Hornblende dehydration reactions during magma ascent at Soufrière Hills Volcano, Montserrat

Abstract

Hornblende phenocrysts in recent andesites of the Soufrière Hills Volcano display reaction rims of microcrystalline plagioclase, pyroxene, Fe-oxides and interstitial glass, formed by decompression during magma ascent. Mass balance calculations give the following reactions with mineral proportions in agreement with modal abundances:

$$\begin{aligned} 100.0\,{\text{hbl}} & \to 49.8\,{\text{cpx}} + 27.6\,{\text{opx}} + 18.2\,{\text{plag}} + 4.5\,{\text{Ti - mag}} \\ 100.0\,{\text{hbl}} & \to 53.9\,{\text{cpx}} + 24.4\,{\text{opx}} + 18.1\,{\text{plag}} + 1.9\,{\text{ilm}} + 1.7\,{\text{mag}} \\ \end{aligned} $$

These reactions require an open chemical system with exchange of selected components with surrounding melt. Volatiles, TiO₂ and alkalis are expelled and SiO₂ and FeO^T are consumed. Matrix glasses fall into two compositional groups. Glasses in pumice are relatively rich in CaO and poor in K₂O and Na₂O compared to glasses in dome samples. The former glasses formed by moderate amounts of groundmass crystallisation of plagioclase, associated with rapid magma ascent in explosive eruptions. The later glasses evolved in response to hornblende breakdown, groundmass crystallisation and mixing of melts from different levels during slow magma ascent and extended residence time in the dome. Interstitial glass compositions in reaction rims reflect the compositions of the surrounding matrix glasses, but show variable compositional differences mostly consistent with the proposed open-system reactions.

Introduction

Hornblende phenocrysts are abundant in volatile-rich orogenic andesites. Many display breakdown textures, including reaction rims of microcrystalline plagioclase, pyroxene and Fe-oxides, forming microcrystalline rims enveloping remnant hornblende cores. This texture can develop during magma ascent and decompression as hornblende becomes unstable in the presence of coexisting melt with decreasing dissolved water content. Reaction rim width is dependent on the over-steps in temperature and pressure. It is also a function of decompression rate; consequently, the potential exists to estimate magma ascent rates from rim width measurements (Rutherford and Hill 1993).

In a pioneering experimental study Rutherford and Hill (1993) measured hornblende reaction rim width as a function of time for the Mount St. Helens (MSH) May 18, 1980, dacite. They then estimated magma ascent rates for the eruption by applying this calibration to interpret reaction rim width data from natural samples. Strictly this method is only directly applicable to the specific conditions of the experimental calibration for the MSH compositions and magma ascent rates obtained for other eruptions by extrapolation outside these conditions may not be quantitatively accurate. More general application of this technique therefore requires an understanding of the kinetic controls on the rates of hornblende dehydration reactions.

One important question concerns the nature and extent of material transport during the reaction, as different elements or species are likely to diffuse at different rates. Does amphibole breakdown in a semi-closed system involve near-isochemical conversion of amphibole to the anhydrous products with the release of volatiles, or does the reaction take place in an open chemical system with diffusive exchanges of chemical components with the surrounding melt? These questions can be addressed by detailed assessment of the material balance between reactants and products.

Melt inclusions in hornblende, matrix glasses in the groundmass and glasses trapped within hornblende reaction rims may offer insights into the evolution of hornblende, and the reaction mechanism of hornblende breakdown. Hornblende reaction during decompression may not be isochemical. Thus, variations of glass chemistry in reaction rims and adjacent areas of the matrix might reflect modifications related to the hornblende breakdown reaction.

We present new compositional data and material balance estimates for hornblende breakdown in andesite from the ongoing eruption of the Soufrière Hills Volcano, Montserrat. We also re-evaluate existing mass balance data for the 1980 eruptions of MSH (Rutherford and Hill 1993). Quantitative accounts of the extent of chemical exchange between hornblendes and their surroundings are established, forming the foundation for quantitative evaluation of the rates of material transport during hornblende dehydration reactions.

Previous work

Amphibole reaction textures have long been recognised and are commonly mentioned in petrological descriptions of andesites (e.g. Williams 1942; Schmidt 1957; Jakes and White 1972; Gust and Johnson 1981; Reagan et al. 1987; Clynne 1999). Hornblende reaction textures from MSH 1980–1986 andesites (Scarfe and Fuji 1987; Rutherford and Hill 1993) and recent Soufrière Hills Volcano andesites (Murphy et al. 2000; Devine et al. 1998a, 1998b; Rutherford and Devine 2003) have received particular attention.

Rutherford and Hill (1993) investigated decompression-induced reaction rims on hornblendes in the post May 18, 1980, MSH andesite. Using hornblende and reaction rim phase compositions in a mass balance petrologic mixing programme (Wright and Doherty 1970) they found that no combination of the product phases in the reaction rim reproduces the reactant hornblende composition. Based on observations that rim growth proceeds inwards from the crystal margins, and that no breakdown occurs in the hornblende interiors, they proposed that, rather than breaking down isochemically, amphibole reacts with the surrounding rhyolitic melt. This conclusion is supported by the observation that there is no breakdown where hornblende is in contact with another crystalline phase. Their mass balance solution indicates that 73 parts amphibole react with 27 parts melt by weight to form the reaction products (orthopyroxene, clinopyroxene, plagioclase and ilmenite):

$$73\,{\text{amph}} + 27\,{\text{melt}} \to 29\,{\text{opx}} + 24\,{\text{cpx}} + 43\,{\text{plag}} + 3\,{\text{ilm}}$$

(1)

In the process the remaining melt becomes enriched in SiO₂, K₂O and Na₂O and depleted in Al₂O₃, FeO and CaO.

This result is problematic because, as shown by Rutherford and Hill (1993), no single combination of hornblende and melt can exactly reproduce the composition of the reaction rim products. Hornblende and the host melt each contain more K₂O than any of the MSH reaction rim phases, thus excess K₂O is an unavoidable by-product of hornblende breakdown. Most other components (SiO₂, TiO₂, Al₂O₃, Fe₂O₃, MnO, MgO, CaO and Na₂O) can only be balanced by reacting hornblende with a proportion of host melt; however, optimum solutions for each component require hornblende and melt to be combined in different proportions. Hornblende breakdown can be alternatively described by a reaction of the form amph → cpx + opx + plag + mag + ilm ± residuals, where residuals convey the compositional differences between hornblende and the best-fit combination of product phases. These differences are conceivably balanced by the surrounding melt. This perspective has a number of advantages. Firstly, melt is not included as a phase in mass balance calculations, so there is no requirement for all elements present in the melt to contribute to the reaction. Secondly, there is no requirement for the composition of the melt to remain fixed. Thirdly, magnetite is present in MSH hornblende reaction rims but absent from the best-fit solution for the reaction amph + melt → cpx + opx + plag + Fe-oxides (Rutherford and Hill 1993). If melt is removed from the equation magnetite is required in the solution. Considering these limitations we cannot know whether existing mass balance calculations accurately reflect material transport behaviour during MSH hornblende breakdown reactions. A more detailed investigation clarifying these issues is necessary as a basis for quantitative treatment of transport processes.

Geological setting

The 1995-to-present (August 2004) eruption of Soufrière Hills Volcano, Montserrat (WI) is the latest in a series of andesitic, dome-forming eruptions in a history dating from >170 ka (Roobol and Smith 1998; Harford et al. 2002). The current eruption is thought to have been triggered by injection of mafic magma into a pre-existing andesitic magma storage region, causing non-uniform reheating and remobilisation of the resident magma (Murphy et al. 1998; Devine et al. 1998a). Injection of hot basaltic magma and local heating of the andesite has continued periodically throughout the eruption (Couch et al. 2001; Devine et al. 2003), and essentially every sample of erupted magma shows evidence of some mingling of a hotter basaltic andesite with the pre-existing andesite (Murphy et al. 1998; Rutherford and Devine 2003). The rate of magma recharge into the chamber controls the eruption rate (Sparks et al. 1998). The current eruption has seen large fluctuations in magma discharge rate, with pauses and pulses in magmatic activity (Sparks et al. 1998; Robertson et al. 2000).

Petrological background

Andesite is the predominant eruption product throughout Montserrat’s history. The lava is a coarse-grained, porphyritic, crystal-rich hornblende–hypersthene andesite (58–61 wt% SiO₂) containing minor ubiquitous mafic inclusions of basaltic andesite (Murphy et al. 2000). The petrology and geochemistry of the andesite have been documented continuously throughout the eruption (Devine et al. 1998a, 2003; Murphy et al. 2000). No significant temporal variation has been observed in either first order petrological features or bulk magma chemistry, although banded lava is common, reflecting local differences in phenocryst proportions, crystallinity, grain size and vesicularity. The andesite contains phenocrysts and microphenocrysts of plagioclase (30–35 wt%), amphibole (6–10 wt%), orthopyroxene (2–5 wt%), titanomagnetite (2–4 wt%) and quartz (<0.5 wt%). Clinopyroxene (<0.5 wt%) occurs as microphenocrysts, and apatite and ilmenite are present in trace amounts. Disequilibrium features consistent with variable heating of the magma storage region are common. Plagioclase phenocrysts exhibit strongly resorbed sieve-textured zones and sodic cores mantled by calcic rims. Some orthopyroxene phenocrysts have Mg- and Ca-rich rims and quartz crystals are often resorbed and surrounded by clinopyroxene reaction rims. Hornblendes range from euhedral to those completely pseudomorphed by pyroxene–plagioclase-oxide intergrowths. All variants of texture and composition can be observed in a single thin section.

There is wide variation in groundmass textures. Pumice clasts are glassy with minor microlites of plagioclase and up to 35 wt% glass, whereas some dome samples have highly crystalline groundmass with as little as 5% glass, abundant plagioclase microlites, oxides, orthopyroxene and crystalline silica mainly in the form of cristobalite (Murphy et al. 2000; Harford et al. 2003).

Hornblende

Explosive pumice ejecta contain predominantly pristine, euhedral or slightly rounded amphibole phenocrysts that are vitreous and green-coloured (Fig. 1a). In more slowly erupted dome samples a range of reaction textures is preserved (Devine et al. 1998a; Murphy et al. 2000). Many crystals have experienced slight to extensive oxidation and are variably replaced at the rims and along cleavage planes by a very fine aggregate of minute (∼1–10 μm) Fe-oxides and pyroxene (opacitisation, cf Garcia and Jacobson 1979; Murphy et al. 2000). These hornblendes appear dark brown and are often opaque (Fig. 1b).

Fig. 1

Photomicrographs of hornblende reaction rims. a Large pristine hornblende phenocryst. b Hornblende phenocryst with fine-grained reaction rim. The crystal is brown due to oxidation and parts of the crystal display opacitisation (cf Garcia and Jacobson 1979). c Large brown hornblende phenocryst displaying extensive areas of opacitisation and decompression-induced reaction rim. d Hornblende phenocryst with extensive coarse-grained reaction rim (after Murphy et al. 2000, with permission). The original outline of the crystal is clearly visible. a–d Photomicrographs from a collection by M. Murphy, of the 1995–1999 eruption products of Soufrière Hills Volcano, Montserrat. All are in transmitted light. Field of view is ∼3 mm in all cases. e–g Backscattered electron (BSE) images of the reacted hornblende phenocryst studies from sample MVO1090B. The reaction rim is only preserved on two sides of the phenocryst. The boxes in (e) and (f) show the reaction rim areas enlarged in following frames. Minerals, glass and voids are represented by different greyscale levels, reflecting compositional differences. The hornblende reaction rim contains pyroxene (px), plagioclase (pl), titanomagnetite (Ti-mag) and glass

Reaction rims are common, surrounding and partially replacing hornblende phenocrysts with fine-grained (∼5–30 μm) intergrowths of clinopyroxene, orthopyroxene, plagioclase and Fe-oxides (Fig. 1c). Rims occur only on hornblende in contact with melt and rim growth proceeds inward from the crystal margins. In general reaction rims preserve the euhedral morphology of the precursor hornblende and are of uniform width around a given crystal. Multiple populations of hornblende phenocrysts with fine-grained reaction rims have been identified at SHV (Devine et al. 1998b). The main population (about 90%) display thin (∼100 μm) reaction rims, while the remainder display thicker (∼200–400 μm wide) reaction rims. These reaction rims form as a result of magma decompression either during magma ascent or during pre-eruptive shallow storage (Garcia and Jacobson 1979; Rutherford and Hill 1993; Rutherford and Devine 2003). Rim width is dependent on time and the magnitude of overstep in temperature and pressure, as well as decompression rate (Rutherford and Hill 1993). Thin-rimmed hornblendes are thought to have ascended rapidly from a deep source within the hornblende stability field, whereas thicker-rimmed hornblendes ascended more slowly. In slowly erupted samples a spectrum of rim widths is common. Qualitatively grain size in these rims increases with rim width. Similar observations were made on MSH samples by Rutherford and Hill (1993) who attribute rim width variability to conduit flow processes; thicker rimmed hornblendes are attributed to slower moving magma near the conduit walls, or possibly even erosion of magma emplaced in the earlier phase of eruption when mass eruption rates were much higher. They propose that the different batches of magma are then intimately mixed during ascent.

Resorbed hornblende phenocrysts and coarse-grained (∼30–200 μm) hornblende reaction rims, mineralogically similar to those described above, are also present in SHV andesites (Fig. 1d). Similar textures described from subvolcanic andesites and basalts from the Sarroch District, SW Sardinia (Conte 1993) have been attributed to magma heating and high-temperature hornblende destabilisation. Any or all of the reaction textures described above may be preserved in close proximity within a single sample. Sometimes individual hornblende phenocrysts display multiple reaction textures. Commonly reaction rims are accompanied by opacitisation (Fig. 1c), presumably due to late-stage oxidation of hornblende cores surviving previous heating or decompression events (Murphy et al. 2000).

Analytical techniques

Sample selection

Polished thin sections from samples of recent Soufrière Hills andesite were selected from the Montserrat Volcano Observatory sub-collection held at the University of Bristol. The chosen suite (Table 1) contains samples derived from a range of eruptive conditions, including explosion-derived pumice and ballistic clasts (October 1997 to March 1999), and lava blocks from dome building episodes during periods of lower extrusion rate (January 1997 to February 2001). In samples containing fine-grained hornblende reaction rims, compositions of the reactant amphiboles and compositions and proportions of the product phases were investigated. Coexisting glasses were also characterised in these samples: matrix glass, melt inclusions in amphibole phenocrysts and glass within reaction rims. Unreacted amphibole phenocrysts were analysed in other contemporaneous samples to determine whether any compositional heterogeneity might exist within the precursor amphiboles. One representative amphibole phenocryst from each sample was chosen for study.

Table 1 List of samples from Soufrière Hills Volcano, Montserrat, used in this study

Electron probe microanalysis

Major element compositions of the minerals were obtained by electron probe microanalysis (EPMA) at the University of Bristol. Most amphibole phenocrysts were analysed on a Cameca Camebax electron microprobe with four wavelength dispersive spectrometers, operated through SAMx automation software. Online data reduction used the ϕ–ρ– z model of Pouchou and Pichoir (1985). Amphibole compositions from sample MVO1217 and mineral phases within all of the amphibole reaction textures were analysed on a JEOL JXA-8600 Superprobe equipped with four spectrometers, LINK analytical X-ray analysis system, LEMAS automation and ZAF online data reduction. All mineral phases were analysed with an accelerating potential of 15 kV, a beam current of 15 nA and minimum beam diameter (∼1 μm). Minimum counting times were typically ≥20 s on peak and 8 s on background.

Major element compositions of glasses were also obtained by EPMA at the University of Bristol, using the JEOL JXA-8600 electron probe. EMPA of hydrous silicate glasses is complicated by alkali migration during prolonged heating of unstable samples under an electron beam. This effect can result in significant underestimation of Na and overestimation of Si contents (Morgan and London 1996). Harford et al. (2003) have determined optimum conditions for EPMA of Soufrière Hills matrix glasses. Using an accelerating voltage of 15 kV and a 2 nA beam rastered over as large an area as possible (typically ∼5×4 μm), with total Na count time limited to 40 s and early analysis of Na, Si and Al, Na-loss can be limited to less than ∼5% relative. The same analytical conditions have been adopted in this study. Interstitial glass in reaction rims occupies small areas. Only areas large enough to analyse using the above conditions are reported and many analyses of smaller areas had to be discarded.

Modal analysis

The distribution and proportions of phases throughout hornblende reaction rims were determined by an element mapping approach. This was performed using a Cambridge S250 Mk3 Stereoscan scanning electron microscope (SEM) fitted with a PCXA energy-dispersive X-ray analyser. Relative X-ray intensities were mapped for several diagnostic elements (Si, Ti, Al, Fe, Mg, Ca and K) known to be concentrated preferentially, either individually or in combination, in one or more phases of interest (i.e. K in volcanic glass, Fe and Ti in magnetite, Al in plagioclase, Mg and Fe in pyroxene together with Ca to distinguish clinopyroxene from orthopyroxene). All elements were scanned in energy-dispersive mode with an operating voltage of 20 kV and dwell time of 100 ms. This technique was only performed after EPMA so that quantitative compositional data would not suffer from sodium mobilisation resulting from mapping. Resulting BSE digital images or maps for each element were then scaled to the best resolution attainable (∼1 μm per pixel) and analysed using image processing software (NIH image) to quantify areas of high BSE count density. Areal fractions were converted to mass fractions using the following phase densities (g cm⁻³): plagioclase = 2.7, clinopyroxene = 3.3, orthopyroxene = 3.5 and Fe-oxides = 5.2.

Reaction rim width measurement

Reaction rim widths were measured and scaled from magnified BSE images of the rims, obtained by SEM. Precise measurements of rim widths are possible by this method because the rims preserve the euhedral morphology of the precursor amphibole and display relatively uniform thickness around a given crystal. Misleading depth effects encountered in transmitted light techniques are avoided as back-scattered electrons sample only the phases exposed near the thin section surface (Rutherford and Hill 1993). Reaction rim widths are displayed most accurately normal to the crystal faces that they replace. These are best displayed and most easily identified in grains cut perpendicular to the c-axis (two good cleavages) and on faces (011), and in grains cut normal to the a axis (one good cleavage and straight extinction). Where hornblendes retain a substantial core crystal orientation can be determined optically.

Mineral chemistry

Amphibole

Amphibole phenocrysts are abundant in the recent Soufrière Hills andesite. These have been described in detail by Murphy et al. (2000) and Rutherford and Devine (2003). Most are large (0.5–1.5 cm) and green to brown in colour, with euhedral to anhedral form. Inclusions of plagioclase and titanomagnetite are common, as well as occasional melt inclusions. All are calcic amphiboles; magnesiohornblendes in the terminology of Leake et al. (1997). Rutherford and Devine (2003) have observed compositional zoning in some hornblende phenocrysts, with zones of higher Mg and Si grading gradually outwards to darker-coloured zones depleted in these elements and enriched in Fe and Al. The total compositional variation is small; ≤2 wt% change each in SiO₂, MgO, FeO and Al₂O₃. Multiple zones are common and up to seven have been identified in large phenocrysts.

The composition required for material balance calculations is that of reacted hornblende. This must be reconstructed from remnant hornblende material. Suitable proxies for the reacted amphibole are either unreacted phenocryst rims, or the outermost region of remnant cores surviving in reacted phenocrysts. The choice between these depends on whether pristine hornblende phenocrysts have the same chemistry as those with reaction rims, and whether compositional zoning in the reactant hornblende is likely. Compositions of unreacted and reacted SHV hornblendes are given in Table 2. Both core and rim compositions are reported, though for reacted phenocrysts the rim composition is for the outermost remaining hornblende, not the original rim material. No compositional zonation is optically discernable in these hornblendes. This could either be because none is present, or because zoning is less visible in near-basal sections, as used in this study, than in crystals sectioned parallel to the crystallographic c-axis (Rutherford and Devine 2003).

Table 2 Representative average amphibole compositions

Comparing core compositions from Table 2, and allowing for uncertainties of ±2σ in the mean composition for each sample, no significant difference between unreacted and reacted phenocryst core chemistry is evident. Compositions correspond most closely to those of the MgO and SiO₂ enriched zones reported by Rutherford and Devine (2003); approximately 48–49 wt% SiO₂, 6–7 wt% Al₂O₃, 14–15 wt% FeO^T and 14.5–15 wt% MgO. Only one phenocryst (MVO1217) displays the MgO and SiO₂-depleted composition, as observed by Rutherford and Devine (2003) towards the outer margins of zones (around 46 wt% SiO₂, 8–9 wt% Al₂O₃, 15–16 wt% FeO^T and 13.5–14 wt% MgO). Compositions of both unreacted and reacted rims are consistent with core compositions within error.

Reaction rim assemblage

Fine-grained (∼5–70 μm) reaction rims are composed of intergrowths of clinopyroxene, orthopyroxene, plagioclase and Fe–Ti oxide(s) (Fig. 1). Glass is present within the reaction rims as interstitial patches between the mineral grains. This glass is compositionally similar to the surrounding matrix glass, perhaps indicating that glass in reaction rims originated as matrix melt that infiltrated permeable hornblende reaction rims, rather than as a direct product of the hornblende breakdown reaction. Alternatively interstitial reaction rim melt could have equilibrated with the matrix melt by diffusion during rim formation.

Pyroxene

Clinopyroxene and orthopyroxene are the most abundant phases in the hornblende reaction rims, comprising approximately 53 and 23% of the mode, respectively. Both cpx and opx crystals display a variety of morphologies, ranging from small (∼10 μm) rounded anhedral grains to larger (∼70 μm) elongate dendritic, skeletal or tabular laths. No preferred orientation of the crystals is observed. Pyroxene compositions within the reaction rims are variable; clinopyroxene has En_40-50 and Wo_27-41 and orthopyroxene has En_61-67 and Wo_3-6 (Table 3). These compositions fall within the range of groundmass pyroxene, and clinopyroxene microphenocryst compositions in recent Soufrière Hills andesite (Murphy et al. 2000), although clinopyroxene in reaction rims has slightly lower average Wo and En (Fig. 2). No systematic variation is observed between pyroxene composition and depth into the reaction rim (Fig. 3a).

Table 3 Representative average compositions of pyroxenes in hornblende reaction rims

Fig. 2

Pyroxene compositions in hornblende reaction rims, plotted as mol% of the simple quadrilateral components. Compositions of the main SHV groundmass pyroxene populations (including clinopyroxene microphenocrysts) determined by Murphy et al. (2000) are also shown (outlined fields) for comparison

Fig. 3

Plot illustrating the variation in product mineral compositions across the width of reaction rims. Distances on the x-axis increase towards the inner reaction rim boundary with hornblende. Each plot symbol refers to a single reaction rim. Each data point represents the composition and position of a single crystal within the reaction rim. Error bars indicate the maximum dimension of each crystal normal to the reaction rim margins. a Pyroxene compositions are expressed in mol% Wo. b Plagioclase compositions are expressed in mol% An. A weak decrease in An with depth into the rim is indicated within MVO177, for which the line of best fit (not shown) is linear with a correlation coefficient (R ²) of 0.6. c Titanomagnetite compositions expressed in mol% Usp

Plagioclase

Plagioclase accounts for approximately 18% of the reaction rim mode. Crystals have subhedral and tabular morphologies, typically being smaller (∼5–50 μm) than the surrounding pyroxenes. Plagioclase compositions vary over the range An_53-64 (Table 4, Fig. 4), consistent with compositions of sodic rims of calcic microphenocrysts (Murphy et al. 2000) and groundmass microlites in both pumices and dome rocks (Murphy et al. 2000; Couch et al. 2003). No systematic variation is observed between plagioclase composition and depth within the reaction rims (Fig. 3b), although a weak increase in Ab deeper in the rim is indicated within MVO177.

Table 4 Representative average compositions of plagioclase in hornblende reaction rims

Fig. 4

Plagioclase compositions in hornblende reaction rims, plot as mol% An

Fe-oxides

In three of the reaction rims studied (MVO177, MVO1006w and MVO1090B) titanomagnetite is the only oxide phase present, accounting for about 5% of the mode. Crystals are typically small (∼5–30 μm) anhedral grains. Compositions are variable, typically spanning the range Usp_10-26; the overall range extending to Usp₅₄. Almost as much chemical variation is present within a single reaction rim as between the three rims studied, and no spatial relationship is observed (Fig. 3c). Magnetite and ilmenite rather than titanomagnetite were found in the remaining reaction rim (sample MVO1217), each comprising approximately 2% of the mode. For both minerals crystals are small (∼5–30 μm) and anhedral. Compositions are shown in Table 5 and Fig. 5.

Table 5 Representative average compositions of Fe–Ti oxides in amphibole reaction rims

Fig. 5

Fe–Ti oxide compositions in hornblende reaction rims, plot in the system FeO–Fe₂O₃−TiO₂ as mol%

Material balance calculations and estimates of material transport

Mass balance calculations were performed between the reactant hornblende and the product reaction rim phases, using the linear least squares petrologic mixing program ‘Petmix‘ of Wright and Doherty (1970). The program was used according to the method for calculation of a chemical mode. Given the bulk chemical composition of a rock, in terms of element oxides, together with the chemical compositions of its constituent minerals, the program will calculate the modal mineral proportions (in weight percent) that best fit the rock composition. For the case of the hornblende breakdown reaction, the reactant hornblende is taken as the bulk composition.

Soufrière Hills Volcano

Two slightly different product assemblages have been observed in SHV hornblende reaction rims so we have considered a separate reaction for each:

$${\text{hbl}} \to {\text{cpx}} + {\text{opx}} + {\text{plag}} + {\text{Ti - mag}}$$

(2)

$${\text{hbl}} \to {\text{cpx}} + {\text{opx}} + {\text{plag}} + {\text{ilm}} + {\text{mag}}$$

(3)

Melt has not been included as a phase in these reactions because neither hornblende alone, nor any single combination of hornblende and melt can exactly reproduce the composition of the reaction rim products. By expressing the reaction as a simple conversion of hornblende to pyroxenes + plagioclase + Fe-oxide(s), the compositional differences between hornblende and the best-fit combination of product phases are clear. It is assumed that these differences are balanced by exchanges with the host melt phase.

The modelling treats minerals as homogeneous phases, so some generalisations about mineral compositions have been made. For the reactant hornblende we have used outermost remnant core compositions as a proxy. This composition (Table 6) is the combined average of all ‘rim’ data from the four reacted phenocrysts studied. The hornblende volatile content has not been analysed here so compositions have been recast to 100 wt% anhydrous. In the absence of any systematic variation in orthopyroxene, clinopyroxene, plagioclase or titanomagnetite composition between or within reaction rims, compositions of each of these phases have been averaged over the three samples displaying reaction 2 and normalised to 100 wt% (Table 6). Compositions of mineral phases in the reaction rim from MVO1217 have been treated separately, and input mineral compositions for reaction 3 products are normalised averages over the individual reaction rim (Table 6).

Table 6 Input parameters and results of Soufrière Hills Volcano hornblende breakdown material balance calculations using the PETMIX petrologic mixing program

Results of the material balance calculations are given in the output section of Table 6. The best-fit solutions indicate the following mass balances for the two reactions:

$$100.0\,{\text{hbl}} \to 49.8\,{\text{cpx}} + 27.6\,{\text{opx}} + 18.2\,{\text{plag}} + 4.5\,{\text{Ti - mag}}$$

(2a)

$$100.0\,{\text{hbl}} \to 53.9\,{\text{cpx}} + 24.4\,{\text{opx}} + 18.1\,{\text{plag}} + 1.9\,{\text{ilm}} + 1.7\,{\text{mag}}$$

(3a)

These results agree well with modal analysis of the reaction rim assemblage converted to mass fractions. However, the crystalline phases in the reaction rims do not exactly reproduce the reactant hornblende composition when combined in the proportions indicated by reactions 2a and 3a (bulk rim compositions, Table 6). The amounts by which the reactant and bulk products differ (residuals) are relatively small (≤1.00 wt% in magnitude in all cases), and vary in magnitude and sense for each element, and between the two reactions. Positive residuals indicate that reactant hornblende is enriched in a given element relative to the bulk reaction rim, and that the residual material is released by the reaction. Negative residuals indicate that the reaction cannot have occurred without addition of material from the chemical system beyond the hornblende phenocryst. However, residuals given by Petmix need to be compared with uncertainties in input compositions to establish whether the apparent differences can be regarded as significant or within uncertainties.

We calculated uncertainties of two standard errors for the reactant and combined product compositions (uncertainties, Table 6) to minimise residuals from Petmix. For most elements, residuals remain beyond these uncertainties (all elements in reaction 2a; SiO₂, TiO₂, Al₂O₃, FeO^T, Na₂O and K₂O in reaction 3a). These results are consistent with reaction between hornblende phenocrysts and the surrounding melt in an open chemical system, with exchange of most elements. The reaction is inferred to consume SiO₂ and sometimes FeO^T from the melt, and release Na₂O, K₂O and perhaps TiO₂ and MgO into the melt. Volatiles (H₂O and Cl) must also be released by the reaction. If any compositional zoning were present in the reactant hornblende, the likely trend would have been towards lower MgO and SiO₂ and higher FeO^T and Al₂O₃ based on the observations of Rutherford and Devine (2003). This would help to reduce shortfalls in FeO^T but increase shortfalls in SiO₂. Finally we note that variations in magnetite composition and in oxide phase proportions (between reactions 2a and 3a) may reflect variations in fO₂ during the reactions. Such variation in fO₂ might relate to detailed differences in degassing histories, since water exsolution can cause strong oxidation (Candela 1986).

MSH 1980–1986 eruptions

We have made material balance calculations for the compositional data of Rutherford and Hill (1993), using the same approach as for SHV above. We tested each of the five amphibole compositions reported by Rutherford and Hill (1993) as the reactant, against the MSH product rim phase compositions of Rutherford and Hill (1993) in material balance calculations using Petmix as before. The smallest residuals were achieved using the SH-084 amphibole rim composition as the reactant (Table 7):

$$100.0\,{\text{hbl}} \to 43.0\,{\text{cpx}} + 24.5\,{\text{opx}} + 23.7\,{\text{plag}} + 5.2\,{\text{mag}} + 3.5\,{\text{ilm}}$$

(4)

Table 7 Input parameters and results of new Mount St. Helens hornblende breakdown material balance calculations using the PETMIX petrologic mixing program

This reaction is markedly different to that proposed by Rutherford and Hill (1993):

$$73\,{\text{amph}} + 27\,{\text{melt}} \to 29\,{\text{opx}} + 24\,{\text{cpx}} + 43\,{\text{plag}} + 3\,{\text{ilm}}$$

(5)

The new calculations here are based on MSH hornblende rim compositions rather than a composition averaged over an entire crystal as in the calculations of Rutherford and Hill (1993). Hornblende phenocrysts in MSH dacites show considerable compositional variation and are commonly zoned: rims are commonly enriched in FeO and SiO₂, and depleted in Al₂O₃ and MgO relative to the cores (Rutherford and Hill 1993). Melt has not been included either as a product or reactant phase in this study. It is assumed instead that exchanges with the melt phase account for any residuals. Thus not all components in the melt have to contribute to the reaction and that the composition of the melt does not need to remain fixed. Magnetite is included in the product assemblage.

Large residuals (≤3.30 wt% in magnitude) are associated with reaction 4, including shortfalls in SiO₂, Fe₂O₃, FeO and TiO₂, and excesses in Al₂O₃, MgO, CaO and Na₂O. Fe₂O₃ has been considered in mass balance calculations for MSH because input reactant and product compositions from Rutherford and Hill (1993) were already adjusted for ferric iron content. Rutherford and Hill (1993) do not report uncertainties for product phase compositions so we were unable to compare these residuals with standard errors. Available uncertainties of two standard errors for the reactant hornblende composition alone are insufficient to account completely for any of the residuals (uncertainties, Table 7). Thus these results are consistent with reaction between hornblende phenocrysts and the surrounding melt in an open chemical system. Based on these results the reaction consumes SiO₂, TiO₂, Fe₂O₃ and FeO from the melt and releases Al₂O₃, MgO, CaO and Na₂O from the hornblende into the melt.

Melt chemistry

Melt inclusions in hornblende

Scarce melt inclusions in SHV hornblende phenocrysts are small (typically ∼<10 μm diameter) and irregularly shaped. Most samples proved unsuitable for reliable EPMA and compositional data reported here are limited. Melt inclusion compositions range from 69.4 to 71.6 wt% SiO₂ (normalised to 100 wt% anhydrous) in a pumice sample and 71.0–74.8 wt% SiO₂ in a dome sample (Table 8). These compositions lie on a trend similar to other natural SHV glasses (Fig. 6) and are the least evolved melts found so far in SHV andesite. Hornblende melt inclusion compositions are similar to estimates of the groundmass compositions (matrix glass plus crystals smaller than 100 μm; Murphy et al. 2000). This is consistent with hornblende phenocryst growth fairly early in the crystallisation sequence, before significant groundmass crystallisation occurred. Comparison with experimental glass compositions of Rutherford and Devine (2003) indicates temperatures of 900–880°C at water pressures close to 130 MPa for formation of these melt inclusions (Fig. 6). These temperatures are consistent with the maximum estimates of hornblende crystallisation temperatures, based on plagioclase–amphibole thermometry between hornblende phenocrysts and plagioclase inclusions (Couch et al. 2003). Volatile contents estimated using the ‘difference method’ (100 wt% minus analytical totals equals volatile content; Devine et al. 1995) range from 8.3 to 5.0 wt% in the pumice sample, and 4.3–3.1 wt% in the dome sample. The lower values in melt inclusions in the dome sample may reflect leakage through cracks or along cleavage planes during prolonged magma decompression and residence time in the dome before collapse.

Table 8 Melt inclusion compositions in hornblende

Fig. 6

Harker diagram for Soufrière Hills Volcano glass compositions. Analyses are given in Tables 7, 8 and 9. Temperatures are from Rutherford and Devine (2003)

Matrix glasses

Matrix glass is rhyolitic with wide compositional variations (Fig. 6). SiO₂ can vary between 72.0 and 81.1 wt% (normalised to 100 wt% anhydrous; Table 9). Two distinctive groups are present based on sample type; pumice glasses are notably low in K₂O, high in Na₂O and high in CaO, whereas dome glasses are relatively high in K₂O, low in Na₂O and low in CaO. Similar groupings have been recognised in previous studies (Harford et al. 2003; Couch et al. 2003). K₂O data in particular indicate a distinct difference between the groups (Fig. 6f).

Table 9 Interstitial melt compositions in hornblende reaction rims

Trends of enrichment in K₂O and depletion in Na₂O with increasing SiO₂ in all matrix glasses are consistent with groundmass crystallisation dominated by plagioclase; the main groundmass mineral. K₂O is nearly incompatible in the SHV groundmass assemblage. Thus higher K₂O in dome glasses than pumices has been explained in terms of more extensive groundmass crystallisation (Harford et al. 2003). This interpretation is consistent with groundmass textures: pumice samples contain 20–35% modal glass, whereas dome samples can have as little as 5% modal glass (Murphy et al. 2000; Harford et al. 2003; Couch et al. 2003). While groundmass crystallisation is certainly important, a difficulty with this interpretation arises because matrix glass in dome lavas and pumices with similar SiO₂ contents (i.e. similar amounts of groundmass crystallisation), have very different K₂O contents (Fig. 6). Hornblende breakdown releases K₂O into the host melt, but not enough to account for the observed differences in K₂O content. Complete hornblende breakdown and partitioning of all K₂O into a small fraction of residual melt (∼10 wt%) could increase dome melt K₂O contents by no more than 0.15 wt% relative to pumice glasses. Another source of glass composition variability is magma mingling between residual melts from different magma batches (Rutherford and Devine 2003).

Normative compositions for the Soufrière Hills glass compositions have been projected onto the An-corrected ternary system Qz′–Ab′–Or′ using the calculation procedure of Blundy and Cashman (2001). The pumice and dome matrix glasses define contrasting trends (Figs. 7, 8). Pumice glasses (Fig. 7a, b) trend steeply away from Ab′ towards the 0.1 MPa quartz–feldspar cotectic at about Qz′ = 50%, Ab′ = 35% and Or′ = 15%. Dome glasses (Fig. 8a, b) follow a shallow trend away from the Ab′–Qz′ join at about Ab′ = 85% and Qz′ = 15% roughly delineating the eutectic position over a range of decreasing pressures. These respective matrix glass trends extend back to the compositions of melt inclusions in hornblendes and the average groundmass composition estimated by Murphy et al. (2000). The pumice trend is as expected for rapid decompression involving plagioclase crystallisation without reaching the silica phase field (Blundy and Cashman 2001). The high Qz′ content of some pumice glasses indicates partial equilibration at low pressures before quenching by explosive eruption. The dome glasses plot to the Or′-rich side of their pumice counterparts, reflecting higher K₂O contents in the melts. This trend plausibly reflects either physical or chemical mixing between melts ascending at different rates from the magma chamber and experiencing different amounts of groundmass crystallisation en route. We have simulated mixing between melts from different levels by combining compositions of evolved low pressure matrix melts and least evolved melts from inclusions in hornblende at 10% increments. Resulting intermediate melt compositions plot along a similar shallow trend to natural melts in dome samples (Fig. 8a), consistent with the premise that dome glasses are mixtures of melts ascending at different rates.

Fig. 7

Qz′–Ab′–Or′ ternary (after Blundy and Cashman (2001) adapted from Tuttle and Bowen (1958)), showing Soufrière Hills Volcano glass compositions in pumice samples. a Sample MVO1006w and b sample MVO1090B. Cotectic lines and compositions of H₂O-saturated minima and eutectics are after Blundy and Cashman (2001) for a range of water pressures. Neither sample contains a silica phase; consequently cotectics corresponding to the positions of these glasses give maximum rather than actual $\rm{P}_{\rm{H_2}O}$ conditions of melt equilibration

Fig. 8

Qz′–Ab′–Or′ ternary showing Soufrière Hills Volcano glass compositions in lava dome samples. a Sample MVO177. The mixing line shows the range of melt compositions that could be achieved by mixing the least evolved melt inclusion material with the most evolved matrix glass. Plot symbols mark increments of 10% in melt proportions. b Sample MVO1217. Silica microlites are present in both the matrix glass and interstitial reaction rim melt (thick rim). This allows quantitative estimates of the pressures at which these melts were last equilibrated (100 to <25 MPa for the matrix and <100 MPa for reaction rim melt). Scatter in Qz′ values (vertical arrows) in interstitial reaction rim melts might result from SiO₂ consumption from the melt during hornblende dehydration

Hornblende reaction rim glasses

Interstitial glass within hornblende reaction rims is rhyolitic with SiO₂ varying between 69.9 and 80.0 wt% (normalised to 100 wt% anhydrous; Table 10). Compositions are heterogeneous and more variable than matrix glasses (Fig. 6). Notably interstitial reaction rim glasses show the same groupings as the corresponding matrix glasses in terms of K₂O, Na₂O and CaO (Fig. 6) and lie along the corresponding trends in the Qz′–Ab′–Or′ projections (Figs. 7, 8). Other components are more variable with significant heterogeneity within single reaction rims. Some interstitial glass is indistinguishable from surrounding matrix glass whereas other glasses are enriched or depleted in some components. Enrichments in FeO^T and TiO₂ are prominent in many interstitial glasses and a small number of reaction rim glasses are rich in MgO and poor in SiO₂.

Table 10 Matrix glass compositions in the recent Soufrière Hills Volcano andesite

We have already shown that hornblende breakdown requires an open system with exchanges of components with the surrounding melt. Some of the interstitial glasses within reaction rims are similar to the host matrix glass and can be interpreted as surrounding melt infiltrating the reaction rims. In there cases the interstitial reaction rim glass faithfully mirrors the matrix glass (Figs. 6, 7, 8). However, other glasses are markedly different and are mostly consistent with the chemical exchanges suggested by the mass balance calculations. In particular the calculations infer releases of TiO₂ and MgO into the melt, which can explain the interstitial glasses enriched in these elements. Exchanges in SiO₂ and Na₂O are more difficult to interpret since these elements vary considerably in matrix glasses, due to variable groundmass crystallisation. FeO^T is the only discrepancy, being consumed in reaction 2a, but enriched in many interstitial glasses relative to the matrix glasses.

Conclusions

Hornblende phenocrysts in recent Soufrière Hills Volcano andesites display reaction rims of microcrystalline plagioclase, pyroxene, Fe-oxides and interstitial glass, formed by hornblende dehydration during magma ascent and decompression. Mass balance calculations give the following reactions, with mineral proportions in agreement with modal abundances expressed as mass fractions:

$$ \begin{aligned} 100.0\,{\text{hbl}} & \to 49.8\,{\text{cpx}} + 27.6\,{\text{opx}} + 18.2\,{\text{plag}} + 4.5\,{\text{Ti - mag}} \\ 100.0\,{\text{hbl}} & \to 53.9\,{\text{cpx}} + 24.4\,{\text{opx}} + 18.1\,{\text{plag}} + 1.9\,{\text{ilm}} + 1.7\,{\text{mag}} \\ \end{aligned} $$

Re-evaluation of the compositional data and material balance estimates for hornblende breakdown in post May 18, 1980, MSH andesites (Rutherford and Hill 1993) gives the reaction:

$$100.0\,{\text{hbl}} \to 43.0\,{\text{cpx}} + 24.5\,{\text{opx}} + 23.7\,{\text{plag}} + 5.2\,{\text{mag}} + 3.5\,{\text{ilm}}$$

Both MSH and Soufrière Hills Volcano hornblende dehydration reactions occurred in open chemical systems with exchange of most components between the hornblende and the melt. The reactions studied reveal deficiencies in SiO₂ and Fe-oxides, and excesses of Na₂O and K₂O, the residuals presumably being accommodated by the surrounding melt. Volatiles (H₂O and Cl) must also be released by the reaction. These results imply that transport of mobile components in the melt must play a part in rim growth. It follows that transport processes may have the potential to control the rate of reaction rim growth.

Matrix glasses fall into two compositional groups. Glasses in pumice are relatively rich in CaO and poor in K₂O and Na₂O compared to glasses in dome samples. The former glasses formed by moderate amounts of groundmass crystallisation of plagioclase associated with rapid magma ascent in explosive eruptions. The latter glasses reflect mixing between melts that experienced extensive groundmass crystallisation and hornblende breakdown during slow magma ascent and extended residence time in the dome, and less evolved melts that ascended more rapidly from the magma chamber. Interstitial glass compositions in hornblende reaction rims reflect the compositions of the surrounding matrix glasses, but show variable compositional differences mostly consistent with the proposed open-system hornblende breakdown reactions.