Bipolar membranes reveal surface-hydroxyl-structure-dependent water dissociation mechanism

Abstract

The high energy penalty for driving fast water dissociation hinders the implementation of emerging catalysis and energy conversion technologies with high sustainability, including electrolysis and fuel cells. This challenge stems in part from a limited understanding of how active-site structures influence water dissociation pathways. Here, we show that the structure of surface hydroxyls governs water dissociation activity in bipolar membranes. Using tin oxide quantum wires as model catalysts, we find that hydroxyl groups associated with bridging oxygen vacancies promote proton transfer and accelerate the alkaline water dissociation pathway at the membrane interface. By increasing the abundance of these hydroxyl structures through controlled synthesis, the bipolar membrane achieves a transmembrane voltage of 0.88 ± 0.02 V at 1 A cm⁻², equivalent to an overpotential of approximately 60 mV. When incorporated into a pure-water electrolyzer, the resulting membrane enables a full-cell voltage of 1.98 ± 0.02 V and stable operation for more than 550 hours at 1 A cm⁻². These findings provide guidance for the design of catalysts for efficient bipolar membrane devices.

Slow kinetics of water dissociation challenge the application of bipolar membranes in emerging energy technologies. Here, the authors show that hydroxyls adsorbed on bridging oxygen vacancies accelerate alkaline water dissociation and report enhanced pure-water electrolysis using bipolar membranes.

Introduction

Water dissociation (WD, H₂O ⇌ H⁺ + OH⁻) is a crucial gateway process for emerging electrochemical energy conversion and storage applications, such as water electrolysis^1–3, CO₂ electroreduction^4,5, ammonia electrosynthesis^6,7, and other solution-phase chemical/electrochemical hydrogenation reactions with high sustainability^8,9. To accelerate WD, catalysts and reverse biases are often applied, which can be understood in terms of two aspects: (i) the catalyst effect via heterogeneous protonation/deprotonation reactions and (ii) the second Wien effect that predicts increased WD rate constants in strong electric fields^2,10–12. In most cases, the catalyst effect plays a more influential role than the second Wien effect in promoting WD¹³.

However, as revealed by studies using bipolar membranes (BPMs), platforms for investigating the WD mechanism and developing efficient WD catalysts for H⁺ and OH^- generation^12,14,15, the WD activity of existing catalysts remains low. This is measured by transmembrane voltages (TMV, i.e., the voltage drops across the BPMs)^1,7,16,17, which are typically high (greater than 1 V experimentally versus the theoretical potential of ≤0.82 V) due to the additional large free energy required to dissociate H₂O and to drive the migration of produced H⁺ and OH⁻¹².

Recent efforts have been invested in deepening the understanding of the WD mechanism and in synthesizing high-performance WD catalysts. The latest research from Boettcher and coworkers identified that the electrical conductivity, surface area, and surface hydroxyl coverage of catalysts are descriptors of their WD activities¹⁸. This physical picture invokes hydroxyl groups as active sites for dissociating water into H⁺ and OH^-. Given the pH difference across a reverse-bias BPM, surface hydroxyls on WD catalysts are likely protonated and exist in the form of −OH₂⁺ near the acidic cation exchange layer (CEL) surface, and those near the alkaline anion exchange layer (AEL) surface would be deprotonated to −O^-. This results in two WD pathways, the acidic (Eqs. 1 and 2) and the alkaline (Eqs. 3 and 4) routes (Fig. 1a)^13,19–22:

$$-{{OH}_2}^{+}+H_{2}O\rightleftharpoons -OH+H_3{O}^{+}$$ 1

$$-OH+H_{2}O\rightleftharpoons -{{OH}_2}^{+}+{OH}^{-}$$ 2

$$-OH+H_{2}O\rightleftharpoons -O^{-}+H_3{O}^{+}$$ 3

$$-O^{-}+H_{2}O\rightleftharpoons -OH+{OH}^{-}$$ 4

Fig. 1. Surface hydroxyl structures and oxygen-vacancy configurations in SnO₂ QWs.

a Schematic illustration of two water dissociation (WD) pathways: the acidic route (red) and the alkaline route (blue) on active sites formed by different metal (M)-hydroxyl structures. b A schematic illustration of the nucleation-to-growth transition of SnO₂ QWs, alongside a representative high-resolution transmission electron microscopy (HRTEM) image presenting their twisted 1D morphology. Scale bars, 5 nm. c EPR spectra of QWs with varying synthesis durations. d Fractions of different O species analyzed from high-resolution O 1 s XPS spectra of these SnO₂ QWs. O_l (lattice oxygen), O_v (vacancy-associated oxygen),O_a (adsorbed oxygen species). e Normalized Sn K-edge XANES spectra of QWs4, QWs7, and QWs-Ann along with the Sn foil, SnO₂, and SnO references. f The average Sn oxidation states derived from K-edge XANES results. g The Raman spectra of QWs4, QWs7, and QWs-Ann, and the SnO₂ reference. h 1D ¹H-¹¹⁹Sn MAS NMR cross-polarization spectra and ¹H MAS NMR spectra of QWs7. Insets are the corresponding ball-and-stick structures. i The ball-and-stick structures of possible hydroxyl structures existing on SnO₂ surfaces, including hydroxyls adsorbed on bridging oxygen vacancies ( − O_brH), in-plane oxygen vacancies ( − O_inH), and five-coordinated Sn sites (Sn_5c−OH). Blue, red, and white spheres represent Sn, O, and H atoms, respectively. Source data for Fig. 1d–h are provided as a Source Data file.

The existence of the two pathways was supported by a strong correlation between the catalyst point of zero charge (PZC) and WD activity, indicating that high-PZC oxides are generally preferred catalysts for WD near the AEL¹. Nevertheless, on the surface of a WD catalyst, hydroxyls are adsorbed at different sites and adopt distinct structures, for instance, terminal hydroxyls bound to a metal center and bridge hydroxyls ( − O_brH) connecting to adjacent metal cations^18,23–27. The correlation between hydroxyl structures and their associated WD pathways (acidic or alkaline) has not been established to date. Notwithstanding that the link between surface hydroxyl coverage and WD activity provides a macroscopic understanding of the WD mechanism, microscopic-level insights assigning the two WD pathways to specific active site structures remain absent. Missing this mechanistic picture, unfortunately, leads to irrational design and low activity of today’s WD catalysts, as well as to their inefficiency in catalysis and energy applications.

In this article, we establish a link between the active site structure and WD activity by identifying that surface hydroxyl groups associated with bridging oxygen vacancies (b-O_vs) are preferred active sites for the alkaline WD pathway on SnO₂ quantum wire (QW) catalysts. The one-dimensional morphology of QWs could enable uninterrupted pathways to facilitate proton transfer between adjacent sites and mitigate transport limitations in nanoparticle-based systems, thereby providing a physical basis for enhanced WD kinetics. This finding yields a synthetic strategy for efficient WD catalysts through surface −O_brH enrichment. By controlling the catalyst synthesis time, surface −O_brH is enriched, as verified by Raman spectroscopy and solid-state nuclear magnetic resonance (SSNMR) studies. Experimental analysis and computational modeling identify that −O_brH is the active site for alkaline WD process, the number of which determines the WD kinetics on SnO₂ QWs. The optimized catalyst with the highest −O_brH population, when applied to BPMs, offers a low TMV of 0.82 ± 0.02 V at 500 mA cm⁻² and 0.88 ± 0.02 V at 1 A cm⁻², corresponding to overpotentials of ~0 and 60 mV and outperforming state-of-the-art WD catalysts and BPMs. Using the BPM equipped with this high-activity SnO₂ QW catalyst, we achieve a full-cell voltage of 1.98 ± 0.02 V at 1 A cm⁻² and 550-h stability for pure-water electrolysis at 60 °C. In BPM water electrolyzers (BPMWEs) using H₂SO₄ and KOH, 1 A cm⁻² is achieved at 1.89 ± 0.02 V, and the cell degradation rate is as low as 35 μV h⁻¹ during a 450-h operation.

Results

SnO₂ synthesis and characterization

SnO₂ QWs were synthesized using a solvothermal method (See details in the Methods section). The QWs are self-assembled by ultrasmall nanoparticles, and reaction times of 4 and 7 h (The samples are labeled QWs4 and QWs7) produce average QW diameters of 1.9 and 2.2 nm, respectively (Fig. 1b and Supplementary Fig. 1). X-ray diffraction (XRD) analysis of the SnO₂ QWs after capping ligand removal reveals diffraction peaks that align with the standard tetragonal SnO₂ phase (P4₂/mnm, JCPDS No. 41-1445)²⁸, confirming the formation of phase-pure crystalline QWs (Supplementary Fig. 2). Broad XRD peaks suggest a reduced coherent scattering domain, consistent with the ultrasmall dimensions of these materials²⁹.

Electron paramagnetic resonance (EPR) spectroscopy (Fig. 1c and Supplementary Fig. 3) detects characteristic sharp peaks at g = 2.002 attributed to superoxide ions adsorbed at surface oxygen vacancies (O_vs)³⁰. The O_v concentration in QWs4 is ~9.16 × 10¹² spins g^-1, and a comparable result of 1.09 × 10¹³ spins g^-1 was confirmed in QWs7 (Supplementary Table 1). These O_vs would transform into surface hydroxyls via water adsorption and dissociation, as further supported by additional machine-learning molecular dynamics (MLMD) simulations showing vacancy healing at both bridge and planar oxygen sites via water-assisted hydroxylation (Supplementary Fig. 4). An air-annealed control (QWs-Ann) was prepared to obtain a lower O_v concentration of ~2.02 × 10¹² spins g^-1 without changing the crystallinity and morphology of the QWs (Supplementary Fig. 5).

Increased O_v concentrations from QWs-Ann to QWs7 are further ascertained by the O 1 s X-ray photoelectron spectra of these materials (Supplementary Figs. 6–7 and Table 2). The fraction of O_v centered at 531.8 eV rises from about 42% in QWs-Ann to 61% in QWs7 (Fig. 1d). X-ray absorption near-edge structure (XANES) spectra at the Sn K-edge evidence distinct Sn oxidation states in these QWs (Fig. 1c), which decrease in the sequence of +3.3 (QWs-Ann) > +2.8 (QWs4) > +2.6 (QWs7) (Fig. 1f). This observation corroborates the formation of more O_vs by extending the synthesis duration, since the materials must keep charge neutral. Extended X-ray absorption fine structure (EXAFS) analyses reveal a reduction in the first Sn–O shell coordination number (CN) from 6.1 (SnO₂ reference) to ~5.1 (QWs 7) arising from O_v formation (Supplementary Figs. 8a, b, and Table 3). In phase with the reduced CN is a weakened Sn-O scattering amplitude presented on the EXAFS wavelet transform (WT) mappings, serving as evidence for the concentrated surface O_vs (Supplementary Fig. 8c)

Raman spectroscopy distinguishes two different O_v structures: in-plane O_vs (i-O_vs) at 571 cm⁻¹ and bridging O_vs (b-O_vs) at 629 cm⁻¹ (Fig. 1g and Supplementary Fig. 9)³¹. Additional Raman bands appear at 477, 692, 785, and 857 cm⁻¹—frequencies that are typically infrared-active in stoichiometric SnO₂ but become Raman-active in the presence of structural defects³². Moreover, all samples consistently exhibit the E_g phonon mode at 481 cm⁻¹ and the B_2g mode at 776 cm⁻¹, corresponding to in-plane oxygen vibrations and Sn–O bond contraction, respectively. Deconvoluting the spectra from 500 to 660 cm⁻¹ produces a positive correlation between the b-O_vs population and synthesis time, rising from ~34% in QWs4 to about 45% in QWs7 (Supplementary Table 4). This peak is almost absent on QWs-Ann surfaces, evincing the low b-O_vs population in this catalyst. Together with the EPR results, we concluded that b-O_vs are preferentially enriched with longer synthesis durations.

Consequently, at least two surface hydroxyl structures are expected as they can form at either the O_vs or the Sn terminals. Indeed, this is verified by 1D ¹H magic-angle spinning (MAS), 1D ¹¹⁹Sn MAS NMR, and 1D ¹H → ¹¹⁹Sn cross-polarization (CP) SSNMR studies (Fig. 1h, and Supplementary Figs. 10 and 11). The peak at ~6.2 ppm was reported to comprise terminal hydroxyl bound to one Sn cation, bridge hydroxyls connecting at least two Sn cations, and adsorbed water molecules^33,34. Further information gained from the ¹H → ¹¹⁹Sn CP NMR spectrum evinces that surface hydroxyls associate with both the five- and six-coordinated Sn sites (Sn_5c and Sn_6c), proving the presence of more than one −OH structure (Fig. 1i). PZC measurements imply that these observed changes in surface O_v and hydroxyls do not noticeably alter the surface charge properties of these SnO₂ QWs (Supplementary Fig. 12)^26,35,36.

The dependence of WD activity on −O_brH population

We then assessed the WD activities of BPMs using these SnO₂ QWs in a BPM electrolyzer equipped with two reference electrodes to sense the TMV (Fig. 2a)¹³. The assembly details are depicted in Supplementary Fig. 13.

Fig. 2. The dependence of WD activity on −O_brH population.

a A schematic of the membrane-potential-sensing setup. b, c TMVs at 1 A cm⁻² for QWs4 and QWs7 presented as a function of catalyst loading under different electrolyte combinations. Measurements were conducted at room temperature (25 °C) without iR correction. d, e Non-iR-corrected full-cell voltages at 500 mA cm⁻² and 1 A cm⁻² as functions of mass loadings of QWs4, QWs7, and QWs-Ann under electrolyte combinations of pH 7 | 7 and pH 7 | 14 at 60 °C. The width of the colored belt represents the standard deviation of three independent measurements. Error bars represent standard deviation (SD). Source data for Fig. 2b–e are provided as a Source Data file.

The correlation between catalyst mass loadings and TMVs falls into a U-type (or a semi-U-type) correlation for all SnO₂ QWs (Fig. 2b, c and Supplementary Figs. 14, 15), arising from the trade-off between the catalyst effect and the electric field effect^16,37,38. As the catalyst loading approaches the optimal, the WD kinetics converge to the Butler-Volmer model with a large exchange current density and a near-zero overpotential^13,22,39. For both QWs4 and QWs7, the optimal catalyst mass loading shifts to lower values for all tested electrolyte combinations, including pure water|pure water (pH 7 | 7), 0.5 M H₂SO₄ catholyte|pure water anolyte (pH 0.3 | 7), and pure water catholyte|1 M KOH anolyte (pH 7 | 14). At the same time, the TMVs at optimal loadings for all electrolytes decrease; for instance, in the case of 7 | 7, the lowest TMV is 1.07 ± 0.02 V for QWs4 and 0.90 ± 0.03 V for QWs7 at 1 A cm⁻², equivalent to a 170-mV overpotential reduction. Given that these two catalysts have comparable total O_v concentrations, but differ in the populations of b-O_vs and thus −O_brH, we tentatively link the improved WD activity of QWs7 to its higher b-O_v and −O_brH fraction on surfaces.

In addition, TMVs obtained using pH 7 | 14 combinations are ~400 mV higher than under the other two conditions for all SnO₂ QWs, indicating that these materials share pH-dependent WD activities. Given that the mechanism of WD on metal oxide surfaces involves proton transfer from water to the catalyst surface, across the catalyst, and from the catalyst surface to water, electrolyte pH is likely to impact the activation energies and equilibria of these steps on the same catalysts. We proposed the following picture to explain this pH dependence: when the anolyte is strongly alkaline (i.e., 1 M KOH), heterogeneous proton transfer from water to the catalyst surface is suppressed because the OH^- anions from the electrolyte push the equilibrium backward. For a similar reason, the pH 0.3 | 7 condition also exhibits slightly higher TMVs at 1 A cm⁻² than pH 7 | 7. The much more pronounced attenuation in WD activity at pH 7 | 14 reveals that water-to-surface proton transfer under alkaline conditions limits the WD rate on SnO₂ QWs.

The dependence of TMVs and electrolyte pH also determines the full-cell potential of BPMWE (Figs. 2d, e, and Supplementary Figs. 16, 17). We found that cell potentials decreased in the order of QWs-Ann > QWs4 > QWs7 across all pH conditions and current densities. Specifically, at 500 mA cm⁻² and 1 A cm⁻², the cell voltages of the pure water BPM electrolyzer based on the optimal SnO₂ QWs7 catalyst loading are 1.86 ± 0.02 V and 2.00 ± 0.02 V, respectively. Taken together, we assigned the markedly reduced TMV in the case of QWs7 to an accelerated alkaline WD pathway, and then to enriched surface −O_brH at bridging O_vs on this catalyst.

Computational studies

We sought molecular insights into the correlation between WD activity and surface hydroxyls to understand why enriching −O_brH dramatically promotes WD on SnO₂. The surface Pourbaix diagram suggests that under water dissociation conditions, SnO₂(110) is the most stable termination (Supplementary Fig. 18). We then compared intrinsic acidities of two prototypical motifs on the SnO₂(110) surface: the protonated terminal hydroxyl on the top of a five-coordinated Sn site (Sn_5c–H₂O⁺) and the one bridging two six-coordinate Sn sites ( − O_brH) (Fig. 3a, Supplementary Fig. 19 and Supplementary Data 1). Two-dimensional free energy surfaces for deprotonation were computed along two reaction coordinates, the protonation state (s_p) and the conjugate acid-base distance (s_d), respectively (Fig. 3b, c, and Supplementary Fig. 20).

Fig. 3. Computational insights into hydroxyl-assisted water dissociation pathways.

a The representative equilibrium configuration of the surface after MD simulations (water molecules omitted for clarity; detailed water adsorption structures are found in Supplementary Fig. 19). Free energy surfaces for deprotonation of (b) Sn_5c–H₂O⁺ and (c) −O_brH sites. s_p and s_d represent the protonation state and conjugate acid–base distance used as reaction coordinates for constructing the two-dimensional free-energy surfaces, respectively. The reference state corresponds to the partially hydroxylated surface in panel a. Energy barriers for (d) protonation and (e) deprotonation processes at two different sites, along with possible species. The proton source was used to classify the acidic and alkaline environments. N.B. denotes a barrierless process. f A schematic illustrating possible acidic and alkaline WD reaction pathways.

We took the partially hydroxylated surface as the reference state located at (0, 0), and the deprotonated state at (1, 10 ~ 15). The deprotonation free energies (ΔG) between these two states were determined to be 22.17 ± 2.25 and 26.7 ± 1.34 kJ mol⁻¹ for Sn_5c–H₂O⁺ and −O_brH, respectively, corresponding to pKa values of 3.88 ± 0.39 for Sn_5c–H₂O⁺ and 4.68 ± 0.23 for −O_brH, via $pKa=△G/2.303k_{b}T$. By taking the midpoint, a pH of 4.28 ± 0.31 at the point of zero charge (pH_PZC) was obtained, consistent with the experimental results.

We then investigated the proton transfer kinetics across various motifs by the proton source on this pKa-versified SnO₂(110) surface (acidic and alkaline conditions; Fig. 3d, e, and Supplementary Figs. 21, 22). Under acidic conditions, the protonation of the Sn_5c − OH and −O_br⁻ is almost barrierless, and the barrier for the reverse process generally has an activation energy of 0.44 eV for Sn_5c − H₂O⁺ and 0.49 eV for −O_brH, which governs the WD process. Converting Sn_5c − OH to Sn_5c − O⁻ is much more kinetically unfavorable due to its higher energy barrier of 0.56 eV. Under alkaline conditions, Sn_5c − H₂O⁺ and −O_brH₂⁺ encounter no barrier since no energy well or plateau is observed for the corresponding deprotonated product. The energy barrier for deprotonating −O_brH to −O_br⁻ is also minimal at 0.03 eV. This implies that the species involved in the alkaline pathway are likely Sn_5c − OH, −O_brH, and −O_br⁻. The reverse protonation on the −O_br⁻ site is energetically favored, with a barrier of 0.46 eV, which is 0.16 eV and 1.37 eV lower than that on Sn_5c − OH and −O_brH, respectively.

Based on these results, we propose that −O_brH is the site for alkaline WD under conditions when pH > pH_PZC (Fig. 3f): proton transfer from water to −O_br⁻ generates free OH⁻ and −O_brH, followed by successive deprotonation of −O_brH and protonation of −O_br⁻ to facilitate H⁺ transfer on SnO₂ surfaces. Conversely, at pH <pH_PZC, sites like Sn_5c − H₂O⁺ and Sn_5c − OH act as acidic WD sites, and deprotonation of Sn_5c–H₂O⁺ to Sn_5c − OH releases free H⁺ to complete the WD process. This chemical picture aligns with our experimental observation that the amount of −O_brH strongly correlates with the WD activity of SnO₂ QWs. In contrast, hydroxyls at i-O_v sites ( − O_inH) exhibit poor WD kinetics, as the protonation of −O_in⁻ to −O_inH requires an extremely high energy barrier of ~2.2 eV (Supplementary Fig. 23); therefore, these sites are spatially inaccessible to water. Accordingly, −O_inH₂⁺, −O_inH, and −O_in⁻ are unlikely to be the active sites for WD. The proposed mechanistic asymmetry in the WD kinetic landscape indicates that proton transfer along the alkaline pathway limits the WD rate on SnO₂ catalysts, an issue that can be addressed by increasing the surface −O_brH. We note that the density functional theory model represents an idealized catalyst surface and does not fully capture the structural complexity, dynamic interfacial hydration, and electric-field effects under practical catalytic conditions. Therefore, the calculations are intended to provide mechanistic insights that are consistent with, rather than definitive proof of, the experimental observations.

Catalysts with higher WD activity

Leveraging the above findings, we pursued a catalyst with greater WD activity. Thus, we extended the solvothermal synthesis time to 10 h to obtain QWs (labeled QWs10) that feature an average diameter of 3.1 nm, a further increased O_v concentration ( ~ 1.46 × 10¹³ spins g⁻¹) and −O_brH population ( ~ 55%) (Supplementary Figs. 24–29 and Supplementary Table 4). This catalyst provides TMVs of 1.09 ± 0.02 V at 500 mA cm⁻² and 1.23 ± 0.03 V at 1 A cm⁻² under pH 7 | 14 conditions at a loading of 19 μg cm^-2 (Fig. 4a, Supplementary Figs. 30 and 31), showcasing improved alkaline WD activity compared to QWs7 catalysts. Using the SnO₂ QWs10 catalyst for WD in pure water, we documented a TMV of 0.82 ± 0.02 V at 500 mA cm^-2 and 0.88 ± 0.02 V at 1 A cm^-2, corresponding to BPMWE full-cell voltages of 1.83 ± 0.02 and 1.98 ± 0.02 V for green H₂ production using pure water electrolysis at 60 °C (Fig. 4b and Supplementary Fig. 32).

Fig. 4. Structure–activity relationship and device performance enabled by bridging hydroxyl motifs.

a TMVs at 1 A cm⁻² for QWs4, QWs7, and QWs10 at the optimal mass loading of 19 μg cm⁻² under pH 7 | 14 electrolyte combinations. Measurements were conducted at room temperature (25 °C) without iR correction. The mean line represents the average of five independent measurements. Error bars represent standard deviation (SD). b Full cell and TMV polarization curves of QWs10 under pure water electrolysis at 60 °C. Error bars represent the standard deviations of three independent measurements. c Normalized physicochemical variables and their correlation with water dissociation activity. All parameters are scaled using min–max normalization. The right panel summarizes the Pearson correlation coefficients (r) between TMV and each variable, with color intensity indicating the magnitude of r. D, S, σ, ρ, Ov, −O_brH, and TMV denote particle diameter, specific surface area, electrical conductivity, density, oxygen-vacancy concentration, bridging hydroxyl groups associated with oxygen vacancies, and transmembrane voltage, respectively. d A schematic illustration of the designed water electrolyzer. e Long-term pure water electrolysis at 1.0 A cm⁻² and 60 °C using different gas diffusion electrodes (GDLs) and catalysts. GDLs employed include Toray 090 carbon paper (Toray-090), stainless steel fiber paper (SSFP), and platinum-coated titanium spacer (PTS). Source data for Fig. 4a–c and Fig. 4e are provided as a Source Data file.

It should be noted that further prolonging synthesis duration to 12 h (i.e., QWs12 catalysts) yields comparable −O_brH population and consequently WD activity (TMV ≈ 0.89 V) (Supplementary Fig. 33). We also confirmed the generality of the dependence of the WD activity on −O_brH population using rutile TiO₂ (R-TiO₂) and H₂-annealed R-TiO₂ (H-Annealed-R-TiO₂) (Supplementary Fig. 34): H-Ann-R-TiO₂ exhibits an order-of-magnitude lower surface hydroxyl concentration (2 % vs. 12%) and markedly inferior WD activity (1.82 V vs. 1.74 V).

To quantitatively resolve the structure–activity relationship, we performed Pearson correlation analysis across seven experimentally accessible catalyst variables, including TMV, −O_brH fraction, O_v concentrations, particle size (D), conductivity (σ), surface area (S), and catalyst density (ρ) (Fig. 4c, Supplementary Figs. 35, 36, and Supplementary Table 5). Among them, the −O_brH fraction exhibits the strongest negative correlation with TMV (r ≈ -0.98), whereas other variables show significantly weaker correlations. Hence, we concluded that the −O_brH fraction is a measurable material variable that determines WD kinetics.

The operational stability of the pure water BPMWE was tested at 1 A cm⁻². When Co₃O₄ serves as the anode, the overall water electrolysis activity degrades fast after 50 h due to the dissolution of Co₃O₄ in the anolyte (Fig. 4d, e, and Supplementary Fig. 37–40). Using IrO₂ catalysts supported by a Pt-coated Ti mesh, we observed a low cell degradation rate of about 0.87 mV h^-1 during continuous electrolysis of more than 400 h at 1 A cm^-2 and 60 °C, and replacing the anode with a fresh electrode restores activity to 2.23 V, only 0.12 V higher than the initial cell voltage (Fig. 4e). This validates the high stability of our BPM, given its activity degradation rate of ~0.28 mV h⁻¹. Post-operation Raman spectroscopy confirms the persistence of signatures associated with −O_brH motifs after electrolysis (Supplementary Fig. 41).

We confirmed improvements in metrics, including current densities, cell voltages, TMVs, stability, and cell degradation rates, as recorded using our BPM, compared with literature reports (Supplementary Table 6)^{1,7,16–19,25,40–42}. In an electrodialysis-like mode for water electrolysis at 25 °C, the BPM also exhibits low degradation rates. Like the pure water electrolysis measurement, the anode was replaced using a fresh Co₃O₄ electrode every 150 h, and the degradation rates are in the range of 35 to 164 μV h⁻¹ (Supplementary Fig. 42). The improved BPMWE performance compared to literature reports highlights that SnO₂ QWs10 enables industrial-level BPM water electrolysis, driven by the enriched hydroxyls at the bridging O sites of SnO₂.

Discussion

This work links the density of −O_brH adsorbed on the bridging O sites of SnO₂ to its WD activity and produces a strategy for efficient WD catalyst synthesis: enriching these −O_brH facilitates the alkaline WD pathway, the rate-limiting process for SnO₂ catalysts, endowing SnO₂-based BPMs to operate with a low TMV of 0.88 ± 0.02 V at 1 A cm⁻². −O_brH groups are identified as the preferred active centers for alkaline WD, owing to their low deprotonation barrier to −O_br⁻ and kinetically favorable reprotonation process. These features enable rapid heterogeneous water-to-surface protonation and proton transfer near the AEL within BPM interfaces. By contrast, terminal hydroxyls primarily mediate acidic water dissociation, and hydroxyls at in-plane oxygen sites remain largely inactive toward H₂O cleavage. Water electrolyzers using BPMs equipped with SnO₂ QWs10 catalysts, featuring the highest −O_brH density and optimal mass loading, enable a full-cell voltage of 1.98 ± 0.02 V at 1 A cm⁻² and stable green H₂ production from pure water for more than 550 h. In the electrodialysis-like mode, the full-cell voltage was further lowered to 1.89 ± 0.02 V at the same current density. These findings bridge the knowledge gap between active site structures and WD reaction pathways, offering design principles for efficient WD catalysts and BPMs that contribute to the development of more efficient and industrially relevant devices and processes in real-world chemical and energy industries.

Methods

Materials

Tin (IV) chloride pentahydrate (SnCl₄·5H₂O, 99%), oleic acid (analytical grade, AR), oleylamine (AR), ethanol (AR), cyclohexane (AR), isopropanol (AR), potassium hydroxide (KOH, AR), sulfuric acid (H₂SO₄, AR), and tetracyclohexyltin (Sn(C₆H₁₁)₄, 99%) were all purchased from Innochem and used as received without further purification. All chemicals were of analytical reagent (AR) grade. Glass-body Ag/AgCl reference electrodes were obtained from CH Instruments and stored in 1 M KCl solution prior to use to maintain a stable reference potential during electrochemical measurements.

Synthesis of SnO₂ QW catalysts

SnO₂ quantum wires (QWs) were synthesized using a modified solvothermal method adapted from a previously reported oleic-acid/oleylamine-assisted synthesis of SnO₂ nanostructures²⁹. Briefly, SnCl₄·5H₂O (1.7 mmol) was added to a mixed organic solvent containing oleic acid (20 mL) and oleylamine (2.5 mL). The mixture was heated under stirring until a transparent precursor solution was obtained. Ethanol (10 mL) was then introduced into the solution, after which the mixture was sealed in a Teflon-lined stainless-steel autoclave and heated at 180 °C. The reaction duration was adjusted to 4, 7, 10, or 12 h to regulate the morphology and defect structure of the resulting QWs. After cooling to room temperature, the products were collected by adding ethanol (40 mL) followed by centrifugation at 8000 × g for 8 min. The precipitates were purified by three washing cycles consisting of redispersion in cyclohexane, ethanol-induced precipitation, and centrifugation. The obtained solids were dried under vacuum at 70 °C overnight. The samples were denoted as QWs4, QWs7, QWs10, and QWs12 according to their respective reaction durations. To prepare an annealed control sample with a lower oxygen-vacancy concentration, QWs4 was annealed in air at 180 °C for 20 min in a muffle furnace, ground into powder, and denoted as QWs-Ann.

Anode and cathode preparation

Gas-diffusion electrodes were prepared by a modified spray-coating procedure based on a previously reported electrode-fabrication protocol¹. To prepare the catalyst ink, 40 mg of catalyst powder was dispersed in 4 mL of isopropanol containing 40 μL of the corresponding ionomer dispersion. Pt/C (60 wt% Pt) with Nafion D521 was used for cathode preparation, whereas Co₃O₄ or IrO₂ was combined with PiperION-A5-HCO₃-EtOH ionomer for anode preparation. The catalyst inks were sonicated for at least 4 h to obtain homogeneous dispersions before coating. Toray 090 carbon paper or platinum-coated sintered titanium frits were used as cathode gas-diffusion layers (GDLs), and platinum-coated sintered titanium frits or stainless steel fiber paper was used as anode GDLs. All GDL substrates had a geometric area of 2.2 × 2.2 cm². The mass of each bare GDL substrate was recorded before catalyst deposition. The catalyst inks were spray-coated onto the GDLs on a hotplate maintained at 85 °C until a dry catalyst loading of 2 mg cm⁻² was obtained. To improve interfacial contact between the electrodes and the ion-exchange layers, a thin ionomer overlayer was further applied to the catalyst-coated GDLs. PiperION-A25-HCO₃ ionomer (5 wt% in ethanol) was sprayed onto the anode side facing the anion-exchange layer, while Nafion D521 ionomer (5 wt% in ethanol) was sprayed onto the cathode side facing the cation-exchange layer. The overlayer amount was controlled to ~10% of the dry catalyst–ionomer loading, corresponding to 0.2 mg cm⁻².

Membranes and ionomers

The cation-exchange membrane, Nafion 211, was purchased from the SCI Materials Hub. The as-received membrane sheets were cut into 4 × 4 cm² pieces. Nafion D521 dispersion (5 wt% in ethanol) was used as an ionomer for the cathode and the BPM intermediate catalyst layer preparation (below). The anion-exchange membrane PiperION-A20-HCO₃ was purchased from SCI Materials Hub. The key physical and electrochemical properties of these membranes are summarized in Supplementary Table 7. The as-received membranes were cut into 4 × 4 cm² pieces and soaked in 0.5 M KOH for 2 h. Afterward, the membranes were rinsed and immersed again in a fresh 0.5 M KOH solution for an additional 24 h to complete the ion-exchange process. Prior to the BPM assembly, the membranes were rinsed with copious amounts of deionized water. The compatible AEL ionomer, PiperION-A25-HCO₃ (5 wt% in ethanol), was used for the anode and the BPM intermediate catalyst layer preparation (see below).

Preparation of BPMs

To fabricate the catalytic interlayer of the BPM, 10 mg of catalyst nanoparticles were dispersed in 4 mL of isopropanol and sonicated at room temperature ( ~ 25 °C) for at least 5 h to form a homogeneous suspension. Two ionomer dispersions were prepared separately: 10 μL of Nafion D521 (SCI Materials Hub) and 10 μL of PiperION-A25-HCO₃ (SCI Materials Hub, 5 wt% in ethanol) were each diluted into 1 mL of isopropanol and sonicated for at least 15 min before use. Nafion was used for the CEL-facing electrodes, while PiperION was used for the AEL-facing electrodes. The fabrication process for BPMs is shown in Supplementary Fig. 43.

The membrane coating was performed by sequentially spraying the Nafion ionomer solution, the catalyst suspension, and the PiperION ionomer solution onto a Nafion 211 membrane, which was affixed to a hotplate preheated to 85 °C. A handheld airbrush with a 0.2 mm nozzle was used for all spraying steps. Air purging was continuously applied during the coating process to promote rapid solvent evaporation and uniform film formation. The volume of ionomer solution sprayed was maintained at 25% of the catalyst suspension volume. The coated area was confined using a 2 cm × 2 cm PTFE stencil, and the membrane loading (L, in µg cm⁻²) was estimated using:

$$\mathrm{L}=\frac{{\mathrm{\omega }}_0\times {\mathrm{R}}_{\mathrm{c}}}{{\mathrm{A}}_{\mathrm{m}}}$$ 5

Where w₀ is the mass of nanoparticles used in the suspension, Aₘ is the membrane coating area (4 cm²), and R_c is the material retention coefficient. After spraying, the coated Nafion 211 membranes were gently removed and transferred to a Petri dish filled with deionized water for hydration and storage before BPM assembly. The extracted first limiting current densities are ~2.0, 2.9, and 2.1 mA cm⁻² for QWs4, QWs7, and QWs10 (Supplementary Fig. 44), respectively, suggesting broadly similar bulk membrane transport properties (e.g., thickness, ion-exchange capacity, and intrinsic ionic conductivity) under the tested conditions. Accordingly, the variations in WD activity are primarily attributed to differences in interfacial catalytic properties, consistent with the structure–activity relationship discussed in this work.

Assembly of BPMEAs

The bipolar membrane-electrode assembly (BPMEA) was fabricated and integrated into the electrolysis setup. First, the cathode gas diffusion layer (GDL, Toray Carbon Paper 090, wet-proofed, SCI Materials Hub) with a Pt catalyst layer was placed onto a platinum-coated titanium backplate featuring serpentine flow channels for ultrapure water supply and H₂ transport. Next, the Nafion membrane with a water dissociation (WD) catalyst layer was carefully positioned on the catalyst-coated GDL, ensuring that the WD-free side was in direct contact with the GDL catalyst layer, while avoiding wrinkles or irregularities. The PiperION anion exchange layer (AEL) was then placed on top of the Nafion® membrane, ensuring that the WD catalyst layer faced the AEL, while maintaining a smooth, wrinkle-free interface. Subsequently, additional gaskets and an anode GDL (stainless steel fiber paper, SCI Materials Hub) coated with either a Co₃O₄ or an IrO₂ catalyst layer were placed on top of the stack, with the catalyst layer facing the AEL. Finally, a corrosion-resistant stainless steel backplate with serpentine flow channels was positioned on top for H₂ removal. The entire electrolyzer assembly was uniformly tightened using a torque wrench (6 Nm), with screws evenly distributed along the edges of the metal block to ensure a secure, uniform compression.

Material characterizations

High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were acquired using a TALOS 200X field-emission transmission electron microscope (FEI) operated at 200 kV. X-ray diffraction (XRD) patterns were recorded on a PANalytical diffractometer equipped with Cu Kα radiation (λ = 1.5406 Å). Electron paramagnetic resonance (EPR) measurements were performed using a Bruker ELEXSYS II spectrometer operating in X-band mode ( ~ 9.20 GHz). X-ray photoelectron spectroscopy (XPS) and ultraviolet photoelectron spectroscopy (UPS) were conducted on an Axis Supra spectrometer (SHIMADZU, Japan) equipped with a monochromatic Al Kα source (1486.6 eV). All binding energies were calibrated using the C 1 s peak at 284.8 eV from adventitious carbon. X-ray absorption spectroscopy (XAS) was carried out in fluorescence mode at the 12-ID multipole wiggler beamline at the Australian Synchrotron (Clayton, Victoria). Reference samples included Sn foil, SnO, SnO₂, QWs4, QWs7, QWs10, and QWs-Ann. Raman spectra were collected using an HR800 confocal Raman microscope (Horiba Jobin Yvon, France) equipped with a 633 nm excitation laser. ¹H, ¹¹⁹Sn, and ¹H − ¹¹⁹Sn cross-polarization (CP) magic angle spinning (MAS) NMR spectra were measured using a Bruker AVANCEIII/WB-400 spectrometer with a magnetic field strength of 9.4 T. Resultant ¹H-¹¹⁹Sn MAS NMR cross-polarization chemical shifts were externally referenced to solid Tetracyclohexyltin. The electrical conductivity of the powdered samples was measured using a ST2722B four-point probe resistivity tester, with applied pressures ranging from 2.0000 MPa to 30.000 MPa.

Electrophoretic mobility measurement via dynamic light scattering (DLS)

Electrophoretic mobility was measured using dynamic light scattering (DLS, Nano ZS90, Malvern Instruments) to determine the isoelectric point of the catalyst samples. Measurements were conducted in automatic acquisition mode with a maximum collection time of 100 s and an acquisition interval of 1 s. Each pH value was assessed through six independent DLS measurements, and the results were averaged for statistical reliability. The applied voltage (10 V) and electric field frequency (10 Hz) were adjusted as needed to ensure stable readings across different pH conditions. Zeta potential values were calculated using the Smoluchowski model. For sample preparation, catalyst powders were dispersed at ~10 mg/L in deionized water, with pH adjustments made using KOH and HCl while maintaining a constant ionic strength across all suspensions. The dispersions were ultrasonicated for 20 min, then immediately transferred to polytetrafluoroethylene (PTFE) vials to minimize adsorption effects. Measurements were performed using Malvern cuvettes to ensure optical consistency. To ensure reproducibility, each measurement ( > 6 DLS acquisitions) was repeated at least three times. Final PZC values were determined by averaging results from multiple independent experiments, ensuring high accuracy and repeatability.

Surface area measurement

The surface area of the samples was determined via N₂ adsorption/desorption isotherms at 77 K using a Micromeritics ASAP 2460 surface area analyzer. Before measurement, samples were dispersed in an isopropanol/water mixture (1:1 wt.%) and then dried under vacuum at room temperature for 18 h. Subsequently, they were activated at 150 °C for 10 h to remove residual solvents and gases. Activation was deemed complete when the outgassing rate dropped below 2.5 μTorr min⁻¹. The actual sample mass was determined by the mass difference between the empty and post-activation loaded sample tube.

Transmembrane voltage measurements

The transmembrane voltage across the bipolar membrane (BPM) junction was measured using a custom-designed membrane potential-sensing setup, adapted from previous work by Chen et al.¹³ The membrane potential sensing configuration was designed to resolve the voltage drop across the bipolar membrane (BPM) junction under water dissociation (WD) conditions.

A cation-exchange layer (CEL) sensing strip (2 cm × 8 cm) was aligned at one end of the hydrogen-evolution-reaction-(HER)-side gas diffusion electrode (GDL) and placed outside the active area. A custom-cut PTFE spacer (0.1 mm thickness) was then positioned to isolate the active region. Subsequently, a CEL membrane loaded with WD catalyst was placed on top of the cathode GDL, with its uncoated side in contact with the GDL. The CEL sensing strip was exposed via a local window in the PTFE spacer to enable electrical connection outside the active region. An anion-exchange layer (AEL) membrane was then stacked on top of the CEL, sandwiching the catalyst layer to construct the BPM structure. Another PTFE spacer was placed above the AEL, and an AEL sensing strip was aligned over the AEL membrane with a locally exposed area allowing contact. The anode GDL (catalyst-coated side facing downward) was placed over the AEL to complete the membrane electrode assembly (MEA). Flow field plates and current collectors were assembled and secured using eight screws, each tightened to 50 inch-pounds (5.6 Nm).

Two O-rings (inner diameter: 1.2 cm; outer diameter: 2.6 cm) were placed above the exposed ends of the CEL and AEL sensing strips. Custom-made flanges (tube diameter: 12 mm; wall thickness: 3 mm; flange diameter: 26 mm; thickness: 5 mm) were mounted above the O-rings and clamped in place to ensure a sealed connection. Ag/AgCl reference electrodes were inserted into each of the O-ring compartments, immersed in 0.1 M H₂SO₄ and 0.1 M KOH electrolytes on the CEL and AEL sides, respectively. A schematic diagram of the layer-by-layer BPM assembly is provided in Supplementary Fig. 13.

During measurements, electrolytes were continuously circulated on both sides of the cell at room temperature under reverse-bias conditions. Electrochemical testing was carried out using a dual-channel potentiostat (DH7003B, Donghua Instruments). For Channel 1, the working electrode (WE) and sensing electrode (SE) leads were connected to the cathode current collector, while the counter electrode (CE) and reference electrode (RE) leads were connected to the anode current collector. Channel 2 was configured with the SE and RE leads connected to the two Ag/AgCl reference electrodes; the WE and CE leads were unused. Both channels were operated synchronously: Channel 1 performed a chronoamperometric test by stepping current densities at 20, 40, 60, 80, 100, 200,…, up to 1000 mA cm⁻² (each step held for 60 s), while Channel 2 recorded open-circuit membrane potential data every 0.5 seconds. The transmembrane potential values at the end of each current step were used for analysis of the polarization curve.

Electrolyzer measurements

For the polarization curves, this study focused on high current density regimes (20 to 1000 mA cm⁻²), where effective water dissociation (WD) is essential for sustaining ionic current and where steady-state voltage readings are reliable. Polarization curves were obtained by incrementally increasing the current density in two regimes: from 20 mA cm⁻² to 100 mA cm⁻² in 20 mA cm⁻² steps, and from 100 mA cm⁻² to 1000 mA cm⁻² in 100 mA cm⁻² steps. During each measurement, both the cathode and anode chambers were supplied with either 18.2 MΩ·cm deionized (DI) water or 0.5 M H₂SO₄ | 1 M KOH solution at 60 °C and a flow rate ~8 ml min⁻¹. Standard electrolyte solutions were used in the electrochemical measurements. The electrolyte configurations are denoted using nominal pH values defined from the electrolyte compositions. Specifically, 0.5 M H₂SO₄ and 1 M KOH were denoted as approximately pH 0.3 and pH 14.0, respectively, while deionized water was denoted as pH 7.0. These pH labels were used to describe the electrolyte configurations, including pH 7.0 | 7.0, pH 0.3 | 7.0, and pH 7.0 | 14.0.

The cell voltage was directly recorded from the current collector plates to eliminate resistive artifacts introduced by cable impedance. At each step, the system was allowed to stabilize for 2 min, and the voltage at the end of this period was recorded to construct the polarization curve. Prior to formal testing, an activation procedure was conducted using DI water at 60 °C circulated to both cathode and anode chambers. The current was initially held at 50 mA for 1 min, followed by stepwise increases of 50 mA increments from 50 to 500 mA, with each step maintained for 60 seconds. The final 500 mA increment was held for 5 minutes. If the activation voltage exceeded 2.5 V at 500 mA cm⁻², the membrane electrode assembly (MEA) was disassembled and reassembled to ensure a proper membrane interface and contact. Initial electrochemical screening was conducted under a symmetric pH environment (pH 7.0 | 7.0), followed by tests under asymmetric pH gradients (pH 7.0 | 14.0 and pH 0.3 | 7.0). Before each test, the BPM device was equilibrated in the corresponding electrolyte configuration for 1 h under continuous circulation using a peristaltic pump to maintain flow and minimize concentration gradients.

During long-term durability tests, deionized water was continuously circulated and refreshed on both the cathode and anode sides to minimize polarization effects and maintain a stable operating environment. For electrodialysis environment testing, the electrolyte feeds were switched to 0.5 M H₂SO₄ (catholyte) and 1 M KOH (anolyte), replacing DI water to simulate harsh acid–base operating conditions. To prevent corrosion under alkaline conditions, a nickel fiber paper electrode (SCI Materials Hub) was employed as the anode gas diffusion layer (GDL) instead of stainless steel. All other operational parameters remained consistent with those used in the standard BPM water electrolysis (BPMWE) tests. Notably, while the cathode and anode GDLs were replaced at each current density transition, the same BPM membrane was used throughout the entire testing sequence to ensure comparability and evaluate membrane-level durability. Long-term stability measurements were performed once for each condition, and degradation rates were estimated by linear fitting of the voltage–time traces.

Theoretical calculation details

Models

Using a symmetric, periodic slab with five O−Sn−O layers, we created the SnO₂(110) surface, which had lateral dimensions of a 4 × 2 surface cell (see Supplementary Fig. 19 and Supplementary Data 1)⁴³. We separated the symmetric slabs by a solvent gap of 25 Å, forming an orthorhombic supercell with dimensions of 12.7 × 13.4 × 41.0 ų. To ensure the bulk water volume in the supercell was roughly cubic, we adjusted the water density in the bulk solvent region to ~1 g/cm³. This resulted in 143 water molecules in the solvent space between the SnO₂ slabs.

Molecular dynamics simulations

Molecular dynamics simulations were performed using LAMMPS⁴⁴ combined with the DeepMD-kit potential⁴⁵. The DPMD model was obtained from ElectroFace, an artificial intelligence-accelerated ab initio molecular dynamics dataset for electrochemical interfaces, developed by Jun Cheng’s group from Xiamen University. The training dataset was generated using the DP-GEN concurrent learning package, which adaptively samples configurations through iterative training, exploration, and DFT labeling stages. The final dataset comprises 6000 structures of the SnO₂(110)/water interface (669 atoms): 600 structures randomly extracted from a 100 ps AIMD trajectory (the initial dataset) and 5400 structures collected via concurrent learning. MD exploration was performed at 330 K, 430 K, and 530 K to ensure sufficient coverage of phase space. The DeepMD-kit potential utilized a pretrained model based on the Deep Potential-Smooth Edition (DeepPot-SE) as the structural descriptor⁴⁶. The cutoff radius for the local environment was set to 6.0 Å, with a cutoff smooth (rcut smth) of 0.5 Å. The final root mean square error (RMSE) for energy was 0.504 meV/atom, and the force error was 66.3 meV/Å. All simulations were conducted in the canonical (NVT) ensemble at 330 K using a Nosé-Hoover thermostat⁴⁷, with a time step of 0.5 fs. The time scale of the molecular dynamics simulation is 1 ns.

Constrained ab initio molecular simulations

Constrained ab initio molecular dynamics (AIMD) simulations were used to evaluate the proton-transfer processes involved in protonation and deprotonation at the SnO₂(110)/H₂O interface. The simulations were carried out using the slow-growth approach, following a previously reported constrained AIMD protocol with modifications for the present SnO₂ surface models. All AIMD calculations were performed with the CP2K/Quickstep package^48,49. The electronic structure was described using the Perdew–Burke–Ernzerhof (PBE) exchange–correlation functional together with Grimme’s D3 dispersion correction to account for van der Waals interactions at the solid–liquid interface^50,51. Core electrons were treated with Goedecker–Teter–Hutter pseudopotentials⁵², and valence electrons were expanded using short-ranged double-ζ basis sets with one set of polarization functions (DZVP). The plane-wave cutoff for the auxiliary electron-density grid was set to 400 Ry, and the self-consistent-field convergence criterion was 3 × 10⁻⁷ atomic units. Molecular dynamics trajectories were propagated in the canonical ensemble with a time step of 0.5 fs. The temperature was maintained at 330 K using a Nosé–Hoover thermostat. Proton-transfer free-energy profiles were obtained by constraining the O–H distance as the collective variable. For protonation, the collective variable described the approach of a proton from interfacial water toward the adsorbate oxygen on the SnO₂ surface. For deprotonation, the collective variable described the transfer of the adsorbate proton toward a solvating water molecule. During the slow-growth simulations, the target value of the constrained collective variable was varied at a rate of 0.0005 Å fs⁻¹.

Voronoi diagram

To track the protonation state of the surface during equilibrium sampling simulations, we used the same collective variable of the protonation state (CV) s_p as in the enhanced sampling simulations⁵³. According to the Voronoi principle, the total number of hydrogen atoms assigned to a specific site i is the fraction of protons that are much closer to that site than to any other site.

$$w_i\left(R_i\right)=\left\{\begin{array}{c}\hfill \begin{array}{cc}\hfill 1,\hfill & \hfill \begin{array}{cc}\hfill \mid R_i-R_j\mid <\mid R_m-R_j\mid ,\hfill & \hfill \forall m\ne i\hfill \end{array}\hfill \end{array}\hfill \\ \hfill \begin{array}{cc}\hfill 0,\hfill & \hfill \begin{array}{cc}\hfill \mid R_i-R_j\mid <\mid R_m-R_j\mid ,\hfill & \hfill \forall m\ne i\hfill \end{array}\hfill \end{array}\hfill \end{array}\right)$$ 6

To reproduce the same behavior as in the above equation without discontinuity or singularity, the index functions w_i are replaced by the softmax functions n_i,

$$n_i=\sum _{j=1}^{N_H}\frac{e^{-\lambda \mid R_i-R_j\mid }}{{\sum }_{m=1}^{N_c}{e}^{-\lambda \mid R_m-R_j\mid }}$$ 7

Where the first sum runs over all H atoms with index j, and m is the index of Voronoi centers. The parameter λ controls the smoothness of the function.

Protonation state

An acid-base site has a reference value n⁰_i that counts the number of hydrogen atoms bonded in the neutral state⁵³. The difference between the current number of hydrogen atoms and this reference value is:

$${\delta }_i=n_i-n_0$$ 8

Specifically, the net protonation state of the SnO₂ surface is given by:

$$s_p=s_{p,cus}+s_{p,br}$$ 9

Where, s_p,cus and s_p,br are defined as:

$$s_k=\sum _{k=0}^{N^s-1}{2}^k\times q_k$$ 10

Here, k corresponds to the indices used to label the reactive sites, q_k is the total proton excess/defect associated with each reactive species, and N is the number of inequivalent reactive sites. The following conventions were used for the Sn_5c and O_br sites⁵⁴:

• For s_p,Sn5c: k = 0 corresponds to the index of the Sn_5c − OH/H₂O sites, k = 1 to the O_br and water;

• For s_p,br: k = 0 corresponds to the index of the O_br sites, k = 1 to the Sn_5c − OH/H₂O sites and water; This defines the reactive sites between which proton transfer can occur, resulting in a unique net surface protonation state for different scenarios.

Conjugate acid-base distance

Once the proton transfer has occurred, the acid-base sites will have an abnormal number of hydrogen atoms within their Voronoi polyhedra. Then it exhibits a large change of 0 upon proton transfer.

Author contributions

Y.W. and H.L. conceived the idea. Y.W. and P.O. supervised the project. H.L. designed and carried out the experiments and wrote the manuscript. C.Y. performed computational studies. M.S. and Y.Y. performed the XAS measurement. Y.Y. and F.L. contributed to data analysis, discussions, and paper preparation. All authors discussed, commented on, and revised the manuscript.

Peer review

Peer review information

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

Funding

This work was financially supported by the National Key Research and Development Program of China (2024YFB4106400), the National Natural Science Foundation of China (22179088), the National Natural Science Fund for Excellent Young Scientists Fund Program (Overseas), Collaborative Innovation Center of Suzhou Nano Science & Technology, the 111 Project, Joint International Research Laboratory of Carbon-Based Functional Materials and Devices, Suzhou Key Laboratory of Advanced Photonic Materials (Grant SZS2023010), the ARC Center of Excellence for Green Electrochemical Transformation of Carbon Dioxide (CE230100017) and a Discovery Project (DP250100576) funded by the Australian Government.

Data availability

Experimental data generated in this study are available in the Supplementary Information and Source Data file. Optimized computational structures and representative initial and final configurations from molecular dynamics simulations are provided in Supplementary Data 1. No restrictions apply to data availability. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at (10.1038/s41467-026-75170-x).