Dynamic hydroxyl mediated charge buffering stabilizes high valence ruthenium edge sites for acidic water oxidation Authors: Tianqing Yan, Kang Xue, Shishi Zhang, Huayang Wang, Zhen-Feng Huang, Minhua Ai, Ruijie Gao, Chengxiang Shi, Wanliang Mi, Xiangwen Zhang, Lun Pan, Ji-Jun Zou Published: 2026 ABSTRACT High valence metal species are essential for driving electrocatalytic acidic water oxidation, but suffer from intrinsic thermodynamic instability. This activity-stability paradox is particularly severe for ruthenium-based catalysts, where highly active Ru(>IV) species are also precursors for rapid dissolution. Here we show that a cooperative host-guest architecture circumvents this trade-off through dynamic, hydroxyl-mediated charge buffering. By anchoring isolated ruthenium atoms at β-MnO2 edge sites, we demonstrate that host-derived surface hydroxyls govern both kinetics and robustness. These hydroxyls optimize intermediate binding to facilitate ideal O-O coupling, accelerating the reaction. Concurrently, the specific coordination environment allows the manganese host to reversibly accommodate excess oxidative charge, preventing irreversible structural degradation and ruthenium dissolution. The resulting Ru0.03Mn0.97O2 catalyst achieves a mass activity 223-fold higher than RuO2 and increases the stability number by three orders of magnitude. As an anode in a proton exchange membrane electrolyzer, it sustains an industrial-level current density of 1 A cm−2 for over 1000 h, comparing favorably to IrO2 while reducing precious metal usage by 80%. Introduction Proton-exchange membrane water electrolyzers (PEMWEs) offer a scalable pathway to a green hydrogen economy, but their application is impeded by the sluggish kinetics and corrosive nature of acidic oxygen evolution reaction (OER) at the anode1,2. The industry-standard iridium oxide catalysts are constrained by cost and scarcity3,4, which has intensified the search for ruthenium-based catalysts as Ru is both more abundant and intrinsically more active. Unfortunately, the high activity of Ru is coupled with structural instability5, because the high valence Ru(>IV) species essential for high turnover rates are also precursors to rapid dissolution under operating potentials6–9. The central challenge lies in stabilizing these hyperactive Ru sites without sacrificing their catalytic efficacy. Current approaches to undermine this activity-stability paradox have to accept a compromise. The first strategy relies on passive protection, such as alloying or deposition on inert supports10,11. While these static shields maintain high activity, they fail to counteract the dynamic accumulation of oxidative equivalents at the active site, ultimately leading to structural failure. The second approach sacrifices activity for stability by suppressing the formation of unstable intermediates, often by kinetically stabilizing a lower oxidation state of Ru12–15. However, a definitive solution requires an architecture that protects the Ru centers from dissolution while preserving the catalytically essential high valence state. Here, we circumvented this trade-off via a dynamic hydroxyl-mediated charge buffering, enabled by a cooperative host-guest architecture. By anchoring isolated Ru atoms at the low-coordination edge sites of acid-stable β-MnO2 host, we created robust Ru-O-Mn motifs where the specific electronic properties of the host fulfilled a dual role. In the statistical aspect, the lattice confinement physically immobilized the Ru centers and electronically pre-oxidized them to the desired high valence state. In the dynamic aspect, the host-derived surface hydroxyls optimized the binding energetics to facilitate hydrogen-bond-assisted O–O coupling, accelerating reaction kinetics. Simultaneously, these species imposed geometric constraints that enable the Mn host to reversibly accommodate oxidative equivalents, thereby preventing the Ru centers from irreversible structural reorganization and oxidative dissolution. This design transforms the host from a passive support into an active host in the catalytic cycle, culminating in a low-precious-metal catalyst that delivers both competitive activity and operational robustness under industrial conditions. Results Rational design of cooperative host-guest architecture To undermine the intrinsic paradox of hyperactive but unstable Ru(>IV) species, we conceptualized a cooperative host-guest architecture wherein isolated Ru atoms are embedded within an active and isostructural host lattice (Fig. 1a). Specifically, the host fuls two critical functions, including dynamically buffering Ru sites against oxidative dissolution and structurally inducing the formation of active high valence states. We employed a two-step density functional theory (DFT) screening to translate this conceptual framework into a material, first identifying the optimal host and subsequently defining the local Ru coordination environments.Fig. 1DFT-guided design of cooperative architecture.a Schematic illustrating the cooperative stabilization concept. b Surface Pourbaix diagrams for RuO2. c Surface Pourbaix diagrams for β-MnO2. Note that here focuses on the key surface states, while all calculated surface states are provided in the Supplementary Information. d Free energy diagrams of possible surface states for RuO2 and MnO2. e Pauling electronegativity of cations in candidate rutile hosts versus that of Ru(IV). f Calculated formation energies for Ru substitution at edge versus terrace sites within β-MnO2. g Formation energy of an edge Mn vacancy in β-MnO2 and that of a Mn vacancy adjacent to a substituted Ru atom. h Calculated dissolution energy of surface Ru atom on RuO2 (110) compared with an edge Ru-doped β-MnO2. DFT-guided design of cooperative architecture. a Schematic illustrating the cooperative stabilization concept. b Surface Pourbaix diagrams for RuO2. c Surface Pourbaix diagrams for β-MnO2. Note that here focuses on the key surface states, while all calculated surface states are provided in the Supplementary Information. d Free energy diagrams of possible surface states for RuO2 and MnO2. e Pauling electronegativity of cations in candidate rutile hosts versus that of Ru(IV). f Calculated formation energies for Ru substitution at edge versus terrace sites within β-MnO2. g Formation energy of an edge Mn vacancy in β-MnO2 and that of a Mn vacancy adjacent to a substituted Ru atom. h Calculated dissolution energy of surface Ru atom on RuO2 (110) compared with an edge Ru-doped β-MnO2. An ideal host should have the capacity for both dynamic charge buffering and electronic stabilization of high valence Ru species. To ensure atomic-level integration and effective host-guest interactions with RuO2, three isostructural rutile-phase oxides (TiO2, SnO2 and β-MnO2) with thermodynamically stable (110) surfaces were selected as representative models (Supplementary Data 1)16–18. Inspired by pseudocapacitance principles19–21, we hypothesized that host surface hydroxyls could buffer oxidative equivalents via rapid and reversible dehydrogenation (M-OH → M = O + H+ + e−). This synchronous proton-electron release maintains local electroneutrality, thereby preventing the over-oxidation of active Ru centers. To validate this buffering capacity, we calculated surface Pourbaix diagrams to analyze potential-dependent protonation/deprotonation behaviors, using RuO2 surface as a reference (Figs. 1b–d, S1 and S2)22,23. As seen, RuO2 exhibited a static deprotonated surface (1/2 μ1-O) across the relevant potential window (1.20–1.70 V), demonstrating the lack of intrinsic buffering capability. Conversely, all three host candidates maintained hydroxylated surfaces. Notably, β-MnO2 exhibited a two-step deprotonation initiated at a low potential of 1.32 V, indicating a highly responsive surface capable of mediating dynamic charge transfer. While all candidates satisfied the buffering prerequisite, the decisive criterion was the host’s electronic modulation of the Ru dopant, governed by cation electronegativity (Fig. 1e). Mn4+ has the highest electronegativity (1.92 vs. 1.84 for Ru4+) and was predicted to withdraw electrons from Ru24. Bader charge analysis quantified that the charge on a Ru atom embedded in β-MnO2 increased to +1.89 |e| (compared to +1.79 |e| in bulk RuO2), implying an elevated oxidation state primed for catalysis (Fig. S3). In contrast, TiO2 and SnO2 featured less electronegative cations and induced negligible electronic perturbation. Consequently, the synergy of dynamic buffering and electronic pre-oxidation capabilities established β-MnO2 as the optimal host. With β-MnO2 selected, we evaluated the thermodynamic stability of Ru substitution at two dominant surface sites, including basal terraces and low-coordination edges. The calculations revealed a strong thermodynamic preference (0.63 eV) for Ru substitution at edge sites compared to terrace sites (1.78 eV), pointing to a self-organizing atomic arrangement (Figs. 1f, S4)22. While creating an initial Mn vacancy at the edge is energetically costly (7.44 eV), the substitution of a single Ru atom dramatically altered the local energetic landscape, lowering the energy required to remove an adjacent Mn atom to 5.03 eV (Figs. 1g, S5). This reduction in vacancy formation energy facilitates sequential Ru incorporation along the edge sites, thereby providing a thermodynamic driving force for the assembly of atomically precise Ru-O-Mn motifs. We further calculated the Ru dissolution energy to quantify the stability of this architecture. The value for a Ru atom in this confined and hydroxyl-covered edge site was 2.06 eV, which is substantially higher than the 1.67 eV for a Ru atom on RuO2 (110) (Figs. 1h, S6). While these DFT insights are based on computational models that may not fully capture operational conditions, they consistently indicate that the specific Ru-O-Mn edge motifs have good potential to provide the necessary structure for the robust retention of active species. Synthesis and structural characterization Guided by the theoretical prediction of thermodynamic preference for Ru substitution at edge sites, we synthesized a series of RuxMn1-xO2 catalysts (where x = 0.01, 0.03, 0.05, and 0.08) using a modified site-selective ion-exchange method25 (Fig. 2a). Powder X-ray diffraction (XRD) patterns verified that all catalysts retained the pure β-MnO2 phase. A systematic shift of the (110) peak to lower angles confirmed the isomorphous substitution of smaller Mn(IV) ions by larger Ru(IV) ions (Fig. 2b)26. Crucially, the optimal Ru0.03Mn0.97O2 catalyst exhibited no phase-separated RuO2, corroborating the atomic-level incorporation of Ru. This substitution-induced local lattice distortion was further evidenced by systematic shifts in the Mn-O vibrational modes in Raman and FT-IR spectra (Fig. S7).Fig. 2Structural and electronic characterization of Ru0.03Mn0.97O2.a Schematic of the synthesis procedure. b XRD patterns of the RuxMn1-xO2 series and β-MnO2. The left panel shows the full diffraction patterns, while the right panel displays an enlarged view of the {110} peak. c AC-HAADF-STEM image of Ru0.03Mn0.97O2. d Corresponding EDS elemental mapping. e Intensity profiles along the selected regions in (c). The top and bottom panels display the profiles corresponding to the red and green dashed rectangles in (c), respectively. f Ru K-edge Fourier-transformed EXAFS spectra of Ru0.03Mn0.97O2, RuO2, and Ru foil. g Normalized Ru K-edge XANES spectra of Ru0.03Mn0.97O2, RuO2, and Ru foil. The inset shows a magnified view of the main absorption edge (highlighted by the dashed box). h Mn 3 s XPS spectra of Ru0.03Mn0.97O2 and β-MnO2. Structural and electronic characterization of Ru0.03Mn0.97O2. a Schematic of the synthesis procedure. b XRD patterns of the RuxMn1-xO2 series and β-MnO2. The left panel shows the full diffraction patterns, while the right panel displays an enlarged view of the {110} peak. c AC-HAADF-STEM image of Ru0.03Mn0.97O2. d Corresponding EDS elemental mapping. e Intensity profiles along the selected regions in (c). The top and bottom panels display the profiles corresponding to the red and green dashed rectangles in (c), respectively. f Ru K-edge Fourier-transformed EXAFS spectra of Ru0.03Mn0.97O2, RuO2, and Ru foil. g Normalized Ru K-edge XANES spectra of Ru0.03Mn0.97O2, RuO2, and Ru foil. The inset shows a magnified view of the main absorption edge (highlighted by the dashed box). h Mn 3 s XPS spectra of Ru0.03Mn0.97O2 and β-MnO2. The overall morphology of the catalyst was first examined by scanning electron microscopy (SEM) (Fig. S8), revealing that the β-MnO2 support possesses a characteristic nanorod structure with average dimensions of approximately 1.17 ± 0.2 μm in length and 100 ± 30 nm in diameter. Aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC-HAADF-STEM) provided direct visual evidence of the cooperative host-guest architecture. Atomically dispersed Ru atoms were clearly resolved as bright Z-contrast spots, preferentially populating the low-coordination edge sites of the β-MnO2 nanorods to form atomically precise Ru-O-Mn edge chains (Figs. 2c, d, S9). The measured Ru-Ru spacings of 2.9 Å and 3.4 Å matched the Mn-Mn periodicities in the host lattice, confirming that Ru atoms directly replaced Mn within the one-dimensional octahedral chains (Fig. 2e). The selective substitution of Ru at the edge sites of β-MnO2 nanorods is governed by a synergy of thermodynamic and electrostatic driving forces during the ion-exchange process. From a thermodynamic perspective, the edge regions of the nanorods possess a higher density of unsaturated metal sites compared to the basal planes. These sites exhibit higher chemical reactivity, confirmed by the lower energy barrier for Ru substitution (Fig. 1d, e)27. More crucially, the electrostatic driving force plays a decisive role in this site-selectivity. As directly visualized by Kelvin probe force microscopy (KPFM) (Fig. S10), the edges of the β-MnO2 nanorods exhibit a higher local surface potential compared to the basal surfaces. This intensified local electric field creates a strong electrostatic attraction for the positively charged Ru precursor cations, effectively guiding them to preferentially anchor and exchange at the edge positions28. This synergistic mechanism ensures the robust and site-selective anchoring of Ru at the host edges, which is instrumental in forming the highly active Ru-O-Mn edge-chain architecture. X-ray absorption spectroscopy (XAS) substantiated the formation of Ru-O-Mn coordination, the linkage within the edge chains (Figs. 2f, S11). The extended X-ray absorption fine structure (EXAFS) spectrum of Ru0.03Mn0.97O2 was dominated by a Ru-O-Mn scattering path at 2.36 Å, while the characteristic Ru-O-Ru signal of RuO2 located at 3.1 Å was absent29,30. Such structural consistency, originating from the isostructural nature of β-MnO2 and RuO2, underscores the formation of a robust embedded Ru-O-Mn architecture that effectively anchors the Ru active sites against migration or detachment. This atomic arrangement enabled the predicted interfacial charge transfer, which was confirmed by both X-ray absorption near-edge structure (XANES) and X-ray photoelectron spectroscopy (XPS). Specifically, the strong inductive effect exerted by the electronegative Mn4+ host via Ru-O-Mn bridges leads to a significant reduction in electron density at the Ru centers. A positive shift of Ru K-edge and Ru 3p3/2 binding energy (~0.9 eV) confirmed the structural induction of intrinsic high valence Ru(>IV) species31 (Figs. 2g, S12 and S13). Meanwhile, this electron withdrawal from Ru was balanced by a partial reduction of the Mn host to an average oxidation state of +3.14 (Figs. 2h, S14–S16). The accumulation of Mn(III) species promoted the formation of surface oxygen vacancies (Figs. S17, S18), which acted as Lewis acid sites to trigger water dissociation32,33. Consequently, this process spontaneously anchored a dense protective hydroxyl layer, priming the surface for the subsequent dynamic charge buffering. Acidic OER performance and PEMWE validation We first evaluated the acidic OER performance using a rotating disk electrode (RDE) setup. Polarization curves revealed that the Ru0.03Mn0.97O2 catalyst exhibited substantial activity, requiring an overpotential of only 232 mV to achieve 10 mA cm−2 (Figs. 3a, S19 and S20), which compared favorably both β-MnO2 and commercial RuO2. The corresponding non-iR-corrected voltammograms are provided in Fig. S19b. The contribution from carbon black was confirmed to be negligible (Fig. S21). This performance is attributed to the high intrinsic activity, as evidenced by a turnover frequency (TOF) of 1.81 s−1 and a mass activity of 6858 A gRu−1 at 1.50 V vs. RHE. This represents a 223-fold enhancement over commercial RuO2, demonstrating a competitive performance among recently reported acidic OER catalysts (Figs. S22, S23 and Table S1). The Faradaic efficiency (FE) for OER was further quantified via the rotating ring-disk electrode (RRDE) method34,35, confirming that the observed anodic current originates almost entirely from oxygen evolution rather than side reactions (Fig. S24).Fig. 3Electrocatalytic OER performance and PEMWE validation.a iR-corrected OER polarization curves in 0.1 M HClO4 (compensated using the measured cell resistance of 22.51 ± 0.52 Ω). All LSV curves were collected at a scan rate of 5 mV s−1 with a rotating speed of 1600 rpm and measured at room temperature (25 °C). b Long-term chronopotentiometry stability test at 20 mA cm−2. Inset: Comparison of the stability number (S-number). Post-reaction AC-HAADF-STEM image of the catalyst. c Polarization curves of a PEMWE device at 80 °C (without iR-compensation). d Long-term stability test of a PEMWE device employing the Ru0.03Mn0.97O2 anode at 1 A cm−2. The inset displays a schematic diagram of the assembled PEMWE device configuration used for the test. Electrocatalytic OER performance and PEMWE validation. a iR-corrected OER polarization curves in 0.1 M HClO4 (compensated using the measured cell resistance of 22.51 ± 0.52 Ω). All LSV curves were collected at a scan rate of 5 mV s−1 with a rotating speed of 1600 rpm and measured at room temperature (25 °C). b Long-term chronopotentiometry stability test at 20 mA cm−2. Inset: Comparison of the stability number (S-number). Post-reaction AC-HAADF-STEM image of the catalyst. c Polarization curves of a PEMWE device at 80 °C (without iR-compensation). d Long-term stability test of a PEMWE device employing the Ru0.03Mn0.97O2 anode at 1 A cm−2. The inset displays a schematic diagram of the assembled PEMWE device configuration used for the test. Crucially, this substantial enhancement in activity did not come at the expense of stability. In a 1750-h chronopotentiometry test at 20 mA cm−2, it exhibited negligible activity degradation, whereas commercial RuO2 failed within 70 h (Figs. 3d, S25). Post-test analysis detected minimal dissolution of Ru (0.43%) and Mn (1.02%). Consequently, the catalyst achieved a stability number (S-number) of 4.37 × 108, exceeding the RuO2 benchmark by over three orders of magnitude. Post-mortem HRTEM confirmed that the Ru-O-Mn edge chains remained well-preserved after prolonged operation (Figs. 3d, S26). We further validated the practical viability of this architecture in a PEMWE assembled via the Catalyst Coated Substrate (CCS) method36,37. Remarkably, our catalyst enabled the device to reach an industrial-scale current density of 3 A cm−2 at a cell voltage of 1.96 V (80 °C). This performance surpassed commercial RuO2 and IrO2 benchmarks while reducing precious metal usage by 80% (Fig. 3e). Furthermore, the electrolyzer sustained continuous operation for over 1000 h at 1 A cm−2 without significant voltage increase (Fig. 3f). This performance and stability are highly competitive compared to recently reported results (Table S2). This combination of robust durability, high mass activity, and minimal precious metal loading establishes a viable pathway toward cost-competitive green hydrogen production. Mechanistic origin of the enhanced activity To elucidate the mechanistic origin of the enhanced activity, we combined in situ spectroscopy with kinetic analysis. In situ Raman spectra revealed the formation of oxygenated intermediates on Ru0.03Mn0.97O2. While the peak at 668 cm−1 corresponds to the bulk metal-oxygen (M-O) vibration, the feature at 575 cm−1 is specifically assigned to the Ru-O* intermediate (Fig. 4a)38–41. This assignment is supported by the absence of the corresponding peak for β-MnO2, which effectively rules out the possibility of Mn-O* species (Fig. S27). Furthermore, the feature is validated by a significant redshift from 575 to 542 cm−1 upon 18O-labeling, whereas negligible shifts are observed in D-labeled electrolytes. Consistently, in situ differential electrochemical mass spectrometry (DEMS) performed on the 18O-labeled catalyst in normal electrolyte revealed negligible formation of 34O2 or 36O2, indicating that lattice oxygen does not participate in O2 evolution (Fig. S28). Notably, unlike β-MnO2, which exhibits a distinct MnO4− signal at ~839 cm−1 at potential above 1.7 V vs. RHE, no such peak was detected for Ru0.03Mn0.97O2 throughout the entire OER process. These results supported an adsorbate evolution mechanism (AEM), consistent with the structural robustness of Ru0.03Mn0.97O2 during OER, as the exclusion of lattice oxygen participation effectively prevents the formation of oxygen vacancies that typically trigger structural reorganization and metal dissolution42.Fig. 4Mechanistic investigation of OER activity.a In situ Raman for Ru0.03Mn0.97O2 under different OER potentials. The top and middle sub-panels display the spectra acquired at 1.6 V in 18O-labeled and D-labeled HClO4 electrolytes, respectively. The bottom sub-panel shows the potential-dependent in situ Raman spectra recorded from OCP to 1.8 V. b Schematic illustration of the Tafel slope isotope effect. c Tafel slope and transfer coefficient α plotted against log(J) derived from the polarization curves in H and D electrolytes. d OER polarization curves of Ru0.03Mn0.97O2 in HClO4 electrolyte from pH 0.4 to 1.3 on the RHE scale. The inset is the proton order estimated by ρRHE = (∂log(i)/∂pH), with ρRHE and i being the proton order and current density at a fixed potential of 1.50 V vs. RHE. e Surface Pourbaix diagrams for Ru0.03Mn0.97O2. f Schematic of the hydrogen-bond-assisted reaction pathway. g Corresponding Gibbs free energy diagrams for the OER. h LSR and derived volcano plot (η versus ΔG*O). Mechanistic investigation of OER activity. a In situ Raman for Ru0.03Mn0.97O2 under different OER potentials. The top and middle sub-panels display the spectra acquired at 1.6 V in 18O-labeled and D-labeled HClO4 electrolytes, respectively. The bottom sub-panel shows the potential-dependent in situ Raman spectra recorded from OCP to 1.8 V. b Schematic illustration of the Tafel slope isotope effect. c Tafel slope and transfer coefficient α plotted against log(J) derived from the polarization curves in H and D electrolytes. d OER polarization curves of Ru0.03Mn0.97O2 in HClO4 electrolyte from pH 0.4 to 1.3 on the RHE scale. The inset is the proton order estimated by ρRHE = (∂log(i)/∂pH), with ρRHE and i being the proton order and current density at a fixed potential of 1.50 V vs. RHE. e Surface Pourbaix diagrams for Ru0.03Mn0.97O2. f Schematic of the hydrogen-bond-assisted reaction pathway. g Corresponding Gibbs free energy diagrams for the OER. h LSR and derived volcano plot (η versus ΔG*O). We further investigated the reaction kinetics of Ru0.03Mn0.97O2 using isotope-dependent Tafel analysis, focusing on how H/D substitution influences the apparent Tafel slope in H2O versus D2O (Fig. 4b). We observed pronounced isotope sensitivity, where the Tafel slope shifted significantly from 42 to 118 mV dec−1 upon replacing H2O with D2O (Figs. 4c, S29). Prior studies43–45 have shown that such isotope-dependent Tafel slopes arise only when the rate-determining step (RDS) involves a concerted proton-electron transfer (CPET) due to the modification of transfer coefficient (α). Accordingly, our isotope-dependent Tafel behavior supported a CPET-type RDS, possibly the *O + H2O → *OOH. This concerted pathway is kinetically advantageous as it couples proton and electron transfer to bypass high-energy charged intermediates, effectively lowering the activation barrier for rapid oxygen evolution46. Unlike RuO2, which typically shows near pH-invariant kinetics on the RHE scale47, Ru0.03Mn0.97O2 exhibited a fractional proton order of 0.68 (Figs. 4d, S30). Notably, the nearly constant ρRHE values (0.63-0.69) across the 1.46–1.50 V vs. RHE range confirm the mechanistic pathway remains invariant within this potential window. Given the concerted nature of RDS, the observed non-ideal pH dependence implies that the apparent kinetics were decoupled from the intrinsic CPET barrier. Instead, reaction rates were modulated by a pH-dependent surface pre-equilibrium, which regulated the hydroxyl coverage and reactivity of key oxygenated intermediates, including *OH, *O, and *OOH48–50. To experimentally capture these active surface hydroxyls, methanol probe experiments were further conducted. As shown in Fig. S31, Ru0.03Mn0.97O2 exhibits a significantly higher methanol oxidation (MOR) current density and a much earlier onset potential compared to RuO2. Given that MOR in acidic media is primarily triggered by the presence of reactive surface *OH, this result provides direct kinetic evidence that our host-guest architecture promotes a higher coverage of active hydroxyl species at the interface. This enriched *OH environment not only facilitates the chemical probe reaction but also acts as the key regulatory factor in optimizing the OER pathway. To identify the origin of this pre-equilibrium, we computed the surface Pourbaix diagram for Ru0.03Mn0.97O2 using DFT calculations. Under OER conditions, a hydroxyl-rich surface termination (1/2 µ1-O + 1/6 µ2-OH) was predicted to be thermodynamically favored (Figs. 4e, S32 and S33). To show the dependence of activity trend with hydroxyl-rich termination, as shown in Fig. 4f, g, we compared the free-energy diagrams of three representative surfaces, including the hydroxylated and non-hydroxylated Ru0.03Mn0.97O2, and RuO2. Notably, all these surfaces exhibited the same linear scaling relation (LSR) between *OOH and *OH (ΔG*OOH = ΔG*OH + 2.9 eV), consistent with a hydrogen-bond-assisted reaction pathway51–54. Indeed, in situ ATR-FTIR spectra detected the distinct shoulder features at 1055/1065 cm−1 attributable to strengthened hydrogen-bonding interactions involving *OOH (Fig. S34). Crucially, this intercept value is significantly lower than the conventional universal scaling of ~3.2 eV, indicating that the conventional activity limitations are effectively circumvented. With the new intercept value fixed, the activity differences were primarily governed by the binding strength of *O. Specifically, the hydroxyl-rich termination tuned the binding strength of *O toward the Sabatier optimum, thereby balancing the two uphill steps (*OH → *O and *O → *OOH) and minimizing the theoretical overpotential to 0.25 V (Fig. 4h). Removing surface hydroxyls increased the overpotential to 0.52 V and shifted the RDS back to *OH deprotonation48,51,55. Moreover, this hydroxyl-assisted surface was energetically more favorable than that of RuO2 (0.68 V) and the alternative oxide pathway mechanism (OPM) (0.57–0.71 V, Fig. S35). Collectively, the surface hydroxyl species circumvent conventional LSR limitations by selectively stabilizing *OOH relative to *OH, while simultaneously tuning the O binding strength to the Sabatier optimum. Mechanistic origin of stability To elucidate the electronic origin of the catalyst’s resilience against oxidative stress, we tracked the surface state evolution using quasi in situ XPS techniques56. For the Ru0.03Mn0.97O2 catalyst, we monitored the spectral changes during a potential cycle from OCP to 1.5 V and back to OCP. The Ru 3 d peak remained invariant throughout this reversible cycle (Figs. 5a, S36a), indicating that the Ru centers maintained their stable valence state regardless of the potential stress. In contrast, the Mn host exhibited a dynamic and synchronized response involving both cation and anion sub-lattices (Fig. 5b, c). As the applied potential increased from OCP to 1.5 V, the Mn 2p peak progressively shifted to higher binding energies (Figs. 5b, S36b), reflecting the absorption of oxidative charge by the Mn host. This phenomenon identifies the MnO2 support as a redox-active charge reservoir, where the Mn sites undergo self-adaptive oxidation to sacrificially dissipate the mounting oxidative charge. Meanwhile, the O 1s spectra showed a synchronous positive shift in the lattice oxygen peak (Fig. 5c), mirroring the increased oxidation state of the metal centers. Crucially, despite this high oxidative stress, the characteristic peak of adsorbed hydroxyls not only remained prominent but exhibited a distinct potential-driven enrichment. According to quasi in situ XPS quantification, the hydroxyl content increased significantly from 21.59% at OCP to 32.75% at 1.3 V and reached a maximum of 36.97% at 1.5 V (Fig. S37). Even after returning to OCP, the hydroxyl coverage remained as high as 36.31%, confirming the robust retention of the protective hydroxyl-rich termination. Upon returning the potential to OCP, both the Mn 2p and O 1s peaks shifted back to their initial positions. This behavior was distinct from that of commercial RuO2 (Figs. 5d, S36c and S38). Under identical conditions, RuO2 showed a prominent and irreversible shift of the Ru 3d and 3p peaks to higher binding energies, demonstrating the rapid accumulation of unstable high valence Ru species that act as precursors to oxidative dissolution. Post-stability characterization further corroborated these in situ observations and revealed that both the Ru valence state and surface hydroxyl content in our catalyst remained virtually unchanged after prolonged operation (Figs. S39–S41). Specifically, the potential-driven enrichment ensures a high coverage of surface hydroxyl species anchored on the MnO2 host, which serves as the fundamental chemical reservoir for the catalyst’s stability. Crucially, the deprotonation of these host-carried hydroxyls acts as the primary mechanism for dynamically buffering the oxidative charge. Consequently, during the OER process, the MnO2 host adaptively absorbs the mounting oxidative charge through a synergistic dual-pathway. In addition to the self-adaptive valence state transitions of the Mn sites, the dynamic evolution of host-carried surface hydroxyls effectively shields the adjacent Ru active sites from detrimental over-oxidation and subsequent leaching.Fig. 5Mechanistic investigation of stability.a–c Quasi in situ XPS spectra of Ru0.03Mn0.97O2 for the a Ru 3d5/2, b Mn 2p3/2, and c O 1s regions during OER. d Quasi in situ XPS spectra of the Ru 3d5/2 region for RuO2 during OER. The detailed peak deconvolution curves for a, b, and d are provided in Supplementary Information. e Gibbs free energy diagrams for Ru dissolution from the surfaces of Ru0.03Mn0.97O2 and RuO2. Mechanistic investigation of stability. a–c Quasi in situ XPS spectra of Ru0.03Mn0.97O2 for the a Ru 3d5/2, b Mn 2p3/2, and c O 1s regions during OER. d Quasi in situ XPS spectra of the Ru 3d5/2 region for RuO2 during OER. The detailed peak deconvolution curves for a, b, and d are provided in Supplementary Information. e Gibbs free energy diagrams for Ru dissolution from the surfaces of Ru0.03Mn0.97O2 and RuO2. To elucidate the mechanistic link between the observed surface hydroxyls and the suppression of Ru oxidative dissolution, we employed DFT calculations to map the dissolution pathway and energetics (Fig. 5e). The dissolution of Ru-based oxides is typically governed by a multi-step process initiated by surface over-oxidation, followed by a critical structural reorganization of local RuO6 octahedron57. As the non-electrochemical distortion represents the kinetic bottleneck in the degradation pathway, we calculated the energy barrier for this specific event. For commercial RuO2, the barrier was moderate at 1.28 eV (Figs. 5e, S42). However, the barrier increased significantly to 2.78 eV for the hydroxylated Ru0.03Mn0.97O2 catalyst. Furthermore, we performed a control calculation on a model of our catalyst stripped of surface hydroxyls to isolate the specific contribution of hydroxyl-mediated charge buffering. This resulted in a barrier drop to only 1.19 eV. Crucially, the surface hydroxyls firmly coordinate to the Ru centers in Ru0.03Mn0.97O2, restricting the geometric distortion required for the release of soluble RuO4 species. As a result, the energy barrier for RuO4 dissolution in hydroxylated Ru0.03Mn0.97O2 is significantly elevated, exceeding that of its unhydroxylated counterpart by 1.04 eV (Fig. 5e). These computational insights indicate that static Ru-O-Mn motifs alone were insufficient to impart the observed stability, and instead, the dynamic charge buffering mediated by the surface hydroxyls effectively prevented the structural distortion and oxidative dissolution for our Ru0.03Mn0.97O2 catalyst. Discussion In summary, we validated a cooperative Ru-O-Mn edge chain architecture that overcomes the activity-stability trade-off for acidic water oxidation. Specifically, we constructed a Ru0.03Mn0.97O2 catalyst that sustains active high valence species via dynamic hydroxyl-mediated charge buffering. This architecture delivered competitive catalytic performance, achieving a mass activity 223-fold higher than that of commercial RuO2 and increasing the intrinsic stability number by three orders of magnitude. In a proton exchange membrane water electrolyzer, the catalyst supported an industrial current density of 3 A cm−2 at 1.96 V and maintained stable operation at 1 A cm−2 for over 1000 h. Notably, this performance surpassed commercial IrO2 and RuO2 benchmarks while reducing precious metal usage by 80%. These results demonstrate that the host in this architecture functions as an active participant rather than a passive support, where the retained surface hydroxyls improve reaction kinetics to the Sabatier optimum while ensuring structural robustness against oxidative dissolution. Our work establishes a rational catalyst design paradigm for oxygen evolution in proton-exchange membrane water electrolyzers. Methods Materials All chemicals were of analytical grade and were used as received without further purification. Manganese(II) sulfate monohydrate (MnSO4·H2O) and absolute ethanol were purchased from Tianjin Kermel Chemical Reagent Co., Ltd. Ammonium persulfate ((NH4)2S2O8) and methanol were sourced from Macklin Biochemical Co., Ltd. Ruthenium(III) chloride was obtained from J&K Scientific Co., Ltd. RuO2 was obtained from Alfa Aesar, and IrO2 was supplied by Umicore. Perchloric acid (HClO4) was purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. Commercial 40 wt% Pt/C catalyst and 5 wt% Nafion® dispersion (D521) were supplied by Johnson Matthey and DuPont, respectively. All aqueous solutions were prepared with ultrapure water (18.25 MΩ·cm, Millipore). Preparation of β-MnO2 In a typical synthesis, manganese(II) sulfate monohydrate (MnSO4·H2O; 1.69 g) and ammonium persulfate ((NH4)2S2O8; 2.28 g) were dissolved in 80 mL of ultrapure water. The resulting transparent and colorless solution was stirred for 30 min before being transferred to a Teflon-lined stainless-steel autoclave. The autoclave was then heated at 140 °C for 12 h. After cooling to room temperature (25 °C), the resulting black precipitate was collected by centrifugation, washed sequentially with ultrapure water and ethanol (three times each), and subsequently dried in an oven at 80 °C for 6 h. The self-supported β-MnO2 electrodes were prepared via an anodic electrodeposition method on a platinum-plated titanium felt (Pt/Ti felt) substrate. The electrodeposition was performed in an aqueous solution containing 0.3 M MnSO4 and 0.36 M H2SO4, maintained at 94 °C. A constant current density of 7 mA cm−2 was applied for 1 h. Following deposition, the as-prepared electrode was annealed in air at 450 °C for 5 h. Preparation of RuxMn1-xO2 The RuxMn1-xO2 catalysts were synthesized via a modified site-selective ion-exchange method25. In a typical procedure, a specific amount of the as-prepared β-MnO2 powder was dispersed in an aqueous solution of RuCl3 with a predetermined concentration. The suspension was stirred continuously at 60 °C for 24 h to facilitate the ion-exchange process. Afterwards, the solid product was collected by centrifugation, thoroughly washed several times with deionized water to remove residual ions, and subsequently dried. Finally, the collected powder was calcined in air at 200 °C for 1 h to yield the final RuxMn1-xO2 catalyst. Catalysts with different Ru contents (x = 0.01, 0.03, 0.05, and 0.08) were prepared by adjusting the initial concentration of the RuCl3 solution accordingly. Structural characterization X-ray diffraction (XRD) patterns were collected using a Bruker AXS D8 Focus (D8-S4) X-ray diffractometer with Cu Kα radiation in the scan range of 10° to 90° at a step of 0.01°. Raman spectra were obtained on Confocal Raman Microscopy (LabRAM HR Evolution) with an excitation wavelength of 532 nm. The scanning electron microscopy (SEM) images were taken on a Field Emission Scanning Electron Microscopy (Hitachi, Regulus 8100). Spherical aberration-corrected images were obtained using scanning TEM (JEM-ARM200F) at 200 kV. Transmission electron microscopy (TEM), high-resolution transmission electron microscopy (HRTEM), and elemental mapping analysis were performed on a JEM-F200 transmission electron microscope at 200 kV. X-ray photoelectron spectroscopy (XPS) was carried out on a ThermoFischer (ESCALAB 250Xi) instrument. The carbon peak at 284.8 eV was used as a reference to correct for charging effects. Fourier transform infrared (FT-IR) spectroscopy was constructed on a Bruker Invenio instrument. Kelvin probe force microscopy measurements The surface potential maps of the MnO2 nanorods were acquired using an Atomic Force Microscope (Bruker Dimension Icon) operating in KPFM mode. For sample preparation, a small amount of MnO2 powder was dispersed in absolute ethanol and ultrasonically treated for 20 min to form a homogeneous suspension. After standing briefly to allow large aggregates to precipitate, the supernatant was drop-casted onto an indium tin oxide (ITO) glass substrate. Once the solvent had completely evaporated, the substrate was secured to the sample stage using double-sided conductive carbon tape to ensure reliable electrical contact between the catalyst and the probe. During the measurements, topographical images and surface potential maps were collected synchronously. In situ ATR-FTIR measurements In situ ATR-FTIR spectroscopy was performed on a Bruker Invenio spectrometer. The spectral resolution of FTIR spectra is 4 cm−1. To obtain the working electrode, the Ru0.03Mn0.97O2 catalyst ink was used to prepare the catalyst film on the Au/Si substrate by the drop casting method (Fig. S43a). The reference electrode and counter electrode were a saturated calomel electrode and a platinum mesh, respectively. In measurements, 0.1 M HClO4 was used as the electrolyte. Signal acquisition is performed at open circuit potentials, 1.1 V, 1.2 V, 1.3 V, 1.4 V, 1.5 V and 1.6 V vs. RHE In situ DEMS with isotope labeling In situ DEMS experiments were performed on a differential electrochemical mass spectrometer provided by Linglu Instruments (Shanghai) Co., Ltd. A three-electrode system was employed for the measurements, consisting of 0.1 M HClO4 as the electrolyte, a saturated calomel electrode (Hg/Hg2Cl2, SCE) as the reference electrode, and a Pt wire as the counter electrode (Fig. S43b). To prepare the working electrode, the Ru0.03Mn0.97O2 catalyst was drop-casted onto an Au-plated porous membrane (Linglu, pore size ≤ 20 nm, thickness 40 μm). Initially, the catalyst was isotopically labeled by linear scanning in an 18O-labeled electrolyte within a potential range of 1.1 to 1.7 V vs RHE. Subsequently, the catalyst surface was rinsed several times with deionized water, dried, and further tested in a standard electrolyte without isotope labeling. In situ Raman spectra with isotope labeling In situ Raman spectra were acquired using a Horiba LabRAM HR Evolution confocal Raman microscope equipped with a 532 nm excitation laser. Using unlabeled HClO4 as the electrolyte, the signal is collected at a potential of 1.1 V to 1.8 V vs RHE (Fig. S43c). During the potentiostatic measurements at 1.6 V vs RHE, the in situ Raman spectra were collected several times at specific intervals to record the change of M-O Raman peaks in 18O-labeled (H218O) and D-labeled (D2O) electrolytes. Quasi in situ XPS measurements Quasi in situ XPS measurements were performed using an X-ray photoelectron spectroscope (Thermo Fischer, ESCALAB 250Xi) equipped with an electrochemical reaction cell interconnected to the analysis chamber. Typically, the catalyst was placed in the reaction cell with 0.1 M HClO4 electrolyte and subjected to electrochemical treatments (Fig. S43d) at various potentials (OCP, 1.0 V, 1.3 V, and 1.5 V vs. RHE). Following each treatment, the reaction cell was evacuated to 2 × 10−6 mbar, and the sample was subsequently transferred to the analysis chamber (<8 × 10−8 Pa) via a vacuum transfer rod without air exposure to preserve its chemical state. All binding energies were calibrated using the C 1 s peak (284.8 eV) as the reference. XAFS measurements Ru K-edge X-ray absorption spectroscopy (XAS) measurements, comprising both near-edge (XANES) and extended fine structure (EXAFS) regions, were performed at the Singapore Synchrotron Light Source (SSLS). The storage ring was operated with an electron beam energy of 2.5 GeV and a maximum current of 200 mA. Additionally, the corresponding Mn K-edge XAS data were acquired at the BL11B beamline of the Shanghai Synchrotron Radiation Facility (SSRF), China. For data analysis, the raw XAS signals were evaluated using the ATHENA program within the IFEFFIT software suite. Standard data reduction protocols were applied: the original absorption spectra underwent baseline removal, post-edge background subtraction, and edge-step normalization. To magnify the high-k oscillations, the normalized χ(k) signals were multiplied by a k3 weight. Finally, the k3-weighted data in the k-space range of 3–12.6 Å−1 were transformed into real space (R-space) via a Fourier transform, employing a Hanning window function with a smoothing parameter (dk) of 1.0 Å. Electrochemical measurements The electrochemical evaluations were conducted at ambient temperature (25 °C) utilizing an IVIUMSTAT electrochemical workstation (Ivium Technologies, Netherlands) equipped with a conventional three-electrode configuration. The working electrode consisted of a glassy carbon (GC) rotating disk electrode (RDE, diameter: 5 mm, geometric area: 0.196 cm2). A platinum mesh and a saturated calomel electrode (SCE, Hg/Hg2Cl2) functioned as the counter and reference electrodes, respectively. We prepared the 0.1 M HClO4 testing solution by diluting concentrated perchloric acid with ultrapure water, which was subsequently kept in dark storage. The pH of the freshly prepared electrolyte was confirmed to be 1.0 ± 0.2 (averaged from three parallel tests). All measured potentials were mathematically converted to the reversible hydrogen electrode (RHE) scale using the relationship: E (vs RHE) = E (vs SCE) + 0.305 V. This conversion factor was empirically determined in a hydrogen-saturated electrolyte using a Pt mesh working electrode, a Pt plate counter electrode, and the SCE reference. Linear sweep voltammetry (LSV) was run at 5 mV s−1, and the zero-current-density crossing point was taken as the thermodynamic potential limit (as depicted in Fig. S44). To formulate the catalytic ink, 5.0 mg of the synthesized catalyst and 1.0 mg of carbon black were mixed into a solvent blend containing 700 μL of absolute ethanol, 270 μL of ultrapure water, and 30 μL of Nafion ionomer. This suspension was ultrasonicated for 3 h to guarantee homogeneous dispersion. Subsequently, a 10 μL aliquot of this ink was drop-cast onto the GC substrate and allowed to air-dry, yielding a final mass loading of approximately 0.255 mg cm−2. OER polarization profiles were recorded through LSV from 1.2 to 1.8 V (vs RHE), with a 95% iR-drop compensation applied. We acquired electrochemical impedance spectroscopy (EIS) data at an applied potential of 1.3 V (vs. SCE) spanning a frequency window from 105 Hz down to 10−1 Hz. The uncompensated ohmic resistance (Rs) was extracted from the first intersection point of the Nyquist plot with the real axis at high frequencies. Averaging three separate measurements yielded an Rs value of 22.5 ± 0.52 Ω, which was consistently applied for the aforementioned iR correction. The electrochemically activity surface area (ECSA) was evaluated via double layer capacitance at the scan rates of 20, 40, 60, 80, 100 mV s−1 based on the Eq. (1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ECSA=\frac{{C}_{dl}\times S}{{C}_{s}}$$\end{document}ECSA=Cdl×SCsWhere Cs is the specific capacitance of the sample, and a general specific capacitance of Cs is usually 0.06 mF cm−2. S is the area of the working electrode (0.196 cm2). The turnover frequency (TOF) number was calculated using the following Eq. (2).2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$TOF=\frac{J\times A}{4\times F\times m}$$\end{document}TOF=J×A4×F×mWhere J is the current density, A is the area of the working electrode (0.196 cm2), F is the Faraday constant (96485.3 C mol−1), m is the number of moles of the total loaded Ru atoms. Here, we assume all Ru atoms are active sites to calculate the lower limit TOF, which provides a conservative and reproducible estimation of the intrinsic catalytic activity. The Faradaic efficiency (FE) was evaluated using a rotating ring-disk electrode (RRDE) system configuration. A Pt ring (0.187 cm2) and a glassy carbon disk (0.248 cm2) loaded with the catalyst served as the working electrode. The electrochemical measurements were conducted in an Ar-saturated 0.1 M HClO4 solution, which was continuously purged with high-purity Ar for at least 30 min prior to the test to eliminate dissolved oxygen. During the OER FE stability test, the rotation speed was maintained at a constant 1600 rpm to ensure steady-state hydrodynamic transport of the evolved oxygen from the disk to the ring. A constant anodic current of 1.0 mA cm−2 was applied to the disk electrode, while the Pt ring electrode was biased at a constant potential of 0.4 V vs RHE to instantly reduce the collected oxygen molecules. The FE for OER was quantified using the following Eq. (3).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$FE=\frac{{I}_{ring}}{{C}_{e}\times {I}_{disk}}\times100\%$$\end{document}FE=IringCe×Idisk×100%Where Idisk is the given current on the disk electrode. Iring is the collection current on the Pt ring electrode. Ce is the oxygen collection coefficient (∼0.2) for this type of electrode configuration. The catalyst was loaded on Ti felt for stability test at a current density of 20 mA cm−2. The stability number of catalysts (S-number) was computed from Eq. (4)58.4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Stability\,number=\frac{Number\,of\,{O}_{2}}{Number\,of\,dissolved\,metal}$$\end{document}Stabilitynumber=NumberofO2NumberofdissolvedmetalWhere the number of O2 is the molar number of oxygen evolved within a period of time. The total dissolved noble metal number was measured via ICP-OES. Fabrication of PEM water electrolyzer The titanium fiber felt and carbon paper loaded with anode and cathode catalysts, respectively, were used to prepare the membrane electrode assembly (MEA). A 130 μm thick Nafion 115 membrane was used for the membrane. Before MEA preparation, the proton exchange membrane was sequentially boiled in 3 wt% H2O2 solution, distilled water, 1 M H2SO4 solution and distilled water for 1 h each as a pretreatment. To assemble the MEA, a self-supported Ru0.03Mn0.97O2 catalyst was directly used as the anode, while 40% Pt/C (with a mass loading of 0.5 mg cm−2) was spray-coated onto the proton exchange membrane to serve as the cathode. After MEA fabrication, the MEA was pressed under 1.2 MPa at 110 °C. H2O was used as the electrolyte, which was circulated through the serpentine channel at a flow rate of 100 mL min−1. The performance of the PEM electrolyzer was evaluated by measuring polarization curves at 80 °C. Long-term PEMWE performances were evaluated by measuring the cell voltage at a constant current of 1 A cm−2. Computational Methods All first-principles spin-polarized computations were executed utilizing the Vienna Ab initio Simulation Package (VASP). To accurately describe the electron-core interactions, the projector augmented-wave (PAW) pseudopotential method was adopted, and the revised Perdew-Burke-Ernzerhof (RPBE) functional was chosen to evaluate the exchange-correlation energy59–62. The plane-wave basis set was truncated with a kinetic energy cutoff of 500 eV. Furthermore, to properly capture the strong electron localization behavior, the DFT + U scheme was incorporated. Specifically, the effective Hubbard U (Ueff) values were explicitly fixed at 3.9 eV for the Mn 3d63 orbitals and 1.2 eV for the Ru 4 d orbitals. The Brillouin zone was sampled leveraging a 3 × 3 × 1 Monkhorst-Pack k-point grid. Structural optimizations were considered complete once the maximum residual forces on each atom dropped below 0.02 eV Å−1 and the global energy variance between electronic iterative steps was strictly converged to 10−5 eV. Additionally, Grimme’s DFT-D3 methodology with Becke-Johnson (BJ) damping was implemented to adequately account for the long-range van der Waals (vdW) dispersion forces64,65. The thermodynamic Gibbs free energies evaluated at the standard conditions of 298.15 K and 1 atm were deduced using Eq. (4):4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varDelta {G}_{ads}=\varDelta {E}_{ads}+\varDelta {E}_{ZPE}-T\varDelta {S}_{ads}$$\end{document}ΔGads=ΔEads+ΔEZPE−TΔSads In this mathematical expression, ΔEads refers to the direct DFT-computed electronic adsorption energy. The parameters ∆EZPE and T∆Sads signify the resulting variations in the zero-point vibrational energy and the entropy equivalent, respectively. Both of these thermodynamic corrections were quantitatively extracted by performing vibrational frequency analyses on the surface-bound intermediates. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Transparent Peer Review file Source data Source Data Source Data