The Pavin trachyandesite exhibits various textures, from dominantly pumice to denser and darker clasts that differ by vesicularity and groundmass microtexture, while their phenocryst assemblage and whole-rock composition do not show significant differences within the limits of our investigations. Major-element whole-rock compositions are given in Table 1 where the Pavin magma composition is compared to a closely analogous composition (Nugère trachyandesite) and a slightly more evolved one (Clierzou trachyte), both from the Chaîne des Puys volcanic field. All rocks show overall similarities, being metaluminous (P.I. in the 0.6–0.9 range) and close to silica saturation in the CIPW norm. The differentiation trend in the Chaîne des Puys has been thoroughly documented and is compositionally marked by a consistent decrease of FeO, MgO, CaO, and increase of alkalis with increasing SiO₂ (Camus, 1975; Martel et al., 2013; Maury et al., 1980). The Nugère (Chaîne des Puys) and Pavin trachyandesites have close contents for most major elements, though the Pavin composition is consistently richer in TiO₂ and K₂O, whereas poorer in P₂O₅. In the Chaîne des Puys differentiation trend, the Na₂O/K₂O ratio does not vary from the Nugère to the Clierzou compositions (Table 1) then it tends to decrease with increasing SiO₂ in the trachyte range, through an increase of K₂O at almost constant Na₂O (Martel et al., 2013 and references therein). The compositional differences between the Pavin and Nugère magmas, notably the K₂O-richer character of the Pavin magma, are sufficient to indicate that the Pavin magma belongs to a specific differentiation trend, i.e. distinct from the Chaîne des Puys one, as also evidenced by comparing trace-element compositions (Villemant et al., 2016)

The phenocryst assemblage of the Pavin magma mostly consists of plagioclase, amphibole, clinopyroxene, and Fe–Ti oxides (Appendix B). Phenocrysts are found both isolated and as small glomeroporphyritic agglomerates. Apatite is a common accessory phase. Scarce biotite can be found as inclusions in amphibole and isolated microphenocrysts. Modal proportions range from 10 to 23% (recalculated to 7 to 20 wt% crystallinity; Table 2). Such a variability of the phenocryst proportions could result from heterogeneous distribution at sample scale (e.g., due to the tendency of phenocrysts to agglomerate) or might be acquired through crystallization along the course of the eruption. No attempt has been made in this study to evaluate the variability of magma crystallinity through the Pavin deposit stratigraphy

To some extent, the phenocryst assemblage of the Pavin trachyandesite can be compared qualitatively to counterparts from the Chaîne des Puys. Although compositionally close to the Pavin sample, the Nugère trachyandesite is almost aphyric. Other trachyandesites in the Chaîne des Puys, although scarce and volumetrically trivial, typically contain as Fe–Mg phenocrysts, both amphibole (more or less stable) and stable clinopyroxene, and lack biotite (Maury et al., 1980). Further, the couple clinopyroxene–amphibole is still present in the less differentiated trachytes, Aumone and Clierzou, which also lack biotite (Boivin et al., 2015). In contrast, biotite is a systematic phase in the more differentiated trachytes, either as the single Fe–Mg phase or accompanied by either amphibole or clinopyroxene, not both (Boivin et al., 2015; Martel et al., 2013). The Pavin phenocryst assemblage thus compares well with the Chaîne des Puys magmas at comparable levels of differentiation. The rare occurrence of biotite in the Pavin magma, at lower SiO₂ than needed for biotite appearance in the Chaîne des Puys series, might reflect the higher amount of K₂O in the Pavin magma compared to the Chaîne des Puys ones to a similar differentiation degree.

The natural phases were analysed following the method given in Appendix A2 and the compositions given in Appendix C

Plagioclase, although frequently fractured, is euhedral with clear rims, and variously zoned with frequent cloudy to sieved core textures. The plagioclase composition spreads from 32 to 58 mol% of anorthite (An32 to An58), with an orthoclase content not exceeding 5% (<Or5) and consistently increasing as An decreases, (Fig. 3a). Plagioclase rims have a restricted compositional range of An35–38, while most of the cores have compositions of An45–62, with the exception of one profile showing a reverse zoning with core composition down to An32. From core to rim, the plagioclases show either smooth An decrease or drastic drops (Fig. 3b). In the Chaîne des Puys, for comparison, plagioclases in the trachyandesites are reported as being mostly in the range of andesine (An30–50; Maury et al., 1980). The plagioclases of the less differentiated Chaîne des Puys trachytes are also andesine, in the range An30–42, with a mean composition of An38 (Boivin et al., 2015; Martel et al., 2013).

Amphibole occurs as 1–2-mm-long, euhedral and unresorbed phenocrysts with common inclusions of oxides, apatite and biotite (Appendix B2). The amphiboles are kaersutites and hastingsites after Leake et al. (1997), showing a rather scattered range in contents of SiO₂ (38.5–40.7 wt%), Al₂O₃ (11.1–13.9 wt%), Al^IV (1.9–2.2), TiO₂ (4.1–5.3 wt%), and Mg# (0.61–0.66) (Appendix C2). Amphibole-rich cumulates are common in the Pavin deposit, notably in the upper part of the deposit stratigraphy. The amphibole from the cumulates is mostly kaersutite that plots separately from the amphiboles analysed in the juvenile pyroclasts (Fig. 4). This is good evidence that the latter form a truly phenocrystic population, without any contribution of xenocrysts of cumulative origin. Within the phenocrystic amphibole population, core and rim compositions do not show systematic variations and are close to each other in the limits of our data set. Thus, the compositional scattering of the amphibole phenocrysts is not related to any significant zoning and likely reflects variations in the crystallization conditions (liquid composition and/or intensive parameters) within the magma reservoir. Amphiboles from the low-silica Clierzou trachyte are also compositionally scattered, but commonly show slightly lower contents in A-site cation number, Al^IV content, and Mg# (Fig. 4).

The clinopyroxenes in the juvenile pyroclasts are 1–2 mm long, have smoothed more or less resorbed rims, and contain frequent glass inclusions. They are salites with 44–48 mol% wollastonite (Wo44–48) and 34–45 mol% enstatite (En34–45) (Fig. 5a), with notable scattering in SiO₂ and Al₂O₃ contents (Fig. 5b; Appendix C3). By comparison, clinopyroxenes from cumulates are mostly less calcic (Wo41–47) and more magnesian (En39–46), and are on average poorer in SiO₂ and richer in Al₂O₃ and TiO₂ (Fig. 5b; Appendix C3). The compositions of the clinopyroxenes from the cumulates are even more scattered than those of the trachyandesite, suggesting large variations in the crystallization conditions. Compared to the Pavin pumice, clinopyroxenes from the Clierzou trachyte are mostly less calcic (Wo41–47) and more magnesian (En39–46; Fig. 5a), and are on average poorer in SiO₂ and richer in Al₂O₃ (Fig. 5b).

Biotite is scarce and contains about 6 wt% TiO₂, 14.5 wt% MgO and 13–14 wt% Al₂O₃, less than 0.4 wt% MnO and F, with Mg# ~0.65 (Fig. 6; Appendix C4). Comparatively, the Clierzou trachyte does not contain biotite. Biotites are only present in the most differentiated trachytes (>65 SiO₂ wt%), in which they are poorer in magnesium compounds (e.g., Mg# ~0.55; Fig. 6) and slightly richer in alumina (14–15 wt% Al₂O₃ in the Vasset trachyte; Martel et al., 2013) than the Pavin biotites.

Fe–Ti oxides are both magnetite–ulvöspinel and ilmenite–hematite solid solutions (Appendix C5). The isolated phenocrysts often show strong textures of exsolution, whereas Fe–Ti oxides found as mineral inclusions in mostly clinopyroxene and amphibole are texturally and compositionally homogeneous. Fe–Ti oxides in amphibole and clinopyroxene do not show systematic compositional differences. Apparently, homogeneous isolated phenocrysts show compositions similar to those found as inclusions, i.e. titanomagnetite with 67–78 mol% magnetite (Mt67–78; with 0.8–1.3 wt% MnO, 79–83 wt% FeO, and 7.5–11.5 wt% TiO₂) and ilmenite with 63–68 mol% ilmenite (Ilm63–68; with 0.8–1 wt% MnO, 55–57 wt% FeO, and 36–38 wt% TiO₂). For comparison, the magnetites and ilmenites from the Clierzou trachytes are Mt- and Ilm-richer, respectively (Mt80–90 and Ilm70–90; Martel et al., 2013).

Glass inclusions are frequently trapped in plagioclases and clinopyoxenes, but are mostly devitrified in the former whereas mostly glassy in the latter, so that all analysed glass inclusions were clinopyroxene-hosted, with one in a magnetite. The compositions are trachytic, showing narrow ranges of 64–66 wt% SiO₂ and 11.5–12.5 wt% alkalis (Appendix C6). The microlite-poor matrix glasses of the Pavin pumices have compositions similar to those of the glass inclusions (Fig. 7). The Pavin glasses overall broadly compare with the bulk-rock compositions of the intermediate-silica trachytes of the Chaîne des Puys (Kilian, Vasset, Sarcouy, Puy de Dôme), except for lower Na₂O and higher K₂O contents (both by ~1 wt%; Fig. 7). The K₂O-richer character of the Pavin glasses conforms to the K₂O-richer content of the Pavin whole-rock trachyandesite compared to Chaîne des Puys analogues.

The H₂O contents of the Pavin glasses are inferred from a calibrated by-difference method (see Appendix A2), which assumes that volatiles other than H₂O can be neglected. Fluorine contents are below 0.5 wt% in the Pavin glass inclusions (Appendix C6), in agreement with the fluorine, chlorine, and sulphur contents below 0.5 wt% for the glass inclusions of the Chaîne des Puys trachytes (Martel et al., 2013). We thus assume that H₂O is by far the most abundant volatile species and that the inferred values cannot be overestimated by more than 0.5 wt%. The range of H₂O content is large, with the matrix glasses showing low contents (<1.5 wt%) and the glass inclusions ranging between 1.1 and 5.3 wt% H₂O (Fig. 8; Appendix C6).

The natural pumice pyroclasts contain phenocrysts of plagioclase, amphibole, clinopyroxene, magnetite, ilmenite, and minor apatite. The scarcity of biotite, only found as inclusions in amphibole or rare isolated microphenocrysts, likely reflects inherited crystals

We selected pumice clasts from an explosive event during which magma was supposed to ascend rapidly from the storage region to the surface. This has the advantage to minimize syn-eruptive crystallization, although the samples do contain some microlites that evidence crystallization after reservoir exit. If such syn-eruptive crystallization is negligible, the pre-eruptive phase compositions are provided by the phenocryst rims in equilibrium with the interstitial glass (or alternatively glass inclusions if they were not compositionally modified after entrapment). The plagioclase phenocrysts show rim compositions of An35–40 (Fig. 3b) that turned out to be impossible to reproduce under realistic storage conditions, as evidenced experimentally below. Rather, the plateau range of An46–52 present in most of the plagioclase phenocrysts seems to represent the pre-eruptive compositions, whereas compositions <An40 may reflect crystallization stages at shallower levels. Contrary to plagioclase, the other phenocrysts do not show significant compositional zonings, possibly reflecting their slower crystallization kinetics or lower compositional sensitivity to (lower) pressure and melt H₂O content. The residual glasses of the erupted products may have been compositionnally modified by syn-eruptive crystallization during magma ascent in the volcanic conduit (e.g. phenocryst rims and microlites). In contrast, melt trapped as inclusions by phenocrysts growing in the reservoir may be more representative of the pre-eruptive melt composition. Yet, microlite-bearing residual glasses and glass inclusions globally cover the same compositional range (Fig. 7), suggesting a similar compositional evolution. In particular, the Fe–Mg partition coefficients between glass inclusion and host clinopyroxene give a mean value of 0.11 ± 0.03, whereas reaching 0.25 ± 0.04 for the experimental melt–clinopyroxene pairs (in agreement with the value of 0.28 ± 0.09 given by Putirka, 2008), which likely suggests some compositional modification of the melt inclusions after entrapment.

Therefore, the experiments should reproduce 7–20 wt% phenocrysts consisting of An49 ± 3 plagioclase, amphibole with 12 ± 1 wt% Al₂O₃ (Am12 ± 1), Wo46 ± 2 clinopyroxene, Mt72 ± 5 magnetite, Ilm65 ± 2 ilmenite, and apatite in a glass with less than 65 ± 1 wt% SiO₂ and up to 5.5 wt% H₂O, as shown by the glass inclusions (keeping in mind though that they were mostly modified after entrapment).

Pairs of coexisting Mt and Ilm, either trapped in phenocrysts (clinopyroxene or amphibole) or as individual phenocrysts, can be used to constrain the temperature (T) and oxygen fugacity (fO2) prevailing at the time of their crystallization, provided they co-crystallized from the same parental melt. Pairs that were positively tested for chemical equilibrium (after Bacon and Hirschmann, 1988) were used to calculate T– fO2 values following the formulation of Sauerzapf et al. (2008). Since this formulation is not calibrated originally for trachyandesitic to trachytic compositions, we tested it against the experimental oxides that crystallized under controlled T and fO2. The results give T estimates up to 40 °C and fO2 up to 1.3 log units, but mostly agree within 0.3 log units with the experimental conditions (Appendix D1), so that we validated the method for the Pavin compositions. The calculations on the natural Mt–Ilm pairs give T between 850 and 970 °C and fO2 in the range from NNO+1.3 to +1.9

No thermodynamic model is adequately calibrated to estimate total pressure (P) for trachyandesite compositions and phase assemblages such as the Pavin ones. Nevertheless, the formulations of Ridolfi and Renzulli (2012), based on amphibole composition, were tested. Compared to the run P, the calculations give significant and systematic P overestimations by over 100 MPa (e.g., 105 to 304 ± 50 MPa; Appendix D2). Such calculations also overestimate T by 41–146 °C, whereas underestimating fO2 by about 1 log unit (Appendix D2). Therefore, we cannot consider the formulations of Ridolfi and Renzulli (2012) relevant for amphiboles crystallizing from the Pavin trachyandesite.

The pre-eruptive H₂O content may be inferred from H₂O dissolved in glass inclusions, assuming volatile tightness after reservoir exit. Although the large variability in H₂O contents (1.1–5.3 wt%; Fig. 8) likely suggests that not all glass inclusions remained sealed after reservoir exit, the H₂O-richer ones suggest entrapment pressures of ~120 MPa (Di Matteo et al., 2004; minimum value if not H₂O saturated).

The juvenile pumice shows a great variability in many petrological features as shown above, notably phenocryst proportions, amphibole and clinopyroxene compositional scatter, and plagioclase zonings. Such heterogeneities likely point to variability in the differentiation process, in terms of liquid/crystal ratio and intensive parameters. This is further demonstrated by the inability of any of our phase-equilibrium experiments to fulfil the observed natural assemblage and phase compositions. Actually, the inferred intensive parameters can only be bracketed, the somewhat limited range of parameters fixed above representing the conditions for the mean assemblage and compositions. Indeed, the delimited P–T field for the Pavin storage conditions is tightly bordered by the clinopyroxene, ilmenite, and biotite appearance curves, so that slight P–T–fO2 variations may drastically modify the phase assemblage and the liquid/crystal ratio (e.g., decreasing T to 950 °C leads to biotite crystallization and total crystallinities of ~70 wt%; Fig. 9c)

The Pavin volcano is an isolated monogenetic crater of trachyandesitic composition, suggesting that it is fed by a short-lived reservoir of likely small volume. Small polycrystalline cumulates and free cumulative crystals are abundant in the uppermost part ash fraction of the pyroclastic sequence, suggesting that the eruption tapped lower cumulative parts of the reservoir. This highlights that the reservoir was likely emptied and thus of small volume. We argue that the observed variability in phase compositions is well explainable if considering a small volume reservoir where magma evolves under strong gradients of physical parameters (notably temperature that in turn controls phase assemblage and magma crystallinity). In this respect, the Pavin case contrasts with many documented large volume silicic eruptions, notably ignimbritic ones, where phase assemblage and compositions are more homogeneous, likely resulting from the buffering effect of large magma volume on the controlling physicochemical parameters.

The incompatibility between the plagioclase rim compositions of An32–40 (Fig. 3b) and the experimental compositions in the 150–400 pressure range (Fig. 9) argues for plagioclase crystallization upper in the conduit. Some of the plagioclase rims show a gradual An decrease, whereas others show a drastic drop (Fig. 3b), suggesting that the magma may have reached this intermediate storage level following different ascent rates. Similarly, some clinopyroxene-hosted melt inclusions preserved the pre-eruptive melt H₂O contents of ~4.5–5.5 wt%, whereas most of them are poorer in H₂O (Fig. 8), likely reflecting imperfect tightness and connection with the residual liquid during magma ascent (Humphreys et al., 2008). Most of the glass inclusions contain ~3 wt% H₂O, suggesting quench pressures around 50 MPa (Di Matteo et al., 2004). Such a melt inclusion re-equilibration pressure is compatible with crystallization of An32–40 plagioclase. Therefore, we assume that parts of the magma crystallized at an intermediate level of ~50 MPa (i.e. 2.0 km depth), whereas other batches ascended fast enough to preserve the pre-eruptive phase compositions and H₂O contents. The 0.3 to 1.2 wt% H₂O of the residual glasses mark a later degassing that is not associated with crystallization, probably due to the slow kinetics and large crystal nucleation delays prevailing at low pressures (Mollard et al., 2012). The residual glass H₂O contents below 1.2 wt% reflect pumice clast quenching at fragmentation depth below 1 km (P ≤ 20 MPa; Fig. 10)