Origin of stabilization and destabilization in solid-state redox reaction of oxide ions for lithium-ion batteries

Further increase in energy density of lithium batteries is needed for zero emission vehicles. However, energy density is restricted by unavoidable theoretical limits for positive electrodes used in commercial applications. One possibility towards energy densities exceeding these limits is to utilize anion (oxide ion) redox, instead of classical transition metal redox. Nevertheless, origin of activation of the oxide ion and its stabilization mechanism are not fully understood. Here we demonstrate that the suppression of formation of superoxide-like species on lithium extraction results in reversible redox for oxide ions, which is stabilized by the presence of relatively less covalent character of Mn4+ with oxide ions without the sacrifice of electronic conductivity. On the basis of these findings, we report an electrode material, whose metallic constituents consist only of 3d transition metal elements. The material delivers a reversible capacity of 300 mAh g−1 based on solid-state redox reaction of oxide ions.

R echargeable lithium-ion batteries (LIBs) are widely used in our daily life because LIBs have the highest gravimetric/ volumetric energy density among commercial energy storage devices. LIBs are used as a power source for zero emission electric vehicles and are expected to be used for grid energy storage 1 . LIBs are becoming a key technology enabling a shift from fossil fuel to renewable electric energy, which potentially realizes green and sustainable energy development in the future. Nevertheless, the possibility of further increase in the energy density is severely restricted because of unavoidable theoretical limits for positive electrodes, such as spinel-type oxides and iron phosphates.
In contrast to these materials, which are currently used in the commercial LIB, there is room for further increase in the energy density for layered oxides, such as LiCoO 2 (ref. 2), LiNi 1/2 Mn 1/ 2 O 2 (ref. 3), Li 2 MnO 3 (ref. 4) and those derivatives [5][6][7] . Among them, Li 2 MnO 3 -based electrode materials have been extensively studied as positive electrode materials in the past decade 6,[8][9][10][11][12][13][14][15] . The reaction mechanism of this material was a controversial subject for a long time. Since the oxidation state of manganese ions is tetravalent, oxidation of manganese ions beyond the tetravalent state is difficult. Instead of manganese ions, negatively charged anions, oxide ions (O 2 À ), donate electrons on charge (electrochemical oxidation). However, oxidation of oxide ions results in partial loss of oxygen as an irreversible process, that is, decomposition reaction 8 . The loss of oxygen induces the formation of trivalent manganese ions on discharge (electrochemical reduction), leading to the unfavourable phase transition in the layered structure 11 . Nevertheless, it has been evidenced that reversible solid-state redox for oxide ions is possible for the Li 2 RuO 3 -based system, which essentially has the same crystal structure as Li 2 MnO 3 , and the contribution of oxide ions has been experimentally evidenced by using an arsenal of characterization techniques 14 and theoretical method 16 . Very recently, the formation of peroxo-like dimers in Li 2 À x IrO 3 has been experimentally visualized by transmission electron microscopy 17 whereas recent experimental 18 and theoretical 19 studies have suggested that an isolated hole is formed on the charge process for lithium-excess electrode materials. Enrichment of lithium ions as the highly ionized cation results in a less covalent character for oxide ions, and thus the oxide ions are more easily oxidized compared with conventional oxides with late transition metals. Therefore, it has been proposed that the holes in the oxygen 2p orbital is effectively stabilized 18,19 .
The use of anion redox, especially oxide ions, is a crucial strategy to design and develop new electrode materials with high gravimetric/volumetric energy density for LIB. Reversible capacity of electrode materials is potentially further increased by the enrichment of lithium contents with less transition metals in the close-packed structure of oxide ions. Recently, our group has reported that Li 3 NbO 4 (refs 20,21) and Li 4 MoO 5 (ref. 22), which have higher lithium contents than those of Li 2 MnO 3 and Li 2 RuO 3 , are potentially utilized as host structures for a new series of high-capacity electrode materials. Similar concepts are also proposed in the literature 23,24 . Manganese-substituted Li 3 NbO 4 , Li 1.3 Nb 0.3 Mn 0.4 O 2 (0.43Li 3 NbO 4 -0.57LiMnO 2 ), delivers large reversible capacity (approximately 300 mAh g À 1 ) with reversible solid-state redox reaction of oxide ions 20 . Similar to pentavalent niobium, a material with pentavalent antimony, Li 4 FeSbO 6 , has been recently reported 25 . Solid-state redox reaction of oxide ions is also activated in Li 4 FeSbO 6 , and a reductive coupling mechanism as an irreversible process has been evidenced in this system. As a non-rocksalt system, the use of Codoped Li 2 O has been also proposed 26 . Although many articles now describes the anion redox for battery materials, the border between reversibility and irreversibility for the solid-state redox reaction of oxide ions remains unclear, and it is a critical point to understand the factors affecting reversibility of anion redox.
In this article, we answer these questions through systematic studies on Li 3 NbO 4 -LiMeO 2 (Me ¼ Fe, Mn and V) binary system. Reversibility of the solid-state redox reaction highly depends on the transition metal elements selected, which correlates with the formation of electrochemically unstable superoxide species because of charge transfer from oxidized oxide ions. Furthermore, on the basis of these findings, we demonstrate a niobium-free high-capacity positive electrode material, which effectively utilizes reversible solid-state redox reaction of oxide ions. These findings can potentially enable highenergy LIBs free of less abundant transition elements, such as cobalt and nickel ions.

Results
Synthesis and characterization of Li 3 NbO 4 -LiMeO 2 system. Figure 1 summarizes characterization of Li 3 NbO 4 -LiMeO 2 (Me ¼ Fe, Mn and V) binary system by synchrotron X-ray diffraction (SXRD) and SEM. Li 3 NbO 4 crystallizes into a cationordered rocksalt-type structure while a lack of d electrons in a conduction band results in an insulating character. Substitution of 3d transition metal ions for Nb/Li ions effectively induces conductive electrons, and colour of samples is also changed from white for Li 3 NbO 4 to black for substituted samples with Mn 3 þ and V 3 þ . Such 3d transition metals can accept electrons from oxide ions. However, a change in the crystal structure is also unavoidable, and formation of a cation-disordered rocksalt-type structure is found. Recently, electrode materials with the cationdisordered rocksalt-type structure, such as Li 1  high-capacity electrode materials. Historically, such cationdisordered rocksalt phase had been regarded as electrochemically inactive as electrode materials because of a lack of the Li migration path in a bulk structure. Nevertheless, formation of percolating network for the Li-excess system (Li 1 þ x Me 1 À x O 2 ) opens the path for Li migration in the cationdisordered rocksalt-type structure 27 .
Electrochemistry of Li 3 NbO 4 -LiMeO 2 system in half-cells. Although as-prepared samples with primary particle size of 2-3 mm show insufficient electrode performance 20 , mechanical ball milling with carbon (reduction of particle size with uniform mixing with carbon is achieved in Supplementary Fig. 1) effectively improves the electrode performance of the samples in Li cells. Electrode performance is further improved at elevated temperature (50°C), and three cation-disordered rocksalt samples, Li 1.3 Nb 0.3 Me 0.4 O 2 (Me ¼ Fe 3 þ , Mn 3 þ and V 3 þ ), deliver large reversible capacities in Li cells, as shown in Fig. 2. It is noted that three Li 3 NbO 4 -based samples with different 3d transition metals show quite different electrochemical behaviour. The Fe system shows a large initial charge (oxidation) capacity of 350 mAh g À 1 that is quite close to that of the theoretical capacity (383 mAh g À 1 ) as defined by the extraction of all Li þ ions (1.3 moles) from the crystal lattice. However, a clear change in a voltage profile is found on the second charge process, which is totally different from the first charge process. A well-defined voltage plateau is observed on the initial charge, but an S-shaped profile is noted after first discharge process. Polarization is small for the S-shaped profile centred at 2.5 V, which will be discussed in the later section. The voltage plateau is not observed for the V system, and the sample shows S-shaped profile from the initial charge process. The observed reversible capacity is comparable to that of a theoretical capacity based on V 3 þ /V 5 þ two-electron redox (236 mAh g À 1 ) and is much smaller than that expected from lithium contents in the structure. The Mn system delivers a large discharge capacity of 300 mAh g À 1 with the appearance of voltage plateau as reported in our literature 20 . Capacity retention of the samples is compared in Supplementary Fig. 2  from 3 to 4 V and slightly large polarization for the plateau region at 4.2 V. On the discharge process, two regions are not distinguished and a continuous S-shaped profile on QOCV is observed. Such behaviour originates from the hysteresis of oxidation/reduction reaction for solid-state redox of oxide ions.
Charge compensation mechanisms in Li 3 NbO 4 -LiMeO 2 system. Three different samples show quite different electrochemical behaviour in Li cells, as shown in Fig. 2. Such behaviour is expected to originate from differences in charge compensation mechanisms depending on the 3d transition metals (Fe, Mn and V) in the Li 3 NbO 4 framework. Charge compensation mechanisms in Li cells were, therefore, examined using a combination of different characterization methods: synchrotron SXRD, hard/soft X-ray absorption spectroscopy (XAS) and X-ray photoelectron spectroscopy (XPS) with the assist of density functional theory (DFT) calculations. Results are described in detail in Supplementary Figs 3 Important findings are summarized as follows: (1) In the Mn system, reversible oxidation of oxide ions, coupled with Mn 3 þ / Mn 4 þ redox, is realized as reported in our previous work 20 .
(2) In the V system, V 3 þ /V 5 þ two-electron redox is active, but the oxidation of oxide ions is not evidenced. (3) In the Fe system, the formation of superoxide (O 2 -) (ref. 29) is observed from the measurement of O K-edge XAS, which is further supported by the DFT study with COOP analysis (Supplementary Fig. 14). Since the most significant changes on charge/discharge curves were observed for the Fe system, changes in soft XAS spectra were examined on the initial cycle and second charge at 50°C (Fig. 4). Formation of superoxide is further pronounced on charge at 50°C, and superoxide is stabilized in the bulk of particle rather than the surface ( Supplementary Fig. 9). However, superoxide species are electrochemically oxidized and decomposed by further charge. The superoxide species disappears as seen in the O K-edge spectra after charge to 4.8 V. This process inevitably results in the oxygen loss and structural reconstruction process (the latter is also supported by transmission electron microscopy as the formation of nanosize grains, as shown in Fig. 4e). Origin of the formation of superoxide is discussed in the later section. The oxygen loss results in the reduction of Fe 3 þ to Fe 2 þ and accumulation of surface deposits on discharge. The voltage plateau is significantly shortened on the second charge, and the S-shaped voltage profile centred at 2.5 V is observed with relatively small polarization. This reaction mainly originates from Fe 2 þ /Fe 3 þ redox, as shown in Fig. 4b, and the formation of superoxide is not found in the second charge. In the conventional Li-excess system, Li 1 þ x Ni y Co z Mn (1-x-y-z) O 2 , a clear voltage plateau is observed at 4.6 V at the initial charge process, but not for the second charge. An S-shaped voltage profile without the plateau region is observed from the second charge with activated Mn 3 þ /Mn 4 þ redox 8,9,11,12 . The reversible contribution of oxide ions for charge compensation is less pronounced in these systems. Oxygen loss on charge is further supported by XPS ( Fig. 4f and Supplementary Fig. 8). Oxygen molecules released in the cells are electrochemically reduced on discharge, leading to the formation of superoxide (on the surface of electrode), which further reacts with electrolyte 11,30,31 . This process is clearly evidenced, especially for the Fe system, as the accumulation of surface deposits on AB and active materials ( Fig. 4f and Supplementary Fig. 8).
In contrast, for Li 1.3 Nb 0.3 Mn 0.4 O 2 , reversible changes in O K-edge XAS spectra are observed (Supplementary Fig. 6). Moreover, a clear voltage plateau is observed even in the 'second' charge ( Fig. 2), indicating that solid-state redox reaction is a reversible process. Note that DFT calculations also support these findings. Formation energy of the charged Mn system is energetically stable, but segregation (decomposition) to LiFeO 2 and LiNbO 3 accompanying O 2 gas evolution is energetically preferable for the Fe system (see equations (s3) and (s4) in Supplementary Methods). Nevertheless, the plateau gradually becomes shorter in the continuous cycles probably because of the increase in polarization. Cyclability as the electrode material is expected to be further improved through the optimization of battery components, for example, electrolyte, binder, surface coating of particle and so on, and controlling charge conditions, as shown in our previous work 20 .
A question remains in the V system. Why are not oxide ions experimentally oxidized in this system? Theoretical prediction in Fig. 3 indicates that (1) two-electron redox of V 3 þ /V 5 þ occurs coupled with vanadium migration to tetrahedral sites, which is consistent with experimental finding (see the Supplementary Methods for theoretical and experimental evidences) and (2) oxidation of oxide ions is also possible in the V system, which was not experimentally observed on charge to 4.8 V, as shown in Fig. 2. This inconsistency for the oxide ion redox simply originates from the difficultly of the electron transfer from oxide ions to V 5 þ for the fully charged state, namely kinetic limitation. DFT calculation clearly supports this fact. Oxide ions are theoretically oxidized, but the hole induced in oxygen 2p orbital is isolated in the structure, as shown in Fig. 3. Vanadium ions in Li 0.5 Nb 5 þ 0.3 V 5 þ 0.4 O 2 (fully charged state), therefore, cannot transfer electrons from oxide ions and oxidation of oxide ions in the V system is kinetically restricted. Similar situation is observed for Li 3 À x NbO 4 as a model material. Calculated voltage by DFT study is estimated to be 4.8 V for Li 3 NbO 4 (Li 3/2 Nb 1/2 O 2 , Fig. 3a). Corresponding partial density of states diagram and partial electron density are also shown in Supplementary Fig. 15. Oxidation of O 2 À ions is indicated during delithiation by DFT study since creation of O 2p hole level is clearly visible just above Fermi level at x ¼ 0.5 in Li 3/2 À x Nb 1/2 O 2 . However, holes are completely isolated in the structure, and thus electron transfer between O 2p and Nb 5 þ is kinetically limited as observed in the experimental study 20 .
Design of high-capacity electrode materials with anion redox. As shown in this article, oxidation of oxide ions is not difficult, and only electron transfer from oxide ions to 3d transition metals is a necessary condition. However, the stabilization of oxidation reaction of oxide ions is not easy. In many materials, oxidation of oxide ions is possible, but this process induces oxygen loss, as in the cases of In general, peroxide/superoxide species are stabilized for sp elements without valence electrons (K þ , Ca 2 þ and so on ) and d 10 closed-shell (Zn 2 þ , Cd 2 þ and so on). In contrast, transition metal oxides often decompose these species. One typical example is a disproportionation reaction of H 2 O 2 catalysed by MnO 2 . This reaction is triggered by electron transfer between peroxide ions and surface manganese ions. Nb 5 þ also has no valence electron, and therefore a similar role is anticipated with the sp elements. Among 3d transition metal elements, Ti 4 þ has a similar electronic configuration with Nb 5 þ . These ions most probably screen off electrons of unstable 'oxidized' oxide ions and thus suppresses charge transfer. In addition, Nb 5 þ and Ti 4 þ are highly ionized ions compared with late transition metal ions with oxide ions, and thus the mixing between metal d orbital and oxygen 2p orbital is less pronounced. Therefore, similar to the lithium enrichment 18,19 , a character of oxide ions becomes more ionic (approaches two minus as a net charge) because of electron donation from Nb and Ti, and this fact would be beneficial to stabilize the oxidation of oxide ions.
To test this hypothesis, a binary system of Li 2 TiO 3 À LiMnO 2 has been examined. One to one composition between Li 2 TiO 3 and LiMnO 2 has been synthesized, which is reformulated as Li 1 Fig. 16). A result of structural analysis by neutron scattering is also provided in Fig. 5a and Supplementary Table 2   to enhance the electrode performance. Thus prepared sample shows a large reversible capacity, as shown in Fig. 5b, and the Nb-free sample delivers more than 300 mAh g À 1 at 50°C. A voltage profile of Li  36 . On the initial charge, Mn 3 þ is oxidized to Mn 4 þ for the slope region to 4 V as evidenced from Mn L-edge XAS spectra, and this fact also influences the profile of O K-edge XAS spectra (Fig. 5d). No change in Mn L-edge XAS spectra is observed for the plateau region at 4.2 V. A change in electronic structures is also not evidenced for Ti L-edge XAS spectra (Supplementary Fig. 17). Similar to Li 1.3 À x Nb 0.3 Mn 0.4 O 2 , a new peak appears at ca. 530 eV, (see Supplementary Fig. 6 for more details), which is further intensified as increase in the charge capacity for the plateau region, and the formation of superoxide is not evidenced. Similar observation is also noted for the recent study for the conventional Li-excess system 18 , but the change is more clearly pronounced in the Ti-Mn system. Although the energy of the new peak is consistent with that of Li 2 O 2 , further study is needed to understand the factor affecting the profile of XAS spectra. Nevertheless, the possibility of the formation of superoxide species can be excluded. Such change potentially originates from two possibilities: the formation of isolated holes 18,19 and/or s-hybridization as theoretically proposed in the Ru-Sn system after delithiation 19 . An almost identical profile with the pristine sample is observed after discharge to 1. The selection of 3d transition metals, which accept electrons from oxide ions on charge, is an essential key to determine whether to stabilize oxidation of oxide ions (the formation and stabilization of isolated holes as proposed in the literature 18,19 ) or to form unstable superoxide-like species. Stabilization/ This process is also called as the reductive coupling 14,16 . The formation of superoxide is further supported by the DFT study with COOP analysis ( Supplementary  Fig. 14). Such superoxide species would be stabilized by coupling with Li þ and Nb 5 þ (Ti 4 þ ) ions. Similar results are expected for Ni 3 þ /4 þ and Co 3 þ /4 þ with heavily hybridized characters for oxide ions near the Fermi level and indeed oxygen loss is experimentally observed for Li-excess electrode materials with these elements. In contrast, oxide ions do not donate electrons to less covalent Mn 4 þ with the d 3 configuration (as the high-spin configuration) associated with energy gap between filled t 2g and empty e g bands, as shown in Figs 3 and 6. Ru 5 þ also have a similar electronic configuration. Since the energy level of t 2g orbital for Mn 4 þ is high enough than that of Fermi level, oxide ions are solely oxidized on further oxidation. The presence of electrons in t 2g orbital is also essential without the sacrifice of electronic conductivity in bulk. Moreover, this reaction is further stabilized by the presence of Nb 5 þ and Ti 4 þ , which donates electrons to oxide ions because of high ionic characters as cations and completely suppresses charge transfer from oxidized oxide ions.
In conclusion, the use of solid-state redox reaction of oxide ions is an effective strategy to further increase energy density of LIBs. We have demonstrated that the stabilization of redox reaction for oxide ions is possible using a combination of only 3d transition metals as the metallic constituents. This serves as a significant proof-of-concept towards practical applications. We expect that by relaxing the constraints posed on materials design by the conventional concept of transition metal redox, many new positive electrode materials with high capacity will appear, similar to Li  Fig. 2, Supplementary Figs 2 and 8 were mixed with 10 wt% carbon. Composite electrodes consisted of 76.5 wt% active materials, 13.5 wt% acetylene black and 10 wt% poly(vinylidene fluoride), pasted on aluminium foil as a current collector. Rest of the data were collected using the samples with 20 wt% carbon. Composite electrodes consisted of 72 wt% active materials, 18 wt% acetylene black and 10 wt% poly(vinylidene fluoride). Metallic lithium (Honjo Metal) was used as a negative electrode. The electrolyte solution used was 1.0 mol dm À 3 LiPF 6 dissolved in ethylene carbonate:dimethyl carbonate (1:1 by volume) (Kishida Chemical). A polyolefin microporous membrane was used as a separator. R2032-type coin cells (Hosen Corp.) or TJ-AC (Tomcell Japan) were assembled in the Ar-filled glove box. The cells were cycled at a rate of 10 mA g À 1 at room temperature or 50°C.  Materials characterization. Soft X-ray absorption (XAS) spectra were collected at BL-11 (O K-edge and Me L II, III -edges) in the synchrotron facility of Ritsumeikan University (Synchrotron Radiation Center) 36 . The absorption spectra were collected with fluorescence yield and electron yield modes. Similar to the measurements for hard XAS, the samples were prepared in the Ar-filled glove box, and thus prepared samples were set on the spectrometer using a laboratory-made transfer vessel without air exposure. Normalization of the XAS spectra was carried out using the program code IFEFFIT (ref. 39). The postedge background was determined using a cubic spline procedure. XPS measurements were carried out with VG ESCALAB 250 spectrometer (Thermo Fisher Scientific K.K.) using monochromatized Al Ka X-ray radiation (1486.6 eV). The system was operated at 15 kV and 200 W. The base pressure of the analysis chamber was less than 10 À 8 Pa. These characterizations were carried our using a laboratory-made transfer vessel to avoid the sample exposure to moisture/air. transmission electron microscopy observation was conducted by using JEM-ARM200F (JEOL) operated at 200 keV. The samples were dispersed in dimethyl carbonate and then supported on a copper grid.

Energy
A neutron diffraction pattern was collected at BL09 (SPICA) in the Material and Life science Facility (MLF) of the Japan Proton Accelerator Research Complex (J-PARC) (ref. 40).
Data availability. The data supporting the main findings of this study are available from the corresponding authors on request.