Voltage- and time-dependent valence state transition in cobalt oxide catalysts during the oxygen evolution reaction

The ability to determine the electronic structure of catalysts during electrochemical reactions is highly important for identification of the active sites and the reaction mechanism. Here we successfully applied soft X-ray spectroscopy to follow in operando the valence and spin state of the Co ions in Li2Co2O4 under oxygen evolution reaction (OER) conditions. We have observed that a substantial fraction of the Co ions undergo a voltage-dependent and time-dependent valence state transition from Co3+ to Co4+ accompanied by spontaneous delithiation, whereas the edge-shared Co–O network and spin state of the Co ions remain unchanged. Density functional theory calculations indicate that the highly oxidized Co4+ site, rather than the Co3+ site or the oxygen vacancy site, is mainly responsible for the high OER activity. Determining catalyst electronic structures during electrochemical reactions is crucial to understand mechanisms. Here authors perform in operando soft X-ray spectroscopy on a cobalt oxide catalyst during O2 evolution and observe voltage and time-dependent valence state transitions.

T he oxygen evolution reaction (OER) plays a crucial role in many modern energy storage and conversion technologies such as solar/electricity-driven water splitting systems, rechargeable metal-air batteries, and regenerative fuel cells. However, the OER still constitutes a bottleneck in these electrochemical devices because of the intrinsically sluggish kinetics involving multistep proton-coupled electron transfer processes 1,2 . The widespread use of alkaline water electrolysis for the OER strongly depends on the availability of low-cost and efficient electrocatalysts. New materials need to be designed to replace the benchmark OER catalysts that typically consist of precious metal oxides such as IrO 2 and RuO 2 . For this purpose, 3d transitionmetal (TM) oxides have recently emerged as promising candidates. They are inexpensive because they are earth abundant and environmentally friendly and have tunable chemical reactivities depending on the type of electronic and crystal structures. During the past decade, many 3d oxides such as spinels 3,4 , perovskites 5,6 , rock salt 7 , and oxyhydroxides [8][9][10][11][12] , have become serious candidates for OER catalysts. The intricate interplay of charge, spin, orbital, and coordination degrees of freedom of 3d ions in oxides makes possible the preparation of highly active catalysts 5,6,13 .
The theoretical description of the OER involves four electron charge-transfer steps on surface metal sites. Very recently, anionic redox processes at lattice-oxygen sites were also considered 14,15 . These are complex processes, and tremendous efforts have been devoted to identifying the key parameters 5,6,12,13,16 . These parameters are usually identified based on the electronic structure of the as-prepared materials. There is, however, an increasing amount of evidence that electrocatalysts also undergo crystallographic changes under OER conditions 8,[17][18][19] . For example, surface amorphization combined with a change in the Co-O network from corner-sharing to edge-sharing was observed for Ba 0.5 Sr 0.5 Co 0.8 Fe 0.2 O 3−δ (Ref. 20 ). Under an electric field, SrCoO 3−δ converts from the perovskite phase to brownmillerite SrCoO 2.5 and further to the unexplored H-SrCoO 2.5 phase 21,22 .
It is clear that in operando experimental studies are needed to elucidate the mechanism of the OER. Many particular intermediate states of 3d TM-oxide catalysts were identified by previous operando infrared spectroscopic studies [23][24][25][26] . The earlier infrared work identified the oxyl radical (Ti-O • ) at the n-SrTiO 3 / aqueous interface 24 , while very recent work indicated both the oxyl radical (Ti-O • ) that terminates the surface, and the bridged intermediate (Ti-O • -Ti) parallel to the surface 26 25 . The objective should not only be to follow the crystal structure changes, but also to record and understand the electronic structure and its modifications during the reaction. In fact, depending on how detailed and specific the information is, knowledge about the electronic structure can be most valuable for identification of the active sites. The soft X-ray absorption spectra at the 3d TM-L 2,3 and O-K edges are highly sensitive to the valence state, spin state, and local environment of 3d TM elements. Note that the valence-state transition of a high-valent TM element does not necessarily indicate a gain of one hole in the metal 3d state. For instance, the Co 4+ oxides BaCoO 3 (Ref. 27 ) and SrCoO 3 (Ref. 28 ) are negative charge-transfer systems, in which the 3d 5 configuration has a contribution of <10%, whereas the dominant configuration is 3d 6 L (L stands for the hole in the ligand O 2p states) with t 2g 5 e g 1 L eg for a low-spin state and t 2g 4 e g 2 L eg for an intermediate-spin state as shown in Ref. 28 , induced by very strong covalence between Co 3d and O 2p. Such O 2p holes can be directly studied by soft X-ray absorption spectroscopy (SXAS) at the O-K edge, especially for an edgeshared Co-O network, which presents a well-separated Co 3+ -and Co 4+ -related spectral feature in the spectra of Co 3+ /Co 4+ mixed valence oxides 29,30 . Thus, the O-K SXAS spectra can generally be used to explore the Co 4+ content via Co 3d-O 2p covalence 27,28 .
Here we apply SXAS at the Co-L 2,3 and O-K edges to determine in operando the valence state, spin state, and local coordination of the Co ions in Li 2 Co 2 O 4 during the OER. Li 2 Co 2 O 4 has the best OER activity among materials with spinel structures and is quite comparable to the well-known benchmark IrO 2 31 . The very detailed information that we were able to obtain about the Co local electronic structure allowed us to deduce that under OER conditions a substantial fraction of the Co 3+ ions was converted into high-valent Co 4+ . This is associated with a delithiation process that also took place. Importantly, after the OER, the lowspin state of the Co ions and the edge-shared Co-O network remain unchanged. By combining these results with those obtained by density functional theory calculations, we can then infer that the Co 4+ sites (with a dominant oxygen ligand hole ground state) are responsible for the high OER activity. Although it is experimentally challenging to separate the electrochemical liquid cell from the ultrahigh vacuum, our study shows that SXAS in operando is the spectroscopic method of choice for the investigation of TM 3d-based catalysts since the atomic-like multiplet structures in the spectra contain highly specific information about the valence, spin, and local coordination of the 3d ions.
Results pH-dependence of the OER activity. We started by studying the pH-dependence of the OER activity, which can provide us with an initial overview of the possible active sites and reaction paths 14,32,33 . It was suggested that a conventional OER involving four concerted proton-electron transfer steps on a surface metalion center, exhibited pH-independent activity, while latticeoxygen oxidation from highly covalent oxides involving nonconcerted proton-electron transfer steps exhibited a strong pHdependent OER activity 14 . The strong pH-independent activity of Ni-based OER catalysts was attributed to the formation of negatively charged surface sites that act as OER precursors originating from a deprotonation 32 or was associated with severe surface degradation in addition to the redox activity of lattice oxygen 33 . In the spinel oxide ZnFe 0.4 Co 1.6 O 4 , the pH-dependent OER activity was assigned as decoupled proton-electron transfers, while the lattice-oxygen contribution was excluded, considering the wide energy gap between the O p-band center and the Fermi level due to the cation-deficient spinel 34 . Figure 1a presents the OER performance of Li 2 Co 2 O 4 . It is apparent that the OER activity of Li 2 Co 2 O 4 has a strong pH-dependence, very similar to that of ZnFe 0.4 Co 1.6 O 4 having a Co 3.34+ valence state but very different from that of LaCoO 3 , which contains pure Co 3+ . This is the first indication that the Co valence of Li 2 Co 2 O 4 may have changed during the OER. We note that surface degradation as the origin of the pH-dependent OER activity can be excluded. Supplementary Fig. 1 shows, for example, that the Li 2 Co 2 O 4 catalyst has excellent electrochemical stability, as indicated by the nearly constant overpotential over 100 h. Our next step was to investigate the electronic structure of Li 2 Co 2 O 4 as a function of applied voltage and time, and for this we analyzed the cyclic voltammograms to select the voltages for the SXAS experiments. Figure 1b shows the cyclic voltammograms recorded at a scan rate of 5 mV s −1 with Li 2 Co 2 O 4 loading on carbon paper. We observed two redox transitions in the cyclic voltammogram: a weak transition at 1.25 V (vs. reversible hydrogen electrode (RHE)) and a relatively strong peak at 1.40 V. In CoO, Co 3 O 4 , and CoOOH, two transitions at 1.4 and 1.5 V were previously attributed to the Co 2+ /Co 3+ and Co 3+ /Co 4+ redox couples, respectively 4 . Since we have only a Co 3+ /Co 4+ redox couple, the relatively strong peak at 1.4 V might correspond a sharp valence-state transition of cobalt ions from Co 3+ to Co 4+ , while the weak feature at 1.25 V was assigned to a minor contribution from the of Co 3+ → Co 4+ transition, as shown in Fig. 1b. The height of the strong peak at 1.4 V decreases with an increasing number of CV scans from the 1st cycle (black) to the 20th cycle (magenta). This might suggest that the number of cobalt ions expected to undergo the valence-state transition gradually decreases in Li 2 Co 2 O 4 . Here, we observed a sharp increase in the current density at 1.55 V, where the material is activated to be a stable, highly active OER catalyst under electrochemical conditions.
In operando SXAS studies under OER conditions. In Fig. 2a, b we present the O-K spectra of Li 2 Co 2 O 4 in operando as a function of the number of scans (taken within 2 min for each scan) and the applied voltages. Upon applying 1.4 V, the current density is still very weak, as shown in Fig. 1b. However, one can already observe in Fig. 2a a clear increase in the intensity of two new spectral features, β and γ, with an increasing number of scans. After 10 scans (20 min), the spectral intensities of β and γ reach a maximum and remain unchanged with further scans as shown in Supplementary Fig. 2. For an applied voltage of 1.6 V, under which there is a sharp rise in the current density, as displayed in Fig. 1b, the spectral intensities of peaks β and γ increase much more quickly, as shown in Fig. 2b. The intensity increase stopped after 20 min (see Supplementary Fig. 2) but reached a much higher maximum value.
To interpret the time-dependent O-K SXAS spectra taken at 1.4 and 1.6 V, we also present for comparison in Fig. 2c the spectra of the layered LiCoO 2 (cyan line), Li 0.66 CoO 2 (orange line, from Ref. 29 ), and the layered Na 0.5 CoO 2 (dark cyan line, from Ref. 30 ) having Co 3+ , Co 3.34+ , and Co 3.5+ valence states, respectively. All three reference materials have the same lowspin state and the same edge-shared Co-O network at a 90°C o-O-Co bond angle. The absolute energy positions of all spectra were calibrated using the sharp pre-edge peak at 531.7 eV in the O-K SXAS spectrum of NiO measured simultaneously 35 . All the peaks below 533 eV represent the unoccupied O 2p states that are mixed into the unoccupied Co 3d states. The sharp single peak α at 530.25 eV for the as-prepared Li 2 Co 2 O 4 in Fig. 2a, b, as well as that for LiCoO 2 in Fig. 2c, reflects a fully occupied t 2g 6 state associated with the low-spin Co 3+ ion 29 . In this case, only transitions to the unoccupied e g states are possible. For the mixed-valent Co oxides Li 0.66 CoO 2 (from Ref. 29 ) and Na 0.5 CoO 2 (from Ref. 30 ), two additional peaks, β and γ, appear at lower energies. They originate from transitions to the unoccupied t 2g and e g orbitals of the Co 4+ ion, respectively, as also observed in BaCoO 3 (see Fig. 2e, f (blue line)), also having the low-spin state and the same 90°Co-O-Co bond angle. Thus, the increase in the spectral intensities of the β and γ features during the OER indicates an increase in the Co valence state. In other words, there is a transition from a Co 3+ state to a Co 4+ state in part of the Li 2 Co 2 O 4 catalyst during the electrochemical reaction. The spectra of Li 2 Co 2 O 4 at different applied voltages are presented in Fig. 2d, where one can observe only a negligible amount of Co 4+ ions at 1.2 V. The spectral intensity of Co 4+ (β and γ) in Li 2 Co 2 O 4 after 20 min under an applied voltage of 1.4 V is only slightly smaller than that of Na 0.75 CoO 2 in the Co 3.25+ state 30 , while after 20 min under 1.6 V, it lies between those of Li 0.66 CoO 2 and Na 0.5 CoO 2 . The Co valence in these two cases can be estimated by constructing a weighted sum of the Li 2 Co 2 O 4 spectrum (blue line) and BaCoO 3 spectrum (cyan line, from Ref. 27 ), as shown in Fig. 2e, f. We find Co 3.2+ and Co 3.4+ valences for these two cases, respectively.
Using the same procedure as shown in Fig. 2e, f, we extract and display in Fig Table 1, we show the lithium content as a function of time. The data sets reveal that the composition of the as-prepared material, i.e., at 0 min in Fig. 3b, is in accordance with the chemical formula of Li 2 Co 2 O 4 , and after 20 min of the OER at an applied voltage of 1.6 V, half of the lithium ions were removed. The current density j (solid line), as shown in Fig. 3c, also has a similar trend with time. This strongly indicates that the OER activity is directly related to the appearance of Co 4+ ions.
We have carried out a number of verification experiments to ensure that our experimental observations are correct. We also carried out O-K SXAS measurements under ultrahigh vacuum conditions. In Fig. 2c we show the in vacuo spectrum of Li 2 Co 2 O 4 after 20 min of OER under an applied voltage of 1.6 V (magenta circles). A comparison with the in operando spectrum (black circles) reveals that the spectra are identical in the energy region of interest. There are differences between the in vacuo and in operando spectra, but these occur only in the energy region above 533 eV. As illustrated in Supplementary Fig. 3, there is extra background in the in operando spectrum above 533 eV, which can be attributed to the oxygen in the electrolyte. Thus, from the relevant features that appear below 533 eV, we can safely conclude that the O-K SXAS experiments (in operando and in vacuo) establish the irreversible valence change of Co during the electrochemical reaction. Next, we also performed SXAS experiments at the Co-L 2,3 edges. In Supplementary Fig. 4e, we show that there is a change and shift to higher energies of the Co spectrum when comparing the Li 2 Co 2 O 4 after the OER with the as-prepared Li 2 Co 2 O 4 , indicating an increase in the Co valence 27,36,37 . We also quantitatively analyzed the Co-L 2,3 spectra in terms of a weighted superposition of the theoretical LS Co 3+ and LS Co 4+ spectra, see Supplementary Fig. 5a. We found 60% Co 3+ and 40% Co 4+ , i.e., an LS Co 3.4+ state, which is fully consistent with the O-K data. We note that the sharp low-energy shoulder S at 779.7 eV in the Co-L 2,3 spectra as displayed in Supplementary Figs. 4e and 5a is a transition to the unoccupied t 2g and a fingerprint of the LS Co 4+ state, as observed in the BaCoO 3 spectrum (wine) in Supplementary Fig. 4a. As a final check, we also measured the Co-L 2,3 and O-K spectra by the more surface-sensitive total electron yield (TEY) method for comparison with the total fluorescence yield (TFY) spectra. Supplementary Fig. 5a, b reveals that the Co-L 2,3 spectra are very similar and thus also extracted the Co valence, i.e., Co 3.35+ in TEY versus Co 3.4+ in TFY. Supplementary Fig. 6b, c shows that  the relative spectral weight of the β and γ features in the O-K SXAS is also only slightly smaller in the TEY than in the TFY, fully consistent with the Co-L 2,3 findings. We note that these small differences between the TEY and TFY spectra may suggest that part of the Co 4+ ions at the surface are somewhat destabilized by the in vacuo conditions under which the TEY spectra were recorded. From all these checks we can safely conclude that the SXAS experiments firmly establish the irreversible change in the Co valence during the electrochemical reaction.
Structure and morphology characterization. We now investigate the effect of OER on the crystal structure of Li 2 Co 2 O 4 using X-ray powder diffraction (XRD). Figure 4a shows the data of the sample as prepared (top) and after the OER (middle) under an applied voltage of 1.6 V for 20 min in an O 2 -saturated 1 M KOH aqueous solution. The XRD patterns can be indexed by the standard Fd-3m symmetry (JCPDS No. 01-080-2159) corresponding to a pure spinel structure. We also observe that the patterns before and after the OER are essentially identical and that there are no additional peaks that otherwise may indicate the presence of an impurity phase. We note that there is a reflection from the carbon paper substrate, which we have indicated by the orange line at the bottom of Fig. 4a. The results demonstrate that the overall crystalline phase of Li 2 Co 2 O 4 remains unchanged after the OER. We can rule out impurity phases such as Co 3 O 4 , which was reported previously after the electrochemical process 38 . To further confirm our point, we have studied the Raman spectrum, which is also highly sensitive to impurity phases. As shown in Fig. 4b, the Raman spectra of Li 2 Co 2 O 4 before and after the OER display four peaks at 450, 487, 590, and 608 cm −1 , corresponding to the A 1g + E g + 2F 2g active modes, indicating again a pure spinel structure for the sample both before and after the OER.
As a next step, we examined the atomic arrangement in the Li 2 Co 2 O 4 nanoparticles in the near surface regions using a highangle annular-dark-field scanning transmission microscope (HAADF-STEM). Figure 4c, d shows the HAADF-STEM images viewed down the [110] axis of the Li 2 Co 2 O 4 samples before and after the OER, respectively. We have observed that the atomic arrangement in the as-prepared nanoparticles matches very well the spinel lattice with the space group Fd-3m and that Co ions and Li ions occupy the 16d and 16c octahedral sites, respectively. The HAADF -STEM images show well-defined lattice fringes in the surface region of Li 2 Co 2 O 4 samples, revealing no change in the crystal structure at the surface of Li 2 Co 2 O 4 due to the OER. From the combination of SXAS, XRD, and HAADF-STEM analysis, we can infer that the low-spin state of the Co ions remains unchanged. We can also conclude that there is no complete change from the edge-shared network to corner-shared network, since the O-K SXAS spectrum of perovskite (La 1−x Sr x CoO 3 ) is very different from that of spinel structure (Li x CoO 2 ), as shown in Supplementary Fig. 7. The strong intersite hopping smears the difference in the spectral features in the O-K edge between Co 3+ and Co 4+ for a corner-shared network in perovskite La 1−x Sr x CoO 3 , while the differences between Co 3+ and Co 4+ are clearly seen in the spinel structure Li x CoO 2 (Ref. 29 ) and Na x CoO 2 (Ref. 30 ) with an edge-shared network. However, we cannot exclude slight changes in the Co-O-Co bond angle. The detailed structural changes within the data presented could be revealed in the future by theoretical studies and operando XRD experiments.
DFT calculations for OER mechanisms and active sites. Having determined the valence state of the Co ions and the crystal structure under the OER condition, we now investigate the underlying mechanism and the active site for the OER performance in Li 2 Co 2 O 4 . For this purpose, we performed DFT calculations to model the reaction pathways. Three scenarios of the reaction mechanism were considered, as shown in Fig. 5a-c, and each involves four proton-electron transfer steps. The first one, shown in Fig. 5a, is the metal-site adsorbate evolution mechanism (MAE). This is the conventional adsorbate evolution mechanism 12 , which considers only the redox activity of the TM sites. The second one, shown in Fig. 5b, is the lattice-oxygen-vacancysite mechanism (LOV). This mechanism was proposed very recently 14 , in which the adsorbates are located at the sites of the lattice-oxygen vacancy. The last one, shown in Fig. 5c, is the metal-and-lattice-oxygen-vacancy-site mechanism (MLOV). Here, we consider the adsorbates to be at both TM sites with different valence states and lattice-oxygen vacancy sites 15 . Both LOV and MLOV scenarios consider the role of the lattice-oxygen vacancies at the surface. However, the effect of the metal sites with a high oxidation state generated during the OER is explicitly taken into account in the MLOV scenario, while in the LOV scenario, the occurrence of a high oxidation state is merely considered the cause for the formation of the oxygen vacancies. We built models for Li 2 Co 2 O 4 and Li 1 Co 2 O 4 to reveal the effects of the occurrence of the Co 4+ state generated during the OER, as shown in Supplementary Fig. 10, and calculated the free energies of the electrochemical OER process.
For Li 2 Co 2 O 4 , the Bader charge (q) of all Co sites at the surface is +1.47, representing Co 3+ according to previous theoretical work 12 , while for mixed-valent Li 1 Co 2 O 4 , there are two different Co sites on the surface: the Co 3+ site (q = +1.49) and the Co 4+ site (q = +1.60) 12 . Figure 5d Fig. 5g-i. In the MLOV route, the Gibbs free-energy difference (ΔG) of step 1 related to the formation of oxygen vacancies is much lower for Li 1 Co 2 O 4 than for Li 2 Co 2 O 4 ; thus, the rate-limiting step is changed to the adsorption of *OH in MLOV.
For Li 1 Co 2 O 4 , the ΔG of step 1 for Co 4+ is~0.3 eV lower than that for Co 3+ in the MLOV route, indicating again that the lattice oxygen around Co 4+ is very unstable and is easily removed from the lattice, forming *OO intermediates. The different ratelimiting steps between Co 3+ and Co 4+ also originate from the different binding energies of *OH. It is worth noting that although both the LOV and MLOV scenarios have considered the effects of the oxygen vacancy, the calculated total overpotential for LOV in Li 1 Co 2 O 4 is much higher than that for MLOV, indicating the important role of the metal Co 4+ site in the OER activity. If we replace distinguishable Co 3+ and Co 4+ with an average Co valence state, the LOV scenario is more favorable than the MAE route, as found previously 14 . The occurrence of highvalent Co 4+ ions in the Co oxide leads to an increase in the O 2p band center, whose positive effects on the OER activity were well studied previously 14,40 .

Discussion
The OER activity of Li 2 Co 2 O 4 and many other Co oxide systems was attributed to the presence of Co ions which have their e g shell occupied by an electron 41 . Starting with an LS Co 3+ state in the as-prepared material, it was proposed that the e g shell acquires an electron by considering the formation of a CoO 5 pyramidal coordination due to the loss of oxygen at the sample surface. We found no evidence for this in our experiments. First, we would like to point out that a loss of oxygen should decrease the valence of the Co ions; i.e., part of the Co 3+ should be converted into Co 2+ . However, our Co-L 2,3 spectra of the Li 2 Co 2 O 4 catalyst material before and after the OER do not show features that otherwise could have indicated the presence of Co 2+ species. As shown in Supplementary Fig. 4f, a Co 2+ ion would have a sharp peak at 777.8 eV, which is totally absent in our Li 2 Co 2 O 4 spectra both before and after the OER. In addition, our DFT calculations indicate that oxygen vacancies under OER conditions are very unstable and that they quickly become filled by OH. Second, if the Co ion remains in the trivalent state when the CoO 5 pyramidal coordination is formed, then there will be not one electron transferred to the e g shell but two electrons. The associated high- 10  spin (HS) state for the t 2g 4 e g 2 configuration is well documented in the literature for Co 3+ ions with pyramidal local coordination [42][43][44][45][46] . The spectrum of such an HS CoO 5 pyramidal system is given in Supplementary Fig. 4d and is very different from our Li 2 Co 2 O 4 spectra. We thus can rule out the presence of e g electrons in our catalyst material before and after the OER. Our in operando and in vacuo SXAS spectra instead showed a pure LS Co 3+ state in Li 2 Co 2 O 4 before the OER and that it is converted to an LS Co 3.4+ state after the OER. This is also reconfirmed by comparing the spectra to those of known Co compounds in the literature. Supplementary Fig. 4c shows that the spectrum of LiCoO 2 , which is an LS Co 3+ system, is identical to that of the as-prepared Li 2 Co 2 O 4 . Supplementary  Fig. 4b displays that the spectrum of Li 0.66 CoO 2 , which is an LS Co 3.34+ material, is very similar to that of Li 2 Co 2 O 4 after the OER.
Our DFT calculations indicate that the Co 4+ site spontaneously created under OER conditions is mainly responsible for the high OER activity. It is well known that the O 2p to Co 3d charge-transfer energy Δ (defined as the energy difference between the center O 2p band and the upper-Hubbard band (UHB), as shown in Supplementary Fig. 11) decreases with increasing valence state of the TM ion 47 . For Co 2+ , it is positive 48,49 ; for Co 3+ , it is close to zero 50 ; and for Co 4+ , it can be negative 27,28,51 . Consequently, the covalency increases, i.e., from rather ionic for Co 2+ , to highly covalent for Co 3+ , and further to Co 4+ , in which the holes mainly reside on the oxygen ligands rather than on the Co 27,28 . This trend can experimentally be directly observed in the O-K SXAS spectra: Supplementary Fig. 6 shows the spectral changes upon converting CoO to the asprepared Li 2 Co 2 O 4 , Li x CoO 2 , and Li 2 Co 2 O 4 after the OER and finally to BaCoO 3 . The figure displays how the spectral weight is shifting quickly to lower energies with increasing valence. The schematic density of states of Co 3+ and Co 4+ oxides (in the limit of no metal-ligand hybridization) is also shown in Supplementary  Fig. 11 for Co 3+ and Co 4+ oxides. The relation between the charge-transfer energies Δ 4+ and Δ 3+ is Δ 4+ = Δ 3+ − U + δ, where U is the on-site Co 3d Coulomb repulsion and δ is an extra gain originating from the lattice relaxation. It was previously found that the shift of the valence band center toward E f upon converting Co 2+ /Co 3+ to the mixed Co 3+ /Co 4+    Therefore, for Co 4+ oxides, we have high unoccupied density states with O 2p ligand character close to the chemical potential in the ground state 28 , which is generally denoted as the ligand oxygen hole (LOH) 27,28 . Thus, the high OER activity of the electrochemical catalysts is intimately linked to the presence of large numbers of LOHs.
Nevertheless, one must also realize that an increase in the numbers of LOHs will make gradually the material unstable 54 . It is well known that Co 4+ oxides such as SrCoO 3 and Sr 2 CoO 4 can easily decompose or become off-stoichiometric, even in air; in particular, x cannot be reduced to zero for the Li x CoO 2 and Na x CoO 2 systems. We therefore suggest that perhaps a valence state of the Co ion in the range of +3.3 -+3.4 would be a good starting point to find Co oxide materials that are highly OER active and yet also structurally stable. La 0.6 Sr 0.4 Co 0.2 Fe 0.8 O 2.9 (Ref. 55 ), SrCoO 2.7 (Ref. 15 ), Pr 0.5 Ba 0.5 CoO 3−δ (Ref. 40  In summary, the origin of the high OER activity of the Li 2 Co 2 O 4 catalyst was revealed by our in operando SXAS measurements as a function of applied voltage and time. We have observed that an irreversible valence-state transition from Co 3+ to Co 3.4+ occurs under OER conditions. This transition is accompanied by a delithiation process. The low-spin state of the Co ions, however, remains unchanged. No complete change from the edge-shared network to the corner-shared network can be observed. The underlying mechanism of the OER activity was then revealed by DFT calculations, which indicated that the adsorbates located at the Co 4+ sites have a lower overall overpotential for the OER process than those located at the Co 3+ sites or oxygen vacancy sites. The microscopic origin is the dominant O 2p hole characteristic of the ground state of Co 4+ having a negative charge-transfer energy. In such cases, the O 2p hole is due to the strong Co 3d-O 2p covalence, which is different from the strong O 2p-O 2p bond occurring in peroxo-like (O 2 ) n− species. We suggest that the ideal valence state of the Co ions for high OER activity is Co 3.3+ -Co 3.4+ . This can be either prepared or spontaneously created under OER conditions. A higher content of Co 4+ may increase the activity but will also lead to a decrease in the structural stability of the catalyst material.

Methods
Synthesis. To obtain a very small average size of the particles that is suitable for in operando soft X-ray absorption experiments, Li 2 Co 2 O 4 was prepared by the citrate sol-gel method as follows: lithium acetate (≥99.9%, Alfa Aesar), cobalt acetate hexahydrate (≥99.9%, Alfa Aesar), citrate (≥99.5%, Aladdin), and urea (≥99%, Aladdin) were dissolved in distilled water with a mole ratio of 1:1:2:2. The solution was evaporated at 80°C with magnetic stirring for 4 h and then transferred to an oven at 180°C overnight to obtain a gel. Afterwards, the gel was calcined at 350°C for 48 h to obtain Li 2 Co 2 O 4 .
Physicochemical characterization. The structure and phases of Li 2 Co 2 O 4 before and after the OER were identified using grazing incidence X-ray diffraction (XRD, Bruker D8-Advance AXS diffractometer with Cu K α irradiation and an incidence angle of 1°) and Raman (Horiba XploRA confocal Raman Microscope with laser wavelength of 532 nm). It is worth noting that the laser power for the Raman experiment was only 0.2 mW to avoid a structure phase transition caused by the laser. The morphology of the samples was characterized by transmission electron microscopy (TEM, FEI Tecnai G2 F20 S-TWIN) and scanning transmission electron microscopy (STEM, FEI aberration corrected Titan G2 60-300 with convergence angle of~21.4 mrad and camera length of~145 mm). The elemental compositions were measured by inductively coupled plasma-optical emission spectrometry (Spectro Arcos SOP).
Electrochemical measurements at home institute. Electrochemical measurements were conducted using an electrochemical workstation (Metrohm Autolab PGSTAT 302N) with a standard three-electrode electrochemical cell. The catalysts loaded on carbon paper with a loading mass of 0.3 mg cm −2 via drop-casting, Hg/ HgO (1 M KOH) and Pt foil acted as the working, reference and counter electrodes, respectively. All electrochemical experiments were performed in freshly prepared O 2 -saturated KOH (pH = 12.5-14; 99.99%, Alfa Aesar). Polarization curves for Li 2 Co 2 O 4 were recorded at a scan rate of 10 mV s −1 , and cyclic voltammograms were recorded at a scan rate of 5 mV s −1 . Unless specifically mentioned, all electrode potentials used in this study were referenced to the RHE and iR-corrected to compensate for the effect of solution resistance.
Soft X-ray absorption spectroscopy. The in operando SXAS experiments at the O-K and Co-L 2,3 edges were carried out at the 11A beam line of the National Synchrotron Radiation Research Center in Taiwan using the TFY mode. NiO and CoO single crystals purchased from Matek Material Technologie & Kristalle GmbH were recorded simultaneously in a separate ultrahigh vacuum chamber in TEY mode to serve as relative energy calibration for the measurements at the O-K and Co-L 2,3 edges, respectively. The SXAS spectra of the as-prepared Li 2 Co 2 O 4 and Li 2 Co 2 O 4 after the OER were measured using both TEY and TFY modes simultaneously. A correction for the self-absorption effects in the TFY spectra was applied 57,58 . There are several reasons why we performed the operando timedependent TFY SXAS only at the O-K edge. First, in the conversion from Co 3+ to Co 4+ , the created holes mainly reside in the O 2p states; therefore, the O-K SXAS spectrum is more sensitive to them than the Co-L 2,3 SXAS. Second, the O-K SXAS spectrum for an edge-shared network presents separated the features for Co 3+ and Co 4+ as well as t 2g -and e g -related holes well, while the increase in Co valence mainly produced a slight energy shift in the TFY Co-L 2,3 SXAS. Finally, in terms of practicality, each Co-L 2,3 spectrum with reasonable statistics took more than 20 min, which is too slow to detect the time-dependent effects for the operando experiments.
For SXAS measurements, the Li 2 Co 2 O 4 catalyst powder was dispersed in ethanol and deionized water and then sonicated for 30 min. The ink was then dropcast on carbon paper with a loading mass of 0.3 mg cm −2 for ex situ SXAS experiments. For in operando experiments, the ink was dropped into the thin membrane window (100 nm silicon nitride with a 1 × 1 mm 2 area coated by 3 nm Ti/10 nm Au from Silson Ltd) with a loading mass of~1 mg cm −2 . This window was used as the working electrode and to separate the liquid and the ultrahigh vacuum environment. The in operando SXAS experiments were performed using an in situ electrochemical liquid cell 59,60 equipped with three electrodes (working, reference, and counter electrodes) under control by a VersaSTAT 3 potentiostat from Princeton Applied Research. Two platinum wires were used as the reference and counter electrodes. Here, we selected a Pt pseudoreference electrode due to space constrictions in the electrochemical cell and calibrated the potential to RHE following the procedure described by Kasem and Jones 61 . Freshly prepared O 2saturated 1.0 M KOH was used as the electrolyte, and the electrochemical liquid cell system also contained a liquid pump, an inlet, and an outlet tube for the electrolyte flow.
TFY was used as the detection method for the absorption signal in the in operando SXAS experiments. A photon escape depth of~200 nm is sufficiently large to overcome the liquid region and the membrane separating the liquid from the ultrahigh vacuum. The particle size and distribution of Li 2 Co 2 O 4 nanoparticles used for the SXAS experiments were determined by high-resolution transmission electron microscopy in Supplementary Fig. 8. The average size of particles was <20 nm, which can ensure sensitivity to the surface region of the catalyst material for SXAS measurements. Assuming that the active region for the OER reaction is within a depth of~5 nm from the surface 14,62 , it can be estimated that~80% of the TFY signal originates from this region. This is illustrated in Supplementary Fig. 9, which shows how the ratio of the surface-to-bulk contribution in the TFY gradually increases as the size of particles decreases.
Density function theory calculations. All DFT calculations were performed with the Vienna Ab initio Simulation Package 63,64 using projector-augmented wave pseudopotentials and the Perdew-Burke-Ernzenhof 65 exchange correlation functional. The energy cutoff of the plane wave was 500 eV. To describe the strong correlation of the localized Co 3d states, the Hubbard U model 66 was applied, and the value of U eff (=U − J) was set to 3.52 eV according to previous work 12 . The experimental lattice constants were adopted, and the geometries were relaxed until a maximum threshold force of 0.02 eV/Å was fulfilled. For computational efficiency, a 3 × 3 × 1 Monkhorst-Pack k-point mesh was used for all calculations. The systems were initiated with Co atoms in a ferromagnetic configuration, which was allowed to evolve during the calculations. A 1 × 1 primitive cell (lattice constants are 7.9825 Å) was employed to build periodic slab models, with four Co sites per surface. There were eight atomic layers in the slab models, and four layers at the bottom were fixed during the relaxation. The thickness of vacuum spacing perpendicular to the surface was~20 Å to prevent spurious interactions.
The overpotentials for Li 2 Co 2 O 4 surfaces were calculated considering three scenarios of the reaction mechanism 12,14,15 . Elementary steps and all the corresponding equations considered for three scenarios of the reaction mechanism are provided in the supplementary information. The Gibbs free-energy differences were calculated using the computational hydrogen electrode model under standard conditions 12 . We considered voltage applied U RHE = 0 and 1.23 V. The zero-point energy (ZPE) and entropy corrections were calculated according to the literature 12,15 (Supplementary Table 2). The experimental Gibbs free formation energies and entropic contributions of H 2 O and H 2 under standard conditions (T = 298 K and P = 1 bar) were obtained from the CRC Hanbook 67 . For the adsorbed species, the ZPEs were calculated for the (001) Li 2 Co 2 O 4 surface. The entropy corrections for the adsorbent on the surface were considered zero, since the main contribution to the entropy is due to the translational entropy. The theoretical overpotential (η) was defined from the Gibbs free-energy differences at each step.
Data availability