pressure (⩽100 MPa) and under relatively dry (⩽3 wt%) and hot (?) conditions.

As stated above, the existence of phonolites, with MgO contents <0.4 wt% containing a mineral assemblage Amp+Cpx+Pl+Fe–Ti oxides [Debeuf 2004, Pelleter et al. 2014], suggests that, even if these magmas evolved from liquids similar to those produced in this work, their composition and mineralogy [low-Mg number of amphiboles, Debeuf 2004] do not necessary reflect the production level: they could reflect shallower final ponding conditions. This observation highlights the need for detailed petrographic data on Mayotte phonolites, especially those on-land, if we are to define more precisely the pre-eruptive conditions of Khartala-type phonolites.

5.3. Various levels of phonolite production/storage at Mayotte

The crystallization experiments, performed on a representative evolved basanitic sample from the 2018–2021 eruption, allowed us to define the best conditions (T–P–fO₂–fH₂O) for producing the characteristical compositional variability of phonolites from the north-central parts of the Grand Terre island, Petite Terre and NE active submarine ridge of Mayotte. In particular, these experiments show the ease of producing phonolitic melts in the shallow crust, as observed in many other settings. Phonolitic melts containing up to ∼3–4 wt% of H₂O_melt are produced by the extensive crystallization of at least 65 wt% of a mineral assemblage consisting of Ol+Pl+Amp+Cpx+Bt+Mt+Ilm+Ap, at crustal levels ⩽12–15 km. Pressure (hence depth) appears to be an important factor controlling the final FeO* content of residual phonolitic liquids. This oxide could be potentially used as a geobarometer (for this type of magma series) to infer the generation/storage depths of corresponding magmas.

A conceptual model summarizing our view on Mayotte phonolites is shown in Figure 10. Considering the above water-contents, the differentiation of basanitic magmas at depths ⩽4–5 km produces residual liquids with trachyte–benmoreite affinities. The Fe-enriched phonolite compositions found in the Mayotte system are instead produced at around 6–8 km, whereas at greater depths, the fractionation process results in the generation of phonolitic melts which are progressively enriched in SiO₂–Al₂O₃ and depleted in FeO*.

Hence, the production of phonolitic melts with the Karthala affinities of Mayotte island occurs at multiple depths above the local Moho [16–17 km, Dofal et al. 2021]. However, the presence of highly evolved phonolitic magmas (i.e. MgO poor) suggests that these magmas experienced a later episode of crystallization and fractionation at shallower levels, prior to their eruption. The occurrence of relatively shallow phonolitic reservoirs below the Horseshoe region (Figure 1C) would have permitted their subsequent rapid emptying and collapse during past eruptions, which would readily explain the presence of a caldera-looking depression at this point, since relatively shallow depths of magma reservoirs favors the formation of caldera structures [e.g. Martíet al. 2008].

Our results contrast with the conditions inferred from mineral geobarometers and MELTS simulations, which for producing the Fe-rich phonolites from Mayotte, predict the crystallization of a dry mineral assemblage at significantly higher pressures than those determined in this work (600 versus 200 MPa). This difference highlights the still poor resolution that these two methods have for predicting the depths of magma evolution in alkaline systems, and calls for additional calibration of these tools.