Mineralogy of the deep lower mantle in the presence of H2O

Abstract Understanding the mineralogy of the Earth's interior is a prerequisite for unravelling the evolution and dynamics of our planet. Here, we conducted high pressure-temperature experiments mimicking the conditions of the deep lower mantle (DLM, 1800–2890 km in depth) and observed surprising mineralogical transformations in the presence of water. Ferropericlase, (Mg, Fe)O, which is the most abundant oxide mineral in Earth, reacts with H2O to form a previously unknown (Mg, Fe)O2Hx (x ≤ 1) phase. The (Mg, Fe)O2Hx has a pyrite structure and it coexists with the dominant silicate phases, bridgmanite and post-perovskite. Depending on Mg content and geotherm temperatures, the transformation may occur at 1800 km for (Mg0.6Fe0.4)O or beyond 2300 km for (Mg0.7Fe0.3)O. The (Mg, Fe)O2Hx is an oxygen excess phase that stores an excessive amount of oxygen beyond the charge balance of maximum cation valences (Mg2+, Fe3+ and H+). This important phase has a number of far-reaching implications including extreme redox inhomogeneity, deep-oxygen reservoirs in the DLM and an internal source for modulating oxygen in the atmosphere.


INTRODUCTION
At depths of about 660-2890 km from the Earth's surface, the lower mantle represents the largest fraction of our planet (>55% by volume). In the lower mantle, the dominant upper-mantle ferromagnesian silicate mineral (Mg,Fe) 2 SiO 4 breaks down into (Mg,Fe)O ferropericlase (Fp) and (Mg,Fe)SiO 3 bridgmanite (Brg) [1,2], and this main mineral assemblage further transforms to Fp and postperovskite (pPv) near the core-mantle boundary (CMB) [3,4]. Recent experiments revealed that the lower half of the deep lower mantle (DLM) beneath 1800 km in depth may be governed by remarkably different pressure-induced chemistry, leading to a set of unexpected mineralogical, geophysical and geochemical behaviors [5][6][7]. In the DLM, water becomes a strong oxidant that can oxidize iron [8] and common iron oxides [9] to form hydrogen-bearing FeO 2 H x (x < 1) with pyrite structure (Py) [10][11][12]: However, pure iron and wüstite (FeO) are not major minerals in the lower mantle, and only become abundant when approaching the iron core; iron in the mantle is mostly present as an endmember component of ferromagnesian silicates or oxides. In the upper mantle and crust, Fe and Mg often form a complete solid solution in silicates, but in the lower mantle, they tend to be separated. Brg can only accommodate a maximum of 15 mole% Fe under the reducing DLM conditions, and the Fe-endmember silicate simply breaks down to simple oxides [13].
The key questions, therefore, are how much Mg can replace Fe to form (  relations of ferromagnesian silicates and oxides in the presence of water and found very distinct mineralogy under the DLM conditions. The polymorphic transition of olivine to wadsleyite at 410 km depth causes a seismic discontinuity that defines the top of the mantle transition zone (MTZ). With increasing depth (and pressure), wadsleyite transforms to ringwoodite and eventually decomposes to Fp plus Brg [2], causing the 660 km seismic discontinuity that defines the boundary between the MTZ and the lower mantle. Wadsleyite and ringwoodite can store 1-3 wt% water [14,15], which could make MTZ the largest water reservoir in the solid Earth. Schmandt et al. [16] explored the dehydration melting of hydrous peridotite at the top of the lower mantle, which may represent one of the key processes to transport water into the lower mantle [17][18][19]. Here, we further investigated the wadsleyite-water system down to the DLM, where dramatically different chemical behaviors lead to the creation of novel hydrous phases [20].

RESULTS
We performed high pressure-temperature (P-T) X-ray diffraction (XRD) experiments using laser-heated diamond anvil cells (DACs). A synthetic wadsleyite sample with composition (Mg 0.8 Fe 0.2 ) 2 SiO 4 (Fo80) containing 2 wt% water and an olivine sample with composition (Mg 0.7 Fe 0.3 ) 2 SiO 4 (Fo70) were used as starting materials and loaded with water into the sample chambers of two DACs. Fo80 was compressed to 134 GPa and heated to 2800 K to mimic conditions of the CMB. The sample assemblage was probed in situ at high P using synchrotron XRD during laser heating (Fig. 1). Fo80 decomposed into pPv-type (Mg, Fe)SiO 3 and a pyrite-structured oxide phase (Mg, Fe)O 2 H x (Py-phase), instead of the well-known decomposition product of Fp in the anhydrous case. The Py-phase was quenchable to room temperature (RT) at high pressure ( Fig. 1C and D). The Fo70 in H 2 O was studied at 98-133 GPa, and the Py-phase was also observed to coexist with (Mg,Fe)SiO 3 (Brg and/or pPv depending on the pressure) upon laser heating (Fig. 2). These results confirm that in the presence of H 2 O, the Py-phase would replace Fp to become an important constituent of the DLM. The dry reaction assemblage, is replaced by the wet mineral assemblage in the presence of water: We proceeded to investigate the P-T dependence of the Mg fraction X Mg = Mg/(Fe+Mg), and hydrogen content x in the wet assemblage. We note that the water in Eq. (4) mainly affects the silica-free oxides. The study can thus be simplified to the silica-free Fe-Mg-O-H system of the Fp to Py-phase conversion by substituting Eq. (3) into Eq. (4): Thus, we studied the effects of water on the highpressure chemistry of Fp. As the most abundant oxide in the mantle, Fp has been extensively investigated for its elastic, optical, thermal and elemental partitioning properties to understand the geophysical and geochemical behavior of the lower mantle (e.g. [21,22]). As a result of the very strong partitioning of Mg in Brg relative to Fp, we chose three starting Fp compositions with X Mg = 80% (Pe80), 70% (Pe70) and 60% (Pe60), which would be in equilibrium with Brg of X Mg = 96%, 92% and 88% [23,24], respectively, in representative pyrolite compositions with bulk X Mg of 90-80%. The Pe70 sample was compressed in H 2 O to 96 GPa and laserheated to 1460 K. Its XRD patterns clearly show coexistence of body-centered cubic ice-VII [17] and face-centered cubic Fp (Fig. 3A). Above 103 GPa and 1800 K, Eq. (5) (Fig. 3B). The Py-phase is quenchable to RT under pressure (Fig. 3C). Releasing pressure at RT, the Py-phase was preserved metastable down to 50 GPa, and is unstable at lower pressures (Supplementary Figs S1 and S2). A similar transition to the Py-phase was observed with Pe60 as the starting material (Fig. 4), but the transition pressure was ∼20 GPa lower than that of Pe70. The transition was not observed in the experiment with Pe80 up to 110 GPa and 2450 K, where Pe80 remained in the ferropericlase phase without the Py-phase.
We performed conclusive crystallographic analysis on the Py-phase using a multigrain algorithm [25,26].  (Fig. 5A). To identify the effects of different variables, we conducted an additional H-free high P-T experiment of (Mg 0.6 Fe 0.4 )O sample loaded in pure O 2 medium without any water. We successfully synthesized H-free Py-(Mg 0.6 Fe 0.4 )O 2 at 84 GPa, and after quenching to RT, determined its volume per f.u. to be 20.127(6)Å 3 (Fig. 5A). Based on the starting composition and reported volume [10] of Mg-free Py-FeO 2 as 20.304Å 3 , the volume dependence of H-free Py-(Mg,Fe)O 2 on X Mg at 84 GPa and RT is represented by the linear relationship: where 0 < X Mg < 1. The hydrogen content x can be estimated by comparing the molar volumes of hydrogen-bearing and hydrogen-free Py-phases of the same X Mg . Following the scheme [9] used for FeO 2 H x , hydrogen content x in (Mg,Fe)O 2 H x synthesized at 83.5 GPa is estimated to be between 0.6 and 0.8 (Fig. 5A). We further examined the cation composition of the recovered sample after complete decompression to ambient pressure. A recovered Py-(Mg 0.6 Fe 0.4 )O 2 H x sample was analyzed using an electron microprobe (Supplementary Fig. S3). Although the sample possibly lost hydrogen and excess oxygen at ambient conditions, the overall surface Mg to (Mg+Fe) ratio (X Mg ) remained uniform at 0.59 ± 0.04, indicating preservation of the cation composition during the conversion of Fp to Py-(Mg, Fe)O 2 H x .

IMPLICATIONS
Our high P-T experiments on the phase relations of Pe60, Pe70 and Pe80 are summarized in Fig. 5B, a projection of Fe-Mg-O-H quaternary phase diagram. The results clearly demonstrate that extensive solid solutions can be formed in Py-(Fe,Mg)O 2 H x , and the maximum Mg limit increases with increasing P. The results indicate that for a wet pyrolite with an average bulk composition of X Mg = 85%, its Fp (Pe70) will transform to the Py-phase at depths of 2100-2300 km depending upon the hot or cold geotherm temperatures [27]. For an iron-rich bulk pyrolite X Mg = 80%, its Fp (Pe60) will transform to the Py-phase at 1700-1900 km depth, while for an iron-poor bulk X Mg = 90%, its Fp (Pe80) may remain stable down to the CMB at 2891 km depth. Nevertheless, with essentially inexhaustible iron from the core, reaction (1) will kick in at the CMB to provide sufficient FeO 2 H x endmember to stabilize (Fe,Mg)O 2 H x .
As a result of the ubiquitous presence of Fp in the lower mantle, its transition to the Py-(Mg, Fe)O 2 H x carries great geological significance. Our experiments suggest that the mineralogical heterogeneity of the deep lower mantle depends upon wet or dry conditions (Fig. 6). How much water (carried by hydrous materials or as free solid or liquid H 2 O) is available in the DLM remains an open question [8,28]. This relies heavily on the primordial water reservoirs and distributions [29] in the DLM as well as water circulation from the MTZ [19,30]. Previous views of a dry lower mantle were largely caused by a lack of hydrous phases that could carry water under high P-T conditions beyond the MTZ. Since the discoveries of the dense hydrous phase-H [20] and δ-AlOOH [31], hydration of the lower mantle has become an increasingly more likely scenario, which has been further strengthened by a series of subsequent observations of high P-T hydrogen-bearing phases [11,[32][33][34]. In addition, the breakdown GPa after quenching back to ambient temperature with red hash marks noting the Py-phase. Uncertainties in pressure were estimated from measurement error and the thermal expansion of high P-T gold [41]. The X-ray wavelength was 0.3445Å. Temperature uncertainty is ±200 K in this experiment. Py, Py-phase.
of water-bearing wadsleyite and ringwoodite to anhydrous Brg and Fp does not prohibit the descent of water; the released H 2 O could be carried down as inclusions in minerals [17], trapped between grain boundaries or transferred to the newly discovered high P-T hydrogen-bearing phases [35].
The remarkable properties and geological implications of the iron endmember Py-FeO 2 H x have been studied extensively [5,11,32,36], and can be used as guidance for the Py-(Mg,Fe)O 2 H x solid solutions. The hydrogen released from the reaction (3) in the DLM can rise freely to sustain the waterhydrogen cycle [9]. The observed low seismic velocities of Py-FeO 2 H x have been used to interpret the seismologically observed ultralow velocity zones (ULVZs) [36]. In fact, the velocities of the iron endmember are too low and some other mantle materials must be added to match the ULVZ. Likewise, Py-(Mg,Fe)O 2 H x is expected to have low velocities, but not as low as that of the iron endmember because of the presence of Mg, thus becoming a better fit for matching ULVZs.

CONCLUSION
Both Py-(Mg,Fe)O 2 H x and Py-FeO 2 H x are nominally oxygen excess phases, which are defined as the total oxygen exceeding the balance of the cations based on the 'normal' maximum valence state assignment of Mg 2+ , Fe 3+ , H + and O 2− . These phases could create highly inhomogeneous and unusual redox environments in the DLM [37], collect into oxygen-rich reservoirs and potentially provide a huge internal source of oxygen [10] for modulating oxygen fluctuations in the atmosphere [38,39]. Our experiments indicate that the Py-type phases cannot retain all excess oxygen after releasing pressure, thus preventing direct recovery from natural samples. They may convert to low-pressure phases with the maximum valence state, and reveal the oxidation history. This provides an alternative explanation for the high ferric inclusions often observed in natural diamonds [37,40]. In conclusion, the presence of water in DLM will create a very different mineralogy enriched in Py-(Mg,Fe)O 2 H x that has a profound impact on the evolution and dynamics of the solid Earth.

Hydrous wadsleyite
Hydrous wadsleyite samples were synthesized with a Kawai-type 1500-ton multi-anvil apparatus at the Center for High Pressure Science and Technology Advanced Research (HPSTAR). The reaction precursors were mixed in the optimized molar ratio (0.4 FeO:1.6 Mg(OH) 2 :1.0 SiO 2 ) and then placed in an alloy capsule made of 75% Au and 25% Pd. The Fo80 samples were synthesized at 16 GPa and heated at 1400 • C for 4 hours. After samples were quenched and decompressed, the recovered samples were checked on the synchrotron beamline by XRD to be single phase wadsleyite (Fig. 1). The water content was measured from infrared-red spectroscopy and determined to be 2.2±0.3 wt% (details in Supplementary Fig. S4).

Olivine (Fo70)
Fo70 sample was synthesized at HPSTAR. The starting materials consisted of reagent grade MgO, Fe 2 O 3 and SiO 2 in the proper ratio. They were loaded into separate crucibles and placed in an oven at 1600 • C for more than 8 hours. After mixing well, synthesis was performed in a gas-mixing furnace for >2 hours at 1300 • C and oxygen fugacity buffer of Fayalite-Magnetite-Quartz -1.
Natl Sci Rev, 2021, Vol. 8, nwaa098 Ma ntle geo the rm S ub du ct in g sl ab ge ot he rm P e 7 0 + P y P e 6 0 + P y   Figure 6. Mineralogy at the lower mantle. The lower mantle is divided roughly in the middle of the lower mantle, below which is the DLM. While H 2 O or dense hydrous phases (blue droplets) are carried down by plate subduction or form from primitive water in shallower lower mantle, they react with Fp to form the Py-phases in the DLM (solid patches). Color gradient in the patches indicates the content of Py-phases. The mineral composition within the wet pockets of the DLM may include Brg, Fp as well as Py-(Mg, Fe)O 2 H x . Fe 2 O 3 powders were mixed according to the target mole ratio of Mg and Fe. The mixture was placed in a platinum crucible and kept in a furnace at 1300 • C for 24 hours for dehydration. A solid solution of cubic (Mg,Fe)O was synthesized from the mixture of oxides at 1 GPa and 1200 • C in a piston-cylinder press using a graphite capsule, which reduced ferric to ferrous iron. The composition and homogeneity of the synthesized samples were confirmed by electron probe microanalysis ( Supplementary Fig. S5).

In situ XRD experiments at high P-T conditions
Angular dispersive XRD experiments were performed at sectors 16ID-B of HPCAT and 13ID-D of GSECARS at the APS, ANL. The x-ray energy was 30.5 keV for the samples of Pe70 in H 2 O, Pe60 in O 2 , 36.0 keV for Pe60 in water and 37.0 keV for Fo70 in water. Samples were pre-compressed in a pair of 1 mm diamond anvils and sample platelets of ∼30 (W) × 30 (L) × 10 (T) μm 3 were loaded into DACs. High pressure was achieved using beveled diamond anvils with 300 μm outer diameter and 150 or 100 μm inner diameter culet. The sample chamber was a hole of one-third of the inner culet diameter, which was drilled in a tungsten or rhenium gasket. For the experiments with H 2 O, a droplet of deionized water was injected into the chamber and then was immediately sealed. For Fp+O 2 experiment, liquefied O 2 was piped into the sample chamber in a liquid-nitrogen cooled container.
One to two pieces of thin gold foil were placed around the sample as pressure calibrant. Combining the uncertainties from indexing (±0.3 GPa) and calibrant (±1.0 GPa), the total pressure uncertainties propagated to ±1.1 GPa. During in situ laser heating, thermal pressure was corrected for thermal effects using a P-V-T equation of states of gold [41] and uncertainty was up to ±6 GPa.
The diameter of laser spot at GSECARS was around 20 μm and flat down to ∼10 μm. At the time of experiment, the incident X-ray beam size was 3 × 4 μm 2 . We estimated temperature uncertainties during laser heating were up to ±200 K.
At HPCAT, the heating temperature was measured by fitting the gray-body radiation curve on both sides of the sample. The diameter of laser spot was approximately 30 μm at 2000 K, which covered the entire sample. We estimated an uncertainty of up to ±200 K on temperature measurements throughout the laser heating experiments [42,43].