Facet-selected oxygen-deficient ruthenium oxide for acidic oxygen evolution catalysis

Abstract

Proton exchange membrane water electrolysis enables sustainable hydrogen production, where ruthenium dioxide catalysts are highly active for the oxygen evolution reaction but suffer from degradation associated with the lattice oxygen oxidation mechanism. Conventional stabilization strategies mainly focus on heteroatom doping, but often diminish intrinsic Ru active sites. Here, we propose a Ru-intrinsic and defect-inheritance strategy that does not rely on doping and yields a reconstructed RuO₂ with surface oxygen-deficiency on preferentially exposed (101) facet (od-RuO₂). These facet-selected oxygen-deficient Ru sites strengthen the hybridization of Ru t_2g orbitals and π-donor oxygen ligands, shifting and broadening the t_2g^* states towards the Fermi level, thereby facilitating OOH bond formation through the adsorbate evolution mechanism dominated pathway. Herein, od-RuO₂ achieves an overpotential of 161 mV at 10 mA cm⁻² and operates for over 3000 hours, and also sustains membrane electrolyser operation, demonstrating its potential for durable acidic oxygen evolution catalysis.

Ruthenium oxide is highly active for acidic water electrolysis, but lattice oxygen oxidation limits durability. Here, the authors engineer oxygen-deficient (101) facet RuO₂ that favours an intrinsic stable pathway, simultaneously delivering high activity and stability and guiding catalyst design.

Introduction

Hydrogen (H₂) is a high-energy-density (120–140 MJ kg⁻¹) fuel with zero carbon emissions as a clean energy carrier, playing a crucial role in decarbonizing sectors such as transportation, industry, and power generation. Although alkaline electrolysis (AEL) has been widely adopted for large-scale hydrogen production because of its cost-effectiveness and technological maturity¹, proton exchange membrane water electrolysis (PEMWE) offers a more advanced alternative owing to its high current density, lower ohmic losses, superior hydrogen purity, compact design, and rapid response to intermittent renewable energy inputs^2–4. However, the large-scale application of PEMWE is constrained by the inherently sluggish anodic oxygen evolution reaction (OER), which is attributable to its complex four-electron transfer mechanism, leading to high overpotential and severe energy losses^5–7. To increase PEMWE efficiency, highly active and durable OER catalysts are needed, as they directly determine the energy consumption and operational lifetime of the electrolyser.

Currently, noble metal oxides such as ruthenium oxides (RuO₂) and iridium oxides (IrO₂) represent the most promising candidates for acidic OER catalysis^8–10, demonstrating superior performance in PEMWE. Owing to its high stability, which is primarily attributed to its operation via the adsorbate evolution mechanism (AEM) pathway, IrO₂ has been widely adopted in commercial PEMWE. However, its moderate intrinsic activity necessitates high loadings at the anode (typically 1.5–2.0 mg cm^-2), thereby restricting the PEMWE scalability for industrial hydrogen production due to the scarcity and high cost of iridium^11–14. To overcome these constraints, research has increasingly turned toward Ru-based catalysts (especially RuO₂), which offer markedly higher intrinsic OER activity and are approximately an order of magnitude less expensive than their Ir-based counterparts^15–17, thereby making them highly attractive for next-generation acidic OER catalysis. However, the lattice oxygen oxidation mechanism (LOM) identified in RuO₂ catalysts¹⁸, which depends on direct lattice O–O coupling, accelerates the activity at the cost of structural degradation arising from lattice oxygen depletion during water electrolysis^11,19,20.

To address this limitation, recent strategies to switch RuO₂ from the LOM towards the more stable AEM pathway have been developed. These strategies are centered on doping or intercalating guest elements, such as V, Nb, Mn, or Pt, into RuO₂ to induce favorable electronic or structural changes during the OER^21–25. However, the performance of these catalysts highly depends on the guest atoms, which introduce structural heterogeneity and induce phase separation, thereby compromising the intrinsic stability. In addition, guest atom incorporation can decrease the number of active Ru sites, thus suppressing the catalytic activity and obscuring the contributions of the mechanisms. Therefore, identifying an intrinsic structural design strategy to reliably promote the AEM pathway in RuO₂ without relying on guest elements to achieve durable and high-performance acidic OER catalysts remains an urgent challenge.

Herein, to increase the stability of Ru-based catalysts while maintaining high activity for the OER, we report a surface oxygen-deficient ruthenium oxide nanocluster (od-RuO₂) catalyst for acidic overall water splitting that is explicitly designed to adhere to the AEM principle. The catalyst was synthesized via a scalable sol-gel and hydrothermal approach. Through regulation of the reconstruction of the SrCo_1-xRu_xO_3−δ perovskite precursor, RuO₂ undergoes confined growth from the cobalt spinel oxide, yielding nanoscale RuO₂ clusters with marked unsaturated coordination Ru sites preferentially exposed on the dominant (101) surface. Theoretical calculations indicate that compared with other common planes with surface oxygen deficiency, this oxygen-deficient (101) lattice plane is thermodynamically more stable. These surface oxygen-deficient Ru sites facilitate preferential adsorption of hydroxide ions (OH^-), enabling efficient electron transfer via the AEM pathway. The catalyst achieved an overpotential at 10 mA cm⁻² (η₁₀) of 161 mV and maintained stable operation for more than 3000 h, positioning it as a competitive acid-stable OER catalyst. In a practical PEMWE device, od-RuO₂ exhibited a low voltage of 1.51 V at 1.0 A cm⁻² and sustained operation for more than 1200 h at 1.0 A cm⁻². By elucidating the intrinsic-structure and mechanism relationship governing the AEM pathway in Ru-based oxides, this work provides a feasible strategy for designing efficient and durable Ru-based catalysts, with strong prospects for practical PEMWE applications.

Results

Ru-O π ligand strengthening strategy

To determine the feasibility of Ru-based catalysts being rationally engineered to achieve both high stability and high activity for the OER, we first investigated their intrinsic electronic properties through theoretical analysis. According to the fundamental principles of the AEM, the low covalency of the metal–oxygen (M–O) bond, to ensure removal of electrons from Ru 4d orbitals rather than O 2p orbitals, is an essential condition for AEM realization²⁶. However, conventional RuO₂ represents a typical high-covalency case, characterized by a small energy gap between the Ru 4d and O 2p states (a small charge transfer energy Δ, as shown in Fig. 1a). As a result, the t_2g^* band retains substantial O 2p character near the Fermi level (E_F)^27,28, which is unfavorable for the AEM because it increases lattice oxygen participation in redox processes. In addition, compared to the σ-type overlap with the Ru e_g orbitals ($d_{z^2}$ and $d_{x^2-y^2}$, which point directly towards the ligands along the bond axes, as shown in Fig. 1b), the π-type overlap with the t_2g orbitals corresponds to a smaller splitting and is weaker (d_xy, d_xz, and d_yz, which are oriented between the oxygen ligands and thus overlap with the O 2p π orbitals sideways relative to the Ru–O bond axes, as shown in Fig. 1c). Consequently, the hybridization of the t_2g and O_2p orbitals is incomplete. These two factors make the lattice oxygen more redox-active, enabling direct participation in the O–O bond formation step (as the LOM). Therefore, to suppress this undesirable pathway and promote the AEM pathway, a strategy to strengthen the hybridization between Ru t_2g orbitals and π-donor oxygen ligands by designing intrinsic structurally engineered RuO₂ catalysts is proposed.

Fig. 1. Fundamental analysis of the catalyst design strategy.

a Molecular orbital diagram of RuO₂. b, c Shape and directionality with respect to the σ and π ligands of the Ru e_g and t_2g orbitals for RuO₂. d Electron configuration of 4d electrons in od-RuO₂, with higher t_2g* occupation via coordinatively unsaturated Ru atoms. e Comparison of the PDOS projected on the p orbitals of O atoms and the d orbitals of Ru atoms for common RuO₂ oxygen-deficient planes. f PDOS projected on the p orbitals of O atoms and the d orbitals of Ru atoms for the RuO₂ saturated (101) plane (top) and oxygen-deficient (101) plane (bottom). g Diagram of the band modulation strategy for switching from the LOM to the AEM pathway. The LOM pathway involves the combination of two lattice oxygen atoms (O_lat), or an O_lat with an adsorbed oxygen atom (O_ads), while the AEM pathway only conducts the combination of two O_ads.

Theoretically, π-donor oxygen ligands can be strengthened by introducing coordinatively unsaturated Ru atoms^29,30, which promote stronger Ru–O π interactions and facilitate charge redistribution near the Fermi level (a relatively lower Ru valance and higher t_2g^* occupation, as shown in Fig. 1d). In addition, when oxygen vacancies, which are ubiquitous in metal oxides³¹, become sufficiently abundant in sufficiently small (nanoscale) particles, the exposed surfaces naturally feature abundant surface oxygen vacancies. These surface oxygen vacancies, corresponding to surface oxygen-deficient Ru atoms, are precisely the targeted active sites, which can provide stronger adsorbate binding than normal oxide sites and assist in adsorbate dissociation at the surface³². This stronger adsorbate binding stabilizes key intermediates (OH^*, O^*, and OOH^*), while promoted adsorbate dissociation helps sustain the AEM pathway by preventing lattice oxygen from directly participating in O–O bond formation. Guided by this insight, to identify a favourable facet for surface oxygen-deficient Ru atoms, theoretical density functional theory (DFT) calculations were conducted. The DFT calculations were performed using idealized models under vacuum conditions. The results showed that the (101) plane of RuO₂ possesses both high surface stability and a favorable electronic configuration for AEM. As the AEM pathway benefits from minimal occupation of O 2p orbitals near the Fermi level³³, projected density of states (PDOS) calculations across all the planes reveal that the (101) surface displays a positive energy band separation (ΔE = +0.29 eV, as the energy difference between the lowest Ru 4d state and the highest O 2p state near the Fermi level), significantly distinct from the negative ΔE of other low-index planes (Fig. 1e and Supplementary Fig. 1). This pronounced band separation effectively reduces Ru–O covalency and suppresses O 2p participation, thereby promoting electron transfer from Ru 4d orbitals rather than lattice oxygen. Such an electronic configuration provides a clear basis for the AEM preference on the (101) plane, accounting for its stability under acidic OER conditions. Moreover, comparative analysis of saturated and unsaturated (101) surfaces indicates that the introduction of surface oxygen vacancies on (101) plane can markedly increase the Ru t_2g and O π hybridization by broadening and upshifting the Ru 4d t_2g band occupation (band center ε_d from –3.09 eV to –2.76 eV) and reducing the electron density of O 2p orbitals (band center ε_p from –2.61 eV to –2.77 eV) near the Fermi level, thereby promoting preferential adsorption of OH⁻ at Ru active sites (Fig. 1f). Overall, oxygen-vacancy formation lowers the Ru valence and increases the Ru-centered 4d electron density, while the designed oxygen-deficient (101) facet induces an upshifted and broadened Ru 4d distribution, increasing the Ru 4d contribution in the frontier-energy region near the Fermi level. Together with the downward shift of O 2p states, this electronic redistribution promotes Ru-centered adsorbate evolution and suppresses lattice oxygen redox participation. Concurrently, the pre-existing surface vacancies on the (101) surface not only enable the OER to efficiently proceed at surface sites but also impart strong resistance to new internal lattice oxygen loss and minimize structural degradation, as evidenced by its highest calculated oxygen vacancy formation energy (E_Ovac = 5.91 eV) among the commonly observed low-index RuO₂ surfaces (Supplementary Fig. 2 and 3).

In general, the orbital hybridization calculations and oxygen vacancy formation energy calculations complement each other and establish the intrinsic electronic and structural basis for AEM-favorable od-RuO₂. Therefore, unlike previous approaches that relied solely on dopant modification or external strain to adjust electronic structures, this method achieves intrinsic modulation of Ru–O orbital interactions by rational structural design. The overall strategy for redox center modulation—combining π-donor ligand strengthening through oxygen vacancy engineering with facet-selective (101) exposure—is schematically illustrated in Fig. 1g.

Defect-inheritance synthesis strategy

To realize the above described surface oxygen-deficient configuration, perovskites are preferentially considered precursors. Typically, in SrCoO_3-δ, the cobalt atoms inherently exhibit highly unsaturated coordination (3-δ ≈ 2.52)³⁴, making SrCoO_3−δ a particularly suitable host for incorporating Ru atoms and initiating controlled lattice reconstruction. Therefore, a tailored defect-inheritance synthesis route involving a straightforward and scalable sol-gel method with SrCoO_3-δ as a precursor, followed by a hydrothermal process, was employed. First, metal nitrates were gelled to form a liquogel, which was air annealed to transform it into a Ru-doped perovskite precursor, SrCo_1-xRu_xO_3-δ (Fig. 2a and Supplementary Fig. 4). Within this perovskite lattice, Ru and Co atoms are strongly bonded within oxygen octahedra, whereas Sr atoms occupy weakly bonded cuboctahedral sites and readily leach under acidic hydrothermal conditions (Fig. 2b). This selective Sr dissolution triggers complete structural reconstruction, during which the confinement effect of the cobalt spinel oxide promotes nanoscale growth of RuO₂. Most residual cobalt oxide species, which exist as independent phases, can be removed during the subsequent rinsing process, as confirmed with inductively coupled plasma mass spectrometry (ICP–MS) results (Supplementary Fig. 5). As a result, the reconstructed RuO₂ inherits the oxygen-deficient environment of the perovskite, yielding the final od-RuO₂ catalyst (Fig. 2c). The aforementioned (101) lattice plane is particularly suited for stabilizing surface oxygen vacancies, as the (101) plane has the lowest surface energy (E_surf) among the commonly observed low-index RuO₂ surfaces when vacancies are primarily distributed at the surface (Supplementary Fig. 6), indicating its thermodynamic stability that favors its preferential formation, verifying the thermodynamic feasibility of forming the oxygen-deficient (101) surface during synthesis reconstruction. By leveraging a rational perovskite-to-oxide transformation, this defect-inheritance synthesis strategy enables precise control over the spatial distribution of surface defects and electronic structure, which ensures the synthesis of od-RuO₂ nanoclusters with a high density of AEM-favorable active sites, integrating structural stability, electronic optimization, and synthetic feasibility into a single high-performance OER catalyst.

Fig. 2. Design and synthesis of surface oxygen-deficient ruthenium oxide.

Design scheme of od-RuO₂, including a perovskite SrCo_1-xRu_xO_3-δ, b hydrothermal process, and c RuO₂ nanoclusters with oxygen-deficient Ru sites on the (101) surface. d HR-TEM image of the od-RuO₂ catalyst as nanoclusters. e HAADF-STEM image of the od-RuO₂ catalyst with a predominant (101) lattice plane. The upper left image is the partial enlargement of the right STEM image. The lower left image is the FFT diffraction pattern of this region. f EELS mapping images of the od-RuO₂ catalyst. The majority of RuO₂ covered over the residual cobalt oxide.

To determine if the developed design strategy could provide an od-RuO₂ nanocatalyst with high stability and high activity for the OER, the nanocatalyst was synthesized, and its structural properties were examined by morphological characteristics. X-ray diffraction (XRD) confirms the precursor with a crystalline phase corresponding to the perovskite SrCoO_2.52 (Supplementary Fig. 7). Whereas, no distinct RuO₂ diffraction peaks were observed in the XRD patterns of od-RuO₂ (only the baseline could be detected, shown as shown in Supplementary Fig. 8), indicating that this special RuO₂ consists of ultrasmall nanoclusters. Previous studies have confirmed that particles having crystalline domain sizes below 5 nm become difficult to analyze, due to both broad peaks and low signal-to-noise ratios³⁵. Next, to further determine the composition of the catalyst, energy-dispersive X-ray spectroscopy (EDS, Supplementary Fig. 9) and electron energy loss spectroscopy (EELS, Fig. 2f) were performed. Although both Ru and Co signals were present, the cobalt oxide only served as original sites for the defect-inheritance of Ru, and most of cobalt also dissolved during the acidic hydrothermal process, as only some residual signals remained that were not completely washed away. The EELS analysis of randomly selected regions revealed that Ru predominantly covered the cobalt oxide, thereby suggesting that the residual cobalt oxide just serves as a constraint or a support for od-RuO₂ recombination and growth, rather than serving as the exposed active sites. The O signal is reduced in regions with the highest Ru intensity, indicating the possible presence of substantial surface oxygen vacancies.

To ensure high exposure of a specific lattice plane, the particle size of od-RuO₂ should be in the nanoscale range. High-resolution transmission electron microscopy (HR-TEM) images confirmed the morphology of od-RuO₂ as dispersed nanoparticles with uniform sizes, averaging 2 – 4 nm in diameter (Fig. 2d and Supplementary Fig. 10). High-angle annular dark-field scanning TEM (HAADF-STEM) images (Fig. 2e) revealed Ru as brighter atomic columns, corroborating the presence of nanoscale RuO₂ clusters, while only negligible Co₃O₄ was observed. The statistical result of fast Fourier transform (FFT) diffraction pattern analysis of identifiable regions across all STEM images (Supplementary Figs. 11 and 12) shows that the majority of these nanoparticles displayed well-defined lattice fringes corresponding to the (101) plane of rutile RuO₂, with a lattice spacing of 0.25–0.26 nm. Notably, this orientation significantly dominated over the common (110) plane typically observed in commercial RuO₂. As designed by theoretical calculations, this preferential exposure of the (101) facet is significant because the (101) surface possesses distinct orbital hybridization characteristics and higher oxygen vacancy formation energy that stabilize surface-deficient configurations and favor the AEM over the LOM. Therefore, given the ultrasmall nanocluster size (possessing a high surface-to-volume ratio) and the statistically and theoretically predominant exposure of the (101) facet, a substantial fraction of oxygen-deficient Ru sites is expected to be enriched at or near the (101) surface. These results suggest that the hydrothermal process not only confines RuO₂ growth to nanoclusters, avoiding overgrowth into larger crystallites, but also favors specific lattice plane exposure, potentially endowing the material with observed catalytic properties. These morphological results preliminarily indicate that a targeted nanocatalyst with the desired surface oxygen-deficient configuration has been synthesized.

Characterization of od-RuO₂

To experimentally identify the oxygen-deficient configuration of od-RuO₂ in depth, its microstructure and electronic properties were further investigated. Electron paramagnetic resonance (EPR) was utilized for preliminary detection of the overall oxygen vacancy state of the catalyst. EPR signals around g ≈ 2 are commonly associated with unpaired electrons trapped at oxygen-vacancy-related defect sites in metal oxides^36–38. More notable peaks were observed for both the perovskite precursor and od-RuO₂ than for commercial RuO₂ (Fig. 3a), indicating that significant oxygen vacancies were introduced into od-RuO₂ from the perovskite, consistent with the design strategy.

Fig. 3. Characterization of od-RuO₂.

a EPR spectra of od-RuO₂, the perovskite precursor, and commercial RuO₂. The typical peaks for oxygen vacancies are at a g-factor of ~2. b Ru K-edge XANES spectra of od-RuO₂, Ru foil, commercial RuO₂, and commercial RuCl₃. The inset image is the partial enlargement at half-peak intensity region, indicating the valence states. c Ru K-edge FT-EXAFS spectra of od-RuO₂ and commercial RuO₂, where R denotes the radial distance. No phase shift correction was applied. The first shell feature is assigned to the Ru–O coordination shell on the basis of the subsequent EXAFS fitting. The higher shell features occur at apparent radial positions comparable to those observed for typical RuO₂^39–41. The Ru K-edge WT-EXAFS of d od-RuO₂, e commercial RuO₂, and f Ru foil, where k represents the wavenumber of the emitted photoelectron. g FT-EXAFS fitting of od-RuO₂, with a fitted coordination number of ~5.5. The window shows the fitting range of the first shell. h O K-edge XANES spectra of od-RuO₂ and commercial RuO₂. i Ru L₂-edge XANES spectra of od-RuO₂ and commercial RuO₂.

X-ray absorption fine structure (XAFS) analysis was further employed to elucidate the local chemical environment and oxidation state of Ru in od-RuO₂. The Ru K-edge X-ray absorption near-edge structure (XANES) spectra (Fig. 3b) were compared with those of references, including Ru foil, commercial RuCl₃, and commercial RuO₂, which correspond to typical oxidation states. The white-line intensity confirmed that the predominant substance was RuO₂, whose spectrum was clearly distinct from that of RuCl₃. The half-peak K-edge energy revealed that the oxidation state of Ru in od-RuO₂ lay between those in commercial RuO₂ and commercial RuCl₃, indicating an intermediate valence between +3 and +4. The first derivative of the Ru K-edge XANES spectra, the Ru L₃-edge X-ray absorption spectra, and also the X-ray photoelectron spectroscopy (XPS) further corroborated this electronic structure (Supplementary Figs. 13, 14, and 15). Notably, as shown in the L₂-edge spectra (Fig. 3i), the t_2g peak, which was prominent in commercial RuO₂, was absent in od-RuO₂, indicating substantial filling of the t_2g orbitals and thereby a lower Ru valence state in od-RuO₂ than in commercial RuO₂, corroborating the K-edge XAS results. This reduced oxidation state of Ru is significant, as it serves as supplementary evidence for the abundant pre-existing oxygen vacancies and an upshift in the Ru 4d band. This reduced Ru oxidation state in od-RuO₂ indicates a markedly higher t_2g^* occupation near the Fermi level, thereby facilitating the targeted stronger Ru–O π interactions and contributing to it following the AEM pathway.

To gain insights into the coordination environment and verify the presence of unsaturated Ru sites implied by the morphological analysis, the Fourier transform extended X-ray absorption fine-structure (FT-EXAFS) spectrum (Fig. 3c) of od-RuO₂ in R space (the radial distance from a central absorbing atom to its neighboring atoms) exhibited a prominent peak at approximately 1.50 Å, corresponding to the Ru–O coordination shell. Additional peaks at ~2.70 Å, ~3.2 Å, and ~4.1 Å were attributed to different Ru–Ru coordination shells^39–41. These Ru–Ru peaks had reduced intensities compared with the commercial RuO₂, consistent with short-range ordering in nanoclusters and limited long-range crystallinity. The wavelet transform of EXAFS (WT-EXAFS) analysis more intuitively substantiates this conclusion (Fig. 3d–f). In contrast to the commercial RuO₂, od-RuO₂ exhibited substantially weaker Ru-Ru scattering signals in Ru K-edge WT-EXAFS, indicating the formation of these nanoscale clusters rather than extended crystalline domains. The resulting high degree of dispersion leads to increased exposed surface area. Therefore, the observed unsaturated coordination Ru atoms could have a greater tendency to emerge on the surface. In addition, for more in-depth illustrating the bonding state, the WT-EXAFS plots of both Ru K-edge and Co K-edge for od-RuO₂, standard RuO₂, and standard Co₃O₄ are demonstrated (Supplementary Fig. 16). As reported previously⁴², the bonding of Ru and Co will induce a significant and characteristic shift of local maximum to a lower k region for Ru K-edge plots or a higher k region for Co K-edge plots relative to the corresponding Ru–Ru and Co–Co scattering features. Notably, the local maximum of od-RuO₂ at the Ru K-edge (even though the Ru-Ru bonding is weak) remains consistent with that of standard RuO₂. Accordingly, no local maximum of the residual Co species in od-RuO₂ emerges at the higher k region than the standard Co-Co region in the Co K-edge spectra. These results collectively indicate that no Ru-Co bonding is observed in od-RuO₂.

Quantitative analysis through FT-EXAFS fitting (Fig. 3g) further confirmed unsaturated coordination of the Ru atoms in od-RuO₂. The coordination number for Ru in od-RuO₂ was determined to be approximately 5.5, with an average Ru–O apparent radial position in the first shell of 1.5 Å. This coordination number was notably lower than the ideal coordination number of 6²⁷ (Supplementary Fig. 17), providing further evidence for the effectiveness of the defect-inheritance strategy, whereby the oxygen-deficient environment of ruthenium and cobalt in the Ru-doped perovskite is successfully transferred to the reconstructed RuO₂ (Supplementary Fig. 18). As the reduced coordination environments in Ru-based catalysts are associated to enhanced adsorption dynamics⁴³, these surface oxygen-deficient Ru sites serve as nucleophilic centers for OH⁻ adsorption, which plays a crucial role in increasing the OER activity. However, unlike other reports that rely on dopant elements, the od-RuO₂ accesses both high reactivity and improved stability by intrinsic structure design with specified tuned local coordination on target surfaces—not merely creating more vacancies.

Moreover, the O K-edge spectra (Fig. 3h) were investigated to confirm the presence of oxygen vacancies and the preference for the AEM pathway. Compared with commercial RuO₂, the od-RuO₂ spectrum exhibited a clear inversion in the relative intensities of the t_2g (530.7 eV) and e_g (533.7 eV) peaks, with the t_2g peak becoming weaker than the e_g peak, in contrast to the trend observed in the commercial RuO₂. This inversion indicates the increased electron occupation of the od-RuO₂ t_2g band (consistent with the Ru L₂-edge results, Fig. 3i), supporting a reduction in the Ru valence. In addition, the intensities of both the t_2g and e_g peaks were considerably decreased in the od-RuO₂ spectrum, suggesting a markedly weakened Ru–O covalency and providing additional evidence for AEM pathway promotion. Furthermore, the narrowed energy gap between the t_2g and e_g peaks also implied an upshift of the t_2g band, which would facilitate more rapid electron transfer.

These findings provide compelling evidence that the hydrothermal reconstruction process effectively transforms the perovskite precursor into surface-distributed RuO₂ nanoclusters. In this synthesis strategy, the reconstruction process spatially confines the growth of od-RuO₂ from the cobalt spinel oxide, thereby preventing overgrowth into larger crystalline domains. The resulting od-RuO₂ (compared with commercial RuO₂) exhibits a relatively low Ru valence state, a weakened Ru–O covalency, and surface unsaturated coordination of Ru. Collectively, these features endow od-RuO₂ with an optimized electronic structure favorable for the AEM-dominated OER pathway, bridging the long-standing trade-off between activity and stability in acidic environments. This design concept not only advances the understanding of structure–activity relationships in Ru-based catalysts but also offers a generalizable strategy for developing next-generation acid-stable electrocatalysts for scalable hydrogen production and related electrochemical energy technologies.

Electrocatalytic performance

Having determined the structural and electronic properties of od-RuO₂, its electrocatalytic performance for the OER was next evaluated to verify the capability advantages imparted by the designed surface structure. The OER performance of the catalysts was systematically assessed in a 0.5 M H₂SO₄ electrolyte at 25°C using linear sweep voltammetry (LSV), with commercial RuO₂ and pure Co₃O₄ (to exclude any contribution from the activity of Co₃O₄) serving as references. An iR compensation of 90% was applied to all polarization curves unless stated otherwise (the non-iR corrected results are presented in Supplementary Fig. 19). The optimized sample of od-RuO₂ catalyst exhibited an overpotential to reach 10 mA cm^-2 (η₁₀) of 161 mV, significantly outperforming commercial RuO₂ (η₁₀ = 352 mV) and Co₃O₄ (η₁₀ = 543 mV, with limited OER activity when Co acted as the active site) (Fig. 4a). Correspondingly, the Tafel slope (a kinetic parameter describing the required amount of overpotential increase for every tenfold rise in current density) of od-RuO₂ was determined to be 62.4 mV dec^-1 (Fig. 4b), indicating more favorable OER kinetics compared with those of commercial RuO₂ (97.4 mV dec^-1) and Co₃O₄ (231.5 mV dec^-1). The electrochemical surface area (ECSA), which reflects truly active catalyst surface area, was evaluated to further understand the intrinsic activity of the catalysts. According to the non-faradaic method, ECSA is proportional to the electrochemical double-layer capacitance (C_dl), which can be obtained from cyclic voltammetry (CV) in a non-faradaic potential region. The od-RuO₂ exhibited a C_dl value of 89.22 mF, markedly higher than those of commercial RuO₂ (27.80 mF) and Co₃O₄ (3.45 mF) (Fig. 4c). The LSV curves normalized to the ECSA (Fig. 4d) indicated that od-RuO₂ demonstrated not only a substantially larger number of electrochemically active sites but also higher intrinsic activity per unit area, which indicates that od-RuO₂ possesses improved intrinsic catalytic activity, not merely a larger surface area, thus demonstrating the effectiveness of the rational surface engineering strategy. In addition, to further clarify that the role of Co₃O₄ in this work is structural rather than catalytic, the OER performance of parallel samples with varying rinsing cycles was evaluated (Supplementary Fig. 20). The samples with lower Co content after different rinsing times not only do not show performance loss, but even show a certain degree of improvement. Overall, the samples with varying rinsing times exhibited essentially consistent high performance, as Ru species themselves are the real active sites.

Fig. 4. OER performance of od-RuO₂.

a OER polarization curves of od-RuO₂, with commercial RuO₂ and standard Co₃O₄ as reference samples. b Tafel slope curves of od-RuO₂, with commercial RuO₂ and standard Co₃O₄ as reference samples. c C_dl plots and d ECSA-normalized polarization curves of od-RuO₂, commercial RuO₂, and standard Co₃O₄. e Chronopotentiometric (CP) response of od-RuO₂ and commercial RuO₂ at 10 mA cm⁻². The inset image is the partial enlargement of the initial stage of stability performance. f AST polarization curves and overpotential at 10 mA cm^-2 of od-RuO₂ after 0 to 65,000 cycles. Unless otherwise specified, all referred tests were systematically conducted in a 0.5 M H₂SO₄ electrolyte at 25 °C. An iR compensation of 90% was applied to all polarization curves unless stated otherwise.

To optimize the synthesis process for achieving the highest catalytic activity, a series of od-RuO₂ samples were prepared with the same method by varying the Ru:Co molar ratio and the hydrothermal pH, temperature, and reaction time. Among these samples, the sample derived from the perovskite with an initial composition of SrCo_0.8Ru_0.2O₃ (20% Ru) exhibited the highest catalytic activity (Supplementary Fig. 21). Consequently, a Ru:Co molar ratio of 20% was identified as optimal and selected as the standard for further characterization. The hydrothermal pH, which is a critical parameter modulating the extent of the reconstruction process, was also investigated. The pH was systematically controlled by varying the amount of acid added, with the tested pH values ranging from 0.1 to 1.6. Both excessive and insufficient acid negatively affected the performance (Supplementary Fig. 22). Insufficient acid likely impeded extraction of Co and Ru from the perovskite, resulting in incomplete recombination. Conversely, excessive acid could have caused overdissolution of Co, leading to insufficient confinement of Ru growth and a subsequent reduction in nanocluster formation. Additionally, highly acidic conditions with strong oxidizing HNO₃ were considered to be associated with overoxidation and loss of Ru (as Ru oxidation state over +4, e.g., RuO₄, will be soluble^19,44), thereby impairing nanocluster formation and further decreasing the catalytic activity. Other synthesis parameters, including the hydrothermal temperature and reaction time, were similarly optimized (Supplementary Fig. 23). While a too low temperature or a too short reaction time resulted in incomplete transformation of the perovskite precursor to the final catalyst and thus reduced activity, extending the reaction time or increasing the temperature to beyond a certain threshold did not result in further performance enhancements. Under optimal conditions (150 °C for 12 h), the reconstruction process was complete, and the resulting od-RuO₂ samples displayed consistent activity (used for all aforementioned characterization). In addition, unless otherwise specified, a standard catalyst loading of 5 mg cm^-2 (with Ru atom content of ~50%, Supplementary Fig. 5) was appropriately set for all electrochemical measurements (Supplementary Fig. 24).

Durability tests in acidic media, which are a key benchmark for practical applications, were performed to determine the long-term stability of od-RuO₂. The chronopotentiometric (CP) response, an electrochemical test that measures the potential change of a catalyst under a constant applied current density, indicates that the catalyst maintains its activity over time, thus serving as a direct measure of durability. Under a constant OER current density of 10 mA cm^-2 in a standard three-electrode water electrolysis cell, od-RuO₂ maintained a nearly constant overpotential during more than 3000 h of continuous operation, with an increase of only 4 mV observed during the first 500 h (Fig. 4e). In contrast, commercial RuO₂ underwent rapid performance decay ( + 100 mV in 10 h) under the same conditions, underscoring the durability of od-RuO₂. Additionally, ICP–MS was used to quantify dissolved Ru species in the filtered electrolyte after the stability tests. The measured concentrations were negligible compared with the theoretical value expected for complete catalyst dissolution ( ~ 20 ppm) as shown in Supplementary Fig. 25, indicating minimal Ru leaching during operation. To further assess structural stability during prolonged operation, post-OER HR-TEM images of the catalyst after long-term acidic OER operation were also statistically analysed (Supplementary Fig. 26 and 27). The catalyst largely retains its nanocluster morphology after extended electrolysis, without obvious degradation or agglomeration. Importantly, the post-OER images also confirmed that the catalyst continues to exhibit a predominantly RuO₂ (101)-oriented structure, consistent with the designed facet-selective structure of original od-RuO₂. The prolonged acidic OER stability of od-RuO₂ was also verified by accelerated stress tests (AST) via fast-scan CV, which revealed only a 5 mV overpotential increase in 0.5 M H₂SO₄ after 200,000 cycles over a voltage range of 1.2 to 1.4 V vs. reversible hydrogen electrode (RHE) (Supplementary Fig. 28), and only an 18 mV increase after 65,000 cycles over a wider potential window of 1.2 to 1.7 V vs. RHE (Fig. 4f and Supplementary Fig. 29). In addition, supplementary electrochemical measurements were also performed on a glassy carbon (GC) electrode to further validate the catalytic performance of od-RuO₂ (Supplementary Fig. 30). As presented in Supplementary Table 1, od-RuO₂ demonstrates competitive catalytic activity and stability compared with state-of-the-art Ru-based catalysts for the acidic OER.

Insights into the OER process

To elucidate the OER mechanism of the od-RuO₂ catalyst, differential electrochemical mass spectrometry (DEMS) with isotope labeling was employed to identify the source of oxygen evolved during the OER process. The theoretical OER pathways and corresponding expected signals are illustrated in Fig. 5a, b. The typical oxygen products expected for both the AEM and LOM are summarized in Supplementary Table 2.

Fig. 5. Tracing of the OER catalytic pathway for mechanism analysis.

Schematic illustration of the DEMS process on the RuO₂ (101) plane and typical products following the a LOM pathway and b AEM pathway. c Operando DEMS signals of O₂ products for od-RuO₂ in the electrolyte with H₂¹⁸O. d Operando DEMS signals of O₂ products for od-RuO₂ in the electrolyte with H₂¹⁶O. e Partial enlargement of the ³⁶O₂ and ³⁴O₂ signals in d. Gibbs free-energy diagrams of od-RuO₂ for the OER under an applied potential of 0 V following f the AEM pathway, g the LOM pathway with O_lat–O_lat coupling, and h the LOM pathway with O_lat–O_ads coupling. The symbols and corresponding inset figures of the successive reaction steps in f: *+H₂O, the initial state comprising a vacant active site and an interfacial H₂O molecule; *OH, the adsorption of OH^-; *O, the surface oxygen intermediate formed by deprotonation; *OOH, the O-O coupling by a second adsorption of OH^-; *+O₂, the regeneration of the vacant active site following further deprotonation and O₂ evolution. The symbols and corresponding inset figures of the successive reaction steps in g: O*O, the initial state with two adjacent lattice oxygen atoms; (${\mathrm{O}}_{-}^{*}\mathrm{O}$), the O-O coupling between two lattice oxygen atoms; *+O₂, the generation of the vacant active site following O₂ evolution; *OH, the adsorption of OH^-; HO*OH, a second adsorption of OH^-; O*OH, the deprotonation; O*O, regeneration of the initial state following a second deprotonation. The symbols and corresponding inset figures of the successive reaction steps in h: O*O, the initial state with one lattice oxygen atom and one adsorbed oxygen atom; (${\mathrm{O}}_{-}^{*}\mathrm{O}$), the O-O coupling between the lattice oxygen atom and the adsorbed oxygen atom; *+O₂, the generation of the vacant active site following O₂ evolution; *OH, the adsorption of OH^-; HO*OH, a second adsorption of OH^-; O*OH, the deprotonation; O*O, regeneration of the initial state following a second deprotonation.

Preliminary ex-situ DEMS results revealed that od-RuO₂ likely followed the AEM pathway, as evidenced by the significantly lower ³⁴O₂ signal than that for commercial RuO₂, as shown in Supplementary Fig. 31. However, ex-situ DEMS still requires supplementation from operando DEMS to definitively confirm the AEM pathway. The key evidence for distinguishing the AEM pathway from the LOM pathway lies in detecting ³⁶O₂ formation.

To further substantiate the AEM pathway, operando DEMS was performed using the setup illustrated in Supplementary Fig. 32a. In this integrated system, gaseous products generated at the electrode surface are instantaneously extracted by the vacuum chamber and directly introduced into the mass spectrometer, enabling real-time detection with high sensitivity and accuracy. The experiments were conducted in two stages (Supplementary Fig. 32b), in which H₂¹⁸O and H₂¹⁶O were utilized as the electrolyte in 0.5 M H₂SO₄. The catalyst was deposited on a porous gas-permeable Au working electrode and subjected to CV between 1.165 V and 2.015 V vs. RHE.

In the first stage, the catalyst underwent 12 CV cycles in the H₂¹⁸O electrolyte. For the AEM pathway, at least one ¹⁸O atom derived from the electrolyte is involved in the O₂ products. As shown in Fig. 5c, mass spectrometry revealed a minor amount of ³⁴O₂ (¹⁶O + ¹⁸O) initially, arising from the original residual surface ¹⁶O coupling with ¹⁸OH. The predominant product was ³⁶O₂ (¹⁸O + ¹⁸O), originating entirely from adsorbed ¹⁸O. Notably, no ³²O₂ (¹⁶O + ¹⁶O) was detected, confirming the minimal involvement of lattice oxygen and supporting the AEM pathway. The catalyst subsequently underwent several additional CV cycles to fully label the surface with ¹⁸O.

In the second stage, the catalyst was washed with deionized (DI) water, dried at 25 °C, and then subjected to CV cycling in the H₂¹⁶O electrolyte. If the AEM mechanism is active, only the remaining residual surface adsorbates (containing ¹⁸O) would combine with ¹⁶OH from the electrolyte, yielding ³⁴O₂ as the product initially. With continued cycling, the predominant product should be ³²O₂ (¹⁶O + ¹⁶O), originating entirely from adsorbed ¹⁶O, while no ³⁶O₂ should be observed. The experimental data confirmed this prediction: ³⁴O₂ signals were detected because of partial ¹⁸O exchange, whereas the ³⁶O₂ signal, which would indicate lattice oxygen coupling, was absent (Fig. 5d, e). The combined evidence from operando and ex-situ DEMS measurements unequivocally demonstrates that the OER proceeds predominantly via the AEM mechanism in od-RuO₂.

To theoretically investigate the OER mechanism and origins of od-RuO₂, the Gibbs free-energy for the AEM and LOM were calculated. The calculations consider adsorption on surface models under vacuum within the computational hydrogen electrode (CHE) framework developed by Nørskov and co-workers⁴⁵. Although this approach does not fully reproduce the realistic electrochemical environment at the catalyst surface under an applied potential, it can still provide a useful basis for comparing the relative energetic trends of the related reaction pathways. The Gibbs free-energy profiles can directly indicate the lowest-energy OER pathway. On the basis of the aforementioned design strategy, a RuO₂ model was constructed with oxygen vacancies on the predominant RuO₂ (101) surface. The effect of surface unsaturated coordination Ru atoms on the OER kinetics was investigated. The Gibbs free energies for both the AEM and LOM pathways were calculated to identify the more energetically preferred OER mechanism. For the AEM pathway, the rate-determining step (RDS) was deprotonation of *OH to form *O, with a calculated free-energy change of 1.76 eV, as shown in Fig. 5f. For the LOM pathway, as at least one lattice oxygen is involved in oxygen coupling, two conditions were considered: 1) O–O coupling between two lattice oxygen atoms (O_lat) and 2) O–O coupling between a lattice oxygen atom and an adsorbed oxygen atom (O_ads). Under the two LOM pathway conditions, the RDS was the formation of O₂ from O*O, with substantially higher free-energy changes of 6.07 eV and 2.23 eV, respectively, as shown in Fig. 5g, h. These results confirmed that the AEM pathway is energetically more favorable than the LOM pathway for the oxygen-deficient (101) surface. Additionally, these calculations confirmed that surface oxygen vacancies facilitate adsorption of oxygenated intermediates, effectively reducing the energy barrier for the OER. These mechanistic insights explain the increased OER stability and activity of the designed (101) plane.

In conclusion, the DEMS and DFT results collectively and mechanistically demonstrate that this od-RuO₂ catalyst, driven by the design strategy of oxygen-deficient Ru sites, predominantly follows the AEM pathway. This mechanistic insight not only elucidates the origin of the catalyst’s activity and durability but also validates the effectiveness of the oxygen-vacancy–engineered surface design strategy.

PEMWE performance

To assess the practicality of od-RuO₂ for water electrolysis under industrial conditions, a PEMWE device was assembled by employing od-RuO₂ and commercial RuO₂ successively as the anode catalysts, commercial Pt/C as the cathode catalyst, Nafion 115 as the proton-exchange membrane (PEM), porous transport layer (PTL), and gas diffusion layer (GDL), as shown in Fig. 6a. The synthesis of the nanoparticulate od-RuO₂ via a scalable modified sol–gel method that is amenable to mass production. To evaluate the electrochemical performance of this PEMWE, the current–voltage (I–V) polarization curves were recorded to determine cell efficiency under different operating current densities, while stability tests were performed to assess long-term durability under constant current operation. The I–V polarization curves demonstrated that the PEMWE device equipped with od-RuO₂ achieved a current density of 1.0 A cm^–2 at a voltage of only 1.5126 V, compared with 2.1483 V for the device using commercial RuO₂ (Fig. 6b), indicating its efficiency. Notably, this od-RuO₂ employed PEMWE device even exhibited a voltage of 1.8535 V for a high current density of 3.0 A cm^-2, which approached the 2026 PEMWE technical targets (Supplementary Fig. 33) proposed by the U.S. Department of Energy (DOE). Furthermore, continuous operation at 1.0 A cm^-2 and 80 °C (conditions representative of typical industrial proton exchange membrane electrolyser operating environments) proceeded for more than 1200 h without observable performance degradation (Fig. 6c). In addition, after a temporary interruption caused by equipment relocation, the original membrane electrode assembly (MEA) was reactivated and subjected to extended durability testing, maintaining stable operation for more than 3000 h (Supplementary Fig. 34). In contrast, the commercial RuO₂-based PEMWE exhibited rapid voltage escalation and deactivation under the same conditions, emphasizing the competitive stability and practical applicability of od-RuO₂ for long-term acidic water electrolysis and hydrogen production. As presented in Supplementary Table 3, od-RuO₂ demonstrates competitive PEMWE performance compared with state-of-the-art Ru-based catalysts.

Fig. 6. PEMWE testing of od-RuO₂.

a Illustration of a single-cell PEMWE device with commercial PEM, PTL, and GDL. b I–V polarization curves with an active area of 25 cm² under 80 °C. c Voltage‒time curves of the PEMWE device with od-RuO₂ and commercial RuO₂ at 1.0 A cm^–2 and 80 °C. The inset image is the partial enlargement of the initial stage of PEMWE durability performance. d SEM image of the MEA cross-section before PEMWE testing. e, f EDS mapping of the MEA cross-section before PEMWE testing. g SEM image of the MEA cross-section after PEMWE testing. h, i EDS mapping of the MEA cross-section after PEMWE testing.

To further evaluate the structural robustness and degradation behavior of the developed catalyst under realistic operation, post-operation analyses were conducted on the MEA following prolonged PEMWE testing. The cross-sectional scanning electron microscopy (SEM) and EDS analyses of the MEA before and after operation (Fig. 6d–i) confirmed a stable ruthenium distribution at the anode. Although the catalyst layer becomes less morphologically uniform compared with its pristine state, a substantial amount of Ru remains firmly anchored on the membrane surface, indicating minimal catalyst degradation or loss during PEMWE operation. Although further industrial-scale evaluations are ongoing to assess the harsh condition long-term stability, Ru utilization efficiency, and performance under dynamic load conditions, these findings position od-RuO₂ as a promising anode catalyst for efficient and durable PEMWE. Future efforts could extend this rational reconstruction approach to other noble or transition metal based systems, offering potential routes to design cost-effective and durable catalysts for large-scale hydrogen production.

Discussion

In conclusion, we report a structure-intrinsic designed od-RuO₂ catalyst featuring nanoclusters with oxygen-deficient Ru sites on the (101) surface. These RuO₂ nanoclusters were obtained via confined growth from a cobalt spinel oxide matrix through a scalable sol-gel method followed by hydrothermal reconstruction. Comprehensive characterizations, supported by DFT calculations, confirmed that the OER on od-RuO₂ proceeds via the AEM pathway. The oxygen-deficient Ru atoms on the (101) surface can enhance Ru t_2g and O π hybridization by broadening the Ru 4d t_2g band occupation and decreasing the electron density of O 2p orbitals near the Fermi level, thereby serving as preferential sites for OH^- adsorption, further facilitating oxygen evolution. This alternative reaction pathway enables od-RuO₂ to avoid the extensive lattice oxygen participation associated with the LOM, thereby mitigating catalyst degradation and increasing the stability of the catalyst. The practical PEMWE incorporating od-RuO₂ delivered competitive activity and steady performance under industrial conditions. These results support the potential of this intrinsic structure engineering strategy for constructing AEM-dominated RuO₂ catalysts, supporting further development of efficient and durable electrocatalysts for acidic water splitting and sustainable hydrogen production. Future work could focus on extending this catalyst design strategy to other noble-metal or transition-metal oxide systems, exploring large-area electrode fabrication, and device-level optimization, to further bridge the gap between laboratory-scale discovery and industrial-scale implementation.

Methods

Chemicals

All chemicals were used without further purification. Ruthenium(III) chloride (RuCl₃, 45-55 wt.% Ru), ruthenium (IV) oxide (RuO₂, 99.9%), strontium nitrate (Sr(NO₃)₂, 99%), cobalt (II) nitrate hexahydrate (Co(NO₃)₂·6H₂O, 98%), commercial platinum on carbon (Pt/C, 50 wt% Pt), ethylenediaminetetraacetic acid (EDTA, 99.4–100.6%), citric acid (CA, 99.5 %), nafion perfluorinated resin solution (5 wt.%), nitric acid (HNO₃, 70%), sulfuric acid (H₂SO₄, 95.0-98.0%), ethanol (CH₃CH₂OH, 99.5%), Water-¹⁸O (H₂¹⁸O, 99 atom% ¹⁸O) were obtained from Sigma‒Aldrich. Deionized (DI) water was used as solvent.

Synthesis

Synthesis of perovskite precursor

The perovskite precursor (SrCo_1-xRu_xO_3−δ) was synthesized using a sol-gel method. Initially, Sr(NO₃)₂, Co(NO₃)₂·6H₂O, and RuCl₃ were dissolved in deionized (DI) water following the desired stoichiometric ratio. Ethylenediaminetetraacetic acid (EDTA) and citric acid (CA) were then added as chelating agents under magnetic stirring, maintaining a molar ratio of 1:1:1.5 for total metal-ions: EDTA:CA. The pH of the solution was adjusted to 8–10 using aqueous ammonia solution, and the mixture was heated at 90 °C until a transparent sol formed. This sol was dried at 200 °C for 24 h. The dried product was ground into a fine powder, transferred to an alumina crucible, and calcined at 800 °C for 6 h to yield the SrCo_1-xRu_xO_3−δ perovskite precursor.

Preparation of od-RuO₂

The od-RuO₂ samples were prepared using a hydrothermal method. A total of 200 mg of SrCo_1-xRu_xO_3−δ was added to a mixture containing varying amounts of HNO₃ and 60 mL of DI water, followed by ultrasonic treatment for 20 min. The resulting solution was then transferred to a Teflon-lined autoclave and heated at 150 °C for 12 h. After cooling, the mixture was centrifuged and washed several times with DI water. The final product, od-RuO₂, was collected and ground into a powder after drying.

Theory calculations

All spin-polarized density functional theory (DFT) calculations were performed using the Vienna Ab initio Simulation Package (VASP). The exchange–correlation energy was described using the Perdew–Burke–Ernzerhof (PBE) functional⁴⁶, and the interactions between valence electrons and ionic cores were treated using Projector augmented-wave (PAW) pseudopotentials⁴⁷. Strong on-site electronic correlations were accounted for using the density functional theory plus Hubbard U approach (DFT + U)⁴⁸, with an effective parameter of U-J = 4.88 eV for Ru. Long-range dispersion interactions were included using the D3 correction with Becke–Johnson damping (DFT-D3(BJ) dispersion correction⁴⁹) in all calculations. The plane wave kinetic energy cutoff was set to 500 eV. Electronic self-consistency and structural relaxation were considered converged when the total-energy difference was below 10⁻⁵ eV and the residual force on each atom was below 0.03 eV Å⁻¹, respectively. Densities of states (DOS) were processed by VASPKIT⁵⁰, and Bader charge analysis was conducted using the implementation developed by the Henkelman group⁵¹. All the atomic structures were visualized using VESTA⁵². The CONTCAR file of the optimized models is provided in Supplementary Data 1.

The surface oxygen vacancy formation energy (E_Ovac) refers to the energy required to remove an oxygen atom from a material’s lattice and then create a vacancy. The oxygen vacancy formation energy is defined as E_Ovac = (E_removed−E_slab + 2μ_O)/2, where E_removed represents the energy of the slab when two oxygen atoms are removed (one oxygen atom from each surface), E_slab is the energy of the slab without defects, and μ_O denotes the chemical potential of the oxygen atom that is removed, referenced to O₂ gas at 0 K.

Generally, the process of catalysts to accelerate chemical reactions is fundamentally a surface-driven process⁵³. Given the limited stability of commercial RuO₂, identifying an alternative lattice plane with the potential to enhance stability is of significant importance. Surface effects are particularly pronounced in nanomaterials, where a high surface-area-to-volume ratio results in properties that differ substantially from those of the bulk material^54–57. The stability of a surface is characterized by its surface energy (E_Surf), which quantifies the excess energy of surface atoms and is important in determining surface structure and reconstruction behavior⁵⁸. The surface energy E_surf is defined as E_surf = (E_slab − E_bulk)/2 A, where E_bulk represents the energy of the slab’s constituents in an ideal bulk configuration, and A denotes the area of one side of the slab.

Reaction free-energy profiles were evaluated using the computational hydrogen electrode (CHE) framework developed by Nørskov and co-workers⁴⁵. The Gibbs free energy change associated with each elementary reaction step was calculated using the equations given below:

[AEM]

$${\mathrm{\Delta }G}_1=\mathrm{G}\left(*\mathrm{OH}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(*\right)-\mathrm{G}\left({\mathrm{H}}_2\mathrm{O}\right)$$ 1

$${\mathrm{\Delta }\mathrm{G}}_2=\mathrm{G}\left(*\mathrm{O}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(*\mathrm{OH}\right)$$ 2

$${\mathrm{\Delta }\mathrm{G}}_3=\mathrm{G}\left(*\mathrm{OOH}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(*\mathrm{O}\right)-\mathrm{G}\left({\mathrm{H}}_2\mathrm{O}\right)$$ 3

$${\mathrm{\Delta }\mathrm{G}}_4=\mathrm{G}\left(*\right)+\mathrm{G}\left({\mathrm{O}}_2\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(*\mathrm{OOH}\right)$$ 4

[LOM]

$${\mathrm{\Delta }\mathrm{G}}_1=\mathrm{G}\left(*\right)+\mathrm{G}\left({\mathrm{O}}_2\right)-\mathrm{G}\left(\mathrm{O}*\mathrm{O}\right)$$ 5

$${\mathrm{\Delta }G}_2=\mathrm{G}\left(*\mathrm{OH}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(*\right)-\mathrm{G}\left({\mathrm{H}}_2\mathrm{O}\right)$$ 6

$${\mathrm{\Delta }\mathrm{G}}_3=\mathrm{G}\left(\mathrm{HO}*\mathrm{OH}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(*\mathrm{OH}\right)-\mathrm{G}\left({\mathrm{H}}_2\mathrm{O}\right)$$ 7

$${\mathrm{\Delta }\mathrm{G}}_4=\mathrm{G}\left(\mathrm{O}*\mathrm{OH}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(\mathrm{HO}*\mathrm{OH}\right)$$ 8

$${\mathrm{\Delta }\mathrm{G}}_5=\mathrm{G}\left(\mathrm{O}*\mathrm{O}\right)+1/2\mathrm{G}\left({\mathrm{H}}_2\right)-\mathrm{G}\left(\mathrm{O}*\mathrm{OH}\right)$$ 9

Materials characterizations

X-ray diffraction (XRD) patterns were recorded using a Bruker D8 diffractometer equipped with Cu Kα radiation (λ = 1.5406 Å). X-ray photoelectron spectroscopy (XPS) measurements were conducted with a Thermo Fisher Scientific ESCALAB 250Xi spectrometer using an Al Kα X-ray source (1486.6 eV). Transmission electron microscopy (TEM) and high-resolution TEM (HR-TEM) images were acquired using a JEOL 3200 fs TEM operating at an acceleration voltage of 300 kV. High-angle annular dark-field scanning TEM (HAADF-STEM), energy-dispersive X-ray spectroscopy (EDS) mapping, and electron energy loss spectroscopy (EELS) analyses were performed on a FEI Themis scanning transmission electron microscope, also operated at an acceleration voltage of 300 kV.

Ruthenium K-edge, Ruthenium L₃-edge, and Ruthenium L₂-edge X-ray absorption fine structure (XAFS) spectra were collected in transmission mode at Australia’s Nuclear Science and Technology Organisation (ANSTO). Energy calibration was performed using a standard ruthenium foil. Cobalt K-edge and Oxygen K-edge XAFS spectra were collected in transmission mode and fluorescence mode respectively at Singapore Synchrotron Light Source (SSLS). Energy calibration was performed using a standard cobalt foil. Wavelet-transform analysis of the EXAFS spectrum was performed in MATLAB using the Continuous Cauchy Wavelet Analysis (CCWT) code developed by Manuel Muñoz and François Farges^59,60. The k²-weighted χ(k) spectra were used as input, and the modulus of the resulting wavelet coefficients was plotted as a function of the photoelectron wavenumber (k) and radial coordinate (R). No phase shift correction was applied.

The ex-situ DEMS measurements were performed at the mass spectrometry lab (MS Lab) of the Chemical, Molecular and Materials Analysis Center (CMMAC) in the Department of Chemistry, National University of Singapore (NUS). The gas samples were collected using a gas-tight three-electrode system. For isotope labeling, 2 mL of 0.5 M H₂SO₄ was prepared using H₂¹⁸O as solvent. The catalyst was operated at 10 mA cm^-2 in ¹⁸O-labeled electrolyte for 18 h to achieve isotope incorporation. The electrodes and cell were then rinsed with ¹⁶O water for 30 min to remove residual H₂¹⁸O. Subsequently, the cell with a 0.5 M H₂SO₄ solution of H₂¹⁶O was bubbled with N₂ for 48 h with the flow rate of 5 mL min⁻¹ to eliminate dissolved oxygen. Finally, during OER at 10 mA cm^-2, the gaseous products were collected into 20 gas sample bags, which were delivered to MS Lab and analyzed by Agilent 5975 GCMS with Agilent 19091S-433 column.

The operando DEMS was carried out on the QAS 100 device (Linglu Instruments Co., Ltd., Shanghai) with a specialized three-electrode electrochemical system. A schematic of DEMS electrochemical cell is provided in Supplementary Fig. 32. The Ag/AgCl and Pt wire were employed as reference and counter electrodes, respectively. The working electrode consisted of a porous Au film supported on a porous polytetrafluoroethylene membrane. L-DP01 membrane from Shanghai Linglu was employed. The catalyst ink was drop-cast onto the Au film and air-dried. For isotope labeling, the electrode was cycled 12 times in ¹⁸O-labeled 0.5 M H₂SO₄ electrolyte with the potential range of 1.165–2.015 V vs. RHE at a scan rate of 5 mV s⁻¹ for isotope labeling, during which the ³²O₂, ³⁴O₂, and ³⁶O₂ signals were continuously monitored by mass spectrometer. The electrodes and cell were then rinsed with ¹⁶O water for 30 min to remove residual H₂¹⁸O, followed by scanning in a 0.5 M H₂SO₄ solution of H₂¹⁶O at 5 mV s⁻¹ from 1.165 V to 2.015 V vs. RHE. The gaseous products were recorded by mass spectrometry.

Electrochemical measurements

Electrochemical measurements were carried out using a VMP3 electrochemical workstation (BiO-Logic Inc.) in a three-electrode configuration at 25 °C. The workstation was equipped with a built-in electrochemical impedance spectroscopy (EIS) analyzer. The data acquisition was performed using EC-Lab software (V11.52).

The working electrode consisted of od-RuO₂ deposited on carbon cloth (CC). To prepare the working electrode, 5 mg of od-RuO₂ powder (with Ru atom content of ~50% as ICP results in Supplementary Fig. 5) was dispersed in a mixture of 400 μL DI water, 300 μL ethanol, and 30 μL of 5 wt.% Nafion solution. The suspension was sonicated for at least 60 min to form a uniform ink. The ink was then pipetted onto the carbon cloth of 1 cm² and air-dried at 25 °C for 12 h before testing. A similar process was applied to the preparation of glassy carbon (GC) electrodes, with the deposited catalyst amount adjusted to the geometric electrode area of GC electrode (0.0707 cm²).

A saturated calomel electrode (SCE) and a platinum electrode were used as the reference electrode and the counter electrode, respectively. Measurements were conducted in 0.5 M H₂SO₄ with a pH of 0.30 ± 0.04. All potentials were converted to the reversible hydrogen electrode (RHE) scale according to: E(RHE) = E(SCE) + 0.241 + 0.0591 × pH, where 0.241 V is the potential of the saturated calomel electrode relative to the standard hydrogen electrode at 25 °C. Unless otherwise stated, the electrochemical measurements in this work were performed once. Linear sweep voltammetry (LSV) was conducted at a scan rate of 0.1 mV s^–1. The accelerated stress tests (AST) were conducted via fast-scan CV from 1.2 to 1.4 V (vs. RHE) at a scan rate of 100 mV s⁻¹ for 200,000 cycles, and from 1.2 to 1.7 V (vs. RHE) at a scan rate of 250 mV s⁻¹ for 65,000 cycles.

Tafel plots were constructed by plotting the overpotential (η, mV) against the logarithm of current density (log [J]). The Tafel slope was derived by linear fitting of the linear region of the plot using the Tafel equation:

$$\eta =\mathrm{b}\log \left[J\right]+a$$ 10

where b is the Tafel slope and a is the intercept.

The uncompensated resistance (R) was determined by EIS for each catalyst–electrode configuration. The EIS measurements were conducted at open circuit potential, with the frequency range set from 10 mHz to 100 kHz. An iR compensation of 90% was applied to all polarization curves unless stated otherwise. The measured resistance values at CC were 1.62 Ω for od-RuO₂, 1.95 Ω for commercial RuO₂, and 2.22 Ω for Co₃O₄ (Supplementary Fig. 35). The resistance values at GC were 5.92 Ω for od-RuO₂ (Supplementary Fig. 36).

The electrochemically active surface area (ECSA) was estimated from the electrochemical double-layer capacitance (C_dl). Cyclic voltammograms were recorded in a non-Faradaic potential window of 0.81–0.91 V vs. RHE in 0.5 M H₂SO₄, with scan rates of 2, 4, 8, 10, 15, and 20 mV s⁻¹. The C_dl was determined from the linear dependence of the capacitive current on the scan rate.

The electrochemically active surface area was calculated as:

$$\mathrm{ECSA}=C_{dl}/C_S$$ 11

where C_s is the specific capacitance of a nominally flat surface. In this study, a value of 35 μF cm⁻² was used for C_s.

PEMWE tests

The as-synthesized od-RuO₂ or commercial RuO₂ was employed as the anode catalyst for PEMWE with a loading of ~3 mg cm⁻² (identified as the most appropriate loading for our PEMWE devices). Commercial Pt/C was used as the cathode catalyst with a loading of ~0.5 mg_Pt cm⁻². Nafion 115 served as the proton-exchange membrane (PEM). Before assembly, the membrane was treated in 0.5 M H₂SO₄ at 80 °C for 30 min, and sequentially rinsed with DI water. Pt-plated Ti fiber felt was used as the anode porous transport layer (PTL). Carbon paper was used as the cathode gas diffusion layer (GDL). The anode catalyst inks were prepared by ultrasonically dispersing the catalyst and 5 wt% Nafion at a catalyst-to-Nafion weight ratio of 1:0.2 in a solvent mixture of isopropanol, 1-propanol, and DI water at weight ratios of 1:1:3. The cathode inks were prepared similarly, with a catalyst-to-Nafion weight ratio of 1:0.35 in a solvent mixture of isopropanol, 1-propanol, and DI water at weight ratios of 1.4:0.95:1. The anode and cathode inks were then uniformly sprayed onto opposite sides of the membrane. The catalyst-loaded membrane was combined with PTL and GDL and then hot-pressed at 145 °C under 5 MPa for 5 min to fabricate the membrane–electrode assembly (MEA). The MEA was incorporated into a PEMWE equipped with Ti bipolar plates containing serpentine flow channels with a geometric area of 25 cm² (5 cm × 5 cm). DI water was circulated to the anode at a flow rate of 170 mL min⁻¹ using a magnetic pump. The steady-state polarization curves were recorded in galvanostatic mode from 0 to 4 A cm⁻² at 80 °C. The PEMWE durability measurements were conducted at a current density of 1.0 A cm⁻² at 80 °C. All data measured in PEMWEs were obtained without iR correction.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

H.W., X.P.W., and J.M.X. conceived the idea. H.W. completed the paper and prepared the revision. H.Y.Z., X.Z., X.P.W., and J.M.X. contributed to the review and editing. H.W. performed synthesis and electrochemical measurements of the samples. Y.F.M., H.A., W.C., X.Z., J.C.Y., and H.C.L. were responsible for the analysis of electrochemical results. Q.Z. and Z.G.Y. carried out DFT simulations. C.L., J.L.L., and J.G.C. performed PEM tests and analysis. J.M.X. oversaw the overall project and preparation of the manuscript.

Peer review

Peer review information

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

Data availability

All the data supporting the findings of this study are available in the paper and its supporting files. 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 https://doi.org/10.1038/s41467-026-76901-w.