Species mass transfer governs the selectivity of gas diffusion electrodes toward H₂O₂ electrosynthesis

Abstract

The meticulous design of advanced electrocatalysts and their integration into gas diffusion electrode (GDE) architectures is emerging as a prominent research paradigm in the H₂O₂ electrosynthesis community. However, it remains perplexing that electrocatalysts and assembled GDE frequently exhibit substantial discrepancies in H₂O₂ selectivity during bulk electrolysis. Here, we elucidate the pivotal role of mass transfer behavior of key species (including reactants and products) beyond the intrinsic properties of the electrocatalyst in dictating electrode-scale H₂O₂ selectivity. This tendency becomes more pronounced in high reaction rate (current density) regimes where transport limitations are intensified. By utilizing diffusion-related parameters (DRP) of GDEs (i.e., wettability and catalyst layer thickness) as probe factors, we employ both short- and long-term electrolysis in conjunction with in-situ electrochemical reflection-absorption imaging and theoretical calculations to thoroughly investigate the impact of DRP and DRP-controlled local microenvironments on O₂ and H₂O₂ mass transfer. The mechanistic origins of diffusion-dependent conversion selectivity at the electrode scale are unveiled accordingly. The fundamental insights gained from this study underscore the necessity of architectural innovations for mainstream hydrophobic GDEs that can synchronously optimize mass transfer of reactants and products, paving the way for next-generation GDEs in gas-consuming electroreduction scenarios.

Introduction

Hydrogen peroxide (H₂O₂) is one of the 100 most important chemicals in the world. Given the advantages of electrolysis controllability and device modularity, the H₂O₂ electrosynthesis route via the two-electron oxygen reduction reaction (2e⁻ ORR) offers enhanced versatility across a spectrum of niche applications, including portable air disinfection and consumer-oriented sewage purification^1,2,3,4. The key bottleneck hindering the industrial implementation of this technology lies in the difficult-to-reconcile contradiction between the underwater gas consumption characteristics of the 2e⁻ ORR and the inherent low O₂ solubility in water^3,5. Gas diffusion electrodes (GDEs), which allow the direct utilization of gaseous feedstocks and greatly shorten the gas delivery length, are widely recognized as the most suitable platform for 2e⁻ ORR^6,7,8,9,10. Particularly for those GDEs with breathing-mimicking designs, the capability to obviate energy-intensive aeration facilities paves the way for further advancements toward commercially viable electrolysis stacks^11,12. In light of these considerations, the material-centric screening of electrocatalysts to direct ORR selectivity toward the desired 2e⁻ pathway, followed by their integration into GDEs for bulk electrolysis, has recently emerged as a priority^13,14,15.

Accurate quantification of H₂O₂ selectivity is crucial and typically depends on different testing methodologies tailored to electrocatalysts and realistic electrodes. For electrocatalysts, the selectivity can be determined by the molar fraction of H₂O₂ in the ORR product, which is obtained from a rotating ring-disk electrode (RRDE) setup. Conversely, for realistic electrodes, the percentage of charge transferred toward H₂O₂ production in bulk electrolysis provides the selectivity data, specifically the Faradaic efficiency (FE)^15,16. Theoretically, the intrinsic RRDE selectivity may always outperform the extrinsic FE, as parasitic hydrogen evolution and H₂O₂ decomposition (via cathodic reduction, bulk disproportionation, and anodic oxidation) are not taken into account in the former^16,17. Nevertheless, counterintuitive phenomena have been recently reported for 2e⁻ ORR-oriented GDEs with tuned or optimized interface/structural properties. For example, interface hydrophobic engineering of carbon black (CB)-based GDE can significantly improve the FE to ~100% by creating an O₂-rich microenvironment, while the RRDE selectivity of CB was only ~80%^18,19. Similar electrode FE far exceeding RRDE selectivity can occur even in open air-fed H₂O₂ production scenarios, as long as conventional planar electrocatalyst films are subjected to spatially discontinuous modulation². These contradictions also raise another significant issue concerning the vast difference in sensitivity of selectivity to potential at catalyst- and electrode-scale. The consistently excellent selectivity of electrocatalysts over a wide potential window demonstrated in RRDE can only be maintained under benchtop conditions for electrode-scale bulk electrolysis (a few to several tens of mA cm⁻²). Deteriorated selectivity of O₂-to-H₂O₂ conversion often occurs upon operation at industrially relevant current densities (>100 mA cm⁻²)^1,20,21. Taken together, the above apparent inconsistencies underscore the profound fact that the core factors governing H₂O₂ selectivity may differ greatly between catalyst- and electrode-scale.

For phase-transition-involving 2e⁻ ORR, besides the extensively explored microscale proton-coupled electron transfer^1,13, mesoscale species transport events, including the transfer-adsorption of gaseous reactants, the desorption-diffusion of liquid products, and the penetration-wetting of electrolytes into the electrode pores, are equally critical for its high-rate implementation. In the case of bench-scale experiments (e.g., RRDE tests and low-current operation), the gas/liquid mass transfer and catalytic conversion behaviors with different process kinetics and response times are inherently easily matched, given the overall low electron flux and sluggish electron transfer at the interface. The priority factor governing H₂O₂ selectivity in this scenario is therefore identified as the intrinsic reactivity determined by the electronic structural features of the electrocatalyst^21,22,23,24. However, once transferred to electrode-scale high-rate electrolysis, the greatly accelerated electron transfer in the strong electric environment is no longer compatible with the suboptimal species transfer process. This can lead to the increase of H₂O₂ current density being intimately related to the mass transport behavior of key species manipulated by the interfacial/architectural properties of GDEs^5,7. Several relevant indications in the referenceable CO₂ reduction reaction (CO₂RR) community well illustrate this conjecture. For example, controllable tailoring of wettability, catalyst loading, and ionomer content of GDEs enables fine-tuning of the multi-carbon product distribution pattern and even spatial variations in selectivity within the catalyst layer (CL), originating from local concentration gradients defined by mass transfer fluctuations of reactants^25,26. However, despite this crucial role, mechanistic insights into the diffusion-dependent apparent selectivity for realistic electrodes in 2e⁻ ORR remains under-discussed and poorly understood. Especially in the context of the chemical instability of H₂O₂ on the product side compared to multi-carbon compounds, which further obscures this interaction and makes it even more elusive.

Here, we elucidate that the transport behavior of kinetically relevant species, including both reactants and products, can significantly manipulate electrode FE during bulk electrolysis, particularly for the more diffusion-limited open air-fed systems studied as a model in this work. This effect is independent of the intrinsic reactivity of the electrocatalyst and tends to become apparent when operating in an industry-relevant current density regime where reaction rates surpass diffusion rates (Fig. 1a). We initiated our investigation using commercial CB as the electrocatalyst, and employed various interfacial/architectural properties of GDEs (wettability and CL thickness), which can effectively tailor the species transport behavior, as probe factors. By combining short/long-term electrolysis, in-situ reflection-absorption imaging, and theoretical calculations, we comprehensively investigated the mechanistic origins of diffusion-dependent H₂O₂ selectivity at the electrode-scale.

Fig. 1: Wettability-controlled H₂O₂ production.

a Schematic of species transfer governing H₂O₂ selectivity at the electrode scale. Red and white spheres represent oxygen and hydrogen, respectively. b Schematic of the preparation of GDE with thin catalyst layer (CL) by spray-coating method (MPL: microporous layer; CC: carbon cloth). Cross-sectional SEM images of GDEs using c Nafion or d PTFE as the binder. e Water contact angle evolution of different GDEs under electric field. f H₂O₂ yield and corresponding FE of spraying-made hydrophilic (CB/Nafon-thin) and hydrophobic (CB/PTFE-thin) GDEs during 10 cycles (1 h per cycle) of continuous operation at a current density of 30 mA cm⁻². g H₂O₂ production performance of CB/Nafon-thin and CB/PTFE-thin at different current densities in 1 h. Source data are provided as a Source Data file.

Results

Wettability-controlled H₂O₂ production

Aiming at universality, unmodified commercial CB was selected as a model electrocatalyst and its intrinsic ORR reactivity was first characterized by means of RRDE, showing a moderate H₂O₂ selectivity of 76.3% on average in the studied potential range (0.47 to 0.07 V vs. RHE, Supplementary Fig. 1). The GDE was assembled by spray-casting catalyst ink (CB + binder) on the surface of a commercial gas diffusion layer (GDL) composed of a microporous layer (MPL) and a carbon cloth (CC) substrate (Supplementary Figs. 2 and 3), as schematically shown in Fig. 1b. Starting with interfacial wettability—which can significantly affect the transport behavior of key species²⁷—as a probe factor, we strive to explore the essential relation between mass transfer and electrode-scale H₂O₂ selectivity. To modulate the wettability of the CL, two types of polymers, perfluorinated sulfonic acid ionomer (Nafion) and polytetrafluoroethylene (PTFE), were used as binders in the ink, with the former containing both hydrophilic (–SO₃⁻) and hydrophobic (–CF₂) groups and the latter containing only hydrophobic (–CF₂) groups (see Supplementary Fig. 4 for the specific chemical structure). Nafion- and PTFE-assembled GDEs are hereafter denoted as CB/Nafion-thin and CB/PTFE-thin, respectively.

The physicochemical properties of GDEs constructed with different binders were examined, revealing that CB/Nafion-thin and CB/PTFE-thin experienced no significant differences in terms of the morphology, pore characteristics, surface element chemical state, and phase structure of the CL (Supplementary Figs. 5–8). The cross-sectional scanning electron microscopy (SEM) images in Fig. 1c, d depict a slight increase in the thickness of the CL for CB/PTFE-thin (20.0 μm) compared to CB/Nafion-thin (17.2 μm), possibly because the higher calcination temperature required in the PTFE case results in wider microcracks within its CL (11.5 μm, Supplementary Fig. 7c) than in the Nafion case (5.3 μm, Supplementary Fig. 7a)²⁸. Although we previously reported the strengthening effect of cracks in the CL of GDE on 2e⁻ ORR², the width of <15 μm here is obviously not within the effective range for activating this effect (at least >100 μm). Regarding the wetting properties, it is important to note that, unlike common heterogeneous catalytic processes, the electrode surface in electrocatalytic processes always tends to be invaded by liquid due to the electrowetting and electrowetting-accelerated salt deposition^2,19,29. Only the interface that stabilizes after a certain electrolysis period can reflect the real wettability of the electrode material². Following this principle, we conducted contact angle (CA) measurements of different electrodes under electric fields. It can be seen from Fig. 1e that after 15 min of electrolysis, the CA of CB/Nafion-thin dropped significantly from the initial 102.1° to 47.9° and then stabilized, while CB/PTFE-thin remained constant hydrophobicity with an insignificant change in CA (141.5° to 128.0°). As such, we define Nafion and PTFE-built GDEs as hydrophilic and hydrophobic categories, respectively³⁰. In view of the above, it can be rationally concluded that wettability should be an isolated variable introduced by changing the binder during GDE preparation. This renders the subsequent investigation of the correlation between H₂O₂ selectivity and wettability reliable.

We then focused on bulk electrolysis using a custom-made undivided cell, which allows the GDL side to be directly exposed to open air (Supplementary Figs. 9 and 10), thus enabling investigation of O₂-to-H₂O₂ performance in air-fed scenarios without aeration facilities. Note that under this condition, our GDE can also be called air self-diffusion electrode³¹ or natural air diffusion electrode¹² proposed in some other reports. Considering the instability of wettability in the early stage of electrolysis, the performance evaluation of GDEs here was performed in 10 consecutive cycles (1 h per cycle) without any intermittent drying treatment. In addition, to prevent the poorly water-repellent GDL-induced rapid flooding from obscuring the understanding of wettability tuning selectivity (Supplementary Fig. 11), we performed additional PTFE hydrophobization modification on the GDL, increasing its CA from 126.0° to 147.2° (Supplementary Fig. 12). In this case, at a moderate current density of 30 mA cm⁻², the Nafion-bonded hydrophilic CB/Nafion-thin can deliver a stable H₂O₂ production similar to that of the PTFE-bonded hydrophobic CB/PTFE-thin during 10 h of continuous operation (Fig. 1f). Notwithstanding, the performance comparison reveals that the FE of CB/Nafion-thin is consistently slightly inferior (~2%) to that of CB/PTFE-thin, despite using the same electrocatalyst. Importantly, this inferiority is observed to become more significant with increasing operating current density. As depicted in Fig. 1g, CB/Nafion-thin demonstrates a pronounced performance degradation in terms of FE compared to CB/PTFE-thin when operating at current densities exceeding 30 mA cm⁻². Especially for industrially relevant electrolysis conditions (>100 mA cm⁻²), the FE of CB/PTFE-thin can be up to 72.7% higher than that of CB/Nafion-thin. These facts suggest the occurrence of electrocatalyst-independent events that can tune electrode selectivity.

Species diffusion behavior tunes electrode FE

The electrochemical properties were characterized to elucidate this wettability-modulated H₂O₂ production. As revealed by the linear sweep voltammetry (LSV) curve in Fig. 2a, CB/Nafion-thin demonstrates a consistently slightly higher ORR current density than CB/PTFE-thin at potentials ranging from 0.6 to 0.0 V vs. RHE, which most likely stems from the fact that the hydrophilicity imparted by Nafion in CB/Nafion-thin enhances water availability and ion transport around the active centers³². This can be evidenced by its more active hydrogen evolution reaction (HER) over the entire examined potential range in N₂-saturated LSV (Fig. 2a) and lower interfacial charge transfer resistance in electrochemical impedance spectroscopy (Fig. 2b) compared to CB/PTFE-thin. Therefore, in view of the insignificant variation in the physicochemical properties of the electrodes during the wettability modulation process and the contradiction between H₂O₂ productivity and electrochemical properties, it can be postulated that the difference in 2e⁻ ORR selectivity between CB/Nafion-thin and CB/PTFE-thin should be intimately related to the wettability of the CL. For the strong diffusion-limiting open air-feeding system studied here, efficient delivery of O₂ to the active centers of the electrode is crucial for the 2e⁻ ORR, which is fundamentally determined by the mass transfer efficiency of O₂ over the wettability-controlled interface structure^7,33. The hydrophobic CL is expected to enable stronger O₂ affinity compared to the hydrophilic counterpart, as demonstrated by adhesion measurements of different surfaces to water and gas. From Fig. 2c, CB/PTFE-thin exhibits only about half the adhesion force to water droplets in the air as that of CB/Nafion-thin, indicating weaker confinement of local liquid water (equivalent to stronger aerophilicity). Consistently, the underwater O₂ bubbles burst immediately once approaching the surface of CB/PTFE-thin, whereas they can be readily pinned on the surface of CB/Nafion-thin (Supplementary Movies 1 and 2). With this aerophilicity, more three-phase interfaces (TPI) with high reactivity can be formed on the electrolyte-facing side of CB/PTFE-thin, favoring H₂O₂ generation².

Fig. 2: Species diffusion behavior tunes electrode FE.

a LSV curves (without iR correction, R = 14.6 ± 0.4 Ω) and b Nyquist plots for different GDEs. Inset: equivalent circuit for EIS fitting. c Water adhesive force measurements in air for different GDEs. The inset shows the water contact angles of different GDEs. d O₂ transport behavior and reaction process in the CL of hydrophilic CB/Nafon-thin (left) and hydrophobic CB/PTFE-thin (right). e Schematic of the experimental setup of the in-situ reflection-absorption imaging (RAI) system. f Mapping of H₂O₂ concentration on the surface of hydrophilic CB/Nafion-thin and hydrophobic CB/Nafion-thin at different current densities measured by in-situ RAI system. Source data are provided as a Source Data file.

To explain the amplification of wettability-dependent electrode FE with increasing operating current density, we calculated the O₂ consumption rate at the active interface based on the Faraday’s law of electrolysis and plotted it versus current density (Supplementary Fig. 13). Obviously, the FE fluctuation of GDE during H₂O₂ production strongly depends on the O₂ availability at the interface. Significant deterioration in FE will occur when the O₂ supply rate cannot keep up with the operating current density. For example, the O₂ supply rate corresponding to 95% FE at 30 mA cm⁻² can only support 30% FE when switching to ~100 mA cm⁻². Therefore, based on the above results, we can elucidate the mechanism by which wettability affects electrode FE, from the perspective of O₂ transport. The Nafion-bonded CL is likely to be completely wetted by the electrolyte due to hydrophilicity (left side of Fig. 2d), for which the transport of air to the CB surface is mediated by aqueous-phase diffusion in its structure. As for the PTFE-bonded CL, the hydrophobicity can repel the liquid electrolyte and retain the gas within the CL (right side of Fig. 2d), thereby forming a developed gas-phase diffusion network for air transport. When operating at mild current densities (≤30 mA cm⁻²), aqueous- or gas-phase diffusion-controlled air transport within hydrophilic or hydrophobic CLs simultaneously enabled a dynamic equilibrium between O₂ consumption and atmospheric replenishment at the reaction interface, due to an overall low local O₂ demand. However, as the current density increased above 30 mA cm⁻², the sluggish liquid-phase diffusion of air could not keep pace with the rapid O₂ consumption kinetics. This mismatch resulted in a large fraction of CB surface within the hydrophilic CL being exposed to the O₂-depleted electrolyte, thereby promoting undesirable H₂ generation (Fig. 2d)³². In stark contrast, the hydrophobic GDE maintained a satisfactory FE up to 150 mA cm⁻², due to the unobstructed gas phase channels within its CL, which allowed for timely replenishment of consumed O₂ (Fig. 2d)³⁴. This advantage was further supported by the higher ORR current of CB/PTFE-thin compared to CB/Nafion-thin in the potential window more negative than 0.0 V vs. RHE in the LSV curve (Fig. 2a).

Note however that besides the reactant side, the transport behavior of the product is equally critical considering the chemical instability of H₂O₂. Ideally, the generated H₂O₂ should be desorbed immediately and transported rapidly away from the cathode surface to prevent its further electroreduction to H₂O³⁵. According to Fig. 1g, the inherent O₂ capture capacity of CB/Nafion-thin (delivering ~95% FE at 30 mA cm⁻²) can theoretically support at least nearly 30% FE when switched to 100 mA cm⁻² (Supplementary Fig. 13), while the experimental value (13.6%) was found to be less than half of this value. Furthermore, if only O₂ mass transfer limitation dominates, the absolute value of H₂O₂ accumulation concentration at CB/Nafion-thin is unlikely to show a downward trend after 100 mA cm⁻². These facts collectively suggest that the concentration of H₂O₂ detected in the bulk solution is not intrinsic but an apparent value after undergoing reductive decomposition, especially for hydrophilic GDEs at high current densities. In principle, the overreduction of H₂O₂ generally occurs in its two diffusion processes with opposite directions: from the reaction boundary to the bulk solution or from the bulk solution back to the cathode surface². However, accurate discrimination of these two processes remains challenging, given the lack of powerful tools to monitor the real-time distribution of H₂O₂ concentration on the electrode surface (rather than simple bulk concentration measurements). To this end, an in-situ reflection-absorption imaging system (Fig. 2e and Supplementary Figs. 14 and 15) was developed, which can track the spatiotemporal evolution of the concentration of target species generated at the electrode surface by monitoring changes in their light absorption intensity^36,37. We selected potassium titanium oxalate as an indicator for H₂O₂ and performed bulk electrolysis to map the instantaneous H₂O₂ concentration distribution on the surface of CB/Nafion-thin and CB/PTFE-thin at different current densities.

As shown in Fig. 2f, unlike the relatively uniform surface distribution of H₂O₂ concentration in the hydrophilic case, local high H₂O₂ accumulation appeared at the edges of the air films on the hydrophobic GDE surface. This is consistent with the previous discussion that these regions, as TPIs, can promote 2e⁻ ORR conversion due to high O₂ accessibility². Importantly, the evolution trend of the average H₂O₂ production rate on the CB/Nafion-thin and CB/PTFE-thin surfaces with current density (Supplementary Fig. 16) is observed to strictly correspond to the concentration trend detected in the bulk solution (Fig. 1g). For example, the H₂O₂ production rate on the CB/Nafion-thin surface at 100 mA cm⁻² is indeed lower than that at 30 and 50 mA cm⁻². This suggests that at ≥100 mA cm⁻², H₂O₂ generated within the hydrophilic CL of CB/Nafion-thin simultaneously undergoes severe reductive decomposition during its migration to the bulk solution (rather than from the bulk solution back to the cathode). Therefore, the favorable transport of product H₂O₂ is as critical as that of reactant O₂, especially for the hydrophilic GDE with TPI located inside the CL at high-rate electrolysis, which unfortunately has been largely ignored in previous studies.

Sensitivity of diffusion-dependent electrode FE to CL thickness

CL thickness can also profoundly affect species transport behavior²⁶, and was therefore employed as another investigational probe here to examine its sensitivity to transport-dependent electrode FE by varying the CB loading. Of particular note, the spray-coating method described above was no longer applicable, as eye-visible cracks (width >100 μm) progressively developed in the CL structure with increasing CB loading (Supplementary Fig. 17). This was observed to alter the production behavior of H₂O₂ (Supplementary Fig. 18), as also demonstrated in our recent report². To prevent this interference, we instead utilized the rolling technique commonly used in chemical/environmental engineering fields³⁸ to fabricate hydrophilic/hydrophobic GDEs with thick CLs. As schematically shown in Fig. 3a, the rolling-made GDE presents a sandwich-like structure composed of a CB-PTFE GDL, a stainless-steel mesh (SSM) substrate, and a CB-binder CL. Note that since the rolling method can hardly achieve precise gradient modulation of CL thickness, we therefore only constructed GDEs with significantly thicker CLs than those in the spray-coating case (2.83 mg cm⁻²) at a fixed high CB loading of 28.3 mg cm⁻² to highlight their performance differences.

Fig. 3: Sensitivity of diffusion-dependent electrode FE to CL thickness.

a Schematic of the preparation of GDE with thick CL by rolling method. Cross-sectional SEM images of b CB/Nafion-thick and c CB/PTFE-thick. d H₂O₂ yield and e corresponding FE of CB/Nafion-thick and CB/PTFE-thick during 10 cycles (1 h per cycle) of continuous operation at a current density of 30 mA cm⁻². f CB loading and ECSA values of different GDEs prepared by spraying and rolling methods. g Schematic representation of the water microenvironment of GDEs with different wettability and different CL thicknesses. h H₂O₂ production performance of hydrophobic GDEs in 1 h with different CL thicknesses at different current densities. i Schematic of the electrowetting effect. j Comparison of H₂O₂ production performance of CB/PTFE-thin and CB/PTFE-thick during long-term electrolysis at a current density of 30 mA cm⁻². Inset: optical images of the air-facing side of different GDEs. Source data are provided as a Source Data file.

The cross-sectional SEM images in Fig. 3b, c reveal that the GDEs prepared by the rolling method, using different binders, achieved an order of magnitude increase in CL thickness compared to those prepared by the spraying method (Fig. 1c, d). Consequently, the samples with Nafion and PTFE as binders are hereafter referred to as CB/Nafion-thick and CB/PTFE-thick, respectively. No obvious cracks were found in their structure due to the release of internal stresses by the rolling operation (Supplementary Fig. 19). Note that although the two methods inevitably caused slight variations in the pore diameter of the CL due to the different operating pressures, its effect on species transport and electrode performance may be negligible in the current study (see Supplementary Fig. 20 for the reason). Meanwhile, the rolling-made GDE experienced a similar CA evolution (Supplementary Fig. 21) as the spraying-made GDE under an electric field, lending credibility to comparisons of the two types of electrodes in terms of H₂O₂ production, with CL thickness as a variable. The H₂O₂ yield and FE profile of rolling-made GDEs in 10 consecutive cycles are given in Fig. 3d, e, showing a phenomenon consistent with the spraying-made GDEs that hydrophilicity is not as effective as hydrophobicity in H₂O₂ production. However, it is noteworthy that the impact of CL thickness on H₂O₂ production varies greatly under different wettability conditions. At an operating current density of 30 mA cm⁻² (first cycle), an increase in CL thickness resulted in significantly reduced FE under hydrophilic conditions, dropping from 95.1% for CB/Nafion-thin to 48.7% for CB/Nafion-thick. However, under hydrophobic conditions, the FE remained nearly unchanged, with 96.3% for CB/PTFE-thin compared to 96.1% for CB/PTFE-thick.

To unravel the origin of this difference, cyclic voltammetry (CV) was performed to quantify the electrochemically active surface area (ECSA) of different GDEs (Supplementary Fig. 22). This measurement helps reflect the proportion of solid CB within the CL that is wetted and accessible by the liquid electrolyte²⁷. Figure 3f plots the ECSA of GDEs prepared by different methods, from which ECSA is positively correlated with CB loading in the hydrophilic case with Nafion as the binder (no correlation in the hydrophobic case with PTFE as the binder). This can be explained by the fact that increasing the thickness of the hydrophilic CL is accompanied by a nearly proportional expansion of the electrolyte-filled region, while this phenomenon cannot occur in the hydrophobic CL as its solid-liquid interface is always located between the outermost side of the CL and the electrolyte (under no electrolysis) (Fig. 3g). As a result, the increased CL thickness in the hydrophilic case creates a distinct barrier to O₂ transport throughout the layer, which can further limit O₂ availability and thus shift the catalytic selectivity towards increased competition by HER³⁹.

In addition, the expansion of the wetting area in the structure of CL corresponds to a wider distribution of the reactive electron flux⁵. This presumably intensifies the reductive decomposition of H₂O₂ during its diffusion in the thicker solid-liquid layer inside the CL of CB/Nafion-thick at 30 mA cm⁻² (similar to CB/Nafion-thin operated at 100 mA cm⁻² in the previous section). Even at a lower current density (10 mA cm⁻²) with lower O₂ demand, CB/Nafion-thick still demonstrated a significantly lower FE than CB/Nafion-thin (Supplementary Fig. 23), indicating that the unfavorable transport of the product H₂O₂ is indeed amplified in the GDE with a thick hydrophilic CL. Compared with CB/Nafion-thin, the accelerated decomposition of H₂O₂ induced by CB/Nafion-thick at 30 mA cm⁻² can further support this conclusion (Supplementary Fig. 24). For the hydrophobic case, although increasing the CL thickness also limits the gas-phase diffusivity of O₂ towards the outermost side of CL as the active interface due to the elongation of the diffusion path, this adverse effect becomes apparent and can be reflected in the H₂O₂ production process only when operating at current densities exceeding 30 mA cm⁻² (Fig. 3h). In other words, under mild electrolysis scenarios (≤30 mA cm⁻²), the omnidirectional hydrophobicity of this CB-PTFE film enables the timely supply of required O₂ at the interface within a film thickness of 158.3 μm herein. In addition, mass transfer limitation of H₂O₂ is also unlikely to occur due to the unchanged TPI position.

Note that the effect of CL thickness on H₂O₂ production of the hydrophobic electrode at 30 mA cm⁻² becomes discernible as electrolysis proceeds to 10 h, with the FE gradually declining for CB/PTFE-thick (Fig. 3d, e) but remaining relatively stable for CB/PTFE-thin (Fig. 1f). We have previously demonstrated that hydrophobicity degradation caused by the electrowetting (Fig. 3i), can significantly impact 2e⁻ ORR stability². However, the progressive thinning and eventual disappearance of the gas film (i.e., hydrophobicity) at the electrode-electrolyte interface induced by electrowetting occurred on both CB/PTFE-thin and CB/PTFE-thick (Supplementary Fig. 25). We therefore consider that the above phenomenon should be related to the difference in microenvironment evolution caused by the interaction between CL thickness and electrowetting. In view of this, a longer period of 250 h of continuous electrolysis was performed to highlight this difference (Fig. 3j and Supplementary Fig. 26). As expected, a more pronounced attenuation trend is revealed for CB/PTFE-thick, with its FE monotonically decreasing from 96.1% in the first cycle to 36.9% after 250 h of continuous operation. Interestingly, CB/PTFE-thin delivered relatively stable H₂O₂ production but experienced rapid failure within a few hours in the later stage of 250 h electrolysis, probably due to the sudden complete wetting of the GDL (inset of Fig. 3j) induced by the severe HER initiated on its air-facing side (see Supplementary Figs. 27–29 for detailed discussion).

Electrowetting-induced evolution of microenvironment and species transport

In-situ ECSA measurements were conducted to real-time monitor the electrowetting-driven electrolyte penetration behavior into the hydrophobic pores of different GDEs during long-term electrolysis (Supplementary Figs. 30 and 31). It was noticed that the increase in double-layer capacitance (C_dl) of CB/PTFE-thin plateaued after about 10 to 50 h of electrolysis, indicating that its CL was almost fully wetted by the electrolyte (Fig. 4a). With further electrolysis extending to 80 h, the “electrolyte perspiration” phenomenon that observed on the air-facing side of CB/PTFE-thin (inset of Fig. 3j) suggests that the liquid has even partially penetrated the GDL. For CB/PTFE-thick, however, a steady increase in C_dl was found within 200 h due to the increased CL and GDL thickness (Fig. 4b), for which “electrolyte perspiration” did not occur until 200 h of electrolysis (inset of Fig. 3j). Also importantly, CB/PTFE-thick demonstrated a consistently higher C_dl than CB/PTFE-thin under the same electrolysis conditions. This is because the denser pore structure (i.e., smaller pore size, Supplementary Fig. 20) within the CL of CB/PTFE-thick can lead to greater liquid penetration depth by inducing stronger capillary forces⁴⁰, once the interface wettability is modulated from hydrophobic to hydrophilic by the electric field. As a result, the gradual advancement of the electrolyte in the CL will switch the local O₂ transport from gas-phase diffusion to liquid-phase diffusion. Note that the overall low CL thickness of CB/PTFE-thin seems to hardly limit the O₂ supply even if complete wetting occurs (≤30 mA cm⁻²), as evidenced by the insignificant difference in H₂O₂ production of hydrophilic CB/Nafion-thin compared to hydrophobic CB/PTFE-thin with similar CL thickness at 30 mA cm⁻² (Fig. 1f). However, the higher liquid seepage depth in the CL of CB/PTFE-thick will cause a portion of CB to be largely devoid of O₂ and this trend will become obvious as the electrolysis proceeds, partially explaining its declining H₂O₂ accumulation rate in Fig. 3j.

Fig. 4: Electrowetting-induced evolution of microenvironment and species transport.

Specific double layer capacitance of a CB/PTFE-thin and b CB/PTFE-thick after electrolysis at 30 mA cm⁻² for 0, 5, 10, 50, 100, and 200 h. c Comparison of electrochemical H₂O₂ dissociation performance of fresh/wetted CB/PTFE-thin and CB/PTFE-thick GDEs. d Through-plane transport behavior of reactants and products within the CL of mainstream GDE (left) and conceived GDE with an ideal architecture (right) during the 2e⁻ ORR process. e Evolution of the mass transfer coefficient of O₂ within the active layer during electrolysis quantified by molecular dynamics (MD) simulations (MSD: mean-squared displacement). Source data are provided in Supplementary Data 1.

In addition, the gradual penetration of electrolyte into the CL would also form an evolving barrier to the transfer of H₂O₂, thereby aggravating its decomposition (just like increasing the thickness of the hydrophilic CL). To examine this effect, we conducted a comparative study on the H₂O₂ decomposition performance of two hydrophobic electrodes before and after 250 h of electrolysis (Fig. 4c). Apparently, compared with the respective fresh cases, the wetted CB/PTFE-thin showed almost unchanged H₂O₂ decomposition, while the wetted CB/PTFE-thick demonstrated a significantly accelerated H₂O₂ electroreduction process. This observation is consistent with the FE evolution of the two GDEs in Fig. 3j. To rationalize this discrepancy, we comprehensively analyzed the through-plane transport behavior of reactants (O₂ and H₂O as a proton source) and products (H₂O₂) within the CL, which is essentially modulated by electric field-driven variations in the wettability microenvironment.

Figure 4d (left) shows a schematic of the mainstream GDE utilized in this work and the transport directions of different species as well as their concentration as a function of distance from the substrate. At the beginning of electrolysis, the CB at the CL-electrolyte interface is exposed to an O₂-rich and H₂O-deficient microenvironment created by hydrophobicity, and both the synthesis and getaway of H₂O₂ products can therefore be facilitated at this active three-phase boundary with optimal gas/liquid equilibrium. However, as the electrolysis proceeds and the accompanying solid-liquid boundary migrates deeper into the CL, unfavorable mass transfer of O₂ and H₂O₂ will be induced simultaneously. The real-time evolution of the mass transfer coefficients of these species quantified by molecular dynamics (MD) simulations can well illustrate this effect (see Supplementary Figs. 32–34 for simulation details and reliability analysis). When the electrolyte advances to one-third or two-thirds of the thickness of the entire CL, the mass transfer coefficient of O₂ inside the layer decreases by 26.0% (2.05 × 10⁻⁴ cm² s⁻¹) or 42.6% (1.59 × 10⁻⁴ cm² s⁻¹), respectively, from 2.77 × 10⁻⁴ cm² s⁻¹ before electrolysis (Fig. 4e). Similar attenuation is even more severe for the product H₂O₂ under the same conditions, with its mass transfer coefficient decreasing by 43.3% (2.19 × 10⁻⁵ cm² s⁻¹) or 73.1% (1.04 × 10⁻⁵ cm² s⁻¹), respectively, from the initial 3.86 × 10⁻⁵ cm² s⁻¹ (Supplementary Figs. 33 and 34), due to the extended tortuous diffusion path. This diffusion limitation implies a prolonged residence time of H₂O₂ in the wetted region of the CL, where the lack of O₂ together with the abundance of electrons and protons increases the likelihood of H₂O₂ reduction reactions². And this adverse effect will become more pronounced for GDEs with thicker CLs, especially CB/PTFE-thick with a CL thickness of 158.3 μm. However, the 20.0 μm thin CL of CB/PTFE-thin appears unable to activate this effect yet unless the current density is significantly increased, consistent with the phenomenon observed in Fig. 3j.

The above analysis also provides inspiration for conceiving GDE configurations with ideal architectures adapted to 2e⁻ ORR characteristics. Briefly, the CL should perhaps evolve toward ultra-thinness, preferably the extreme case of a single-layer electrocatalyst with optimal material utilization (right side of Fig. 4d). In this configuration, the entire structure of CL can always exhibit the highest proximity to the open-air feed, even when encountering severe electrowetting, meaning that the H₂O₂ production process is no longer limited by gas mass transfer. More intriguingly, the undesired H₂O₂ reduction process is also extremely suppressed, as the generated H₂O₂ can diffuse directly into the bulk solution without traversing the high electron/proton region that favors H₂O₂ reduction. Note, however, that despite these attractive advantages, such configurations simultaneously pose a significant challenge for the GDL in terms of waterproof performance, especially under industrial-relevant harsh electrolysis. For example, in our study, in stark contrast to the relatively stable H₂O₂ production at 30 mA cm⁻² (Fig. 3j), CB/PTFE-thin demonstrated a rapidly decaying FE when switched to 100 mA cm⁻² (Supplementary Fig. 27), due to severe flooding caused by the deteriorated hydrophobicity of the GDL. Although commercial PTFE membranes are being actively explored in the CO₂ reduction reaction (CO₂RR) community as an alternative GDL, they continue to suffer from issues related to poor electrical conductivity and mechanical instability^32,34. In this context, rational architectural innovations that enable thick hydrophobic GDEs to serve for long periods without experiencing selectivity degradation caused by species mass transfer limitations (as encountered with CB/PTFE-thick), may also be desirable. In any case, more extensive and thorough future research efforts to overcome the selectivity-stability dilemma of GDEs for electrolyte-involved electroreduction communities, such as 2e⁻ ORR, are urgently needed.

Molecular-level understanding of diffusion-dependent electrode FE

So far, we have focused exclusively on uncovering the central role of macroscopic mass transfer behavior of key species in governing electrode FE during bulk electrolysis. In fact, diffusion at the molecular level can also significantly alter reaction selectivity, which however has been largely overlooked. Therefore, we next resort to computational tools to better illustrate and understand this phenomenon. Combining the categories of reactive oxygen groups identified in previous studies⁴¹ and XPS analyses of CB-based GDEs here (Supplementary Fig. 8), we constructed a graphene layer with ether groups (C–O–C, the most reactive site toward 2e⁻ ORR) on the basal plane as a representative model (Supplementary Fig. 35). Figure 5a depicts the general schematic of the ORR process at this active site: with *OOH as a common reaction intermediate, ORR can proceed through either a 2e⁻ pathway to produce H₂O₂ by breaking the *–O band (first row), or a 4e⁻ pathway to directly yield H₂O by breaking the O–O band (second row)⁴². However, the product H₂O₂ in 2e⁻ ORR can also undergo further dissociation to form H₂O, making the process an indirect 4e⁻ ORR. Furthermore, this step is thermodynamically spontaneous, as evidenced by its strongly exothermic nature in the free energy profiles of reaction intermediates in the corresponding pathway (Supplementary Fig. 36). This aligns precisely with the above-observed preference of the wetted CB/PTFE-thick for H₂O₂ decomposition in bulk electrolysis. Therefore, creating a favorable microenvironment that promotes microscale migration of H₂O₂ away from the active centers is also crucial, which fundamentally determines the concentration of H₂O₂ detected in the bulk solution and subsequent calculation of the electrode FE.

Fig. 5: Molecular-level understanding of diffusion-dependent electrode FE.

a Schematic of ORR along the 2e⁻ or 4e⁻ pathway on the ether group. Blue, red, and white spheres represent carbon, oxygen, and hydrogen, respectively. Mapping of surface absorbed light intensity (representing OH⁻ concentration) as a function of electrolysis time for (b) hydrophilic CB/Nafion-thin and c hydrophobic CB/PTFE-thin at 100 mA cm⁻². d Ab initio molecular dynamics (AIMD) simulated motion trajectory of H₂O₂/HO₂⁻ molecules along the z-axis when different numbers of OH⁻ are placed at the interface (snapshots of the final state structures). e Relative concentration distribution of H₂O₂/HO₂⁻ molecules near the interface under different alkaline conditions simulated by MD. f LSV curves (without iR correction) in neutral (R = 14.6 ± 0.4 Ω) or alkaline electrolytes (R = 11.6 ± 0.1 Ω) with/without H₂O₂ using CB/Nafion-thin as a representative. g Calculated adsorption energies of O₂ and HO₂⁻ at ether groups in a local alkaline environment. h In-situ ATR-SEIRAS spectra of CB electrocatalyst at various potentials (OCP: open circuit potential) under hydrophobic conditions (PTFE binder). Source data are provided as a Source Data file.

Recently, the local alkalization effect at GDE interface has drawn significant attention in the field of gas-consuming electroreduction due to its ability to deeply affect reaction pathways while mediating the transport of reaction species^18,34,43. For example, the increase in local pH on or near the electrode surface during the CO₂RR process favors the dimerization of *CO by strengthening the dipole attraction between neighboring *CO molecules, thus leading to the selective conversion of CO₂ into C₂ products³⁴. For 2e⁻ ORR, it is emphasized that local alkalinization can induce the deprotonation of H₂O₂ to form HO₂⁻, and then its immediate separation from the reaction interface occurs due to cathodic repulsive interactions¹⁸. However, for the system studied here, given that different cathode surfaces produce nearly identical repulsive forces at the same potential, we shifted our focus to examining the generation rates of hydroxide (OH⁻) ions on different electrode surfaces to explain the differences in their H₂O₂ production performance. To this end, the same in-situ reflection-absorption imaging system (Fig. 2e) was employed with phenolphthalein as an indicator for OH⁻, to monitor and compare the local alkalinization at the interface of hydrophilic and hydrophobic GDEs in a real-time manner.

As shown in Fig. 5b, c, both hydrophilic and hydrophobic GDEs demonstrate rapid accumulation of OH⁻ on their surfaces, particularly within the first 3 s of the reaction, resulting from the local consumption rate of protons via 2e⁻ ORR exceeding their mass transfer from the bulk. However, it was also noted that the trend of increasing absorbed light intensity on the CB/PTFE-thin surface was significantly stronger and faster than that on the CB/Nafion-thin surface. The average light intensity in these two-dimensional imaging was 2.5 times higher at CB/PTFE-thin than at CB/Nafion-thin (Supplementary Fig. 37). This indicates that the OH⁻ concentration generated on the surface of CB/PTFE-thin exceeds that of CB/Nafion-thin by 2.5 folds, because of the insufficient contact between the bulk and local electrolyte caused by hydrophobicity⁴⁴. Similar phenomena were also observed on CB/Nafion-thick and CB/PTFE-thick with thicker CL (Supplementary Figs. 38 and 39). We believe that this significant difference in OH⁻ concentration at the interface will greatly change the migration behavior of H₂O₂ molecules. Once H₂O₂ desorbs from the active centers, its deprotonation to HO₂⁻ occurs rapidly through interactions with the surrounding abundant OH⁻. Thereafter, the electrostatic repulsive interactions conferred by these OH⁻ are expected to enable pushing HO₂⁻ away from the active centers, with the degree of this effect being controlled by the OH⁻ concentration.

To evaluate the difference in this electrostatic repulsion effect on the two GDEs, we conducted ab initio molecular dynamics (AIMD) simulations based on an explicit solvation model. Neutral and various alkaline reaction microenvironments were modeled by placing different numbers of OH⁻ ions (0, 2, and 4) on the surface of a graphene layer featuring a C–O–C group. Figure 5d and Supplementary Figs. 40–42 display the calculated results of the displacement distance of H₂O₂/HO₂^– molecules and the corresponding snapshots of the initial and final state structures. In the absence of OH^– ions on the surface, the H₂O₂ molecules drift inconspicuously along the z-axis, about 0.39 Å after 2.8 ps. However, once OH⁻ ions are introduced, HO₂⁻ molecules can be quickly repelled away from the initial position. And the repulsion distance tends to increase with the increase of OH⁻ concentration, from 1.36 Å at 2 OH⁻ to 1.86 Å at 4 OH⁻. Similar phenomena were also observed in the relative concentration distribution of H₂O₂/HO₂⁻ molecules near the reaction interface obtained by MD simulation (Fig. 5e and Supplementary Fig. 43). These results clearly demonstrate that the generated OH⁻ can strongly repel local HO₂⁻ through electrostatic interactions, thereby significantly reducing the likelihood of its further dissociation into H₂O. To validate the computed results experimentally, we also comparatively studied the electrochemical reduction of H₂O₂ in neutral and alkaline electrolytes (Fig. 5f). Interestingly, an opposite change pattern for the LSV response current was observed in neutral (increased) and alkaline (decreased) electrolytes when H₂O₂ was introduced, implying that the presence of OH⁻ effectively suppressed the electroreduction of H₂O₂ at the cathode. This is essentially in line with recent computational studies that reported a significant increase in the H₂O₂ decomposition energy barrier with increasing solution alkalinity³⁵. Therefore, the superior FE of hydrophobic GDE over hydrophilic GDE can be attributed to the former’s ability to promote product diffusion by inducing stronger local alkalinization—a phenomenon that, to the best of our knowledge, has rarely been discussed in the literature before.

Note that from a reaction kinetics perspective, the swift escape of the product positively shifts the reaction equilibrium. As shown by density functional theory (DFT) calculations, the adsorption of O₂ at the active site is thermodynamically more favorable than that of HO₂⁻ by 0.13 eV (Fig. 5g). This preferential adsorption of O₂, coupled with the previously demonstrated robust O₂ capture capability endowed by hydrophobicity, allows O₂ reactants to quickly refill into the active centers exposed once the product has drifted away, thereby facilitating the formation and stabilization of key intermediates (*O₂, *OOH, and *H₂O₂)²³. In-situ attenuated total reflectance surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) analysis provides reliable evidence for this speculation. As shown in Fig. 5h (hydrophobic case), the intensity of the absorption band corresponding to *OOH (~1258 cm⁻¹)⁴⁵ gradually increases with the negative shift of the applied potential. However, for the hydrophilic case (Supplementary Fig. 44), this enhancement stagnates at −0.5 V vs. RHE, and the band intensity even begins to decrease after −0.7 V vs. RHE. In addition, the hydrophobic case also exhibits more distinct and identifiable absorption bands that can be assigned to *O₂ (~1481 cm⁻¹) and *H₂O₂ (~1406 cm⁻¹)^45,46, compared with the hydrophilic case. Accordingly, this favorable reaction microenvironment can greatly improve the utilization efficiency and turnover frequency of reaction sites, explaining the previously observed anomaly that the electrode FE is higher than the intrinsic selectivity of electrocatalyst CB before exceeding a specific current density threshold, especially for the hydrophobic case (Fig. 1g).

The above discussion may raise an interesting issue regarding the priority between hydrophobicity accelerating O₂ diffusion and local alkalization optimizing H₂O₂ transfer. We think that ensuring intrinsically high H₂O₂ generation rates by improving O₂ mass transfer may be a key prerequisite for the subsequent enhancement of H₂O₂ accumulation concentration in the bulk solution by creating a local alkaline microenvironment. For example, monotonically increasing current density, despite intensifying the local alkalinization effect, cannot always favor H₂O₂ production due to eventual limitations in O₂ mass transfer. In fact, this phenomenon may already occur at several hundreds of mA cm⁻² (the specific value depends on the CL thickness, Fig. 3h) for the open air-fed system studied here. Thus, this deficiency, together with the aforementioned electrowetting-induced species transport degradation, suggests that the current mainstream strategy of merely modulating the hydrophobicity of the CL is clearly insufficient. Future architectural innovations for hydrophobic GDEs that can completely circumvent the O₂ supply limitation (e.g., tailoring of ultra-large artificial gas channels) while enhancing local alkalinization (e.g., construction of high-curvature micro/nanostructures) are urgently needed, to enable the synchronous optimization of reactant and product transport for long-period GDE service at ampere-level current densities. These efforts may constitute critical milestones for the commercialization of open air-fed H₂O₂ electrosynthesis technology.

Discussion

In summary, we underscore the pivotal role of electrocatalyst-independent mass transfer behavior of key species, including reactants and products, in governing electrode selectivity toward open air-fed H₂O₂ electrosynthesis. Utilizing wettability and CL thickness as probe factors, we discover that efficient gas-phase diffusion induced by hydrophobicity can promptly replenish the O₂ consumed at the interface, aligning with the electron flux and thereby facilitating relatively stable H₂O₂ production up to 150 mA cm⁻² (only 50 mA cm⁻² for the hydrophilic counterpart). On the product side, hydrophobicity fosters a favorable local alkaline microenvironment that propels H₂O₂ away from the active site by leveraging electrostatic repulsion. This not only mitigates the risk of H₂O₂ overreduction but also promotes a positive shift of the reaction equilibrium, achieving electrode selectivity that even surpasses the intrinsic selectivity of the electrocatalyst. We also demonstrate that this beneficial transport of reactants and products is severely disrupted as electrowetting-driven liquids infiltrate the hydrophobic regions, leading to a continual decline in apparent selectivity, especially for those GDEs with thicker CLs. Future research should therefore concentrate on architectural innovations for hydrophobic GDEs with sustained optimal mass transfer of O₂ and H₂O₂, aiming for long-term H₂O₂ production process with ultrahigh rate (multi-amperes per square centimeter) and propelling the commercialization of 2e⁻ ORR.

Methods

Chemicals and materials

Carbon black (CB) (Vulcan XC-72) and polytetrafluoroethylene (PTFE) (60%) were supplied by Shanghai Aladdin Biological Technology Co., Ltd. Nafion ionomer solution (5 wt%) was purchased from Sigma-Aldrich. W1S1011 GDL was purchased from CeTech. SSM was provided by Kunshan TengErHui Electronic Technology Co., Ltd. Anhydrous sodium sulfate (Na₂SO₄, 99%), sodium hydroxide (NaOH, 95%), and potassium titanium (IV) oxalate (K₂TiO(C₂O₄)₂, ≥98%) were purchased from Shanghai Macklin Biochemical Co., Ltd. All the chemicals were of analytical grade unless otherwise specified.

Fabrication of working electrodes

The GDE used in this work was fabricated by depositing CB electrocatalyst on the surface of W1S1011 or SSM substrate via different techniques, as detailed below.

Spraying method: 20 mg of CB particles were ultrasonically dispersed in anhydrous ethanol for 30 min, and then a certain amount of Nafion or PTFE binder was dropped into the solution (at a CB-to-binder mass ratio of 5:3) and subjected to additional ultrasonic treatment for 30 min. The as-prepared catalyst ink was transferred to an airbrush gun and sprayed evenly on the microporous layer (MPL) side of W1S1011 GDL with a diameter of 3 cm (CB loading 2.83 mg cm⁻²) using N₂ gas at 0.6 bar, followed by calcination for 30 min (120 °C for Nafion and 360 °C for PTFE) to obtain working electrodes with different binders. Nafion- and PTFE-assembled GDEs are denoted as CB/Nafion-thin and CB/PTFE-thin, respectively, due to their overall thin catalyst layer (CL). In addition, GDL was hydrophobically modified to improve the waterproof performance by soaking it in the PTFE solution (10 wt%) for 30 min. The catalyst ink was then deposited on hydrophobic GDL via the same spraying step described above. Note that the GDEs assembled with Nafion and PTFE were also marked as CB/Nafion-thin and CB/PTFE-thin, respectively, because their CL structures are the same as those of the above GDEs.

Rolling method: the rolling-made GDEs used in this work were all sandwich-like structures including a CB-binder functional layer, a SSM substrate, and a CB-PTFE waterproof layer (also serving as a GDL). Specifically, 80 mg of CB particles were ultrasonically dispersed in anhydrous ethanol for 30 min, followed by the dropwise addition of PTFE (at a CB-to-PTFE mass ratio of 1:3) and vigorous stirring at 70 °C until a wet paste was formed. The CB-PTFE paste and SSM (diameter 3 cm) were placed in a mold, cold-pressed at 18 MPa for 10 min and calcined at 360 °C for 30 min to obtain GDL. The same procedure was then used to prepare the CL (the mass ratio of CB to Nafion or PTFE was 5:3) on the other side of the SSM with a CB loading of 28.3 mg cm⁻² (i.e., 200 mg of CB), and the sheet was calcined at 120 °C (or 360 °C for the PTFE binder) for 30 min to obtain the final electrode material. The GDEs prepared by this method with Nafion or PTFE as the binder were named CB/Nafion-thick and CB/PTFE-thick, respectively, due to their thicker CL compared to the spraying-made GDEs.

Characterization

SEM and energy-dispersive X-ray spectroscopy mapping were performed on a Hitachi S4800 field emission scanning electron microscope. X-ray photoelectron spectroscopy (XPS) was acquired by a Thermo Scientific K-Alpha Nexsa. X-ray diffraction spectra were collected on an X-ray diffractometer (Rigaku MiniFlex 600) with a Cu Kα radiation (λ = 0.154 nm). The contact angles of different electrodes were recorded using a contact angle instrument (DropMeterA-100P). The adhesion forces between water or oxygen and the surfaces of different electrodes were measured using a tensiometer (Dataphysics DCAT 25, Germany). The pore size distributions of the active layers of different GDEs were measured by N₂ adsorption–desorption isotherms at 77 K on a Micromeritics ASAP 2460, and calculated by the Brunauer–Emmett–Teller (BET) and Barrett–Joyner–Halenda (BJH) methods.

H₂O₂ yield in bulk electrolysis

H₂O₂ electrosynthesis tests were performed in a custom-made undivided electrolyzer containing 100 mL of 0.05 M Na₂SO₄ (pH = 7) as a supporting electrolyte. A Reference 3000 electrochemical workstation (Gamry Instruments, USA) was used to apply the galvanostatic mode to the electrolytic cell for electrolysis experiments. The prepared GDE (30 mm in diameter) and a Pt foil (10 mm × 10 mm × 0.1 mm) were used as cathode and anode, respectively, with a mutual distance of 20 mm. Throughout the electrolysis process, the GDL side of the cathode was exposed to open air without the assistance of any external aeration equipment. Aliquots of the solution (1 mL) were collected from the reactor at pre-established time points and analyzed to determine the H₂O₂ concentration. All error bars represent the standard deviation of three independent measurements.

Quantification of H₂O₂ concentration

The concentration of H₂O₂ was quantified via a UV-vis spectrophotometer (UV759, Shanghai Instrument Analysis Instrument Co., Ltd) at a wavelength of 400 nm using the potassium titanium (IV) oxalate method. Potassium titanium (IV) oxalate can react with H₂O₂ to form a stable yellow-orange titanium (IV)-peroxide complex, with its color depth being proportional to the H₂O₂ concentration¹⁷. The standard curve for H₂O₂ quantification was shown in Supplementary Fig. 10. The Faraday efficiency (FE) for H₂O₂ electrosynthesis was calculated through the following equation:

$${{\rm{FE}}}\left(\%\right)=\frac{{nFCV}}{{It}}\times 100\%$$

(1)

Where n is the number of electrons transferred for O₂ reduction to H₂O₂ (n = 2), F is the Faraday constant (96486 C mol⁻¹), C is the concentration of H₂O₂ (mol L⁻¹), V is the volume of electrolyte (L), I is the current (A), and t is the electrolysis time (s).

Calculation of interface O₂ consumption rate

The consumption rate of O₂ (${C}_{{{\mbox{O}}}_{2}}$, mol s⁻¹) at the reaction interface can be calculated by using the Faraday’s law of electrolysis as follows:

$$[{C}_{{{{\rm{O}}}}_{2}}]_{{{{\rm{H}}}}_{2}{{{\rm{O}}}}_{2}}=\frac{j\eta S}{{nF}}$$

(2)

$$[{C}_{{{{\rm{O}}}}_{2}}]_{{{{\rm{H}}}}_{2}{{\rm{O}}}}=\frac{j\eta S}{{nF}}$$

(3)

Where j is the current density (A cm⁻²), η is the FE for H₂O₂ (or H₂O) production (%), S is the electrode effective area (cm²), n is the number of electrons transferred for O₂ reduction to H₂O₂ (n = 2) or H₂O (n = 4), and F is the Faraday constant (96486 C mol⁻¹).

Electrochemical measurements

The ORR selectivity and activity of electrocatalyst CB were characterized using the rotating ring-disk electrode (RRDE, E7R9) (Pine Instruments, USA) technique. The RRDE assembly included a glassy carbon rotation disk electrode (disk area: 0.2475 cm²) and a Pt ring (ring area: 0.1866 cm²) with a collection efficiency (N) of 0.37. The catalyst ink was obtained by mixing 5 mg of CB, 950 μL of ethanol, and 50 μL of Nafion (5 wt%). After 30 min of ultrasonic treatment, 5.0 μL of catalyst ink was drop-cast onto the disk electrode. Pt sheet and Hg/HgO were used as counter electrode and reference electrode, respectively. Specifically, LSV was performed in an O₂-saturated KOH (0.1 M, pH = 13) solution between −0.1 and 1.1 V vs. RHE, at a scan rate of 10 mV s⁻¹ and a rotation speed of 1600 rpm with a ring potential of 1.4 V vs. RHE. The corresponding electron transfer number (n) and H₂O₂ selectivity during ORR were calculated based on the disk current and ring current as follows:

$$n=\frac{4{I}_{d}}{{I}_{d}+\frac{{I}_{r}}{N}}$$

(4)

$${{{\rm{H}}}}_{2}{{{\rm{O}}}}_{2}\left(\%\right)=\frac{200\frac{{I}_{r}}{N}}{{I}_{d}+\frac{{I}_{r}}{N}}$$

(5)

Where, I_d is the disk electrode current (mA), I_r is the ring electrode current (mA), and N is the collection efficiency (0.37).

The electrochemical properties of different GDEs were tested in the H₂O₂ electrosynthesis electrolyzer described above using the same electrochemical workstation. A three-electrode system including GDE as the working electrode, a Pt sheet as the counter electrode, and an Ag/AgCl electrode as the reference electrode was established. LSV curves were collected at a scan rate of 20 mV s⁻¹ over a potential window of 0.6 to −0.7 V vs. RHE in 0.05 M Na₂SO₄ solution (pH = 7). Electrochemical impedance spectroscopy (EIS) was carried out under an open circuit potential in Na₂SO₄ (0.05 M, pH = 7) electrolyte solution, with a frequency range from 10⁵ to 10⁻² Hz and an amplitude of 5 mV. EIS data were fitted using ZView software with an appropriate equivalent circuit. The ECSA of different GDEs was determined by a double-layer capacitance method. Cyclic voltammetry (CV) measurements were performed in the double-layer region, ranging from 0.77 V to 0.97 V vs. RHE, with scan rates of 20, 40, 60, 80, and 100 mV s⁻¹. 0.1 M deoxygenated KOH (pH = 13) was used as the supporting electrolyte. Pt sheet and Hg/HgO were used as counter electrode and reference electrode, respectively. The double-layer capacitance (C_dl) was estimated by plotting the ∆j = (j_a ‒ j_c)/2 at 0.87 V vs. RHE against the scan rate. The ECSA value was calculated from the equation:

$${{\rm{ECSA}}}=\frac{{C}_{{{\rm{dl}}}}}{{C}_{{{\rm{s}}}}}\times S$$

(6)

where C_s is the specific capacitance of a flat standard electrode with 1 cm² of real surface area (determined to be 40 μF cm⁻² according to several previous studies^21,47), S is the actual geometric area of the GDE electrode.

The same method was used to conduct in-situ ECSA monitoring during the long-term production of H₂O₂ with different GDEs, by using another custom-made electrolyzer, as shown in Supplementary Fig. 9. All the electrode potentials were converted to reference the reversible hydrogen electrode (RHE) using:

$${E}_{{{\rm{RHE}}}}={E}_{{{\rm{Ag}}}/{{\rm{AgCl}}}}+0.197+0.059\times {{\rm{pH}}}$$

(7)

$${E}_{{{\rm{RHE}}}}={E}_{{{\rm{Hg}}}/{{\rm{HgO}}}}+0.098+0.059\times {{\rm{pH}}}$$

(8)

The resistance of the electrochemical cell was measured by iR compensation on a CHI660E electrochemical station (Chenhua, Shanghai). The specific resistance values corresponding to the RRDE (5.6 ± 0.2 Ω, 0.1 M KOH), LSV (14.6 ± 0.4 Ω, 0.05 M Na₂SO₄), and CV (5.6 ± 0.2 Ω, 0.1 M KOH) techniques are different due to the different electrolytes used.

In-situ reflection-absorption imaging experiment

The in-situ reflection-absorption imaging sensing system consists of three main modules: the incident light module, the sensing module (including three-electrode system), and the detection module, as shown in Supplementary Fig. 14. Considering the discoloration speed of phenolphthalein, we chose purple light (410 nm) with the least absorption sensitivity. The purple light emitted from a light-emitting diode is projected onto the 25× objective lens, focused onto a 0.5 mm aperture to form a point light source, and then collimated into a parallel beam through an achromatic lens. The parallel beam enters through one surface of the prism in the sensing module and shines onto the electrode material. These lights are absorbed by the phenolphthalein solution (or titanium (IV)-peroxide complex) on the surface of the working electrode, generating reflected light. The reflected light is collected by the prism and detected by the camera. During the experiment, the camera collected the changes in reflected light intensity caused by the color change of phenolphthalein (or titanium (IV)-peroxide complex), which indirectly reflected the two-dimensional distribution (approximately 2 mm × 2 mm) data of the changes in hydroxyl ion (or H₂O₂) concentration on the surface of different GDEs during the reaction process. Note that in order to highlight the difference in local alkalinization at the interface between hydrophilic and hydrophobic GDEs, the electrolysis experiments were performed at a current density of 100 mA cm⁻².

In-situ ATR-SEIRAS measurements

In-situ attenuated total reflectance surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) experiments ware performed using a Bruker VERTEX 70 v spectrometer cooled by liquid nitrogen. The relevant electrochemical experiments were performed in a custom-made undivided three-electrode cell with an Ag/AgCl and a Pt serving as reference and counter electrodes, respectively. 50 μL of the catalyst ink was dropped onto the surface a Si wafer coated with Au thin layer and dried in air at room temperature to obtain the working electrode. The catalyst ink for the hydrophilic case was obtained by mixing 5 mg of CB, 950 μL of ethanol, and 50 μL of Nafion (5 wt%). And the catalyst ink for the hydrophobic case was obtained by mixing 5 mg of CB, 950 μL of ethanol, and 50 μL of PTFE (5 wt%). In-situ ATR-SEIRAS spectra of the electrolysis process were recorded in O₂-saturated 1.0 M KOH (pH = 14) at open circuit potential or a potential stepwise from 0.9 V to −0.9 V vs. RHE with continuous O₂ bubbling.

Computational details

The spin-polarized DFT calculations were performed using the Vienna ab-initio simulation package (version 5.4.1) to study the ORR process of CB electrocatalysts^48,49. The exchange correlation energy was modeled by using Perdew–Burke–Ernzerhof (PBE) functional⁵⁰ within the generalized gradient approximation (GGA). The projector augmented wave (PAW) pseudo-potentials⁴⁹ were used to describe the ion–electron interactions. The cutoff energy of 400 eV was adopted in the calculations. The DFT-D3 method was used to correct the influence of van der Waals interactions⁵¹. One layer of carbon (001) facet was cleaved with a vacuum layer of 15 Å to build the surface structures. An oxygen atom was substituted for a carbon atom to build an ether group, and the unsaturated carbon was saturated with hydrogen atoms. All the atoms were relaxed until the energy change was less than 10⁻⁵ eV for the electronic relaxation and the forces on each atom were less than 0.02 eV Å⁻¹ for the ionic relaxation. A (2 × 2 × 1) k-point grid was employed for the simulation models. The adsorption energy (E_ads) of O₂ and H₂O₂ at this active center was calculated according to the following equation:

$${E}_{{{\rm{ads}}}}={E}_{{{\rm{total}}}}-{E}_{{{\rm{substrate}}}}-{E}_{{{\rm{adsorbate}}}}$$

(9)

Where E_total, E_substrate, and E_adsorbate represent the energy of adsorption structure, substrate, and adsorbate, respectively. Based on the computational hydrogen electrode (CHE) model⁵², the Gibbs free energies of different intermediates in different ORR pathways were calculated as follows:

$$G={E}_{{{\rm{DFT}}}}+{E}_{{{\rm{ZPE}}}}-{TS}$$

(10)

Where G, E_DFT, E_ZPE, and TS represent the Gibbs free energy, electronic energy calculated by DFT, zero-point energy estimated within the harmonic approximation, and entropy at 298.15 K, respectively.

The AIMD simulations based on DFT were performed to analyze the interaction between HO₂⁻ and OH⁻ molecules on the surface of the CB electrocatalyst. AIMD simulations in this study were performed on the CP2K/Quickstep code using the mixed Gaussian and plane waves method⁵³. The core electrons were represented by the Goedecker–Teter–Hutter (GTH) pseudopotential and the molecularly optimized GTH double-ζ valence single-polarization short-range (DZVP-MOLOPT-SR-GTH) basis set with a cutoff energy of 400 Ry⁵⁴. The valence electrons were treated explicitly by the PBE exchange-correlation functional with Grimme’s D3(BJ) dispersion correction⁵¹. Each system was performed in the canonical (NVT) ensemble at the desired temperature using the canonical sampling through a velocity rescaling (CSVR) thermostat chain with a time step of 0.5 fs⁵⁵. The simulation time was set to 2.8 ps.

To obtain the specific quantitative values of the mass transfer coefficients of O₂ and H₂O₂ in the electrode and electrolyte system during real-time electrolysis, molecular dynamics (MD) simulations were performed using the Forcite functional module in Materials Studio⁵⁶. Considering the huge difference in scale between MD simulation and real experiment, it is almost impossible to completely reconstruct the experimental environment⁵⁷. Therefore, a substitution module was carefully constructed, including a layer of randomly arranged graphite structures with hydrogenated edges to represent our carbon electrocatalyst layer, 500 O₂ molecules and 500 H₂O molecules to represent open air and electrolyte, respectively. Note that the carbon layer was constructed through AC functional module by using a typically graphite structure downloaded from Cambridge Crystallographic Data Centre, and cubic form was used for the construction method. Three systems were constructed based on this module: H₂O molecules did not penetrate into the graphite layer, corresponding to the initial unwetted GDE; H₂O molecules penetrated into one-third and two-thirds of the thickness of the entire graphite layer, corresponding to GDEs with different wetting degrees at different electrolysis times. Note that H₂O₂ molecules are always placed at the interface between the H₂O molecule layer and the graphite layer (i.e., the three-phase interface, TPI). The ramp density was set from 0.30 g cm⁻³ to 2.02 g cm⁻³ to prevent the spearing of aromatic units. Energy minimization was performed using the Newton methods in cascade to eliminate local nonequilibrium. During the minimization, the convergence level was set to 0.001 kcal mol⁻¹ Å⁻¹. The geometry optimization was performed to obtain the proper position of each atom, and the maximum iteration was set to 50,000. Before the simulation process started, a 20 ps-NVT process was performed to verify the most realistic and stable structure of the modules in the constructed cell. The time-step was set to 0.7 fs and one frame was output every 250 steps. The necessary fixed coordinates were then performed for the carbon, and free volume calculations were performed to check for possible pores for water entry. Finally, a 20 ps-NVT process was performed to check for energy fluctuations. The mean-squared displacements of O₂ and H₂O₂ molecules in different systems were calculated (Eqs. (11) and (12)) to quantify their corresponding mass transfer coefficients⁵⁸. All pressures were set to atmospheric pressure, the same as in the experimental process. The Electrostatic and Van der Waals summation method was set as the force field, with a cutoff distance of 12.5 Å and a buffer width of 0.5 Å. Regarding the calculation of the relative concentration of H₂O₂ on the electrode surface under different alkalinity conditions, the same calculation indicators as above were used. The established module consisted of a 6 Ų graphene layer, 50 H₂O molecules, and different amounts of H₂O₂ (or HO₂⁻) and OH⁻ species. The Compass III force field was applied during all the calculation process⁵⁹.

$${MSD}=\frac{1}{N}{\varSigma }_{i=1}^{N}{|\varDelta {\overrightarrow{r}}_{i}(t)|}^{2}$$

(11)

$$D=\frac{1}{6}\frac{d}{{dt}}{{\mathrm{lim}}}_{t\to \infty }{MSD}(t)$$

(12)