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

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.

H ydrogen 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 place 1,2 . 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 chemistry [3][4][5] ; these characteristics make them attractive alternatives to the high-cost benchmark noble metal-based oxides, i.e., IrO 2 and RuO 2 . Depending on the composition, two spinel structures can be formed: (i) spinel, whereby a divalent cation, e.g., Co II , is located at the tetrahedral site, and trivalent Fe III at the octahedral site, and (ii) inverse spinel, whereby Co II is located at the octahedral site and Fe III at both the tetrahedral and octahedral sites 6 . The addition of small amounts of Fe in Co 3 O 4 has been found to reduce the overpotential, while excess Fe increases the overpotential 7,8 . 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 debate 4,5,[9][10][11][12][13][14] . Additionally, although surface chemical and structural rearrangement of Co-based spinel oxides has been recently observed 4,[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 Co 2 FeO 4 spinel and CoFe 2 O 4 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 Co 2 FeO 4 and CoFe 2 O 4 nanoparticles during cyclic voltammetry (CV) measurements under OER conditions. Comprehensive information regarding the surface state changes is obtained by the scalebridging 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 Co 2 FeO 4 and most likely in most mixed Co x Fe (3-x) O 4 spinel oxides when x is in the range of 1.1-2.7 due to the miscibility gap 21,22 . Interestingly, hydroxyl groups were trapped at the interface between the nanodomains, possibly yielding a significantly enhanced OER activity of pristine Co 2 FeO 4 compared to CoFe 2 O 4 . During OER, different levels of Fe dissolution occur in the nanodomains of Co 2 FeO 4 , along with concurrent irreversible structural transformation towards Co IV O 2 , leading to a substantial decrease in the OER activity. In contrast, negligible Fe loss was observed for CoFe 2 O 4 . Instead, (Fe III , Co III ) 2 O 3 was formed on the surface, further decreasing the OER activity of CoFe 2 O 4 . 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. Co 2 FeO 4 and CoFe 2 O 4 nanoparticles were synthesised by a hydrothermal method (see Methods). Both pristine nanoparticles have the standard cubic spinel structure (Fd 3m 23 ), as confirmed by X-ray powder diffraction (XRD) (Supplementary Fig. 1 The difference in the lattice constants originates from the differences between Co/Fe contents and their radius (Co 3+ has a radius of 0.61 Å, which is slightly less than the Fe 3+ radius of 0.65 Å 24 ); this is consistent with the XRD data (in Supplementary  Fig. 1), whereby the diffraction peaks of Co 2 FeO 4 are shifted to higher 2θ values compared to those of CoFe 2 O 4 .
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 Co 2 FeO 4 and CoFe 2 O 4 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 N 2 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 (iR s ) correction (R s 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 system 5 .
The LSV plots, shown in Fig. 1a, b, reveal that pristine Co 2 FeO 4 exhibits a higher OER activity than pristine CoFe 2 O 4 , since the overpotential of Co 2 FeO 4 (359 mV at 10 µA/cm 2 ) is lower than that of CoFe 2 O 4 (432 mV at 10 µA/cm 2 ). Pristine Co 2 FeO 4 has a Tafel slope of 43 ± 1 mV/dec (Fig. 1e), while pristine CoFe 2 O 4 has a much larger Tafel slope of 79 ± 2 mV/dec (Fig. 1f), indicating that OER charge transfer kinetics are faster on pristine Co 2 FeO 4 than on pristine CoFe 2 O 4 . The measured Tafel slope of pristine Co 2 FeO 4 nanoparticles is also lower than most pristine Co 3 O 4 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 Co 2 FeO 4 and CoFe 2 O 4 ( Fig. 1a, b, e, f). In particular, the Tafel slope of Co 2 FeO 4 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 CoFe 2 O 4 increases slightly to 83 ± 1 mV/dec. Thus, despite the high OER activity of pristine Co 2 FeO 4 , its OER activity drops as the number of CV cycles increases, eventually reaching similar values as detected for the less active CoFe 2 O 4 .
Furthermore, Co 2 FeO 4 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), respectively 26,27 . 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 Co 3 O 4 26-29 . The cathodic peak C1 at 1.1 V for the Co(II)/Co(III) process 26,27 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 Co 2 FeO 4 , nearly no redox couples were observed for CoFe 2 O 4 , 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 Co 2 FeO 4 and CoFe 2 O 4 , we first performed XPS to examine the oxidation state of Co and Fe on the surface of Co 2 FeO 4 and CoFe 2 O 4 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 2p 1/2 and 2p 3/2 peak locations for Co(II), Co(III) and Co(IV) 31-34 , 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 Co 3 O 4 -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 Co 3 O 4 -like Co increases (as indicated by the depth analysis using peak deconvolution of the Co 2p peak 35 shown in Supplementary Fig. 6b). Our CV data for Co 2 FeO 4 , 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 Co 3 O 4 -like Co is similar to that of CoO 2 -like Co(IV) 36 13 .
To further verify the irreversible change in the oxidation state of Co 2 FeO 4 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 Co 37 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 Co 2 FeO 4 and CoFe 2 O 4 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 Co 2 FeO 4 , viewed along the [001] zone axis. Interestingly, the interplanar spacing of d 220 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 Co 2 FeO 4 , 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 (bluedotted 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 3m, hexagonal 38 , Supplementary Table 1), which agrees with observations of a previous study 19 . Therefore, we hypothesise that epitaxial growth of (Co, Fe)OOH occurs on Co 2 FeO 4 , with an orientation relationship of (010) Co 2 FeO 4 //(1-101) (Co, Fe)OOH. A previous study proposed that Co II ions at the tetrahedral site are oxidised to form amorphous Co III oxyhydroxides 16 . Here, we observed crystalline Co III oxyhydroxides, possibly formed by crystallisation of the amorphous Co III 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 Co 2 FeO 4 spinel oxide. The reflection spots in the FFT image (Fig. 3c, bottom inset) match well with the CoO 2 phase (P 3m1, hexagonal 39 , Supplementary Table 1), with an octahedrally coordinated Co (IV). The formation of CoO 2 is also confirmed by additional reflection spots that correspond well to (11)(12)(13)(14)(15)(16)(17)(18)(19)(20) CoO 2 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)  presence of Co (IV) by XPS (Fig. 2a). These results most likely imply an irreversible transformation towards (Co, Fe)O 2 phase on the surface of Co 2 FeO 4 after 1000 cycles. Additionally, CoO 2 is the stable phase at higher potentials, in accordance with the Co Pourbaix diagram 40 .
In contrast with Co 2 FeO 4 , we observed no significant structural changes on the CoFe 2 O 4 nanoparticle surface after 100 cycles (Fig. 3d, e). After 1000 cycles, a structural transformation of the surface of [001]-orientated CoFe 2 O 4 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  Notably, we observed an amorphous layer on the surfaces of aggregated Co 2 FeO 4 and CoFe 2 O 4 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 Co 2 FeO 4 (Fig. 3b), can be regarded as an active intermediate for OER 27 , the activity of Co 2 FeO 4 decreased after 100 cycles (Fig. 1a, e). The cause of the decrease in the OER activity of Co 2 FeO 4 (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 APT 44 , a mass-spectrometry technique with sub-nanometre spatial resolution in three dimensions 45,46 , 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 D 2 O in order to use APT to examine the distribution of hydroxyl groups after OER 47 . Figure 4a exemplifies a cross-sectional atom map of pristine Co 2 FeO 4 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 CoO x and FeO y 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 Co 2 FeO 4 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' Co 2 FeO 4 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 Co 2 FeO 4 nanoparticle. In contrast, non-segregated pristine Co 2 FeO 4 nanoparticles have uniformly distributed Fe and Co (Fig. 5a,e). The compositions of the Fe-rich and Co-rich nanodomains in pristine Co 2 FeO 4 , 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 H 2 temperature-programmed reduction (H 2 TPR) and the effect of laser pulse energy on the measurement of oxide stoichiometry 48 Tables 1 and 2). The formation of Fe-rich and Co-rich nanodomains in the pristine Co 2 FeO 4 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°C 21,22 = 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 studies 21,22 . Despite this, we unambiguously reveal nanoscale compositional modulation of Co 2 FeO 4 nanoparticles. In contrast, we did not observe any segregation for the CoFe 2 O 4 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 decomposition 21,22 .
Next, we examined elemental redistribution of 'segregated' Co 2 FeO 4 nanoparticles after 100, 500 and 1000 cycles under OER  (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 d 220 measured from two regions of pristine Co 2 FeO 4 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 Co 2 FeO 4 and CoFe 2 O 4 observed by XRD shown in Supplementary Fig. 1). The hydroxyl groups can be considered as 'fingerprints' that indicate the regions where OER occurs 49 . 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 Co 2 FeO 4 nanoparticles. More importantly, we observed a dramatic compositional change within the nanodomains of Co 2 FeO 4 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 Corich 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 Co 2 FeO 4 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' Co 1.7 Fe 1.3 O 3.3 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)O 2 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 CoFe 0.75 Al 1.5 O 4 by electron energy loss spectroscopy 5 , 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 Co 2 FeO 4 after 100 and 500 cycles (Fig. 1a, c). After 1000 cycles, surface formation of stable (Co IV , Fe III )O 2 further decreases the activity of Co 2 FeO 4 .
For comparison with Co 2 FeO 4 , the compositional changes of the CoFe 2 O 4 nanoparticles after 100, 500 and 1000 cycles are The error bars for the ratio were calculated from R where R is the ratio of Co/Fe, a and b is Co and Fe concentration and σa and σb is the standard deviation of Co and Fe concentration. (Source data is provided in a Source Data file).   Table 2). No evident elemental redistribution was observed after 100 cycles, while 2-3 nm oxygen-rich regions were seen on the surface of CoFe 2 O 4 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 Co II was oxidised to Co III while Fe III remained stable even after 1000 cycles ( Fig. 2 and Supplementary  Fig. 7). Therefore, we speculate that the oxygen-rich regions possibly correspond to the (Fe III , Co III ) 2 O 3 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 Co 2 FeO 4 , whereas one semicircle is observed for the inverse spinel CoFe 2 O 4 . The distribution of relaxation times for Co 2 FeO 4 (inset of Fig. 7a) reveals three distinct peaks, and one peak is discerned for CoFe 2 O 4 (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 fitting 50,51 , more details in Supplementary Note 4 and Supplementary Figs. 24 and 25). The equivalent circuit for Co 2 FeO 4 , shown in Fig. 7c, contains the double layer capacitor C 1 and the pseudocapacitors C 2 and C 3 . The corresponding pseudocapacitive properties are based on changing oxidation states of electrochemically accessible cobalt sites and adsorptive discharge of oxygen-containing species in the electrolyte 52,53 . So, in course of a catalytic cycle, intermediate catalytic (re-)transformations occur via pseudocapacitive charging/discharging through the faradaic resistors R 1 and R 2 54 . In parallel, the OER reaction proceeds via OH − -to-O 2 conversion steps at the solution side 55 . We can see from Fig. 7c that the resistance increases with the number of CV cycles, particularly for   (Fig. 7b), which may relate to its low OER activity compared to Co 2 FeO 4 , as indicated by Tafel plots (Fig. 1e, f). For both Co 2 FeO 4 and CoFe 2 O 4 , 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 Co 2 FeO 4 and CoFe 2 O 4 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' Co 2 FeO 4 whose compositional modulation is   Fig. 7 Electrochemical sub-processes of Co 2 FeO 4 and CoFe 2 O 4 during OER. Electrochemical impedance spectroscopy data in complex plane representation (Nyquist plot) and determined distribution of relaxation times (insets) of electrodes covered by a Co 2 FeO 4 nanoparticles at 1.63 V vs. RHE (≈E LSV,initial (6 mA/cm 2 geom )) and b CoFe 2 O 4 nanoparticles at 1.73 V vs. RHE (≈E LSV,initial (6 mA/cm 2 geom )) in the pristine state and after 100, 500 and 1000 cycles. c, d Corresponding changes of the resistances (solid lines) and capacitances (dashed lines) as the number of CV cycles increases (obtained by equivalent circuit fitting to the displayed model circuits). Source data are provided as a Source Data file. driven by spinodal decomposition 21,22 , and 'non-segregated' Co 2 FeO 4 in pristine Co 2 FeO 4 (Fig. 8a, b). We speculate that such composition modulation is present in mixed Co x Fe (3-x) O 4 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' Co 2 FeO 4 (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, (Co III , Fe III )OOH is formed epitaxially on the surface of non-segregated Co 2 FeO 4 ( Fig. 3b and Supplementary Fig. 9a, b), and possibly in the nanodomains of segregated Co 2 FeO 4 nanoparticles (Fig. 8d). Despite the formation of active (Co III , Fe III )OOH 13 , the activity of Co 2 FeO 4 decreases as a consequence of the irreversible structural transform towards (Co IV , Fe III )O 2 (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 FeO 4 2in alkaline electrolytes 13,43,56 (Fig. 8c, e-f). Therefore, we conclude that the concurrent structural transformation towards a stable (Co IV , Fe III )O 2 phase and Fe dissolution lead to a decrease in OER activity of Co 2 FeO 4 as the number of CV cycles increases. For CoFe 2 O 4 , Co II present on the surface of pristine sample (Fig. 8g) is oxidised to Co III at the onset of OER (Fig. 8h), as shown by XPS data (Fig. 2b). After 1000 cycles, the (Fe III , Co III ) 2 O 3 phase, whose dimension is 4-5 nm, is likely formed on the surface of CoFe 2 O 4 (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 Co II catalyses the OER of both Co 2 FeO 4 and CoFe 2 O 4 , since the Co oxidation state increases (indicated by XPS in Fig. 2 and XANES in Supplementary Fig. 7). The Co II in the tetragonal sites (termed Co II Td ) are thought to be active sites as they are responsible for the formation of active Co III OOH species 17 ; 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) 57,58 ). These observations provide a strong indication that Fe, which can actively catalyse OER, seems to be associated with the number of Co II Td 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 potentials 56 , and Co III OOH provides an electrically conductive support 9 . A previous study reported that Fe can serve as an indirect active site, for example by changing the spin and charge state of Co 11 , or by assisting 5 or stabilising the active Co III OOH intermediate 59 , 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 OER 4 . Our study indicates that Fe promotes the formation of active species for OER, possibly Co III Fe III OOH, whose OER activity is significantly better than that of Co III OOH 13 (although its activity drops rapidly due to Fe dissolution). Therefore, in comparison to CoFe 2 O 4 , Co 2 FeO 4 has a higher OER activity most likely due to the formation of more Co III Fe III OOH yielded by its optimal ratio of Fe content to Co II Td sites. Also, the interface between with Fe-rich and Co-rich nanodomains and b non-segregated Co 2 FeO 4 nanoparticles. The interface between Co-rich and Fe-rich nanodomains of segregated Co 2 FeO 4 is thought to provide trapping sites for c hydroxyl groups, thereby contributing to a lower overpotential of pristine Co 2 FeO 4 than that of g pristine CoFe 2 O 4 . At the beginning of electrolysis, concurrent structural transformation to c, d Co III OOH and c, e, f Fe dissolution occurred for both segregated and non-segregated Co 2 FeO 4 . In addition, f stable (Co IV , Fe III )O 2 is formed on the surface of Co 2 FeO 4 , further degrading its activity. The OER activity of CoFe 2 O 4 also decreases as the number of CV cycles increases, which is due to h oxidation of Co(II) to Co(III) and i the formation of the stable (Fe III , Co III ) 2 O 3 phase on the surface. (The left axis is overpotential at 10 µA/cm 2 , the top axis is the average O/M ratio, the right axis is the average Co/Fe ratio, and the bottom arrows point to the trend of increasing CV cycles). the nanodomains of segregated Co 2 FeO 4 traps hydroxyl groups, providing additional active regions for OER, thereby further enhancing the OER activity of Co 2 FeO 4 . Therefore, we conclude that the presence of Co III Fe III OOH, promoted by Fe and Co II Td , coupled with the nanosized defect features, leads to Co 2 FeO 4 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 oxides 7,8 .
In summary, our study provides atomic-scale insights into the evolving surface structure of Co 2 FeO 4 and CoFe 2 O 4 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 CO 2 reduction. O 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.

Methods
XRD measurements. The XRD analysis of the pristine Co 2 FeO 4 and CoFe 2 O 4 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°a nd 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 8-10.
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 cm 2 ). The glassy carbon electrode was first cleaned and polished to a mirror finish with 50 nm Al 2 O 3 . 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 D 2 O for APT specimen preparation and measurements. The deuterium oxide (D 2 O with 99.9 at.% D) and potassium deuteroxide solution (40 wt.% KOD in D 2 O 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  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 × 10 12 photons/s at the sample position. X-rays were focused to achieve an approximate beam size of 100 × 300 μm 2 (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 CoFe 2 O 4 . Pre-edge baseline corrections were done using Larch XAS Viewer 61 . 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 D 2 O 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 D 2 O) 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 water 62 . 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. 12-22 and Supplementary Note 3.
H 2 TPR measurements. The H 2 TPR measurements were conducted in a flow setup 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% H 2 (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 H 2 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 Figshare 63 . Source data are provided with this paper.