Mass transfer in the Earth’s interior: fluid-melt interaction in aluminosilicate–C–O–H–N systems at high pressure and temperature under oxidizing conditions

Characterization of the processes that govern the budget and recycling of volatile components in the Earth’s interior is central to our understanding of the formation and evolution of the solid Earth, its oceans, and atmosphere (e.g., Kasting et al. 1993; Jambon 1994; Tolstikhin and Marty 1998; Deines 2002). However, our knowledge of the nitrogen abundance in the Earth’s mantle and in mantle-derived magma is not well established with model-dependent mantle abundance suggested between about 1 and 35 ppm (Marty, 1995; Javoy, 1997). The nitrogen abundance likely also is heterogeneously distributed with considerably greater nitrogen contents in subduction zone environments where abundances above 0.1% N₂ have been reported (Fisher et al. 2002). This latter environment likely also is where most of the exchange of volatiles (including nitrogen) occurs between the solid Earth and the atmosphere (Schmidt and Poli, 2003; Bebout, 2007).

Transfer of nitrogen from the Earth’s interior to the surface and atmosphere is accomplished via partitioning of nitrogen between solid mantle, partial melts and perhaps multicomponent fluids with melt/mineral and fluid/mineral N₂ partition coefficients exceeding 1. Mass transport is at the core of these processes.

Characterization of transport processes requires data and models that describe physico-chemical properties of silicate melts and C–O–H–N fluids at high temperature and pressure under controlled redox conditions. Redox conditions are important because even within the redox range of natural magmatic processes in the Earth’s interior (Carmichael and Ghiorso 1990; Wood et al. 1990), carbon and nitrogen, in particular, may occur in different oxidation states (Libourel et al. 2003; Kadik et al. 2004 2010; Mysen et al. 2008 Mysen et al. 2011; Li and Keppler 2014). The nitrogen solubility and solution mechanism(s) in silicate melts are sensitive to the redox state of nitrogen (Libourel et al., 2003; Mysen and Yamashita, 2010; Kadik et al. 2015) and needs to be controlled, therefore, during experiments in order to describe the behavior of nitrogen during melting and crystallization in the Earth’s interior.

Nitrogen is but one of several volatile components in the Earth’s interior with the system C–O–H–N comprising the main components (Jambon, 1994). Solubility and solution mechanisms of C–O–H–N components in silicate melts and of silicate components in C–O–H–N fluid are, therefore, necessary, and may be carried out experimentally (e.g., Franz 1998; Newton and Manning 2005 2006; Li and Keppler 2014; Kadik et al. 2015). However, most of that information is from analysis of glasses formed by temperature-quenching of volatile-containing melts to form a glass at ambient temperature at high pressure followed by decompression and, finally, examination of these glasses. Application of such data to characterize high-temperature/pressure melts and fluids relies on the assumption that there are no changes of these materials as they are brought from high temperature and high pressure to ambient conditions. The structure of a glass formed is this way is not that of the melt. Rather, it is the structure frozen in at the glass transition (e.g., Moynihan et al. 1976; Richet and Bottinga 1986; Dingwell and Webb 1990; Richet et al. 1992) because melt structure is temperature (and pressure) dependent (e.g., Stebbins 1988; Mysen et al. 2003; Agee 2008; Yamada et al. 2011; Wu et al. 2014). Moreover, in melts that contain volatiles, their speciation can be temperature- and pressure-dependent (Nowak and Behrens, 1995; Mysen and Yamashita 2010; Wu et al. 2014). In addition, during quenching to ambient temperature and pressure, some or all of the volatiles may exsolve from the cooling melt because the solubility of volatiles in melts varies with both temperature and pressure. Therefore, neither the volatile content nor the speciation of a glass is that which existed in its molten state at high temperature and pressure.

An additional problem associated with examination of quenched material as a proxy of high-temperature/pressure is that the structure of C–O–H–N fluids in equilibrium with silicate at high temperature and pressure cannot be quenched for quantitative structure study at ambient conditions. This is partly because silicate and other oxides dissolved in fluids at such high temperatures and pressures often precipitate during quenching (e.g., Zhang and Frantz 2000; Mysen and Armstrong 2002; Kessel et al. 2004), and partly because the C–O–H–N fluid structure may change as pressure and temperature are reduced from the experimental to ambient conditions (e.g., Mysen and Yamashita 2010; Wu et al. 2014). In summary, it is necessary, therefore, to study melts and fluids at the temperature and pressure of interest in order to attain quantitative understanding of their structure and their structure-property relations.

Recent experimental studies have addressed structural and compositional aspects of silicate–COH fluid + melt systems while samples were at high temperature and pressure (Ni and Keppler 2013; Mysen 2015a, b). Less is known about chemically more complex situation in which the C–O–H–N system of volatiles in equilibrium with aluminosilicate melts. The objective of the present study is to address this latter type of system in order to provide experimental data with which to further our understanding the structure of fluids and melts of aluminosilicate–C–O–H–N systems relevant in the temperature and pressure range of the Earth’s crust and upper mantle under oxidizing conditions.

The experiments were conducted under conditions where nitrogen exists in its oxidized state, N₂. We note, though, that nitrogen may be reduced to amide and ammine forms at and above hydrogen fugacities between those defined by the NNO–H₂O (nickel–nickel oxide) and MW–H₂O (magnetite–wüstite) redox conditions (Mysen et al. 2008), which are redox conditions within the range encountered in the Earth’s interior (Wood et al. 1990; Frost and McCammon, 2008). This information is, therefore, relevant only to the most oxidized part of this range of redox conditions.

Optical and vibrational spectroscopic data from the samples were recorded while these were at the desired high pressure and temperature while contained in externally heated, hydrothermal diamond anvil cells (HDAC) of the type originally described by Bassett et al. (1996). In this cell, the upper and lower diamonds are surrounded by Mo wrap-around wire heaters. Two thermocouples in contact with the upper and lower diamonds, respectively, are located within less than 1 mm from the sample. The samples, themselves, were contained in the cell within 125 μm thick Ir gaskets with a 500 μm diameter sample hole. Iridium was used as gasket material because it does not react with the sample (melt and fluid) at the temperature and pressure conditions of these experiments.

The temperature was measured with K-type thermocouples that were calibrated against the melting temperature of NaCl (800.5 °C). Temperature error was about ± 3 °C with the thermal gradient across the sample volume less than 2 °C. Pressure was generated by the fluid at high temperature in the cell. Pressure is, therefore, a dependent variable in these experiments. It was determined from the one-phonon Raman shift of ¹³C synthetic diamond, which was used to avoid interference with the Raman signals of ¹²C from the diamond cell itself (Bassett et al. 1996; Schiferl et al. 1997; Mysen and Yamashita 2010). Ten to 30 μm across diamond pieces were grown from ¹³CH₄ by using CVD techniques (Liang et al. 2009). Pressure uncertainty at high temperature is ± 110 MPa and precision ± 40 MPa determined from the precision of the Raman spectrometer (± 0.1 cm⁻¹ by using Ne lines as internal spectrometer calibrant) and the uncertainty in the calibration by Mysen and Yamashita (2010).

The samples in the diamond cell were examined through optical microscopy and via confocal micro-Raman and microFTIR methods while at the temperature and pressure of interest. In the Raman spectroscopic studies, the samples (melt and fluid) were excited with the 490 nm line of a solid state laser in a JASCO model NRS-3100 confocal micro-Raman spectrometer. The spectrometer is equipped with a single monochromator, a holographic notch filter, and holographic gratings with 600, 1200, and 2400 grooves/mm. The laser was focused to about 1 μm diameter at the sample with power at the sample typically near 30 mW. Spectra were recorded through a × 50 magnification/0.42 N.A. long-working distance Mitutoyo objective lens. An Andor Model DV401-F1 1024 × 128 pixel (25 μm pixel size) Peltier-cooled CCD was used for signal detection. By using the confocal optics, > 85% of the signal was recorded from within 30 μm of a sample surface (Mysen 2015a). For infrared spectroscopy, a JASCO model IMV4000 FTIR microscope system operating in transmission mode with 10X objective and condenser cassegrain lenses, an InSb detector, CaF₂ beam splitter, and a halogen light source was employed.

Staring materials were aluminosilicate glass chips of composition (Na₂Si₄O₉)₉₀(Na(NaAl)₄O₉)₁₀ (denoted NA10) together with solid Ag₂C₂O₄ and AgN₃ as sources of CO₂ and N₂, respectively. No reducing agent was added to the starting materials so that all components, including N₂ and CO₂, remain in their oxidized form. The (Na₂Si₄O₉)₉₀(Na(NaAl)₄O₉)_1o glass was from the batch originally synthesized by Mysen (2007), but remelted and quenched to glass to be used for starting materials here in order to eliminate incipient crystals and adsorbed and water that may contributed to the sample during its storage. The Ag₂C₂O₄ breaks down to metallic Ag and CO₂ gas in between 150° and 200 °C and AgN₃ to metallic Ag and N₂ gas under similar temperature conditions (Boettcher et al. 1973; Roskosz Mysen and Cody, 2006). A few experiments were also conducted with Ag₂C₂O₄ + H₂O only in order to generate reference spectra of volatiles in the COH fluid system.

The sample volume defined by the central hole in the Ir gaskets is about 0.025 mm³, which poses a major challenge for quantitative addition of individual components to the starting material. To overcome some of the problems posed by these very small samples, a mixture of NA10 + Ag₂C₂O₄ + AgN₃ with 8.7 wt% CO₂ and 8.0 wt% N₂ was prepared prior to sample loading. Double-distilled and deionized H₂O was added with the aid of a micro-syringe, but the proportion of H₂O relative to solid materials could not be determined quantitatively because the sample masses are on the order of 0.0x μg. It was, nevertheless attempted to add total solid (NA10 + Ag₂C₂O₄ + AgN₃) and liquid H₂O visually estimated to about 1:1 volume proportions.

The FTIR measurements were conducted with the same sample contained in the diamond cell as that used for Raman measurements. This sample was subjected to the same temperatures and pressures by using the same protocol as the Raman measurements and was carried out after the Raman measurements were completed. Pressure could not be measured with the diamond shift in the FTIR system. It was assumed, therefore, that for given temperature, the pressure recorded with the ¹³C diamond in the Raman measurement series was the same during the subsequent FTIR measurements. The validity of this assumption was assessed by monitoring the temperature of gas exsolution from the fluid phase during the Raman and FTIR experiments. Gas exsolution at the same temperature at constant sample volume implies the same pressure at each temperature. In the present experiments, this exsolution temperature was within 10°–15 °C in the Raman and FTIR measurement series. This uncertainty corresponds to a ± 0.02 g/cm³ density uncertainty for pure H₂O, which for the H₂O density of the experiments reported here corresponds to ± 20 MPa. Equation-of-state data for the fluid mixture are not available, but given that H₂O is a dominant volatile component, ± 20 MPa uncertainty seems reasonable. This conclusion implied a temperature/pressure evolution in the FTIR experiments within error the same as in the Raman measurement series.

The Raman and infrared transmission spectra include contributions from the diamond anvils. This background signal was eliminated in the FTIR spectra by normalizing to those of an empty diamond cell at the same temperature after the sample was removed. For the Raman spectra, the procedure is more complex, and the diamond background was subtracted by using the procedure of Dalou et al. (2015). Briefly, this required recording Raman spectra of the diamond cell away from the sample hole in the Ir gasket but within the same focal plane as the sample measurement at the same temperature and pressure. This spectrum was then subtracted from the sample spectra (melts, fluids, and supercritical fluids), which results in a spectrum without signal contribution from the second-order Raman signal intensity from the diamond cell itself.

The subtraction method used for the Raman spectra could not be used over more than several hundred cm⁻¹ frequency range in part because the sensitivity of the CCD detector varies with frequency. Subtraction over wider ranges could, therefore, result in artificial intensity variations. Moreover, in frequency regions with rapid intensity variations, the spectrometer positioning uncertainty (2–3 cm⁻¹) could result in artificial sharp intensity variations. Care was, therefore, exercised to narrow the frequency ranges of Raman spectra to within those of particular interest when conducting the background correction. The infrared spectra suffered from complications resulting from major variations in intensity across the spectra so that in order to obtain reasonable signals from CO₂ and from the overtone/combination bands of water species (see also below) portions of the spectra could not be recorded as the signals commonly saturated the detector. Sample thickness is governed by the gasket thickness and could not, therefore, be varied greatly. It is for this reason that only portions of the infrared spectra are shown.

The solubility mechanisms of volatile species in silicate–C–O–H–N melts qualitatively resemble those in the simpler systems such as silicate–H₂O and silicate–COH. Addition of nitrogen to the system under oxidizing conditions results in formation of molecular N₂, which does not interact chemically with the other volatile components.

Water in the form of molecular H₂O dominates in C–O–H–N fluids with OH groups forming less than 3% of the total water concentration even at the highest temperatures and pressures (Fig. 3). This is less than the proportion of OH groups in pure silicate–H₂O systems at similar temperatures and pressures, where this proportion often exceeds 10% (Mysen 2010). In fact, by using the relationship between proportion of OH in fluid, expressed as A₄₅₀₀/(A₄₅₀₀ + A₅₂₀₀) from the FTIR data, and fugacity of H₂O for the same silicate composition and pure H₂O (Mysen 2010) (see Fig. 12a), the much lower proportion of OH groups in the C–O–H–N–silicate fluid implies very considerably dilution of water because the f_H2O of pure H₂O at the pressures and temperatures of these experiments is in the 200–700 MPa range.

Equivalent OH/H₂O for melt were not obtained for reasons discussed above. However, the overall degree of silicate polymerization as a function of water fugacity, using the NA10–H₂O data from Mysen (2010) for calibration, results in an equivalent f_H2O quite similar to that of pure H₂O (Fig. 12b). This observation may suggest that the presence of C and N components in the melt has a much smaller influence on activity composition relations in the (C–O–H–N)-saturated melt than in fluid. This difference reflects the much smaller solubility if silicate components in the fluid than in the melt.

Fluid/melt partition coefficients can be obtained from the ratio of integrated intensities of bands in spectra of coexisting melts and fluids assigned to specific vibrations provided that the instrumental parameters such as acquisition time, slit settings, and focal depth are same during acquisition of both spectra. This can be done because in a ratio of integrated intensities from bands assigned to the same vibration, the remaining variables affecting Raman intensity is temperature, which, of course, is the same for coexisting fluid and melt. Then, the partition coefficient of component, i, between fluid and melt, f/m, is;

where A _i ^f and A _i ^m are the integrated areas from the Raman spectra of a band assigned to a specific type of vibration in component, i.

The partition coefficient for total H₂O (which includes all OH-bearing species) can be derived with Eq. (3) from the 3600 cm⁻¹ band intensity in spectra of fluids and melts (Fig. 5). This partition coefficient is always greater than unity, which is similar to the partition coefficient in silicate–H₂O systems (Mysen 2013; Webster et al. 2014). From the linear fit to the ln K_H2O^f/m vs/ 1/T (T is temperature in kelvin), the enthalpy from this partioning is − 6.5 ± 1.5 kJ/mol (Fig. 13), which is quite similar to the results in the simpler silicate–H₂O (Mysen 2013) where ∆H = − 7.6 ± 0.7 kJ/mol. That fact that this ∆H value differs from 0 reflects, most likely, the different extent and type of speciation of H₂O in the coexisting melt and fluid. We should also note that the decreasing partition coefficient with increasing temperature actually reflects an increase in pressure as well, which results in increased silicate solubility in fluid and decreased silicate content of hydrous melt (its water content increases). In the isobaric (0.2 GPa) experiments by Webster et al. (2014), no temperature-dependence was observed. That notwithstanding, their fluid/melt partitions fall naturally on the extension of the present data and those of Mysen (2013).

For carbon species in melts and fluids in the silicate–C–O–H–N system, there are also similarities between the data and those in the simpler silicate–COH system (Mysen 2015a). However, there also are differences. For example, in the silicate-saturated C–O–H–N fluid, we have evidence only of molecular CO₂ and CO₃² groups, whereas in the silicate–COH system, Mysen (2015a) also observed evidence of HCO₃^- groups in the fluid. This difference implies that in the C–O–H–N fluids studied here, the simple carbonate equilibrium;

is shifted further to the left than was the case in the COH fluids studied by Mysen (2015a). The implication is that the C–O–H–N fluids examined here are more alkaline that those examined by Mysen (2015a).

For C–O–H–N-saturated melts, on the other hand, there is spectroscopic evidence for both CO₃² and HCO₃^- groups. No evidence could be found for molecular CO₂ in the spectra of melt. A simple solution mechanism involving an equilibrium between CO₃² and HCO₃^- is (Mysen 2015a):

In Eq. (5), Q ⁿ , and Qⁿ⁻¹, are silicate species with, respectively, n and n−1 bridging oxygen. The CO₃^2-/HCO₃^- abundance ratio in melts increases with increasing temperature with ∆H = 114 ± 22 kJ/mol (Fig. 14). The HCO₃^- abundance in melt diminishes rapidly with increasing temperature and can barely be observed at 750 °C and cannot be seen in spectra of single phase fluids at 800 °C. In comparison, in the N-free system silicate–COH, also examined in the hydrothermal diamond anvil cell with similar protocol, the CO₃/HCO₃^- ratio was determined for the fluid to be considerably less sensitive to temperature with a ∆H = 17 ± 3 kJ/mol (Fig. 14).

The fluid/melt partition coefficients for CO₃^2- groups are in the range of ~ 0.2 to ~ 0.7 as a function of increasing temperature and pressure with an enthalpy of 52 ± 14 kJ/mol (Fig. 15). These values fall in the range reported by Webster et al. (2014) and Mysen (2015b) even though the compositions in those experiments differed significantly from those of the present compositions.

Even though nitrogen exists in coexisting fluid and melt only as molecular N₂, there are differences in the interaction between the C–O–H–N fluid matrix in fluid and N₂ and silicate matrix in melts and N₂. This difference results in distinctly temperature-dependent N₂ partition coefficients with an enthalpy change (− 20 ± 4 kJ/mol; see Fig. 16) that is considerably greater than the − 6.5 ± 1.5 kJ/mol observed for H₂O partitioning between fluid and melt, for example (Fig. 13). The nitrogen partition coefficients also are more than 20% greater (and increasing with decreasing temperature) than for H₂O. It would seem, therefore, that the activity-composition relations of N₂ in melts, and perhaps fluids, differ from those of H₂O. This difference likely is because H₂O may dissolve both as molecular H₂O and as OH groups forming bonding the silicate network, whereas there is no evidence that N₂ behaves in such a manner.

It follows from the observations here and in the literature (Mysen et al. 2008; Li and Keppler 2014) that where redox conditions in the Earth are more oxidizing than those defined by the NNO–H₂O and FMQ–H₂O buffer nitrogen released into fluid or melt will most likely exist as molecular N₂. In other words, there is no evidence to indicate chemical interaction between N₂ and other volatile and silicate species under such circumstances.

Redox conditions similar to these are those observed in subduction zones, for example (Wood et al. 1990; Cottrell and Kelley 2011). Under these conditions, nitrogen effectively serves as a diluent of the volatile species. Dilution is indicated, for example, by the low OH/H₂O ratio of the C–O–H–N fluid (Fig. 12), which indicates lesser water activity than in silicate–H₂O fluids. This much lower OH/H₂O ratio probably this because silicate solubility in C–O–H–N fluids must be considerably less than in the absence of N₂. This follows from simple relationships of the form;

Where Si⁴⁺ is a proxy for silicate components and Si(OH)³⁺ a proxy for hydroxylated silicate. This means that as the activity of silicate decreases, the equation shifts to the left thus reducing the proportion of OH groups in the system,

It also appears, therefore, that a decrease of silicate activity in solution in the fluid resulting from the dilution of H₂O by other fluid components enhances fluid alkalinity. This feature, in turn, may alter the solubility of pH-dependent minor and trace elements in the fluid (e.g., high field strength cations; see Ayers et al. 2012) and, therefore result in different element patterns resulting from C–O–H–N fluid release than from release of fluids without N₂, where the silicate solubility is greater and the fluid likely is less alkaline. These features are, of course, also dependent on redox conditions because under reducing conditions nitrogen may form ammonia or ammine-type complexes (Mysen and Fogel, 2010; Li and Keppler 2014), which results in significant changes the behavior of nitrogen in both fluids and melt.

It appears that C–O–H–N-saturated melts are less sensitive to the presence of N₂ than C–O–H–N fluids as both the degree of melt polymerization, NBO/T, and silicate species abundance, appear insensitive to the existence of molecular N₂. However, as is the case for fluids, when magmatic activity occurs where conditions are sufficiently reducing to nitrogen to ammonia and ammine groups (somewhere near the hydrogen fugacity defined by the magnetitie–wüstite–H₂O redox buffer; see Mysen et al. 2008), this may change because the solubility of reduced nitrogen in silicate melts is much greater than oxidized nitrogen. Furthermore, the solution mechanism of reduced nitrogen as NH-complexes resembles that of dissolved H₂O (Mysen et al. 2014). Thus, nitrogen in magmatic liquids under such reducing conditions would tend to enhance melt properties that are sensitive to melt polymerization.