Improving the oxygen redox reversibility of Li-rich battery cathode materials via Coulombic repulsive interactions strategy

The oxygen redox reaction in lithium-rich layered oxide battery cathode materials generates extra capacity at high cell voltages (i.e., >4.5 V). However, the irreversible oxygen release causes transition metal (TM) dissolution, migration and cell voltage decay. To circumvent these issues, we introduce a strategy for tuning the Coulombic interactions in a model Li-rich positive electrode active material, i.e., Li1.2Mn0.6Ni0.2O2. In particular, we tune the Coulombic repulsive interactions to obtain an adaptable crystal structure that enables the reversible distortion of TMO6 octahedron and mitigates TM dissolution and migration. Moreover, this strategy hinders the irreversible release of oxygen and other parasitic reactions (e.g., electrolyte decomposition) commonly occurring at high voltages. When tested in non-aqueous coin cell configuration, the modified Li-rich cathode material, combined with a Li metal anode, enables a stable cell discharge capacity of about 240 mAh g−1 for 120 cycles at 50 mA g−1 and a slower voltage decay compared to the unmodified Li1.2Mn0.6Ni0.2O2. Tailoring the oxygen redox reactivity in Li-rich cathode is crucial for developing high-energy batteries. Here, the authors report a strategy to obtain a flexible crystal structure and enhance the oxygen redox reversibility.

L ithium-ion batteries (LIBs) play a vital role as rechargeable power sources in various electronic devices and facilities today; in particular, they are used in electric vehicles (EVs) owing to their high specific energy, high potential, and light weight characteristics [1][2][3] . In LIBs, cathode materials largely determine the specific energy, lifespan, and safety because they have a lower specific capacity and unstable structure compared with anode materials. Therefore, the development of highperformance cathode materials is extremely urgent. In all kinds of cathode materials, Li-rich Mn-based oxides are one of the most promising candidates for next-generation cathode materials due to their high specific capacity (>300 mAh g −1 ) and high specific energy (~1000 Wh kg −1 ), which originate from the charge compensation of both the anion and cation redox chemistry in Li-rich Mn-based oxides. However, some critical issues, such as structural instability, poor rate capability, and voltage decay, need to be resolved. In particular, continuous voltage decay and irreversible oxygen release are the main stumbling blocks for commercialization since they gradually reduce the energy content and lifetime during cycling. Safety concerns caused by oxygen evolution are also critical for Li-rich Mn-based oxides. To overcome these issues, numerous research efforts have been conducted to unravel the reasons for voltage fade and irreversible anion redox chemistry, and search for modification strategies.
Currently, it is well accepted that the layered to spinel phase transition caused by transition metal (TM) migration is the primary origin for voltage decay [4][5][6] . It is worth noting that this process is closely related to the oxygen release induced by the irreversible oxygen redox chemistry because the structural instability and oxygen release from oxygen redox chemistry can accelerate voltage fading [7][8][9][10] . Therefore, the study and modulation of anion redox chemistry is urgent. To better modulate the anion redox activity and reversibility, scientists have proposed covalence theory 11,12 and unhybridized oxygen 2p bands 13 . In particular, the latest O nonbonding states theory has been widely accepted because it can effectively rationalize anion redox chemistry 12,13 . Recent studies report that the electrochemical property of this O nonbonding state is related to the relative positions of the antibonding (TM-O)* and O 2p nonbonding bands. However, the band position is determined by the introduction of the d-d Coulomb interaction term U and charge transfer (CT) term Δ from solid-state physics 12,14,15 . According to this theory, electrons are extracted from the filled lower-Hubbard bands (LHB) caused by Mott-Hubbard splitting in many oxides and fluorides due to U << Δ (Fig. 1a), which is known as cationic redox chemistry. In contrast, electrons are removed from the unhybridized O 2p state located above the LHB and are accompanied by serious oxygen release when U >> Δ for most Li-rich cathode materials, which is called irreversible oxygen redox (Fig. 1b). For the middle case, U ≈ 2Δ, electrons can exchange from both the nonbonding O 2p state and (TM-O)* band due to the overlap between the LHB and unhybridized O 2p bands (Fig. 1c). Therefore, the capacity can be increased from both the anionic and cationic redox reactions while mitigating oxygen release, meaning that oxygen redox is more reversible 12,15 . Hence, to obtain reversible anion redox and decrease structural damage, strategies to regulate U and Δ are urgently needed. One is to choose a TM with a higher d state, such as Ru and Ir, to reduce the U term; as a result, the LHB can overlap with the nonbonding O 2p band better, and the anion redox chemistry will be enhanced while reducing O 2 escape. The other is to substitute O atoms with less electronegative elements, such as S and Se, which can make ligands with nonbonding p states approach the TM band. However, neither strategy is suitable for practical application due to their high cost or low voltage plateau. It is worth noting that Li 1.2 Mn 0.6 Ni 0.2 O 2 with oxygen vacancies and/or a spinel phase can improve the properties of cathode materials. Nonetheless, until now, to the best of our knowledge, the study of mechanisms is still insufficient, especially from the direction of the band position combined with experiments.
Herein, we reveal that the enhancement of oxygen redox reversibility and the mitigation of TM migration in Li-rich Mnbased oxide (Li 1.2 Mn 0.6 Ni 0.2 O 2 ) materials with oxygen vacancies are closely related to the adjustment of d-d Coulomb interaction U. The enhanced d-d excitation of Mn and the lowering of term U can be verified by means of resonance inelastic X-ray scattering (RIXS) with Li 1.2 Mn 0.6 Ni 0.2 O 2 due to the reduction of Mn and the incorporation of oxygen vacancies, which illustrates the strength of reversible oxygen redox while reducing O 2 release. Both of the above results are validated by electrochemical energy storage characterization and differential electrochemical mass spectrometry (DEMS) measurements. We also report that the softened crystal structure is favorable to the reversible distortion of TMO 6 octahedra in TM layers rather than the migration of TMs caused by a broken rigid octahedral structure; therefore, the voltage fade can be mitigated. Furthermore, based on the experimental results, the general enhancement mechanism for the reversible anion redox chemistry is rationalized by the schematic density of state (DOS) patterns.

Results
Creation of oxygen vacancies and reduced manganese. To study the effect of oxygen vacancies and spinel-like structures in the pristine Li 1.2 Mn 0.6 Ni 0.2 O 2 (LRMO) cathode, a modified sample Li 1.2 Mn 0.6 Ni 0.2 O 2-δ (M-LRO) with oxygen vacancies and reduced Mn is synthesized by a more uniform liquid-solid method. First, we used X-ray powder diffraction (XRD) to determine the crystal structure of pristine and Li 1.2 Mn 0.6 Ni 0.2 O 2-δ , and the refined patterns of both samples are shown in Supplementary Fig. 1. The results illustrate that both samples are composed of layered rhombohedral phases with the space group of R 3m, but peaks of the spinel phase are not found. There are some small superlattice peaks at approximately 20°, which are caused by the monoclinic phase (C2/m); however, these are not refined because of their low content. Moreover, neutron powder diffraction (NPD) is used to quantitatively detect the amount of oxygen vacancies due to its sensitivity to light elements. Figure 1d, e show the refined patterns, and f and g show the crystal models and corresponding lattice parameters. As a result, oxygen vacancies account for approximately 5% of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample. Based on the refined results, we simulated the effect of different oxygen vacancy concentrations on the structure, as shown in Supplementary Fig. 1d. By comparing with Supplementary Fig. 1c, we find that 5% oxygen vacancies have almost no significant effect on the crystal structure. Furthermore, the lattice parameters and atomic occupation information of both samples are shown in Supplementary Table 1 and Supplementary Table 2, respectively. We can find that the lattice parameters and TM-O bond length are increased for Li 1.2 Mn 0.6 Ni 0.2 O 2-δ due to the reduction expansion 16,17 , which indicates the existence of crystal distortion. Moreover, the increased Li + diffusion channel is beneficial to a high rate capability. Spherical aberration-corrected scanning transmission electron microscopy (SAc-STEM) was used to observe the atomic arrangement. As Supplementary Fig. 2 shows, the high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images indicate that the bulk structure is not destroyed by hydrazine aqueous solution, but some zigzag atomic ladders are produced on the surface of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample due to the oxygen vacancies, which can provide more active sites and reduce the thickness of the cathode electrolyte interphase (CEI) during extended cycling.
Moreover, there are relatively more dislocations in the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample ( Supplementary Fig. 2c, d); therefore, although this requires further study, we think that these boundary defects and dislocations are the causes of crystal structure softening 18,19 . The 7 Li magic angle spinning (MAS) solid-state nuclear magnetic resonance (ssNMR) measurements are shown in Supplementary Fig. 3, where the centre peaks at 0 ppm arecaused by Li 2 CO 3 , LiOH and Li 2 O on the surface and located in a diamagnetic environment 20 . The oxygen vacancy could be illustrated by the widened sideband pattern in the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample. Because oxygen vacancies affect the asymmetric distribution of the electron cloud around the nucleus, the local magnetic field around the nucleus is perturbed, thereby affecting the anisotropy of the chemical shift of the nucleus, which will broaden the spectrum peak 21,22 . In addition to the production of oxygen vacancies, tetravalent manganese can also be reduced, which can be confirmed by the calculation results of the fitted Mn 3 s spectra, obtained by X-ray photoelectron spectroscopy (XPS) 23 , and Mn L-edge spectra of soft X-ray absorption spectroscopy (XAS, Supplementary Fig. 4). According to the formula between the Mn valence (ν Mn ) and splitting energy of Mn 3s 24 Fig. 4c). It is worth noting that the bulk structure is uninfluenced for both samples, as shown by the partial fluorescence yield (PFY) mode because of its detection depth of 100 nm.
Electrochemical characterizations. Based on the materials and analysis above, the electrochemical characterizations of both samples were performed. First, we tested the rate capacity of different samples treated with different concentrations of aqueous hydrazine for different times. We find that the sample obtained after treatment with a 2 M hydrazine aqueous solution for 1 h exhibits the best rate capacity ( Supplementary Fig. 5). Hence, we only focus on this sample and the pristine material in the following tests to study the mechanisms. There is a slightly visible plateau at~2.4 V for the modified sample in the initial discharge curves and the dQ/dV profile in Fig. 2a, b, which is related to the reduction or activation of Mn 4+/3+ and confirms the existence of Mn 3+ in the initial state 25,26 . Figure 2c shows the rate capacity of both samples. The specific capacity contribution is more than 260 mAh/g at a specific current of 25 mA/g for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, which is ascribed to the enhanced anion and cation redox chemistry. In comparison, the discharge capacity of the pristine sample is only~200 mAh/g at the same specific current. The leading discharge capacity is still maintained at a high specific current for the modified material. In addition to the excellent capacity in the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, the voltage decay is significantly mitigated at different specific currents for Li 1.2 Mn 0.6 Ni 0.2 O 2-δ (Fig. 2d). Specifically, the median discharge voltage of the pristine sample is lower than that of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample at any specific current. The electrochemical performance is also illustrated by the cycling stability at different specific currents. Figure 2e shows the cycling property, voltage decay behavior and Coulombic efficiency of both samples at a specific current of 50 mA/g. A negligible capacity fade can be observed during 120 cycles for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ material. The modified Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample has less voltage decay with an attenuation value of only 1.7 mV per cycle for the, than the pristine sample (2.4 mV per cycle). This improvement in electrochemical performance can also be reflected by the smaller fluctuation in the Coulombic efficiency of the modified sample in the right panel of Fig. 2e. In addition to the cycling performance at a low specific current, the cycling performance at specific currents of 250 mA/g and 1250 mA/g are investigated (Supplementary Fig. 6a, b). As the results show, the capacity of Li 1.2 Mn 0.6 Ni 0.2 O 2-δ is higher and more stable than that of the pristine sample. For the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, the voltage decay is only 1.0 and 0.54 mV/cycle at specific currents of 250 mA/g and 1250 mA/g, respectively. In contrast, the voltage attenuation of the pristine sample is more serious, and the decay value reaches 1.8 and 0.68 mV/cycle at the same specific current. Additionally, the Coulombic efficiency patterns at different specific currents are shown in Supplementary Fig. 6c, d. Clearly, the modified sample has better Coulombic efficiency with an average greater than 99% after long-term cycling. The above rate performance is in good agreement with the result of the potentiostatic intermittent titration technique (PITT) in Supplementary  Fig. 7 because the large lithium-ion diffusion coefficient allows the battery to undergo fast charging; the detailed discussion is shown in Supplementary Fig. 7.

Reversible distortion driven by reduced Mn and O vacancies.
To reveal the mechanisms of the mitigated voltage decay and increased capacity, in situ XRD measurements are carried out to monitor the crystal structure change in real time. The overall contour patterns of the pristine and Li 1.2 Mn 0.6 Ni 0.2 O 2-δ samples are presented on the left of Fig. 3a, b. The asterisk indicates the Be/BeO peaks from the window of the cell mould. The representative diffraction peaks are indexed with (003), (101) and (104). The change in amplitude of these three peaks is smaller for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample than for the pristine sample. To clearly observe the change, they are extracted from the overall pattern and inspected, as shown in the middle of Fig. 3a, b. For both samples, the (003) peak shifts to a lower 2θ degree; however, the (101) and (104) peaks shift to a higher degree during the charging process, which is attributed to the increased lattice parameter c caused by the promoted electrostatic repulsion of adjacent oxygen layers and the shrinkage of the TM-TM distance caused by the shortened TM-O bond due to Li + extraction, respectively 27 . Moreover, the oxidation of low valence state TMs with larger radii will also cause the layer spacing to contract 28 . It is worth noting that the peaks are almost stable for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample from 4.5 to 4.8 V, which indicates less oxygen release and more active reversible oxygen redox chemistry than the pristine sample. During discharge, the diffraction peaks of both samples shift to a reverse direction accompanied by Li + insertion. However, the peaks of the pristine sample cannot move to the original position, especially for the dashed line shown, which is caused by irreversible oxygen loss and irreversible TM migration from the octahedral sites of the rigid TM layers to the octahedral sites of the Li + layers, resulting in structural changes and voltage decay. In comparison, although it is thermodynamically favorable for TMs to migrate from TM layers to Li + layers, the softened crystal structure can bind TM migration in the form of a TMO 6 (MnO 6 /NiO 6 ) octahedral distortion to prevent TMs from migrating to the Li + diffusion tunnels. This stabilizes the crystal structure and oxygen, which is confirmed by the Mn migration barrier in the following discussion. As a result, the voltage fade is mitigated. In addition to the crystal structure, to illustrate the voltage decay mitigation by the flexible distortion of the softened structure due to oxygen vacancies, the stable electronic structure can be used to describe the reasons for voltage fade mitigation. The Mn K-edge spectra of both samples at different charge states are given in Fig. 4a, b. Accompanied by Li + deintercalation during charging, there is a weak continuous change in the Mn Kedge spectra of the pristine sample in Fig. 4a, which means that the electronic structure has been changed slightly but the valence state of manganese is stable. In contrast, the change in the absorption edges is more obvious for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample in Fig. 4b, which indicates that the electronic structure and valence state of the Mn ions have both been changed when charged. The insets show the pre-edge absorption peaks, which illustrate that the MnO 6 octahedral distortion is caused by the hybridization of the 3d−4p orbitals 29,30 . We find that the change is irreversible when the voltage is returned to 2.0 V from the charged 4.8 V in the pristine sample. We also quantitatively calculated the change in the pre-edge peaks at the charge-discharge states ( Supplementary Fig. 8a). The intensities of both samples increase continuously with Li + deintercalation and reach a maximum when charged to 4.8 V. However, the change in amplitude of the pristine sample is more serious than that of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample. In particular, the signal cannot return to the original position when the pristine sample is discharged to 2.0 V. Moreover, we define here that the intensity of structural distortion is proportional to the change in intensity of the pre-edge peaks; therefore, the corresponding extent of distortion is calculated according to the intensity of pre-edge peaks, and the distortion of the open circuit voltage (OCV) state is set to one ( Supplementary Fig. 8b) 29 . The distortion of the pristine and Li 1.2 Mn 0.6 Ni 0.2 O 2-δ samples reaches approximately 195% and 160% when charged to 4.8 V, respectively. When discharged to 2.0 V, the distortion of the pristine sample is still as high as 140%; in sharp contrast, the distortion of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample is almost the same as that in the OCV state. This distortion illustrates that the Mn ions have migrated and stayed in the Li + layers from the broken rigid MnO 6 octahedron, which is caused by the excessive distortion of the pristine sample; hence, the local structure is irreversibly changed. In comparison, the distortion of the Li 1 Fig. 9b). Therefore, the voltage decay can be significantly suppressed. To further reveal that the softened structure can inhibit the migration of Mn ions, ex situ Mn K-edge spectra of both samples with different cycles are given in Fig. 4c Fig. 9c, d).
In addition, soft XAS is sensitive to the oxide states and orbital information of different elements. The ex situ Ni and Mn L-edge spectra and O K-edge spectra of both samples are collected in total electron yield (TEY) mode, which can detect to a depth of 5-10 nm (Supplementary Fig. 10)   discharged to 2.0 V, notably, there is also some Mn 3+ in the pristine sample, which is caused by the activation of the Li 2 MnO 3 component and structural rearrangement ( Supplementary  Fig. 10a, b). The Ni L 3,2 -edges are located at approximately 853 and 870 eV, respectively; in particular, the L 3 -edge shows asymmetric split peaks at higher and lower energies. We find that the lower energy split peak begins to decrease, while the intensity of the higher energy peak increases when charged to 4.55 V, which illustrates that Ni 2+ is oxidized to Ni +3/4 . However, when charge to 4.8 V, the pristine sample shows Ni +4 ; in contrast, the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample shows partially reduced Ni 4+ , which is caused by a reductive coupling mechanism (RCM) 31,33 . Specifically, the O 2is oxidized to peroxo-like species, and the electrons are transferred to the LHB closer to the O 2p nonbonding band (Fig. 1c), thus reducing O 2 release, as shown in Fig. 5 (the corresponding test device diagram is in Supplementary Fig. 11). This phenomenon can also be observed in the bulk with FY mode (Supplementary Fig. 12). Regarding the O K-edge spectrum, the peaks below 534 eV are ascribed to the hybridization of O 2p with the TM 3d band, while the peaks above this point are attributed to the mixing of O 2p with the 4sp band 34 . The arrows indicate the e g orbital ( Supplementary  Fig. 10e, f). The changes in the e g peaks with increasing voltage indicate structural distortion, and the comparison shows that the distortion is larger in the pristine sample when charged to 4.8 V; moreover, when discharged to 2.0 V, the pristine e g peak cannot be restored, indicating that the distortion is irreversible due to the irreversible migration of TMs from the damaged rigid octahedral structure ( Supplementary Fig. 10e, f and Supplementary Fig. 13). This result further emphasizes the importance of flexible distortion caused by oxygen vacancies.
Suppressed cation migration during charge-discharge can be observed in the HAADF-STEM images ( Supplementary Fig. 14). The atomic-resolution image reveals that TMs significantly migrate to Li layers from the octahedral sites of the TM layers for the pristine sample, as shown by the rectangular box in Supplementary Fig. 14a. Moreover, the dark region indicates the existence of nanovoids, which are caused by cation and anion migration and oxygen release accompanied by oxygen redox chemistry; this promotes the degradation of the lattice structure and voltage decay 35 . In contrast, there is almost no change, outside of the slightly twisted TM arrangement, for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample when charged to 4.8 V. This result is consistent with the above analysis. The migrating TMs can be bound in the flexible lattice; thus, distortion is driven by the reduced Mn and oxygen vacancies. When the pristine sample is discharged to 2.0 V, the dumbbell-shaped TM-TM structure becomes blurred ( Supplementary Fig. 14c), while some TMs remain in the Li + layers, which could be associated with the side reaction of oxygen with the electrolyte and the irreversible migration of TMs from the broken rigid TMO 6 octahedron. However, the crystal structure of Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample remains intact because of its reversible distortion and oxygen redox ( Supplementary Fig. 14d). Moreover, compared with the retention of the superstructure peaks of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample 36 , the disappearance of the pristine sample shows that the TMs migrate irreversibly from the disturbed rigid crystal structure ( Supplementary Fig. 15), which is in line with the above results. The softened crystal structure mitigates TM migration by allowing reversible distortion.
Oxygen redox enhanced by the flexible distortion of O vacancies. The Li 1.2 Mn 0.6 Ni 0.2 O 2-δ electrochemical energy storage performance is also associated with reduced oxygen release and enhanced oxygen redox chemistry. The amount of oxygen release was measured by operando DEMS (Fig. 5 and Supplementary  Fig. 11). We find that the oxygen production of pristine and Li 1.2 Mn 0.6 Ni 0.2 O 2-δ samples is 0.31 and 0.16 μmol mg −1 , respectively. The reduced oxygen production can be attributed to the oxygen vacancies, and it can reduce the parasitic reaction between oxygen and electrolyte; therefore, less corrosion of the cathode material reduces the dissolution and migration of TMs, which can be verified by the thinner CEI ( Supplementary Fig. 16). As a result, the voltage fade is strongly mitigated. Although the oxygen release is reduced according to the DEMS results of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, oxygen redox is enhanced, which can be rationalized by the presence of peroxo-like (O 2 ) 2− species 33,37,38 and molecular oxygen 36,39 . First, the peroxo-like species caused by oxygen redox can be detected in the O 1 s XPS spectra (Supplementary Fig. 17). To detect the bulk structure, samples are etched for different times. When the etching time exceeds 9 min, the peak of the electrolyte oxidation species (533 eV) disappears, illustrating that the influence of the surface electrolyte is removed. In the figures, the purple region (530.5 eV) indicates the existence of lattice oxygen (O − /O 2À 2 ) caused by the oxygen redox chemistry. The larger the area is, the more peroxo/ superoxo-like species, therefore, we can find that the oxygen redox activity of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample is stronger, and the different etching times can also demonstrate this enhanced phenomenon in the bulk. Second, the high-resolution O K-edge RIXS spectrum can prove the existence of molecular O 2 in Fig. 6a-d. In the low energy loss region, vibrational peaks are observed when both samples are charged to 4.8 V. Specifically, the energy loss at~0.2 eV corresponds to a vibrational frequency of 1600 cm −1 , matching the vibration of O 2 molecules caused by oxygen redox 36,39,40 . However, we speculate that the more confined change in the inelastic peaks of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample is related to the distortion of octahedral TMO 6 . The softened crystal structure preserves the octahedral TMO 6 structure due to flexible distortion; therefore, the vibration of molecular oxygen will be confined in the undisturbed structure, which can be demonstrated by the redshift of the peaks between −1.0 and −0. cathode material due to oxygen redox chemistry. Moreover, the peak located at an energy loss of 8 eV also reinforces the existence of oxygen holes caused by the removal of electrons from the O 2p band 41 . The stronger peak means that the oxygen redox activity is higher for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample (Fig. 6a, c). The oxygen holes can also be confirmed by the O K-edge soft XAS spectrum [42][43][44] . Figure 7a, b show the ex situ O K-edge soft XAS patterns of the pristine and Li 1.2 Mn 0.6 Ni 0.2 O 2-δ samples in FY mode, respectively. As the dashed line shows, the intensities of the e g peaks weaken during charging, which indicates that structural distortion is irreversible for the pristine sample during cycling due to the structural arrangement accompanied by oxygen release. In comparison, the modified sample still shows a peak with weak intensity (Supplementary Fig. 13), which means that the softened structure results in flexible distortion to inhibit TM migration and structural arrangement, which is in line with the TEY mode results (Supplementary Fig. 10). The integrated intensity of the O K-edge XAS spectra between 525 and 534 eV demonstrates the distribution of oxygen hole states above the Fermi level 34 . Figure 7c shows the variation in the integrated intensity calculated according to the shaded area in Fig. 7a, b at different charge-discharge states. The results show that the number of oxygen holes gradually increase as charging progresses and reaches a maximum of 4.8 V for both samples. However, the integrated intensity of Li 1.2 Mn 0.6 Ni 0.2 O 2-δ is greater than that of the pristine sample, meaning that the oxygen redox activity is greater for the modified sample. It is worth noting that, although there are more holes in the modified sample, the emissive electrons from oxygen can be accepted by the (TM-O)* band instead of O 2 release; hence, the structure could be considered as stable 33 . In addition to absorption spectroscopy, the O K-edge RIXS map also illustrates the oxygen redox chemistry due to its sensitivity to the oxidation state of oxygen 45 . The two distinct features at the emission energy of 525 eV are attributed to TM(3d)-O(2p) mixing in Fig. 7d, e. In addition, the feature in the encircled area represents the fingerprint of oxygen redox due to its high resolution to the oxidation state of the oxygen ion 17,46 . This fingerprint feature is more distinguishable in the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample charged to 4.8 V than that of the pristine sample, which illustrates that there is more oxygen redox activity in the modified sample. This is in good agreement with the electrochemical energy storage performances and XAS data. To further illustrate the nature of the enhanced reversible oxygen redox due to the oxygen vacancies and reduced Mn, the Mn L-edge RIXS of both samples are obtained as shown in Fig. 7f, g. To compare the intensity change, the spectra are normalized to one, which can demonstrate the number of scattering photons or difficulty in scattering these photons. The elastic peaks located at 0 eV energy loss are attributed to the refilling of the 2p core holes by the incident electron, the yellow area represents the d-d excitation between 1 and 5 eV, which is caused by the refilling of 2p core holes by electrons occupying the 3d valence or conduction band, and the blue region means the charge transfer (CT) excitation [47][48][49] . First, the CT intensity is related to the extent of hybridization of the Mn-O bond. Therefore, the hybridization of Mn-O bonds decreases with increasing voltage in both samples. However, the peak intensity of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample can revert back to the original state compared to the pristine sample, which may be caused either by flexible distortion or less oxygen release in the modified sample 50,51 . Then, the peak intensity of the d-d transition represents the number of emitted photons; specifically, if the d-d Coulombic interaction (repulsion) term U is too large, electrons can be localized in the d atomic orbital. Hence, fewer photons can be detected during de-excitation. The d-d transition can be seen at the OCV for both samples, as the dotted line shows; however, the intensity of the transition for the pristine sample is weaker than that for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, indicating that U is smaller in the modified sample. With the extraction of Li + , although the repulsion gradually increases in both samples, electrons are more easily removed from the O 2p nonbonding band accompanied by oxygen release in the pristine sample, as shown in Fig. 7h, i. When Li + is reinserted into the structure, the d-d excitation peak of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample can again revert back to the original intensity at the OCV state; in contrast, the peak of the pristine sample almost disappears in the limit of resolution of the ex situ Mn L-edge RIXS spectra, which further illustrates that the MnO 6 octahedral structure is irreversible and destroyed due to rigid distortion.
Unified picture of enhanced oxygen redox in oxides with oxygen vacancies and reduced TMs. We use DFT to uncover the higher migration barrier of Mn ions for the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, which can explain why Mn is stable during electrochemical cycling of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ . The calculation models are shown in Supplementary Fig. 18. According to these models, the diffusion channel is given in Supplementary Fig. 19a, b. First, the Mn-ion migration path is more tortuous from the octahedral MnO 6 structure in the TM layers to the tetrahedral LiO 4 structure in the Li layers of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample than that of the pristine sample, meaning that migration is difficult during electrochemical cycling. Quantitatively, the Mn-ion migration barrier of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample is 2.50 eV; in contrast, the migration barrier of the pristine sample is lower (2.23 eV) ( Supplementary Fig. 19c), which is in good agreement with the above experimental analysis. In particular, the HAADF-STEM images (Supplementary Fig. 14) show that this result is consistent with the previous calculation 52 . The flexible octahedral MnO 6 structure distortion is also mimicked, and the Mn atom highlighted with a dashed circle close to the oxygen vacancy is used as a simulation target to study the degree of distortion  ( Supplementary Fig. 20) 53 . The results are shown in the histogram, and the distortion of the pristine sample is larger than that of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample with lithium extraction. More significantly, the distortion of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample can almost be restored to the original state; in contrast, the distortion of the pristine sample is still very large during Li + reinsertion, which is mainly ascribed to the oxygen vacancies softening the crystal structure and preventing damage to the structure caused by rigid distortion, which is in line with the XAS results. The electronic band structure is evaluated using the density of states (DOS), as calculated by DFT at different states (Supplementary Fig. 21), and the schematized patterns that are combined with the RIXS results are shown in Fig. 8. First, for the originating state, the Mott-Hubbard splitting, namely, the d-d Coulombic interaction term U, is smaller in the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample due to the existence of a small amount of Mn 3+ and the distortion caused by it. Because U is inversely proportional to the orbital volume, the larger Mn 3+ radius relative to Mn 4+ will also result in a smaller U 12 . Then, in the state of complete delithiation, the Li-O-Li band is irreversibly damaged. Therefore, the analysis is ignored at this state. However, in terms of qualitative analysis, the Mn ions are oxidized to +4 in both samples, and the smaller radius and half-filled t 2g 3 states both cause an increase in U. At this state, electrons are easily removed from the O 2p nonbonding band due to the O 2 release in both samples when comparing this state with the other states. The difference is that this trend is more obvious in the pristine sample, which is in line with the DEMS and RIXS results. Last, when Li + is reinserted back into the structure when discharged to 2.0 V, the crystal structure of the Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample can recover to the initial state; therefore, U decreases. In comparison, the broken structure prevents the decrease in term U for the pristine sample, for instance, irreversible Mn migration to Li + layers along with oxygen release, which is in good agreement with Mn L-edge RIXS results. Moreover, the reduced d band width and the d band centre being closer to the Fermi level, which is due to oxygen vacancies, facilitate the adjustment of U and Δ. ( Supplementary  Fig. 21). Therefore, the calculation results and the experimental results, especially the RIXS results, reveal why the reduced TMs (Mn) and oxygen vacancies can improve performance.

Discussion
With advanced characterization techniques, we studied the inherent nature of enhanced reversible oxygen redox caused by reduced Mn and oxygen vacancies. On the one hand, the reduced Mn can take part in charge compensation; on the other hand, oxygen vacancies can not only decrease O 2 release and mitigate Mn migration but can tolerate the structural distortion that occurs during electrochemical cycling by allowing flexible distortion. Therefore, the voltage decay is significantly suppressed. Specifically, the d-d Coulombic interaction term U can be decreased by the introduction of reduced Mn and oxygen vacancies, which can adjust U and prevent it from being much larger than Δ; thus, oxygen release can be decreased and reversible anion redox chemistry can be enhanced. Furthermore, flexible distortion can prevent the irreversible migration of TMs and maintain the integrity of the crystal structure; therefore, U can reversibly return to its initial state. As a result, the enhanced O redox chemistry becomes reversible. In summary, this work demonstrates an interesting approach to mitigate voltage decay and enhance reversible oxygen redox chemistry in Li-based cathode active materials by chemical reduction strategies and defect engineering.

Methods
Materials synthesis. The precursor Ni 0.25 Mn 0.75 CO 3 was synthesized by the coprecipitation method. First, a certain amount of MnSO 4 ·5H 2 O and NiSO 4 ·6H 2 O were dissolved into 100 ml of deionized water, and then 50 ml of Na 2 CO 3 (2 M) was slowly added to the above mixed solution and stirred for 12 h. Finally, the stirred solution was washed three times with deionized water and twice with ethanol and placed in an oven to dry at 80°C. To obtain the pristine sample, an excess of 5% LiOH·H 2 O was mixed with the corresponding proportion of the precursor. Then, the mixture was transferred to a crucible and heated to 500°C at a rate of 2°C/min in a muffle furnace, kept isothermal for 300 min, heated at 5°C/ min to 900°C for 720 min, and finally naturally cooled to room temperature. For the modified Li 1.2 Mn 0.6 Ni 0.2 O 2-δ sample, the synthesized pristine sample was dispersed in different concentration of hydrazine hydrate solutions and stirred for different times. Then, the mixed solution was washed twice with deionized water by centrifugation and twice with ethanol. Finally, the sample was dried in an oven at 80°C and then dried in a vacuum oven at 120°C for 12 h.
Electrochemical measurements. The prepared material was mixed with polyvinylidene fluoride (PVDF, Arkema, French) and carbon black (Lion Specialty Chemicals Co., Ltd) at a weight ratio of 80%:10%:10%, the mixture was ground with agate mortar for 10 ± 2 min in an air environment, and N-methyl pyrrolidinone (Shanghai Macklin Biochemical Co., Ltd.) was added during the grinding process. The obtained slurry was coated on aluminium foil (Shenzhen Kejing Star Technology Co., Shenzhen) at a thickness of 16 μm and with a purity of more than 99.35%. Next, the sample was dried for 12 h under vacuum at 120°C. The active material mass, diameter and thickness of the positive electrode were 2-3 mg/cm 2 , 9 mm and 20-30 μm, respectively. Lithium metal with a purity greater than 99.9% was used as the counter and reference electrode, and its thickness and diameter was 0.45 mm and 15.6 mm, respectively (China Energy Lithium Co., Ltd. Tianjin). The assembly and disassembly of the CR2025-type coin cells were carried out in a glove box filled with argon gas and the content of H 2 O and O 2 is less than 0.1 ppm (Etelux Inertia Gas System (Beijing) Co., Ltd.). A Whatman GF/D glass microfibre filter was used as the separator. A highvoltage electrolyte was purchased from Beijing Institute of Chemical Reagents, and the typical electrolyte formula was LiPF 6 (1 M) in a 1:1 volume ratio of fluoroethylene carbonate and dimethyl carbonate. The amount of electrolyte for each cell was 60-80 μL. PITT was performed with an electrochemical workstation (Metrohm-Autolab, PGSTAT 302N). The sample was gradually charged from OCV to 4.8 V with a voltage step of 0.1 V. During this process, charging occurred at a constant voltage for 1800 s, and was then relaxed for the same amount of time to obtain the chronoamperometric curve (I-t). Galvanostatic charge-discharge was performed from 2 to 4.8 V at specific currents with a battery tester (NEWARE CT-4000) in an environmental chamber (22 ± 3°C).
Physicochemical characterizations. The crystal structures of materials were characterized by performing powder XRD with a Smartlab XRD (Rigaku, a CuKα radiation source) in a 2 theta range of 10-80°. The in situ XRD electrode was assembled in a cell mould with a beryllium window (Beijing Scistar Technology Co., Ltd), the positive electrode without a current collector is cut into a 1 × 1 cm 2 square, and the counter electrode and separator were the same as the coin cells. The specific current was 25 mA/g for the in situ XRD cell, and data were collected every 15 min. Ex situ TEM (Tecnai G2 F30) was used to observe the CEI and morphology after cycling. Ex situ spherical aberration-corrected transmission electron microscopy (JEM ARM200F, JEOL, Tokyo, Japan) was used to observe the atomic structure of materials at different states. Ex situ XPS with a monochromatic X-ray source (Al Kα) is used to detect the chemical state and electronic state of the material in different states. Ex situ 7 Li NMR spectra were collected by a solid-state NMR Spectrometer (Bruker AVANCE III, 400 MHz) under MAS rate of 12.8 kHz. For the ex situ measurement, the cells were disassembled, washed with dimethyl carbonate, packaged in aluminiumplastic bags in an Ar-filled glove box, and then brought to the testing site for rapid sampling loading and testing. The overall exposure time in air is less than 2 min. Oxygen production was recorded by operando DEMS (Linglu Instruments Co., Ltd. Shanghai). The DEMS cell was assembled with a Swagelok-type cell, where the diameter and thickness of electrode disc were 12-19 mm and 30-50 μm, respectively. The assembled cell was connected to the gas path of the mass spectrometer (Pfeiffer, OminiStar GSD 320). Note that the total carrier gas (Ar) was 3 mL/min, and the flow was 0.5 mL/min through the Swagelok cell. The cell was continuously ventilated for 10-15 h until the baseline was stable and then charged and discharged at a specific current of 25 mA/g. The overall operating diagram and Swagelok cell are shown in Supplementary Fig. 11a, b. Synchrotron radiation characterization. Hard XAS was conducted at the KMC2 beamline of the synchrotron BESSY II at Helmholtz-Zentrum Berlin für Materialien und Energie (HZB, Germany) 54 . O K-edge soft XAS was performed at the Russian-German Beamline in the TEY mode at HZB, the O K-edge spectra in the FY mode were collected at beamline BL08U1-A in the Shanghai Synchrotron Radiation Facility (SSRF), and all other TM L-edge spectra in the TEY and FY modes were tested at beamline 20A1 of the Taiwan Light Source (TLS) at HsinChu, Taiwan, China. The RIXS maps and spectra of O and Mn were recorded at the U41-PEAXIS beamline at HZB 55,56 .
Neutron powder diffraction. Data was obtained at the Vulcan beamline spallation neutron source (SNS) at Oak Ridge National Laboratory 57 . The size of the incident beam was 5 mm × 12 mm, the diameter of the receiving collimator was 5 mm, and the bandwidth of the incident electron beam was 0.7-3.5 Å. Double-disk choppers at a speed of 30 Hz were employed, and a 0.5-2.5 Å d-spacing was allowed in the diffraction pattern of the θ ± 90 detector banks. In high resolution mode, Δd/d was~0.25%. The power of SNS was 1400 kW. At a temperature of 25°C, NPD data acquisition took 3 h, and VDRIVE software was used to reduce the data. A full-pattern Rietveld refinement was implemented using the GSAS program with an EXPGUI interface 58 .
DFT calculation. We used the hexagonal Li 3 Mn 3 O 6 structure with the R-3m space group as the initial structure 59 . We built a 3 × 3 superstructure of Li 3 60 . The generalized gradient approximation (GGA) with the Perdew-Burke-Ernzerhof (PBE) functional was used to describe the electronic exchange and correlation effects [61][62][63] . Uniform G-centred k-point meshes at a resolution of 2π × 0.04 Å −1 and Methfessel-Paxton electronic smearing were adopted for integration in the Brillouin zone for geometric optimization. The simulation was run with a cut-off energy of 500 eV throughout the computations. These settings ensured convergence of the total energies to within 1 meV per atom. Structural relaxation proceeded until all forces on the atoms were less than 1 meV Å −1 and the total stress tensor was within 0.01 GPa of the target value. To describe the on-site Coulombic interaction, the DFT + U approach was used to calculate all the elementary reaction steps, and the U terms of Mn-3d and Ni-3d were 3.9 and 6.4 eV, respectively. The octahedral MnO 6 distortion (4) was calculated by the equation 4 ¼ 1 6 ∑ R i ÀR av R av 2 , where R av is the average bond length and R i is the individual bond length of Mn−O in the octahedral MnO 6 . The migration barriers of Mn were calculated using the climbing image nudged elastic band (ciNEB) method 64 .
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request. The curves and graphs of the data generated in this study have been deposited in the figshare database under accession code https:// figshare.com/10.6084/m9.figshare.18093398. Source data are provided with this paper.