Stabilization of O–O Bonds by d0 Cations in Li4+xNi1–xWO6 (0 ≤ x ≤ 0.25) Rock Salt Oxides as the Origin of Large Voltage Hysteresis

Multinary lithium oxides with the rock salt structure are of technological importance as cathode materials in rechargeable lithium ion batteries. Current state-of-the-art cathodes such as LiNi1/3Mn1/3Co1/3O2 rely on redox cycling of earth-abundant transition-metal cations to provide charge capacity. Recently, the possibility of using the oxide anion as a redox center in Li-rich rock salt oxides has been established as a new paradigm in the design of cathode materials with enhanced capacities (>200 mAh/g). To increase the lithium content and access electrons from oxygen-derived states, these materials typically require transition metals in high oxidation states, which can be easily achieved using d0 cations. However, Li-rich rock salt oxides with high valent d0 cations such as Nb5+ and Mo6+ show strikingly high voltage hysteresis between charge and discharge, the origin of which is uninvestigated. In this work, we study a series of Li-rich compounds, Li4+xNi1–xWO6 (0 ≤ x ≤ 0.25) adopting two new and distinct cation-ordered variants of the rock salt structure. The Li4.15Ni0.85WO6 (x = 0.15) phase has a large reversible capacity of 200 mAh/g, without accessing the Ni3+/Ni4+ redox couple, implying that more than two-thirds of the capacity is due to anionic redox, with good cyclability. The presence of the 5d0 W6+ cation affords extensive (>2 V) voltage hysteresis associated with the anionic redox. We present experimental evidence for the formation of strongly stabilized localized O–O single bonds that explain the energy penalty required to reduce the material upon discharge. The high valent d0 cation associates localized anion–anion bonding with the anion redox capacity.


■ INTRODUCTION
Li-containing rock salt oxides form one of the most studied families of positive electrodes for rechargeable lithium ion batteries. Following the success of the first commercial Li-ion systems using LiCoO 2 as a cathode material, extensive exploration of other LiMO 2 analogues has led to the discovery of higher-performing materials, among which are the current state-of-the-art NMC phases (LiNi x Mn y Co 1−x−y O 2 ). The maximum capacity of these materials is, however, still limited because of the 1:1 Li:M ratio, and great hope is being placed in the "Li-rich" rock salt oxides (Li x M y O x+y with x > y) which can be accessed by partial substitution of the transition metal with excess Li associated with an increase in the mean metal oxidation state above +III (e.g., Li(Li 1/3 Mn 2/3 )O 2 or Li 2 MnO 3 ). Such compounds can deliver capacities exceeding 200 mAh/g, making them promising cathode materials for future Li-ion rechargeable batteries. These capacities exceed expectations based on purely metal-based redox within the expected manifold of oxidation states in cathode materials with Li/M > 1. This is now well understood as the contribution of oxygen to the redox activity, made possible by the presence of nonbonding oxygen states close to the Fermi level in Li-rich rock salt oxides because some O 2p orbitals do not take part in the formation of M−O bonding and antibonding orbitals. 1−3 Oxygen-based redox reversibility can be enhanced by moving from 3d to 4d and 5d cations, which avoids the formation of gaseous O 2 upon anion oxidation. 4 Chemistries that enable reversible oxygen oxidation with earth-abundant components would bring these capacities within practical reach.
Despite their large diversity of structures and compositions, few Li-rich rock salt oxides have been studied as cathode materials. By carefully selecting the transition metals used, these compounds can exhibit complex cation-ordered superstructures 5−8 and their Li content can be varied extensively (LiMO 2 → Li 2 MO 3 → Li 3 MO 4 → Li 4 MO 5 ...), 9,10 resulting in a considerable diversity of candidates for new cathode materials and a great opportunity to explore the role played by different cations in triggering oxygen redox. Following this idea, Yabuuchi and co-workers have shown the possibility of using a highvalence d 0 cation such as Nb 5+ or Mo 6+ to increase the Li content in rock salt oxides, create nonbonding oxygen states close to the Fermi level and activating the redox activity of oxygen. 11,12 Interestingly, several of these materials show extremely large voltage hysteresis with most of the discharge capacity recovered below 2 V, but its origin has not been investigated so far. Although such low energy efficiency is detrimental to their use in practical Li-ion batteries, understanding the origin of this feature will help in the design of future cathode materials because the voltage hysteresis issue is pervasive in materials showing anionic redox activity, including the long-sought-after Li-rich NMC phases. 13 −15 In this work, we investigate a family of Li-rich rock salt oxides Li 4+x Ni 1−x WO 6 (0 ≤ x ≤ 0.25) derived from the Li 4 NiWO 6 phase reported by Mandal et al. 16 and vary the Li/Ni ratio to tune the structural and electrochemical properties of the material. Using a combination of spectroscopic techniques (XPS, XAS, and Raman), the role of Ni and O in the charge compensation during electrochemical (de)lithiation was investigated, revealing the influence of redox-inactive W 6+ (5d 0 ) cations on the formation of stable, discrete O−O bonds upon oxidation of oxide and the associated voltage hysteresis.

■ RESULTS
Synthesis and Structure of Li 4+x Ni 1−x WO 6 Materials. The initial synthesis experiments to prepare the stoichiometric Li 4 NiWO 6 phase were carried out by a classical solid-state method using NiO, WO 3 , and a 10% molar excess of Li 2 CO 3 (1:1:4.4) to compensate for Li 2 O volatility during hightemperature synthesis (Methods). This resulted in a darkbrown powder which initially appeared pure from laboratory PXRD. However, upon further analysis by synchrotron PXRD, neutron powder diffraction, and TEM-EDX, the sample turned  4 NiWO 6 , the cubic unit cell of the NaCl structure is shown in addition to the C2/c cell to highlight the rock salt ordering of the structure. Both structures can be described by a staggered stacking sequence of a unique layer with a similar ordering of the highest oxidation state cation (W 6+ in orange and Ta 5+ in gray). The unique layer for each structure is rotated by 90°and displayed in the lower part of the figure. The different contents of Ni (blue) and Li (green) coupled with the off-center displacement of W 6+ result in different orderings of the remaining cation sites in the two materials, with site 2 (site 3) of the Fddd structure being split into sites 2 and 6 (sites 3 and 5) of the C2/c structure. The partial occupation of the different cation sites is indicated in the figure.

Journal of the American Chemical Society
Article out to be nonstoichiometric and contain a small amount of cubic Li 1−x Ni x O impurity, thus providing evidence for the existence of a solid solution of the form Li 4+x Ni 1−x WO 6 . By revising the initial synthesis conditions, using different Ni/W ratios and sacrificial powder to better control the Li stoichiometry (Methods), we were able to obtain the stoichiometric pale-green Li 4 NiWO 6 phase without impurity as well as the Li 4+x Ni 1−x WO 6 solid solution with 0.05 ≤ x ≤ 0.25. The compositions of the different phases were confirmed by ICP−OES and TEM−EDX analysis (Note 1 in Supporting Information (SI)). Although all members of the nonstoichiometric Li 4+x Ni 1−x WO 6 solid solution could be approximately indexed with the structural model reported by Mandal et al., 16 the diffraction pattern of the stoichiometric Li 4 NiWO 6 (x = 0) could not be matched with the reported structure and turned out to be a new structure type. A thorough structural investigation was therefore undertaken on the Li 4 NiWO 6 and Li 4.1 Ni 0.9 WO 6 samples using synchrotron X-ray and neutron powder diffraction data (detailed structural analysis in Notes 2 and 3 in the SI).
The structure of the stoichiometric Li 4 NiWO 6 phase was solved in the C2/c space group (a = 5.84579(10) Å, b = 17.58769(35) Å, c = 5.109138(9) Å, β = 124.768(1)°, V = 431.506(2) Å 3 ; see Table S1 in Supporting Information for full structural parameters) and can be thought of as a monoclinic distortion of the Li 3 Ni 2 TaO 6 (Fddd) archetype based on the same ordering pattern of the almost fully occupied W sites, which follows that of Ta in the archetype (Figure 1b) but a different Ni/Li ordering. 17 The relationship between the two cells is shown in Figure S1 in Supporting Information. Pauling's rule of electroneutrality prevents two highly charged W 6+ cations from being coordinated to the same oxygen. In other words, W 6+ cannot be situated in the first two cation neighboring shells of any W 6+ cation at translations T 1 = (a/2, a/2, 0) and T 2 = (a, 0, 0), where a is the rock salt cubic cell dimension. As a result, they form isolated WO 6 octahedra arranged to minimize electrostatic repulsions between W 6+ cations, separated by T 3 = (a, a/2, a/2) (Figure 3a), adopting 1 of the 10 such arrangements noted by Hauk et al. 18 As a consequence, all oxygen atoms in the structure are coordinated to only one W atom. Li 3 Ni 2 TaO 6 and the new Li 4 NiWO 6 structure can be described as a staggered six-layer repeat sequence of a single mixed cation layer with six distinct octahedral sites (Figure 1, bottom). The reduction in symmetry from Li 3 Ni 2 TaO 6 to Li 4 NiWO 6 , with a lower Ni content and the higher charge of the d 0 W 6+ cations, arises from an enhanced Li/ Ni ordering, which changes the number of distinct cation sites from four to six in Li 4 NiWO 6 by splitting site 2 (site 3) of the Fddd structure (Figure 1b) into sites 2 and 6 (sites 3 and 5) of the C2/c structure (Figure 1a). The W 6+ cations in Li 4 NiWO 6 are almost fully ordered onto site 1, with 4.7% Li + on this site and the remaining 4.7% W 6+ distributed between two of the remaining five cation sites (sites 4 and 6). The Ni 2+ occupancy of all other sites is driven by their interaction with highly charged W 6+ cations in this particular arrangement. Sites 2 and 6 have only five W 6+ cations in the three neighboring cation shells at T 1 , T 2 , and T 3 (Table S2 in Supporting Information), compared to eight for the other sites, making them more likely to contain Ni 2+ as a result of lower electrostatic repulsion. The d 0 configuration of W 6+ leads to asymmetric displacement from the center of its octahedron, moving away from site 6 and toward the edge shared with site 2 (Tables S3 and S4 in Supporting  Information), attributed to second-order Jahn−Teller effects. 19,20 This destabilizes one site at the expense of the other, leading to a high occupation of site 6 by Ni 2+ (57%) and a low occupation of site 2 (8.9%). For the same reason, site 3 is further away from the W 6+ cations and has a higher Ni 2+ content (27.6%) compared to that of site 5 (4.2%). This contrasts with the nearly statistical distribution of cations in sites 2 and 3 in Li 3 Ni 2 TaO 6 , where the d 0 cation has a lower charge and is not displaced from the octahedron centroid. Finally, site 4 is almost fully occupied by Li + (94%) as a result of electrostatic repulsion from its W 6+ neighbors and the large distortion enforced by the two edge-sharing and two corner-sharing neighboring WO 6 octahedra ( Figure S2 and Table S4 in Supporting Information), which would be less favorable to high-spin Ni 2+ than the other less distorted sites. To obtain the final Rietveld fit (Figure 4a), it was necessary to include an Fddd minority Li 4 NiWO 6 phase isostructural with Li 3 Ni 2 TaO 6 of refined weight fraction 11.8(6) %, which can be interpreted as either regions of lithium and nickel site disorder within the material or a consequence of small Li/Ni ordered domain sizes resulting in a significant antiphase boundary content. The asymmetric displacements of W atoms occur in opposing directions along [010] and [01̅ 0] axes, resulting in no overall dipole in the material (Figure 5a). Such displacements are absent (or disordered) in the less ordered Fddd analogue, where the average positions of the tungsten sites are located at the centroids of the octahedra at Wyckoff site 8a (which has point symmetry 222 and thus no dipole).

Journal of the American Chemical Society
Article an increased Li content, Li 4.1 Ni 0.9 WO 6 moves further away from the Li 3 Ni 2 TaO 6 stoichiometry and adopts a monoclinic structure with four distinct cation sites, similar to the archetypal layered Li 5 ReO 6 rock salt superstructure. 21 The WO 6 octahedra in Li 4.1 Ni 0.9 WO 6 are again isolated from each other, with similar W−W distances to Li 4 NiWO 6 ( Figure 3) that also correspond to the third cation shell at T 3 = (a, a/2, a/2) in order to minimize electrostatic repulsions but in a motif distinct from that found in Li 4 NiWO 6 . Cations order through the adoption of a two-layer structure where (Li/Ni)O 6 octahedra surround WO 6 units in a "honeycomb" arrangement in one layer (layer 1 in Figure 2), which alternates with a mixed Li/Ni layer (layer 2 in Figure 2).
Ni is distributed over three of the four cation sites (sites 2, 3, and 4), with a preference for site 2 in the W layers. Site 2 has the lowest electrostatic repulsion from W 6+ cations, with only five W 6+ in the first three cation shells compared to eight for sites 3 and 4 (Table S6 in Supporting Information). This Li/Ni ordering is similar to that in related Li 4 NiMoO 6 but differs from that in Li 4 NiTeO 6 , in which Ni is localized only at site 2. 12,22 The similar W−W network and influences on Li/Ni order suggest that the two structures are close in energy, with the increased Li content in Li 4.1 Ni 0.9 WO 6 favoring the layered structure of Li-rich Li 5 ReO 6 . Furthermore, oxygen atoms in both structures have similar coordination environments, each belonging to one WO 6 octahedron, although the partial Ni/Li ordering in the two materials could lead to several local coordination configurations around oxygen (OWNiLi 4 , OWNi 2 Li 3 , or OWLi 5 ), which will necessarily have an effect on the electrochemical properties. These two variants of W ordering derived from the C2/m and Fddd space groups have also been reported for Os in Li 4 MgOsO 6 and Re in Li 4 MgReO 6 , 23,24 depending on synthesis conditions. As for the Li 4+x Ni 1−x WO 6 system studied here, the difference between the two variants in these systems could be related to a tight balance in the Li/Mg ratio. The Rietveld model in C2/m gave a poor fit to the SXRD and NPD in comparison to the Le Bail fit (Note 3 in SI). The symmetry was reduced to the noncentrosymmetric Cm by the displacement of the tungsten site off-center toward one apex of the WO 6 octahedron ( Figure  5b and Table S7 in Supporting Information), thereby removing the 2-fold rotational axis, maintaining the total number of cation sites at four but doubling the number of anion sites from two (Wyckoff sites 8j and 4i, point symmetry 1 and m respectively) to four (sites 4b and 2a, point symmetry 1 and m, respectively). Here, the off-centering of W differs from that in the C2/c Li 4 NiWO 6 phase, and the displacement propagates ferrodistortively throughout the lattice, resulting in an overall permanent dipole. The final structural parameters of the combined SXRD and ND Rietveld refinement (Figure 4b), including the 2.7(2) wt % Li 0.3 Ni 0.7 O impurity, can be found in Table S5 in Supporting Information with the final cell parameters (a = 5.113747(23) Å, b = 8.791326(40) Å, c = 5.093213(23) Å, β = 110.1564(12)°). While exploring the Li 4+x Ni 1−x WO 6 solid solution, we found that the same Cm structure is preserved up to x = 0.25 and observed a linear decrease in lattice parameters with increasing value of x ( Figure  S3 in Supporting Information). This is consistent with the increasing content of smaller and more highly charged Ni 3+ when Ni is substituted by Li. For x ≥ 0.2, impurity reflections corresponding to Li 4 WO 5 were observed, suggesting that the limit of solid solution lies between nominal values of 0.15 ≤ x ≤ 0.20.
Electroche mical Properties. Li 4 NiWO 6 and Li 4.15 Ni 0.85 WO 6 (x = 0.15) samples were chosen for initial electrochemical studies in order to maximize the differences between the two structures and compositions while avoiding the presence of the Li 4 WO 5 impurity for samples with x ≥ 0.2. Ball milling of the active materials to reduce the particle size from 5−20 μm to 0.5−3 μm ( Figure S4 in Supporting Information) was found to help both in the preparation of the casting and in achieving better electrochemical performance. The first charge/ discharge cycles for the two materials are compared in Figure 6a. Both materials show very high capacities of 270 mAh/g when charged to 5 V, with some irreversible capacity on discharge. This is more pronounced for Li 4 NiWO 6 , whose reversible capacity is less than 140 mAh/g on the first cycle (52% Figure 3. W ordering. The arrangement and distances between W atoms in Li 4 NiWO 6 (a) and Li 4.1 Ni 0.9 WO 6 (b). The distances (in Å) from the central W atom are indicated next to each atom. The two structures have similar W−W distances, corresponding to the translation vector T 3 = (a, a/2, a/2), where a is the dimension of the rock salt cubic cell shown in the figure. However, the ordering differs between the two structures, resulting in a layered W ordering in Li 4.1 Ni 0.9 WO 6 (b), highlighted by the orange planes, and a tridimensional W ordering for Li 4 NiWO 6 (a), with some of the W atoms situated between the gray planes.

Journal of the American Chemical Society
Article reversibility) and decreases to 110 mAh/g on the 10th cycle, whereas a higher reversible capacity was observed for Li 4.15 Ni 0.85 WO 6 with 200 mAh/g on the first discharge (74% reversibility), decreasing to 150 mAh/g after 10 cycles ( Figure  6b). Except for the lower reversible capacity of the stoichiometric phase, which could simply be related to the higher amount of Ni it contains, both materials show very similar features on charge/discharge, suggesting similar redox processes. This is particularly clear when comparing the differential capacity curves of the two materials for the first two cycles ( Figure S5 in Supporting Information). Because Li 4.15 Ni 0.85 WO 6 displays a higher capacity and its structure is comparable to that of previously investigated materials, 12,22 further characterizations were performed on this sample only. Figure 6c shows the differential capacity curve of the first and second cycles of Li 4.15 Ni 0.85 WO 6 , with three redox processes occurring at 3.7, 4.2, and 4.7 V vs Li + /Li on the first charge and most of the redox activity on discharge happening at 1.7 V. Such a large voltage hysteresis (2 to 3 V) is uncommon for rock salt oxide cathode materials and was therefore investigated further. In the next cycles, different processes are observed, with ∼0. 35 Li + deintercalated at 2 V and the remaining 2.25 Li + at 3.7 and 4.4 V. The irreversible capacity on the first cycle and subsequent modification of the voltage−capacity curve indicate a change in the material upon initial charging, similar to the first activation cycle of most Li-rich rock salt oxides, including Li-rich NMC, 25 which could be related to a rearrangement of cations or oxygen release from the material. To understand the origin of the irreversibility, several cells were cycled with different maximum cutoff voltages, as shown in Figure 6d, with the corresponding differential capacity curves in Figure S6 in Supporting Information. For a 4 V cutoff, less than 100 mV polarization is observed between the oxidation and reduction processes, and the discharge capacity of 100 mAh/g slightly exceeds the charge capacity of 85 mAh/g, suggesting that extra lithium is inserted upon reduction of the material at low voltage. Excess electrochemical insertion of lithium is not uncommon in rock salt oxides 3,26,27 and can be expected here by the partial

Journal of the American Chemical Society
Article o x i d a t i o n o f n i c k e l i n t h e p r i s t i n e m a t e r i a l (Li 4.15 Ni 3+ 0.15 Ni 2+ 0.7 WO 6 ), which upon reduction can lead to the insertion of an extra 0.15Li + . After charging at 4.5 V with a capacity of 225 mAh/g, several broad reduction peaks are spread between 3.4 and 4.4 V, and an intense peak appears at 1.7 V corresponding to the low-voltage plateau described previously. The modification is even more obvious when the material is fully charged to 5 V, with only a broad reduction feature left at 4 V and most of the capacity recovered on the 1.7 V plateau. One can note the appearance of two small features at 2.2 and 2.8 V (inset of Figure S6 in Supporting Information) when the sample is charged at 4.5 V, which could indicate a small fraction of cations (Ni or W) in different coordination environments. More importantly, it clearly appears that the irreversible capacity in the first cycle is due to the redox process at 4.2 V.
Ex situ samples were prepared along the first two cycles, and their XRD patterns were measured to look for structural changes during cycling (Figure 7b). In addition to significant broadening of the peaks, a reversible shift of the peak positions is observed upon charge and discharge. Superstructure peaks are preserved, suggesting that the W ordering is not affected, and all patterns can be indexed with the same Cm space group as the pristine material using the LeBail method, with the refined lattice parameters shown in Figure 7a and listed in Table S8 in Supporting Information. At 4 V, the slight decrease in in-plane lattice parameters a and b and increase in out-of-plane parameter c are consistent with the contraction of the honeycomb layer containing the majority of Ni (layer 1 in Figure 2) upon Ni oxidation and removal of Li + from the interlayer space. When charged further, all parameters decrease, with a large jump between 4.5 and 5 V, leading to a total contraction of the unit cell volume of 7%. The cell size increases back to close to its initial value on the next discharge, with slight differences in cell parameters that reflect the nonreversibility of the redox process at 4.2 V on initial charging. A similar evolution is observed in the second cycle with a less pronounced contraction at 5 V. Surprisingly, the most drastic volume change happens between 4.5 and 5 V on charge and between 5 and 3 V on discharge, whereas this corresponds to less than 0.5Li + (de)inserted from the material, suggesting that another structural process participates in the substantial contraction of the unit cell.
Charge Compensation Mechanism. The high initial charge capacity (270 mAh/g) and reversible discharge capacity (200 mAh/g) cannot be explained solely by the Ni 2+ /Ni 4+ redox couple, which can only provide up to 116 mAh/g on the first cycle ( Ni is partially oxidized in the p ristine Li 4.15 Ni 3+ 0.15 Ni 2+ 0.7 WO 6 ) and 127 mAh/g on the next cycles. To understand the role played by Ni, W, and O in the different electrochemical processes, we measured selected ex situ samples by several spectroscopic techniques. X-ray absorption spectros-

Journal of the American Chemical Society
Article copy at the W L 3 -edge ( Figure 8a) shows a very slight change in shape of the absorption peak from 4 to 4.5 V that cannot be assigned to a change of oxidation state but rather to a distortion of the W local environment. This is consistent with an irreversible transformation of the material associated with the 4.2 V process, as the double peak initially observed seems to coalesce and remains unchanged on further cycling. The Ni Kedge was used to probe the activity of Ni, with NiWO 4 and LiNiO 2 as standards for +2 and +3 oxidation states, respectively. The edge position of the pristine sample is similar to that of the NiWO 4 standard (Figure 8b), as expected on the basis of the predominance of the Ni 2+ oxidation state in the starting material. Upon charging to 4 V, the edge gradually shifts to higher energies and finally reaches the edge position of the LiNiO 2 standard at 4.5 V, indicating the oxidation of Ni 2+ to Ni 3+ . Slight changes can be more easily visualized by plotting the difference between two successive spectra ( Figure S7 in Supporting Information). Further oxidation to 5 V does not affect the edge, which suggests that the Ni oxidation is limited to Ni 3+ , making Li 4.15 Ni 0.85 WO 6 a rare example of a cathode material where the average oxidation state of nickel is considerably less than +4 after charging to 5 V. This limitation cannot be attributed to reduced sensitivity of the Ni K-edge to oxidation states higher than 3+ as several studies have shown a clear shift between trivalent and quadrivalent nickel compounds, 28,29 including an in situ study of the Li 1−z Ni 1+z O 2 (z ≤ 0.02) cathode. 30 On discharge, Ni is partially reduced back to an oxidation state between Ni 2+ and Ni 3+ at 3 V and to Ni 2+ when discharged to 1 V. During the second cycle, the oxidation of Ni seems to start at a higher voltage, between 4 and 4.5 V (Figure 8c), but does not reach the full Ni 3+ oxidation state. This is possibly a consequence of structural reorganization during the first cycle that leaves some Ni 2+ ions inactive in the material, resulting in a broadening of the Ni K-edge maximum but no clear shifts of the edge position.
At this point, it is very surprising that only the Ni 2+ /Ni 3+ redox couple is active in the material because it should theoretically account for about 64 mAh/g whereas the reversible capacity reaches 200 mAh/g, leaving two-thirds of the capacity unexplained. To investigate the possible participation of oxygen redox in the charge compensation mechanism, O 1s XPS spectra were collected to probe the presence of oxidized oxygen species. Figure 9b compares the XPS results for samples charged and discharged at different voltages on the first cycle. Up to 4.5 V, the spectra are dominated by the signal of oxygen O 2− at 530.6 eV and show little variation, with only a weak contribution at higher binding energy for 4 and 4.5 V. This feature appears very clearly for the sample charged at 5 V as an intense contribution at higher binding energy (532.2 eV), which can be explained by the formation of oxidized oxygen species, written as O n− (0 ≤ n < 2), until more information about their chemical nature is obtained as discussed below. Other contributions arising from deposited species are also very intense for this sample (∼ 534 eV) because we expect some electrolyte decomposition at the cathode surface at high voltage. Upon discharge, the relative intensity of the O n− peak decreases compared to that of the O 2− one but does not fully disappear, suggesting that some of the O n− species formed are not reduced back to oxide. In the second cycle ( Figure S8 in Supporting Information), the contribution at 532.2 eV increases again at 4.5 V, a lower voltage compared to that in the first cycle. The same observations can be made by monitoring the W 4f doublet position, which is very sensitive to changes in the local environment of W ions (Figure 9b). At 5 V, the doublet shifts to higher binding energy, suggesting an increase in the ionicity of W−O bonds with decreasing Li

Journal of the American Chemical Society
Article content, and a new doublet appears, whose evolution closely follows that of the O n− peak in the O 1s spectra. This new doublet is therefore assigned to a new W environment coordinated by the oxidized O n− species, and its relative intensity is correlated with the amount of oxidized oxygen next to the surface of the sample. To strengthen this assignment, we measured the XPS spectra of two reference samples with W 6+ cations in different coordination environments, namely, WO 3  confirming that W is in the +6 oxidation state at all points along the charge/discharge curves. It is quite unexpected that most oxidized oxygen species appear above 4.5 V on the first charge, whereas Ni does not oxidize beyond +3 according to XAS, leaving a gap in our understanding of the charge compensation mechanism between OCV and 4.5 V. Because XPS is a surfacesensitive technique and mostly probes the first 5 nm below the surface, it is possible that oxygen is oxidized at lower potential in the bulk of the material, as is suggested by the low-intensity   31 and Raman spectroscopy (light green in Figure 10b). The most intense peak at 983 cm −1 is generally assigned to the short W−O bond stretching mode, 33 and the less intense peak at 920 cm −1 is assigned to the stretching of the peroxo O−O bond. Computed Raman spectra (dark green) using DFT calculations can fairly well reproduce these two features (961 and 939 cm −1 , respectively) and further demonstrate that both peaks correspond to combinations of stretches of the short W−O and peroxo O−O bonds. This result shows that the presence of a peroxo bond does not necessarily result in one but possibly several vibrational modes, which is the case for Na 2 O 2 , MgO 2 , and ZnO 2 , 34 and that the DFT calculation can provide decisive insight for the correct assignment of the Raman peaks.
To check the formation of O−O bonds in the bulk structure of Li x Ni 0.85 WO 6 , we recorded Raman spectra of ex situ samples at different states of charge ( Figure 10a) and compared them to those obtained by DFT calculations on two model systems Li 3 Ni 3+ WO 6 and Li 3 Ni 2+ WO 5 (O 2 2− ) 1/2 with delithiation mediated by Ni redox and O redox, respectively (Note 4 and Figure S9 in Supporting Information). The pristine material has one main broad Raman band at 810 cm −1 corresponding to the W−O stretching mode and less intense peaks (327, 460, 515 cm −1 ) at lower frequency, where one expects bending and deformation modes. Upon charging to 4 V, the intensity of the main peak decreases and a sharper peak appears at 930 cm −1 , joined by a second peak at 910 cm −1 at 4.5 V. Finally, the fully charged sample (5 V) shows a very well defined Raman spectrum, with relatively sharp and intense peaks, suggesting that the local structure of the fully charged sample is fairly ordered. Focusing on the peaks in the 800−1000 cm −1 region in the fully charged sample, we observe a very intense peak at 890 cm −1 and a second peak at 930 cm −1 that appears as a shoulder. The peak at 930 cm −1 is then present in all subsequent spectra, including the fully discharged samples, whereas the most intense band progressively disappears again upon discharge and is replaced at 1 V by a broader band centered at 830 cm −1 . In the second cycle, a perfectly reversible evolution is observed, consistent with the improved electrochemical reversibility after the first cycle. The position of the peak at 930 cm −1 is in excellent agreement with that at 920 cm −1 in WO 2 (O 2 )H 2 O and could result from the formation of a peroxo bond. Calculation of the Raman modes was therefore performed with DFT to confirm

Journal of the American Chemical Society
Article the assignment of the peaks, using the pristine structure and two models of delithiation mediated by Ni redox (blue spectra in Figure 10b) and O redox (red spectra) respectively, the latter being characterized by the formation of a peroxo (O 2 ) 2− unit (d O−O = 1.46 Å, Figure S10 in Supporting Information). The calculated spectra are shown in Figure 10b 35,38 highlighting that different processes can happen at the surface and in the bulk of the particles. The frequencies we find are closer to the frequencies encountered for MgO 2 (864 and 934 cm −1 ) and ZnO 2 (847 and 944 cm −1 ) compared to those of Li 2 O 2 (790 cm −1 ) and Na 2 O 2 (736 and 791 cm −1 ). 34 Such differences are likely to arise from different O−O bond orders, which may greatly vary depending on the metal center: here we emphasize the role of W 6+ in coordinatively stabilizing the peroxide species. The agreement between experimental and calculated Raman spectra, and more importantly the comparison with the structurally and spectroscopically well-defined hydrated tungsten peroxide WO 2 (O 2 )H 2 O, provides strong evidence for the formation of localized peroxo oxygen−oxygen single bonds upon full charge of Li 4.15 Ni 0.85 WO 6 . This is consistent with the oxidized oxygen signal obtained from XPS, which also agrees with measurements on the same peroxidecontaining WO 2 (O 2 )H 2 O standard, and shows that the redox of oxygen is not only happening at the surface but is indeed a bulk effect. The formation of quasi-molecular (O 2 ) 2− species explains the shift of the W 4f doublet at higher binding energy ( Figure  9b) with the increased ionicity of W-(O 2 ) 2− bonds compared to W−O 2− bonds. This can also explain the strong contraction of the unit cell observed by diffraction at the end of charge ( Figure   7a) in which the formation of peroxo bonds would be expected to strongly perturb interatomic distances in the material. Finally, the fact that the peak at 930 cm −1 does not completely disappear at the end of the first cycle is also consistent with the XPS data and indicates that some peroxo species are formed irreversibly and maintained through subsequent cycling, thus explaining some of the irreversible capacity of the first cycle.

■ DISCUSSION
By combining different techniques, we have obtained some insights into the electrochemical behavior of Li 4.15 Ni 0.85 WO 6 . One of the most surprising results is the impossibility of accessing the Ni 3+ /Ni 4+ redox, which suggests close competition between the redox activities of Ni and O. This results stands out from the behavior of the related Li 4 NiTeO 6 , which uses the full Ni 2+ /Ni 4+ redox couple with less than 200 mV hysteresis between charge and discharge. 22 Moving from a main group cation (Te 6+ ) to a d 0 cation (Mo 6+ , W 6+ ) as non-redox-active elements not only affects the cation ordering in the structure but also drastically changes the electrochemical properties of the material. Here, the activation of redox of oxygen-based states is demonstrated by the alignment of both Raman and XPS spectra of electrochemically delithiated samples with tungsten peroxide standard WO 2 (O 2 )H 2 O. DFT calculations were carried out to investigate the electronic structure of delithiated Li 4 NiWO 6 using both the VASP PBE+U and HSE06 total energies. (See Note 4 in SI for details on the calculation.) An idealized cationordered Li 4 NiWO 6 structure was used for the calculations, with the experimental tungsten positions fixed and Ni/Li ordering with the lowest-energy configuration. The two models of delithiation previously described were then explored ( Figure  S11 in Supporting Information), mediated by Ni redox and O redox, respectively. The computed voltage for x = 1 in Li 4−x NiWO 6 is found to be 0.2−0.4 V higher for the anionic redox than for the cationic Ni 2+ /Ni 3+ model (Table S9 in Supporting Information), which is consistent with our observation of Ni oxidation at 3.7 V, and highlights the small difference in energy between cationic and anionic redox in this material. The stabilization of peroxide upon further oxidation could prevent the activation of the Ni 3+ /Ni 4+ redox couple or else could be mediated by removing electrons from Ni e g states to form transient Ni 4+ species, according to the reductive coupling mechanism proposed for Ru-based materials. 1,39 However, no change in the oxidation state of Ni beyond 3+ is observed experimentally, so this second hypothesis cannot be confirmed. Because irreversibility increases when charging above 4 V, the material probably undergoes some irreversible oxidation of oxygen and gas release from the surface, corresponding to the process at 4.2 V on the first charge. This would explain why oxidized oxygen species are not observed by XPS in ex situ samples until 5 V. Nevertheless, the reversible formation of peroxo (O 2 ) 2− species in the second cycle, measured by Raman and XPS, proves that the large reversible capacity on discharge (200 mAh/g) is due to both cationic and anionic redox, contributing respectively to one-third and twothirds of the capacity. If we assume that the Ni 2+ /Ni 3+ redox is fully used, then 1.75e − are to be accounted for by oxygen (i.e., 0.29e − /O after normalizing by the number of oxygen in the formula unit), which is 16% higher than in Li-rich NMC (0.25e − /O) according to Luo et al., 40 reflecting the possibility of attaining considerable reversible anionic redox capacity from localized O−O bond formation.

Article
Turning to the large voltage hysteresis observed upon activation of oxygen redox, it should be underlined that this is a true concern for the development of Li-rich cathode materials 41 because it results in a large penalty in the roundtrip energy efficiency (<90% for Li-rich NMC). Such hysteresis has been correlated with oxygen redox activity and cationic migrations, 13,15,42 which both result in different thermodynamic pathways on charge and discharge. One pending question is how the choice of a transition metal affects the bond order and reversibility of O−O bond formation upon charge and discharge. The use of high-valence d 0 cations to increase the lithium content leads to higher capacities, thanks to the combined participation of cationic and anionic redox. However, the XPS and Raman evidence on and almost certainly differing extents of oxygen hole localization compared to that of the 3d n / 5d 0 system studied here. This would lead to less destabilized antibonding σ* states, thus improving the reversibility of the process and resulting in relatively low voltage hysteresis because of the reduced energy penalty for occupying these lessantibonding states. 26,39,44 To take full advantage of the high capacity of d 0 -based Li-

■ CONCLUSIONS
In this work, the structural and electrochemical properties of a family of Li-rich rock salt oxides Li 4+x Ni 1−x WO 6 were explored. The cation ordering is very sensitive to the Li/Ni ratio and generates two new structure types derived from Li 3 Ni 2 TaO 6 for Li/Ni = 4 and Li 5 ReO 6 for Li/Ni > 4. Similar electrochemical processes were found regardless of the cation ordering, but a higher reversible capacity was measured for the Li 4.15 Ni 0.85 WO 6 sample compared to the stoichiometric one. Surprisingly, Ni could not be oxidized beyond Ni 3+ even at high voltage (5 V), accounting for only 64 mAh/g (i.e., one-third of the total reversible capacity (200 mAh/g)), thus leaving the large reversible capacity mostly unexplained by a classical cationic redox mechanism. The capacity associated with oxygen redox in this material is thus considerably larger than that obtained from the oxidation of the 3d n cations. Raman spectroscopy combined with DFT calculation on the cycled material and a known tungsten peroxide standard compound represents definitive evidence of the formation of true peroxo (O 2 ) 2− anions with localized O−O single bonds, stabilized by the presence of 5d 0 W 6+ cations consistent with the solution coordination chemistry of d 0 peroxo complexes. This results in good reversibility of the anionic redox process after the first activation cycle, with a very large voltage hysteresis between charge and discharge that we attribute to the difficulty in reducing the stabilized (O 2 ) 2− species, highlighting the importance of balancing electrochemical access to oxygen oxidation with enabling its reversal. The electronic structure and bonding characteristics of both redox-active (Ni) and redox-inactive (W) components of the structure play key roles in controlling the capacity accessible by stabilizing localized O−O bonds and the reversibility of the bond formation that defines the efficiency of the energy storage process. Such findings can be generalized to other d 0 -containing Li-rich rock salt oxides, which also struggle with large voltage hysteresis and low energy efficiency, although the origin of this issue had not been investigated until now. Having clarified this point, we expect that this work will help in defining new strategies to decrease the voltage hysteresis and bring highcapacity rock salt oxides closer to potential applications.

■ METHODS
Synthesis. Pale-green powder samples of Li 4 NiWO 6 were prepared by a conventional ceramic route. A stoichiometric mixture of Li 2 CO 3 (99.997%, Sigma-Aldrich, dried at 250°C for 12 h), NiO (99.999%, Sigma-Aldrich, dried at 250°C for 12 h), and WO 3 (99.995%, Sigma-Aldrich, dried at 250°C for 12 h) was ball-milled in propanol for 2 h, dried, pressed into a pellet, and loaded into an alumina crucible to produce ∼1 g of material. Part of the precursor mixture was used as a sacrificial powder in order to minimize the effects of Li volatility. The samples were fired twice at 1000°C for 24 h with a heating/cooling rate of 5°C/min, with intermediate hand grinding of the pellet and sacrificial powder between firings. Brown-black powder samples of Li 4+x Ni 1−x WO 6 (0 ≤ x ≤ 0.25) were prepared as above but with a 10% molar excess of lithium in the appropriate ratio of starting materials (e.g., Li/Ni/W = 4.51:0.9:1 for x = 0.1). For the neutron powder diffraction (NPD) experiments, both Li 4 NiWO 6 and Li 4.1 Ni 0.9 WO 6 were enriched with 7 Li using 7 Li 2 CO 3 (99% 7 Li atom, Sigma-Aldrich dried at 250°C for 12 h), with the synthesis scaled up to 4 g of product. WO 2 (O 2 )H 2 O·nH 2 O was prepared following the method described by Journal of the American Chemical Society Article Pecquenard et al., 31 followed by a thermal treatment at 120°C for 12 h to prepare reference sample WO 2 (O 2 )H 2 O.
Diffraction. Phase identification of the Li 4+x Ni 1−x WO 6 samples was conducted using laboratory powder X-ray diffraction data obtained using a PANalytical X'Pert Pro diffractometer (Co Kα 1 ) in Bragg− Brentano geometry. This instrument was also used for the determination of lattice parameters in the Li 4+x Ni 1−x WO 6 (0 ≤ x ≤ 0.25) series by the inclusion of crystalline Si as an internal standard. For detailed structural analysis of the as-synthesized materials, synchrotron X-ray powder diffraction (SXRD) data were collected at the I11 beamline at the Diamond Light Source (Oxfordshire, U.K.), with an incident wavelength of 0.82588(1) Å using five multianalyzer crystal detectors. The samples were loaded into borosilicate capillaries of 0.1 mm diameter to minimize the absorption cross-section of the sample and mounted on a capillary spinner during data collection. Highresolution time-of-flight (ToF) neutron powder diffraction (NPD) data were collected at room temperature using the HRPD instrument at ISIS (Oxfordshire, U.K.). The structural models were refined by the Rietveld method 45,46 using the TOPAS software.
Electrochemical Testing. To evaluate the performance of Li 4+x Ni 1−x WO 6 (x = 0.15) as a Li-ion battery cathode material, composite electrodes were fabricated by casting a mixture of active material/Super C carbon/poly(vinylidene fluoride) (PVDF) binder (Kynarflex, Arkema) (80:10:10 by wt) onto an aluminum foil current collector. Coin cells (CR2025) were assembled in an argon-filled glovebox using 1 M LiPF 6 in ethylene carbonate/dimethyl carbonate (BASF) 1:1 by volume as the electrolyte, impregnated onto a glass fiber separator (Whatman) with a lithium metal counter electrode. Electrochemical characterization was carried out with a Maccor Series 4000 battery cycler, at 25°C, using a C rate of C/10 (defined as 1 Li + exchanged in 10 h) between 1 and 5 V. For the preparation of ex situ samples, electrodes were cycled to different charge/discharge states, recovered, washed with DMC in the glovebox, and dried.
Ex Situ PXRD of Charged Cathodes. Samples mixed with Super C carbon (80:20 by wt) were cycled to different states of charge/discharge in Swagelok cells. They were recovered, washed with DMC, and dried under vacuum before being loaded into borosilicate tubes (0.3 mm diameter). The capillaries were sealed with wax prior to removal from the glovebox and then heat sealed with a blowtorch. PXRD of the samples was collected at the I11 beamline with an incident wavelength of 0.826212 Å using a wide-angle position-sensitive detector.
X-ray Absorption Spectroscopy (XAS). The X-ray absorption spectra were recorded at the B18 beamline at the Diamond Light Source (Oxfordshire, U.K.). X-ray absorption near-edge structure (XANES) spectroscopy was used for the analysis of the nickel and tungsten oxidation states, near the Ni K-and W L III -edges. The data was calibrated using Ni/W metal foil references and normalized with the Athena software. 47 Electrodes for ex situ XAS were prepared by pressing a pellet (∼60 mg) of active material with super C carbon and PVDF binder (74:13:13 by wt). After washing the samples, the active electrode pellets were mixed with an appropriate amount of cellulose (Sigma-Aldrich), pressed to the appropriate density, and sealed into polyethylene-lined foil bags (Sigma-Aldrich) to prevent air exposure. Samples of uncycled active material were prepared in an analogous way, including the carbon and binder. Materials used as reference standards including NiWO 4 (Ni 2+ and W 6+ standard), WO 3 (W 6+ standard), and WO 2 (W 4+ standard) were prepared by mixing appropriate masses of each with cellulose and pressed as pellets but were not sealed in polyethylene foil bags.
X-ray Photoelectron Spectroscopy (XPS). Measurements were performed in a standard ultrahigh vacuum surface science chamber consisting of a PSP Vacuum Technology electron energy analyzer (angle integration ±10°) and a dual-anode Mg Kα (1253.6 eV) X-ray source. The base pressure of the system was 2 × 10 −10 mbar, with hydrogen as the main residual gas in the chamber. The spectrometer was calibrated using Au 4f 7/2 at 83.9 eV. All spectra were calibrated by aligning the main peak in the O 1s region to that measured from the reference WO 3 compound at 530.6 eV. XPS spectra were fitted using Voigt functions (30% Gaussian/70% Lorentzian) after Shirley background removal. The analysis was performed with the minimum number of components required to explain the data. In the case of W 4f spectra, the relative intensities of the 4f 5/2 and 4f 7/2 peaks were constrained from the ratio found for the pristine spectra. Samples were transported from the glovebox to the instrument using a dedicated transfer chamber to avoid contact with air.
Raman Spectroscopy. Ex situ measurements were performed at different stages of galvanostatic cycling. To avoid oxygen and moisture contamination, an airtight Raman cell (ECC-Opto-STD, El-Cell, GmBH) was assembled inside an argon-filled glovebox. Raman spectra were collected with a 633 nm wavelength laser using a Raman system (Renishaw inVia Reflex) with a microscope focused through a 50× objective lens (Leica). The estimated power on the sample was 0.43 mW with 200 s exposure time and two accumulations. The baseline of the spectra was corrected for clarity.
Computational Methods. Plane-wave-based density functional theory (DFT) calculations were performed with VASP, 48 using the PBE functional 49 augmented by an on-site Hubbard correction (PBE+U) using a rotationally invariant scheme 50 with a U eff of 7 eV applied to the Ni 3d states. Core electrons were treated using the projected augmented wave approach, 51 with the Li 1s, Ni 2p, and W 5s and 5p semicore states treated as valence states. A 700 eV plane-wave cutoff energy was used for calculations of final structures and properties, with a 7 × 7 × 4 k-point grid. A 600 eV plane-wave cutoff and a reduced kpoint grid of 5 × 5 × 3 were used when screening the energies of many configurations, which was sufficient to obtain relative energies within 20 meV of those obtained using the more accurate settings. In general, the unit cell and atomic positions of structures were optimized until all forces were below 0.01 eV/Å, with a tighter convergence of 0.001 eV/Å used for the equilibrium structures of normal mode calculations.
Raman spectra are computed using the vasp_raman.py code. 52 A full normal mode calculations was performed at the Γ point using finite differences with displacements of 0.01 Å and the harmonic approximation. Following this, the change in the macroscopic dielectric tensor for each mode was calculated using density functional perturbation theory and was used to compute the Raman activity of the mode. Projection operators were evaluated in reciprocal space for these calculations to provide a high-enough accuracy.
Hybrid DFT calculations were performed with a Gaussian basis set using CRYSTAL14. 53 The HSE06 functional 54 was used to obtain a better prediction of delithiation potentials. The pob-TZVP triple-ζ valence with polarization basis functions was used for Li, Ni, and O, 55 and a modified Hay−Wadt double-ζ basis set was used for W, which had previously been shown to perform well for WO 3 . 56 A k-point grid of 7 × 7 × 4 was used for each model, and the unit cell and atomic positions were optimized until the forces were below 0.01 eV/Å.

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/jacs.8b13633.
Composition/structure analysis and computational summary; structural parameters, evolution of cell parameters upon cycling, and the calculated delithiation potential; structural information, SEM images, differential capacity plots, XPS for the second cycle of Li 4.15 Ni 0.85 WO 6 , and results for the calculated structure and associated DOS (PDF) Li 4 NiWO 6 (CSD 1884635) (CIF) Li 4.1 Ni 0.9 WO 6 (CSD 1884637) (CIF)