Introduction

The global uptake of electric vehicles is driving demand for lithium-ion batteries with greater energy densities1, hence the need for new cathodes with higher capacities2. Lithium-rich cathode materials with lithium:transition metal (TM) ratios >1, including layered Li1+x(Ni,Mn,Co)1−xO2 and disordered rocksalts such as Li2MnO2F, offer increased capacities over conventional cathodes such as LiCoO2 and LiFePO43,4,5,6,7,8. These high capacities are possible because Li-rich cathodes can exhibit reversible redox of bulk oxide ions, termed ‘oxygen-redox’, as well as transition metal ion redox9,10,11. O-redox allows Li-rich cathode materials to achieve theoretical capacities exceeding 300 mA h g−1 ref. 9, which opens up a new frontier in battery chemistry.

However, one critical issue is that O-redox is almost always associated with a large voltage hysteresis in the first-cycle electrochemical load curve12,13,14,15,16,17. The degree to which fundamental atomic-scale mechanisms of O-redox in Li-rich cathodes contribute to voltage hysteresis is not fully understood and remains a topic of considerable debate18,19,20,21,22,23,24,25,26,27,28,29,30,31,32; in particular the nature of the oxidised O species formed on charge and the role of TM rearrangements are unclear33. To harness Li-rich cathodes for technological use, the interrelations between O-redox, TM migration and voltage hysteresis must be fully understood so that strategies can be found to mitigate the loss of energy density.

The disordered rocksalt cathode Li2MnO2F exhibits a large capacity, comparable to that of Li-rich ordered layered oxides34,35. Previous studies using resonant inelastic X-ray scattering (RIXS) and density functional theory (DFT) have identified molecular O2 formed and trapped within the bulk structure when charged to 4.8 V (approximately Li0.75MnO2F)35. Li2MnO2F and some other disordered rocksalts6,34,36,37,38 display a smaller first cycle voltage hysteresis than ordered layered cathode counterparts10,39, which raises important questions: what is the atomic-scale O-redox mechanism within these disordered materials, and how does their local structure facilitate a more reversible O-redox process than in Li-rich layered cathode materials? In the highly ordered Li-rich layered cathodes, it is established that TM migration is necessary for O2 formation14,16,21,22,40,41. In disordered rocksalt materials such as Li2MnO2F, however, it is still unclear what role, if any, TM migration plays in O2 formation.

To address these questions about transition metal migration, O2 formation and voltage hysteresis in Li2MnO2F, we have conducted a multi-technique study of the charge-storage mechanism using DFT and ab initio molecular dynamics (AIMD) simulations, high-resolution RIXS mapping and galvanostatic intermittent titration technique (GITT) electrochemical measurements. Our DFT results reveal that in the highly charged (delithiated) state, molecular O2 species are thermodynamically favoured over superoxide and peroxide species. This result is confirmed by new high-resolution RIXS mapping data, which shows vibrational features from molecular O2 only, with no evidence for superoxide and peroxide species. Using AIMD, we resolve an O–O dimerisation mechanism that involves TM migration and features peroxide and superoxide species as short-lived (picosecond timescale) reaction intermediates, before ultimately forming the thermodynamic end-product, molecular O2. We implicate irreversible Mn migration as a contributor to first cycle voltage hysteresis in Li2MnO2F and discuss how fully-reversible Mn migration could provide a route to stable O-redox cycling through O2 formation without voltage loss.

Results

Oxygen environments and short-range order in Li2MnO2F

In disordered rocksalt-structured Li2MnO2F, octahedrally coordinated cations (Li, Mn) and anions (O, F) occupy two interpenetrating face-centred cubic (fcc) sublattices (Fig. 1a). Disordered rocksalts do not display long-range order, but do exhibit short-range cation order; i.e., preferential local structural motifs37,42,43,44,45,46. Understanding how short-range order affects the local structure around oxygen anions in Li2MnO2F is important because O-redox activity in Li-rich cathodes has previously been attributed to the preferential oxidation of specific lattice O2− ions with Li-rich coordination environments47,48,49. O ions with a higher number of Li neighbours have a lower Madelung site potential, which indicates a lower energy required to localise an electron hole (O2− → O− + e−)49, while oxygen ions with linear Li–O–Li bonding configurations have O 2p states at the top of the valence band that are susceptible to oxidation on charge47,48. Furthermore, the possible presence of extremely lithium-rich oxygen-coordination (i.e. O−Li6) is of particular interest, because removal of these Li during charging may leave these O undercoordinated with no directly bonded Mn neighbours, and potentially allowing O–O dimerisation without requiring Mn–O bond breaking or Mn migration.

Fig. 1: Structure and short-range order of Li2MnO2F.
figure 1

a Representative relaxed structure of disordered Li2MnO2F in a (2 × 2 × 2) expansion of the conventional rocksalt unit cell. b Frequency of O−LixMn6–x octahedra in Li2MnO2F at T = ∞ K (upper panel) derived from a binomial distribution with n = 6, p = 2/3, representing the fully-random limit and at T = 2000 K (lower panel) obtained from cluster-expansion based Monte Carlo simulations, representing the pristine ‘as-prepared’ material.

Full size image

We first quantify the frequency of different oxygen coordination environments (O−Li6, O−Li5Mn, etc) in pristine, as-prepared Li2MnO2F. We used DFT calculations to parameterise a cluster-expansion Hamiltonian to describe the short-range interactions in Li2MnO2F and ran lattice Monte Carlo simulations at T = 2000 K to approximate the ball-milling synthesis conditions of the pristine, as-prepared material34 (for computational details, see Methods section). Figure 1b shows the predicted frequencies of different O-ion coordination octahedra, O−LixMn6–x, obtained by sampling structures from these simulations. We also show data for a model at T = ∞ K, which represents a hypothetical fully random arrangement of the rocksalt lattice, i.e., with no short-range order (Supplementary Note S2.1).

The fully random (T = ∞ K) model of Li2MnO2F features a binomial distribution of O−LixMn6–x (Fig. 1b), which is skewed towards O-octahedra with a high number of Li neighbours due to the 2:1 ratio of Li:Mn in the material. In this fully random model, 8.5% of the O-environments are O−Li6. In the system approximating the ball-milling synthesis conditions (T = 2000 K), the distribution of O-environments deviates from a fully random binomial distribution (Fig. 1b), which indicates short-range order in pristine Li2MnO2F. O−Li6 environments are predicted to have a very low abundance (<0.05%), and F-ions preferentially occupy anion sites with high numbers of Li neighbours (Supplementary Fig. 1)43,45. Because oxygen ions in O−Li6 coordination appear with a very low frequency in the pristine material, the molecular O2 that is observed in experiments upon cycling34 cannot originate uniquely from starting O–Li6 oxygen ion sites. Instead, the O2 molecules must arise from O–LinMn6–n (where n ≤ 5) sites in the pristine material, or O–Lix▢6–x configurations (where ▢ is a vacancy) that could form during charge due to O, Li or Mn displacement.

Stable structures and nature of oxidised oxygen on lithium extraction

Having characterised the anion short-range order in pristine Li2MnO2F, we now investigate the charge mechanism by examining structures of highly delithiated Li0.67MnO2F. This stoichiometry corresponds to structures charged past the limit of Mn3+ to Mn4+ redox and provide new insights into the thermodynamics of different oxidised O species and the possibility of Mn migration.

First, we perform a random structural search at a stoichiometry of Li0.67MnO2F by generating 150 rocksalt configurations with random distributions of anions and cations, plus cation vacancies, over their respective sublattices. This random structure search allows an unbiased sampling of all hypothetical rocksalt-structured configurations of Li0.67MnO2F, without imposing any conditions on the pristine Li2MnO2F structure or on the kinetic pathway that might be required to reach these delithiated structures in experimental samples. The random structure search therefore generated structures that could be obtained under unrestricted TM and anion rearrangement during charge; we label this the ‘Mn-rearrangement’ model. Second, we model delithiation of the pristine material, under the condition that no TM ion migration is permitted, denoted as the ‘constrained-Mn’ model. Here, we used our cluster expansion model at T = 2000 K to generate 150 Li2MnO2F structures, representative of the pristine material. These structures were delithiated either (i) by removing random lithium ions or (ii) removing Li based on a ranking of site energies from electrostatics, to reach a composition of Li0.67MnO2F and then relaxed with DFT. The relaxations quench the structures to a local potential energy minimum and do not allow for significant atomic rearrangements such as Mn migration.

To assess whether O–O dimerisation occurs in any of the structures in either of the two models, we consider the distributions of O–O distances in the relaxed structures (Fig. 2a); distances shorter than 1.7 Å are indicative of covalent O–O dimerisation. In the constrained-Mn model structures, we find no O-dimers of any kind. Charge compensation beyond the Mn4+ limit in these models is predominantly from lattice On− ions. In contrast, a large proportion of structures from the Mn-rearrangement model show some O–O interatomic distances <1.7 Å, indicating short covalent O–O bonds (Fig. 2a).

Fig. 2: Search for O–O bonds and thermodynamics of O-dimer formation in delithiated Li0.67MnO2F.
figure 2

a Search for O–O dimers by an analysis of O–O interatomic distances in delithiated structures of the pristine material obtained from the cluster-expansion (‘constrained’ Mn), and in ‘Mn rearrangement’ models. b Energetics of the structures from a with the structures from the ‘Mn-rearrangement’ model separated into those containing O–O bonds (<1.7 Å) and those not containing O–O bonds. The right section of each panel shows the kernel density estimations of the probability of energies in the left panel, where each dot is the energy of one structure.

Full size image

Insight into the thermodynamics of O–O dimerisation in Li0.67MnO2F is provided by considering the calculated energies of all the relaxed structures for the constrained Mn and Mn-rearrangement models (Fig. 2b). For the Mn-rearrangement model, structures that contain covalent O–O bonds are on average ∼0.4 eV per formula unit more stable than those that do not contain covalent O–O bonds. Furthermore, the lowest energy structures across both models contain covalent O–O bonds. The results therefore suggest a thermodynamic driving force may exist for pristine Li2MnO2F to undergo a framework transformation upon delithiation to allow O–O dimerisation.

The lack of dimerisation in the constrained Mn model is because O–O dimerisation is an activated process that requires either Mn–O bond breaking, or Mn/O displacement. All O atoms in the constrained Mn model start with at least one Mn neighbour, and the structures relax to the nearest local energy minimum, rather than overcoming the barrier needed for O–O dimerisation. In contrast, in the Mn rearrangement model, some O atoms begin with no Mn neighbours and can dimerise without an activation barrier, so covalent O–O bonds (dO–O < 1.7 Å) form during the geometry relaxations.

There is an ongoing debate over the bond length and oxidation state of O–O dimers in charged O-redox cathodes; peroxide O22− (dO–O ∼1.44 Å) or (O···O)n− species with long interatomic separations of ~2.4 Å are sometimes invoked to explain capacity from oxidised O11,21,50. In relaxed structures from the random structure search, the probability density for covalent O–O bonds has maxima at ∼1.22 Å, ∼1.30 Å and ∼1.45 Å, (Fig. 2a, inset) indicating that molecular O2(0), superoxide O2− and peroxide O22− species could all form in Li0.67MnO2F. In Fig. 3a, we compare the thermodynamic stability of structures containing the different types of O–O dimers, classified, as is convention, according to bond length: (peroxide O22− (1.35 Å ≤ dO–O < 1.70 Å), superoxide O2− (1.24 Å ≤ dO–O < 1.35 Å) and molecular O2 (dO–O < 1.24 Å))51. Structures containing molecular O2 species are most frequently obtained and are, on average, the most stable, indicating that molecular O2 is the thermodynamic product in charged Li2MnO2F.

Fig. 3: Computational and experimental evidence for molecular O2 formation in charged Li2–xMnO2F.
figure 3

a Comparison of the stability of structures from the Mn rearrangement model, containing the different types of O–O dimers, classified according to bond length: (peroxide O22– (1.35 Å ≤ dO–O < 1.70 Å), superoxide O2– (1.24 Å ≤ dO–O < 1.35 Å) and molecular O2 (dO–O < 1.24 Å). The panels show the kernel density estimations of the probability of energies for the classified structures. The energies were calculated relative to the most stable configuration from the entire search. Thus, zero energy corresponds to the most stable obtained from the Mn rearrangement model. Calculations used the SCAN functional. b High-resolution resonant inelastic X-ray scattering (RIXS) map of Li2–xMnO2F charged to 5.0 V, showing the vibrational features corresponding to molecular O2 only, and an energy loss feature at ∼7.5 eV.

Full size image

To investigate the possible presence of peroxide or superoxide species in charged Li2MnO2F, we performed high-resolution O K-edge RIXS mapping of Li2–xMnO2F charged to 5.0 V. In RIXS, incident radiation excites electrons from the O 1 s core-level states to empty O 2p valence states, creating core-holes. Relaxation of electrons from the filled O 2p valence states back into the core hole (O 1 s), results in emission photons, whose energy are measured, providing a direct probe of the oxygen valence states. The RIXS results are presented in terms of excitation energy, and (emission) energy loss. At an excitation energy of 531 eV, a series of energy loss features near the elastic (zero energy loss) peak can be resolved, arising from transitions to different vibrational energy levels of an O2 molecule (Fig. 3b), with the peak separation of the lowest energy-loss peaks corresponding to the fundamental molecular O2 vibrational frequency (ν) of 1550 cm−1 Refs. 14, 16, 52. No vibrational features from peroxide (ν = 750 cm−1) or superoxide (ν = 1100 cm−1) species are observed. The combined DFT and RIXS results support the idea that molecular O2 is the thermodynamically favoured oxidised O product in highly delithiated Li2MnO2F, and in particular, highlight that molecular O2 is favoured over peroxide and superoxide species.

Mechanism of O–O dimerisation in delithiated Li2–xMnO2F

The structural analysis presented above reveals a strong thermodynamic driving force for O2 formation in highly delithiated Li0.67MnO2F, which is consistent with the experimental observation of molecular O2 trapped in the bulk structure35, and also highlights the necessary role of Mn, O or F migration for this O-redox process to occur. Although this analysis provides valuable thermodynamic insights, it does not give direct information about the atomic mechanisms involved.

We therefore performed ab inito molecular dynamics (AIMD) simulations on a selection of delithiated structures, which allows us to probe their structural evolution as a function of time53,54. We examined nine different charged Li0.67MnO2F structures, which were obtained as Li2MnO2F from the cluster-expansion, then delithiated (see Methods section). We ran AIMD simulations on each structure at 500 K, a slightly elevated temperature with respect to experiment to allow better sampling of kinetically allowed processes within the accessible simulation timescale (~60 ps). For our detailed analysis here, we focus on one exemplar structure (Fig. 4); structures for the other simulations are shown in the Supplementary Information (Supplementary Fig. 7).

Fig. 4: Reaction mechanism to form O–O dimers in Li0.67MnO2F from ab inito molecular dynamics (AIMD) and geometry relaxations.
figure 4

a O···O interatomic separation of O species forming O–O dimers from GGA+U AIMD simulations at 500 K. b Total energy of selected structures along the AIMD trajectory, fully relaxed with DFT using the HSE06 functional. c Geometry of the relaxed structures from b, indicating Mn migration events and O–O dimerisation. Mn ions displaced from their octahedrally coordinated rocksalt sites are indicated by pink polyhedra.

Full size image

Within this simulation trajectory, three O2 molecules form spontaneously. Further analysis highlights three key points. First, we show that molecular O2 formation is preceded by Mn ion migration (Fig. 4c), with several Mn ions migrating from their initial octahedrally coordinated sites to interstitial sites. These are either also octahedrally coordinated sites, located at the shared edge between two pairs of octahedral sites in the original rocksalt lattice or are fivefold coordinated sites. The displaced Mn ions are stabilised in these interstitial positions due to large off-site relaxations of the anion sublattice (Supplementary Fig. 8). Second, these Mn migration events leave some O ions in undercoordinated environments (fewer than two Mn nearest-neighbours); a concurrent displacement of the anion sublattice allows some O ions to approach each other which then permits O–O dimerisation. Third, the mechanism to form molecular O2 involves peroxide and superoxide species appearing as short-lived (picosecond timescale) reaction intermediates; details of these intermediate O2n− species can be resolved by tracking the O–O interatomic separation of the pairs of O that form the O2 molecules (Fig. 4a).

Closer analysis shows that the formation of molecule O2(1) involves rapid dimerisation (within 2 ps), briefly appearing as a peroxide intermediate with an O–O bond of 1.40 Å, before detaching from neighbouring Mn ions and relaxing to molecular O2 (dO–O = 1.22 Å). O2(1) can be classified clearly as molecular O2 for the majority of the 60 ps simulation, except for a brief interaction with framework O atoms (Supplementary Fig. 9), which cause temporary (∼4 ps) lengthening of the O–O distance to ∼1.30 Å (superoxide).

The formation of molecules O2(2) and O2(3) shows different behaviour; these dimerise after 3 ps and 7.5 ps respectively, and initially bridge between two or three Mn ions (Supplementary Fig. 5). Molecule O2(2) initially has a O–O separation of ∼1.3 Å, corresponding to a superoxide species, before shortening slightly to an average of 1.25 Å, which is intermediate between the equilibrium superoxide O–O distance and molecular O2 species. Molecules O2(2) and O2(3) remain in this intermediate state, until ∼55 ps, at which point the O–O distances for both dimers shorten to 1.22 Å (molecular O2), coinciding with each dimer moving away from their neighbouring Mn ions, into a Li-vacancy nanovoid in the structure.

The change in potential energy that accompanies this process of coupled Mn-migration and molecular O2 formation is illustrated in Fig. 4b, which shows DFT energies for selected structures along the AIMD trajectory that were fully relaxed at the hybrid functional level. Each selected structure relaxes to a local energy minimum, and following the reaction pathway leads to an overall stabilisation of the system. The superoxide species found in the AIMD simulations are found to be short-lived metastable reaction intermediates on a ps timescale, and exist in a shallow potential well on the energy surface, in agreement with the observation of some superoxide species from the random structure search (Fig. 2). The final product along the AIMD trajectory, containing molecular O2 is confirmed to be the most stable configuration, in agreement with the experimental RIXS data (Fig. 3).

Charge-discharge process and voltage hysteresis

Understanding the important relationship between voltage hysteresis, TM migration and O2 formation in Li-rich cathodes requires investigating both the charge and discharge processes, and considering possible structural changes during the first cycle. Our results show that in Li2MnO2F, molecular O2 formation at the top of charge drives a reconfiguration of the cathode Mn-host framework. This may then lead to a different reaction pathway and different energetics for discharge compared to charge, and a new structure after the first cycle.

To investigate the role of molecular O2 formation and Mn-migration in voltage hysteresis, we calculated the voltage curve upon charging to Li0.67MnO2F, considering the following two possible alternative end-points (i) metastable structures containing lattice On− ions, and (ii) a structure containing O2. The structures containing lattice On− ions were obtained by removing Li from the pristine Mn-host framework. The structure containing O2 was obtained by taking the most stable structure containing On− and performing the minimum number of Mn hops that would leave an O atom undercoordinated and allow O2 to form (Fig. 5). The calculated charge voltage curves were compared with an experimental first charge/discharge curve obtained from GITT measurements (see Methods section). GITT provides a voltage profile much closer to the thermodynamic equilibrium than cycling at a conventional C-rate (i.e. 0.1 C).

Fig. 5: Experimental and calculated voltage curves.
figure 5

The calculations are based on charging to structures containing lattice On–, or to a structure containing O2, compared against the experimental first-cycle voltage-capacity curve from galvanostatic intermittent titration technique (GITT) measurements.

Full size image

The results in Fig. 5 reveal two key features. First, the calculated charge voltage curve to structures containing lattice On− ions exceeds 5 V, which is clearly inconsistent with experiment. Second, the delithiated structure containing O2 has a lower predicted voltage, providing a better agreement with experiment. The average calculated charge voltage from the pristine structure to the structure containing O2 is 3.65 V (Supplementary Note S2.3), which compares well with the average experimental voltage of 3.65 V.

We then investigated the discharge process and how this is affected by a structural rearrangement during charge, by re-inserting lithium ions back into the structure containing O2 and calculating the average discharge voltage. Re-inserting Li into the structure containing O2 results in a new discharged structure that contains two O−Li6 environments; the O atoms of the O2 molecule formed during charge are re-incorporated into the lattice as On− ions. This new discharged structure is significantly higher in energy (∼0.4 eV f.u.−1) than the pristine Li2MnO2F structure, due to the relative instability of the trapped O−Li6 environments (Supplementary Fig. 18). The average calculated discharge voltage is 3.35 V; this is 0.3 V lower than the calculated charge voltage, in accord with the ~0.16 V hysteresis observed experimentally from a mid-point potential rest experiment (Supplementary Fig. 20). In other words, the calculations indicate there is a voltage hysteresis, and this arises from the irreversible structural transformation during charge to form O2.

Strategies to harness reversible O-redox

Preventing voltage hysteresis and voltage fade is critical for the development of practical O-redox cathodes with high energy densities. Hysteresis can have several sources, including kinetic limitations such as cathode polarisation due to slow lithium diffusion, or first-order phase transitions, both of which should disappear in the limit of extremely slow charging rates, where thermodynamic equilibrium is approached. Another type is ‘path-dependent’ hysteresis which can arise from irreversible structural changes or slow mobility of host TM cation or anion species55. One strategy proposed18 to suppress path-dependent hysteresis in O-redox cathodes is to prevent O2 formation by inhibiting TM migration16,56,57. In our AIMD simulations, we observe several Mn migration events. Mn migration occurs when Li ions are removed from the structure adjacent to Mn, and the Mn ions then move into new sites, made possible by these vacancies. The relatively large number of Mn ions that migrate is because there are many lithium vacancies at high levels of delithiation in such a lithium-rich system Li2MnO2F. This implicit link between high Li:TM ratios and ease of Mn migration implies a trade-off in disordered rocksalt cathodes between theoretical capacity and preventing voltage hysteresis. Lower levels of Li-excess, and a more contiguous and connected 3D framework of edge-sharing TM ions is expected to help prevent TM migration.

A similar principle has been proposed in layered Li-rich cathodes, where the superstructure ordering within TM layers affects the stability of those layers; greater stability is achieved with more contiguous TM connectivity16. Interesting 3D examples that illustrate this principle include ‘partially ordered’ spinel-type Mn-oxyfluoride cathodes, which display small voltage hysteresis38. These spinel-type materials have a relatively low Li:TM ratio (∼1.5 compared with 2 here in Li2MnO2F), which is likely to have a relatively well-connected network of edge-sharing Mn octahedra. We suggest that these features will tend to reduce the magnitude of off-lattice anion displacements at high states of charge and inhibit TM migration. Achieving a lower level of Li-excess in Li2MnO2F, if the level of fluorination is kept constant, opens the possibility of using low-valent dopants such as Mg2+ and Zn2+58, in contrast to high-valent d0 dopants such as Ti4+ and Nb5+ that are often used in disordered rocksalt cathodes39.

Another approach to the design of Li-rich disordered rocksalt cathodes would be to allow O2 formation and then aim to have fully-reversible TM migration, where the TM ions return to their original sites18. By allowing O2 formation, the cathode is rendered much more stable on charge, since metastable lattice On− species are not trapped in the structure. To allow O2 formation, the roles of local structural rearrangements and reversible transition ion migration become important. The displacement of octahedral cations into tetrahedral sites, as has recently been described for Cr3+ to Cr6+ in the disordered rocksalt cathodes Li1.2Mn0.2Ti0.4Cr0.2O2 ref. 59 and Li2Mn0.75Cr0.25O2F ref. 60, and for Fe ions in Li1.17Ti0.33Fe0.5O2 ref. 15 may permit O–O dimerisation while partially recovering the original structure on discharge. In our study, we show that large off-lattice displacements of anions can permit the migration of Mn ions into interstitial sites. Fully-reversible migration of these Mn ions back to their original sites on discharge could recover the starting structure, and may result in suppressed voltage hysteresis61.

Discussion

In conclusion, we have shown that, despite exhibiting limited structural order, Li-rich disordered rocksalt cathodes such as Li2MnO2F, require transition metal migration before molecular O2 is able to form in the bulk structure on charge. Our results unify the behaviour of disordered rocksalt cathodes with Li-rich ordered layered cathodes, where it is already known that the O-redox process also involves transition metal migration and O2 formation, leading to voltage hysteresis. The nature of hysteresis in disordered materials does differ from the layered systems, since they exhibit solid solution rather than two-phase behaviour during the first charge. Hence, Mn migration and O2 formation occur throughout the rocksalt rather than only in a fraction of the layered cathode that is charged.

Ab initio molecular dynamics combined with DFT calculations show that both TM migration and O2 formation are thermodynamically favoured processes and occur rapidly in charged Li2MnO2F. High-resolution RIXS mapping confirms that molecular O2 forms (rather than superoxide or peroxide species) and is trapped within lithium vacancy clusters or nanovoids in the bulk material. On discharge, the O2 molecules are reduced back to O2− lattice ions. When migration is irreversible, these O2− ions occupy sites with a different coordination environment than the pristine compound, which leads to a loss of voltage (~0.3 V) in the first charge/discharge cycle, consistent with the degree of voltage hysteresis observed experimentally. If Mn displacement is fully-reversible, the original Li2MnO2F structure and coordination around O2− are recovered, which facilitates O-redox cycling without path-dependent voltage hysteresis. The results presented here suggest that promoting reversible transition metal displacement or suppressing migration altogether in Li-rich disordered rocksalt cathode materials would provide effective strategies to harness O-redox without loss of energy density.