Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# 3D atomic-scale imaging of mixed Co-Fe spinel oxide nanoparticles during oxygen evolution reaction

## Abstract

The three-dimensional (3D) distribution of individual atoms on the surface of catalyst nanoparticles plays a vital role in their activity and stability. Optimising the performance of electrocatalysts requires atomic-scale information, but it is difficult to obtain. Here, we use atom probe tomography to elucidate the 3D structure of 10 nm sized Co2FeO4 and CoFe2O4 nanoparticles during oxygen evolution reaction (OER). We reveal nanoscale spinodal decomposition in pristine Co2FeO4. The interfaces of Co-rich and Fe-rich nanodomains of Co2FeO4 become trapping sites for hydroxyl groups, contributing to a higher OER activity compared to that of CoFe2O4. However, the activity of Co2FeO4 drops considerably due to concurrent irreversible transformation towards CoIVO2 and pronounced Fe dissolution. In contrast, there is negligible elemental redistribution for CoFe2O4 after OER, except for surface structural transformation towards (FeIII, CoIII)2O3. Overall, our study provides a unique 3D compositional distribution of mixed Co-Fe spinel oxides, which gives atomic-scale insights into active sites and the deactivation of electrocatalysts during OER.

## Introduction

Hydrogen has long been proposed as a clean energy carrier within sustainable energy infrastructure. Although water electrolysis is a key technology in the production of hydrogen, it remains inefficient, and there are many complex challenges to improve its efficiency. One of the major hurdles is the limitation in the performance of anode electrocatalysts, where the oxygen evolution reaction (OER) takes place1,2. Optimisation of OER electrocatalysts requires a detailed understanding of the correlation between the surface composition of electrocatalysts and their activity and stability. However, it is notoriously challenging to perform a full three-dimensional (3D) structural and chemical characterisation of the topmost atomic layers of electrocatalysts, especially for catalyst nanoparticles <100 nm in diameter. In addition, the electrocatalyst surfaces undergo drastic structural and compositional changes during OER. Therefore, to develop high-performance OER electrocatalysts, it is imperative to thoroughly evaluate the contribution made by individual atoms during reactions to the relationships between catalytic activity and stability.

Mixed 3d transition metal oxides, such as mixed Co-Fe spinel oxides, have attracted much attention in the context of OER electrocatalysts due to their high abundance, low cost and rich redox chemistry3,4,5; these characteristics make them attractive alternatives to the high-cost benchmark noble metal-based oxides, i.e., IrO2 and RuO2. Depending on the composition, two spinel structures can be formed: (i) spinel, whereby a divalent cation, e.g., CoII, is located at the tetrahedral site, and trivalent FeIII at the octahedral site, and (ii) inverse spinel, whereby CoII is located at the octahedral site and FeIII at both the tetrahedral and octahedral sites6. The addition of small amounts of Fe in Co3O4 has been found to reduce the overpotential, while excess Fe increases the overpotential7,8. However, the role of Fe of mixed Co-Fe oxides or (oxy)hydroxides in catalysing OER is poorly understood, being the subject of ongoing and intense debate4,5,9,10,11,12,13,14. Additionally, although surface chemical and structural rearrangement of Co-based spinel oxides has been recently observed4,15,16,17,18,19,20, the surface reconstruction or phase transformation responsible for the change in OER activity and stability has not yet been studied in-depth. Therefore, this study aims to (i) correlate changes in OER performance with structural and compositional evolution of Co2FeO4 spinel and CoFe2O4 inverse spinel, thereby elucidating their deactivation processes during OER, and (ii) pinpoint the role of Fe in the OER activity of mixed Co-Fe oxides.

In this work, we use atom probe tomography (APT), in conjunction with X-ray photoelectron spectroscopy (XPS), X-ray absorption spectroscopy (XAS), high-resolution transmission electron microscopy (HRTEM) and electrochemical impedance spectroscopy (EIS) to characterise the evolution of the oxidation state, structure and composition on the surfaces of Co2FeO4 and CoFe2O4 nanoparticles during cyclic voltammetry (CV) measurements under OER conditions. Comprehensive information regarding the surface state changes is obtained by the scale-bridging method, including oxidation state measurements of bulk volume and top surface layer (5–10 nm) of nanoparticles by XAS and XPS, respectively, along with nanoscale and atomic-scale elemental and structural characterisation of individual nanoparticles by APT and HRTEM. Our study reveals the presence of Co-rich and Fe-rich nanodomains, created by spinodal decomposition, in pristine Co2FeO4 and most likely in most mixed CoxFe(3-x)O4 spinel oxides when x is in the range of 1.1–2.7 due to the miscibility gap21,22. Interestingly, hydroxyl groups were trapped at the interface between the nanodomains, possibly yielding a significantly enhanced OER activity of pristine Co2FeO4 compared to CoFe2O4. During OER, different levels of Fe dissolution occur in the nanodomains of Co2FeO4, along with concurrent irreversible structural transformation towards CoIVO2, leading to a substantial decrease in the OER activity. In contrast, negligible Fe loss was observed for CoFe2O4. Instead, (FeIII, CoIII)2O3 was formed on the surface, further decreasing the OER activity of CoFe2O4. Overall, our 3D atomic-scale data, combined with X-ray- and electron-based microscopy and electrochemical data, show great promise for improving understanding of the complex structure-activity-stability relationships of electrocatalysts.

## Results

### Structure, size and morphology of spinel oxide nanoparticles

Co2FeO4 and CoFe2O4 nanoparticles were synthesised by a hydrothermal method (see Methods). Both pristine nanoparticles have the standard cubic spinel structure (Fd$$\bar{3}$$m23), as confirmed by X-ray powder diffraction (XRD) (Supplementary Fig. 1). The size of pristine Co2FeO4 and CoFe2O4 nanoparticles is 10.3 ± 2.6 nm and 10.4 ± 2.7 nm, respectively, and both have a spherical shape (Supplementary Figs. 2a–d and 3a–d). The lattice constants of Co2FeO4 and CoFe2O4 nanoparticles, as measured from the selected area electron diffraction patterns shown in Supplementary Figs. 2e and 3e, is 8.60 Å and 8.69 Å, respectively. The difference in the lattice constants originates from the differences between Co/Fe contents and their radius (Co3+ has a radius of 0.61 Å, which is slightly less than the Fe3+ radius of 0.65 Å24); this is consistent with the XRD data (in Supplementary Fig. 1), whereby the diffraction peaks of Co2FeO4 are shifted to higher 2θ values compared to those of CoFe2O4.

### Electrochemical performance

The electrocatalytic activity was measured by linear sweep voltammetry (LSV), using a scan rate of 10 mV/s on a rotating disk electrode (RDE), on Co2FeO4 and CoFe2O4 nanoparticles in the pristine state and after various CV cycles in 1.0 M KOH under OER conditions, Fig. 1a–d. Tafel slopes were derived from the LSV data, see Fig. 1e, f. The current density was normalised to surface areas determined by the Brunauer–Emmett–Teller (BET) method from N2 physisorption measurements (Supplementary Fig. 4, additionally, the current density normalised to the geometric surface area of glassy carbon electrodes was provided in Supplementary Fig. 5). Ohmic drop (iRs) correction (Rs extracted from Nyquist plots) was applied to compensate for a lowering of the actual potential resulting at the electrode as compared to the nominally applied one due to current flux in the highly resistive system5.

The LSV plots, shown in Fig. 1a, b, reveal that pristine Co2FeO4 exhibits a higher OER activity than pristine CoFe2O4, since the overpotential of Co2FeO4 (359 mV at 10 µA/cm2) is lower than that of CoFe2O4 (432 mV at 10 µA/cm2). Pristine Co2FeO4 has a Tafel slope of 43 ± 1 mV/dec (Fig. 1e), while pristine CoFe2O4 has a much larger Tafel slope of 79 ± 2 mV/dec (Fig. 1f), indicating that OER charge transfer kinetics are faster on pristine Co2FeO4 than on pristine CoFe2O4. The measured Tafel slope of pristine Co2FeO4 nanoparticles is also lower than most pristine Co3O4 and Co-based spinel oxide nanoparticles (~60 mV/dec)19,25. Increasing the number of CV cycles leads to a gradual deterioration in activity of both Co2FeO4 and CoFe2O4 (Fig. 1a, b, e, f). In particular, the Tafel slope of Co2FeO4 increases to 83 ± 2 mV/dec after 1000 cycles, which is almost double the Tafel slope of the pristine state, while the Tafel slope of CoFe2O4 increases slightly to 83 ± 1 mV/dec. Thus, despite the high OER activity of pristine Co2FeO4, its OER activity drops as the number of CV cycles increases, eventually reaching similar values as detected for the less active CoFe2O4.

Furthermore, Co2FeO4 exhibits pronounced redox couples during CV measurements (inset of Fig. 1c). Specifically, during the first CV cycle, two broad anodic peaks are observed at ~1.19 V (A1) and ~1.48 V (A2), possibly corresponding to the oxidation process of Co(II)/Co(III) and Co(III)/Co(IV), respectively26,27. The cathodic sweep exhibits a relatively strong cathodic peak at ~1.44 V (C2), which is usually attributed to the Co(IV)/Co(III) couple in Co3O426,27,28,29. The cathodic peak C1 at ~1.1 V for the Co(II)/Co(III) process26,27 is almost negligible after the first CV cycle, suggesting that the Co(II)/Co(III) process is likely not to be fully reversible. Additionally, the A2 and C2 peaks become less pronounced and nearly indiscernible after 1000 cycles, which indicates that the Co(III)/Co(IV) oxidation is likely irreversible. The gradual formation of irreversible Co(III) and Co(IV) surface species possibly results in the A2 peak gradually shifting to higher potentials, which leads to the increased Tafel slope (Fig. 1e)30. In comparison with Co2FeO4, nearly no redox couples were observed for CoFe2O4, with a slight anodic shift after 1000 cycles (dark purple curve, insert in Fig. 1d), most likely suggesting the occurrence of an irreversible oxidation process.

### Oxidation state on the surfaces

To investigate the reasons for the activity changes of Co2FeO4 and CoFe2O4, we first performed XPS to examine the oxidation state of Co and Fe on the surface of Co2FeO4 and CoFe2O4 in their pristine state as well as after 100, 500, and 1000 cycles. XPS measures the average oxidation state of approx. 100 µm × 100 µm × 5 nm of the surface region of the nanoparticles deposited on glassy carbon. Given the closeness of 2p1/2 and 2p3/2 peak locations for Co(II), Co(III) and Co(IV)31,32,33,34, the satellite features and their intensity change during OER, i.e., 786.5 eV for CoO-like Co(II)35 and 789.5 eV for Co3O4-like Co (II, III), were analysed (peak fitting shown in Supplementary Fig. 6a). We can see from Fig. 2a that the intensity of CoO-like Co(II) satellite features decreases after 100 cycles, suggesting the oxidation of Co(II) to Co(III). As the number of CV cycles increases, the contribution of CoO-like Co(II) decreases significantly, and Co3O4-like Co increases (as indicated by the depth analysis using peak deconvolution of the Co 2p peak35 shown in Supplementary Fig. 6b). Our CV data for Co2FeO4, shown in Fig. 1c, indicate an irreversible oxidation of Co(II)/Co(III) in the first cycle and of Co(III)/Co(IV) after 1000 cycles. While the presence of Co(IV) is difficult to be confirmed by XPS, the possibility cannot be excluded since the spectrum of Co3O4-like Co is similar to that of CoO2-like Co(IV)36. Thus, we conclude that Co(III) and Co(II) are present on the surface of Co2FeO4 after 100 and 500 cycles, while Co(III) and Co(IV) are likely present after 1000 cycles. In contrast to Co2FeO4, the CoO-like Co(II) satellite peaks of CoFe2O4, in Fig. 2b, decreases less significantly as the CV cycles increases. Co(III) seems to be present on the surface of CoFe2O4 after 500 and 1000 cycles, as shown in Supplementary Fig. 6c. In addition to Co, the Fe 2p spectra, shown in Fig. 2b, d remains unchanged, indicating that Fe(III) is present on the surface of both Co2FeO4 and CoFe2O4 as the number of CV cycles increases, in agreement with observations in most previous studies4,13.

To further verify the irreversible change in the oxidation state of Co2FeO4 in their pristine state and after 1000 cycles, we performed XAS that allows spectral detection of a bulk volume of approx. 100 μm × 300  μm × 1 mm (penetration depth) of nanoparticles deposited on glassy carbon (Supplementary Note 1 and Supplementary Fig. 7). We observed a subtle shift of Co K-edge towards higher energy values (Supplementary Fig. 7a), possibly suggesting that only a small volume fraction, potentially on the surface regions, has an increase in oxidation state to Co(IV), which is in agreement with the XAS data of Calvillo et al.20. Additionally, more octahedrally coordinated and less tetrahedrally coordinated Co37 was observed after 1000 cycles (inset, Supplementary Fig. 7a). By relating XPS and electrochemical data (Figs. 1c and 2a), we, therefore, speculate that tetrahedrally coordinated Co(II) irreversibly oxidises to octahedrally coordinated Co(III) or (IV) in the course of 1000 cycles, yielding a decrease in activity.

### Structural changes

Next, HRTEM was employed to evaluate the structural changes of the Co2FeO4 and CoFe2O4 nanoparticles after OER, first in their pristine state and then after 100 and 1000 cycles. Figure 3a shows a background-subtracted HRTEM image of pristine Co2FeO4, viewed along the [001] zone axis. Interestingly, the interplanar spacing of d220 varies slightly in different regions, as presented in Fig. 3a (details in Supplementary Fig. 8). After 100 cycles, a distinct structural change occurs on the surface of Co2FeO4, as highlighted by the blue-dotted areas in Fig. 3b (more HRTEM images shown in Supplementary Fig. 9a, b). The Fast Fourier filtered transform (FFT) images, shown in insets of Fig. 3b, indicate that the motifs observed on the surface (blue-dotted region) are aligned at 45° to the atomistic arrangement of the spinel structure. The reflection spots from the surface region (Fig. 3b, bottom inset) correspond well with β-CoOOH (R$$\bar{3}$$m, hexagonal38, Supplementary Table 1), which agrees with observations of a previous study19. Therefore, we hypothesise that epitaxial growth of (Co, Fe)OOH occurs on Co2FeO4, with an orientation relationship of (010) Co2FeO4//(1-101) (Co, Fe)OOH. A previous study proposed that CoII ions at the tetrahedral site are oxidised to form amorphous CoIII oxyhydroxides16. Here, we observed crystalline CoIII oxyhydroxides, possibly formed by crystallisation of the amorphous CoIII oxyhydroxides in the absence of potentials and electrolytes (also under high vacuum in TEM). Furthermore, after 1000 cycles, a 5–6 nm surface region, highlighted by the dark-blue-dotted area in Fig. 3c (Supplementary Fig. 9c, d), undergoes a phase transformation, since it contains a distinct lattice fringe from the [112]-oriented Co2FeO4 spinel oxide. The reflection spots in the FFT image (Fig. 3c, bottom inset) match well with the CoO2 phase (P$$\bar{3}$$m1, hexagonal39, Supplementary Table 1), with an octahedrally coordinated Co (IV). The formation of CoO2 is also confirmed by additional reflection spots that correspond well to (11–20) CoO2 in the selected area electron diffraction (SAED) pattern after 1000 cycles (Supplementary Fig. 2f); the SAED pattern was recorded from a 500 nm × 500 nm area containing more than 100 nanoparticles. We observed an increase in octahedrally coordinated Co by XANES (Supplementary Fig. 7a, inset), the presence of irreversible Co(IV) by electrochemical data (Fig. 1a) and possible presence of Co (IV) by XPS (Fig. 2a). These results most likely imply an irreversible transformation towards (Co, Fe)O2 phase on the surface of Co2FeO4 after 1000 cycles. Additionally, CoO2 is the stable phase at higher potentials, in accordance with the Co Pourbaix diagram40.

In contrast with Co2FeO4, we observed no significant structural changes on the CoFe2O4 nanoparticle surface after 100 cycles (Fig. 3d, e). After 1000 cycles, a structural transformation of the surface of [001]-orientated CoFe2O4 spinel nanoparticles was discerned, as indicated by the purple dotted lines in Fig. 3f (Supplementary Fig. 9e, f). The reflection spots in the bottom inset of Fig. 3f correspond to either [412]-oriented FeOOH (I4/m, tetragonal41, Supplementary Table 1) or [0001]-orientated Fe2O3 phase (R$$\bar{3}$$c, hexagonal42), with the latter having a better fit. Also, the SAED pattern of CoFe2O4, in Supplementary Fig. 3f, reveals additional reflection spots after 1000 cycles, indicating the presence of (30-30) Fe2O3 or (202) FeOOH. According to the Fe Pourbaix diagram43, Fe2O3 forms as the potential increases. Thus, the surface of inverse spinel CoFe2O4 is thought to undergo a phase transformation to the (Fe, Co)2O3 phase after 1000 cycles.

Notably, we observed an amorphous layer on the surfaces of aggregated Co2FeO4 and CoFe2O4 nanoparticles after 100, 500 and 1000 cycles (exemplified in Supplementary Fig. 10). However, these amorphous layers were most likely the result of carbon contamination under electron beam in TEM (details in Supplementary Fig. 10).

### Compositional evolution and correlation with electrochemical performance

Although the (Co, Fe)OOH, which is observed on the surface of Co2FeO4 (Fig. 3b), can be regarded as an active intermediate for OER27, the activity of Co2FeO4 decreased after 100 cycles (Fig. 1a, e). The cause of the decrease in the OER activity of Co2FeO4 (Fig. 1a, e) at the beginning of OER remains unclear. Other factors, such as chemical composition change, potentially lead to decreases in OER activity. Therefore, we used APT44, a mass-spectrometry technique with sub-nanometre spatial resolution in three dimensions45,46, to investigate the compositional evolution of oxide nanoparticles after OER. All electrochemical measurements were carried out in proton-free, deuterated electrolyte (i.e., 1.0 M KOD in D2O in order to use APT to examine the distribution of hydroxyl groups after OER47.

Figure 4a exemplifies a cross-sectional atom map of pristine Co2FeO4 nanoparticles embedded in a Ni matrix (APT specimen preparation is detailed in Supplementary Note 2/Supplementary Fig. 11 and additional APT data is shown in Supplementary Fig. 12a). The oxide nanoparticles were detected by APT in the form of O ions and Co- and Fe- containing complex molecular ions (see mass spectra in Supplementary Fig. 13). All Co- (in blue) and Fe- (in magenta) containing molecular ions were shown as CoOx and FeOy respectively, in Fig. 4a (separate Co, Fe, O and Ni atom maps are shown in Supplementary Fig. 14). Our detailed APT analysis of 48 pristine Co2FeO4 nanoparticles (Fig. 4a and Supplementary Fig. 12a) reveals that 26 of them have nanoscale compositional modulation, while the remaining 22 exhibits a relatively uniform elemental distribution, as exemplified by Figs. 4e and 5a, selected from the black and red dashed boxes in Fig. 4a and Supplementary Fig. 12a, respectively; we term these nanoparticles ‘segregated’ and ‘non-segregated’ Co2FeO4 nanoparticles, respectively. The 2D Fe composition map (Fig. 4i), plotted from the nanoparticle data of Fig. 4e, clearly reveals separate Fe-rich and Co-rich nanodomains, whose dimensions are in the range of 4–5 nm, in the segregated pristine Co2FeO4 nanoparticle. In contrast, non-segregated pristine Co2FeO4 nanoparticles have uniformly distributed Fe and Co (Fig. 5a,e).

The compositions of the Fe-rich and Co-rich nanodomains in pristine Co2FeO4, obtained by 1D composition profiles in Supplementary Figs. 15a and 16a and data from 25 segregated nanoparticles in Supplementary Fig. 12a, were plotted as composition histograms of Co/Fe and oxygen/(Co+Fe) (termed O/M) ratios in Fig. 4m, n. The average ratios of Co/Fe and O/M from all nanoparticles are detailed in Tables 1 and 2. Note that the oxygen content was normalised by the value measured by using H2 temperature-programmed reduction (H2 TPR) and the effect of laser pulse energy on the measurement of oxide stoichiometry48 was detailed in Supplementary Note 3 and Supplementary Figs. 22 and 23. The Fe-rich and Co-rich nanodomains have Co/Fe ratios of 1.2 ± 0.1 and 2.2 ± 0.1, respectively, and O/M ratios of 1.2 ± 0.1 and 1.3 ± 0.1, respectively. Thus, although the Fe-rich and Co-rich nanodomains have similar oxygen contents, they have significantly different Co/Fe ratios; the stoichiometry of the ‘segregated’ Co2FeO4 nanoparticle is Co2.1Fe0.9O3.9 in the Co-rich nanodomain and Co1.65Fe1.35O3.6 for the Fe-rich nanodomain. Similarly, the stoichiometry of the non-segregated nanoparticles is Co1.7Fe1.3O3.3 (based on values given in Tables 1 and 2). The formation of Fe-rich and Co-rich nanodomains in the pristine Co2FeO4 nanoparticles is most likely the result of spinodal decomposition that is driven by the miscibility gap in the composition range 0.37 < Co/(Co+Fe) < 0.9 at temperatures below 700 °C21,22. According to the CoFe2O4-Co3O4 phase diagram21,22, Co2FeO4 is expected to decompose to Co1.4Fe1.6O4 (Co/Fe = 0.875) and Co2.4Fe0.6O4 (Co/Fe = 4). The discrepancy of compositions between our study and previous work possibly arises from the fact that these nanoparticles do not reach thermodynamic equilibrium after synthesis, and we expect the phase stability of nanoparticles to deviate from that of bulk materials in previous studies21,22. Despite this, we unambiguously reveal nanoscale compositional modulation of Co2FeO4 nanoparticles. In contrast, we did not observe any segregation for the CoFe2O4 nanoparticles, as shown in the atom map of Fig. 6a and the 2D Fe compositional map of Fig. 6e, since the Co/(Co+Fe) ratio (0.33) falls outside the composition window of spinodal decomposition21,22.

Next, we examined elemental redistribution of ‘segregated’ Co2FeO4 nanoparticles after 100, 500 and 1000 cycles under OER conditions (Fig. 4b–d, f–h). Importantly, we observed that the hydroxyl groups, shown as dark blue spheres in 2D Fe compositional profiles of Fig. 4j–l, are preferentially located at the interface between Fe-rich and Co-rich nanodomains after 100 and 500 cycles, and dominantly in Fe-rich regions after 1000 cycles. The trapping of hydroxyl groups at the interface of two nanodomains is most likely induced by the elastic strain that results from the difference in lattice constants of these two domains. This is confirmed by the difference in interplanar spacing of d220 measured from two regions of pristine Co2FeO4 nanoparticles (Fig. 3a and Supplementary Fig. 8), which is ascribed to the difference in Co/Fe content of these nanodomains (similar to the difference of lattice constants of pristine Co2FeO4 and CoFe2O4 observed by XRD shown in Supplementary Fig. 1). The hydroxyl groups can be considered as ‘fingerprints’ that indicate the regions where OER occurs49. Therefore, we hypothesise that interfaces between two nanodomains provide active sites and accelerate OER kinetics, thereby possibly contributing to the high OER activity of pristine Co2FeO4 nanoparticles.

More importantly, we observed a dramatic compositional change within the nanodomains of Co2FeO4 as the number of CV cycles increased (Supplementary Figs. 15b–d and 16b–d along with all other nanoparticle data in Supplementary Fig. 12b–d were summarised in Fig. 4m, n and Tables 1 and 2; the number of nanoparticles for composition histograms was detailed in Supplementary Table 2). Specifically, the Co/Fe ratio in the Co-rich nanodomains remained at ~2.2 after 100 cycles but increased to ~2.8 after 500 cycles and 2.9 after 1000 cycles (Table 1). This result suggests a gradual Fe loss in the Co-rich nanodomains during OER (as also confirmed by 1D profiles of atomic counts in Supplementary Fig. 15e–h). The O/M ratio in Co-rich nanodomains increases gradually to ~1.4 after 100 and 500 cycles, and ~1.6 after 1000 cycles (Table 2 and Fig. 4n). The O/M ratio in the Fe-rich nanodomains also increases to ~1.5 after 1000 cycles. The gradual increase in the O/M ratio of the segregated Co2FeO4 nanoparticles in both Co-rich and Fe-rich nanodomains suggests the occurrence of oxidation in both nanodomains. A more pronounced Fe loss was observed in the Co-rich nanodomains compared to that of the Fe-rich nanodomains (Table 1 and Fig. 4m), possibly suggesting that OER occurs more rigorously in the Co-rich nanodomains than that in Fe-rich nanodomains.

For the ‘non-segregated’ Co1.7Fe1.3O3.3 nanoparticles, we observed 2–3 nm oxygen-rich surface regions after 100 cycles by comparing the 2D Fe and O compositional maps (Fig. 5f, j) of the same nanoparticle. The O/M ratio in the oxygen-rich regions increases from 1.4 ± 0.2 after 100 cycles to 1.8 ± 0.2 after 1000 cycles (see Table 2 and Fig. 5n, which was measured from the 1D concentration profiles in Supplementary Figs. 17 and 18, and all other nanoparticle data in Supplementary Fig. 12b–d). Based on the O/M ratios listed in Table 2, we speculate that the oxygen-rich surface regions possibly correspond to (Co, Fe)OOH after 100 cycles, and (Co, Fe)O2 after 1000 cycles, as observed by HRTEM (Fig. 3b, c and Supplementary Fig. 9a–d). Additionally, we observed a subtle Fe loss in the oxygen-rich region after 1000 cycles, as the Co/Fe ratio increases (Table 1 and Fig. 5m). Previous work also observed an increasing Co/Fe ratio on the surface of CoFe0.75Al1.5O4 by electron energy loss spectroscopy5, attributing it to the formation of Co oxyhydroxide. Here, we speculate that the increasing Co/Fe ratio is most likely due to Fe loss during structural transformation under the OER conditions. Concurrent structural transformation and Fe dissolution most likely lead to the overall reduction in OER activity of Co2FeO4 after 100 and 500 cycles (Fig. 1a, c). After 1000 cycles, surface formation of stable (CoIV, FeIII)O2 further decreases the activity of Co2FeO4.

For comparison with Co2FeO4, the compositional changes of the CoFe2O4 nanoparticles after 100, 500 and 1000 cycles are provided in Fig. 6b–l. The oxide stoichiometry of pristine CoFe2O4 is measured as CoFe2O3.6, based on values in Tables 1 and 2 (measured from 39 nanoparticles in Supplementary Fig. 19a and Supplementary Table 2). No evident elemental redistribution was observed after 100 cycles, while 2–3 nm oxygen-rich regions were seen on the surface of CoFe2O4 after 500 and 1000 cycles, as indicated by the 2D Fe and O compositional maps in Fig. 6g, h, k, l. The O/M ratio in the oxygen-rich regions increased to ~1.7 after 500 cycles, reaching up to ~1.8 after 1000 cycles, while Co/Fe increased only slightly (Tables 1 and 2 and Fig. 6m, n, derived from 1D concentration profiles in Supplementary Figs. 20 and 21 and all other data in Supplementary Fig. 19b–d). The XPS and XANES data suggest that CoII was oxidised to CoIII while FeIII remained stable even after 1000 cycles (Fig. 2 and Supplementary Fig. 7). Therefore, we speculate that the oxygen-rich regions possibly correspond to the (FeIII, CoIII)2O3 phase, as also observed by HRTEM (Fig. 3f and Supplementary Fig. 9e, f).

### Electrochemical sub-processes during OER

To further understand the deactivation of both spinel oxide nanoparticles, we employed EIS to reveal the electrochemical sub-processes during OER. The Nyquist plots, in Fig. 7a, b, show two distorted semicircles for Co2FeO4, whereas one semicircle is observed for the inverse spinel CoFe2O4. The distribution of relaxation times for Co2FeO4 (inset of Fig. 7a) reveals three distinct peaks, and one peak is discerned for CoFe2O4 (inset of Fig. 7b). Figure 7c, d contain the resistances and capacitances for both spinel oxides in the pristine state and after 100, 500 and 1000 cycles (derived by equivalent circuit fitting50,51, more details in Supplementary Note 4 and Supplementary Figs. 24 and 25). The equivalent circuit for Co2FeO4, shown in Fig. 7c, contains the double layer capacitor C1 and the pseudocapacitors C2 and C3. The corresponding pseudocapacitive properties are based on changing oxidation states of electrochemically accessible cobalt sites and adsorptive discharge of oxygen-containing species in the electrolyte52,53. So, in course of a catalytic cycle, intermediate catalytic (re-)transformations occur via pseudocapacitive charging/discharging through the faradaic resistors R1 and R254. In parallel, the OER reaction proceeds via OH-to-O2 conversion steps at the solution side55. We can see from Fig. 7c that the resistance increases with the number of CV cycles, particularly for R1, accompanied by a slightly decreasing C2. This result suggests inhibited kinetics of adsorption of OH and an occurrence of irreversible oxidation of Co, which is consistent with our TEM and APT observation of structural transformation towards (Co, Fe)OOH and (Co, Fe)O2 as the number of CV cycles increases. In contrast, no pseudocapacitive behaviour was observed for CoFe2O4 (Fig. 7b), which may relate to its low OER activity compared to Co2FeO4, as indicated by Tafel plots (Fig. 1e, f). For both Co2FeO4 and CoFe2O4, the increasing faradaic resistances (Fig. 7c, d) explain that the overpotential increases with the number of CV cycles (Fig. 1a, b), which is most likely arisen from irreversible structural transformation towards inactive phases, as revealed by our TEM (Fig. 3c, f and Supplementary Figs. 2f and 3f) and APT data (Tables 1 and 2).

## Discussion

Our study demonstrates that the deactivation process of Co2FeO4 and CoFe2O4 is closely associated with their structural and compositional evolution during OER, as schematically summarised in Fig. 8. Our APT data (Fig. 4i, m) unprecedentedly reveals ‘segregated’ Co2FeO4 whose compositional modulation is driven by spinodal decomposition21,22, and ‘non-segregated’ Co2FeO4 in pristine Co2FeO4 (Fig. 8a, b). We speculate that such composition modulation is present in mixed CoxFe(3-x)O4 spinel oxides when x is in the range of 1.1–2.7 due to the miscibility gap. The interface between the Co-rich and Fe-rich nanodomains of ‘segregated’ Co2FeO4 (Fig. 4j, k) traps the hydroxyl groups, possibly due to the elastic strain induced by the difference in lattice constants (Fig. 8c). At the onset of OER, (CoIII, FeIII)OOH is formed epitaxially on the surface of non-segregated Co2FeO4 (Fig. 3b and Supplementary Fig. 9a, b), and possibly in the nanodomains of segregated Co2FeO4 nanoparticles (Fig. 8d). Despite the formation of active (CoIII, FeIII)OOH13, the activity of Co2FeO4 decreases as a consequence of the irreversible structural transform towards (CoIV, FeIII)O2 (Fig. 3c and Supplementary Fig. 2f) along with gradual loss of Fe (Tables 1 and 2, Figs. 4m, n and 5m, n) via the formation of soluble ferrate ions FeO42- in alkaline electrolytes13,43,56 (Fig. 8c, e–f). Therefore, we conclude that the concurrent structural transformation towards a stable (CoIV, FeIII)O2 phase and Fe dissolution lead to a decrease in OER activity of Co2FeO4 as the number of CV cycles increases. For CoFe2O4, CoII present on the surface of pristine sample (Fig. 8g) is  oxidised to CoIII at the onset of OER (Fig. 8h), as shown by XPS data (Fig. 2b). After 1000 cycles, the (FeIII, CoIII)2O3 phase, whose dimension is 4–5 nm, is likely formed on the surface of CoFe2O4 (Fig. 8i), as evidenced by both HRTEM and electron diffraction pattern (Fig. 3f and Supplementary Fig. 3f) and oxygen-rich regions revealed by APT (Fig. 6k, l, n), which slightly decreased the OER activity.

Importantly, our study demonstrates that OER is catalysed concurrently by multiple active regions in a single nanoparticle. Firstly, we observe that CoII catalyses the OER of both Co2FeO4 and CoFe2O4, since the Co oxidation state increases (indicated by XPS in Fig. 2 and XANES in Supplementary Fig. 7). The CoII in the tetragonal sites (termed $${{Co}}_{{Td}}^{{II}}$$) are thought to be active sites as they are responsible for the formation of active CoIIIOOH species17; this is confirmed by our TEM data (Fig. 3b and Supplementary Fig. 9a, b) and the decreased number of tetrahedral sites by XANES data (inset, Supplementary Fig. 7a) as well as EIS data (Fig. 7a) after OER. Secondly, and most importantly, our results demonstrate that, Fe, when in the presence of $${{Co}}_{{Td}}^{{II}}$$, plays a key role in the OER activity. Specifically, the amount of Fe loss is thought to indicate how rigorously OER occurs. Our APT data (Table 1) reveals that different regions have various levels of Fe dissolution, i.e., oxygen-rich regions of CoFe2O4 < oxygen-rich regions of non-segregated Co2FeO4  ≤ Fe-rich nanodomains in segregated Co2FeO4 < Co-rich nanodomains in segregated Co2FeO4. The number of $${{Co}}_{{Td}}^{{II}}$$ sites in these regions also increases in a similar manner (as indicated by Supplementary Table 3, estimated according to refs. 57,58). These observations provide a strong indication that Fe, which can actively catalyse OER, seems to be associated with the number of $${{Co}}_{{Td}}^{{II}}$$ sites. Therefore, we hypothesis that the presence of Co oxyhydroxide is imperative for the activation of Fe, since FeOOH has low electrical conductivity and stability at lower potentials56, and CoIIIOOH provides an electrically conductive support9. A previous study reported that Fe can serve as an indirect active site, for example by changing the spin and charge state of Co11, or by assisting5 or stabilising the active CoIIIOOH intermediate59, thereby enhancing OER activity. Another study proposed that di-µ-oxo bridged Co-Fe sites act as active sites above a transition voltage, below which di-µ-oxo bridged Co-Co sites catalyse OER4. Our study indicates that Fe promotes the formation of active species for OER, possibly CoIIIFeIIIOOH, whose OER activity is significantly better than that of CoIIIOOH13 (although its activity drops rapidly due to Fe dissolution). Therefore, in comparison to CoFe2O4, Co2FeO4 has a higher OER activity most likely due to the formation of more CoIIIFeIIIOOH yielded by its optimal ratio of Fe content to $${{Co}}_{{Td}}^{{II}}$$ sites. Also, the interface between the nanodomains of segregated Co2FeO4 traps hydroxyl groups, providing additional active regions for OER, thereby further enhancing the OER activity of Co2FeO4. Therefore, we conclude that the presence of CoIIIFeIIIOOH, promoted by Fe and $${{Co}}_{{Td}}^{{II}}$$, coupled with the nanosized defect features, leads to Co2FeO4 having an increased OER activity. This potentially explains how the addition of a small amount of Fe improves the OER activity of mixed Co-Fe spinel oxides7,8.

In summary, our study provides atomic-scale insights into the evolving surface structure of Co2FeO4 and CoFe2O4 nanoparticles during OER and reveals how those structural and compositional changes alter the activity and stability. We demonstrate the importance of 3D atomic-scale imaging and quantitative compositional analysis of nanoparticles in both their pristine state and at various stages of electrochemical reaction when seeking to understand their activity and stability. We believe that APT, when combined with X-ray- and electron-based characterisation techniques, has enormous potential to better understand the reaction and degradation mechanisms of oxide or metallic catalyst nanoparticles during important reactions, such as OER and CO2 reduction.

## Methods

### Synthesis of nanoparticles

Iron (III) nitrate (Fe (NO3)3·9H2O), cobalt (II) nitrate (Co(NO3)2·6H2O), ammonia (NH3·H2O) and polyethene glycol (PEG, Mn = 400) were purchased from Sinopharm Chemical Reagent (Shanghai, China). The CoFe2O4 nanoparticles were prepared by dissolving 3.2 g of Fe(NO3)3·9H2O, 1.2 g of Co(NO3)2·6H2O and 0.2 g of polyethene glycol (PEG, Mn = 400) in 40 mL of ultrapure water (0.055 μS cm−1) under vigorous stirring for 30 min60. For Co2FeO4 nanoparticles, 1.6 g of Fe(NO3)3·9H2O, 2.3 g of Co(NO3)2·6H2O and 0.2 g of PEG were mixed in 40 mL of ultrapure water. 5 mL of ammonia diluted with 5 mL of ultrapure water was slowly added dropwise into the solution mixture. The obtained suspension was subsequently transferred into a 100 mL Teflon-lined stainless autoclave, maintained at 180 °C for 3 h. The product was washed several times with ultrapure water via centrifugation, dried in an oven at 80 °C for 12 h.

### XRD measurements

The XRD analysis of the pristine Co2FeO4 and CoFe2O4 nanoparticles was performed on an Rigaku Ultima IV diffractometer with Cu Kα radiation (λ = 0.15418 nm) at a scanning speed of 4°/min, scanning step of 0.02° and operating voltage of 40 kV. The XRD data was given in Supplementary Fig. 1.

### TEM measurements

TEM and HRTEM experiments were carried out in an aberration-corrected JEOL JEM-2200FS operating at 200 kV, and TEM/EDX data was acquired with an Oxford X-max detector. The TEM samples were prepared by dispersing a small amount of nanoparticles into anhydrous ethanol via ultrasonication, followed by dropping nanoparticle solutions on Cu TEM grids and dried at room temperature. The HRTEM images were processed by using Gatan Digital Micrograph. Additional TEM images were shown in Supplementary Figs. 2, 3, and 810.

### Electrochemical measurements

Electrochemical measurements were performed in a three-electrode system at an electrochemical workstation (PalmSens3), where a Pt wire and Ag/AgCl (3 M KCl) served as counter and reference electrodes. The OER performance was studied by using a rotating disk electrode (10 mm diameter, 0.785 cm2). The glassy carbon electrode was first cleaned and polished to a mirror finish with 50 nm Al2O3. In all, 32 μL of dispersion was transferred onto the glassy carbon disk and then dried at room temperature. The dispersion was prepared by dispersing 5 mg of nanoparticle powder in 1 mL ultrapure water, followed by ultrasonication for 30 min. The LSV curves were recorded with a scan rate of 10 mV/s in 1.0 M KOH solution from 0 V to 0.8 V (vs Ag/AgCl) at a rotating speed of 1600 rpm. The CV measurements were performed at a scan rate of 50 mV/s from 0 V to 0.65 V (vs Ag/AgCl). Electrochemical data normalised to electrode geometric surface area was provided in Supplementary Fig. 5. Another set of electrochemical measurements was carried out at the same conditions except for the KOD solution in D2O for APT specimen preparation and measurements. The deuterium oxide (D2O with 99.9 at.% D) and potassium deuteroxide solution (40 wt.% KOD in D2O with 98 at.% D) were purchased from Sigma-Aldrich. EIS was performed under OER conditions by applying a sine wave signal with a 10 mV amplitude in the frequency range from 6 kHz to 0.2 Hz after equilibrating 5 s at 0.6 V vs. Ag/AgCl for Co2FeO4 and 0.7 V vs. Ag/AgCl for CoFe2O4. The software RelaxIS 3 (rhd instruments) was used for data treatment and analysis (details were given in Supplementary Note 4, Supplementary Figs. 24 and 25).

### N2 Physisorption measurements

The N2 sorption experiments were carried out in a physical adsorption analyser (TriStar II 3020, Micromeritics). In all, 87 mg CoFe2O4 and 72 mg Co2FeO4 nanoparticles were degassed at 150 °C for 8 h under vacuum before measurement. The BET surface area was calculated within the relative pressure range of 0.05–0.3 (p/p°). Data was shown in Supplementary Fig. 4.

### XPS measurements

XPS measurements were performed on a VersaProbe II (Ulvac-Phi) using a monochromatic Al X-ray source (1486.6 eV) operating at 15 kV and 13.2 W. The emission angle between the analyser and the substrate surface is 45°. The binding energy scale was referenced to the main C 1 s signal at 284.8 eV. Detailed Analysis of the spectra was carried out with the software CasaXPS. Peak fitting was revealed in Supplementary Fig. 6.

### XAS measurements

Co Kβ High Energy Resolution Fluorescence Detected (HERFD) XAS and Co Kβ XES were collected at beamline I20 at the Diamond Light Source (3 GeV, 300 mA). A Si (111) double crystal monochromator was used for energy selection of the incident beam, and a rhodium-coated mirror was used for harmonic rejection, delivering a flux of 4 × 1012 photons/s at the sample position. X-rays were focused to achieve an approximate beam size of 100 × 300 μm2 (VxH). A Johann-type XES spectrometer was used with two Ge (444) crystals aligned by setting the maximum of the Kβ emission line of a Co foil to 7059.1 eV. The incident energy was calibrated by setting the first inflection point of the Co XAS spectra to 7709.0 eV for a Co foil. Co Kβ XES spectra were collected from 7620 to 7670 eV, with a step size of 0.2 eV. The HERFD XAS edge spectra were collected with the spectrometer fixed at the maximum of the Kβ emission energy while scanning the energy of the incident monochromator. Co Kβ-detected HERFD XAS spectra were collected from 7690 to 7745 eV, with a step size of 0.25 eV over the edge region (7690−7725 eV) and steps of 0.5 eV over the EXAFS region (7725−8500 eV). Co and Fe K-edge XAS spectra were collected in fluorescence mode using a 4-element Vortex Si-drift detector for all samples but the Co Kb detected HERFD for CoFe2O4. Pre-edge baseline corrections were done using Larch XAS Viewer61. XAS data are discussed in Supplementary Note 1 and shown in Supplementary Fig. 7.

### APT measurements

Before preparing the needle-shaped APT specimen, a bulk material containing nanoparticles was fabricated by the following procedure. A nanoparticle suspension was prepared by mixing 5.0 mg of nanoparticles in 1.0 ml of D2O water, followed by ultrasonication for 30 min. 20 μL of nanoparticle solutions were dropped on a clean glassy carbon and dried at room temperature overnight. The CV measurements were performed on the glassy carbon electrodes in 1 M KOD (in D2O) solution at a scan rate of 50 mV/s from 0 V to 0.65 V (vs Ag/AgCl) for 100, 500 and 1000 cycles, respectively. Afterwards, the glassy carbon electrode was covered by Ni electrodeposition at a constant voltage of −1.5 V for 300 s in an electrolyte mixed with 3.0 g nickel sulphate, 0.5 g nickel chloride, and 0.5 g boric acid in 10 ml DI water62. This bulk material was subsequently used to prepare needle-shaped APT specimens by a lift-out procedure using a focus ion beam/scanning electron microscope (FEI Helios G4 CX) (details are shown in Supplementary Note 2 and Supplementary Fig. 11). The APT experiments were conducted in CAMECA LEAP 5000XR instrument in laser pulsing mode at a specimen temperature of 57 K, laser energy of 30 pJ, pulse frequency of 125 kHz, and detection rates of 0.5. The APT data are reconstructed and analysed using the commercial IVAS 3.8.2 software. Additional APT data and analysis were listed in Supplementary Figs. 1222 and Supplementary Note 3.

### H2 TPR measurements

The H2 TPR measurements were conducted in a flow set-up consisting of a gas supply, a stainless-steel U-tube reactor heated in a ceramic tube furnace, and a thermal conductivity detector (TCD, Hydros 100). For the measurement, 116.6 mg catalyst nanoparticles were pre-treated in 50 Nml min−1 He (99.9999%) at 400 °C for 1 h. After cooling to 60 °C, the set-up was flushed with 50 Nml min−1 4.58% H2 (99.9999%)/Ar (99.9995%) for 1 h. The furnace was heated to 800 °C with a heating rate of 10 K min−1. The maximum temperature was kept constant for 1 h. The temperature of the sample was measured every two seconds by a thermocouple placed inside the reactor. The arising water was condensed in a cold trap. The measured consumption of H2 and temperature were plotted against the measurement time. Experimental data was shown in Supplementary Fig. 23 and Supplementary Note 3.

## Data availability

The raw datasets generated and/or analysed during the current study are available in Figshare63Source data are provided with this paper.

## References

1. Grimaud, A. et al. Activation of surface oxygen sites on an iridium-based model catalyst for the oxygen evolution reaction. Nat. Energy 2, 16189 (2016).

2. Grimaud, A. et al. Activating lattice oxygen redox reactions in metal oxides to catalyse oxygen evolution. Nat. Chem. 9, 457–465 (2017).

3. Li, M. et al. Facile synthesis of electrospun MFe 2 O 4 (M=Co, Ni, Cu, Mn) spinel nanofibers with excellent electrocatalytic properties for oxygen evolution and hydrogen peroxide reduction. Nanoscale 7, 8920–8930 (2015).

4. Smith, R. D. L. et al. Spectroscopic identification of active sites for the oxygen evolution reaction on iron-cobalt oxides. Nat. Commun. 8, 2022 (2017).

5. Wu, T. et al. Iron-facilitated dynamic active-site generation on spinel CoAl2O4 with self-termination of surface reconstruction for water oxidation. Nat. Catal. 2, 763–772 (2019).

6. Sickafus, K. E., Wills, J. M. & Grimes, N. W. Structure of spinel. J. Am. Ceram. Soc. 82, 3279–3292 (1999).

7. Budiyanto, E. et al. Tailoring morphology and electronic structure of cobalt iron oxide nanowires for electrochemical oxygen evolution reaction. ACS Appl. Energy Mater. 3, 8583–8594 (2020).

8. Hung, S. F. et al. Unraveling geometrical site confinement in highly efficient iron-doped electrocatalysts toward oxygen evolution reaction. Adv. Energy Mater. 8, 1701686 (2018).

9. Burke, M. S., Kast, M. G., Trotochaud, L., Smith, A. M. & Boettcher, S. W. Cobalt–iron (oxy) hydroxide oxygen evolution electrocatalysts: the role of structure and composition on activity, stability, and mechanism. J. Am. Chem. Soc. 137, 3638–3648 (2015).

10. Enman, L. J. et al. Operando X-ray absorption spectroscopy shows iron oxidation is concurrent with oxygen evolution in cobalt–iron (oxy) hydroxide electrocatalysts. Angew. Chem. Int. Ed. 57, 12840–12844 (2018).

11. Wang, X. T. et al. Redox-inert Fe3+ ions in octahedral sites of Co-Fe spinel oxides with enhanced oxygen catalytic activity for rechargeable zinc–air batteries. Angew. Chem. 131, 13425–13430 (2019).

12. Feng, C. et al. Fe-based electrocatalysts for oxygen evolution reaction: progress and perspectives. ACS Catal. 10, 4019–4047 (2020).

13. Chung, D. Y. et al. Dynamic stability of active sites in hydr(oxy)oxides for the oxygen evolution reaction. Nat. Energy 5, 222–230 (2020).

14. El Arrassi, A. et al. Intrinsic activity of oxygen evolution catalysts probed at single CoFe2O4 nanoparticles. J. Am. Chem. Soc. 141, 9197–9201 (2019).

15. Zhang, M., De Respinis, M. & Frei, H. Time-resolved observations of water oxidation intermediates on a cobalt oxide nanoparticle catalyst. Nat. Chem. 6, 362–367 (2014).

16. Bergmann, A. et al. Reversible amorphization and the catalytically active state of crystalline Co3O4 during oxygen evolution. Nat. Commun. 6, 8625 (2015).

17. Wang, H. Y. et al. In operando identification of geometrical-site-dependent water oxidation activity of spinel Co3O4. J. Am. Chem. Soc. 138, 36–39 (2016).

18. Wang, H. Y. et al. In situ spectroscopic identification of mu-OO bridging on spinel Co3O4 water oxidation electrocatalyst. J. Phys. Chem. Lett. 7, 4847–4853 (2016).

19. Hsu, C. S. et al. Valence- and element-dependent water oxidation behaviors: in situ X-ray diffraction, absorption and electrochemical impedance spectroscopies. Phys. Chem. Chem. Phys. 19, 8681–8693 (2017).

20. Calvillo, L. et al. Insights into the durability of Co–Fe spinel oxygen evolution electrocatalysts via operando studies of the catalyst structure. J. Mater. Chem. A 6, 7034–7041 (2018).

21. Takahashi, M., Guimaraes, J. & Fine, M. Spinodal decomposition in the system CoFe2O4-Co3O4. J. Am. Ceram. Soc. 54, 291–295 (1971).

22. Dinh, T. M. C. et al. FIB plan view lift-out sample preparation for TEM characterization of periodic nanostructures obtained by spinodal decomposition in Co1.7Fe1.3O4 thin films. Cryst. Eng. Comm. 20, 6146–6155 (2018).

23. Smith, W. & Hobson, A. The structure of cobalt oxide, Co3O4. Acta Crystallogr. Sect. B 29, 362–363 (1973).

24. Wu, G. et al. A strategy to promote the electrocatalytic activity of spinels for oxygen reduction by structure reversal. Angew. Chem. Int. Ed. 55, 1340–1344 (2016).

25. Hamdani, M., Singh, R. & Chartier, P. Co3O4 and Co-based spinel oxides bifunctional oxygen electrodes. Int. J. Electrochem. Sci. 5, 556–577 (2010).

26. Palmas, S., Ferrara, F., Vacca, A., Mascia, M. & Polcaro, A. Behavior of cobalt oxide electrodes during oxidative processes in alkaline medium. Electrochim. Acta 53, 400–406 (2007).

27. Bergmann, A. et al. Unified structural motifs of the catalytically active state of Co (oxyhydr) oxides during the electrochemical oxygen evolution reaction. Nat. Catal. 1, 711–719 (2018).

28. Alex, C., Sarma, S. C., Peter, S. C. & John, N. S. Competing effect of Co3+ reducibility and oxygen-deficient defects toward high oxygen evolution activity in Co3O4 systems in alkaline medium. ACS Appl. Energy Mater. 3, 5439–5447 (2020).

29. Nkeng, P. et al. Characterization of spinel-type cobalt and nickel oxide thin films by X-ray near grazing diffraction, transmission and reflectance spectroscopies, and cyclic voltammetry. J. Electrochem. Soc. 142, 1777 (1995).

30. Shinagawa, T., Garcia-Esparza, A. T. & Takanabe, K. Insight on Tafel slopes from a microkinetic analysis of aqueous electrocatalysis for energy conversion. Sci. Rep. 5, 13801 (2015).

31. Oku, M. & Hirokawa, K. X-ray photoelectron spectroscopy of Co3O4, Fe3O4, Mn3O4, and related compounds. J. Electron Spectrosc. Relat. Phenomena 8, 475–481 (1976).

32. Liu, B., Zhang, Y. & Tang, L. X-ray photoelectron spectroscopic studies of Ba0. 5Sr0. 5Co0. 8Fe0. 2O3− δ cathode for solid oxide fuel cells. Int. J. Hydrog. Energy 34, 435–439 (2009).

33. Madhu, R. et al. Honeycomb-like porous carbon–cobalt oxide nanocomposite for high-performance enzymeless glucose sensor and supercapacitor applications. ACS Appl. Mater. Interfaces 7, 15812–15820 (2015).

34. Menezes, P. W. et al. Uncovering the prominent role of metal ions in octahedral versus tetrahedral sites of cobalt–zinc oxide catalysts for efficient oxidation of water. J. Mater. Chem. A 4, 10014–10022 (2016).

35. Biesinger, M. C. et al. Resolving surface chemical states in XPS analysis of first row transition metals, oxides and hydroxides: Cr, Mn, Fe, Co and Ni. Appl. Surf. Sci. 257, 2717–2730 (2011).

36. Dahéron, L. et al. Electron transfer mechanisms upon lithium deintercalation from LiCoO2 to CoO2 investigated by XPS. Chem. Mater. 20, 583–590 (2008).

37. De Groot, F., Vankó, G. & Glatzel, P. The 1s x-ray absorption pre-edge structures in transition metal oxides. J. Phys. Condens. Matter 21, 104207 (2009).

38. Delaplane, R. G., Ibers, J. A., Ferraro, J. R. & Rush, J. J. Diffraction and spectroscopic studies of the cobaltic acid system HCoC2–DCoO2. J. Chem. Phys. 50, 1920–1927 (1969).

39. Amatucci, G., Tarascon, J. & Klein, L. CoO2, the end member of the Li x CoO2 solid solution. J. Electrochem. Soc. 143, 1114 (1996).

40. Bajdich, M., García-Mota, M., Vojvodic, A., Nørskov, J. K. & Bell, A. T. Theoretical investigation of the activity of cobalt oxides for the electrochemical oxidation of water. J. Am. Chem. Soc. 135, 13521–13530 (2013).

41. Post, J. E. & Buchwald, V. F. Crystal structure refinement of akaganeite. Am. Mineral. 76, 272–277 (1991).

42. Finger, L. W. & Hazen, R. M. Crystal structure and isothermal compression of Fe2O3, Cr2O3, and V2O3 to 50 kbars. J. Appl. Phys. 51, 5362–5367 (1980).

43. Beverskog, B. & Puigdomenech, I. Revised pourbaix diagrams for iron at 25–300 C. Corros. Sci. 38, 2121–2135 (1996).

44. Gault, B., Moody, M. P., Cairney, J. M. & Ringer, S. P., Atom Probe Microscopy. Vol. 160 (Springer Science & Business Media, 2012).

45. Li, T. et al. Atomic imaging of carbon-supported Pt, Pt/Co, and Ir@ Pt nanocatalysts by atom-probe tomography. ACS Catal. 4, 695–702 (2014).

46. Ji, Z., Li, T. & Yaghi, O. M. Sequencing of metals in multivariate metal-organic frameworks. Science 369, 674–680 (2020).

47. Li, T. et al. Atomic-scale insights into surface species of electrocatalysts in three dimensions. Nat. Catal. 1, 300–305 (2018).

48. Devaraj, A., Colby, R., Hess, W. P., Perea, D. E. & Thevuthasan, S. Role of photoexcitation and field ionization in the measurement of accurate oxide stoichiometry by laser-assisted atom probe tomography. J. Phys. Chem. Lett. 4, 993–998 (2013).

49. Kasian, O. et al. Degradation of iridium oxides via oxygen evolution from the lattice: correlating atomic scale structure with reaction mechanisms. Energy Environ. Sci. 12, 3548–3555 (2019).

50. Schiller, C., Richter, F., Gülzow, E. & Wagner, N. Validation and evaluation of electrochemical impedance spectra of systems with states that change with time. Phys. Chem. Chem. Phys. 3, 374–378 (2001).

51. Wan, T. H., Saccoccio, M., Chen, C. & Ciucci, F. Influence of the discretization methods on the distribution of relaxation times deconvolution: implementing radial basis functions with DRTtools. Electrochim. Acta 184, 483–499 (2015).

52. Conway, B. & Liu, T. Behaviour of surface intermediate states in anodic O2 evolution electrocatalysis at Co3O4 on Ni and Ti substrates. Ber. Bunsenges. Phys. Chem. 91, 461–469 (1987).

53. Dionigi, F. et al. Intrinsic electrocatalytic activity for oxygen evolution of crystalline 3d-transition metal layered double hydroxides. Angew. Chem. Int. Ed. Engl. 60, 14446–14457 (2021).

54. Liu, T.-C., Pell, W. & Conway, B. Stages in the development of thick cobalt oxide films exhibiting reversible redox behavior and pseudocapacitance. Electrochim. Acta 44, 2829–2842 (1999).

55. Linnemann, J., Kanokkanchana, K. & Tschulik, K. Design strategies for electrocatalysts from an electrochemist’s perspective. ACS Catal. 11, 5318–5346 (2021).

56. Burke, M. S. et al. Revised oxygen evolution reaction activity trends for first-row transition-metal (oxy) hydroxides in alkaline media. J. Phys. Chem. Lett. 6, 3737–3742 (2015).

57. Carraro, F. et al. In operando XAS investigation of reduction and oxidation processes in cobalt and iron mixed spinels during the chemical loop reforming of ethanol. J. Mater. Chem. A 5, 20808–20817 (2017).

58. Jung, I.-H., Decterov, S. A., Pelton, A. D., Kim, H.-M. & Kang, Y.-B. Thermodynamic evaluation and modeling of the Fe–Co–O system. Acta Mater. 52, 507–519 (2004).

59. Hung, S.-F. et al. Identification of stabilizing high-valent active sites by operando high-energy resolution fluorescence-detected X-ray absorption spectroscopy for high-efficiency water oxidation. J. Am. Chem. Soc. 140, 17263–17270 (2018).

60. Chen, Z. & Gao, L. Synthesis and magnetic properties of CoFe2O4 nanoparticles by using PEG as surfactant additive. Mater. Sci. Eng. B 141, 82–86 (2007).

61. Newville, M. Larch: an analysis package for XAFS and related spectroscopies. in Journal of Physics: Conference Series (IOP Publishing, 2013).

62. Kim, S.-H. et al. A new method for mapping the three-dimensional atomic distribution within nanoparticles by atom probe tomography (APT). Ultramicroscopy 190, 30–38 (2018).

63. Xiang, W. et al. Raw data. Figshare. https://doi.org/10.6084/m9.figshare.17142935.v2 (2021).

## Acknowledgements

W.X. and T.L. thank Deutsche Forschungsgemeinsschaft (DFG) for financial support (project number 407513992). T.F., M.M., J.L., K.T., O.R., S.D., T.L., U.H. and M.H. thank DFG (Projektnummer 388390466 – TRR 247/ A1, A2, B6, C4 and S projects) for financial support. W.X. and T.L. would like to thank Zentrum für grenzflächendominierte Höchstleistungswerkstoffe (ZGH) at Ruhr University Bochum for the access to infrastructure (FEI Helios G4 CX SEM/FIB and Cameca LEAP 5000 XR). O.R. and S.D. thank the Max Planck Society for support. J.L. and K.T. acknowledge the Center for Solvation Science (ZEMOS). T.L. thanks Mr. Arjun BalaKrishnan and Mr. Naiyu Qi for assistance with processing figures.

## Funding

Open Access funding enabled and organized by Projekt DEAL.

## Author information

Authors

### Contributions

T.L. designed, supervised and coordinated the project. N.Y. and X.L. synthesised the nanoparticles and conducted the XRD and BET measurements. W.X. performed the electrochemical measurements, J.L. and K.T. contributed to data analysis and interpretation. U.H. performed the XPS measurements and analysed the XPS data. O.R., M.A. and S.D. performed XAS measurements and data analysis. M.H. performed the TEM measurements and T.L. analysed the TEM data. T.F. and M.M. performed the H2/TPR experiment. W.X. performed the APT experiments, W.X. and T.L. analysed the APT data. All authors agreed on the contents and conclusion of the paper.

### Corresponding author

Correspondence to Tong Li.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

## Peer review information

Nature Communications thanks Douglas Kauffman, Daniel Perea and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

Xiang, W., Yang, N., Li, X. et al. 3D atomic-scale imaging of mixed Co-Fe spinel oxide nanoparticles during oxygen evolution reaction. Nat Commun 13, 179 (2022). https://doi.org/10.1038/s41467-021-27788-2

• Accepted:

• Published:

• DOI: https://doi.org/10.1038/s41467-021-27788-2