Water Solubility in Fe‐Bearing Wadsleyite at Mantle Transition Zone Temperatures

Wadsleyite can store significant amounts of H2O in its crystal structure as hydroxyl. However, H2O solubility in Fe‐bearing wadsleyite remains poorly constrained at mantle transition zone temperatures. Previous studies (e.g., Demouchy et al., 2005 [https://doi.org/10.2138/am.2005.1751]; Litasov et al., 2011 [https://doi.org/10.1007/s00269-010-0382-3]) focused primarily on Fe‐free systems, which do not represent the Earth's interior because Fe may affect the H2O solubility. Here, we investigated the temperature dependence of H2O solubility in Fe‐bearing and Fe‐free wadsleyite at 1500–2100 K. The results indicate that H2O solubility in Fe‐bearing wadsleyite is higher than in Fe‐free samples at 1800–1900 K, corresponding to transition zone geotherm, but there is no clear Fe content dependence in the Fe‐bearing samples. Wadsleyite can contain approximately 1.0 wt.% H2O at transition zone temperatures. The H2O solubility in wadsleyite is lower than ringwoodite along a plume geotherm, which may result in dehydration melting at the 520‐km discontinuity by upwelling flow in plumes.

by Bolfan-Casanova et al. (2018) under comparable conditions. Kohlstedt et al. (1996) reported a H 2 O solubility of 2.1-2.4 wt.% in Fe-bearing wadsleyite, but their experiments were performed at 1370 K, substantially lower than transition zone temperatures (1800-1900 K at 410-520 km depth Katsura et al., 2010]). By considering the temperature dependence of H 2 O solubility in minerals Litasov et al., 2011) and possible Fe effect on H 2 O incorporation, a systematic study regarding H 2 O solubility in Fe-bearing wadsleyite is required to understand the H 2 O incorporation and storage capacity in the mantle transition zone.
In this study, we measured H 2 O solubility in Fe-bearing wadsleyite at 17.5 and 21 GPa as a function of temperature from 1500 to 2100 K, which covers the entire temperature range for wadsleyite in the Earth's interior from cold slabs to hot spots. H 2 O solubility in Fe-free wadsleyite was also investigated for comparison. Note that oxygen fugacity (f O2 ) may affect H 2 O solubility in upper mantle minerals (Liu & Yang, 2020;Yang, 2016). However, as we investigated recently (Druzhbin et al., 2021), H 2 O solubility in Fe-free wadsleyite is independent from f O2 . It should be also the case for Fe-bearing wadsleyite because dissociation of H 2 O is negligible under transition zone conditions (Druzhbin et al., 2021). This is proved by the positive correlation between Fe 3+ /ΣFe and f O2 , but no correlation between Fe 3+ /ΣFe and H 2 O-solubility (McCammon et al., 2004;Mrosko et al., 2013); with increasing Fe 3+ /ΣFe from 28% to 96%, the H 2 O-solubility in wadsleyite does not show any variation (McCammon et al., 2004). Therefore, the f O2 in this study is self-buffered by the samples without additional buffering materials.

Starting Material and High-Pressure Experiments
MgO, SiO 2 , FeO, and Mg(OH) 2 (purities >99.99%) were used as the starting materials. The MgO and SiO 2 were dried at 1370 K in an ambient-pressure furnace, whereas FeO and Mg(OH) 2 were dried at 400 K in a vacuum furnace prior to weighing. Mixtures with a bulk composition of Mg 2 SiO 4 and (Mg,Fe) 2 SiO 4 + 5 or 15 wt.% H 2 O (Table 1) were prepared by grinding in ethanol using an agate mortar. The powders were stored in a vacuum furnace at 400 K prior to use (Fei & Katsura, 2020  High-pressure experiments were performed using the multi-anvil technique. To reduce the amount of Fe absorbed by the sample capsule, Pt-Rh capsules were used. The Fe-content should be within 2.0 wt.% in a 1 ∼ 2-μm layer on the inner wall of the capsule and less than 0.2 wt.% at 6 μm away from the wall (Fei & Katsura, 2020), which is negligible in comparison with the total amount of Fe in the samples. The starting materials were welded in Pt-Rh capsules with inner diameter of 1.0 mm, outer diameter of 1.2 mm, and length of 1.0-1.3 mm. One or two capsules were loaded into an MgO sleeve in each multi-anvil cell assembly with a LaCrO 3 heater, ZrO 2 thermal insulator, and Cr 2 O 3 -doped MgO octahedron with an edge length of 10 mm. The assembly was compressed to 17.5 or 21 GPa at room temperature using eight tungsten carbide cubes with truncated edge lengths of 5 (for 17.5 GPa runs) or 4 mm (for 21 GPa runs). The temperature was sequentially increased to the desired value using a ramping rate of ∼100 K/min measured by a D-type (W/ Re) thermocouple and annealed for 30-300 min ( Table 1). The assembly was then quenched to ambient temperature by switching off the heating power, and the pressure was reduced to ambient condition over a duration of more than 10 h.

Sample Analysis
Cross sections of the recovered capsules were prepared by polishing with sandpaper and diamond powder. The textures in the cross sections were observed by scanning electron microprobe (SEM) using a backscattering electron detector (BSE) with an acceleration voltage of 15 kV and an energy-dispersive spectrometer (EDS).
The concentrations of MgO, SiO 2 , and FeO in wadsleyite were measured using a JEOL JXA-8200 electron probe microanalyzer (EPMA) with a wavelength-dispersive spectrometer (WDS). A forsterite single crystal was used as the standard for MgO and SiO 2 , and a metallic Fe standard was used for FeO. The acceleration voltage and beam current were 15 kV and 15 nA, respectively, with a counting time of 20 s for each analysis point using a focused beam. The (Mg + Fe)/Si atomic ratios and Fe contents [Fe/(Mg + Fe)] were calculated from the EPMA results (Supporting Information).
The C H2O in wadsleyite was analyzed by unpolarized Fourier transformation infrared spectroscopy (FTIR) using a Bruker IFS 120 high-resolution spectrometer coupled with a Bruker IR microscope. The sample capsules were double-side polished to a thickness of approximately 20 μm for FTIR analysis. To examine the possible saturation of infrared absorption, some samples were further polished to a thickness of 12 μm (Table 1). One or two hundred scans were accumulated for each FTIR spectrum at a resolution of 2 cm −1 . Crystal boundaries and cracks were avoided. Five or six spectra from multiple grains with random crystallographic orientations were obtained for each sample.
After baseline correction, the C H2O in wadsleyite was calculated from the Beer-Lambert law, H2O is the molar weight of H 2 O (18.02 g/mol), ε is the absorption coefficient (69,000 ± 7,000 L/ (mol cm 2 ) and 67,000 ± 5,000 L/(mol cm 2 ) for Fe-free and Fe-bearing wadsleyite, respectively [Bolfan-Casanova et al., 2018]), τ is the sample thickness, and ρ is the density (3,500 g/L [Jacobsen et al., 2005]). The integration is performed in the range of 3,000-4,000 cm −1 . The uncertainty of C H2O is calculated from the standard deviation from the multiple spectra, uncertainty of ε, and uncertainty of the thickness measurements (±1 μm). Note that if using ε = 73,000 ± 7,000 L/(mol cm 2 ) reported by Deon et al. (2010), the retrieved C H2O values are systematically lower by 5%-8% (Table 1).

Sample Textures
The samples appear as wadsleyite single crystals with grain sizes of 20-500 μm coexisting with or without enstatite [(Mg,Fe)SiO 3 or MgSiO 3 ], and quenched crystallized melt as confirmed by SEM and EDS ( Figure 1). Superhydrous phase B appeared instead of enstatite in H4898 likely because of the low temperature stability of the dense hydrous magnesium silicate phase.

FTIR Spectra
Raw data of FTIR spectra without baseline subtraction and thickness normalization are shown in Figure 2a. The raw data can be directly compared because all of the samples have a comparable thickness (from 19 to 22 μm). A plateau at 3,300-3,350 cm −1 does not appear, which indicates no infrared absorption saturation (Druzhbin et al., 2021). This is also confirmed by the identical infrared absorptance (identical C H2O ) between the samples with thickness of 21-22 and 12 μm after thickness normalization ( Figure 2b; Table 1).
All of the spectra show sharp peaks at wavenumbers of 3,610, 3,581, 3,355, and 3,325 cm −1 , which agree with previous studies (e.g., Bolfan- Casanova et al., 2018;Deon et al., 2010;Druzhbin et al., 2021;Jacobsen et al., 2005;Litasov et al., 2011;Smyth et al., 2005). The peak at 3,610 cm −1 in the Fe-bearing samples is significant, whereas that in the Fe-free wadsleyite is negligible. In contrast, the 3,581, 3,325, and 3,355 cm −1 peaks are comparable between FEI AND KATSURA 10.1029/2021GL092836 4 of 9  The solid red and blue curves denote spectra at thickness of 22 and 12 μm, respectively. The dash curves are after thickness normalization to 10 μm. The agreement of infrared absorbance after thickness normalization indicates no saturation of infrared absorption. All spectra are raw data without baseline subtraction but vertically shifted for visibility. Sinusoid shape wave interference occurs because of the small sample thickness. FTIR, Fourier transformation infrared spectroscopy. the Fe-free and Fe-bearing systems (Figure 2a). Therefore, the infrared absorption at 3,610 cm −1 is likely caused by protons associated with Fe 3+ that formed due to the substitution of one Si 4+ or two Mg 2+ atoms by Fe 3+ and H + .

C H2O /Solubility in Fe-Free Wadsleyite
The C H2O in the Fe-free wadsleyite show a clear temperature dependence: decreasing from 2.4 wt.% at 1500 K to 0.6 wt.% at 2000 K ( Figure 3; Table 1). Because three phases (wadsleyite + enstatite/superhydrous phaseB + melt) coexist in the Fe-free system that has three components (MgO + SiO 2 + H 2 O), the C H2O in wadsleyite should be uniquely constrained at a given pressure and temperature conditions based on the phase rule, and therefore represent the H 2 O solubility despite the variable H 2 O contents in the starting material. The consistency of C H2O values in the runs at 17.5 and 21 GPa indicates a relatively small pressure dependence of H 2 O solubility. The H 2 O solubility in Fe-free wadsleyite determined in this study generally agrees with previous results (Bolfan-Casanova et al., 2018;Demouchy et al., 2005;Druzhbin et al., 2021;Inoue et al., 1995;Jacobsen et al., 2005;Litasov et al., 2011) (Figure 3).

C H2O /Solubility in Fe-Bearing Wadsleyite
Similar to the Fe-free system, the C H2O in Fe-bearing wadsleyite also show a negative temperature dependence: decreasing systematically from 2.2 wt.% at 1500 K to 0.65 wt.% at 2100 K, but the temperature dependence is slightly smaller than that in Fe-free wadsleyite (Figure 3). At 1500 K, the C H2O values in Fe-free and Fe-bearing wadsleyite are essentially the same. But at 1900-2100 K, the Fe-bearing wadsleyite has clearly FEI AND KATSURA 10.1029/2021GL092836 5 of 9  Druzhbin et al. (2021) include the standard deviation of C H2O from the multiple spectra, uncertainty of infrared absorption coefficients (Bolfan-Casanova et al., 2018), and uncertainty of the thickness measurements (±1 μm). The error bars for previous studies are based on their reported values, and with a correction factor of 100/55 if Paterson (1982) calibration was used. The mantle transition zone geotherm at 410-520 km depth is from Katsura et al. (2010), whereas the plume geotherm is assumed to be 200 K higher. FTIR, Fourier transformation infrared spectroscopy.
higher C H2O than the Fe-free samples (Figure 3). The C H2O of the samples synthesized from Fo90 + 15 wt.% H 2 O is within uncertainty identical to that from Fo75 + 15 wt.% H 2 O (Figure 3).
The C H2O in Fe-bearing wadsleyite measured in this study is higher than the values reported by Bolfan-Casanova et al. (2018). Note that the H 2 O content in the starting material in Bolfan-Casanova et al. (2018) was relatively small (∼5%). Since only two or three phases (wadsleyite, melt, enstatite/superhydrous phaseB) coexist in the Fe-bearing system which has four (MgO + SiO 2 + H 2 O + FeO) or five (+Fe 2 O 3 ) components, the number of phases is smaller than the number of components. Therefore, even though the pressure and temperature are fixed, the C H2O in the samples are not uniquely constrained and may vary with bulk H 2 O content in the starting material. As demonstrated in Fei and Katsura (2020), if the starting material contains insufficient amount of H 2 O (e.g., Fo90 + 5 wt.% H 2 O), the C H2O in the samples may not have reached its maximum (i.e., the solubility). Instead, with 15 wt.% H 2 O, the C H2O is indendent from the bulk H 2 O content in the starting material and represents the H 2 O solubility (Fei & Katsura, 2020). Sun et al. (2018) reported C H2O = 1.0 wt.% in Fe-bearing wadsleyite synthesized from Fo90 + 10 wt.% H 2 O at 1720 K, which is slightly lower than C H2O determined in this study. Note that the Paterson (1982) calibration was used in Sun et al. (2018), which may underestimate C H2O . By adjusting their results to a newer calibration using a correction factor of 100/55 given by Bolfan-Casanova et al. (2018), the C H2O in Sun et al. (2018) becomes 1.7 ∼ 1.8 wt.%, which is slightly higher than in this study (∼1.4 wt.% at 1720 K) (Figure 3). However, this difference could be caused by experimental uncertainties, including the uncertainty of the correction factor between different infrared calibrations, experimental temperature, and FTIR spectra baseline subtraction. Additionally, Kawamoto et al. (1996) reported exceptionally high C H2O at 1700 K (C H2O ≈ 3.0 wt.%), their data points do not show a temperature dependence, which suggests some errors (C H2O overestimated or temperature overestimated) because H 2 O solubility should decrease with increasing temperature (Figure 3).

(Mg + Fe)/Si Ratio in Hydrous Wadsleyite
To confirm the validity, the C H2O values obtained by FTIR in this study are independently examined by the (Mg + Fe)/Si atomic ratio because protons are primarily incorporated into the Mg sites in wadsleyite and the (Mg + Fe)/Si ratio should therefore decrease with increasing C H2O (Sano-Furukawa et al., 2011;Smyth, 1987). The (Mg + Fe)/Si ratio determined by EPMA in this study is plotted as a function of C H2O and compared with previous studies of both Fe-free and Fe-bearing systems (Bolfan-Casanova et al., 2018;Demouchy et al., 2005;Druzhbin et al., 2021;Inoue et al., 1995;Litasov et al., 2011) in Figure 4. Although different analytical methods were used to determine C H2O in these studies, including FTIR, secondary ion mass spectroscopy (SIMS), and elastic recoil detection analysis (ERDA), the (Mg + Fe)/Si ratios follow the same C H2O -(Mg + Fe)/Si relation of two protons substituting on one Mg site. This supports the robustness of the C H2O values determined by FTIR spectroscopy in the present study.
The slopes of the (Mg + Fe)/Si data are slightly steeper than that for pure Mg 2+ = 2H + substitution (Figure 4). This difference may be caused by experimental uncertainty (e.g., uncertainty of the infrared absorption coefficients, uncertainty of the FTIR spectra baseline subtraction, and EPMA uncertainty). On the other hand, the (Mg + Fe)/Si ratio in Fe-bearing wadsleyite is slightly higher than the Fe-free samples, which is likely caused by an additional proton incorporation mechanism in Fe-bearing wadsleyite. Because Fe in wadsleyite can be partially ferric (McCammon et al., 2004), substitution mechanisms of Fe 3+ -H + exchange on Si 4+ sites may occur in additional to the Mg 2+ = 2H + substitution, which will slightly increase the (Mg + Fe)/Si ratio in comparison with the pure Mg 2+ = 2H + substitution mechanism.

Implications for H 2 O Storage Capacity in the Mantle Transition Zone
Although H 2 O solubility in wadsleyite decreases with increasing temperature, Fe-bearing wadsleyite can still contain approximately 1.0 wt.% H 2 O in its crystal structure in the upper part of the mantle transition zone corresponding to 410-520 km depth with a geotherm of 1800-1900 K (Figure 3). This value is essentially the same as that of ringwoodite at 520-660 km depth in the lower part of the mantle transition zone with a geotherm of 1900-2000 K (Fei & Katsura, 2020). This implies that the entire mantle transition could be H 2 O-rich. Reducing conditions in the Earth's interior (Frost & McCammon, 2008) may limit the H 2 O content in some minerals (e.g., Yang et al., 2016;Zhu et al., 2019), however, such limitation may not occur in the mantle transition zone because H 2 O dissociation is negligible in the stabilized wadsleyite field as demonstrated by Druzhbin et al. (2021). A H 2 O-rich mantle transition zone is therefore compatible with low oxygen fugacity and high temperature conditions. This agrees well with previous predications of a H 2 Orich mantle transition zone (e.g., Fei et al., 2017) and findings of naturally formed H 2 O-rich ringwoodite and ice-VII inclusions (Pearson et al., 2014;Tschauner et al., 2018).
In the case of mantle plumes where temperatures are approximately 200 K higher (∼2100 K at 520-km depth) than the ambient mantle, the H 2 O solubility in Fe-bearing wadsleyite is approximately 0.65 wt.%. On the other hand, 1.0 wt.% H 2 O can be dissolved in Fe-bearing ringwoodite at 2100 K ( Figure 3). If ringwoodite in the lower part of the mantle transition zone is nearly H 2 O-saturated (Fei et al., 2017), dehydration melting should occur at the 520-km discontinuity caused by the phase transformation from ringwoodite to wadsleyite under H 2 O-saturated conditions driven by upwelling flow in mantle plumes. By assuming 1.0 wt.% H 2 O in ringwoodite (Fei & Katsura, 2020), 0.65 wt.% H 2 O in wadsleyite (this study), a H 2 O content of ∼16% in hydrous silicate melt at 2100 K (Fei, 2021), and a ringwoodite or wadsleyite volume fraction of ∼55% (Frost, 2008), a mass balance calculation indicates a melt fraction of 1.2 vol.% in the dehydration melting layer. Such a high fraction is sufficient to completely wet the grain boundaries of wadsleyite and thus reduce its viscosity and seismic velocity. Therefore, in addition to previously reported dehydration melting layers at the 410-and 660-km discontinuities (Revenaugh & Sipkin, 1994;Schmandt et al., 2014;Vinnik & Farra, 2007), it is predicted that the 520-km discontinuity is also associated with the low velocities and low viscosities near plumes, although the thickness of the dehydration melting layer might be small, and significant amounts of H 2 O are already extracted by plume upwelling, thus seismically difficult to image.