The influence of water in silicate melt on aluminium excess in plagioclase as a potential hygrometer

Measuring water contents of magmas is fundamental to resolving a number of geological questions, such as the mechanisms of silicic magma evolution, the triggering of volcanic eruptions, and the formation of porphyry copper deposits. This study focuses on the correlation between apparent deviations from stoichiometry of plagioclase crystals and high water concentration in the magmatic melt from which they grew. We considered this relationship as a potential geo-hygrometer (water activity indicator). To test and potentially calibrate this new technique, a range of natural and experimental plagioclase crystals were analysed, with particular care taken to identify and avoid analytical bias and artefacts. In contrast to recently published material, we found no systematic aluminium excess in plagioclase, irrespective of the water concentration of the silicate melt it crystallised from. This suggests that aluminium excess in plagioclase cannot serve as a geo-hygrometer. The high likelihood of misinterpreting analytical artefacts (due to alkali migration and imprecise standardisation) as small deviations from stoichiometry, also requires its application as a mineral exploration tool to be treated with caution.

appears to occur dominantly in plagioclase from mineralised calc-alkaline plutons and was hypothesised to be related to a high partial pressure of water (P H2O ) in the melt 29 . Geochemical correlations support coupled substitution of AlAl 3 SiO 8 and []Si 4 O 8 and occupation of the generated vacancy by H 2 O as incorporation mechanism for water and Al into plagioclase 29,30 . However, studies applying Fourier-transform infrared spectroscopy (FTIR) showed that igneous plagioclase typically only contains OH, not H 2 O, although lattice positions of H incorporation could not always be identified ( 22 and references therein). Moreover, no correlation between H content and feldspar composition or geodynamic setting was found (summary in 31 , their Table 1), thereby supporting fast diffusion of H.
A hygrometer based on Al* would potentially bear the advantages of: (a) wide applicability due to the ubiquitous presence of plagioclase in most igneous rocks, (b) ease of measurement with standard-based wavelength dispersive X-ray spectrometry (WDS) attached to an electron-probe micro-analyser (EPMA), (c) good preservation potential due to slow diffusion of Al, Ca, and Na compared to H (e.g. [32][33][34][35], and (d) potential to resolve changing concentrations recorded in mineral zoning. Therefore, the scope of this study is to test, if the incorporation of Al* in plagioclase can be developed into a geo-hygrometer, using both natural samples from a range of environments and experimentally-grown plagioclase crystals precipitated from melts with contrasting water contents. In the course of this evaluation, care was taken to minimise potential analytical artefacts that could lead to erroneous interpretations.

Results
Sample selection. Plagioclase crystals from a wide variety of samples (see electronic supplementary material) were selected for analysis with EPMA with the aim to: (a) reproduce the dichotomy in Al* between barren (samples without any concrete evidence for an associated porphyry copper deposit at depth) and mineralised lithologies as presented in 29 , (b) study the effect of P H2O in the magma on Al* in plagioclase, and (c) explain other factors -geological-geochemical and analytical -influencing the measured Al* in plagioclase. Unaltered, natural plagioclase crystals can be found in barren magmatic rocks from both plate margin and intraplate environments, but also in some rocks related to porphyry copper deposits (electronic supplementary material). If the hypothesis of discernible Al* in mineralised and barren rocks as postulated in 29 (their Fig. 1) is true, the mineralised sample group should exhibit more pronounced Al* than the barren one. The difference in Al* should be particularly pronounced between mineralised deposits and dry magmas from hotspots if P H2O indeed controls the Al* in plagioclase. As an additional test, plagioclase crystals were produced experimentally from wet and dry melts in this study to further test the correlation between Al* and dissolved water concentration in the melt under controlled conditions. Mineral chemistry of natural plagioclase from different geological settings. From the analysis of 299 data points on plagioclase crystals from 14 natural samples (Fig. 1), it can be shown that: (a) Plagioclase compositions from different samples largely overlap in Al* (at similar anorthite content) and are narrowly distributed on both sides of the stoichiometry line, irrespective of the presumed water content in the magma. The only exception is plagioclase from a basaltic sample from Iceland (HEID), which plots mostly below the stoichiometry line. In agreement with the findings of Williamson et al. 29 , this may be related to its relatively high Fe content (>0.02 Fe (apfu)). Furthermore, no systematic variations in Al* in variably mineralised porphyry sample sets from Batu Hijau (samples BH-181a and BH-184) and Bajo de la Alumbrera (samples BA-SR-55, BA-LP3, BA-YB-38 EP3 and BA-YB-40 P2) could be observed. In addition, we note that partial albitisation of sample BH-181a is reflected in bimodal anorthite contents of plagioclase. Despite the fact that altered zones (quartz-magnetite and potassic) of With respect to analytical conditions, data treatment, and potential artificially-induced errors, it has been observed that: (a) inclusion of cations other than Ca, Na, and K at the M site, such as Ba and Sr, does not change the positions of most data sets visibly. (b) Analysis with normal beam current (ca. 20 nA) is necessary to yield sufficient intensities and uncertainties low enough to observe small differences in Al*. Larger uncertainties, on the other hand, result from reduced beam current for focused analyses and are sufficient only to resolve large differences in Al*. (c) A correction based on repeated analyses of the albite and anorthite standards as unknowns was necessary and resulted in all data to plot close to the stoichiometry line.

Synthetic plagioclase crystallised from melts of contrasting water concentrations. The main
differences between the experimental runs with wet and dry melts are the following: In the high-temperature, water-poor experiments, plagioclase formed crystals of high aspect ratio and was the only mineral that crystallised. In experiment C (1240 °C), single crystals were ca. 150 μm long with hollow interiors (occupied by glass) and occurred only in the upper part of the capsule. In experiment E (1190 °C), plagioclase crystals rarely exceeded ca. 30 μm length, were homogeneously distributed throughout the capsule, and occurred in random orientations (Fig. 2a). The resulting total crystallinity was ca. 50 area%.
In the low-temperature, water-rich experiment (experiment D, 925 °C), plagioclase co-crystallised with acicular to short-prismatic clinopyroxene (Fig. 2b), typically in the form of clusters. Plagioclase was more stubby in shape than in hot, dry experiments, and commonly up to 25 μm long, although a few crystals reached up to 60 μm length. Bubbles of variable size were distributed throughout the experimental capsule and cover ca. 20 area%. The overall crystallinity of this experiment was less than 50 area%.
Plagioclase of all experiments was of similar, intermediate anorthite content (Fig. 3; average compositions in Table 1). Most plagioclase crystals were fairly homogeneous according to quantitative analyses and lack of SEM-BSE contrast. In the low-temperature, water-rich experiment (experiment D), however, plagioclase crystals in the core of clusters were of higher anorthite content. Iron oxide and K 2 O concentrations in plagioclase from all experiments were commonly <0.1 wt% and <0.05 wt%, respectively, i.e. below detection limit. Clinopyroxene and glass were homogeneous throughout the experimental capsule.
Our results from experimentally crystallised plagioclase support the findings based on natural plagioclase: Through analysis of both natural and synthetic plagioclase from dry versus wet melts, large differences in Al* in plagioclase as presented in 29 (their Fig. 1a) could not be reproduced.

Sodium diffusion in plagioclase during analysis as the main mechanism to explain excess aluminium.
Small differences in the composition of plagioclase can only be resolved using homogeneous and geochemically well constrained standards that match the matrix of the analysed mineral (e.g. 36 ). Here, well-characterised end-member compositions of feldspars with nearly perfect recalculated stoichiometry assure that no analytical effect from false standard compositions can affect the results. In addition, instrumental and lab environment conditions (including room temperature, emission current, focal point, etc.) must be stable during an analytical session. In albite-or orthoclase-rich feldspars, alkali element migration under the electron beam of the EPMA is a prominent cause of analytical deviation from the actual composition, as the following example of repeated analysis with focused beam and 20 nA beam current shows (Fig. 4): Elements such as Ca and Fe stay relatively unaffected (around 8.8 ± 0.2 wt% CaO and 0.2 ± 0.02 wt% FeO) during repeated measurement with small beam diameter at high beam current at the EPMA. However, accumulated counts of Na and thus apparent Na 2 O concentrations markedly decrease (from 6.1 to 0.9 wt%) while apparent SiO 2 and Al 2 O 3 concentrations increase (from 56.5 to 62.5 wt% and from 27.3 to 29.4 wt%, respectively). Moreover, the plagioclase composition did not recover even after several minutes without irradiation. Consequently, (a) the totals of the analyses, which are often used as an indicator of analysis quality, only increase slightly (from 99.6 to 101.7 wt%), (b) calculated apfu (based on 8 oxygens) of plagioclase decrease (from 5.00 to 4.63), (c) measured anorthite contents in plagioclase are increased artificially (here from ca. An 43 to An 82 ), and (d) data are shifted away from the stoichiometry line into the "fertile" field in the diagram Ca/(Ca + Na + K) vs. Al/(Ca + Na + K). This extreme example implies that repeat analyses and too long analysis times with a small beam diameter and high beam current can lead to small deviations from the stoichiometry line that may then be misinterpreted as Al*.   We suspect that element diffusion may have occurred in the literature database presented by Williamson et al. 29 , their figure 1 (Fig. 5). In particular, the dataset highlighted in red, which shows apparently high Al*, follows a trend that corresponds well to ca. 15% Na loss (dashed line). Other datasets appear systematically offset as well, but follow trends (sub-)parallel to the stoichiometry line. This, in turn, might be an artefact related to the standardisation.

Discussion
The study of both natural and synthetic samples indicates that Al* in plagioclase is not suitable as a geo-hygrometer due to analytical limitations. In particular, differences in Al* in plagioclase are too small to be resolved by EPMA-WDS routinely. Furthermore, the aluminium excess versus deficiency in plagioclase as described in 29 is unlikely to be related to P H2O in the melt because no differences between plagioclase compositions in samples from dry or wet melts were found. Independent of its incorporation mechanism, Al* in plagioclase does not appear to be a robust means to explore for mineralisation, as the mineralised samples studied here do not show any Al* variations beyond uncertainty. Seemingly large differences of Al* in plagioclase, as observed in literature compilations 29 , may instead be analytical artefacts due to inappropriate standardisation or Na loss during EPMA analysis. Hence, the correlation between mineralisation and Al* in plagioclase remains uncertain, and other factors that influence the mineral composition need to be studied.

Methods
Piston cylinder experiments. Plagioclase was crystallised from synthetic melts in the system diopside (Di)albite (Ab) -An ± water 37 . The Di component was included for depolimerisation of the melt and to promote growth of larger crystals. Other components such as Fe and K were avoided, as they potentially disturb the incorporation of Al* in plagioclase 29 . Water concentrations were initially high in one experiment, intended to impose a   standards were used, respectively. Higher-order excitation lines were suppressed using energy windows in the single channel analyser. Standardisation was improved by repeated analysis of the main standards as unknowns. Plagioclase crystals from natural samples were analysed with 15 kV acceleration voltage (to excite the Fe-Kα line), 20 nA beam current, and 10 μm beam diameter -conditions at which alkali migration is negligible during the analysis time even in albite-rich samples. The small size of plagioclase crystals from experiments precluded analysis with a large beam diameter, so that a focused beam was used and the acceleration voltage was reduced to 12 kV (to lower the excitation volume). Consequently, the beam current was decreased to ca. 4.5 nA. Suitable analysis conditions for plagioclase at the EPMA were determined by monitoring of count rates over time. The uncertainties were calculated from the relative count statistical error (r) based on the signal intensities of the standard compared to the samples, which was then translated into the uncertainty of the apfu.
Compositions of experimental glasses and pyroxenes were analysed with standard-based energy-dispersive X-ray spectroscopy (EDS) using a JEOL JSM-6390 LA scanning electron microscope (SEM), equipped with a Thermo Fisher NSS7 EDS system (30 mm 2 silicon drift detector) and an in-column Faraday cup attached to an Ampere-meter, at the Department of Earth Sciences, ETH Zurich. Analysis conditions were: 15 kV acceleration voltage, ca. 2.3-2.4 nA beam current (recorded before each measurement), corresponding to ca. 20-25% dead time, and 30 s acquisition time. A ZAF matrix correction was applied to the analyses. Expected uncertainties of the measurements are in the range of a few percent for major elements.
Availability of materials and data. All data generated during this study are included in this published article and its Supplementary Information file.