For the synthesis of NH4-phengite, we followed the method described in Watenphul et al. (2009) using 25 % NH4OH solution in excess. They succeeded in the synthesis of phengite of composition (NH4)(Mg0.5Al1.5)(Al0.5Si3.5)O10(OH)2 at 4.0 GPa, 700 ∘C. A homogenous mixture of SiO2, γ-Al2O3 and MgO oxides in the ratio 3.5 : 1 : 0.5 was used as solid starting material, which corresponds to an NH4-phengite of the composition given above. Two Au capsules of 10 mm length, 3 mm in diameter and 0.2 mm wall thickness were filled with NH4OH solution and about 7 mg of the solid starting material (of different fluid / solid ratios of about 0.4 (run NAD1) and 0.8 (run NAD2)) and immediately closed by cold welding. The two capsules were placed in pyrophyllite-NaCl-steel-furnace assemblies of 12.7 mm (1/2 in.) diameter in a piston-cylinder press. Experiments were performed at 4.0 GPa and 700 ∘C for 24 h. P was calibrated according to the quartz–coesite transition (Mirwald and Massonne, 1980). The estimated P uncertainty is about 0.04 GPa. The T was monitored by a chromel–alumel thermocouple, with a precision of ±2 ∘C. The estimated uncertainty in experimental T is about 10 ∘C. At the end of the experiment, the sample was quenched isobarically to a T below 200 ∘C in less than 15 s, followed by slow release of P. The two capsules were cleaned, reweighed and then opened. The solid products were washed in pure water to avoid the presence of any adsorbed NH4+ on the mineral surface and prepared for optical investigations, powder X-ray diffraction (XRD) and spectroscopic analyses.

T-dependent infrared (IR) investigations were also performed on synthetic and natural ammonium-free phengite samples. The synthetic phengite is sample 10 from the study by Melzer and Wunder (2000), containing 99 % phengite of composition (K0.85Cs0.05)(Al1.58Mg0.46)(Al0.54Si3.46)O10(OH)2 together with traces of quartz. Natural phengite from the Gran Paradiso area (sample AL 17; see Chopin and Maluski, 1980), of the composition (K0.9Na0.01)(Al1.45Fe0.12Mg0.43)(Al0.43Si3.56Ti0.01)O10(OH)2, was investigated.

Solid run products were mounted on metal stubs with double-sided stick tape, coated with carbon and characterised optically via scanning electron microscopy using the Gemini ultra-high-resolution FE-SEM from Zeiss located at GFZ Potsdam. The SEM was operated at an accelerating voltage of 3.00 kV, a working distance (WD) of 2.3 mm and an aperture size of 30 µm.

The run products were ground for several minutes in an agate mortar and mixed with Elmer's white glue, before being spread evenly on a zero-scattering Mylar polymer circular foil. Preferential orientation of the crystals was minimised by constantly stirring the sample during drying. The sample was covered with another clean foil and mounted onto a transmission sample holder. Powder XRD patterns were recorded in transmission with a fully automated STOE STADI P diffractometer using Cu K-α1 radiation at 40 kV and 40 mA, a take-off angle of 6∘, a primary monochromator, and a 7∘ wide position-sensitive detector (PSD). The intensities were recorded between 5 and 125∘ 2θ with a detector step size of 0.1∘ and a resolution of 0.02∘. Collected patterns were processed using the GSAS software package (Toby and Von Dreele, 2013) with the Rietveld method for phase identification, phase proportions and unit cell parameters. The refinement procedure for the NH4 phases followed that of Pöter et al. (2007). Initial crystal structures of the respective K+-endmember polytypes were taken from the Inorganic Crystal Structure Database (ICSD, FIZ Karlsruhe, http://icsdweb.FIZ-karlsruhe.de, last access: 20 February 2020), with structures modified by replacement of K+ with NH4+, using the lattice parameters of NH4-phengites given by Watenphul et al. (2009).

Prior to all measurements, samples were kept in the oven at 170 ∘C for a few days to eliminate adsorbed water. Micro-FTIR absorbance spectra were recorded using a Bruker VERTEX 80v FTIR spectrometer in combination with a Bruker HYPERION 2000 microscope equipped with 15× Cassegrain objectives. For the measurements down to 400 cm-1 (i.e. in the 4000–400 cm-1 wave number range), a globar light source, a KBr beam splitter and a mercury cadmium telluride (MCT) detector were used. For the measurements in the OH stretching region (i.e. in the 4000–2500 cm-1 wave number range), a tungsten light source, a CaF2 beam splitter and an InSb detector were used. Dried air was used to purge the IR microscope, and the main optical bench of the system was kept under vacuum during the IR measurements. The ambient and T-dependent spectra were collected with a spectral resolution of 2 cm-1 averaged over 256 scans, with an aperture size of 50µm×50µm. Such an aperture size was possible even for the fine-grained samples as the measurements were performed on prepared thin films and not on the fine-grained single crystals (see below). The P-dependent spectra were accumulated with 4 cm-1 spectral resolution, averaged over 512 scans, with the aperture size set to match the sample at about 20µm×30µm. T- and P-dependent spectra were collected on the same selected area both for sample and background measurement. The spectra were analysed with the program PeakFit v4.11.

Powder X-ray diffraction data revealed the synthesis of different NH4-phengite polytypes (2M2, 2M1, 3T, 2Or and 1M in decreasing order of their abundance) in various proportions. For the NAD1 sample, coesite and kyanite were formed as additional minor phases. In the experimental run with the higher fluid / solid ratios (NAD2), NH4-phengites are exclusively synthesised with no such additional minor phases.

The observed and calculated patterns of the Rietveld refinement of X-ray data for both samples are shown graphically in Fig. 1. The lattice parameters of the two main NH4-phengite polytypes (2M2, 2M1) calculated from the refinement are shown in Table 1, as well as the lattice parameters of NH4-phengite polytypes produced by Watenphul et al. (2009) and the K analogues by Massonne and Schreyer (1986) for comparison.

FE-SEM micrographs of run products of sample NAD1 are shown in Fig. 2. Thin flakes of well-crystallised NH4-phengite crystals are formed and show characteristic pseudohexagonal form in cross section. Chemical analysis by electron microprobe was not applied (a) due to the fact that NH4 phases tend to decompose under the electron beam during long-term exposure (Mingram and Bräuer, 2001; Watenphul et al., 2009) and (b) because the grain sizes of our NH4-phengite crystals are too small (≤ 1 µm). However, NH4+ incorporation into phengite was confirmed by FTIR spectroscopy (see below).

The lattice modes of phengite have been assigned by means of IR and Raman spectroscopy (e.g. Goryainov et al., 2017; McKeown et al., 1999). In addition to the lattice modes, other vibrational modes in NH4-phengite come from the NH4+molecule located at the interlayer cation site, as well as from the OH situated in the octahedral layer (OH stretching vibration). The IR spectrum of the NAD2 sample at ambient conditions is shown in the Supplement (Fig. S1). An isolated NH4+ molecule with ideal Td symmetry has four normal vibrational modes, namely symmetric (ν1) and antisymmetric (ν3) stretching vibrations, as well as symmetric (ν2) and antisymmetric (ν4) bending vibrations (Herzberg, 1966; Kearley and Oxton, 1983). Two of those vibrations are IR-active, i.e. the triply degenerate antisymmetric stretching (ν3) and bending (ν4) vibrations. Previous studies of NH4-bearing silicates (e.g. Busigny et al., 2003; Liu et al., 2019; Vennari et al., 2017; Watenphul et al., 2009; Wunder et al., 2015; Yang et al., 2017) show that the band at 3610 cm-1 is assigned to the OH stretching vibration in phengite, and the bands at 3291 and 1430 cm-1 correspond to the NH4+-ν3 and ν4 vibrations, respectively. The assignment of the ν3 and ν4 modes is based on the assumption that the NH4+ molecule behaves almost as a free tetrahedral molecule in the interlayer cation site in the phengite structure (Watenphul et al., 2009). The band at 3036 cm-1 is related to a combination of symmetric and antisymmetric bending vibrations (ν2+ν4 – the Raman-active ν2 mode occurs at 1669 cm-1, Price et al., 1991) and the band at 2818 cm-1 is ascribed to an overtone of the ν4 bending vibration (2ν4). All of the FTIR bands of the NH4-phengite including the lattice vibrations and the respective references are summarised in Table S1.

The quantification of the amount of NH4+ in micas was done using IR spectroscopy applying the method described by Higashi (2000). In his method, the ratio of the linear absorbances of the NH4-ν4 band to the OH stretching vibration (i.e. A(NH)/A(OH)) is used to quantify NH4+ in muscovite-tobelite solid solutions and endmembers; this ratio was used because the intensity of the ν4 band is a function of the NH4+ content, whereas the intensity of the OH band is constant. Accordingly, we used the linear absorbances of the OH stretching vibration and the NH4+-ν4 band of the background-subtracted spectra of the thin-film samples of NAD2 and of a sample from Watenphul et al. (2009) diluted in KBr (450 mg KBr : 1 mg sample) for the ratio calculation. The calculated (NH4)2O in both samples is 8.2±0.2 wt %. Using another method, the equation given by Pöter et al. (2007) for NH4+ quantification in muscovite-tobelite solid solution, (NH4)2O content is 9.6±0.2 wt % (equivalent to the calculated XNH4+mica=1.4±0.03 pfu). An alternative approach for NH4+ quantification in muscovites proposed by Busigny et al. (2003) is only calibrated for traces of NH4+ and thus cannot be applied for phengites with several wt % of NH4+ (wt % in the present paper vs. ppm in their study). Bentabol and Cruz (2016) reported calibrations to estimate the amount of NH4+ in muscovite-tobelite solid solutions using FTIR and X-ray diffraction methods. Their calibration methods also cannot be applied for our samples.

FTIR spectra in the range of 4000–400 cm-1 of NH4-phengite (NAD2) measured with increasing P at ambient T are shown in Fig. 6. With increasing P, the NH4+-ν3 antisymmetric stretching vibration is shifted to higher wave numbers, accompanied by an increase in its absorbance. In contrast, the OH stretching band shifts to lower wave numbers, becomes broader and appears to vanish. Beyond 25 GPa, the OH and ν3 bands merge and can no longer be distinguished. The P-induced wave number shifts of the OH stretching band and the NH4+-ν3 bands are plotted in Fig. 7a; results are shown only up to 25 GPa because spectra measured at Ps above 25 GPa cannot be reliably fitted. The slope of the regression lines fitted through the shift in wave number – at least that of the NH4+-ν3 band – as a function of P shows a small but discernible change at Ps above 8.6 GPa. The full widths at half maximum (FWHMs) of NH4+-ν3 and the OH stretching vibrations as a function of P are shown in Fig. 7b, in addition to the regression lines fitted through. Changes in the slopes of these regression lines show an inflection point around 8.6 GPa. Despite the large scatter of data, for the OH vibration, there is a general positive trend of the slope on either side of this inflection point but with larger magnitude above 8.6 GPa, whereas for the NH4+-ν3 band, the slope is positive below 8.6 GPa and negative above. The linear absorbances of NH4+-ν3 and the OH stretching vibrations as a function of P shown in Fig. 7c indicate a noticeable discontinuity at Ps between 5 and 8.6 GPa. At Ps lower than 5 GPa, both vibrations show comparable decrease in absorbance with P. At Ps above 8.6 GPa, the absorbance of the NH4+-ν3 vibration increases and the absorbance of the OH stretching decreases.

The evolution of the NH4+-ν4 band with P up to 42 GPa is shown in Fig. 7d. The splitting of the NH4+-ν4 antisymmetric bending vibration band into three peaks is observed with increasing P at ambient T. Similar splitting but only to two peaks was observed previously with cooling (see Fig. S3). This splitting and sharpening of the peaks starts at Ps of about 8.6 GPa, and the slopes (dν/dP) are shown on the regression lines fitted through. In Fig. 7e and f, the FWHMs and linear absorbances of the the NH4+-ν4 vibration are plotted vs. P. The discontinuity in the slopes of FWHMs and linear absorbances of the OH stretching and NH4+-ν3 vibrations at 8.6 GPa (Fig. 7b and c) is also observed for the NH4+-ν4 vibration even more explicitly (Fig. 7e and f).

NH4-phengite endmember with composition (NH4)(Mg0.5Al1.5)(Al0.5Si3.5)O10(OH)2 was successfully synthesised at 700 ∘C, 4.0 GPa and 24 h. NH4+ incorporation into phengite is confirmed by both XRD and IR methods. The lattice dimensions (Table 1) of the main polytypes synthesised here (2M2 and 2M1) are enlarged compared to its K analogues synthesised by Massonne and Schreyer (1986) and are comparable to the NH4-phengite polytypes produced by Watenphul et al. (2009) at identical synthesis conditions. This enlargement is mostly along the c direction owing to the slightly larger ionic radius of NH4+ compared to K+ in the interlayer between the silicate sheets.

The IR spectrum at ambient conditions (Fig. S1) shows the IR-active modes of the NH4+ molecule as well as the expected phengite lattice vibrations (Table S1). The OH stretching vibration has an identical wave number as in ammonium-free phengites (K-phengites), suggesting that the incorporation of NH4+ has no visible effect on the OH vibration band at ambient conditions, which is reasonable from a structural point of view; both NH4+ and K+ have the same charge and approximately the same ionic radii when 12-fold coordinated, so the neighbouring structure should be comparable in both cases.

Quantification of NH4+ from IR spectra using different available calibration methods was quite challenging. To our knowledge, no method is calibrated for the quantification of NH4+ in phengites, neither in tobelite–NH4-phengite nor NH4-phengite–phengite solid solutions. Therefore, we used the calibration methods available for muscovite-tobelite solid solutions.

According to charge balance, the maximum amount of NH4+ that could be present in phengite is 1 pfu or 6.9 wt % (NH4)2O. The calculated (NH4)2O amount of 8.2±0.2 wt % using the Higashi (2000) IR calibration and the calculated (NH4)2O amount of 9.6±0.2 wt % using the Pöter et al. (2007) method exceed the maximum amount of NH4+ that could be present in phengite. Therefore, we assume that the interlayer cation sites are fully occupied with NH4+, and, accordingly, we assume that the phengite produced in the current study has the composition (NH4)(Mg0.5Al1.5)(Al0.5Si3.5)O10(OH)2. This suggests that the IR calibration of Higashi (2000) and Pöter et al. (2007), which were established for the muscovite-tobelite solid solution series, overestimates the NH4+ content in phengite. This holds as a confirmation to the importance of the establishment of a calibration method for NH4+ quantification in phengite as previously suggested by Bentabol and Cruz (2016), who highlighted that mineral-specific calibrations for NH4+ quantification are required for different types of mica minerals and different solid solutions.

The splitting of the NH4+-ν3 stretching and NH4+-ν4 bending vibrations at 3291 and 1430 cm-1 respectively with decreasing T observed here in in situ experiments (Figs. 3, 4a, S2 and S3) has also been observed at low Ts in the IR spectra of other NH4+-bearing micas like synthetic tobelite (Mookherjee et al., 2002a), synthetic phlogopite (Mookherjee et al., 2002b) and in natural NH4+-bearing phengites (Yang et al., 2017). This splitting as well as the sharpening of the NH4+-ν3, ν4, the overtone (2ν4) and the combination bands (ν2+ν4) is attributed by the formerly mentioned studies to a low-T ordering process of the NH4+ molecule based on the proposal by Pauling (1930), who first suggested such orientational ordering of tetrahedral molecules (like NH4+) and the possibility of its transition from free rotation on cooling. Furthermore, with increasing T up to 600 ∘C, the decrease in absorbances of NH4+-ν3 and the OH stretching vibrations cannot be attributed to OH or NH4+ loss, since the spectra measured at ambient conditions before and after heating are comparable and thus indicate a reversible process (Fig. 3). The integral absorbance of the ν3 band at 3291 cm-1 (Fig. 4b), however, decreases by 86.9 % upon heating to 600 ∘C despite the concentration of NH4+ being the same as at ambient conditions. This decrease in the integral absorbance of the NH4+-ν3 band despite no loss of NH4+ can be explained by (i) the NH4+ molecule in the interlayer cation site getting disordered with increasing T and/or (ii) a T-induced decrease in the molar extinction coefficient of the NH4+ molecule. Yang et al. (2017) has assigned this to a change in molar extinction coefficient

The reader is referred to the fact that Yang et al. 2017 used the term absorption coefficient to express the molar extinction coefficient (ε). This is different from our usage for the term absorption coefficient in the current study.

The wave number shift of NH4+-ν3 at 3291 cm-1 with increasing T (Fig. 4a) observed here reflects the lengthening of the N–H interatomic distance of the NH4+ molecule with increasing T up to 600 ∘C. This expansion is reversible upon cooling back to room temperature. Similar shifting to lower wave numbers and lengthening of the N–H bond was also observed in the in situ T-dependent study performed by Yang et al. (2017) but only up to 500 ∘C on natural samples containing few parts per million of NH4+. At Ts higher than 500 ∘C, they observed no further N–H bond lengthening and they proposed that one hydrogen atom is detached from the NH4+ molecule and linked to one of the basal oxygen atoms in the tetrahedral sheets, initiating the devolatilisation processes. Evidently, then, NH4-phengite endmember (studied here) and phengite containing only trace amounts (ppm) (such as that used in Liu et al., 2019, and Yang et al., 2017) of NH4+ behave differently during heating in terms of NH4+ expansion. Accordingly, from this in situ experiment, OH and NH4+ seem to be stable and are still retained in the structure while heating up to 600 ∘C at least for a short period of time (about 60 min). On the other side, our quenched long-term (24 h) annealing experiments (Figs. 5, S4) showed that loss of NH4+ and dehydration starts at Ts of 550 ∘C. This variation in Ts at which devolatilisation takes place with the experimental heating duration confirms that the kinetic effects partly control the destabilisation of NH4+ in phengite in our experiments.

Comparing rates of devolatilisation in quenched long-term (24 h) annealing experiments, NH4-phengite endmember (NAD2) thin film (Figs. 5a, S4a) starts to lose NH4+ and dehydrates at Ts of 550 ∘C. This T is much lower than the T of NH4+ loss in single crystals of natural NH4+-bearing phengite with traces of NH4+, in which dehydration and NH4+ loss were observed to start at 800 ∘C (Liu et al., 2019; Yang et al., 2017). This difference in the T might be due to the nature of the sample (i.e. thin film vs. single crystal, synthetic vs. natural). Our comparative study on thin films (NH4- and ammonium-free) vs. single crystals indicates that the devolatilisation at lower Ts is due to the sample preparation (thin film vs. single crystal). Moreover, NH4+ loss occurs at the same Ts as the dehydration (Fig. 5d). The synchronous decrease in the intensities of the vibrations in which the OH is involved at 910 cm-1 with the decrease in the NH4+ and OH bands confirms dehydration. The fact that also the vibrations at 750–830 cm-1 decrease let us suggest that H must be involved in these vibrations. However, it seems that even complete loss of OH and NH4+ at 650 ∘C (thin films) does not lead to breakdown of the phengite structure as the main vibrational units can still be observed (Fig. S4).

The shift of the OH band to lower wave numbers with increasing P as observed here (Fig. 7a) indicates the weakening of the O–H bond in the hydroxyl group through a lengthening of the O–H distance caused by shortening of the O⋯H and O⋯O hydrogen bond distances (Libowitzky, 1999). The OH stretching band becomes broader and seems to decrease in its absorbance (Figs. 6, 7c). The FWHM of the OH band displays a change in the slope from 1.7 cm-1 GPa-1 at Ps up to 8.6 GPa to 6 cm-1 GPa-1 for higher Ps (Fig. 7b). The wave numbers of the NH4+-ν3 band shift with increasing P to higher values. The shift to higher wave numbers indicates an increase in N–H bond strength (Harris and Bertolucci, 1989; Nakamoto, 1978, as cited in Beran et al., 2004) and hence shortening of the bond length generated by compression. This shortening of the bond length is much more pronounced at Ps above 8.6 GPa; the shift of peak positions (dν/dP) increases from 1.5 cm-1 GPa-1 at Ps below 8.6 GPa to 5.0 cm-1 GPa-1 at higher Ps (Fig. 7a).

Recalling the concept of dipole moment, in the NH4+ molecule, there are four similar N–H bonds (dipoles) that could be stretched antisymmetrically in relation to the Td symmetry elements giving rise to three identical stretching vibrations (NH4+-ν3), and there are H–N–H angles that could be bent in relation to the Td symmetry elements giving rise to three identical bending vibrations (NH4+-ν3); if any of these H–N–H angles change, then the resultant vector for the dipole of each N–H bond changes, the resultant partial dipole moment will change and the tetrahedral shape of the molecule (Td) will consequently change. This concept is of relevance with respect to the continuous increase in the absorbance of the NH4+-ν3 vibration observed between 5 and 10 GPa (Figs. 6, 7c). This increase with increasing P can then be explained by a P-induced increase in the partial dipole moments of the NH4+ molecule. Most probably the orientation of the dipoles (angles) changes with increasing P such that partial dipole moments increase. This change in the orientation of the partial NH4+ dipoles is further emphasised by the splitting of the ν4 vibration into three peaks at Ps > 8.6 GPa (Fig. 7d). The P-induced discontinuity in the changes in the FWHM, the band positions, and the intensities of the OH and NH4+-ν3 bands, as well as the observed splitting of the NH4+-ν4 at Ps of about 8.6 GPa, all indicate local structural rearrangement affecting/reducing/lowering the Td symmetry of the NH4+ molecule. The mechanism and type of symmetry reduction as a response to P are different from those involved with T.

Our finding that NH4+ remains stable in phengites at higher Ts and Ps than previously thought has important implications for the cycles of nitrogen and water between the Earth's crust and the mantle. Comparison between the composition of the bulk Earth and of carbonaceous chondrites shows that the Earth is depleted in nitrogen (Marty, 2012). Recent studies (e.g. Zedgenizov and Litasov, 2017) have suggested that this depletion can be due to “unaccounted” nitrogen that is stored in the Earth's deep interior. However, nitrogen has been reported in two forms: firstly in the form of NH4+ within for example micas (Busigny et al., 2003, 2004; Higashi, 1982) and feldspar (e.g. Erd et al., 1964) in terrains that experienced high-grade metamorphism (Hall et al., 1996), and secondly in the form of N2-bearing micro- and nano-sized inclusions that may persist to great depths (Rudloff-Grund et al., 2016). In this regard, knowledge of the extent to which nitrogen that is subducted is released from metamorphic rocks during devolatilisation is extremely important in answering questions like the following. (1) Does nitrogen in the mantle have a primordial or recycled origin? (2) If it is recycled, how and what transports it to great depths? (3) What are the different P–T conditions controlling the efficiency of this recycling? The current study provides insights on the behaviour of NH4+ and its connection to the hydroxyl (OH) in phengite, as one of the most important hosts and transporters for NH4+. As we show, complete breakdown of phengite does not occur until great depths, at high Ps and high Ts. As such we can demonstrate that crustal recycling and subduction could constitute a feasible mechanism for delivery of recycled N to the mantle, indicating that N stored in the mantle may not necessarily be primordial in its source. For the hydrogen cycle, according to previous studies (Ono, 1998; Schmidt, 1996; Schmidt and Poli, 1998), water transported into the mantle via phengite should not be lost until the complete breakdown of the mineral down to 300 km depths. However, the effect of NH4+ on the stability and dehydration rates of phengite and other NH4+-bearing hydrous minerals has up to now remained uncertain. Liu et al. (2019) suggested that the presence of NH4+ would trigger dehydration at shallower depths – in that the presence of NH4+ increases the efficiency of phengite dehydration. Based on our data, it is tempting to suggest that NH4+ does not have such an effect on recycling efficiency. However, the observed effect could also be attributed to the sample prepared as a thin film vs. single crystal rather than to the ammonium content. For better constraining of the P and T conditions at which NH4-phengite transports nitrogen into the deep levels of the upper mantle, spectroscopic investigations at simultaneous pressure and temperature conditions are needed.