Breathing mode in Nd-CoO_x for active and stable proton exchange membrane water electrolysis

Abstract

Proton-exchange membrane water electrolysis in acidic solutions holds great promise for advancing the green-hydrogen economy but limited by using precious-metal catalysts. Here, we develop a low-cost Co₃O₄-based catalyst with oxygen vacancies and Nd-insertion (Nd-CoO_x). This catalyst achieves a low overpotential of 317 mV at 10 mA cm⁻² for acidic oxygen evolution reaction and remains stable over 800 h. The oxygen vacancies in Nd-CoO_x enhance its catalytic activity, while Nd enables the long-term stability through a flexible “breathing mode” during oxygen evolution reaction, as revealed by in-situ methods such as X-ray absorption spectroscopy and Raman spectroscopy. The presence of Nd also improves the catalytic activity with the oxide path mechanism. This favorable strategy can be extended to other lanthanides. When used as the anode, Nd-CoO_x delivers good performance in water electrolysis to achieve 200 mA cm⁻² at a low cell voltage of 1.69 V with low degradation over 100 h, and can reach 1 and 2 A cm⁻² at 1.93 V and 2.12 V, respectively.

Proton-exchange-membrane water electrolysis is crucial for advancing the green-hydrogen economy but is limited by the use of precious-metal catalysts. Here, the authors report a Nd-CoOx catalyst in which Nd confers long-term stability via a flexible atomic “breathing mode.”

Introduction

Water splitting through electrolysis offers a promising route to produce green hydrogen. Various methods have been applied for water splitting, while proton-exchange membrane water electrolysis (PEMWE) in acidic solutions stands out. It provides good hydrogen purity, high current density, and efficient energy conversion^1,2. However, the slow oxygen evolution reaction (OER) kinetics remains a major challenge. This reaction includes four proton-coupled electron transfer steps and needs catalysts with both high activity and stability in acidic solutions^3,4. Up to now, only rare and expensive Ir- or Ru-based catalysts meet these requirements^1,5. Unfortunately, their high cost and limited reserve result in significant barriers to large-scale implementation. Developing low-cost, noble-metal-free catalysts with high efficiency and long-term stability for acidic OER is therefore an urgent priority.

A large amount of 3 d transition metal oxides have been developed as efficient OER catalysts^6–10. Their multiple chemical states allow them to enhance catalytic activity across a wide pH range. However, most 3 d transition metal oxides suffer from severe structural degradation in acidic environments. Recently, Co-based catalysts have shown significant progress for acidic OER^11–15. They offer both high activity and improved stability after modification. Theoretical calculations suggest that spinel-type Co₃O₄ could achieve OER activity comparable to that of Ir- and Ru-based oxides¹⁶. Unfortunately, Co₃O₄ faces challenges in acidic conditions. It undergoes severe dissolution during operation¹⁷. Its catalytic performance is further limited by poor conductivity and a lack of sufficient active sites². To address these problems, various strategies have been developed to modify Co-based catalysts. These include defect engineering, hetero-structure formation, and elemental doping. For instance, introducing oxygen vacancies in Co₃O₄ was used to improve catalytic activity, while careful control of vacancy content could enhance the stability for acidic OER². Huang et al. demonstrated that incorporating CeO₂ into Co₃O₄ would modify the local bonding environment, improving stability under acidic conditions¹¹. By incorporating Mn into the spinel lattice of Co₃O₄, the Co₂MnO₄ material was prepared and showed a highly extended lifetime in acid environment¹⁸. Recently, La and Mn co-doping in Co₃O₄ achieved a small overpotential of 353 mV at 10 mA cm⁻². This system maintained low degradation for over 360 h in an H-cell, with La stabilizing the surface and Mn improving bulk conductivity¹³. Beyond La and Mn, W doping also showed notable success^12,14. Especially, by introducing W into Co oxide to form a CoWO₄ catalyst with vacancies, water-hydroxide networks would fill these vacancies, enabling stable operation for 175 h at 10 mA cm⁻² in an H-cell and for 600 h at 1 A cm⁻² in PEMWE¹⁴. These advancements highlight the potential of modified Co-based materials for acidic OER.

In this work, we reveal the presence of a “breathing mode” in the atomic structure of Nd-modified Co₃O₄ with oxygen vacancies (Nd-CoO_x). This feature enables stable and efficient OER in acidic conditions. The catalyst exhibits a low overpotential of 317 mV at 10 mA cm⁻² in an H-cell and remains stable for at least 800 h. The oxygen vacancies in Nd-CoO_x enhance its catalytic activity. At the same time, the “breathing mode” around Nd provides structural flexibility, which allows the atomic structure to adapt during OER and prevents structural failure under external potentials. We observed these dynamic structural changes by using in-situ methods such as X-ray absorption spectroscopy (XAS) and Raman spectroscopy. The insertion of Nd also improves the acidic OER activity by creating the oxide path mechanism (OPM). When used as the anode, Nd-CoO_x enables PEMWE to obtain a current density of 200 mA cm⁻² at a small cell voltage of 1.69 V, with minimal degradation over 100 h. The system also delivers 1 A cm⁻² at 1.93 V and 2 A cm⁻² at 2.12 V. These performances surpass those of the IrO₂ anode, which requires 2.05 V and 2.27 V to reach the same current densities, respectively. These results emphasize the potential of Nd-CoO_x as an efficient and cost-effective catalyst for acidic OER in PEMWE systems.

Results

Morphology and structure characterization

The Nd-CoO_x catalyst was prepared using a two-step synthesis method combining microwave-assisted heating (MAH) and solution combustion synthesis (SCS) (Fig. 1a). In the first step, carbon cloth (CC) was immersed in a solution containing Co²⁺ and Nd³⁺ ions, with microwave heating applied to assist the process. In the second step, the wet CC substrate, loaded with Co²⁺ and Nd³⁺ ions, was treated multiple times using the SCS method. This treatment created a rapid oxidation environment, resulting in the formation of the Nd-CoO_x catalyst on the CC substrate. The oxidation process produced oxidized Co species, while the rapid treatment resulted in the generation of oxygen and cobalt vacancies. Additionally, the incorporation of Nd ions likely resulted in Nd-insertion within the oxidized Co species. More preparation information can be observed in the Methods section. Scanning electron microscopy (SEM) images (Supplementary Fig. 1) reveal the morphology of Nd-CoO_x and control samples (CoO_x and NdO_x). All samples exhibit porous sheet-like structures on the surface of the CC substrate (ribbons). Synchrotron radiation-based X-ray diffraction (SR-XRD) patterns (Supplementary Fig. 2) show that the primary crystalline phase in both Nd-CoO_x and CoO_x is Co₃O₄ (JCPDS No. 42-1467). No distinct peaks for Nd-based materials are observed, suggesting that Nd is likely inserted into the Co₃O₄ structure. The atomic ratio of Co to Nd in Nd-CoO_x is determined to be approximately 15:1 using inductively coupled plasma optical emission spectrometry (ICP-OES, Supplementary Table 1). Transmission electron microscopy (TEM) images of Nd-CoO_x (Fig. 1b) confirm its porous sheet morphology. High-resolution TEM (HRTEM) images (Fig. 1c) show the crystal structure for Co₃O₄, with lattice spacing values of 0.245 nm and 0.467 nm for the (311) and (111) planes of Co₃O₄, respectively². A similar Co₃O₄ structure is observed in the CoO_x control sample (Supplementary Fig. 3). Dark-field energy-dispersive X-ray spectroscopy (EDX) elemental mapping images (Fig. 1d) show the homogeneous distributions of Co, Nd, and O elements in Nd-CoO_x.

Fig. 1. Morphology and structural information of Nd-CoO_x.

a Schematic illustration for the preparation of Nd-CoO_x. b–d TEM (b), HRTEM (c), and dark-field EDX elemental mapping (d) images of Nd-CoO_x. e The aberration-corrected HAADF-STEM image of Nd-CoO_x. f The corresponding intensity profiles for the red (Line 1) and yellow (Line 2) arrows in Fig. 1e. g The diagram of compressive stress field induced by Nd.

The atomic structure of Nd-CoO_x was carefully probed by using aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM). In HAADF-STEM, atoms with higher atomic numbers typically produce stronger signal intensities¹. As exhibited in Fig. 1e, bright spots corresponding to Nd atoms, which have a higher atomic number than Co, are clearly visible. In contrast, the lighter O atoms are not detectable under the current imaging conditions. These observations strongly suggest that Nd atoms are atomically dispersed within the Co₃O₄ lattice in Nd-CoO_x. Atomic distances around the Nd atoms are analyzed, specifically along Line 1 and Line 2 in Fig. 1e. The Nd-Co distance in Nd-CoO_x is measured to be approximately 0.30 nm, while the neighboring Co-Co distance is slightly smaller at about 0.27 nm (shown in Fig. 1f). This difference is likely due to the insertion of larger Nd atoms into the Co₃O₄ lattice. As illustrated in Fig. 1g, the substitution of Co atoms with larger Nd atoms compresses the local atomic structure, reducing the Co-Co distance in the lattice. Such local structural modifications may play a critical role in maintaining structural stability during the OER^3,4.

X-ray photoelectron spectroscopy (XPS) has been used to probe the electronic structures of the catalysts (Supplementary Fig. 4). In the Co 2p spectra of commercial Co₃O₄ (Com. Co₃O₄), two distinct peaks are observed at 795.5 eV and 780.5 eV (Supplementary Fig. 4a). These peak positions align well with previously reported values^2,5. In comparison, the Co 2p peaks for both CoO_x and Nd-CoO_x shift to lower binding energy. This shift indicates a lower chemical state of Co in these samples, likely due to the presence of oxygen vacancies (V_O) created during the fast oxidation process. The Nd 3 d XPS spectra (Supplementary Fig. 4b) reveal a lower chemical state of Nd in Nd-CoO_x compared to NdO_x. This change is attributed to the influence of CoO_x. In the O 1 s spectra (Supplementary Fig. 5a), features at 531.3 eV for both CoO_x and Nd-CoO_x are associated with V_O in the metal oxides^19,20. The presence of V_O introduces excess electrons, which can occupy the metal d-orbitals, leading to a lower oxidation state²¹. Further evidence of V_O is provided by electron paramagnetic resonance (EPR) measurements (Supplementary Fig. 5b). CoO_x and Nd-CoO_x exhibit much stronger EPR signals at g = 2.003 compared to Co₃O₄, suggesting the presence of V_O. These V_O are expected to enhance OER kinetics, thereby improving catalytic activity^2,22.

Synchrotron radiation-based XAS has been used to examine the local electronic structure of Nd-CoO_x. Fig. 2a displays the normalized Co K-edge X-ray Absorption Near Edge Structure (XANES) spectra for CoO_x, Nd-CoO_x, CoO, and Co₃O₄. Both CoO_x and Nd-CoO_x show different XANES spectra with a comparison to that of standard Co₃O₄. The white-line peak positions for CoO_x and Nd-CoO_x are lower than that of Co₃O₄ but higher than that of CoO. This suggests a reduced Co oxidation state in the Co₃O₄ structure, likely caused by V_O. As displayed in the inset of Fig. 2a, the average valence states of Co in CoO_x and Nd-CoO_x are calculated from the edge positions, which are approximately 2.2 and 2.3 in CoO_x and Nd-CoO_x, respectively²³. The XANES spectra of CoO_x and Nd-CoO_x also exhibit a pronounced pre-edge peak A at approximately 7709 eV (quadrupolar transition from 1 s to 3 d), which might be attributed to the lowered symmetry with oxygen vacancies and defects^24,25.

Fig. 2. Electronic structure of the catalysts.

a, b Co K-edge XANES spectra (a) and the corresponding Fourier-transform curves of EXAFS spectra (b) of CoO, Co₃O₄, CoO_x and Nd-CoO_x. c WT plots of Co₃O₄, CoO_x and Nd-CoO_x at Co K-edge. d, e Nd L₃-edge XANES spectra (d) and the corresponding Fourier-transform curves of EXAFS spectra (e) of Nd₂O₃, NdO_x and Nd-CoO_x. f Structure models for Co₃O₄, CoO_x and Nd-CoO_x with Nd-insertion.

The Fourier-transform (FT) curves of extended X-ray absorption fine structure (EXAFS) spectra are displayed in Fig. 2b. Strong FT peaks at 1.49 Å and 1.63 Å (the actual fitted bond length values are 1.93 and 2.12 Å in Supplementary Table 2) correspond to the Co-O bonds in Co₃O₄ and CoO, respectively, as expected from their different bond lengths, consistent with previous literature²⁶. Compared to the Co-O FT peak in Co₃O₄ at 1.49 Å, the CoO_x sample shows a longer FT bond length of 1.52 Å (fitted value of 1.95 Å) with reduced peak intensity. This shift is attributed to the presence of V_O, which results in fewer Co–O bonds and a lower chemical state for Co. Interestingly, the Co-O FT peak for Nd-CoO_x is shorter, appearing at 1.45 Å (fitted value of 1.91 Å). This shift is attributed to Nd-modified electronic and atomic structures. As identified in Fig. 1, when the larger Nd atom replaces a Co atom, the local atomic structure becomes compressed. This compression reduces the Co-Co distance and shortens the surrounding Co-O bond, as observed in Fig. 2b. The detailed R-space (Fourier-transform curves of EXAFS spectra) fitting curves and bond lengths are presented in Supplementary Fig. 6 and Supplementary Table 2. The fitted Co-O bond length for CoOₓ is 1.95 Å, which is longer than Co₃O₄ at 1.93 Å. However, in Nd-CoO_x, the Co-O bond length is significantly reduced to 1.91 Å. These results align well with the Fourier-transform curves of EXAFS data. The progression of local structural changes—from typical Co₃O₄ to CoOₓ with V_O (longer Co-O bonds), and finally to Nd-CoOₓ with both vacancies and Nd-insertion (shorter Co-O bonds due to compression)—is illustrated in Fig. 2f. Wavelet transform (WT) plots of Nd-CoOₓ, CoOₓ, and Co₃O₄ are also shown in Fig. 2c. The WT plots provide two-dimensional information for both k-space and R-space data, facilitating the discrimination of backscattering atoms²⁷. The heavier atom (with a higher atomic number) typically exhibits a better ability to scatter photoelectrons, leading to their oscillations occurring at higher frequencies (high k values)²⁸. In Fig. 2c, Co₃O₄ shows the Co-O peak at a k value of ~8.0 Å^−1 29, while for Nd-CoO_x, this peak shifts to a higher k value (~8.3 Å⁻¹), indicating the Nd-insertion.

Figure 2d presents the Nd L₃-edge XANES spectra for Nd-CoO_x, NdO_x, and Nd₂O₃. Nd-CoO_x shows the lowest white-line intensity, indicating a relatively lower valence state for Nd. The corresponding Fourier-transform curves of EXAFS curves are shown in Fig. 2e. For NdO_x, the Nd-O FT peak appears at 2.01 Å, similar to that of standard Nd₂O₃. This indicates an oxidized Nd structure in NdO_x. In contrast, the Nd-O FT peak for Nd-CoO_x sample occurs at 1.94 Å, standing for a shorter Nd-O bond. This reduction in bond length is attributed to the insertion of Nd into the smaller Co₃O₄ lattice (primarily on the surface Co vacancy sites). The limited space at the Co sites in the Co₃O₄ lattice forces the Nd-O bond to shorten. The longer Nd-O bonds also exert compressive forces on the surrounding Co-O bonds, as illustrated in Fig. 2f. This local electronic and structural modulation may play an important role in improving the OER performance.

Evaluation of OER performance

The electrocatalytic OER activities of Nd-CoO_x and reference samples were evaluated in a standard three-electrode cell using 0.1 M HClO₄ electrolyte. Commercial Co₃O₄ and IrO₂ were used as reference catalysts. The Nd concentration in Nd-CoO_x was optimized, as exhibited in Supplementary Fig. 7. As illustrated in Fig. 3a, NdO_x is almost inactive for OER. Commercial Co₃O₄ displays an overpotential of 509 mV to obtain a current density of 10 mA cm⁻². Introducing oxygen vacancies (V_O) in CoO_x significantly enhances its performance, reducing the overpotential to 418 mV at 10 mA cm⁻². This improvement highlights the critical role of V_O in boosting OER activity. However, both catalysts show lower activity compared to the benchmark IrO₂ catalyst, which requires 335 mV at 10 mA cm⁻². Remarkably, the introduction of Nd into CoO_x to form Nd-CoO_x further improves the OER performance. Nd-CoO_x exhibits an overpotential of just 317 mV at 10 mA cm⁻², even lower than that of commercial IrO₂ (335 mV). The LSV curves without iR correction and the corresponding performance errors are summarized in Supplementary Fig. 8. Since pure NdO_x shows negligible OER activity, the good performance of Nd-CoO_x is primarily assigned to the synergistic effect of Nd and CoO_x.

Fig. 3. Electrochemical OER performance of Nd-CoO_x catalyst.

a Polarization curves of the prepared electrocatalysts in 0.1 M HClO₄ (pH: 1.08 ± 0.02) at room temperature (~25 °C) with 90% iR correction (5 mV s⁻¹ scan rate). The compensation resistances are shown in Supplementary Table 3. b Chronopotentiometric response of Nd-CoO_x at 10 mA cm⁻² for over 800 h (the inset shows the stability of CoO_x and commercial Co₃O₄). c Catalytic durability of Nd-CoO_x catalyst with an initial polarization curve and the curve after 5000 cycles. d Current-voltage polarizations (with 90% iR- correction) of the PEMWE cell with Nd-CoO_x as the anodic catalyst (commercial Pt/C as the cathode catalyst, IrO₂ for comparison) at 80 °C. The compensation resistances of Nd-CoO_x ║ Pt/C and IrO₂ ║ Pt/C cells are 0.1 ± 0.02 Ω and 0.09 ± 0.01 Ω, respectively. e Chronopotentiometry testing of Nd-CoO_x at 200 mA cm⁻² in the PEMWE cell for over 100 h.

Tafel slopes of the catalysts are shown in Supplementary Fig. 9. Nd-CoO_x exhibits a smaller Tafel slope of 82 mV dec⁻¹ compared to CoO_x (89 mV dec⁻¹), IrO₂ (97 mV dec⁻¹), and Co₃O₄ (96 mV dec⁻¹). This indicates more favorable OER kinetics for the Nd-CoO_x³⁰. Electrochemical impedance spectroscopy (EIS) curves (see Supplementary Fig. 10 and Supplementary Table 3) reveal that Nd-CoO_x has the smallest charge transfer resistance (R_ct) of 8.9 Ω, indicating quick charge transfer properties. Double-layer capacitance (C_dl) measurements are applied to estimate the electrochemically active surface area (ECSA) of the catalysts. Nd-CoO_x shows a significantly higher C_dl value when compared to CoO_x and NdO_x (Supplementary Fig. 11), indicating the presence of more active sites due to Nd-incorporation. The combination of a smaller Tafel slope, lower charge transfer resistance, and higher active site density confirms that Nd-CoO_x is a competitive catalyst for the acidic OER.

Except for the high activity, the Nd-CoO_x catalyst demonstrates robust stability for acidic OER. As displayed in Fig. 3b, Nd-CoO_x maintains its activity at the current density of 10 mA cm⁻² for at least 800 h. The overpotential increases by only 76 mV after 800 h, corresponding to a degradation rate of just 95 μV h⁻¹. This result strongly confirms the good stability of Nd-CoO_x under acidic conditions^1,12. In comparison, commercial Co₃O₄ and CoO_x (without Nd) exhibit much shorter stabilities, lasting only 30 h—more than one order of magnitude shorter than Nd-CoO_x. Furthermore, Nd-CoO_x shows minimal degradation during cycling. After 5000 cyclic voltammetry (CV) cycles, the overpotential increases by just 15 mV at 10 mA cm⁻² (Fig. 3c). The stability of Nd-CoO_x also extends to large current densities. At 100 and 200 mA cm⁻², the catalyst operates stably for over 100 h without significant performance loss (Supplementary Fig. 12). In contrast, CoO_x without Nd shows rapid decay at 100 mA cm⁻², failing within 20 h (Supplementary Fig. 13b). The beneficial effect of Nd in CoO_x for acidic OER is also observed with other lanthanides, such as La and Pr. As shown in Supplementary Fig. 13a, Pr-CoO_x and La-CoO_x exhibit similar OER activities. Both catalysts maintain stability at 100 mA cm⁻² (Supplementary Fig. 13b) and operate for over 200 h at 10 mA cm⁻² (Supplementary Fig. 14). The results highlight the versatility of lanthanide incorporation in enhancing OER performance and stability.

To assess the practical application of Nd-CoO_x for electrochemical water electrolysis, a proton-exchange membrane electrolyzer was used (Supplementary Fig. 15a). The setup used Nd-CoO_x as the anode (for OER) and commercial Pt/C as the cathode (for HER). A Nafion 117 membrane was employed as the proton-exchange membrane. The current-voltage polarization curves (with iR correction) are shown in Fig. 3d. The Nd-CoO_x ║ Pt/C cell demonstrates higher catalytic activity when compared to the reference IrO₂ ║ Pt/C cell. Notably, the Nd-CoO_x ║ Pt/C electrolyzer needs only 1.93 V to obtain a large current density of 1 A cm⁻² at 80 °C, and 2.12 V for 2 A cm⁻². Both values surpass those of the IrO₂ ║ Pt/C electrolyzer, which requires 2.05 V at 1 A cm⁻² and 2.27 V at 2 A cm⁻². The current-voltage polarization curves (no iR correction) are also presented in Supplementary Fig. 15b. The stability of the PEM electrolyzer with Nd-CoO_x was measurd under constant current conditions (Fig. 3e). At the high current density of 200 mA cm⁻², the electrolyzer exhibited a stable operating voltage of approximately 1.69 V for over 100 h. Nd-CoO_x exhibits high efficiency and robust stability, making it competitive among noble-metal-free catalysts for OER in acidic environment, as summarized in Supplementary Table 4. These results highlight the potential of Nd-CoO_x as a good anode catalyst for practical water electrolysis systems.

In-situ investigations

It is well established that V_O can enhance the catalytic activity of Co₃O₄ for acidic OER, as observed with the CoO_x sample^2,11,13, However, the specific role of Nd in CoO_x, which significantly improves both OER activity and stability, remains unclear. This is particularly intriguing because a high concentration of vacancies often leads to structural failure during operation, resulting in poor catalyst durability², However, the Nd-CoO_x catalyst in this study demonstrates good OER stability, maintaining performance for at least 800 h. This robust stability likely arises from the critical role of Nd in stabilizing the CoO_x crystal structure. To investigate the working mechanism of Nd-CoO_x for stable and active acidic OER, in-situ XAS experiments have been conducted at both the Co K-edge and Nd L-edge during the OER process (Supplementary Fig. 16a). Figure 4a, b show the in-situ Co K-edge XANES and Fourier-transform curves of EXAFS spectra of CoO_x, respectively. The XANES spectra for CoO_x remain largely unchanged during OER. However, the corresponding Fourier-transform curves of EXAFS data reveal changes in bonding under applied potentials (Fig. 4b). Specifically, the Co-O bond length in CoO_x gradually shortens as the applied potential increases from the open-circuit potential (OCP) to 1.65 V vs. RHE. This shortening indicates a decrease in the Co-O distance and suggests a higher Co oxidation state under the applied potential. Similar behavior has also been observed for Co₃O₄ in the literature^11,13. Interestingly, as shown in the magnified view in Fig. 4b, the shortened Co-O bond does not fully recover to its original length after OER (during the OCP again measurement). This indicates some irreversible structural changes that likely contribute to the poor structural stability of CoO_x during acidic OER, as previously reported in the literature¹¹.

Fig. 4. In-situ XAS and Raman measurements.

a, d In-situ XANES spectra of CoO_x (a) and Nd-CoO_x (d) at Co K-edge under different potentials in 0.1 M HClO₄, respectively. b, e The corresponding Fourier-transform curves of EXAFS data for CoO_x (b) and Nd-CoO_x (e), respectively. c, f In-situ Raman spectra of CoO_x (c) and Nd-CoO_x (f) under different potentials, respectively. g Structure models for Nd-CoO_x during the OER process. h Bond length (Co-O and Nd-O) changes and the Raman shift of A_1g peak as a function of applied voltage.

The in-situ Co K-edge XANES and Fourier-transform curves of EXAFS spectra of Nd-CoO_x have been presented in Fig. 4d, e, respectively. The XANES spectra reveal a positive energy shift with increasing applied potential. This shift indicates a higher chemical state of Co under applied potential, which enhances OER activity^14,30. Notably, the energy shift fully recovers after the OER process, as shown in Fig. 4d. Further insights can be drawn from the Fourier-transform curves of EXAFS data in Fig. 4e. The Co-O bond length in Nd-CoO_x changes progressively with increasing potential. Interestingly, the Co-O bond shows an increase in length, opposite to the behavior observed for CoO_x (see the magnified region from 1.45 to 1.51 Å). Typically, as the chemical state of Co increases with higher potential, the Co-O bond shortens, as seen in CoO_x. However, the presence of Nd alters this behavior by modifying the local atomic structure to maintain structural stability. As discussed in Figs. 1 and  2, Nd-insertion significantly compresses the Co-O bond to a shorter initial length than that in normal Co₃O₄. During applied potentials, the Nd-O bond also undergoes changes, influencing the surrounding Co-O bonds. As shown in Supplementary Fig. 16d, the initial Nd-O FT peak at 1.94 Å shifts to a shorter FT length of 1.89 Å under applied potential. This reduced N-O distance implies a higher chemical state of Nd. The shortened Nd-O bond consequently exerts a stretching effect on the surrounding Co-O bonds, resulting in the observed elongation of the Co-O bond in Fig. 4e. Remarkably, both the elongated Co-O bond and shortened Nd-O bond fully recover to their initial states after the OER process (Fig. 4e and Supplementary Fig. 16d). This recovery highlights the good ability of Nd-CoO_x to maintain structural stability during OER, presenting a significant advantage for long-term operation.

To clearly illustrate the bond length changes during OER, structural models of Nd-CoO_x with and without applied potential are presented in Fig. 4g. These models highlight the “breathing mode” behavior, where Co-O and Nd-O bonds undergo squeezing and stretching processes to balance atomic-level structural stress during OER. The bond length changes in Nd-CoO_x and CoO_x are compared in Fig. 4h. For CoO_x, only the Co-O bond changes, decreasing in length with increasing potential. Importantly, this bond does not fully recover after OER, indicating irreversible structural changes. In contrast, for Nd-CoO_x, the Nd-O bond plays an important role. Changes in the Nd-O bond significantly influence the Co-O bond length, which increases along with applied potential but fully recovers after the reaction. This difference likely explains the disparity in OER stability between Nd-CoO_x and CoO_x. In CoO_x, gradual structural changes during acidic OER result in structural stress and eventual collapse under long-term operation. In Nd-CoO_x, the Nd-insertion introduces a controllable and reversible structure change, allowing it to maintain good stability. R-space fitting curves and results for Nd-CoO_x during OER, presented in Supplementary Fig. 17 and Supplementary Table 5, further confirm these bond length changes. This dynamic structural adaptability underscores the robust stability of Nd-CoO_x in challenging acidic environments.

In-situ Raman spectroscopy was performed to further investigate the “breathing mode” described above (Supplementary Fig. 16b). The Raman spectra for both CoO_x and Nd-CoO_x, shown in Fig. 4c, f, exhibit characteristic peaks corresponding to the typical E_g, F_2g, and A_1g features of Co₃O₄, as reported in the literature^30,31. With increasing potential from the open-circuit potential (OCP) to 1.65 V vs. RHE, the main A_1g peak shifts significantly. For CoO_x, the A_1g peak shifts from 694.1 cm⁻¹ to 694.8 cm⁻¹, indicating lattice contraction due to shortened Co-O bonds under applied potential^11,30. However, after the OER process, this blue shift does not recover to its initial value (Fig. 4c). This irreversible shift suggests continuous structural stress in the harsh acidic environment, leading to structural failure and limited stability. This behavior aligns with observations reported in previous studies¹¹. For Nd-CoO_x, a similar blue shift of the A_1g peak is observed, from 692.6 cm⁻¹ to 694.7 cm⁻¹, corresponding to lattice contraction. However, unlike CoO_x, the peak fully recovers once the applied potential is removed, as shown in Fig. 4f. This recovery is attributed to the controllable and reversible structural changes enabled by Nd-insertion, which contributes to the catalyst’s long-term stability. The A_1g peak shifts with applied potential for both CoO_x and Nd-CoO_x are compared in Fig. 4h. In CoO_x, the peak shows irreversible changes, while in Nd-CoO_x, it exhibits flexible and fully recoverable shifts across a relatively larger range. The initial lower A_1g value for Nd-CoO_x (692.6 cm⁻¹) is likely due to lattice expansion caused by the incorporation of larger Nd atoms into the structure. The in-situ Raman results are consistent with the in-situ XAS findings, further confirming the “breathing mode” in Nd-CoO_x. This dynamic mode allows for structural flexibility and stability under harsh OER conditions. Atomic-level structural changes during long-term operation for both Nd-CoO_x and CoO_x are illustrated in Supplementary Fig. 18a, b, respectively. These results highlight the critical role of Nd-insertion in enhancing structural adaptability and stability.

The XAS results for Nd-CoO_x, recorded before and after 800 h operation at 10 mA cm⁻², are presented in Supplementary Fig. 19. Measurements at both the Co K-edge and Nd L₃-edge reveal that Co and Nd exhibit more oxidized states after the long-term stability test. Despite this oxidation, the catalyst maintains its overall electronic structure, consistent with its robust stability. Additional structural analyses, including XRD, SEM, and TEM results for Nd-CoO_x after 800 h of operation, are shown in Supplementary Fig. 20. These analyses reveal that the morphology and crystalline structure of Nd-CoO_x remain largely unchanged compared to its initial state before OER. This strongly supports the critical role of the flexible “breathing mode” enabled by Nd-insertion in maintaining structural integrity and stability during extended OER operation.

DFT calculations

Density functional theory (DFT) calculations have been performed to explore the OER working mechanism of Nd-CoO_x. According to the aberration-corrected HAADF-STEM result in Fig. 1e and XRD patterns in Supplementary Fig. 2, Co₃O₄ (311) is the dominant facet, which has thus been used for the calculations. The CoO_x catalyst was mimicked by a Co₃O₄ (311) surface (standard spinel structure), which incorporated oxygen vacancy sites. The computational model of Nd-CoO_x was constructed by replacing surface Co atoms with Nd atoms. The configuration of Nd-CoO_x is displayed in Supplementary Fig. 21, and the optimized atomic structures around Co and Nd are shown in Fig. 5a. The Co-O bond length in CoO_x is approximately 1.915 Å, consistent with values reported in the literature^2,32,33. In Nd-CoO_x, the incorporation of Nd shortens the Co-O bond to 1.878 Å and extends the Nd-O bond to approximately 2.366 Å. The shortened Co-O bond results from structural distortions induced by Nd, which has a larger atomic radius. These changes in Co-O bond length align closely with the XAS results in Fig. 2. We further perform the calculation to simulate the status with applied potential based on the constant electrode potential model (with structural relaxation and self-consistent field calculations). When an external potential correction of 1.55 V (vs. vacuum energy level) is simulated using the constant electrode potential model, the Co-O bond in Nd-CoO_x stretches slightly to 1.885 Å, while the Nd-O bond shortens to 2.325 Å (Fig. 5a). These computational findings strongly support the presence of a flexible “breathing mode” in Nd-CoO_x, as revealed by in-situ XAS and Raman spectroscopy.

Fig. 5. DFT calculations and the working mechanism.

a Optimized computational models of CoO_x and Nd-CoO_x without applied potential, along with the model of Nd-CoO_x with potential correction (1.55 V with reference to the vacuum energy level). b The DEMS signals of ³²O₂, ³⁴O₂ and ³⁶O₂ tested for Nd-CoO_x in H₂¹⁸O acidic solution. c Gibbs free energy diagrams for Nd-CoO_x and CoO_x through AEM and OPM pathways at 0 V and 1.55 V (with constant electrode potential corrections). d Illustration of the OPM pathway on Nd-CoO_x. e Schematic diagrams of the structural variations of CoO_x and Nd-CoO_x before and after OER.

Differential electrochemical mass spectrometry (DEMS) was used to experimentally identify the OER reaction pathway (Supplementary Fig. 22a). The detailed analysis procedures are illustrated in the Supplementary Information. As shown in Fig. 5b, the Nd-CoO_x catalyst, after five cycles of cyclic voltammetry in an acidic solution containing H₂¹⁸O, produces three gaseous products: ³²O₂, ³⁴O₂, and ³⁶O₂. According to the literature^34,35. The adsorbate evolution mechanism (AEM) typically generates only ³⁴O₂ and ³⁶O₂, without producing ³²O₂. The detection of ³²O₂ in this study indicates a reaction pathway different from AEM, including direct coupling between surface-adsorbed ¹⁶O atoms on Co sites and lattice ¹⁶O, or ¹⁶O atoms on adjacent metal (such as Nd) sites. This behavior is characteristic of the lattice-oxygen-mediated mechanism (LOM) or the oxide path mechanism (OPM)^1,36. Further evidence comes from tetramethylammonium (TMA⁺) experiments, shown in Supplementary Fig. 22b. The results show minimal changes in the reaction when TMA⁺ is added, effectively ruling out the LOM pathway^37,38. Considering all experimental results, the evidence suggests that the OER process on the Nd-CoO_x catalyst follows the OPM pathway. This mechanism has been illustrated in Fig. 5d.

DFT calculations have been further used to probe the reaction mechanism of OER on Nd-CoO_x. Gibbs free energy (ΔG) diagrams of the AEM and OPM pathways are presented in Fig. 5c and Supplementary Data 1. In the AEM pathway, the reaction proceeds via a sequence of intermediates: OH*, O*, OOH*, and finally O₂ formation on the active sites. The potential-determining step (PDS) in this pathway should be the conversion of OH* to O*, which requires a significant free energy uphill of 2.166 eV. In contrast, the OPM pathway follows a different sequence: OH* → OH*-OH* → O*-OH* → O₂. The PDS in this case is the transition from O*-OH* to O₂ (see Fig. 5c), which involves a substantially lower ΔG of 1.879 eV. These findings suggest that, thermodynamically, the OPM pathway is much more favorable than the AEM pathway. The impact of applied bias voltages on the OER free energy profiles was also studied using constant electrode potential models. At a higher bias voltage of U = 1.55 V, all steps in the OPM pathway become thermodynamically downhill, while the AEM pathway still requires an energy uphill of 0.301 eV. By comparing the energy profiles of the two pathways, it is clear that the sluggish kinetics of the AEM pathway arise from the slow formation process of O* intermediate with high-energy. Meanwhile, the OPM pathway prevents such high-energy intermediates entirely, resulting in more efficient kinetics. For comparison, the Gibbs free energy diagrams for CoO_x through AEM and OPM pathways are also provided in Fig. 5c and Supplementary data 1. Schematic illustrations of the adsorbed intermediates on optimized CoO_x structure through the AEM and OPM pathways can be found in Supplementary Fig. S23 and S24, respectively. The AEM pathway is more favorable for CoO_x according to the calculations. However, the PDS energies for CoO_x following the AEM pathway are 2.205 eV (0.00 V) and 0.266 eV (1.55 V), which are much higher than those for Nd-CoO_x following the OPM pathway (1.879 eV (0.00 V) and −0.013 eV (1.55 V)). These results demonstrate that Nd incorporation modifies the fundamental OER mechanism from AEM to OPM and then significantly enhances the catalytic performance by lowering the reaction energy barriers. This conclusion is well-aligned with the enhanced catalytic activity of Nd-CoO_x, as demonstrated in Fig. 3a.

The charge transfer analysis can be observed in Supplementary Fig. 25, in which O* adsorption on CoO_x results in a large electron transfer (0.71 e^-) from the catalyst, with an adsorption free energy of 1.366 eV (far from the ideal value of 2.46 eV). In contrast, O* adsorption on Nd-CoO_x exhibits a weaker interaction, with only 0.53 e^- transferred, leading to a much higher adsorption free energy of 2.754 eV (close to the ideal value of 2.46 eV). The electronegativity difference between Nd and Co induces the surface charge redistribution, thereby affecting the adsorption energy of intermediates and triggering the OPM pathway. A comparison between Nd-CoO_x (with oxygen vacancies) and Nd-Co₃O₄ (without oxygen vacancies) is also shown in Supplementary Fig. 25c and 25d, demonstrating that oxygen vacancies further reduce the adsorption of the O*-OH* intermediate (reduced electron transfer and longer Co-O bonds). The calculated energy values on CoO_x and Nd-CoO_x are provided in Supplementary Table 6.

The working mechanism of Nd-CoO_x is illustrated in Fig. 5e. For the CoO_x catalyst, OER activity is enhanced by the presence of vacancies. However, under acidic conditions and high applied potentials, the active Co sites become further oxidized. This oxidation shortens the Co-O bond length, leading to continuous lattice contraction. Over time, unevenly distributed structural stress, especially around the vacancies, causes structural failure (Fig. 5e). As a result, CoO_x displays limited OER stability. In contrast, the insertion of Nd into CoO_x introduces a flexible “breathing mode” in the atomic structure. This flexibility helps resist structural failure under applied potential, ensuring sustained durability during acidic OER. Additionally, the presence of Nd near Co enhances the OER activity by enabling the reaction to proceed via the OPM pathway. This results in a catalyst that is both stable and highly active, making it suitable for PEMWE. The beneficial role of Nd in enhancing the performance of CoO_x offers a universal strategy that could be extended to other lanthanides, such as La or Pr, as discussed in Fig. 3f. This approach highlights the potential of lanthanide incorporation for improving catalytic performance in acidic OER.

Discussion

In this work, we report the Nd-CoO_x catalyst as a stable and efficient material for acidic OER. It achieves a low overpotential of 317 mV at 10 mA cm⁻² and demonstrates robust durability, operating stably for at least 800 h. The incorporation of Nd introduces a “breathing mode”, characterized by the squeezing and stretching of Nd-O and Co-O bonds during OER. This dynamic behavior, identified through in-situ XAS and Raman spectroscopy, balances atomic-level structural stress, significantly enhancing stability under acidic conditions. DFT calculations further reveal a low energy barrier of the catalyst for acidic OER via the OPM pathway. When used as the anode in a PEMWE, the Nd-CoO_x catalyst delivers good performance. It requires only 1.69 V to obtain a high current density of 200 mA cm⁻² over 100 h and reaches 1 A cm⁻² at a cell voltage of 1.93 V (2.12 V at 2 A cm⁻²).

Methods

Chemicals and reagents

Commercial Co₃O₄ (99.9%), Co(NO₃)₂·6H₂O (99%), Nd(NO₃)₃·6H₂O (99.9%), La(NO₃)₃·6H₂O (99.9%) and Pr(NO₃)₃·6H₂O (99.9%) have been obtained from Shanghai Aladdin Industrial Co., Ltd. Ethylene glycol (EG, 99%), Ethanol (EA), Isopropyl alcohol (IPA, 99.5%) and IrO₂ (99.9%) have been obtained from Shanghai Macklin Biochemical Industrial Co., Ltd. HClO₄ (70-72%) has been obtained from Yonghua Chemical CO., LTD. Carbon cloth (CC) has been obtained from Suzhou Keshenghe Metal Materials Co., Ltd. Nafion (5 wt%) has been obtained from Sigma-Aldrich (Shanghai) Trading Co., Ltd. Platinum carbon loaded on carbon paper (20 wt% Pt/C) and cation exchange membrane (Nafion 117) have been obtained from Suzhou Sinero Technology Co., Ltd. Deionized water (resistivity ≥18.2 MΩ) was produced using an ultrapure water purification system. All reagents were of analytical grade and used directly without additional purification.

Preparation of CoO_x and M-CoO_x porous nanosheets on carbon cloth

The CoO_x catalysts were prepared through microwave heating and the solution combustion method. Carbon cloth (CC) (2 cm × 4 cm) was carefully cleaned under ultrasonication treatment to remove the surface oxides. Then 3.6 mM Co(NO₃)₂·6H₂O and 5 mL Ethylene glycol (EG) were mixed to form a homogeneous solution. The cleaned CC was put in the above solution for 20 s with microwave heating. Subsequently, the CC was taken out and ignited directly with an alcohol lamp, and the final catalyst was obtained after repeating the ignition several times. M-CoO_x was synthesized by using the same steps, except for the addition of 0.4 mM (M(NO₃)₃·xH₂O (M = La, Pr, Nd). The optimal loading of Nd-CoO_x catalysts was around 7.5 mg cm⁻² on CC. Nd-CoO_x electrocatalyst prepared on CC was used as a working electrode in a three-electrode system. Nd-CoO_x electrocatalyst powder was scraped from the CC and used as an anode catalyst in PEMWE.

Preparation of IrO₂ electrodes

7.5 mg IrO₂, 20 μL Nafion (5 wt%), 250 μL ethanol, and 230 μL deionized water were mixed and then ultrasonicated for 30 min to form a homogeneous ink. 50 μL ink was dropped on carbon cloth and then dried at 60 °C in va acuum for 12 h to form the IrO₂ electrode (0.25 cm⁻²). It was around 3 mg cm⁻² IrO₂ on carbon cloth.

Proton-exchange membrane (PEM) cell test

Nd-CoO_x (or IrO₂) has been used as the anode with Pt/C on carbon paper (Pt/C-CP) as the cathode in a PEM cell. 5 mg Nd-CoO_x was added into a mixture of isopropanol (500 μL), deionized water (450 μL) and Nafion solution (50 μL, 5 wt%). After sonication in a water bath for 60 min, a homogeneous anode ink was obtained. To construct the membrane electrode assembly (MEA), the Nafion 117 membrane was sprayed with anodic ink under vacuum adsorption to prevent swelling, with the anodic loading about 5 mg_cat cm⁻². Subsequently, the Pt/C-CP, 0.4 mm Ti plate and the Nafion 117 membrane (loaded with anodic ink) were hot pressed at 363 K for 30 s, obtaining the membrane electrode assembly (MEA). The PEM cell was measured in pure water for 60 min at 80 °C (effective area around 1.0 cm⁻²).

Materials characterization

SEM (ZEISS G500), HRTEM (TALOS 200X), ICP-OES (Avio 200), XPS (Thermo Fischer ESCALAB 250X), and XRD (Empyrean powder diffractometer) have been employed for structural characterizations. XAS measurements were performed at the Shanghai Synchrotron Radiation Facility (https://cstr.cn/31124.02.SSRF.BL11B) and National Synchrotron Radiation Laboratory (https://cstr.cn/31131.02.HLS.XMCD.a and https://cstr.cn/ 31131.02.HLS.XMCD.b). The XAS data were analyzed by using the ATHENA and ARTEMIS software³⁹. The XANES spectra were firstly normalized by subtracting pre-edge and post-edge background, and then treated by calibrating the energy shift based on E₀ of the reference samples. The EXAFS results were analyzed by using the Fourier transform of k-space data in a window (Hanning window with dk = 1) of 2.0−11.0 Å⁻¹ for Co K-edge, and k-weighting was set to three without phase shift. The selected k ranges (2.0−11.0 Å⁻¹ for Co K-edge) were also used for wavelet transform analysis (Morlet function, kappaMorlet = 15, sigmaMorlet = 1). The first shell peaks of the FT spectra were fitted by using shell structural models based on the crystal structures of Co₃O₄.

Electrochemical measurements

A standard three-electrode system (CHI 760E) was used at room temperature ( ~ 25 °C). 0.1 M HClO₄ solution was used in all tests with Pt wire, Ag/AgCl, and the synthesized samples (0.5 × 0.5 cm) as counter, reference, and working electrodes, respectively. Backward scan (from positive to negative potentials, 5 mV s⁻¹) was utilized to avoid the oxidative peaks. A scan rate of 100 mV s⁻¹ was applied during the long-term cyclic voltammetry (CV) measurements. The potential was converted to the value versus the reversible hydrogen electrode (RHE) according to Eq. (1)

$$E_{RHE}=E_{Ag/AgCl}+E_{Ag/AgCl}^0+0.059\times pH$$ 1

where E⁰_Ag/AgCl is the difference value between standard hydrogen electrode and Ag/AgCl electrode. The pH value of the electrolyte was measured multiple times, and the average value obtained was approximately 1.08 ± 0.02. E⁰_Ag/AgCl was determined by referencing to a Pt/C working electrode in a H₂-saturated solution, and the corrected potential was about 0.2 V. The electrochemical impedance spectroscopy measurement (EIS) measurements were conducted at a current density of 10 mA cm⁻², with an additional alternating voltage of 5 mV applied over the frequency range of 0.1 Hz to 100 kHz.

CV test was performed within a small potential window (0.86 – 0.96 V vs Ag/AgCl) at scan rates from 20 to 100 mV s⁻¹ (20 mV s⁻¹ intervals) to determine the double-layer capacitance (C_dl), from which the electrochemically active surface area was estimated. A 90% iR compensation was applied to the polarization curves obtained at high current densities. Chronopotentiometry (CP) measurements were performed to assess the long-term stability under various current densities.

Preparation of 0.1 M HClO₄ solution: Measuring approximately 8.6 mL of concentrated HClO₄ (70%–72%) and slowly adding it, under stirring, to deionized water (500 mL) in a beaker. Following cooling, the solution was quantitatively transferred to a 1000 mL amber volumetric flask, diluted to the mark with deionized water, and homogenized. Finally, the electrolyte was stored in a dark, cool environment at ambient temperature.

DFT calculations

All the DFT calculations were conducted by using Vienna ab initio Simulation Package (VASP 5.4.4)^40,41. The electron-ion interactions were described in terms of the projector augmented-wave (PAW) potentials⁴². The exchange and correlation interactions between electrons were described by the revised Perdew-Burke-Ernzerhof (PBE) within the generalized gradient approximation (GGA-RPBE)⁴³. The Co-Nd system was constructed as a model for theoretical calculations. The plane wave cutoff for these systems was 500 eV. The criterion of convergence of energy was set to 1 × 10⁻⁴ eV. The internal coordinates of each system were fully optimized until the residual Hellmann-Feynman forces were less than 0.02 eV/Å per atom. The lattice parameters in the direction perpendicular to the surface were set as 30 Å, which could thus avoid interactions between periodic slabs. The Brillouin zones were sampled by 1 × 2 × 1 k-point meshes generated according to the Monkhorst-Pack scheme. The Hubbard U (DFT + U) corrections for 3 d and 4 f species were made. Based on the previous works, the U_eff of Co was set as 3, and the U_eff value of Nd (4f-transition metal) was set as 5^44,45. The reaction Gibbs free energies (ΔG) could be obtained using Eq. (2):

$$\mathrm{\Delta }G=\mathrm{\Delta }E+\mathrm{\Delta }{E}_{ZPE}-T\mathrm{\Delta }S$$ 2

In which E was the energy that could be directly obtained from the DFT calculations.${\text{E}}_{\text{ZPE}}$ was zero-point energy. T was 298.15 K in this work, and S was the entropy. Constant potential calculations were employed to unravel the properties of electrochemical processes as a function of the electrode potential. Therefore, the reaction free energies for Nd-CoO_x were calculated under constant electrode potentials. The electrode potential (U) referenced to that of SHE was given by Eq. (3)⁴⁶:

$$U=\frac{-\mu \left(e^{-}\right)-{\mathrm{\Phi }}_{SHE}}{e}$$ 3

In which -μ (e⁻) and ${\mathrm{\Phi }}_{\text{SHE}}$ represented the chemical potential of electrons and work function of the SHE, respectively. We used 4.4 eV for ${\mathrm{\Phi }}_{\mathrm{SHE}}$⁴⁷. The information energy ($E_f$) was calculated by Eq. (4):

$$E_f=E_{total}-E_{substrate}-E_{Co}$$ 4

Where $E_{total}$, $E_{substrate}$, and $E_{Co}$ were the energies of the whole catalyst, the system without Co atom, and an isolated Co atom in vacuum, respectively. The dissolution potentials of cobalt ($U_{diss}$) was obtained by Eq. (5):

$$U_{diss}=U_{diss}^0-E_f/Ne$$ 5

Where $U_{diss}^0$ was the standard dissolution potential of Co atom, and N was the transferred electron number during the dissolution⁴⁸.

Author contributions

J.Z. designed the overall project. K.F. and Y.F. performed most of the experiments, including material synthesis, characterization, and electrochemical tests. H.P.L., X.F. and J.H.W. contributed to the DFT calculations. B.X., J.B.X., S.L., and C.L. assisted with the XAS experiments. All the authors discussed the results and contributed to writing the paper.

Peer review

Peer review information

Nature Communications thanks Paola D’Angelo and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data that support the plots within this article and other findings of this study are available within the article and its Supplementary Information. Any additional relevant data is available from the corresponding authors upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-66408-1.