Altered chemistry of oxygen and iron under deep Earth conditions

A drastically altered chemistry was recently discovered in the Fe-O-H system under deep Earth conditions, involving the formation of iron superoxide (FeO2Hx with x = 0 to 1), but the puzzling crystal chemistry of this system at high pressures is largely unknown. Here we present evidence that despite the high O/Fe ratio in FeO2Hx, iron remains in the ferrous, spin-paired and non-magnetic state at 60–133 GPa, while the presence of hydrogen has minimal effects on the valence of iron. The reduced iron is accompanied by oxidized oxygen due to oxygen-oxygen interactions. The valence of oxygen is not –2 as in all other major mantle minerals, instead it varies around –1. This result indicates that like iron, oxygen may have multiple valence states in our planet’s interior. Our study suggests a possible change in the chemical paradigm of how oxygen, iron, and hydrogen behave under deep Earth conditions.

O xygen and iron are Earth's most abundant elements by number of atoms and by mass, respectively. They form compounds dictating major chemistry of our planet 1 . It is conventionally accepted that the oxygen anion has an unvarying −2 valence state in mantle ferropericlase and bridgmanite throughout the deep interior, where the oxygen fugacity decreases with increasing depth. The redox states are mostly controlled by the 3d transition element Fe which could vary among three valence states, metallic Fe 0 , ferrous Fe 2+ , and ferric Fe 3+ . Recently, a series of new iron oxides have been found with varying O/Fe stoichiometry ranging from the endmember Fe 3+ 2 O 3 on our planet's highly oxidized surface to the other end-member Fe 2+ O which should be stable at the highly reduced conditions in the deep lower mantle down to the coremantle boundary, which include Fe 5 O 7 , Fe 4 O 5 , and Fe 5 O 6 (refs. [2][3][4]. High pressures in the deep lower mantle would promote the crystal field splitting of 3d orbitals of iron and cause the electronic spin-pairing transition, which can affect the physical, chemical, and transport properties of mantle phases [5][6][7] . The conventional wisdom, however, is facing a change in light of the recent discovery of the high-pressure pyrite-structured iron superoxide FeO 2 which has O/Fe ratio even higher than Fe 2 O 3 and can hold a varying amount of hydrogen (denoted as "Py-FeO 2 Hx" with x from 0 to 1) [8][9][10][11][12][13][14][15] . With the subducted plate carrying down water to react with the iron core to form Py-FeO 2 Hx and release hydrogen, oxygen-rich reservoirs could be accumulated in the very reducing core-mantle boundary region [13][14][15] . Such reservoirs at the mid-point (2900 km depth) of the Earth's radius (6370 km) will certainly play a pivotal role in the global chemistry, including the generation of our present day aerobic atomosphere 9,16 . A number of key solid-state chemistry questions on Py-FeO 2 Hx must be understood: What are the valence states of Fe and O? What is the nature of their chemical bonding? What are the effects of hydrogen on the valence and bonding of Fe and O? Is the iron superoxide metallic? Are its delectrons spin-paired?
Assuming oxygen in the common O 2− state as in other iron oxides, the valence state of iron would be ferryl (Fe 4+ ) for FeO 2 . On the other hand based on the analogy to FeS 2 pyrite, in which Fe remains ferrous (Fe 2+ ) with the sulfur forming S 2 2− dimer, would FeO 2 also consist of Fe 2+ cations and O 2 2− dimer? Streltsov et al. 17 predicted the valence state of iron to be +3 for the hydrogen-free FeO 2 Py-phase where oxygen ions do not form O 2 2− dimers, suggesting FeO 2 "in between" the usual dioxides and peroxides, based on the density functional and dynamical mean-field theories (DFT + DMFT). By contrast, Jang et al. 18 employed DFT + DMFT approaches with treating FeO 2 to presumably have O 2 2− dimer as a complete analogue of pyritestructured FeS 2 . Thus far, theoretical calculations cannot conclude the nature of chemical bonding and state of FeO 2 without experimental constraints. Answers must be based on in situ measurements of the micron-sized Py-FeO 2 Hx sample which cannot be preserved after releasing pressure for normal electrochemical analysis.
We take the challenge by combining a variety of recently enabled synchrotron X-ray spectroscopic techniques 19 , and obtain intriguing answers from direct high-pressure experimentation supplemented by first-principles simulations with the self-consistent linear-response approach. Here we find many unexpected chemical behaviors in Py-FeO 2 Hx samples (Supplementary Figures 1-5), where iron is in the reduced, spin-paired ferrous state and the valence state of oxygen varies around −1, instead of at −2 in common iron oxides, while hydrogen is weakly bonded in the structure. Our results suggest that other than iron, oxygen can occupy multiple valence states in our planet's interior, leading to a chemical paradigm change in the deep Earth.

Results
Valence states of iron and oxygen in FeO 2 Hx. The valence state of Fe can be characterized by near-edge X-ray absorption spectroscopy (XAS) features at energy slightly below the iron K-edge 20 . Because of the strong absorption of 7 keV X-ray by diamond anvils, the XAS spectra were collected in the partial fluorescence yield (PFY) geometry for the iron superoxide at 53-133 GPa. That is, the incident X-ray beam passes through one diamond anvil, and then the X-ray fluorescence exits in the radial direction through the Be gasket to avoid double diamond absorption. The peak position and intensity distribution are diagnostic probes for valence state, spin state, geometry, and the ligand-ligand interactions for Fe atoms 20 . The iron K pre-edge feature that is related to the 1s → 3d electronic transition, is particularly sensitive to the valence state of iron 21 . Our XAS-PFY measurements of other representative iron compounds at ambient conditions explicitly show that the position of iron K pre-edge absorption increases with increasing valence state (Fig. 1a top). The position of iron K pre-edge of the iron superoxide FeO 2 is very close, but slightly below, that of Fe 2+ compounds. That is, the valence state of iron in FeO 2 is plausibly close to +2. Its position remains almost constant from 53 to 81 GPa, suggesting a minimal pressure dependence on the valence state of iron 22 , consistent with mantle ferropericlase and bridgmanite. Furthermore, it is intriguing that the iron valence state of the hydrogenfree iron superoxide is similar to that of the hydrogen-bearing Py-FeO 2 Hx with x = 0.5-0.7 at 133 GPa. These observations indicate that the hydrogen content has minimal effects on the valence state of iron in the superoxide (Fig. 1a).
With the valence state of iron slightly under +2, the valence state of oxygen for the Py-FeO 2 would have to be nearly −1, suggesting that oxygen atoms must interact with each other in the iron superoxide. Such interaction can be observed experimentally from the oxygen K-edge spectroscopy. We note that the edge at 540 eV is in a very soft X-ray region inaccessible to XAS in DAC experiments. Instead, we can access the oxygen K-edge by X-ray Raman scattering (XRS) using hard X-ray in and hard X-ray out method, where the energy loss is measured due to inelastic Raman shift 23,24 . For XRS measurements on the oxygen K-edge, we deliberately used LiF as a pressure transmitting-medium, instead of water or O 2 , to assure XRS signals of oxygen only from the Py-FeO 2 Hx (Supplementary Figure 3). In the XRS at 110 GPa, we observed a sharp oxygen π* peak near 531 eV and a broad σ u * peak at 540-550 eV (Fig. 1b). These peaks represent the O electron transition from 1s core orbitals to the antibonding 1π g * and 3σ u * orbitals, respectively (termed as π* and σ* transitions). The strong π* transition indicates the presence of the O-O dimer.
Magnetic and electronic spin states of iron in FeO 2 Hx. Magnetic ordering and spin state of Fe in iron superoxide can be investigated using synchrotron Mössbauer scattering (SMS) and X-ray emission spectroscopy (XES), respectively. 57 Fe is a Mössbauer active isotope with nuclear resonant energy of 14.41 keV. A series of SMS spectra of the iron superoxide were collected upon decompression in steps from 118 GPa down to 32 GPa (Fig. 2a), and the experiment was repeated in three separated runs ( Supplementary Figures 6-8) of FeO 2 Hx with different x (Supplementary Table 1). SMS spectra were analyzed using CONUSS program 25 to extract the quadrupole splitting (QS) and isomer shift (IS) of iron in the Py-phase which is compared with that of standard samples of ferrous pyrite FeS 2 , ferrous Fe 0.48 Mg 0.52 O, and ferric Fe 2 O 3 (Fig. 2b). The QS of iron in the Py-phase barely increased with increasing pressure, whereas the IS displayed an opposite trend. The pressure dependence of QS and IS for iron in the superoxide is comparable to that for ferrous iron (Fe 2+ ) in ferropericlase (Mg, Fe)O in the low-spin state (refs. 26,27 ), likely due to the ferrous nature of the iron superoxide. It is noted that the spin transition could largely affect the QS of iron in (Mg, Fe) O (refs. 27,28 ). Furthermore, the values of the QS and IS of iron in the superoxide extrapolated to ambient conditions are about 0.27 and 0.55 mm s −1 , in excellent agreement with that in pyrite FeS 2 (ref. 29 ).
Our SMS spectra in Fig. 2a consistently showed that the Pyphase remained in the non-magnetic state represented by the clean, single beat spectra, regardless of the content of hydrogen in the lattice. SMS data with multiple time beats indicating magnetic ordering appeared only below 40-50 GPa when the iron superoxide decomposed into a magnetic phase under decompression. We note that the full width at half maximum (FWHM) for the diffraction peaks for the iron superoxide broadened significantly below 65 GPa and the peak intensity continually decreased until they finally disappeared around 40 GPa upon decompression at room temperature. Furthermore, the QS and IS values of iron in the magnetic phase are close to that in hematite (Fe 2 O 3 ) at 32 GPa. It is consistent with our recent X-ray diffraction (XRD) observations where the new diffraction peaks can be indexed to hematite below 40 GPa.
The magnetic spin states of 3d electrons are also directly probed by XES spectra of the Fe K β fluorescence lines 5,30 of the iron superoxide between 40 and 133 GPa (Fig. 3). The intensity of the K ′ β satellite peak of iron in the superoxide decreased between 45 and 60 GPa and disappeared completely at 60-65 GPa where the total spin momentum (S) of iron in the Py-phase approached zero, indicating that iron in the Py-phase undergoes a gradual electronic spin-pairing transition below 65 GPa from high-spin to low-spin states (Supplementary Figure 9). We note that the XES spectra were collected with decreasing pressure and thus the spin transition pressure may be higher due to pressure hysteresis on the spin crossover upon decompression 31 . These observations also suggest that the iron superoxide is not stable in the high-spin state and would decompose into oxygen and hematite across the electronic spin transition of iron, associated with the valence change in oxygen and iron at the same time.
Oxygen bonding and charge transfer in FeO 2 and FeO 2 H. For pyrite-type structures, the covalent nature is mainly controlled by the anion-anion distance (D O ). We followed previous strategies 32, 33 and used the fractional coordination y of oxygen to compare O-O distance at different pressures: where a is the lattice parameter as a function of pressure. The archetypical pyrite FeS 2 with y = 0.385 forms solid anion dimers with divalent Fe cations at ambient conditions. In contrast, conventional dioxides with tetravalent cations (e.g. RuO 2 and OsO 2 ) have a smaller value of 0.34-0.35 for y in the pyrite-type structure 33 . Our previous experimental XRD data showed y = 0.371 with D 0 = 1.937 Å in the iron superoxide at 75 GPa which lies between conventional dioxides and peroxides 9,33 (Table 1). It is known that Bader analysis may underestimate the charge transfer between anions and cations 36 . Our calculations may not indicate the exact valence values, but clearly show charge transfer between anions, which has also been seen in the calculations of archetypal pyrite FeS 2 (ref. 37 ). In hydrogen-bearing Py-FeO 2 H, hydrogen is equally bonded with two nearest oxygen atoms 38 . As a result, hydrogen is calculated to be +0.64 at 100 GPa. Oxygen atoms adopt electrons from hydrogen and become −1.13 while the valence state of iron remains around +1.63 for the Py-FeO 2 H. Therefore, hydrogen only donates electrons to oxygen atoms and has a minimal effect on the valence state of iron.

Discussion
Looking at individual properties, FeO 2 and FeO 2 Hx may appear similar to pyrite FeS 2 in their crystal structure, cation and anion valences, and spin-pairing state. Assimilating the comprehensive experimental and theoretical investigations, however, points to fascinating changes of high-pressure crystal chemistry above 70 GPa which corresponds to roughly the pressures beyond 1700 km depth, which is the midpoint separating the Earth's mass into two equal halves. At higher pressures, the valence of oxygen in oxides is no longer fixed to −2, but becomes variable. The valence of iron becomes more dependent on the crystal structure and pressure rather than the oxygen stoichiometry. Hydrogen is no longer bonded to oxygen as OH, but becomes mobile. FeO 2 and FeO 2 Hx do not exactly belong to the pyrite FeS 2 type structure with interactive anion dimer, nor the PdF 2 type structure without interactive anion dimer, but represent a new structure type in the Pa 3 MA 2 family with   Fig. 3 High-pressure XES spectra of pyrite-structured FeO 2 Hx at 300 K. The integrated intensity of the XES spectra was normalized to unity. Inset: the K ′ β satellite emission peak between 7030 and 7048 eV after the iron main emission peaks was aligned to that at 133 GPa. The changes of the satellite peak intensity in FeO 2 Hx are attributed to the high-spin to low-spin transition, as the disappearance of the satellite peak has been used as a robust criterion for the electronic spin-pairing transition of iron in other iron-bearing compounds 30 partially interactive anion dimer. These issues have been previously noted as separate enigmas in other studies of O, Fe, and H under high pressures, such as formation of O 8 molecules 39 , the redox paradox of ferric iron 40 , and mobility of hydrogen in ice 41,42 . With the integrated study here, now the overall picture clearly indicates a chemical paradigm change under deep Earth conditions.
At zero pressure without external constraint, the O-O interaction type can be empirically determined by their distance 33 24,39,43 ), the reduced valence of oxygen in FeO 2 and FeO 2 Hx, and the formation of new type of Pa 3 peroxide are a natural consequence of high-pressure chemistry.
At zero pressure, the valence of iron or ferric/ferrous ratio is generally correlated to the oxygen fugacity. The correlation is weakened by the additional control by the crystal structure of high-pressure phases. For instance, the paradoxical correlation of lower oxygen fugacity with higher Fe 3+ abundance 40 reflects the ability of the high-pressure perovskite structure to accommodate more Fe 3+ iron 44 . Here the new high-pressure iron superoxide structure is able to keep its iron in ferrous state by reducing the oxygen valence, and fits well in the reducing environment of the core-mantle boundary. The spin-paired non-magnetic Fe 2+ also fits the high-pressure chemistry of the deep lower mantle 5,30 .
At zero pressure, hydrogen is present as H + and replaces other cations without affecting the anion valence. Here, the addition of hydrogen has negligible effects on the chemical and magnetic properties of the iron in FeO 2 Hx, but mainly affects the oxygen valence. This shows the change of hydrogen from strong OH covalent bond at low pressures to weak interaction in the Py-FeO 2 Hx lattice. Hydrogen chemistry in H 2 O is known to have a very important change at high pressures to become symmetrically hydrogen bonded at low temperature 42 and superionic at high temperature 41,45 ; both result in weakening the OH bonding. The present observation indicates the weakened interaction of hydrogen is not limited to H 2 O but is a general high-pressure chemical trend that has many important consequences, including loss of hydrogen in the formation of Py-FeO 2 Hx (ref. 38 ), spontaneous splitting of H 2 O into hydrogen and oxygen 13 , and separation of water and hydrogen cycles in the deep Earth 10 .
Our study demonstrates that under extreme pressures of the deep mantle, the structure type becomes a main controlling factor of valence states, and the conventional sense of valence of elements could be altered to fit the structure. The new type of structure has the unexpected valence state of oxygen that is different from the O 2− state in primary mantle oxides and silicates, and the ferrous, spin-paired, and non-magnetic state of iron that would affect Mg-Fe partitioning between mantle phases 46,47 . We may expect the new iron superoxide in pyrite structure to accommodate and form solid solutions with other divalent and trivalent cations such as Mg 2+ and Al 3+ (ref. 48 ) and with anions, such as sulfur and halogens. We may further expect that the Pa 3 peroxide structure is only one example of the possible compositions in the deep Earth, and that additional structure types with different valence and spin, such as the recently discovered hexagonal phase 48 , will emerge.

Methods
Sample synthesis and characterization. The five samples of the pyrite-type phase have been synthesized and further probed by X-ray spectroscopic techniques in this study (Supplementary Table 1). Iron superoxide samples of FeO 2 or FeO 2 Hx were synthesized from goethite (FeOOH) or 57 Fe-enriched hematite (Fe 2 O 3 ) mixture with O 2 and water H 2 O, respectively, at 1800-2200 K under target pressures in laser-heated DACs at 16IDB, 13ID-D and High Pressure Synergetic Consortium (HPSynC) of the Advanced Photon Source of Argonne National Laboratory. In order to avoid any signal contamination from other phases in the sample and to assure complete transition, the whole starting materials of goethite (FeOOH) or hematite (Fe 2 O 3 ) have been laser-heated repeatedly for 2-3 h at target pressures to synthesize a pure pyrite-structured FeO 2 Hx (x = 0-1) phase.
XRD patterns were collected with the mapping steps of 4 μm for the wholesynthesized samples at beamlines 13ID-D or 16ID-B of Advanced Photon Source (APS), Argonne National Laboratory (ANL), confirming that the whole sample is the pyrite-type phase without any detectable amount of starting materials goethite or hematite left in the DACs (Supplementary Figures 1-5). The hydrogen content (x) of the iron superoxide samples that were synthesized from FeOOH or  with H 2 O was estimated to be 0.5-0.8 based on the unit-cell parameter calibration 10 (Supplementary Table 1). In particular, a relatively small size of starting materials were selected to be 25-30 μm in diameter with a thickness of 6-8 μm, which can reduce the synthesis time. We note that the X-ray beam was focused down to 5-7 μm in the full-width at half-maximum (FWHM) at the sample position. Thus the sample center was deliberately probed in order to achieve a relatively high efficiency of the spectroscopic measurements. That is, about one fifth of the volume of the synthesized samples was probed in the spectroscopic measurements.
Synchrotron X-ray diffraction experiments. XRD experiments were carried out at beamlines 13-IDD and 16-IDB of the Advanced Phonon Source, Argonne National Laboratory. A highly monochromatized incident X-ray beam was used with an energy of 30.49 keV (0.3738 Å), 33.17 keV (0.3738 Å), or 37.08 keV (0.3344 Å). The incident X-ray beam reached the sample position with a beam size of 2-5 μm in FWHM. For laser-heating XRD, two infrared laser beams were focused down to~15 μm and 25-30 μm in the flat top area of the heating profile on both sides of the sample at beamlines 13-IDD and 16-IDB, respectively. With the help of the X-ray induced luminescence on the sample and/or ruby, both laser beams were co-axially aligned with the incident X-ray beam. The temperature of the samples was calculated through fitting the measured thermal radiation spectra with the gray-body approximation 49,50 .
The temperature uncertainty is 100-200 K. A tiny piece of gold was placed in the sample chamber and its lattice parameters of gold were applied to determine pressure and uncertainty 51 . X-ray Raman scattering experiments. XRS measurements on oxygen K-edge spectra of the Py-phase were conducted at 110 GPa by using the XRS technique at beamline 16ID-D, APS, ANL. With the new 320 × 400 mm IDT mirror, the incident monochromatic X-ray beam were focused down to 5-7 µm in FWHM at the sample position. With the sample in a diamond-anvil cell mounted on the rotation center of the XRS spectrometer, oxygen K-edge spectra were collected by scanning the incident beam energy from 518 to 578 eV above the analyzer elastic scattering energy of 10.4467 keV, with an energy interval of 0.5 eV. The spectra were collected and averaged over 40 h of irradiation time. The Py-phase for the XRS measurements was synthesized from FeOOH (goethite) at 110 GP and 2000-2200 K using laser-heating systems at 13ID-D and HPSynC. LiF platelets were deliberately used as pressure-transmitting medium and thermal insulating layers. LiF does not contain any oxygen and would not contribute any signals to XRS measurements on oxygen K-edge spectra of the Py-phase.
X-ray emission spectroscopy experiments. High-pressure XES measurements of the Py-phase at 300 K were performed at beamline 16-IDD, APS, ANL using a Rowland circle configuration with a spherical silicon (4 4 0) analyzer crystal with resolution of~1 eV. A helium tube was used to reduce scattering by air. An incident X-ray beam with an energy of 11.3 keV was used to excite the emission. The collection time for each XES spectrum was~1 h. The 4-7 spectra were added for good statistics at a given pressure. Raman spectra of the diamond anvils were collected as a pressure gauge before and after each measurement while the ruby scale was cross-checked below 60 GPa. The Py-phase samples were synthesized from Fe 2 O 3 and H 2 O approximately at 100-110 GPa and 2000 K. The XES spectrum of the Py-phase at the highest pressure of 133 (±4) GPa was used as the reference for the low-spin state. The total spin moment was then evaluated using the integrated spectral area from 7030 eV to 7048 eV with respect to that of the low-spin reference (Supplementary Figure 9).
Theoretical simulations. Now with experimental knowledge of materials as the base, theory can successfully provide accurate and quantitative details in the extended pressure-temperature regime inaccessible to experiment 54 . The structure of FeO 2 has attracted many computational studies. However, an accurate description about its electron-correlation part in calculating the total energy still remains a challenging task. Based on previous studies, O 2p orbitals may hybrid with Fe 3d orbitals at the Fermi surface. It may cause O electrons to contribute to the electron-correlation portion in calculating the total energy. We note the linearresponse method is excellent in dealing with the exchange-correlation energy for one atom species (e.g. Fe), although it may encounter issues for two types of atoms at the same time. Therefore, the use of Hubbard parameter U gives a first-order approximation to this energy portion. Combined with the self-consistent linearresponse approach 55,56 , many experimental observations can be reproduced by theory.
First-principles calculations based on density function theory using the package VASP were carried out using the projector-augmented wave pseudopotential and a plane-wave cutoff energy of 600 eV. The exchangecorrelation functional employs the parameterization by Perdew, Burke, and Ernzerhof under the Generalized Gradient Approximation. We follow our previous computational recipe 18 to approximate the on-site Coulomb interaction by the Hubbard method (U = 5.0 and J = 0.8 eV). In parallel, we also employed the Heyd-Scuseria-Ernzerhof screened hybrid functional (HSE) that was recently adopted for a variety of iron-oxides and calculated reliable structural, magnetic, and electronic properties 57 . For HSE-type functional, the screening variable was set to 0.15 since it produces reasonable values in other Fe-O minerals like wüstite, magnetite, hematite, and goethite. We generated kpoint in gamma centered Monkhorst-Pack grids of 7 × 7 × 7 throughout our simulation. The structures were relaxed for atomic position, cell shape, and volumes for target pressures until the force acting on each atom was smaller than 0.01 eV/Å. Following the Bader charge division scheme, we are able to determine the valence state of Fe and O atoms from the electron charge density (Table S1). Calculated valence states definitely are against ferric (Fe 3+ ) or ferryl (Fe 4+ ) state of iron. Iron and oxygen atoms in pyrite-structured FeO 2 exhibit partial covalent feature, which is recognized as a common feature in pyrite FeS 2 model 37 . Therefore, our simulation results confirm that the valence state of iron in the Py-phase is close to Fe 2+ with O-O bonding.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding authors and J.L. (Jin.Liu@hpstar.ac.cn).