Deep lower-mantle water reservoir implied by the stability of a mixed-valence hydrous iron-rich oxide

The lower mantle, containing both primordial and recycled water, is the most massive potential water reservoir in the Earth. Geophysical and geochemical evidence combined have suggested that the largest heterogeneities in the deep lower mantle may serve as primitive deep-mantle reservoirs hosting a variety of incompatible species including hydrogen. To understand water storage in the deep lower mantle, we conducted experiments in the Fe-O-H, Fe-Al-O-H and Fe-Al-Mg-Si-O-H systems under high pressure-temperature (P-T) conditions, and discovered a previously unknown hexagonal phase (referred to as “H1-phase”) in all the systems. The single-crystal structure of the H1-phase was determined at 79 GPa with a unit-cell of a=10.022(2 )Å and c=2.6121(9) Å and the space group of P63/m, and its composition was obtained as Fe12.76O18H3.7 combining the structure determination and chemical analysis on the recovered sample. More importantly, about 20 mol% of MgO, Al2O3 and SiO2 can be incorporated into the H1-phase in a realistic mantle system Fe-Al-Mg-Si-O-H and its stability eld is extended to at least 2400 km along a normal geotherm, implying that the H1-phase can store primordial water in the deepest lower mantle. Therefore, plume-generation zones originated from the deepest lower mantle provide a potential source for higher water contents in basalts associated with mantle plume components.


Introduction
The lower mantle, containing both primordial and recycled water 1 , is the most massive potential water reservoir in the Earth. Geophysical and geochemical evidence combined have suggested that the largest heterogeneities in the deep lower mantle may serve as primitive deep-mantle reservoirs hosting a variety of incompatible species including hydrogen 2,3 . To understand water storage in the deep lower mantle, we conducted experiments in the Fe-O-H, Fe-Al-O-H and Fe-Al-Mg-Si-O-H systems under high pressuretemperature (P-T) conditions, and discovered a previously unknown hexagonal phase (referred to as "H1phase") in all the systems. The single-crystal structure of the H1-phase was determined at 79 GPa with a unit-cell of a=10.022(2 )Å and c=2.6121 (9) Å and the space group of P63/m, and its composition was obtained as Fe 12.76 O 18 H 3.7 combining the structure determination and chemical analysis on the recovered sample. More importantly, about 20 mol% of MgO, Al 2 O 3 and SiO 2 can be incorporated into the H1-phase in a realistic mantle system Fe-Al-Mg-Si-O-H and its stability eld is extended to at least 2400 km along a normal geotherm, implying that the H1-phase can store primordial water in the deepest lower mantle. Therefore, plume-generation zones originated from the deepest lower mantle provide a potential source for higher water contents in basalts associated with mantle plume components 1,4 .
Global tomography revealed the Paci c and African superplumes in the lower 1000 km of the mantle 5, 6 .
Combined evidence from geophysics, geochemical and dynamic calculations have suggested that the superplumes are dense and have a chemical origin 7,8,9 . The origin of the chemical heterogeneities is still an area of active debate. Formation of a deep magma body early in the Earth's history would be expected to concentrate a variety of incompatible species including hydrogen in the deepest heterogeneities 2, 4 .
The sites of both large igneous province (LIP) eruption and active hotspot volcanoes lie vertically above the edges of the African and Paci c superplumes 10 201, implying a possible lower mantle source for the primordial noble gases and higher water contents in basalts associated with mantle plume components 1,3,11 . Characterization of the volatile contents of different mantle components indicates that recycled oceanic crust is e ciently dehydrated during subduction and therefore water in ocean island basalts is a combination of primordial and recycled water 1 . Recent calculations suggested that water would preferentially partition into the iron liquid at the core mantle boundary 12 . Despite large uncertainties in the budget and distribution of water in the Earth, water, even in extremely small amounts, can signi cantly affect mantle phase equilibrium and play a critical role in mantle dynamics 13 .
Phase equilibrium data on water-bearing multicomponent systems under the extreme high pressuretemperature (P-T) conditions can provide constraints on water storge and effects of water on the phase relations in a lower mantle system. The water solubility in bridgmanite, the dominant mineral in the lower mantle, remains a debated issue 14,15  , may transport water into the lower mantle through subduction 16,17 . Experimental data on the Al-bearing multicomponent systems have showed that the thermal stability of the phase H is greatly enhanced by incorporation of alumina under P-T conditions throughout the lower mantle 17,18,19,20 . The phase H, δ-AlOOH, ε-FeOOH and CaCl 2 -type SiO 2 all adopt the similar orthorhombic structure under the P-T conditions of the lower mantle except that ε-FeOOH transforms to a cubic pyrite-structured FeOOH x (0<x≤1) (referred to as "py-phase") above 85 GPa 21,22  Iron (Fe) is the most abundant transition metal in the mantle and the dominant element in the core. Iron reacts with water to form wüstite (FeO) and iron hydride (FeH) when iron is over saturated in the system 24,25 . The run products would be different depending on water availability in a Fe-O-H system 25  There have been few phase equilibrium data in hydrous multicomponent systems mainly due to the di culties in characterization of coexisting phases in a multiphase assemblage synthesized at extremely high P-T conditions. Combining in-situ multigrain X-ray diffraction (XRD) with ex-situ transmission electron microscopy (TEM) analysis, we are able to determine chemical composition and crystal structure of 100-300 nm individual grains in a multiphase assemblage synthesized at high P-T conditions 26,27 . In this study, we conducted  Table 1), identi ed a new hexagonal hydrous Fe-rich phase with a mixed valence state, and discussed the water effects on the chemical heterogeneities in the deep lower mantle.
Discovery and characterization of the H1-phase First, the starting material Al 0.2 Fe 0.8 (OH) 3 was loaded in a Ne medium (Run# Sa083, Supplementary Table 1). When the sample was heated to 1600 K at 79 GPa, only the diffraction peaks of the orthorhombic δ-phase appeared. When T was increased to 1800 K, a new set of peaks appeared in addition to those of the δ-phase. We observed a gradual growth of the new phase at the expense of the δphase and it became dominant at 2100 K, with a hexagonal unit-cell of a=10.022(2) Å and c=2.6121(9) Å at 79 GPa and after T quench (referred to as the "H1-phase"), as shown in Fig. 1. The results indicate that the Fe-bearing δ-phase (Fe,Al)OOH is unstable under high P-T conditions of the deep lower mantle along a normal geotherm 28 .
The H1-phase was reproduced in two separate runs coexisting with the δ-phase using the same starting material Al 0.2 Fe 0.8 (OH) 3 (Run# Sb211, Sb343a, Supplementary Table 1 and Fig. 1). To identify the chemical composition of the H1-phase, the sample synthesized at 80 GPa and 2100 K (Run Sb #343a) was recovered to ambient conditions for phase identi cation and chemical analysis by TEM. We found that the Al content in the H1-phase was negligible containing only ~1.5 at.% Al, but the Fe/(Fe+Al) ratio was reduced from 80 at.% in the starting material to ~13 at.% in the δ-phase ( Supplementary Fig. 2). As shown by the contrast between the coexisting H1-phase and δ-phase (Fe,Al)OOH in the elemental mapping of oxygen, the H1-phase contains slightly less oxygen content than (Fe,Al)OOH. However, the exact oxygen content in the H1-phase cannot be determined by chemical analysis on the recovered sample due to the presence of residual water in the system. When the P-T conditions were increased to 86 GPa and 2300 K, we found that the H1-phase coexisted with the δ-phase and the py-phase (Run# Sb069), indicating that the H1-phase is a low-pressure phase relative to the py-phase.
The H1-phase was again successfully synthesized in the Fe-O-H system with the starting material Fe(OH) 3 (equivalent to Fe 2 O 3 +3H 2 O) in a Ne medium at 78 GPa and 2000 K (Run# Sb343b). A nearly pure H1-phase was again obtained with a=10.014(2) Å and c=2.6158(6) Å coexisting with a small amount of ε-FeOOH. The sample was further heated at 2200 K for another 10 minutes to test its thermal stability and the H1-phase remained stable only with better crystallization (Fig. 2A).
The Earth's mantle contains both ferric and ferrous iron. For comparison, we conducted another experiment on FeO loaded in a saturated water medium. The sample was compressed to 80 GPa and then heated at 2100 K for 10 minutes (Run# Sb390) and the H1-phase was again obtained coexisting with ice-VII (Fig. 2B), indicating that formation of the H1-phase is independent of the valence state of iron in a hydrous deep lower mantle. The H1-phase was slowly decompressed and the XRD pattern was  Fig. 3).
Determination of the chemical composition and crystal structure of the H1-phase is a challenge due to its unquenchable crystal structure. Fortunately, spottiness of the XRD pattern of the H1-phase in the Fe-Al-O-H system (Run# Sa083) allows in-situ multigrain indexation and single-crystal structure determination 27,29 . The structural information can be used to constrain its chemical formula because the H1-phase contains mainly Fe and O where the Al content (~1.5 at.%) is negligible. Hydrogen is undetectable by XRD if present. The sample was aligned to the rotation center and a dataset was collected by rotating the DAC from -26.0 to 25.0° in small incremental steps of 0.25°. The X-ray wavelength was 0.3445 Å and the exposure time was 5s/frame. Through the determination of crystallographic orientations for individual grains 30 , we indexed 28 individual grains belonging to the H1-phase, and three of the grains were selected for further structure determination and re nement. The XDS package was used for single-crystal data reduction 31 . In total 774 re ections have been merged from the three grains. A reasonable R1 = 5.59% was obtained for all data. The re ections obtained from the powder XRD and the re ned atomic coordinates determined from the single-crystal structure are given in Table 1  where Fe(OH) 3 was used as the starting material (Run# Sb343b) and the run product was a nearly pure H1-phase ( Supplementary Fig. 4 and Fig. 2A) Fig. 3A, the diffraction peaks from the multiple phases overlap heavily in the powder XRD pattern where several characteristic peaks of the H1-phase were visible but weak in intensity. To identify the H1-phase and other coexisting phases in the multiphase assemblage, we applied the multigrain method 27 and successfully indexed multiple individual grains belonging to the H1-phase, post-perovskite (pPv) structured Fe 2 O 3 phase, bridgmanite (bdg), and the δ-phase, respectively. Supplementary Table 2 Table 2).
In another experiment on 65.6 mol% SiO 2 -12.3 mol% Al 2 O 3 -16.4 mol% MgO-5.7 mol% Fe 2 O 3 (MASFH6, Run# Sb307b) that is similar to a simpli ed MORB composition, the H1-phase was again obtained at 96 GPa and 2300 K containing about 30 at.% (Mg+Si+Al) (Fig. 3B and Table 2). In contrast to ~1.5 at.% Al in the H1-phase at ~80 GPa, the H1-phase contains ~11.2 at.% Al at 96 GPa ( Table 2), implying that the composition of the H1-phase is highly dependent on pressure and possibly mantle compositions. The results suggest that incorporation of Al 2 O 3 , MgO and SiO 2 into the H1-phase extends its stability eld from 86 GPa in Fe-Al-O-H system to at least 96 GPa (~108 GPa after accounting for the thermal pressure) at 2300 K in SiO 2 -bearing lower mantle system, corresponding to about 2400 km depth in the deep lower mantle. The H1-phase (Mg0.63Si1.61Al1.43Fe9.10)O18H3.7 has an estimated density of 6.862 g/cm 3 at 2400 km depth, relative to 5.32 g/cm 3 of the Preliminary Reference Earth Model (PREM) density 38 .

Geophysical and geochemical implications
The preliminary results of this study con rmed the stability of the H1-phase in realistic multicomponent systems over a P-T range of 80-108 GPa and 2000-2300 K (after accounting for the thermal pressure).
Discoveries from mineral physics support a physicochemical boundary at the depth of 1600-1800 km 39 where a series of phase transitions have been discovered in Fe-bearing minerals 21,40,41 , consistent with a chemical boundary at the middle of the lower mantle 5,7,8,42,43 . The H1-phase is another Fe-bearing phase stable under P-T conditions of this region and its hydrous nature adds further complexities to the chemical heterogeneities. The stability and chemical composition of the H1-phase (Mg,Si,Al,Fe) 12.76 O 18 H 3.7 implies that it may store primordial water depending on actual chemical composition and water concentration in the deep lower mantle. Further detailed exploration in the Mg-Si-Fe-Al-O-H systems will be needed to determine the relationship between water storage minerals and chemical composition including its water concentration under high P-T conditions of the deep lower mantle.
The mixed valence state in the hydrous H1-phase has important implications for the geophysical and geochemical properties of the deep lower mantle. Addition of water and escape of H 2 to the atmosphere has been regarded as an important oxidation mechanism in the mantle 44,45,46 and ferric iron can be produced by water-induced oxidation of ferrous iron at shallow depths 44 . Under the deep lower mantle conditions, however, a different mechanism is indicated in this study when the same mixed-valence hydrous iron oxide H1-phase can be produced from chemical reaction between either FeO or Fe 2 O 3 with water, implying that high pressure and high temperature in the deep lower mantle stabilizes the crystalchemistry of the H1-phase. Geophysical and geochemical evidence combined have suggested that the superplumes located at the base of the lower mantle may serve as primitive deep-mantle reservoirs hosting a variety of incompatible species including hydrogen and have been regarded as the largest heterogeneities in the deep mantle 2, 3, 43 . The edges of these heterogeneities seem to have controlled mantle plumes that have generated LIPs and major hotspot volcanoes 4,10 . Therefore, plume-generation zones in the deepest lower mantle provide a potential source for high water contents in basalts associated with mantle plume components 1, 47 .

Methods
The starting materials Al0.2Fe0.8(OH)3, Al0.8Fe0.2(OH)3, Fe(OH)3 , FeO+H 2 O and MgO-Fe2O3-SiO2-Al2O3-H2O gels were used in this study and the details of preparation and characterization of the sample were described elsewhere 23 . Composition and homogeneity of the starting materials were con rmed by a scanning electron microscope (SEM) coupled with an energy dispersive X-ray spectroscopy (EDS) analysis ( Supplementary Fig. 6). LH-DAC experiments were performed between 70-108 GPa and 1600 K-2400 K. In most runs, Ne was used as insulation layer and pressure medium to generate quasi-hydrostatic pressure, and its equation of state was used to calibrate pressure 48 . All the sample pressures were measured after T quench unless noted otherwise. The thermal pressures are estimated as P th =(T-300)*0.0062 (GPa) from a previous study under similar P-T conditions 23 . In the Runs Sb390 and Sb347, water was used as pressure medium and pressure was calibrated by the Raman shift of diamond 49 .
The in-situ laser-heating coupled with XRD measurements were conducted at High Pressure Collaborative Access Team (HPCAT), 16-ID-B beamline of Advanced Photon Source (APS), Argonne National Laboratory (Argonne, IL) (Runs# Sa083). XRD experiments with ex-situ laser-heating were conducted at 15UI beamline of Shanghai Synchrotron Radiation Facility (SSRF) (Runs# Sb211, Sa069, Sb343a, Sb343b, Sb307a, Sb390) and the P02.2 beamline of the PETRA III synchrotron at Deutsches Elektronen Synchrotron (DESY) (Run# Sb335, Sb307b and Sb347). The thin-sections of the recovered samples were prepared using a focused ion beam (FIB). Chemical analysis of recovered samples was performed using a TEM operating at 200 kV which is equipped with a EDS system. Further details about the experiments and data analysis can be found in the supplementary information (Note 1-5). Declarations