A natural fluorophlogopite sample from the Cardiff Uranium Mine, Ontario, Canada, from the collection of the Museum of Nature Hamburg – Mineralogy, was used in this study. The Cardiff Uranium mine lies within the Precambrian Grenville Province, which hosts several similar fluorine- and uranium-rich metasomatic deposits. Phlogopite at Cardiff occurs within fluorine-rich metasomatic and hydrothermal veins developed along syenite gneiss–marble contacts, where F-bearing fluids produced assemblages including phlogopite, fluorapatite, fluorite, scapolite, and humite-group minerals (Robinson and Hewitt, 1959). A euhedral single crystal with dark-brownish colour and a pseudo-hexagonal tabular shape (dimensions ∼ 26 × 24 × 6 mm³; see Fig. 1) was selected for the experiments. Rectangular cuts parallel to the cleavage plane were prepared with a diamond wire saw. The size of the cuts was approximately 3 × 2 × 0.15 mm³, with the shorter edge parallel to the monoclinic (110) crystallographic plane.

The chemical composition of an untreated specimen, as well as of the specimen heated to 1450 K (the maximum temperature reached in our experiments), was determined via wavelength-dispersive electron microprobe analysis (WD-EMPA). Data were collected with a Cameca SX-100 scanning electron microscope system operating at 15 kV accelerating voltage and 20 nA beam current. To avoid sample overheating and subsequent mobilization of light elements, the beam was defocused to a spot with a linear size of ∼ 10 µm on the sample surface. The employed standards were LiF for F (F K_α); albite for Na (Na K_α); MgO for Mg (Mg K_α); corundum for Al (Al K_α); andradite for Si (Si K_α), Ca (Ca K_α), and Fe (Fe K_α); vanadinite for Cl (Cl K_α); orthoclase for K (K K_α); MnTiO₃ for Ti (Ti K_α) and Mn (Mn K_α); NiO for Ni (Ni K_α); olivenite for Cu (Cu K_α); Pb-containing glass for Zn (Zn K_α); SrTiO₃ for Sr (Sr L_α); and Ba-containing glass for Ba (Ba L_α) and Cr₂O₃ for Cr (Cr K_α). The acquisition times were 20 s for Mg, Al, Si, K, Ca, and Fe; 30 s for Na, Cl, and Ti; 60 s for Mn, Ni, Cu, Zn, Sr, Ba, and Cr; and 120 s for F. The detection limits are 250 ± 50 ppm for Na, Mg, Al, Cl, K, Ca, Ti, Cr, and Ni; 350 ± 50 ppm for Si and Cu; 500 ± 50 ppm for Mn, Zn, Sr, and Ba; 800 ± 50 ppm for Fe; and 4200 ± 200 ppm for F. Measurements were done in 20 separate points. The results in average oxide wt % are given in Table S1 in the Supplement. The amounts of ^MFe³⁺ and ^X(OH)⁻ were calculated using the charge balance approach of Li et al. (2020), as described in detail in Aspiotis et al. (2022); titanium (Ti) oxidation state was assumed to be 4+, after Scordari et al. (2013). The calculated crystallochemical formula of the untreated specimen (see Table 1) confirms that the sample studied here is OH-rich fluorophlogopite, according to the currently accepted IMA nomenclature (Rieder et al., 1998).

The atomic structure of pristine fluorophlogopite, as well as of the specimen recovered from 1450 K, was analysed by single-crystal XRD. Measurements were conducted using a Nonius KappaCCD single-crystal diffractometer with graphite-monochromated Mo K_α radiation. A full sphere of diffraction data was measured for both the untreated and heated crystal up to θmax=30°. Pixel intensities have been integrated and corrected using the Eval15 suite of programs (Schreurs et al., 2010). Crystal structure refinement was carried out using Jana2006 (Petříček et al., 2014) based on scattering factors for the uncharged atoms. The structure could be refined in C2/m, suggesting that the studied fluorophlogopite sample belongs to the most common 1M polytype (e,g. Deer et al., 2013). The unit-cell parameters are given in Table 1. Further refinement details are summarized in Table S2 in the Supplement. The crystallographic information files (CIFs) are also included in the Supplement. Minor OH content was refined assuming full occupancy of the X site by fluorine and oxygen. The O–H distance was restrained to 1 Å, and isotropic hydrogen displacement was coupled to the displacement of ^XO by a factor of 1.2. In the heat-treated sample the hydrogen position was fixed using a geometric constraint involving the neighbouring M1 and M2 sites. The M1 and M2 sites were assumed to be fully occupied by Mg²⁺ and Fe²⁺, while the interlayer site was assumed to be fully occupied by K and Na. The tetrahedral-site occupancy was kept constant at 75 % Si and 25 % Al.

According to group-theory analysis, biotite of C2/m symmetry exhibits 30 Raman-active modes: 16 A_g + 14 B_g. The H⁺ cations participate in 2A_g + B_g modes, while K⁺ and M1-site cations do not participate in Raman-active modes (Kroumova et al., 2003; Aspiotis et al., 2022). The parallel polarized Raman spectra are dominated by the fully symmetrical A_g modes, which have four non-zero, non-equivalent polarizability tensor components (αxx, αyy, αzz, and αxz), and, hence, the Raman peak intensities may vary depending on the crystal orientation in relation to k_i and E_i. However, our previous study on series of biotite samples revealed that when -k_i || k_s ⊥ (001), as with the backscattering geometry used here, the parallel polarized Raman spectra remain nearly the same upon rotation of the crystal around the laser beam direction, indicating that αxx≈αyy (Aspiotis et al., 2022). Indeed, the room-temperature parallel polarized spectra measured with E_i || [110], E_i || [010], or E_i || [100] exhibit only subtle differences in the relative intensities (see Fig. S1 in the Supplement).

The observed Raman peaks were attributed to specific atomic vibrations following the peak assignment in previous studies (Loh, 1973; Tlili et al., 1989; McKeown et al., 1999; Scordari et al., 2006; Lacalamita et al., 2011; Aspiotis et al., 2022). The OH-stretching peak at 3710 cm⁻¹ is generated by hydroxyl groups in MgMgMg–OH⁻–K–OH⁻ local configurations. The broad weak peak at 3698 cm⁻¹ is most probably dominated by OH stretching of MgMgMg–OH⁻–K–F⁻ local configurations, but Fe²⁺MgMg–OH⁻–K–OH⁻ local configurations can also contribute to this Raman scattering (Lacalamita et al., 2011; Aspiotis et al., 2022). Lattice dynamics calculations of phlogopite (McKeown et al., 1999) demonstrated that the peaks in the range of 900–1200 cm⁻¹ are generated by TO₄-stretching modes, those in the range of 600–800 cm⁻¹ are dominated by T–O_bridging–T bending and bending and/or stretching vibrations, whereas the MO₆ vibrations contribute mainly to the range below 350 cm⁻¹. Note that in phyllosilicates the TO₄ tetrahedra form a two-dimensional network of six-membered rings, and, therefore, the T–O_bridging–T-bending vibrations can also be considered to be TO₄ ring modes, as labelled in Fig. 2; in particular the strongest peak at 684 cm⁻¹ results from the ring O_bridging-breathing mode of A_g symmetry (Aspiotis et al., 2022). The lowest-energy peak at 92 cm⁻¹ is attributed to the A_g mode involving T₄O₁₀ sheet and X anion translations perpendicular to the cleavage plane (McKeown et al., 1999), and, hereafter, we will refer to this mode as layer out-of-plane translations (see Fig. 2).

Previous studies indicated that the electrical conductivity below 1000 K, as well as the chemical diffusion of F⁻, is strongest along [110] (Li et al., 2016; Zhang et al., 2022). Thus, we have chosen to perform detailed temperature runs with the E_i || [110]. Due to prolonged acquisition time at each temperature, the Raman-scattering measurements were done on two separate days on two different specimens (one day per run, runs 1 and 2; see Table 2), but the data were fully reproducible within the overlapping temperature range. Figure 3a shows selected temperature-dependent Raman spectra measured between 300 and 1450 K. The Raman spectra for the complete heating–cooling sequences for run 1 (300–1100 K) and run 2 (300–1450 K) are presented in Figs. S2–S5 in the Supplement. All temperature-induced spectral changes up to 1100 K were fully reversible on subsequent cooling down (see Fig. 3b) apart from the tiny increase in the wavenumber of the peaks at ∼ 739 and 788 cm⁻¹. A list of all peak positions and FWHMs, along with their uncertainties, is presented in the spreadsheets “Parameters run 1” and “Parameters run 2” in Table S3 of the Supplement.

To track the response of the crystal atomic dynamics to a temperature increase, we have considered phonon expandabilities defined as αω=-1ω0dωdT (Rösche et al., 2022), which can be calculated from linear fits to ωT data points, using the relation ωT=ω0+dωdTT. The complete dataset of the calculated αω values, where ω0 and dωdT are accompanied by their uncertainties, designated as Δω0 and Δ(dωdT), is provided in the spreadsheet “Phonon expandabilities” of Table S3 in the Supplement. A few peaks showed distinct ωT trends in the range of 300–500 and 550–1000 K and, correspondingly, different αω (see Fig. 2, upper plot). It should be noted that, as defined, positive αω values correspond to a trivial positive volume thermal expansion αV=1V0dVdT as ω of the corresponding phonon mode is downshifted with the temperature increase due to enlargement of the interatomic distances and consequent weakening of the atomic bond strength. In the temperature range of 300–500 K, most of the analysed Raman-active phonon modes exhibit conventional positive phonon expandabilities, the values of which decrease with the increase in phonon wavenumber, similarly to magnesian amphiboles (Rösche et al., 2022). Only the TO₄ stretching mode at ∼ 1061 cm⁻¹, comprising mainly vibrations of the TO₄ apical oxygen atoms (McKeown et al., 1999), exhibits a negative αω value in the range of 300–500 K, indicating strengthening of the T–O_apical bond with temperature increase. As apical oxygen atoms are shared between MO₆ octahedra and TO₄ tetrahedra (see Fig. 1), we suggest that heating up to ∼ 500 K initiates weakening of the M–O_apical–T interactions as a consequence of enhanced enlargement of MO₆ volume. Above 500 K the phonon expandability of T–O_apical bond stretching at ∼ 1061 cm⁻¹ becomes the same as that of the TO₄ stretching at 1097 cm⁻¹, which consists of T–O_basal bond stretching (McKeown et al., 1999). Below 550 cm⁻¹ the lowest-energy mode at ∼ 92 cm⁻¹ (layer out-of-plane translations) exhibits a phonon expandability, which is 1 order of magnitude higher than the αω values of the higher-energy modes and becomes negative above 550 K. Moreover, the phonon expandability of the T–O–T bending mode at 739 cm⁻¹, whose wavenumber is sensitive to the A-site occupancy (Aspiotis et al., 2022), also becomes negative above 550 K, which further suggests ongoing dynamical instabilities in the temperature range 550–1000 K. It is worth noting that, despite the low intensity of the Raman peaks at ∼ 739 and 788 cm⁻¹, their positions remain clearly distinct within uncertainties at any measured temperature between 550 and 1000 K (see Table S3 in the Supplement).

Further insights into the temperature-induced transformation processes can be derived from the temperature dependences of the wavenumber and FWHM of the layer out-of-plane translational mode (Fig. 4). The ω92(T) trend shows a plateau-like minimum between 450–550 K, and FWHM₉₅(T) shows a maximum near 600–650 K, which is a classic example of a phonon-mode-driven phase transition (Husson, 1998; Margaritescu et al., 2018; Morana et al., 2023). This observation, coupled with the change in sign of the αω values at 550 K for this Raman vibrational mode, clearly indicates the occurrence of phonon-driven structural transformations at 550–650 K. It is most probable that this phase transition was mirrored by the hard phonon modes reported by Tutti and Lazor (2008) and Zhang et al. (2016). The excess in FWHM₉₂ with respect to a parabolic baseline vanishes at 1000 K, where the anomalous increase in ω92 with increasing temperature also stops, indicating that the transformation processes are complete. Another interesting point is that, between 300 and 1000 K, the baseline of FWHM₉₂ decreases anomalously. Commonly, the phonon FWHM should increase with temperature because of the phonon–phonon interactions and the ensuing phonon decay (Kuzmany, 2009). A decrease in the FWHM would indicate a structural ordering process, which is hardly probable at elevated temperatures. Another plausible explanation is that at 300 K the peak at 92 cm⁻¹ consists of two overlapping components, the energetic difference of which (i.e. Δω) is less than the FWHM of the individual components, and, as a result, they appear as one broad peak. With a temperature increase the two components equalize in energy (i.e. Δω=0), effectively leading to a narrower peak. Given that the peak at 92 cm⁻¹ arises from layer out-of-plane translations, one can speculate that the initial subtle splitting in energy is due to coexisting polytypes or lower translation symmetry along the c axis on a relatively short length scale, below the sensitivity of in-house XRD, which diminishes on heating. Indeed, the structural differences between the unit layers of 1M and 2M1 polytypes consist of tiny displacements of the apical oxygen atom (O3), as well as of the X-site anion (O4) (Takeda and Ross, 1975). The former would affect the intra-layer octahedral–tetrahedral interactions, whereas the latter would impact the interlayer interactions as the A-site cations lie just above and below the X site (Takeda and Ross, 1975). The anomalous negative phonon expandability of the T–O_apical bond stretching at 1061 cm⁻¹ up to 500 K does reveal initial weakening of the octahedral–tetrahedral linkages, while the ω92(T) and FWHM₉₂(T) trends clearly show ongoing interlayer structural instabilities between ∼ 300–1000 K, which should also affect the interlayer K⁺ cations.

Raman-active phonon modes involving vibrations of A-site cations are symmetry-forbidden; however, one can indirectly derive information about the thermal behaviour of the K⁺ cations from the wavenumber of the T–O_bridging–T mode at ∼ 739 cm⁻¹, which is sensitive to the A-site occupancy (Aspiotis et al., 2022). As can be seen in Fig. 5a, ω739(T) has a clear minimum near 650 K, and it anomalously increases with increasing temperature up to 1050 K. Previously we have demonstrated that ω739 increases with the increase in A-site vacancies (Aspiotis et al., 2022). Hence, the ω739(T) trend suggests that, in the range of 650–1050 K, the interactions between the K⁺ cations and T₄O₁₀ sheets weaken considerably due to the ongoing rearrangements in the layer staking, which, in turn, leads to K⁺ mobilization in the interlayer space. This highlights that, between ∼ 650 and ∼ 1050 K, delocalized K⁺ cations can act as charge carriers enhancing the electrical conductivity (Li et al., 2016). On cooling down, ω739(T) shows hysteresis and does not completely recover. Using the relation between ω739 and the amount of ^AK⁺ (xK) for phlogopite, ω739=1529-918xK (Aspiotis et al., 2022), the loss of ^AK⁺ after cooling down from 1450 K is approximately 0.02 apfu, i.e. 2 % K⁺ loss. Although the slight decrease in K content cannot be observed by EMPA due to the typical uncertainty of 1 %–2 % (see Table S1), the apparent ω739 upshift implies such a loss of K. For the sample recovered from 1100 K the change in ω739 corresponds to a K⁺ loss less than 0.01 apfu.

The temperature dependence of the ratio between the sum of the integrated intensities of the OH stretching and the sum of the integrated intensities of the framework vibrations, i.e. of all peaks between 15–1215 cm⁻¹, indicates a subtle but steady increase with temperature up to 1100 K for run 1. This trend could not be observed for run 2 due to the different kinetics resulting from the smaller number of steps within the 300–1100 K temperature range despite the fact that the heating rate was kept constant at 20 K min⁻¹. Nevertheless, a sharp increase can be spotted between 1100 and 1150 K during run 2 (red symbols in Fig. 5b). We attribute this to intensified amplitudes of the H⁺ vibrations just before H⁺ cations to de-bond from the X-site O²⁻, similarly to the thermally induced H⁺ delocalization in amphiboles (Della Ventura et al., 2018; Rösche et al., 2022). Above 1150 K the OH-stretching peaks completely disappear, but they reappear on subsequent cooling down, indicating that between 1150 and 1450 K all H⁺ cations are mobilized and can contribute considerably to phlogopite conductivity. In addition, the intensity ratio also exhibits a maximum near 550–650 K, which could further mark the onset of dynamic instability related to interlayer rearrangements and mobilization of ^AK⁺ cations. Within uncertainties, the intensity ratio IOH-stretching/Iframework completely recovers on cooling down from 1100 K, but after cooling from 1450 K it decreases to 0.035(8) against the initial value of 0.051(6). The change in the intensity ratio indicates 35 ± 10 % reduction in OH⁻ groups. The onset temperature of dehydroxylation (1100 K) is close to the temperature of 1150 K, above which potassium leakage is enhanced (see Fig. 5a), which suggests that the release of OH⁻ groups facilitates the escape of K⁺ ions from the mica structure. There is a fractional amount of lost OH⁻ that exceeds that of ejected K⁺ cations, indicating additional high-temperature processes causing a reduction in hydroxyl groups.

In magnesian amphiboles the occurrence of Fe2+→ Fe³⁺ oxidation could be detected via the appearance of an extra RRS signal near 600 cm⁻¹ (Rösche et al., 2022), originating from FeO₆-related polaron (longitudinal optical polar phonon coupled with electron). Under normal (non-resonance) conditions, polar optical phonons in centrosymmetrical crystals, such as phlogopite (C2/m), are infrared-active but not Raman-active. They become allowed by symmetry when coupled with excited electrons because then the selections rules are changed, and, for Fe-rich amphiboles, this leads to a complete change of the spectrum profile when the polarization of the incident light E_i is parallel to the polaron dipoles (Mihailova et al., 2021; Bernardini et al., 2023, 2024). For Fe²⁺-bearing Mg-rich hydrous silicates, as in the fluorophlogopite sample studied here, RRS signals are expected to appear as additional peaks without changing the overall spectral profile (Rösche et al., 2022).

Extra RRS signals could not be resolved in the high-temperature spectra of phlogopite, but a weak extra RRS signal at 602 cm⁻¹ was detected at room temperature after cooling down from 1450 K (see Fig. 6a), indicating irreversible Fe2+→ Fe³⁺ oxidation. The oxidation of Fe in complex silicates containing TO₄ rings is also mirrored by the minimum in the temperature dependence of the wavenumber of the TO₄ ring breathing mode, which is caused by a local change in the ring puckering to adopt to the smaller size of adjacent Fe³⁺O₆ octahedra (e.g. Watenphul et al., 2017; Della Ventura et al., 2018; Mihailova et al., 2022). Such a minimum is observed for Fe-containing fluorophlogopite at 1300 K (see Fig. 7), suggesting that this is the temperature of the beginning of irreversible Fe oxidation in the studied single-crystal sample. Since only 5 % of the M sites are occupied by Fe²⁺ in the pristine phlogopite sample studied, the MO₆-related phonon modes may not show detectable anomalies at the temperature of Fe oxidation. Yet, a minimum in ω325(T) at 1300 K (see Fig. 7b), as well as a subtle maximum in FWHM₁₉₅ cm⁻¹ (Fig. 7c), was detected, indicating a local structural transformation, most probably related to the shrinking of the MO₆ octahedra when Fe²⁺ irreversibly transforms into Fe³⁺. These two MO₆-related phonon modes also exhibit anomalies near 580 K (see Fig. 7). The ω325(T) trend shows a weak minimum, while the mode near 195 cm⁻¹ exhibits a kink in its FWHM near 580 K (see Fig. 7), indicating a change in the channel of phonon decay (Kuzmany, 2009), and its phonon expandability is reduced above 550 K (see Fig. 2). These anomalies around 580 K of the MO₆-related phonon modes might be due to the interlayer instabilities or the onset of polaron formation, i.e. the onset of reversible Fe2+↔ Fe3+ reaction.

To double check whether reversible onset of Fe²⁺ ↔ Fe³⁺ exchange can be detected above 550 K, we have performed in situ Raman-scattering experiments with a different orientation of the incident light polarization E_i. As has been said above, the Raman-active phonon modes of biotite do not show any anisotropy in intensities within the plane parallel to the cleavage plane. However, biotite is well known for its pleochroism, i.e. strong anisotropy in optical absorption along [100] and [010]. The anisotropy in the electron density of states may affect the orientation of polaron dipoles and, thus, the appearance of thermally activated RRS signals related to reversible Fe²⁺ ↔ Fe³⁺ oxidation in the Raman spectra. Hence, we heated the sample up to 1000 K when E_i was along [010] (run 3), as well as along [100] (run 4). However, no RRS signal was resolved at 1000 K (Fig. 6b). Further studies on biotite with a larger amount of ^MFe²⁺ are necessary to thoroughly check if reversible Fe²⁺ ↔ Fe³⁺ oxidation occurs in mica, similarly to amphiboles (Della Ventura et al., 2018; Mihailova et al., 2021; Bernardini et al., 2023, 2024, 2025).

The room-temperature Raman spectrum of fluorophlogopite heated up to 1450 K shows two weak extra peaks at ∼ 825 and 858 cm⁻¹ (see Fig. 6a), which are assigned to forsterite (Mg₂SiO₄) with an end-member composition (Kuebler et al., 2006; do Nascimento-Diaz et al., 2021). A thermal decomposition of fluorophlogopite to forsterite has been previously detected by Raman spectroscopy and XRD, starting around 1273 K for both F-poor and F-rich phlogopite and being completed at ∼ 1373 K for F-poor phlogopite only (Sun et al., 2022). We analysed the specimen recovered from 1450 K by EMPA and singe-crystal XRD. The results are given in Tables 1 and 2, as well as in Tables S1 and S2 in the Supplement. Thermally induced cracks can be clearly seen in the backscattered electron image (compare Fig. 8b and c), but, overall, the specimen appears to be chemically homogenous, and spatial areas rich in forsterite were not detected within a spatial resolution of 10 µm. Olivine was also not detected by XRD. The XRD analysis indicates only a subtle reduction in the unit-cell volume, which is consistent with the partial oxidation of Fe and a reduction in hydroxyl groups derived from EMPA (see Table 1). Simulations of the measured room-temperature Raman spectrum of fluorophlogopite recovered from 1450 K with weighted spectra of pristine phlogopite and forsterite measured under the same experimental conditions indicate that only a minor fraction of ∼ 1 % of fluorophlogopite decomposed to forsterite (Fig. 8a). The fact that forsterite cannot be resolved by XRD suggests that forsterite grains are of small size, on the nanometre scale, below the detection limit of in-house XRD experiments. An open question remains regarding how exactly phlogopite decomposes to forsterite as no other crystalline phases resulting from the phlogopite structure breakdown were detected (Sun et al., 2022). A simple chemical reaction for the phlogopite end-member composition would be KMg₃AlSi₃O₁₀(OH)2→ Mg₂SiO4+ amorphous KMgAlSi₂O7+ H₂O↑. Although, in general, Raman spectroscopy can probe non-crystalline phases, we could not detect any additional Raman signals that might be related to amorphous silicate, probably because of the weak signal of amorphous material, if any, and the small fraction of decomposed fluorophlogopite. It is worth noting that Trønnes (2002) has suggested that phlogopite in mineral assemblages can break down to forsterite and pyrope, but our experiments on single-crystal fluorophlogopite did not show any traces of garnet as a solid-phase decomposition product, in accordance with the results by Sun et al. (2022).

In order to better follow the spatial distribution of forsterite, we have performed Raman mapping at room temperature on the fluorophlogopite specimen heated up to 1450 K using a microscope with enhanced confocality. Quite surprisingly, at first glance, the spectra collected on the surface with a confocal WITec alpha300R Raman spectrometer and 100 × (NA = 0.90) objective showed much stronger forsterite peaks (Fig. 9a) than the spectra collected with a long-working-distance objective 50 × (NA = 0.35) in in situ temperature-dependent Raman-scattering experiments (see Sect. 2.2.1). It should be emphasized that at room temperature we have collected multiple single-point spectra with the Horiba T64000 spectrometer and objective 50 × to verify the reproducibility of the spectra. To check if the difference in the spectra measured with the two spectrometers was due to different spatial resolution, we have conducted in-depth z-line scans and (x,y)-2D mapping, with an in-depth step size between 0.45 and 1 µm and a lateral step size of 0.5 µm, using a WITec alpha300R spectrometer. These analyses reveal that forsterite is formed mainly on the surface parallel to the cleavage plane, and an exponential fit to the experimentally determined intensity ratio ρ=(I825Fo+I858Fo)(I825Fo+I858Fo+I684Phl) against the distance from the sample surface (Fig. 9b) indicates that the forsterite fraction already decays at a depth of ∼ 0.56 ± 0.03 µm, as determined by the exponential depth decay constant (see figure caption of Fig. 9). Lateral (x,y) Raman maps on the intensity ratio conducted on the sample surface (Fig. 9c) indicate frail submicron-level inhomogeneous distribution of forsterite, even when a threshold ρ value of the intensity ratio in the Raman map is chosen to correspond to a superficial layer of 100 nm in thickness (Fig. 9d). If a ρ value of 0.1, corresponding to a 1 µm thick layer, is considered then the inhomogeneity in the Raman map vanishes, emphasizing that the olivine is formed on the surface.

Overall, understanding the factors that govern the stability and thermal activation of charge carriers of fluorophlogopite is essential for interpreting geological processes in deep-Earth and subduction zones, where hydrous silicates can be regarded to be highly effective metasomatic agents for the lithospheric mantle wedge. The initial weakening of tetrahedral–octahedral sheet interactions below 500 K, triggering the potassium mobilization at 550–650 K, may facilitate the onset of K release into the mantle wedge at higher temperatures. Progressive H⁺ delocalization at ∼ 1150 K and irreversible ^MFe2+→ ^MFe³⁺ oxidation starting at ∼ 1300 K reflect the development of possible charge carriers prior to fluorophlogopite decomposition. Furthermore, the development of local structural and crystallochemical defects, as revealed by in situ Raman spectroscopy, implies that metasomatism of the mantle wedge may develop over a much wider temperature range and consequently span a larger depth profile than previously thought based mainly on established pressure–temperature stability fields of defect-free minerals. Further substitution of F⁻ for OH⁻ will bring these crystallochemical transformations to even greater depths compared to stoichiometric phlogopite, critically extending fluorophlogopite stability fields during subduction and modulating deep volatile cycling (Sun et al., 2022; Steenstra et al., 2024).