Petrological and experimental evidence for differentiation of water-rich magmas beneath St. Kitts, Lesser Antilles

St. Kitts lies in the northern Lesser Antilles, a subduction-related intraoceanic volcanic arc known for its magmatic diversity and unusually abundant cognate xenoliths. We combine the geochemistry of xenoliths, melt inclusions and lavas with high pressure–temperature experiments to explore magma differentiation processes beneath St. Kitts. Lavas range from basalt to rhyolite, with predominant andesites and basaltic andesites. Xenoliths, dominated by calcic plagioclase and amphibole, typically in reaction relationship with pyroxenes and olivine, can be divided into plutonic and cumulate varieties based on mineral textures and compositions. Cumulate varieties, formed primarily by the accumulation of liquidus phases, comprise ensembles that represent instantaneous solid compositions from one or more magma batches; plutonic varieties have mineralogy and textures consistent with protracted solidification of magmatic mush. Mineral chemistry in lavas and xenoliths is subtly different. For example, plagioclase with unusually high anorthite content (An≤100) occurs in some plutonic xenoliths, whereas the most calcic plagioclase in cumulate xenoliths and lavas are An97 and An95, respectively. Fluid-saturated, equilibrium crystallisation experiments were performed on a St. Kitts basaltic andesite, with three different fluid compositions (XH2O = 1.0, 0.66 and 0.33) at 2.4 kbar, 950–1025 °C, and fO2 = NNO − 0.6 to NNO + 1.2 log units. Experiments reproduce lava liquid lines of descent and many xenolith assemblages, but fail to match xenolith and lava phenocryst mineral compositions, notably the very An-rich plagioclase. The strong positive correlation between experimentally determined plagioclase-melt KdCa–Na and dissolved H2O in the melt, together with the occurrence of Al-rich mafic lavas, suggests that parental magmas were water-rich (> 9 wt% H2O) basaltic andesites that crystallised over a wide pressure range (1.5–6 kbar). Comparison of experimental and natural (lava, xenolith) mafic mineral composition reveals that whereas olivine in lavas is predominantly primocrysts precipitated at low-pressure, pyroxenes and spinel are predominantly xenocrysts formed by disaggregation of plutonic mushes. Overall, St. Kitts xenoliths and lavas testify to mid-crustal differentiation of low-MgO basalt and basaltic andesite magmas within a trans-crustal, magmatic mush system. Lower crustal ultramafic cumulates that relate parental low-MgO basalts to primary, mantle -derived melts are absent on St. Kitts. Electronic supplementary material The online version of this article (10.1007/s00410-017-1416-3) contains supplementary material, which is available to authorized users.


Introduction
Arc magmatism above subduction zones involves chemical differentiation of mantle-derived basaltic magmas Abstract St. Kitts lies in the northern Lesser Antilles, a subduction-related intraoceanic volcanic arc known for its magmatic diversity and unusually abundant cognate xenoliths. We combine the geochemistry of xenoliths, melt inclusions and lavas with high pressure-temperature experiments to explore magma differentiation processes beneath St. Kitts. Lavas range from basalt to rhyolite, with predominant andesites and basaltic andesites. Xenoliths, dominated by calcic plagioclase and amphibole, typically in reaction relationship with pyroxenes and olivine, can be divided into plutonic and cumulate varieties based on mineral textures and compositions. Cumulate varieties, formed primarily by the accumulation of liquidus phases, comprise ensembles that represent instantaneous solid compositions from one or more magma batches; plutonic varieties have mineralogy and textures consistent with protracted solidification of magmatic mush. Mineral chemistry in lavas and xenoliths is subtly different. For example, plagioclase with unusually high anorthite content (An ≤100 ) occurs in some plutonic xenoliths, whereas the most calcic plagioclase in Communicated by Othmar Müntener.

3
98 Page 2 of 32 to intermediate and silicic compositions. Differentiation involves a combination of fractional crystallisation, and crustal melting and assimilation, the relative importance of which varies within and between arcs. Erupted arc magmas represent the integration of differentiation processes that begin with melting in the mantle wedge and, consequently, provide information only on an end product. A complementary approach is to focus on xenoliths brought to the surface during eruptions (e.g. Baker 1968;Arculus and Wills 1980;Conrad et al. 1983;Kay and Kay 1985;Hickey-Vargas et al. 1995;Ducea and Saleeby 1998;Costa et al. 2002;Dungan and Davidson 2004;Mcleod et al. 2013;Yamamoto et al. 2013;Smith 2014;Haase et al. 2014;Price et al. 2016). Such xenoliths may include cumulate residues from crystal fractionation, plutonic equivalents of erupted magma, or fragments of crustal rocks. Here we focus on xenoliths found on the Lesser Antilles island of St. Kitts. By integrating insights from xenoliths with the geochemical record of volcanic rocks and with high pressure-temperature phase equilibrium experiments, we develop an image of the subvolcanic magma plumbing system beneath St. Kitts.

Geological setting and previous work
St. Kitts lies within the central Lesser Antilles volcanic arc formed by westwards subduction of the Atlantic oceanic lithosphere. The geology of the island is described in some detail by Baker (1968Baker ( , 1984 and Toothill et al. (2007). In brief, St. Kitts comprises four volcanic centers, the most prominent of which is Mt. Liamuiga (formerly Mt. Misery) rising 1157 m above sea-level. The oldest volcanic rocks of the island have been dated at ~ 1-2 Ma (Maury and Westercamp 1990), although radiometric ages are rather sparse. The last dated eruption, from Mt. Liamuiga, was 1800 years BP and there have been no documented eruptions since settlement in 1624. Erupted rock types range from basalt to rhyolite, with basaltic andesites and andesites dominant (Baker 1984), both as pyroclastic deposits and as lavas. Magnesium-rich basalt (≤ 7 wt% MgO) lavas, with phenocrysts of olivine, clinopyroxene, and plagioclase, occur on the north-east coast at Black Rocks. These are putative parental magmas (Turner et al. 1996;Toothill et al. 2007), although their Mg# (molar Mg/[Mg + Fe]) ≤ 0.64 are too low to have been derived directly from mantle wedge peridotite (Toothill et al. 2007). Isotopic data indicate that differentiation was dominated by fractional crystallisation processes with negligible assimilation of older sialic crust and limited (< 10%) assimilation of biogenic sediments (Toothill et al. 2007). Consequently, St. Kitts represents one extreme of arc magmatic differentiation in which the role of crustal melting is minimal.
In keeping with other Lesser Antilles volcanic islands (Arculus and Wills 1980), St. Kitts yields a large number of magmatic (or cognate) xenoliths, entrained in pyroclastic rocks. Xenoliths from St. Kitts were first described by Fels (1903) and Earle (1925). Baker (1968) notes that St. Kitts xenoliths occur originally in pyroclastic rocks, but are also preferentially weathered out and accumulate in river drainage channels (or "ghuts"). Although such samples lack geological context, in terms of their parent eruptions, their accumulation provides a means to sample a great variety of textural and chemical types. Xenoliths, with or without fragments of host lava, occur as rounded clasts ranging in size from few cm to a half a metre.
This study is based on a total of 35 St. Kitts xenoliths collected during a field campaign in 2009 and augmented by five xenoliths from the collection at Durham University. Petrological data from St. Kitts xenoliths are studied in conjunction with published whole-rock geochemical analyses of St. Kitts lavas (Baker 1984;Turner et al. 1996;Toothill et al. 2007) and new and published experimental petrology data.

Analytical
Initial petrographic analyses were carried out on forty St. Kitts xenoliths. These were subsequently divided into thirteen representative types based on mineral assemblages and textures (Table 1). Xenoliths were classified using the British Geological Society Rock Classification Scheme (1999), and Streckeisen (1976). Modal abundances of the major mineral phases for each of the 13 representative xenoliths were obtained by point counting (Table 1, Fig. 1) using a Pelcon Automatic point counter 1.8 coupled to an optical microscope. Between 1250 and 1750 points were counted for each xenolith. Volume modes were converted into mass Page 3 of 32 98 modes using mineral densities, modified where appropriate for solid solution (as determined from mineral analyses), from Deer et al. (1992).
Thin sections were imaged by Hitachi S-3500N scanning electron microscope (SEM) using backscattered electrons (BSE) obtained at 15 or 20 kV. Major element analyses were performed on a five-spectrometer Cameca SX-100 electron microprobe, calibrated on a variety of oxide and mineral standards. Analytical conditions were 15 or 20 kV acceleration voltage, and 10 nA focused beam for crystals. To minimize alkali loss during analyses of hydrous melt inclusions (MI) and interstitial glass the beam current was dropped to 4 nA and beam diameter increased to10 μm. Ferric iron contents were estimated using the stoichiometric methods of Droop (1987) for spinel, Wood and Banno (1973) for clinopyroxene, and Holland and Blundy (1994) for amphibole.

Experiments
The objective of the high pressure and temperature experiments was to investigate the differentiation conditions that led to the observed geochemical diversity in lavas and xenoliths. The most magnesian (primitive) basalts on St. Kitts have up to 7 wt% MgO and phenocrysts of plagioclase, olivine (Fo <70 ) and clinopyroxene (Toothill et al. 2007). Fractional crystallisation of these phases from such basalts can produce St. Kitts' lower MgO basalts and basaltic andesites (Toothill et al. 2007). Orthopyroxene and Fe-Ti oxide phenocrysts do not appear until the host lava is basaltic andesite with ~ 4 wt% MgO. Given the presence of orthopyroxene in many St. Kitts xenoliths, we chose an experimental starting material with slightly more evolved composition than the most MgO-rich St. Kitts basalt. In this way, we increased the likelihood of attaining multiple saturation with a variety of minerals observed in xenoliths, and generating sufficiently large pools of experimental melt to analyse by electron microprobe. Conversely, the abundance of liquidus olivine was reduced. The selected starting composition (K56) is basaltic andesite lava from Black Rocks with microphenocrysts of plagioclase (28 vol%), olivine (1.5 vol%), oxides (1 vol%) and traces of clinopyroxene and orthopyroxene (Baker 1984). The sample KS_BR1 used in this study is identical to K56 and was kindly provided by Rob Watts. The major element compositions of K56 and KS_BRl are presented in Table 2. K56 is chemically similar to a basaltic andesite from Mont Pelée, Martinique (031-22b1; Table 2) studied experimentally by Pichavant et al. (2002a, b), allowing the two sets of experimental results to be considered together.
Powdered KS_BRl was dried at 100 °C for 4 h and then fused in a 1 atm gas mixing furnace at fO 2 = NNO + 1 log unit in a platinum crucible. Two cycles of melting (2 and 4 h duration) and grinding were carried out to produce a chemically homogeneous glass, as determined by electron microprobe (Table 2). Gold capsules of 2.5-3.0 mm OD were filled with the crushed glass (15-20 μm grain size), to which volatiles were added as H 2 O ± Ag 2 C 2 O 4 to create three different starting materials with the following initial molar fractions of H 2 O/(H 2 O + CO 2 ): XH 2 O = 1, 0.66 and 0.33. The total added volatile content in each experiment was 9.5-10 wt%.
Experiments were carried out at 2.4 kbar in internally heated vessels at Université d'Orleans pressurized with Ar-H 2 mixtures (Pichavant and Macdonald 2007). Temperature was measured with two S-type thermocouples with uncertainty ± 5 °C. The thermal gradient for a 3 cm-long capsule was < 5 °C. Run duration, following Pichavant et al. (2002a, b), was between 22 and 6 h. Three experimental capsules, plus redox sensor capsule, were placed together in a thin alumina tube held by a Pt wire at the furnace hot spot.
The wire was fused electrically at the end of an experiment to achieve isobaric drop-quench at ~ 100 °C/s. Experiments were performed at a hydrogen fugacity (fH 2 ) corresponding to NNO + 1. fH 2 and fO 2 are related via the dissociation of H 2 O: An Ni-Pd fO 2 sensor (e.g. Scaillet et al. 1995;Pichavant and Macdonald 2007) was used to measure fO 2 during each run. The sensor was composed of two pellets of NiPd alloys plus Ni metal to give different initial Ni/Pd ratios. The pellets were loaded into a Pt capsule together with excess H 2 O. The metal phase was analysed after the experiment to calculate fO 2 of the sensor. For any individual charge that is H 2 O-undersaturated (aH 2 O < 1) fO 2 differs from that of the sensor by 2log aH 2 O, according to Eq. (1). We calculated aH 2 O for each charge using the method of Burnham (1979) at the measured (or estimated) H 2 O content of the glass (see below). Experimental fO 2 was then calculated from aH 2 O and fH 2 .
Quenched experimental charges were imaged by SEM and analysed for major elements by Cameca SX100 and JEOL JXA8530F electron microprobes. Larger glass pools and metallic sensors were analysed on the SX100, using the same analytical procedure as for natural samples. Run product crystals and smaller glass pools were analysed on the JXA8530F. Typical analytical conditions for minerals were 10 kV, 10 nA and 100 nm beam size; for glasses 10 kV, 2 nA and10 μm beam size. In experiments with aH2O < 1 (1) glass pools were not big enough to analyse with a defocused beam and beam size was reduced to 4 μm; for some glasses (Runs 3-3 and 4-2) we were obliged to use a focused beam, leading to significant alkali loss (< 37% relative, Table 3). Modal proportions of phases were obtained by mass balance calculations and presented in Table 3.

Petrography
Based on petrographic observations of 40 thin sections (Table A3 Supplementary) we subdivided the xenoliths into cumulates, with relatively high variance mineral assemblages that represent instantaneous solid compositions (Morse 1976), and plutonics, whose texture and composition is consistent with complete solidification of an aliquot of crystal-rich magma during cooling to the solidus [i.e. total solid composition of Morse (1976)]. This bipartite distinction is used throughout the paper. Note that the bulk compositions of cumulates are not equivalent to any magma type (Arculus and Wills 1980); their nomenclature follows Wager et al. (1960). Xenolith mineralogy is dominated by calcic plagioclase and hornblende. Minor olivine (≤ 5 vol%) is widespread, but is frequently observed enclosed by pyroxene and/or amphibole. An important feature of the St. Kitts xenoliths is the presence of two pyroxenes in more than half of the collected samples, although orthopyroxene is more common in plutonics than in cumulates. Oxide minerals occur in all samples with the exception of cumulate xenolith KS-8. Nearly half of the analysed xenoliths contain co-existing ilmenite and magnetite, although ilmenite is confined to plutonic varieties. Plutonic olivine-gabbro xenolith KS-3 is unique: it has exsolved Fe-Ti oxide pairs, with single grains showing discrete ilmenite-magnetite lamellae. Rare biotite was found in a single plutonic xenolith . Quartz was found in five plutonic samples, although it is not always easy to distinguish igneous quartz from xenocrystic/inherited quartz phenocrysts. Sulphide is a widespread accessory phase. Melt inclusions are common and were found in amphibole, orthopyroxene, plagioclase, and oxides.
The relative crystallisation order of plutonic and cumulate xenoliths, determined from textural observations, shows a consistent pattern. Olivine, when present, is always the first phase to crystallise, with the sole exception of an olivinenorite (KS-11) sample where magnetite precedes olivine. Elsewhere, magnetite is the next phase to crystallise and occurs throughout the crystallisation sequence of all xenoliths. The order of orthopyroxene and clinopyroxene appearance varies: they either co-crystallise or clinopyroxene precedes orthopyroxene, which may reflect slight differences in crystallisation temperature between the samples (e.g. Leuthold et al. 2014, Fig. 19). Unusually, in KS-3 clinopyroxene crystallises after orthopyroxene. Plagioclase never crystallises before pyroxenes and typically appears either prior to amphibole or co-crystallises with it. There are two xenoliths, KS-22 and KS-6, in which amphibole crystallised before plagioclase. Amphibole often demonstrates a two-stage crystallisation: early crystallisation alongside pyroxene or plagioclase, and late-stage, interstitial crystallisation. Where present, biotite, apatite and quartz are always last in the crystallisation sequence. Sulphide occurs only as inclusions in magnetite. Ilmenite in plutonic xenoliths cocrystallises with late-stage magnetite.
Plutonic and cumulate xenolith types display striking variation in modal proportion of minerals, textures and amount of interstitial glass. A key difference between the two types is the strong mineral zoning observed in plutonic xenoliths (Fig. 2e). Textures suggestive of mineral-mineral and mineral-melt reactions, such as symplectic and poikilitic textures, as well interstitial amphibole (Fig. 3e), are more common in plutonic xenoliths. In contrast, cumulate xenoliths are minimally zoned, with adcumulate textures and euhedral crystals (Figs. 2f and 3a). Based on textural observations and modes, plutonic and cumulate xenoliths were subdivided into seven diagnostic rock types; note that amphibole gabbros and olivine-amphibole gabbros occur as both plutonic and cumulate xenolith types.

Plutonic xenoliths
Amphibole gabbros are hypidiomorphic granular (1-2 mm grain size) with fabrics that range from well-foliated to isotropic. Amphibole occurs as both subhedral and intergranular crystals. Subhedral crystals define the foliation where present (e.g. KS-6). Intergranular amphiboles contains inclusions of plagioclase in their cores and orthopyroxene in their rims. Oxides occur as inclusions in plagioclase and amphibole but also form intergranular crystals in some samples (KS-16) with inclusions of plagioclase, amphibole and Sulphide. There is widespread evidence of amphibole-pyroxene reaction. In some samples, amphibole rims contain abundant orthopyroxene inclusions; in others, amphibole forms reaction rims around orthopyroxene and clinopyroxene. Plagioclase is strongly zoned with both concentric and patchy variants. Large plagioclase grains commonly contain inclusions of amphibole. In some samples (KS-4, KS-16) plagioclase (± orthopyroxene) forms a mortar texture composed of small, interlocking grains around larger crystals of amphibole and zoned plagioclase. Interstitial quartz may be associated with small plagioclase crystals. Melt inclusions are common.
Olivine amphibole gabbros are isotropic, hypidiomorphic granular, with grain size ≤ 0.7 mm. Orthopyroxene Page 7 of 32 98 mainly forms reaction rims around olivine (Fig. 2c), typically as symplectite intergrowths with sub-micron opaque oxides. Amphibole reaction rims around clinopyroxene grains are common (Fig. 3b) and there is persuasive textural evidence for infiltration of amphibole-forming fluids along grain boundaries. Clinopyroxene often has sieve textures with abundant melt inclusions. Plagioclase is euhedral with normal zoning and often intergrown with amphibole ( Fig. 3c). Oxides are either anhedral inclusions in amphibole or form subhedral grains up to 0.8 mm (Fig. 2c). Ilmenite forms euhedral and anhedral grains, although it may also be interstitial. Small, glassy melt inclusions are ubiquitous and present in all minerals apart from olivine (Fig. 2f).
Olivine gabbro, as represented by KS-3, is a texturally complex, amphibole-free gabbro in contact with the host lava. Lava and xenolith display similar mineralogy. The

Cumulate xenoliths
Amphibole gabbro is a well-equilibrated, layered adcumulate (KS-24, Fig. 3e) made up of two distinct layers with similar mineralogy. The coarse layer comprises euhedral amphibole (≤ 3 mm) and plagioclase (≤ 2 mm), and subhedral magnetite (≤ 2 mm). Large plagioclase grains are moderately zoned with one or more resorption rims containing abundant melt inclusions. The same minerals in the fine-grained layer have a mortar texture with grain size less than 0.3 mm around a few relict larger crystals with similar textures to the coarser layer, suggesting partial recrystallisation. Clinopyroxene forms small subhedral crystals with grain size < 0.2 mm and may also form halos around amphibole phenocrysts (e.g. KS-24; Fig. 3f).
Olivine amphibole gabbros can be subdivided texturally into mesocumulate and orthocumulate variants. Orthocumulates (KS-8, KS-15, KS-21) comprise large interlocking amphibole and large euhedral plagioclase 1-5 mm across. Olivine forms chadacrysts in amphibole (Fig. 3a) and is usually fresh apart from occasional iddingsitised grains in KS-15. Minerals in KS-8 and KS-15 are well equilibrated and rarely zoned. In KS-15 abundant magnetite forms large (1-3 mm) euhedral crystals, whereas KS-8 is distinguished by a lack of oxides. Medium-grained mesocumulate (KS-7) contains a small amphibole-rich, oxide-free xenolith enclave with mineral compositions similar to those of the host xenolith. Amphibole is poikilitic with chadacrysts of clinopyroxene, plagioclase, olivine and Al-rich magnetite. Magnetites range from large interstitial grains (≤ 1.5 mm diameter) to small euhedral grains of ≤ 200 μm. Both variants contain pockets of interstitial microvesicular glass with microlites of clinopyroxene and plagioclase.
Olivine norite (KS-11) is dominated by sub-euhedral adcumulate plagioclase, magnetite and olivine with wellequilibrated 120º grain boundaries. Poikilitic orthopyroxene is interstitial to plagioclase. There are additionally a few larger, anhedral orthopyroxenes with grain size ≤ 2 mm. Olivine is iddingsite-free and contains glassy melt inclusions. Plagioclase twin planes have some slight flexure, suggestive of deformation.
Olivine amphibole gabbronorite (KS-12) is a texturally complex, transitional type of xenolith. It displays cumulative textures, but with strongly zoned, altered minerals and although grouped with cumulates equally be described as plutonic. Euhedral plagioclase and clinopyroxene are enveloped by poikilitic amphibole. Clinopyroxene grains commonly show reaction to amphibole along grain boundaries and fractures (Fig. 3d). Conversely, some anhedral amphiboles are mantled by complex coronas of intergrown clinopyroxene and plagioclase. Magnetite comprises anhedral blebs and subhedral crystals, both as inclusions in silicate phases and interstitial grains.

Mineral and glass chemistry
Representative mineral analyses are presented in Table A1 (Supplementary). For pyroxene and amphibole, Mg# is expressed as Fe denotes total iron. Sulphides are present in many St. Kitts xenoliths, but are too small to be analysed. There is a relatively wide variation in amphibole and plagioclase mineral composition across the different xenolith types, but relatively limited chemical variation in pyroxenes, spinel and olivine.
Olivine is well preserved, rarely iddingsitized (the exception is KS-15), and ranges in composition from Fo 77 to Fo 59 . Olivine from cumulate xenoliths tends to have higher Fo. The range in xenolith olivine is comparable to that of phenocrysts in St. Kitts lavas (Fo 82-62 ; Fig. 4). Individual xenolith grains are typically homogeneous (≤ 4 mol% variation in Fo). The greatest range is found in plutonic varieties (Figs. 1 and 4) where olivine is rimmed by orthopyroxene-oxide symplectites (Fig. 2a, c). Similar textures have been attributed by Johnston and Stout (1984) to oxidation of olivine.
In terms of minor components, NiO is consistently ≤ 0.07 wt%; CaO varies between 0.01 and 0.26 wt% (Fig. 4). There is no clear correlation between Fo and CaO or NiO. The majority of olivine in cumulates overlap the CaO contents (0.15-0.22 wt%) of phenocrysts in lavas (Fig. 4a). Conversely, the CaO contents in plutonic olivine tend to be much lower (0.01-0.11 wt%), although there are two cumulate xenoliths, KS-12 and KS-11 with olivine Fo ≤70 and < 0.10 wt% CaO, akin to plutonics (Fig. 4a). MnO ranges from 0.2 to 0.9 wt%, and is negatively correlated with Fo ( Fig. 4b). MnO contents in cumulate olivine are very similar to those from lavas, whereas plutonic olivine is displaced to higher MnO. Unlike CaO, MnO in KS11 and KS12 olivine does not deviate from the rest of the cumulates.
Oxides are ubiquitous in St. Kitts xenoliths, reaching 12 vol% in KS-15. Only olivine-amphibole gabbro KS-8 is oxide-free. Oxides occur as inclusions in silicate phases, along grain boundaries, in interstitial melt and inside some melt inclusions. Individual euhedral oxides may be up to 1.5 mm (Fig. 2f).
Clinopyroxene is present in three cumulate and four plutonic xenoliths with modal proportions from 0.8 to 18 wt%. Texturally, clinopyroxene can be subdivided into the following groups: homogeneous crystals typical of cumulates (Fig. 2d); normally zoned clinopyroxene in plutonics with diopside cores and augite rims (KS-4); and clinopyroxene with non-systematic sectoral compositional variations. The latter are Al-and Ca-rich diopside and augite that are also high in Fe 3+ /ΣFe (as calculated from stoichiometry), ranging from 0.1 to 0.5.
Clinopyroxene phenocrysts from lavas and plutonic xenolith have a wide range of Mg# (69-80) with a relatively small array of Ca (0.66-0.86 apfu; Fig. 6a). The range of Mg# in clinopyroxene from xenoliths is muted (70-75). Both Ca and tetrahedral aluminium (Al iv ) decrease with decreasing Mg# (Fig. 6a, b), although the Al iv decrease is non-linear, with an abrupt drop at Mg# of 75. In contrast to olivine and spinel, there is no systematic difference between plutonic and cumulate clinopyroxenes and both varieties overlap with phenocrysts from lavas. There is a small increase of TiO 2 (0.3-1.0) with increasing Mg#. TiO 2 content in KS-7 is notably higher than all other clinopyroxenes (1.5-1.8 wt%).
Plagioclase is modally dominant (≤ 80%) in all but one xenolith, KS-21 ( Fig. 1a) and very calcic in composition, similar to other xenoliths from the Lesser Antilles (Baker 1968;Lewis 1973      Ca Fig. 6 Clinopyroxene (a, b) and orthopyroxene (c, d) compositions from lava phenocrysts (Toothill et al. 2007), xenoliths and experiments in terms of Ca (a, c) and Al iv (b, d), expressed as cations per formula unit, versus Mg#. Note high-Ca and low-Ca trends for clinopyroxene phenocrysts from lavas (a). A single experimental pigeonite with high Ca is labelled Contrib Mineral Petrol (2017) 172:98 Page 13 of 32 98 2012; Cooper et al. 2017). However, St. Kitts plagioclase attains the most calcic compositions yet recorded, reaching almost pure anorthite (≤ 99.9 mol %) in two plutonic xenoliths (KS-16 and KS-3). There are five main textural varieties of plagioclase: (1) euhedral, high-An plagioclase without obvious zoning (e.g. KS-8, Fig. 3a and Table A2 Supplementary) or melt inclusions, confined to cumulate xenoliths; (2) plagioclase with calcic cores (An >90 ) separated abruptly from less calcic (An <75 ) rims with fine-scale oscillatory zoning (amplitudes of ± 10 mol % An) and abundant glassy melt inclusions (Fig. 3c); (3) normally-zoned plagioclase with almost monotonic decline from calcic core (An >90 ) to An 50 rims, punctuated by occasional high amplitude (≤ 30 mol % An) calcic spikes; (4) plagioclase lacking a high-An core, but with patchy irregular zoning and tabular textures; (5) subhedral bytownite with irregular zoning. A striking difference between plutonic and cumulate xenoliths is the range in composition. Cumulates are characterised by a limited range in An (< 15 mol %), whereas plutonics may show exceptional ranges (≤ 50 mol % An), even within a single crystal, consistent with protracted in situ crystallisation. K 2 O concentrations in plagioclase range up to 0.14 wt%. Overall the variation in phenocryst compositions from lavas (core An 95 to rim An 60 ) and cumulates is less than in plutonic xenoliths (Fig. 7).
Amphibole is the second most abundant mineral in St. Kitts xenoliths, with modal proportion from 10 to 60% (Fig. 1a, Table 1), but is exceedingly rare in St. Kitts volcanics (Baker 1968;Toothill et al. 2007). Texturally, amphibole can be divided into two groups regardless of whether the xenolith is cumulate or plutonic. In the first group, amphibole forms an interlocking network of discrete, inclusion-poor grains, appearing to crystallise as a relatively late primocryst phase (Fig. 2c, e). Rarely, amphibole has a halo of clinopyroxene, suggestive of subsequent breakdown. This is especially evident where xenoliths are in direct contact with the host lava. (e.g. KS-24, Fig. 3f). In the second group, amphibole is an interstitial phase. In many samples, the interstitial amphiboles form large, optically continuous poikilocrysts, up to several mm across. Inclusions of olivine, oxides, clinopyroxene and plagioclase are common, and evidence of amphibole-clinopyroxene reaction is widespread around grain margins, along cleavage planes, or within poikilocryst interiors (Fig. 3d). This texture, which is widespread in cumulate xenoliths, is reminiscent of the distribution of residual melt, and suggestive of percolation of reactive hydrous melt or fluids through an anhydrous crystal mush, as documented in xenoliths from Martinique (Cooper et al. 2017), Grenada (Stamper et al. 2014) and the Solomon Islands (Smith 2014).
According to the classification scheme of Leake et al. (1997Leake et al. ( , 2004, most St. Kitts amphibole is magnesiohastingsite with lesser tschermakite (KS-16, KS-31 and KS-4). Mg#, calculated with Fe total, is between 52 and 76. Subhedral and euhedral amphibole is normally zoned with < 5% variation in Mg#. There is no difference in amphibole composition between clinopyroxene-bearing and clinopyroxenefree xenoliths, although plutonic and cumulate amphibole differs markedly (Fig. 8a). Plutonic amphibole has lower Mg# (64-54, Fig. 8a) and Al IV (≤ 1.8 apfu), whereas cumulate amphibole has higher Mg# (76-60) and higher Al IV (1.6-2.2 apfu). Titanium contents are in the range 0.15 and 0.40 a.p.f.u. with trends of increasing or decreasing Ti with Mg# ( Fig. 8b) according to the nature of the coexisting oxide mineralogy. Amphiboles on the increasing Ti trend come from xenolith that contain only magnetite (shown by arrows on Fig. 8b), whereas those on the decreasing trend come from samples with ilmenite. Viewed as a suite, amphibole trends resemble a fractionation sequence with Ti increasing to the point of ilmenite saturation and then decreasing.
Glass occurs as melt inclusions in crystals and as interstitial pockets, often vesiculated, in both xenolith varieties. Inclusions, ranging in size from ≤ 10 to 200 μm and from clear to brown in colour, are common in plagioclase and amphibole, and less so in olivine, pyroxenes and oxides. Inclusions often contain gas bubbles with typical volume fractions from 0 to 10%, and rarely 20-50% (Fig. 2f). Inclusions in olivine are usually devitrified or too small to analyse.
Eleven melt inclusions and 13 interstitial glasses were selected for analysis (Table A2 and Table A4 (Table A4 Supplementary). Melt inclusions from plutonic xenoliths are systematically more evolved (66.0-74.8 wt% SiO 2 ) than those from cumulates, which are consistently andesitic (63.1-66.0 wt% SiO 2 ). The occurrence of andesitic melt inclusions is surprisingly rare in global compilations of melt inclusions in volcanic rocks (Reubi and Blundy 2009). In fact, three melt inclusions with > 6 wt% H 2 O and 65-67 wt% SiO 2 lie in the compositional gap shown by Reubi and Blundy (2009).
Overall, melt inclusions describe a fractionation trend from andesite to dacite (Fig. 9). Melt compositions overlap the silica-rich end of the whole-rock compositions of erupted lavas of St. Kitts and so provide information on the more evolved end of the liquid line of descent ( Fig. 9). A striking feature of melt inclusions with < 3 wt% MgO is the trend to lower Na 2 O. All melt inclusions were analysed using a defocussed electron beam and, as there is no correlation between Na 2 O and H 2 O contents, we do not consider this to be an analytical artefact. Instead, in the absence of any Na-rich crystallising phase, this behaviour is suggestive of sequestration of Na 2 O into an exsolving volatile phase. In that case the Na 2 O maximum (~ 4.5 wt%) in whole-rocks and melt inclusions at around 3 wt% MgO would correspond to the onset of significant volatile exsolution.
Water content in melt inclusions, as measured by SIMS, ranges from 8.5 wt% to below detection. Melt inclusions in plagioclase (An 86-79 and An 59 ) show a broad range from 8.2 to 2.5 wt%, with the lower H 2 O associated with the less calcic hosts (Table A2 Supplementary , Fig. 10). CO 2 contents range from below detection to over 1000 ppm, but do not correlate with H 2 O (Fig. 10). Notably, two interstitial glasses (not plotted on Fig. 10) from KS-15 have high CO 2 (> 1000 ppm) but low H 2 O (< 1.6 wt%). Inclusions with lower H 2 O or CO 2 are not associated with the bigger vapour bubbles. The high H 2 O content of many inclusions argues against significant H 2 O loss by diffusion through the host mineral, consistent with rapid pre-eruptive entrainment of the xenoliths in their host magma.
Chlorine contents are consistently high, reaching 3000 ppm in some melt inclusions (Table A3 Supplemen tary). Sulphur contents are low (< 350 ppm) consistent with the presence of sulphide minerals in many samples. There is no clear correlation between H 2 O and any other volatiles species. However, there is a positive correlation between sulphur and chlorine and CO 2 indicative of degassing. Chlorine increases with decreasing MgO, consistent with the incompatible behaviour of Cl and limited partitioning of Cl into exsolving fluids (cf. Blundy et al. 2008).

Thermometry and oxybarometry
Studies of xenoliths from other Lesser Antilles islands (Cooper et al. 2017;Melekhova et al. 2015;Stamper et al. 2014;Tollan et al. 2012) show that T-P-aH 2 O-fO 2 conditions under which xenoliths formed can vary widely, even for a single island. Nonetheless, the majority of xenoliths formed at pressures ≤ 4 kbar from magmas with initial H 2 O content between 2.5 and 4.5 wt%, and fO 2 ranging from 0.5 to 4.5 log units above NNO. Here we use mineral and melt chemistry of St. Kitts xenoliths to constrain intensive parameters with the following oxythermobarometers (  Putirka (2016), and magnetite-ilmenite oxythermometers of Ghiorso and Evans (2008) and Andersen and Lindsley (1985). Note that amphibole-plagioclase thermometry is limited to plagioclase less calcic than An 90 (Holland and Blundy 1994), a requirement met by only three xenoliths.
With a few exceptions, we find good agreement between different thermometers (± 40 °C) applied to the same sample. Plutonic xenoliths generally record lower temperatures (890-770 °C) than cumulate xenoliths, consistent with their more complex textures and mineralogy and more evolved melt inclusions. The presence of exsolved Fe-Ti oxides in KS-3 (Fig. 2b) yields subsolidus temperatures suggestive of protracted cooling. Cumulate xenolith temperatures are in the range 1020-890 °C.
Oxygen fugacity (fO 2 ) for plutonic xenoliths, calculated from coexisting Fe-Ti oxides, lies between NNO and NNO + 1 for all but KS3, which records a significantly higher fO 2 (NNO + 2), consistent with the interpretation of symplectites around olivine (Fig. 2a) as products of oxidation, possibly associated with cooling. The calculated fO 2 values for St. Kitts lavas estimated by Toothill et al. (2007) have a similar range, ΔNNO ± 1, to the un-oxidised plutonic xenoliths. It was not possible to calculate fO 2 for the cumulate xenoliths due to lack of an appropriate assemblage. Fig. 9 Chemical composition of experimental melts (exp. melt) and solid residues (exp. solid) from this study compared with bulk-rock major element variations in St Kitts lavas (Toothill et al. 2007;Turner et al. 1996;Baker 1984), melt inclusions in phenocrysts from cumulates (this study) and lavas (Toothill et al. 2007), and cumulate xenoliths (calculated from mineral modes and EMPA data). Experimental melts from Pichavant et al. 2002a, b (Pich. et al. 2002

Apatite saturation temperatures
The melt inclusions show a trend of decreasing P 2 O 5 with increasing SiO 2 , consistent with apatite saturation. We have used the algorithm of Harrison and Watson (1983) to calculate apatite saturation temperatures for each melt inclusion. Values range from 870 to 994 °C (Table A3 Supplementary), and are generally in good agreement with mineral thermometry. For the four plagioclase-hosted melt inclusions that have retained their volatile contents, we have also calculated plagioclase-melt temperatures using the algorithm of Putirka (2005). These range from 839 to 972 °C, within 40 °C of the apatite saturation temperatures.

Volatile saturation pressures
We calculated H 2 O and CO 2 saturation pressures for melt inclusions and interstitial glasses at the calculated apatite saturation temperatures (Table A3 Supplementary) and fO 2 = NNO + 1 using the MagmaSat algorithm of Ghiorso and Gualda (2015). Values range from 0.5 to 3.8 kbar, with fluid compositions ranging from XH 2 O of 0-1. There is no correlation between calculated pressures and host mineral or xenolith textural type. However, it is striking that the interstitial glasses from KS-15 with high CO 2 and low H 2 O record similar pressures (2.6 and 2.8 kbar) to the melt inclusions. This behaviour is suggestive of flushing of the magma with CO 2 -rich fluids prior to eruption (e.g. Blundy et al. 2010), displacing the interstitial melts along isobars, but without affecting the melt inclusions due to very slow intracrystalline diffusion of CO 2 .

Multiple reaction barometry
Two of the St. Kitts cumulate xenoliths (KS17, KS7) have the assemblage spinel-clinopyroxene-olivine-plagioclase, which has been developed as a geobarometer for mafic rocks using a multiple reaction method (Ziberna et al. 2017). We calculated pressures using mineral analyses from immediately adjacent, texturally equilibrated grains. The values are 6.4 ± 1.8 kbar for KS-7 and 6.0 ± 1.2 kbar for KS-17, lying at the upper limit of volatile saturation pressures from other xenoliths. Unfortunately, there are no melt inclusions in KS17, KS7 with which to make direct comparisons of these two methods.

Experimental petrology
Experimental run conditions and phase proportions are given in Table 3, along with relative Na loss, and H 2 O and CO 2 content of quenched glasses analysed by SIMS. For runs in which volatile contents of quenched glass could not be analysed because of an abundance of crystals and very small pool sizes, H 2 O and CO 2 concentrations were estimated using MagmaSat (Ghiorso and Gualda 2015). In the four experiments where SIMS analyses were possible, the measured values and those calculated using Mag-maSat are in good agreement (Table 3). All experiments are vapour-saturated as evidenced by presence of vapour bubbles in quenched run products. In three water-saturated runs (aH 2 O = 1.0), H 2 O concentration in the melt is around 6 wt% with 16-30 ppm CO 2 , likely introduced as a trace contaminant in the starting materials. The obtained data are in very good agreement with water solubility data of Botcharnikov et al. (2006) for andesitic melts at 200 MPa and 1100-1300 °C. The composition of the coexisting fluid was determined from the composition of the fluid added to the starting material, the measured glass volatile contents and the glass fraction determined by mass balance. As expected, the equilibrium fluid composition is H 2 O-poor compared to the starting material because of the greater solubility of H 2 O compared to CO 2 .
All experiments are crystallisation experiments where crystals nucleate and grow from the melt. No reversals were performed. Consequently, equilibrium in each individual experiment cannot be proven unequivocally. However, systematic variations in melt chemistry, melt fraction and mineral assemblages with changing temperature, together with homogeneous phase compositions, suggest close approach to equilibrium. Likewise, observed crystal textures and morphologies argue against significant problems with nucleation. Phases are homogeneously distributed throughout the capsule and crystals have euhedral shapes, although in two experiments (Run1#2 and Run2#2), skeletal orthopyroxene is suggestive of rapid growth. Residuals from mass   balance calculations show that bulk silicate composition was maintained successfully with apparent Fe loss or gain less than 1%. Sodium loss is apparent in some crystal-rich runs with reduced aH 2 O. This is likely a result of focused-beam EPMA, although we cannot rule out some Na loss to the fluid, as inferred from Na systematics in natural St. Kitts glasses. No quench crystals occur in any of the run product glasses. Crystalline phases and glass were analysed by microprobe (Table 5) in all but two experiments Run3#3 and Run4#3 where the crystals and glass pools were ~ 1 micron. Amphibole in Run3#2 was too small to gain reliable analyses.

Phase relations
A phase diagram for experimental series with different aH 2 O is shown in Fig. 11a ; Fig. 11b). The 2.4 and 4 kbar phase diagrams have very similar topology. Their watersaturated crystallisation sequences are alike (olivine-clinopyroxene-plagioclase-amphibole), although the 4 kbar experiments lie closer to multiple saturation, with the latter three phases appearing within ~ 10 °C of the liquidus. Magnetite is present in all experiments and is a near-liquidus or liquidus phase. Clinopyroxene reacts out between 975 and 950 °C at both 2.4 and 4 kbar. In this region, amphibole is the only ferromagnesian mineral at elevated H 2 O melt .
The H 2 O-undersaturated liquidus is displaced to higher temperatures, but was not determined at either 2.4 or 4 kbar. With decreasing aH 2 O, plagioclase saturation is displaced to higher temperatures and amphibole to lower temperature. The limiting H 2 O melt content for amphibole stability is 6 wt% at 4 kbar. At 2.4 kbar, the lower H 2 O melt content for amphibole stability is less well defined, but likely lies just below 4 wt%. The field of clinopyroxene stability expands with decreasing H 2 O melt . A striking effect of reduced aH 2 O is the appearance of orthopyroxene for H 2 O melt contents below 5 wt% at 2.4 kbar and 6.5 wt% at 4 kbar. The olivine stability field is very restricted, reflecting the relatively low MgO content of the starting composition. Olivine is present only at 1025 °C, 2.4 kbar and at 1000 °C, 4 kbar with H 2 O melt of 3.8 and 6.8 wt%, respectively. Ilmenite is a low temperature phase, appearing below 960 °C at both pressures, but stable to higher H 2 O melt at 4 kbar. The compositions of experimental run products show strong variation with intensive parameters ( Table 5). The degree of crystallisation increases non-linearly with decreasing aH 2 O (Table 3).

Oxides
Titaniferous magnetite (TiO 2 ~ 7-18%) is the dominant oxide phase forming euhedral crystals 1-10 μm in size in all experiments. In Run4#3, magnetite coexists with ilmenite. The TiO 2 content of magnetite increases with decreasing H 2 O melt (Table 5) and there is a slight negative correlation of TiO 2 with temperature. Al# in magnetite increases and Fe 3+ # decreases with increasing temperature and H 2 O melt . Al# and Fe 3+ # concur with the data of Pichavant et al. (2002a, b) at comparable fO 2 (Fig. 5). However, magnetite from this study is higher in TiO 2 compared to magnetite from Pichavant et al. (2002a, b), probably because of slight differences in TiO 2 content of starting compositions (Table 2). Magnetite in experiments HAB21, HAB20 and HAB23, HAB24 of Pichavant et al. (2002a, b) has a low-Al composition similar to that from KS-3 (Fig. 5) and distinct from the rest of the experimental magnetites. These four experiments were run under relatively oxidised conditions (ΔNNO + 3.1 to + 3.8) consistent with the textural inference that KS-3 experienced oxidation during differentiation, driving spinel towards magnesioferrite composition. Experimental magnetite compositions match natural phenocrysts.
Olivine (Fo 65 ) is present in one experiment at 2.4 kbar. It is high in CaO and relatively high in MnO (Fig. 4). The 4 kbar olivine (run HAB7) of Pichavant et al. (2002a, b) is slightly more magnesian (Fo 69 ) but a little lower in CaO and MnO. Experimental olivines match phenocrysts from St. Kitts lavas and cumulate xenoliths, but differ from plutonic xenoliths in terms of CaO and MnO contents (Fig. 4).

Pyroxenes
Clinopyroxene compositions range from diopside to augite. In Run2#3, diopside coexists with pigeonite. Clinopyroxene is high in Al 2 O 3 (Al IV from 0.08 to 0.21 apfu) and correlates positively with Mg# (Fig. 6b) and H 2 O melt , which probably reflects the delay in the onset of plagioclase crystallisation. Ca contents are strongly correlated with Mg# (Fig. 6a). Clinopyroxene in the 4 kbar experiments of Pichavant et al. (2002a, b) has higher Ca and Mg# than at 2.4 kbar (Fig. 6a). Orthopyroxene is enstatite with low Al 2 O 3 (1.8-2.7 wt%). There is a decrease of Al IV and Mg# and an increase of TiO 2 with deceasing aH 2 O (Table 5, Fig. 6c). Ca increases with decreasing Mg# (Fig. 6d). In general, clinopyroxene that coexists with orthopyroxene has lower Ca contents, due to buffering along the pyroxene solvus. Coexisting pyroxenes from Runs #1-2 and 2-2 yield two-pyroxene temperatures, using Eq. (37) of Putirka (2008), of 1056 and 1028 °C, respectively. The Mg# range of experimental and  Pichavant et al. 2002a, b) and at 2.4 kbar, 1000 °C (Run2#1). Plagioclase compositions range from An 58 to An 74 at 2.4 kbar (Fig. 1b) and from An 62 to An 89 at 4 kbar (Pichavant et al. 2002a, b). Plagioclase shows a negative correlation between K 2 O wt% and An (Fig. 7). Pichavant et al. (2002a, b) used a starting composition that had about 50% more K 2 O than KS_BR1 (Table 2), resulting in elevated plagioclase K 2 O content. Anorthite content increases with increasing H 2 O melt at a given temperature, but does not change significantly with temperature at fixed H 2 O melt (Fig. 11) Fig. 11b plagioclase is stable to higher temperatures than 950 °C at 8.5 wt% H 2 O melt , so plausibly even more calcic plagioclase could form under these conditions. However, we consider it unlikely that plagioclase with more than 99 mol % An could ever crystallise from Na-bearing silicate melts, unless the topology of the plagioclase binary changes dramatically at An-rich compositions, as proposed by Nekvasil et al. (2015).
Melt compositions change systematically from basaltic andesite through andesite to dacite with increasing crystallinity (Fig. 9). The effect of aH 2 O on plagioclase saturation imparts a strong influence on Al 2 O 3 and CaO contents of experimental melts. At a given MgO content, Al 2 O 3 and CaO in the melt increase linearly with H 2 O melt . Overall there is very close agreement between experimental melts  (Table 3), this makes melt composition strongly dependent on redox state. For example, the highly oxidised experiments (> NNO + 3) of Pichavant et al. (2002a, b) produced a magnetite-rich solid assemblage and correspondingly Fe-poor melt. The trend of decreasing Na 2 O at MgO contents < 3 wt% observed in lavas and natural glasses is reproduced in the experiments, although as noted above this may have its origins in Na loss from low melt fraction experimental glasses.

Petrogenesis of St. Kitts xenoliths
Xenoliths, lavas and melt inclusions combined with experimental results provide constraints on magmatic processes and conditions beneath St. Kitts. Thus far, we have shown that: 1. The lavas, xenoliths, and experimental assemblages have a similar mineralogy, though lavas lack amphibole phenocrysts, orthopyroxene is more common in plutonic xenoliths and lavas than in cumulate xenoliths, and olivine is rare in experiments; 2. The prevalence of zoning in minerals in plutonic xenolith and lavas testify to a more complex magmatic his-tory than the relatively unzoned cumulate xenolith minerals; 3. Some minerals (e.g. olivine, amphibole, oxides) are compositionally similar in experiments and specific sets of natural samples (i.e. lavas, cumulates or plutonics), whereas others (plagioclases, pyroxenes) are not; 4. Intensive parameters for xenoliths and lavas are very variable, as are estimates of fluid composition (XH 2 O); 5. Experimental melt compositions are broadly consistent with lavas regardless of P, T or aH 2 O.
In this section, we draw inferences from these similarities and differences in terms of magmagenesis on St. Kitts. Alkali contents are very sensitive to melt fraction and crystallising assemblages. Extensive crystallisation of pyroxene and delayed crystallisation of amphibole in runs with aH 2 O = 0.66 led to higher total alkalis. The trend of decreasing Na 2 O in natural samples with < 2.5 wt% MgO may be a consequence of partitioning of sodium into a coexisting vapour phase. Unfortunately, our experimental data, some of which are compromised by Na-loss during EMP analysis, alone do not allow us to quantify Na fluid-melt partition coefficients with any precision. Toothill et al. (2007) demonstrated that melt inclusions in clinopyroxene, orthopyroxene, amphibole and plagioclase from lavas plot at higher SiO 2 and total alkali abundances and lower Al 2 O 3 , CaO, TiO 2 and FeO total than the host lavas. Cumulate-hosted melt inclusions in orthopyroxene, plagioclase, amphibole and ilmenite replicate the melt inclusion lava trends. Overall, melt inclusions in phenocrysts from lavas and cumulates show compositions consistent with them being fractionated products of basaltic andesitic magmas similar to the experimental starting composition. Toothill et al. (2007) and Macdonald et al. (2000) identified two distinct lava trends in St. Kitts, one characterised by higher Al 2 O 3 and CaO and the other with lower Al 2 O 3 and CaO (Fig. 9). The two trends diverge at approximately 4.5 wt% MgO, but both lineages extend to andesitic compositions ( Fig. 9). At ~ 3 wt% MgO the high-Al group haa ~ 21 wt% Al 2 O 3 , while the low-Al group has ~ 18 wt% (Table 6 of Toothill et al. 2007). Macdonald et al. (2000) proposed that high-Al 2 O 3 trend is the result of delayed plagioclase crystallisation, whereas enrichment in CaO is due to delayed clinopyroxene crystallisation. Conversely, Toothill et al. (2007) suggested that high-Al group derives from a "genuinely more aluminous [parent] magma type". Neither set of experiments was able to reproduce the highest Al and Ca basalts reported by Toothill et al. (2007). However, at a given MgO content our experimental melts show a striking correlation between Al 2 O 3 and H 2 O melt owing to the ability of water to suppress plagioclase saturation and hence prevent Al enrichment in derivative melts (Pichavant and Macdonald 2007, their Fig. 6). Taking only experimental melts with 3.0-4.2 wt% MgO (on an anhydrous basis) we find the following linear relationship (concentrations expressed at wt%): Toothill et al's (2007) high-Al group would require H 2 O melt of 10.6 ± 1.0 wt% and their low-Al group 4.6 ± 0.4 wt% (Fig. 12a). Thus, the primary control on the different trends observed by Toothill et al. (2007) can be ascribed simply to differences in magmatic H 2 O contents obviating the need to invoke more (and less) aluminous parent magma types. The two groups may originate from the same H 2 O-rich parental magma differentiated under watersaturated conditions at different crustal depths (pressures). For a typical St. Kitts basalt with 3.5 wt% MgO, 10.6 wt% H 2 O melt would correspond to saturation at 5.8 kbar and 4.6 wt% H 2 O melt to 1.6 kbar, based on calculations using MagmaSat (Ghiorso and Gualda 2015). On this basis, we propose that the different Al 2 O 3 trends observed at St. Kitts represent different differentiation pressures of one or more water-rich basaltic andesite magmas that are themselves products of differentiation of more magnesian basalts generated in the mantle wedge. The proposed pressure range is consistent with that determined from cumulate (1.0-6.4 kbar) and melt inclusion (0.5-3.9 kbar) barometry. Our failure to reproduce the high-Al lava trend experimentally then stems simply from a failure to perform water-saturated experiments at sufficiently high pressure.

Xenolith whole-rock chemistry
We show also in Fig. 9 whole-rock major element compositions of plutonic and cumulate xenoliths calculated from point-counted mineral modes and mineral compositions, as well as the experimental solid residues recalculated from phase proportions and mineral compositions. As expected, experimental residues lie on the low-SiO 2 extrapolation of tie-lines linking experimental melts to the relevant starting composition. In contrast, neither plutonic nor cumulate xenolith compositions overlap the experimental solids or any natural rock compositions from St. Kitts, indicating that the xenoliths are not simply crystal extracts driving the observed liquid lines of descent. Chemically the xenoliths cluster into two groups in keeping with the textural criteria used to subdivide them.
Cumulate xenoliths are significantly richer in Al 2 O 3 and CaO and poorer in Na 2 O and SiO 2 than experimental residues (Fig. 9). This discrepancy, which is surprising in light of the close match of the experimental liquids to St. Kitts lavas, arises because the experimental assemblages consistently have higher proportions of clinopyroxene, at the expense of amphibole, than the cumulates (Fig. 1). We consider three possible explanations for the discrepancy: in the first the cumulates are products of crystallisation of a parent magma less evolved than the basaltic andesite experimental starting material; in the second differentiation took place at pressures higher than the experiments (see above); finally, cumulate compositions have been modified by cryptic chemical interaction with migrating melts and/or fluids in the magmatic system, which converted early formed clinopyroxene into amphibole. All three explanations have merit, and will be explored more fully below. The plutonic xenoliths also plot outside the range defined by whole rock data of St. Kitts lavas, having lower SiO 2 and total alkalis, and higher Al 2 O 3 and CaO (Fig. 9). Consequently, the plutonic xenoliths cannot be simply solidified versions of erupted lavas. Nonetheless, the complex mineral zoning and reaction textures and diverse mineral assemblages of the plutonic xenoliths are suggestive of protracted crystallisation of magma and reaction with trapped melt. Interestingly, the bulk compositions of the plutonic xenoliths lie intermediate between the cumulate xenoliths and the lavas themselves. Thus, a simple explanation for the plutonic xenoliths is that they represent mixtures of cumulates (irrespective of their origin) with varying proportions of trapped melt of the type represented by the diverse erupted melts. The plutonic xenoliths are, in effect, fragments of magmatic mush, in which migrant melts have become trapped, driving a plethora of reactions. Protracted cooling and crystallisation of these mushes within the sub-volcanic reservoir can account for both their complex textures and their relatively low crystallisation temperatures. Temperatures of olivine-clinopyroxene equilibrium for experiments, xenoliths and lavas calculated using Ca-in-olivine thermometer of Shejwalkar and Coogan (2013). Symbols as in Fig. 4 with grey field to denote olivine + clinopyroxene-bearing lavas. Note the lower temperatures of plutonic xenoliths compared to lavas, experiments and most cumulate xenoliths. c Temperatures of coexisting orthopyroxene and clinopyroxene from experiments, lavas and xenoliths plotted in terms of Ca, Mg and Fe cations per formula unit. Tie-lines connect coexisting pyroxenes from experiments (labelled with experimental temperature). A single tie-line for a cumulate xenolith (KS-12) is shown. Note the lower temperature of lava and plutonic xenolith clinopyroxenes relative to experiments as shown by relatively high Ca. d Calculated An content of plagioclase as a function of H 2 O saturation pressure (pH 2 O) for three different molar Ca/Na ratios using the Kd Ca-Na parameterisation in Eq. (3a). The upper axis indicates corresponding H 2 O melt calculated for a representative basaltic andesite melt at 1000 °C using MagmaSat (Ghiorso and Gualda 2015). Grey bars denote plagioclase cores from cumulate xenoliths and phenocryst rims from lavas 98 Page 24 of 32

Mineral compositions
There are significant mismatches between experimental mineral assemblages and compositions, lavas and xenoliths. In this section, we explore possible explanations for these differences using compositions of phenocrysts from St. Kitts lavas (Toothill et al. 2007;Turner et al. 1996;Baker 1984), minerals from xenoliths, and experimental run products.
Olivine phenocrysts in lavas are in the range of Fo 83-63 with CaO content > 0.13 wt%, whereas xenolith olivines lie in the range Fo 80-58 , but with much more variable CaO (0.02-0.21 wt%; Fig. 1a). The lowest CaO content olivines are found in plutonic xenoliths. Experimental olivines lie within the field defined by phenocrysts and most cumulate xenoliths, although their Fo contents (65-70 mol % Fo) are at the lower end of the natural range. Evidently some xenolith and lava olivines crystallised from magmas slightly more magnesian than our basaltic andesite starting composition.
Olivine CaO depends on a variety of magmatic parameters, e.g. melt composition, temperature, pressure (e.g. Jurewicz and Watson 1988;Köhler and Brey 1990;Kamenetsky et al. 2006;Mysen 2004). However, in the presence of clinopyroxene, the CaO content of olivine is controlled primarily by temperature (Köhler and Brey 1990;Shejwalkar and Coogan 2013). In Fig. 12b we have contoured the CaO versus Fo plot (Fig. 4a) for temperature using the thermometer Eq. (12) of Shejwalkar and Coogan (2013). The high CaO contents of olivine in experiments, lavas, and some cumulate xenoliths are consistent with crystallisation from relatively high-temperature melts (1025-1125 °C). Thus, phenocrystic olivines appear to be high-temperature primocrysts crystallised at or close to the liquidus. Conversely, all plutonic xenoliths and some cumulates have lower CaO contents consistent with re-equilibration with clinopyroxene down to temperatures as low as 800 °C. This supports our contention that plutonic xenoliths (and some cumulates) have undergone cooling and solidification in the sub-volcanic reservoir. In the case of KS-3 olivine oxidation, to form orthopyroxeneoxide symplectites, may have increased olivine Fo content slightly (Johnston and Stout 1984). Oxidative increase in Fo can also account for the displacement of these olivines on the MnO-Fo plot (Fig. 4b). Thus, we suggest that the distinctive behaviour of plutonic olivine is a consequence of partial modification of their Fo and CaO contents during both cooling and oxidation.

Oxides
There is close compositional correspondence between experimental spinels and those in cumulates, although the latter extend to much higher Al# and lower Fe 2+ /(Mg + Fe 2+ ) (Fig. 5). The higher Al# of cumulate spinels likely reflects the higher Al content of the coexisting melts, which we attribute above to elevated magmatic H 2 O contents suppressing plagioclase crystallisation relative to olivine and clinopyroxene. Thus, cumulate spinel testifies to crystallisation from slightly wetter and more primitive magmas than the basaltic andesite starting material. Spinel phenocrysts in lavas overlap with low Al# end of the experimental and cumulate spinel, and provide a good match with spinel in plutonic xenoliths that lack evidence for oxidation.
Pyroxenes in cumulate and plutonic xenoliths and lavas show almost complete overlap in terms of Mg# for clinopyroxene (Fig. 6a) and orthopyroxene (Fig. 6c). Experimental clinopyroxenes intersect the xenolith array at high Mg#, but define differentiation trends to lower Ca and higher Al IV . Experimental orthopyroxene is displaced to slightly higher Ca and Al IV contents. The Ca content of coexisting pyroxenes is buffered by the pyroxene solvus, which forms the basis of two-pyroxene thermometry (e.g. Wood and Banno 1973;Wells 1977;Lindsley 1983). Lower Ca in clinopyroxene and higher Ca in orthopyroxene reflect higher equilibration temperatures. Tie lines linking selected experimental and xenolith pyroxene pairs (Fig. 12c, Table 4) are consistent with Kd Fe-Mg between orthopyroxene and clinopyroxene being slightly greater than one (Putirka 2008). Figure 13c indicates that the experimental pyroxene pairs are consistent with higher crystallisation temperatures (1000-1025 °C) than cumulates (− 900 °C), plutonics or lavas. The latter show very scattered compositions, even within a single sample (Toothill et al. 2007), that overlap with those of plutonic xenoliths, whose two-pyroxene crystallisation temperatures are in the range 770-940 °C ( Table 4). The Ca-rich nature of phenocrystic clinopyroxene is therefore consistent with them being xenocrysts appropriated from cooler pockets within the magmatic mush system. A few lava clinopyroxene phenocrysts extend to lower Ca, suggestive of high-temperature crystallisation from the host magma. However, in contrast to olivine, the majority of phenocrysts record temperatures too low to reflect an origin as true primocrysts. The higher Al IV contents of experimental pyroxenes (Fig. 6b, d) may also reflect higher crystallisation temperatures.

Amphibole
St. Kitts lavas are almost all amphibole-free, a common feature of Lesser Antilles islands with the exception of Grenada, Guadeloupe, and Saba (e.g. Arculus 1976;Westercamp and Mervoyer 1976;Baker 1980). Amphiboles in plutonic and cumulate xenoliths are texturally similar, but compositionally distinct, notably in Mg# and Al IV , which are proxies for melt composition (and/or fO 2 ) and crystallisation temperature, respectively (Blundy and Holland 1990). Plutonic amphiboles have lower Mg# and Al IV (Fig. 8a, b)  low Mg# and Al IV of the cumulate amphibole array. Based on textural evidence, we propose that cumulate xenolith amphiboles (± calcic plagioclase) originate by percolative reaction between H 2 O-rich melts or fluids and anhydrous, pyroxene-dominated residues, similar to those in the experiments. The hydrous fluid may also lead to the observed olivine iddingsitisation in some cumulate xenoliths. A similar metasomatic origin for cumulate amphibole was advanced by Smith (2014) for Solomon Islands xenoliths, although their lack of plagioclase likely reflects higher temperatures and/or H 2 O contents.
We can explore possible mechanisms for generating Anrich plagioclase by considering the exchange of Ca and Na between plagioclase and melt from the experiments. For our new experiments and those of Pichavant et al. (2002a, b) we find the following exponential relationship between Kd Ca-Na (= [Ca/Na] plag /[Ca/Na] melt ) and H 2 O melt (Fig. A1  Supplementary): There is no significant temperature effect over the range of experimental temperatures studied. The slope of this relationship is similar to that reported by Sisson and Grove (1993a, b), although their data are offset to slightly higher Kd Ca-Na than ours, due to differences in bulk composition.  Grove et al. (1997) and Laumonier et al. (2017). Grey solid lines with An content of plagioclase are based on weighted least-squares regressions of all plotted plagioclase-bearing experi-ments. H 2 O saturation at run conditions was calculated with Magma-Sat (Ghiorso and Gualda 2015). Assemblages shown in orange and green refer to typical plutonic and cumulate xenolith assemblages, respectively, from St Kitts. Note effect of temperature on olivine and orthopyroxene stability and universal stability of clinopyroxene at 1000 °C. Mineral abbreviations as in Fig. 11 98 Page 26 of 32 We emphasise that Kd Ca-Na is very sensitive to melt composition (e.g. Sisson et al. 2005;Hamada and Fujii 2007), so Eq. (3a) should only be used for basaltic andesites similar to our experimental starting materials.
In Fig. 13d, we show the calculated variation in plagioclase An content for three different melt Ca/Na contents: the two starting materials (Table 2) and the maximum molar Ca/Na (= 2.7) of any St. Kitts lava (Kit59) as reported by Toothill et al. (2007). Kit59 is a Black Rocks basalt with Mg# = 59 and therefore in equilibrium with Fo 82 olivine, the most Mg-rich observed in any St. Kitts natural sample (Fig. 4). The calculated variation in An with H 2 O melt indicates that the highly calcic plagioclase of St. Kitts lavas and xenoliths requires magmatic H 2 O contents of 9-13 wt% (5-8 kbar saturation pressure), in good agreement with the value estimated independently from lava Al 2 O 3 contents. The lower An rims of lava phenocrysts would require H 2 O melt of < 1 wt%, consistent with them being primocrysts grown from their host magma during magma ascent and degassing.
Although high H 2 O melts may be responsible for very calcic plagioclase, it is worth mentioning other possibilities. For example, the An content of plagioclase can also be influenced by exchange of Ca and Na with a fluid phase, similar to that invoked for metasomatic amphibole crystallisation. Thus, the most calcic plagioclases in St. Kitts xenoliths may have a partially metasomatic origin that reflects dissolution and reprecipitation of plagioclase in the presence of migrating Na-poor melts or fluids. Conversely, Devine and Sigurdsson (1995) suggest that very calcic plagioclase (An >95 ) in the Lesser Antilles is the result of crustal assimilation, although Sr isotopic data (Toothill et al. 2007) argue against this on St. Kitts.

Phase assemblages
In keeping with other Lesser Antilles islands (Arculus and Wills 1980) the diagnostic St. Kitts xenolith assemblage is calcic plagioclase + amphibole + clinopyroxene + spinel ± orthopyroxene ± olivine (Fig. 1a). A particular petrogenetic challenge is identifying a stability field for calcic plagioclase + orthopyroxene + amphibole ± olivine, an assemblage that was not generated under any of the experimental conditions investigated (Fig. 11). As cumulates are snapshots of the overall differentiation process, albeit modified by metasomatic reaction, it seems likely that some part of the P-T-H 2 O evolution was not captured experimentally. To explore possible conditions under which the St. Kitts xenolith assemblages were formed, in Fig. 13 we have compiled all published experimental data on bulk compositions similar to those of St. Kitts basaltic andesite (Grove et al. 1997;Pichavant et al. 2002a, b;Almeev et al. 2013;Erdmann et al. 2016;Laumonier et al. 2017; this study) at temperatures of 950 and 1000 °C (Fig. 13). The experimental starting compositions have Mg# 41-54 and molar Ca/Na from 2.3 to 3.8. The water saturation curve in Fig. 13 was calculated using MagmaSat (Ghiorso and Gualda 2015) for a representative basaltic andesite at the experimental temperature. At each experimental temperature we regressed plagioclase An content as a function of P and H 2 O melt and contoured the plots accordingly.
Amphibole + plagioclase + clinopyroxene coexist over a wide P-H 2 O melt range (Fig. 13). Clinopyroxene is stabilised at higher temperatures; with decreasing temperature clinopyroxene and amphibole are in a reaction relationship which is terminal to clinopyroxene crystallisation at pressures of 2-5 kbar. Olivine stability is not clearly defined at either 950 or 1000 °C, although we suspect that all melts are close to olivine saturation and that for slightly higher MgO starting materials the olivine field would expand significantly. Orthopyroxene stability is diminished at high fO 2 and high H 2 O melt (Sisson and Grove 1993b). At 1000 °C orthopyroxene is only stable when H 2 O ≤ 5wt % at 4 kbar ( Fig. 13b). At 950 °C the upper pressure limit of orthopyroxene stability is unconstrained, but certainly exceeds H 2 O melt = 8 wt% and 5 kbar (Fig. 13a). The effect of H 2 O melt on plagioclase saturation is evident at 1000 °C, where plagioclase does not crystallise from melts with more than 8 wt% H 2 O, whereas at 950 °C plagioclase can crystallise from melts with almost 9 wt% H 2 O. The typical plutonic and cumulate xenolith assemblages are shown with coloured symbols. In both cases they lie at P ≥ 4 kbar. However, experimental plagioclase compositions match only the more sodic rims in the natural rocks rather than the calcic cores (Fig. 1).
Although the mineral assemblages can be reproduced at 4 kbar, 950-1000 °C and H 2 O melt ≈ 10 wt% the experimental plagioclase remains less calcic (An 82-76 ) than observed in the cumulates (An 92-96 ). As noted above, this reflects both the Ca/Na ratio of the experimental starting materials, and the tendency of plagioclase An content to increase with increasing P, T and H 2 O (Fig. 13). Anorthite content increases more rapidly with increasing H 2 O at 1000 °C compared to 950 °C. The highest An contents occur at high pressures and high H 2 O melt : An 92-96 plagioclase is stable for H 2 O melt in excess of 10 wt% at 950 °C and 6.5 wt% at 1000 °C. Co-precipitation of these calcic plagioclases with orthopyroxene, as observed in many St. Kitts xenoliths, is problematic, based on the available experimental data. The closest approach is at 950 °C at P > 5.5 kbar and H 2 O melt ≥ 10 wt%. Crystallisation of An-rich plagioclase is enhanced from melts with high molar Ca/Na, although the maximum value on St. Kitts, based on lavas is 2.7. Lava compositions need not necessarily represent true liquids, especially when crystals are entrained, yet this process will only serve to artificially elevate Ca/Na ratios. Instead we propose, as inferred above Page 27 of 32 98 based on plagioclase composition and lava Al 2 O 3 content, that xenolith assemblages on St. Kitts require crystallisation from very H 2 O-rich basaltic or basaltic andesite magmas.
To explore whether such wet magmas can crystallise plagioclase we have compiled existing experimental data on a broad range of hydrous arc magma compositions at pressures ranging from 1 to 12 kbar (Fig. 14, Fig. A2 Supplimentary) (Almeev et al. 2013;Alonso-Perez et al. 2009;Beard and Lofgren 1990;Blatter and Carmichael 2001;Blatter et al. 2013;Brooker unpublished;Cawthorn et al. 1973;Eggler and Burnham 1973;Erdmann et al. 2016;Green and Ringwood 1968;Grove et al. 1997Grove et al. , 2003Holloway and Burnham 1972;Laumonier et al. 2017;Melekhova et al. 2015, Müntener et al. 2001Nandedkar et al. 2014;Panjasawatwong et al. 1995;Pichavant et al. 2002a, b;Prouteau et al. 2001;Rapp and Watson 1995;Sisson and Grove 1993a, b;Sisson et al. 2005;Winther 1990 (thesis); Yoder and Tilley 1962). The experimental starting compositions include high magnesium basalts (HMB), low magnesium basalt (LMB), high alumina basalt (HAB), basaltic andesite (BA) and andesite (A). In Fig. 14a we subdivide plagioclase-free melts from melts saturated with plagioclase or plagioclase + orthopyroxene. Plagioclase is stable in melts with as much as 20 wt% H 2 O, provided temperatures are low. Our subdivision agrees well with the formulation of Almeev et al. (2012) shown as an orange line on Fig. 14a. At 950 °C the upper H 2 O content of plagioclase-saturated melts is 9 wt%, consistent with inferences from Fig. 13. Orthopyroxene coexists with plagioclase under these conditions. In Fig. 14b, we have subdivided the experiments on the basis of their starting materials. Plagioclase stability is clearly independent of magma composition for the different magma types considered. Assemblages identical to those of St. Kitts xenoliths form at pressures of 2-9 kbar. Basaltic andesites with 8-10 wt% H 2 O crystallise an assemblage of amphibole + plagioclase + orthopyroxene or amphibole + plagioclase + clinopyroxene at temperatures of 850-950 °C over a wide range of crustal pressures. Hydrous high alumina basalts under the same conditions crystallise the assemblage amphibole + plagioclase + olivine. Thus, the mineral assemblages of St. Kitts xenoliths are broadly consistent with crystallisation from H 2 O-rich basalts and basaltic andesites at pressures in excess of 4 kbar.

Implication for arc magmatism
Petrological, mineralogical and experimental data provide four independent lines of evidence for differentiation of water-rich (5-11 wt% H 2 O) basaltic andesite magmas at mid-crustal depths beneath St. Kitts: highly calcic plagioclase; the mineral assemblage of cumulate xenoliths; mineral barometry; and the high-alumina group of magmas. In this section we interpret these findings in the context of a vertically extensive ("transcrustal") magmatic mush within which hydrous, basaltic magmas supplied from the underlying mantle wedge crystallise and degas, and residual melts reactively percolate upwards  The elevated magmatic water contents are consistent with findings at a number of other arc volcanoes. Edmonds et al. (2016) propose 6-9 wt% H 2 O in andesites from Soufrière Hills Volcano, Montserrat, on the basis of H 2 O dissolved in orthopyroxene phenocrysts. Grove et al. (2003) proposed ≥ 11 wt% H 2 O in some primitive magmas from Mt. Shasta, Cascades, on the basis of amphibole chemistry and melt inclusions. Laumonnier et al. (2017) used a combination of geophysical and petrological evidence to argue that the andesitic partial melt contained within the Altiplano-Puna Magma Body contains 9-11 wt% H 2 O.
Our proposal for elevated H 2 O contents in St. Kitts basaltic andesites requires proportionately higher H 2 O in parental basalts. Experimental data (e.g. Nandedkar et al. 2014;Melekhova et al. 2015) show that basaltic andesites can be generated from mantle-derived MgO-rich basalts by approximately 40-60 wt% crystallisation, predominantly of anhydrous mafic silicates. In that case basaltic andesite H 2 O contents of ≤ 11 wt% would equate to ≤ 6.6 wt% H 2 O in the parental basalt. This value is within the range (≤ 7.2 wt%) of H 2 O contents of olivine-hosted basaltic melt inclusions from a broad selection of subduction zones (Plank et al 2013), for which the mean H 2 O content is 3.9 ± 0.5 wt%. Basaltic andesite H 2 O contents of > 7 wt% requires that differentiation of their parent magmas occurs at sufficient depth in the crust to hold H 2 O in solution. In the case of St. Kitts basaltic andesites ≤ 11 wt% H 2 O corresponds to pressures of ≤ 6.5 kbar (at 1050 °C). For a mean crustal density of 2660 kg/m 3 for the Lesser Antilles arc (Christeson et al. 2008) this equates to lower crustal depths of ≤ 25 km.
The basaltic andesite starting materials from our experiments have been shown to provide a good match to parental melts for some St. Kitts xenoliths, although slightly higher Mg# parents are required to generate the observed olivine Fo contents. However, adding a small amount of olivine to the basaltic andesites will not render them sufficiently magnesian to be in equilibrium with mantle peridotite. Mantle-derived magnesian basalts would need to crystallise 40-60 wt% to generate the low-magnesium basalts or basaltic andesites found on St. Kitts. The crystalline residues from this differentiation would be olivine + clinopyroxene + spinel (Müntener et al. 2001;Müntener and Ulmer 2006;Nandedkar et al. 2014;Melekhova et al. 2015) with little or no plagioclase or amphibole. We suggest that cumulate rocks with this mineralogy underlie the St. Kitts magmatic system, at depths greater than 25 km. It is worth noting that the density of seismic velocities of such ultramafic cumulate rocks would make them very hard to distinguish geophysically from mantle rocks (Müntener and Ulmer 2006). It is perhaps for this reason that the Moho discontinuity is not clearly resolved beneath the Lesser Antilles (Christeson et al. 2008;Kopp et al. 2011).
We propose that water-rich, relatively oxidised (NNO + 1 to NNO + 2) low-Mg basalts and basalt andesites generated by olivine + clinopyroxene + spinel fractionation in the deep crust ascend to 18-25 km where they begin to differentiate, producing the xenolith varieties observed on St. Kitts. Elevated H 2 O contents are required to generate the calcic plagioclase that is diagnostic of xenoliths on St. Kitts and elsewhere in the Lesser Antilles. St. Kitts xenoliths record snapshots of this polybaric differentiation process, whereas erupted magmas represent the integrated products, modified by migration through the thick crystal mush pile. Our experiments show that most of the lavas erupted on St. Kitts can be generated at pressures of 2.4-4 kbar, with H 2 O contents down to 2-6 wt%, i.e. significantly lower than those inferred for the basaltic andesite parents. The tendency for erupted magmas to match low-pressure experimental data is consistent with the vertically extensive mush concept; as melts migrate upwards they will continually equilibrate with crystals in the mush. Their apparent equilibration pressure will correspond to the top of the mush, even though differentiation began at much greater depths. Only occasionally do melts bypass the mush column and so preserve their highpressure chemistry upon eruption; the Al-rich series on St. Kitts is one such example.
We propose that the hallmark of magmas generated in transcrustal mushes will be multiple saturation with four or more of the following mineral phases: olivine, clinopyroxene, orthopyroxene, amphibole, plagioclase, and spinel. Conversely, the crystals represent snapshots of the entire, polybaric differentiation process and are, therefore, much more sensitive to local conditions. As basaltic andesite or low-Mg basalts magmas entering the base of the mush are H 2 O (and possibly CO 2 ) rich, so the melts leaving the top of the mush will be volatile-saturated. Once a significant mass of mush is established in the arc crust, it tends to buffer melt compositions, becoming relative immune to new additions of slightly different composition. We suggest that the ability of the mush to buffer melts along polybaric, multi-phase cotectics gives rise to relatively well-defined liquid lines of descent. Recent experimental studies of a Mount St Helens dacite by Blatter et al. (2017) are a case in point; they find multiple saturation with amphibole-plagioclase-orthopyroxene-clinopyroxene-oxide at 4 kbar. The abundance of amphibole within the crystal mush column drives amphibole saturation with attendant trace element signatures, despite the paucity of amphibole phenocrysts in many erupted arc magmas (Davidson et al 2007). There is abundant textural evidence in St. Kitts xenoliths for percolative, reactive flow of hydrous fluids/melts, including reactions of early-formed clinopyroxene to amphibole, mineral zoning, poikilitic crystals and disequilibrium textures. The consequences of percolative reactive flow may be especially complicated for trace Page 29 of 32 98 elements, as observed in xenoliths from Martinique (Cooper et al. 2017).
As melts ascend through transcrustal mush they will exsolve volatiles and interact with pre-existing crystals. Interactions include metasomatic reactions, such as those observed in plutonic xenoliths from St. Kitts. Ascending melts may also rip up mush fragments, either as crystal clots or as disaggregated xenocrysts. We have demonstrated that calcic plagioclase, spinel and pyroxene phenocrysts in St. Kitts lavas are typically xenocrystic (or more accurately, antecrystic) in origin. This crystal cargo may be significantly overgrown or overprinted by liquidus phases that crystallise from the host magma as it leaves the top of the mush and ascends to the surface. Olivine phenocrysts and intermediate plagioclase rims in St. Kitts lavas are examples of low-pressure primocrysts. In arc magmas, unraveling the relative proportions of xenocrysts, antecrysts and primocrysts is notoriously difficult.
The release of exsolving fluids from ascending magmas within the mush may also lead to reactions with preexisting crystals, including dissolution and reprecipitation, leaching of fluid-mobile elements (such as Na) and production of hydrous mineral phases (e.g. Smith 2014). Because of their different densities, it is likely that fluid and melt movement are decoupled (Christopher et al. 2015). The compositions of the exsolved fluid changes with depth, being more CO 2 (and SO 2 ) rich at higher pressures. Percolation of fluids from the top of the mush that have been released across a considerable vertical depth range may explain the apparent flushing of shallow-stored magmas with CO 2 . There is widespread evidence of this in melt inclusions from arc volcanoes (e.g. Blundy et al. 2010) as well as in some interstitial glasses from St. Kitts xenoliths. Discharge of SO 2 -rich fluids from the top of a transcrustal mush may play an important role in hydrothermal mineralization (Mavrogenes and Blundy 2017).
The process of melt migration through the transcrustal magmatic mush can operate on two different timescales. Slow, percolative flow can account for many of the observed xenolith textures and the tendency for erupted magmas to be chemically buffered. However, the presence of xenoliths containing volatile-rich melt inclusions indicates that a more rapid process may also operate, whereby melts disaggregate and entrain mush shortly before eruption. Christopher et al. (2015) suggest that large-scale gravitational destabilisation of the mush is one means to rapidly aggregate and release magmas to shallow levels. Conditions under which mushes become gravitational unstable, and the timescales of consequent magma ascent, are rich areas for further study.