Roles of copper(I) in water-promoted CO₂ electrolysis to multi-carbon compounds

Abstract

The membrane electrode assembly (MEA) is promising for practical applications of the electrocatalytic CO₂ reduction reaction (CO₂RR) to multi-carbon (C₂₊) compounds. Water management is crucial in the MEA electrolyser without catholyte, but few studies have clarified whether the co-feeding water in cathode can enhance C₂₊ formation. Here, we report our discovery of pivotal roles of a suitable nanocomposite electrocatalyst with abundant Cu₂O−Cu⁰ interfaces in accomplishing water-promoting effect on C₂₊ formation, achieving a current density of 1.0 A cm⁻² and a 19% single-pass C₂₊ yield at 80% C₂₊ Faradaic efficiency in MEA. The operando characterizations confirm the co-existence of Cu⁺ with Cu⁰ during CO₂RR at ampere-level current densities. Our studies reveal that Cu⁺ works for water activation and aids C‒C coupling by enhancing formations of adsorbed CO and CHO species. This work offers a strategy to boost CO₂RR to C₂₊ compounds in industrial-relevant MEA by combining water management and electrocatalyst design.

Water plays a key role in electrocatalytic CO₂ reduction, but the role of catalyst in water management is unknown. Here, the authors report the importance of catalyst in determining the function of water and demonstrate that the catalyst containing Cu⁺ promotes water activation and C₂₊ formation.

Introduction

The electrocatalytic CO₂ reduction reaction (CO₂RR) using renewable electricity to produce multi-carbon (C₂₊) compounds, in particular ethylene and ethanol, which are widely used in the current chemical and energy industries, is one of the most promising routes to recycle CO₂ and reduce carbon emission^1–3. The electrolyser is the core of CO₂ electrolysis⁴. Among different electrolysers, the zero-gap membrane electrode assembly (MEA), which has been successfully employed for the polymer electrolyte membrane-based water electrolysis and fuel cells, holds great potential as an energy-efficient and scalable device for CO₂ electrolysis^5–7. In the MEA electrolyser for CO₂RR, the cathode catalyst is directly pressed onto a membrane without a catholyte, thus alleviating the problems faced by the alkaline flow cell, such as the flooding of gas diffusion electrode (GDE), carbonate formation and CO₂ loss^3,4.

The lack of the catholyte solution makes the management of water, which is the proton source and is essential to CO₂RR, in the MEA electrolyser more crucial than in the conventional flow cell. In MEA, the H₂O molecules needed for the cathode reaction usually comes from the anolyte by permeation through the membrane and the problem of H₂O deficiency may occur at a high current density, worsening the CO₂RR performance⁵. Co-feeding H₂O with CO₂ in the cathode has been proven to enhance CO₂RR to C₁ products (CO or formate)^5,8–11. For the formation of C₂₊ products, the H₂O consumption becomes approximately twice that for the formation of C₁ products (Supplementary Fig. 1), posing an increased need for H₂O participation. However, only very few studies have been devoted to H₂O management for MEA-based CO₂RR to C₂₊ compounds^5,12. Choi et al. reported that the protons in C₂H₄ originated dominantly from the H₂O molecules of the anolyte coming across the membrane and that co-feeding H₂O with CO₂ in the cathode with a metallic copper catalyst rather enhanced the hydrogen evolution reaction (HER) instead of CO₂RR¹². The cathode catalyst is expected to participate in the activation of H₂O co-fed with CO₂, but the role of the electrocatalyst in H₂O management in CO₂RR is under explored.

Actually, the lack of catholyte as well as useful insights for the design of cathode catalyst for efficient water management has resulted in limited C₂₊ formation performances of the MEA-based CO₂RR. The current density to attain a C₂₊ Faradaic efficiency of ≥ 80% is usually lower than 400 mA cm⁻², regardless of using a cathode-dry or H₂O vapour-feeding MEA system (Supplementary Table 1)¹³. On the other hand, the current density could attain ≥ 1.0 A cm⁻² at ≥ 80% C₂₊ Faradaic efficiency (FE) in a flow cell^3,14, where the alkali and/or hydroxide ions in the alkaline catholyte favoured CO₂ activation and subsequent C−C coupling¹⁵. It remains challenging to achieve simultaneously a high current density and a high C₂₊ FE in the full-cell MEA system.

Here, we report our discovery of a suitable electrocatalyst in fulfilling the role of co-feeding H₂O in enhancing formation of C₂₊ compounds during CO₂RR in the MEA system. We demonstrate that the co-feeding of H₂O has no effect on C₂₊ formation over a Cu⁰ nanoparticle catalyst, but a nanocomposite containing abundant Cu₂O−Cu⁰ interfaces shows significantly positive effect of co-feeding H₂O on CO₂RR to C₂₊ compounds. Under an optimized H₂O pressure, we achieve a current density of 1.0 A cm⁻² at 80% FE of C₂₊ compounds (mainly C₂H₄ and C₂H₅OH) in the full-cell MEA electrolyser. The molar carbon-based selectivity and yield of C₂₊ compounds (C_2-3 olefins and oxygenates) reach 83% and 19%, respectively, better than those attained in thermocatalytic systems for CO₂ hydrogenation under harsher reaction conditions¹⁶. Our operando spectroscopic characterizations confirm the presence of Cu⁺ in the nanocomposite during CO₂RR operated at an ampere-level current density in MEA. We propose an in situ re-oxidation mechanism for the formation of Cu⁺ sites. The catalyst with Cu⁺ sites promotes the activation of H₂O, resulting in enhanced formation of adsorbed CO and CHO intermediates for C−C coupling, and thus showed significant promoting effect of co-feeding H₂O for C₂₊ formation. This work paves the way for constructing an efficient MEA system for CO₂RR to C₂₊ compounds.

Results

Roles of catalysts in determining the effect of co-feeding H₂O

The Cu₂O−Cu⁰ nanocomposite, which was synthesized by solvothermal followed by electroreduction, was composed of the mixed phases of Cu₂O and Cu⁰ resulting from a CuO−Cu₂O composite precursor (Supplementary Fig. 2a). The reference catalyst consisting of only Cu⁰ nanoparticles (Supplementary Fig. 2b) was prepared by H₂ reduction of the CuO−Cu₂O precursor followed by electroreduction. The full-cell CO₂RR performance was evaluated in a MEA electrolyser (Supplementary Fig. 3 and Fig. 1a) employing IrO_x as an anode catalyst (Supplementary Fig. 4) and 0.10 M KHCO₃ as an anolyte for O₂ evolution. In the MEA cell (Fig. 1a), H₂O molecules needed for the cathode reaction can come from the anolyte across the anion-exchanging membrane or can be added into CO₂ flow by a H₂O vapour generator. In the case of co-feeding H₂O in the cathode, the partial pressure of H₂O, P(H₂O), in the inlet of the MEA is kept at <3.1 kPa to ensure that the H₂O vapour does not condense in the pores of electrode (Supplementary Fig. 5a). The sustaining of hydrophobicity of the electrode after the reaction (Supplementary Fig. 5b) further suggests that the gas flow with H₂O vapour can be transported effectively through the GDE¹⁷.

Fig. 1. Effect of co-feeding water on CO₂RR in the MEA electrolyser.

a Schematic diagram of the water source in the MEA electrolyser. GDE: gas diffusion electrode. b Effect of partial pressure (P_H2O) of co-feeding water on the FEs of products and the current density over the Cu₂O−Cu⁰ catalyst at a cell voltage of 3.8 V. c FEs of products and the current density over the Cu₂O−Cu⁰ catalyst in the presence of co-feeding water with a pressure of 2.7 kPa. d FEs of products and the current density for the Cu⁰ catalyst in the presence of co-feeding water with a pressure of 2.7 kPa. e The ratio of water consumption rate in CO₂RR to that in HER with and without co-feeding water over the Cu₂O−Cu⁰ and Cu⁰ catalysts. f ECSA-normalized C₂₊ formation rates with and without co-feeding water over the Cu₂O−Cu⁰ and Cu⁰ catalysts. All error bars represent the standard deviation calculated from at least 3 data points. The cell voltages for the MEA cell were presented as non-IR corrected values. Source data are provided as a Source Data file.

The CO₂RR to C₂₊ compounds (including ethylene, ethanol, acetate and n-propanol) could proceed without co-feeding H₂O. Many operation parameters, such as the flow rate of cathode feed, the reaction temperature and the ionomer (a quaternary ammonia polysulphone, aQAPS-S₁₄, see Supplementary Fig. 6 for the structure) coated on the cathode catalyst, which could accelerate the conduction of OH^‒ ions generated¹⁸, might affect the CO₂RR performance by influencing the availability of H₂O on the cathode in the absence of co-feeding H₂O^19–21. Our evaluation suggests that the influences of these parameters are limited in boosting the C₂₊ formation rates (Supplementary Fig. 7).

We found a significant promoting effect of co-feeding H₂O with an appropriate P(H₂O) over the Cu₂O−Cu⁰ catalyst. The increase in P(H₂O) from 0 to 2.7 kPa increased both the current density and the C₂₊ FE and suppressed the HER, but a too high P(H₂O) (3.0 kPa) was unbeneficial to the CO₂RR, probably because of the hindrance of the adsorption and activation of CO₂ (Fig. 1b and Supplementary Fig. 8a–d). A current density of 1.0 A cm⁻² and a C₂₊ FE of 80% were achieved at a P(H₂O) of 2.7 kPa at a cell voltage of 3.8 V over the Cu₂O−Cu⁰ catalyst (Fig. 1b, c), and meanwhile the C₂₊ formation rate reached 2.5 mmol cm⁻² h⁻¹ (Supplementary Fig. 8e). However, the positive effect of co-feeding H₂O on C₂₊ formation was not observed over the Cu⁰ catalyst (Supplementary Fig. 9a and Fig. 1d). The FE of C₂₊ compounds rather decreased and that of H₂ increased by the presence of H₂O over Cu⁰ at most of the cell voltages (Supplementary Fig. 9b, c). Thus, in spite of slight increases in current densities, the C₂₊ formation rate was almost unchanged by co-feeding H₂O at all the investigated cell voltages over this catalyst (Supplementary Fig. 9d).

The rate of H₂O consumption (including CO₂RR and HER) was increased to some extent by co-feeding H₂O in particular at a higher cell voltage over both Cu₂O−Cu⁰ and Cu⁰ catalysts (Supplementary Fig. 10). The former catalyst showed a higher H₂O consumption rate, in agreement with its higher activity. We observed a significant difference in the change of ratio of H₂O consumption in CO₂RR to that in HER, denoted as f(CO₂RR)/f(HER), after the co-feeding of H₂O between the two catalysts. The f(CO₂RR)/f(HER) was always larger over Cu₂O−Cu⁰ than that over Cu⁰ irrespective of the presence of co-feeding H₂O (Fig. 1e). Over the Cu⁰ catalyst, the f(CO₂RR)/f(HER) values in the absence and presence of co-feeding H₂O were almost the same at all the cell voltages investigated. On the other hand, it is of interest that the f(CO₂RR)/f(HER) over Cu₂O−Cu⁰ was significantly enhanced by co-feeding H₂O at each cell voltage (Fig. 1e), suggesting that the co-feeding H₂O is beneficial to CO₂RR but not HER over this catalyst.

The electrochemical surface area (ECSA) evaluated by measuring the double-layer capacitance (Supplementary Fig. 11) did not change by the presence of H₂O for either Cu₂O−Cu⁰ or Cu⁰ (Supplementary Table 2). The ionic conductivity also remained in the presence of H₂O as reflected from the electrochemical impedance spectroscopy (EIS) (Supplementary Fig. 12). Thus, the enhancement in the C₂₊ formation rate and f(CO₂RR)/f(HER) by co-feeding H₂O over Cu₂O−Cu⁰ should not arise from the changes of the catalyst ECSA or the ionic conductivity of the system.

The ECSA-normalized C₂₊ formation rate on the Cu₂O−Cu⁰ catalyst was higher than that on Cu⁰ even without co-feeding H₂O (Fig. 1f). Thus, the Cu₂O−Cu⁰ catalyst is intrinsically more active toward the C₂₊ formation. More significantly, in the presence of co-feeding H₂O, the difference in the normalized C₂₊ formation rate became more pronounced between the two catalysts, because Cu⁰ had almost no effect of co-feeding H₂O on C₂₊ formation, whereas remarkable H₂O promoting effect was observed on the Cu₂O−Cu⁰ catalyst.

The voltage breakdown analysis using a four-electrode system (Supplementary Fig. 13), which could enable voltage drop analysis at different positions (Supplementary Fig. 14), showed that the main difference in cell voltage at a fixed current density between the Cu₂O−Cu⁰ and Cu⁰ catalysts mainly originated from the cathode non-ohmic overpotential (Supplementary Fig. 15). The lower cathode overpotential generated during CO₂RR in the presence of co-feeding H₂O on the Cu₂O−Cu⁰ catalyst than that on Cu⁰ is responsible for its higher CO₂RR activity, confirming that the intrinsic characteristic of the catalyst is the primary contributor to the observed difference in CO₂RR performances.

The outstanding performance of the Cu₂O−Cu⁰ catalyst is also demonstrated by the single-pass CO₂ conversion, C₂₊ selectivity and C₂₊ yield on a molar carbon basis, which are straightforward parameters for comparison with thermocatalytic CO₂ hydrogenation and are crucial for practical applications. At a current density of 1.0 A cm⁻², the single-pass yield of C₂₊ compounds including C₂H₄, C₂H₅OH, CH₃COO⁻ and C₃H₇OH was 19% with 23% CO₂ conversion and 83% C₂₊ selectivity at a CO₂ flow rate of 40 mL min⁻¹ (Fig. 2a and Supplementary Fig. 16). Such a high selectivity of C₂₊ olefins and oxygenates at about 20% CO₂ conversion is hard to achieve by the thermocatalytic hydrogenation of CO₂ under harsh conditions¹⁶. The present Cu₂O−Cu⁰ catalyst is also quite stable and the performance could be maintained at least for 350 h at a current density of 250 mA cm⁻² (Fig. 2b), better than other catalysts reported to date at an industrial-relevant current density ( ≥ 200 mA cm⁻²) in the MEA-based full-cell CO₂RR to C₂₊ compounds (Supplementary Table 1). A further comparison with some top MEA-based CO₂RR systems shows that the present Cu₂O−Cu⁰ catalyst with co-feeding H₂O could achieve simultaneously a high current density and a high C₂₊ FE, thus resulting in a record-breaking C₂₊ formation rate of 2.5 mmol cm⁻² h⁻¹ (Fig. 2c and Supplementary Table 1). The combination of activity, selectivity and stability suggests that the present MEA system with the Cu₂O−Cu⁰ catalyst in the presence of co-feeding H₂O holds great potential for future applications.

Fig. 2. Full-cell CO₂RR performances in the MEA electrolyser.

a CO₂RR performances as a function of CO₂ flow rate at a constant current density of 1.0 A cm ^− 2. CO₂ conv.: CO₂ conversion. b Stability of the Cu₂O−Cu⁰ catalyst at a constant current density of 250 mA cm ^− 2. The gas flow rates of CO₂ and N₂ were both 30 mL min ^− 1. c Comparison of some typical results using the MEA electrolyser for CO₂RR to C₂₊ compounds (see Supplementary Table 1). d Full-cell energy conversion efficiency (ECE) of C₂₊ compounds over the Cu₂O-Cu⁰ and Cu⁰ catalysts. Reaction conditions: anolyte, 0.10 M KHCO₃ or KOH (Fig. 2d), partial pressure of water in the cathode gas flow, 2.7 kPa. All error bars represent the standard deviation calculated from at least 3 data points. The cell voltages for the MEA cell were presented as non-iR corrected values. Source data are provided as a Source Data file.

The energy conversion efficiency of C₂₊ compounds (C₂₊ ECE), i.e., the ratio of chemical energy stored in the C₂₊ products to the electrical energy input, was also much higher on the Cu₂O−Cu⁰ catalyst than that on Cu⁰ (Fig. 2d). The change of anolyte from 0.10 M KHCO₃ to 1.0 M KOH lowered the cell voltage for ~0.5 V needed to reach a similar current density owing to the decreased cell resistance, and thus the full-cell C₂₊ ECE increased and reached 31% at a cell voltage of 3.1 V (Fig. 2d), where the current density and C₂₊ FE were about 0.80 A cm⁻² and 80%, respectively (Supplementary Fig. 17).

In short, we discovered that the role of co-feeding H₂O in the MEA-based CO₂RR to C₂₊ compounds depends on the catalyst. The Cu₂O−Cu⁰ catalyst shows a significantly positive role of co-feeding H₂O in the formation of C₂₊ compounds by increasing the current density and the C₂₊ FE while suppressing the HER. On the other hand, the co-feeding H₂O has almost no effect on CO₂RR to C₂₊ compounds over the Cu⁰ catalyst. Our finding also implies that the difference in catalytic behaviours for the MEA-based CO₂RR to C₂₊ compounds between different catalysts may arise from their different behaviours toward the co-feeding H₂O.

Structure of catalysts under working conditions

The operando X-ray diffraction (XRD) measurements were performed in a home-designed operando electrochemical-XRD cell (Supplementary Fig. 18) to monitor the evolution of active phases during CO₂RR. The precursor of the nanocomposite catalyst consisted of mixed Cu₂O and CuO phases under open circuit potential (OCP) (Fig. 3a). Following the CO₂RR at a cell voltage in a range of 3.0–3.8 V, the XRD peaks assignable to CuO vanished and those to metallic Cu appeared and became dominant. The peaks assignable to Cu₂O became weakened with an increase in the cell voltage but could still be observed at 3.8 V, which corresponded to a current density of about 1 A cm^-2. Cu₂O was observable even after 5 h of CO₂RR at 3.8 V, revealing that Cu₂O is present throughout the entire CO₂RR process. In agreement with the ex situ XRD result (Supplementary Fig. 2a), the operando XRD results provide solid evidence that Cu⁰ and minor Cu₂O phases co-exist in the nanocomposite catalyst under working conditions.

Fig. 3. Characterizations of the Cu₂O−Cu⁰ catalyst.

a Operando XRD patterns at different cell voltages and at 3.8 V for 5 h. OCP: open circuit potential. b SEM image of the Cu₂O−Cu⁰ catalyst. c HRTEM image of the Cu₂O−Cu⁰ catalyst. d Operando XANES spectra at different cell voltages and at 3.8 V for 5 h. e Operando Raman spectra at different cell voltages and at 3.8 V for 5 h. f Schematic diagram of dynamic reduction and re-oxidation mechanism of copper for the Cu₂O−Cu⁰ catalyst during CO₂RR. Source data are provided as a Source Data file.

The scanning electron microscopy (SEM) showed that the Cu₂O−Cu⁰ nanocomposite after electroreduction has a sesame ball-like morphology with sizes in a range of 150–200 nm (Fig. 3b and Supplementary Fig. 19a, b). The Cu⁰ catalyst has a smoother and irregular morphology with similar sizes (Supplementary Fig. 19c, d). The high-resolution transmission electron microscopy (HRTEM) images of the Cu₂O−Cu⁰ catalyst displayed lattice fringes with interplanar spacings of 0.247 and 0.208 nm, which could be ascribed to the Cu₂O (111) and Cu (111) facets, respectively (Fig. 3c). The lattice fringes of the Cu⁰ catalyst were attributable to the Cu (111) facet (Supplementary Fig. 20).

The operando X-ray absorption spectroscopy (XAS) studies were performed using an operando electrochemical-XAS cell (Supplementary Fig. 21). The X-ray absorption near-edge structure (XANES) result showed that the Cu K-edge for the Cu₂O−Cu⁰ nanocomposite catalyst shifted slightly toward that for Cu foil on increasing the cell voltage from 3.0 to 3.8 V, but it remained between those for Cu foil and Cu₂O even after 5 h of CO₂RR at 3.8 V (Fig. 3d and Supplementary Fig. 22), indicating an average oxidation state of copper between 0 and +1. The Cu K-edge for the Cu⁰ catalyst under the same electrolysis conditions was almost the same as that for Cu foil, confirming that Cu is in metallic state in this catalyst (Supplementary Fig. 22). The wavelet-transform analysis of the extended X-ray absorption fine structure (EXAFS) spectrum at Cu K-edge at a cell voltage of 3.8 V revealed that the Cu₂O−Cu⁰ nanocomposite catalyst had both Cu−Cu and Cu−O coordination shells, whereas the Cu⁰ catalyst had only a Cu−Cu coordination shell (Supplementary Fig. 23). The EXAFS fitting result indicates that the bond length of Cu−Cu in the working Cu₂O−Cu⁰ nanocomposite catalyst is longer than that in Cu foil but shorter than that in Cu₂O, while the length of its Cu−O bond is longer than that in Cu₂O (Supplementary Fig. 24 and Table 3). The coordination numbers of Cu−Cu and Cu−O shells in the Cu₂O−Cu⁰ nanocomposite are 8.80 ± 0.25 and 0.31 ± 0.19, respectively (Supplementary Table 3). These results confirm that the nanocomposite catalyst is composed of both Cu⁺−O and Cu⁰ domains under working conditions.

We further performed operando Raman spectroscopic studies in a home-designed Raman-MEA cell that enables Raman measurements at ampere-level current densities (Supplementary Fig. 25). Four Raman bands at 410, 520, 620 and 710 cm⁻¹, attributable to CuO_x/OH_y, Cu₂O T_2g, Cu₂O T_1u and Cu−OH, respectively^22,23, were clearly observed during CO₂RR at a cell voltage of 2.6–3.8 V (Fig. 3e). These Raman bands remained clearly seen even after 5 h reaction at 3.8 V. On the other hand, only a very weak Raman band at ~710 cm⁻¹ was observed in the operando Raman spectra for the Cu⁰ catalyst (Supplementary Fig. 26). The survival of Raman bands at 520 and 620 cm⁻¹ attributable to Cu₂O provides evidence for the stable existence of Cu⁺ sites on the surface of Cu₂O‒Cu⁰ under reaction conditions.

Raman spectroscopy using ¹⁸O-labelled water (H₂¹⁸O) instead of traditional H₂¹⁶O was performed in a half-cell to probe the source of oxygen associated with Cu⁺ in the catalyst. When the applied potential was shifted from OCP to −1.0 V versus reversible hydrogen electrode (RHE), the Raman bands belonging to Cu₂O at 520 and 620 cm⁻¹ shifted gradually to lower wavenumbers and eventually stabilized at 500 and 600 cm⁻¹, respectively (Supplementary Fig. 27a), indicating the changing of Cu₂¹⁶O into Cu₂¹⁸O during the reaction²⁴. The returning to OCP kept the Raman bands of Cu₂¹⁸O. These results suggest that the oxygen associated with Cu⁺ at the reaction potential is not the remaining oxygen in the CuO_x precursor but originates from water.

We propose a dynamic interchanging mechanism between the reduction and the oxidation states of copper (Fig. 3f). It is known that OH^‒ can be generated on the cathode during CO₂RR and the OH^‒ species favours the oxidation of Cu⁰ to Cu⁺ (Ref. ²⁵). The OH^‒ species along with a small amount of O₂ detected in the cathode (Supplementary Fig. 27b), which might come across the membrane from the anode by the oxidation of H₂¹⁸O (Ref. ²⁶), is proposed to account for the generation of Cu⁺ via re-oxidation. We speculate that the higher local concentration of OH⁻ species over the Cu₂O−Cu⁰ catalyst with rougher surfaces owing to the unique sesame ball-like morphology (Fig. 3b) would be beneficial to the in situ oxidation to generate Cu⁺. In other words, the stable existence of Cu⁺ on the working Cu₂O‒Cu⁰ catalyst is related to the high concentration of OH^‒ generated over the surface and the unique catalyst structure.

Mechanistic studies by density functional theory calculations

Based on the models in the literature^27–29 and the experimental facts, such as the exposed facets (Fig. 3c and Supplementary Fig. 20), the bond lengths and the coordination numbers gained from EXAFS (Supplementary Table 3), we adopted the Cu(111) facet to model the Cu⁰ surface and the extracted half of the oxygen atoms from a pristine Cu₂O(111) slab to simulate the Cu₂O−Cu⁰ surface for density functional theory (DFT) calculations (Supplementary Fig. 28 and Supplementary Data 1).

The current consensus suggests that the mechanism for CO₂ reduction to C₂₊ products on copper surfaces involves CO as the major intermediate, followed by subsequent C−C coupling and hydrogenation/hydrogenolysis³⁰. The first step is the activation of linear CO₂ molecule to the bent adsorbed COO (*CO₂) species (Supplementary Fig. 29), which is then converted to adsorbed CO (*CO) likely via *COOH intermediate on both Cu(111) and Cu₂O−Cu⁰ surfaces (Supplementary Fig. 30). Our DFT calculations show that the activation of a CO₂ molecule to *CO₂ on Cu(111) is an endothermic reaction requiring a reaction energy of 1.06 eV, whereas the reaction energy on Cu₂O−Cu⁰ decreased to 0.12 eV (Fig. 4a and Supplementary Table 4). This could result from the more energy-favourable CO₂ activation on the oxidic Cu surface²⁹. The Bader charge analysis reveals that the Cu atom at the edge of Cu₂O domain on Cu₂O−Cu⁰ has a positive charge of 0.7 eV, and the strong electrostatic interaction between the positively charged Cu atom and the O atom of CO₂ can largely facilitate its adsorption (Fig. 4b). We further calculated the adsorption energy of H₂O, the other major reactant and the proton donor for CO₂RR to C₂₊ compounds, and the result revealed that the H₂O adsorption was significantly stronger with an adsorption energy of −0.58 eV on Cu₂O−Cu⁰ than that on Cu(111) (adsorption energy, 0.08 eV) (Fig. 4c and Supplementary Fig. 31). This is likely because of the hydrogen bond between H₂O and the surface oxygen atoms^31,32. Moreover, the reaction energy for H₂ evolution from the adsorbed hydrogen (*H) species on Cu₂O−Cu⁰ (−0.02 eV) was less negative than that on Cu(111) (−0.16 eV), indicating that HER would not be accelerated on the Cu₂O−Cu⁰ surface irrespective of the enhanced H₂O activation (Supplementary Figs. 32, 33). Consequently, on the Cu₂O−Cu⁰ catalyst, there is a concurrent enhancement in adsorption and activation of CO₂ and H₂O molecules, without facilitating HER.

Fig. 4. DFT calculations of CO₂RR on Cu(111) and Cu₂O−Cu⁰ model surfaces.

a The reaction energy diagram for CO₂ reduction to adsorbed CO. b Bader charges of Cu atoms on the surfaces with CO₂ adsorption (above) and with *OCCHO intermediate adsorption (below). c The Gibbs free energies (ΔG_ads) of water adsorption. d The reaction energy diagram for the dimerization of *CO to *OCCO, the hydrogenation of *CO to *CHO and the subsequent coupling between *CHO and *CO to *OCCHO. The blue, red, white and brown balls represent Cu, O, H and C atoms, respectively. Source data are provided as a Source Data file.

Regarding the C−C coupling mechanism, the direct dimerization of *CO to *OCCO has widely been proposed^3,30. Our DFT calculations show that the reaction energies for the direct *CO dimerization are 1.48 and 0.98 eV on Cu(111) and Cu₂O−Cu⁰, respectively (Fig. 4d and Supplementary Fig. 34). On the other hand, the hydrogenation of *CO to *CHO in the first step followed by the coupling between *CO and *CHO to *OCCHO is an energetically more favourable pathway on both model surfaces (Fig. 4d and Supplementary Fig. 35). In particular, on Cu₂O−Cu⁰, the reaction energies for *CO to *CHO and *CO − *CHO coupling are only 0.39 and 0 eV, respectively, which are significantly lower than those on Cu(111) (Fig. 4d). The Bader charge analysis indicates that the Cu^δ+ site at the edge of Cu₂O domains on Cu₂O−Cu⁰ can bind the O atom of *OCCHO more facilely, thus stabilizing the *OCCHO intermediate (Fig. 4b).

In short, the DFT calculations indicate that CO₂RR over our catalysts proceeds through *CO via *CO₂ and *COOH, followed by the hydrogenation of *CO to *CHO and the coupling between *CO and *CHO to C₂₊ compounds via *OCCHO. The Cu₂O−Cu⁰ surface facilitates the adsorption and activation of both CO₂ and H₂O molecules as well as the formation of key intermediates, in particular *CO, *CHO and *OCCHO species, thus contributing to the higher C₂₊ formation rate.

Mechanistic studies by spectroscopic characterizations

The activation of co-feeding H₂O is a crucial step in our work. Raman spectroscopy has been proven to be a powerful tool for elucidating the structure of interfacial water on electrodes in recent years³³. We conducted in situ Raman spectroscopic measurements in a half-cell with an applied potential range of −0.3 to −1.0 V versus RHE, corresponding to the cell-voltage range used in the MEA cell as reflected by the voltage drop analysis (Supplementary Fig. 15). The Raman spectra for the O−H stretching modes (ν_O-H) in a range of 3000–3800 cm⁻¹ for the Cu₂O−Cu⁰ and Cu⁰ catalysts could be deconvoluted into three bands at approximately 3200, 3400 and 3600 cm⁻¹ (Fig. 5a), which could be attributed to four-coordinated H-bonded (4-HB) water, two-coordinated H-bonded (2-HB) water and weak hydrogen-bonded dangling O−H (weak-HB) water, respectively^33,34. The frequencies of the three bands strongly depended on the applied potential (Fig. 5a), suggesting that the recorded Raman signals originated from the first few water layers close to the catalyst surface³³. The activation energy for water dissociation has been reported to follow the order of weak-HB water <2-HB water <4-HB water³⁵. Thus, the weak-HB water would favour the formation of adsorbed hydrogen (*H) species from H₂O, which in turn affects the subsequent CO₂RR performance³⁶. The relative fraction of the weak-HB water on Cu₂O−Cu⁰ became significantly higher than that on the Cu⁰ catalyst as the applied potential becomes more negative (Supplementary Fig. 36), implying that the Cu₂O−Cu⁰ catalyst has stronger H₂O dissociative adsorption ability under the conditions for efficient CO₂RR. Further, we found that the Raman band of the weak-HB water on Cu₂O−Cu⁰ occurred at a relatively lower wavenumber than that on Cu⁰ at a negative applied potential (Fig. 5a), also indicative of stronger interactions between H₂O and the Cu₂O−Cu⁰ surface under working conditions^34,37. Upon increasing the absolute value of the applied potential, a more significant red shift in Raman band for the weak-HB water was observed for the Cu₂O−Cu⁰ catalyst, indicating a larger Stark effect (slope, 159 cm⁻¹ V⁻¹) as compared to Cu⁰ (slope, 86 cm⁻¹ V⁻¹) (Fig. 5b). This further demonstrates the stronger adsorption of H₂O on the Cu₂O−Cu⁰ catalyst³⁸. Therefore, the Raman spectroscopic results agree well with the DFT calculations and provide experimental evidence for the stronger adsorption of H₂O molecules on the Cu₂O−Cu⁰ catalyst.

Fig. 5. In situ Raman spectra.

a The O–H stretching vibration region for the Cu₂O−Cu⁰ and Cu⁰ catalysts with adsorbed water at different applied potentials (versus RHE). 4-HB: four-coordinated H-bonded water. 2-HB: two-coordinated H-bonded water. weak-HB: weak hydrogen-bonded dangling O−H water. b The linear relationships between the shifts of Raman bands and the applied potential for the dangling (weak-HB) O−H stretching vibration for the Cu₂O−Cu⁰ and Cu⁰ catalysts with adsorbed water. c In situ Raman spectra for CO₂RR over the Cu₂O−Cu⁰ catalyst at different applied potentials (versus RHE). d In situ Raman spectra for CO₂RR over the Cu⁰ catalyst at different applied potentials (versus RHE). All applied potentials were not IR corrected. Source data are provided as a Source Data file.

In situ Raman spectroscopy was further employed to identify possible reaction intermediates during CO₂RR. On scanning the applied potential from OCP to −0.4 V versus RHE over the Cu₂O−Cu⁰ catalyst, Raman bands appeared at 1140, 1340 and 2080 cm⁻¹ (Fig. 5c), which could be attributed to adsorbed CO₂⁻ species (v_sCOO⁻, stretching of CO bond coordinated to surface), OCO⁻ species (ν_sOCO⁻, symmetric stretching of carboxylate species) and adsorbed CO species (vCO, stretching vibration of CO bond), respectively^39,40. The appearance of adsorbed CO₂⁻ implies the favourable CO₂ activation on the Cu₂O−Cu⁰ catalyst. A further scan to −0.5 V versus RHE increased the intensity of the band attributed to *CO dramatically (Fig. 5c), indicating that *CO is a major reaction intermediate under working conditions. In addition, Raman bands at 1160, 1455 and 1580 cm⁻¹, which could be ascribed to *OCCHO species^41,42, were observed. When the potential further shifted to a more negative value, the Raman band of *CO became weakened and those attributable to *OCCHO were enhanced, suggesting that *CO undergoes reduction and C−C coupling to *OCCHO⁴¹. Over the Cu⁰ catalyst, Raman bands for *CO and *OCCHO species appeared from a more negative potential (−0.7 V versus RHE) and became significantly weaker, suggesting the weaker adsorption or lower concentrations of these species on Cu⁰ (Fig. 5d). This is supported by the electrocatalytic result that the Cu⁰ catalyst showed a significantly lower C₂₊ formation rate than the Cu₂O−Cu⁰ catalyst at the corresponding applied potential (Supplementary Fig. 8e, 9d). The observed blue shift of the adsorbed CO band at 2090 cm⁻¹ indicates a stronger CO bond and a weaker interaction between CO molecules and the catalyst⁴³, and thus weakened CO adsorption on Cu⁰ (Fig. 5d). The analysis of Raman bands of *CO on Cu₂O−Cu⁰ and Cu⁰ catalysts at different potentials revealed a larger Stark effect on the Cu₂O−Cu⁰ catalyst^44,45, indicating a stronger CO adsorption on this catalyst (Supplementary Fig. 37). Moreover, we conducted CO stripping measurements, and found that both the intensity and the position of electrochemical CO stripping peaks on Cu₂O−Cu⁰ were larger than those on Cu⁰ (Supplementary Fig. 38). In agreement with the DFT calculations, these experimental findings all affirm that Cu₂O−Cu⁰ has a higher CO adsorption ability than Cu⁰. Therefore, the catalyst composed of Cu⁺−O and Cu⁰ could enhance the activation of H₂O and CO₂ and facilitate the formation and adsorption of CO as well as the formation of *CHO and *OCCHO species, contributing to promoting the formation of C₂₊ compounds.

Discussion

We discovered that the cathode catalyst played a crucial role in determining the role of the cathode co-feeding water in CO₂RR to C₂₊ compounds. While almost no promoting effect on C₂₊ formation was observed on a Cu⁰-only catalyst, the Cu₂O−Cu⁰ nanocomposite consisting of Cu₂O and Cu⁰ domains demonstrated a significant boosting effect of co-feeding water. Both the current density and the C₂₊ FE were enhanced while the H₂ evolution was suppressed by co-feeding H₂O with a proper partial pressure over the Cu₂O−Cu⁰ catalyst. This catalyst achieved a current density of 1.0 A cm⁻² and a single-pass yield of C₂₊ compounds of 19% at 80% C₂₊ FEs under an optimised partial pressure of co-feeding H₂O. A full-cell ECE of 31% could be attained for C₂₊ formation at current density of 800 mA cm⁻² and the system was stable at 250 mA cm⁻² at least for 350 h. Our operando spectroscopic characterizations confirmed the co-presence of Cu⁺ with Cu⁰ on the nanocomposite catalyst during CO₂RR with an ampere-level current density and an in situ re-oxidation mechanism was proposed for the presence of Cu⁺. Both the experimental and DFT computational studies unveiled that Cu⁺ accelerates the activation of H₂O molecules, enhances the formation of adsorbed CO and CHO species, thus favouring the C−C coupling to C₂₊ compounds via *OCCHO. This work not only presents a new insight into the crucial role of the cathode catalyst in water management to boost CO₂RR to C₂₊ compounds in the MEA electrolyser but also offers spectroscopic evidence for the co-presence of Cu⁺ during CO₂RR at an ampere-level current density and the unique function of Cu⁺ for the activation of H₂O to aid C−C coupling.

Methods

Chemicals and materials

Copper nitrate (Cu(NO₃)₂·2.5H₂O, 98%) was purchased from Sigma-Aldrich (Shanghai) Trading Co, Ltd. Potassium bicarbonate (KHCO₃, 99.95%) and potassium hydroxide (KOH, 99.99%) were purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. N, N-dimethyl formamide (C₃H₇NO, 99%) was purchased from Sinopharm Chemical Reagent Co., Ltd. Copper nanoparticles (sizes: 150–200 nm), deuterium oxide (D₂O, 99.8%), dimethyl sulfoxide (C₂H₆OS, 99.9%) and ¹⁸O-labelled water (H₂¹⁸O, 98%) were purchased from Energy Chemical Co. The CO₂ (99.999%) and CO (99.999%) gases were purchased from Messer Gas Products Co., Ltd. The gas diffusion layer (GDL, YLS-30T), Nafion binder (5 wt%, DuPont 520) purchased from Suzhou Sinero Technology Co., Ltd. The IrO_x-coated Ti mesh (thickness: 300 μm, 300 mesh size) was purchased from Hebei Anping Hongyun Metal Products Co., Ltd. Quaternary ammonia polysulfone ionomer (2 wt%, aQAPS-S₁₄) was purchased from Hephas Energy Co. Ltd. Anion exchange membranes (AEMs, QAPPT, dry, 15 μm) were purchased from EVE Institute of New Energy Technology (Alkymer). The 5 cm² CO₂ MEA electrolyser, purchased from Dioxide Materials Co., features a titanium anode flow field and a 904 L stainless steel cathode flow field. The ultrapure water used in the experiments was prepared by an ultrapure water system, with a resistivity of 18.2 MΩ·cm.

Fabrication of the Cu₂O−Cu⁰ catalyst

The CuO_x precursor was first synthesized by a solvothermal method. Typically, Cu(NO₃)₂·2.5H₂O (500 mg) was added into a solution of N, N-dimethyl formamide (DMF) (75 mL). After vigorous stirring for 15 min, the mixture was transferred into a 100 mL Teflon-lined autoclave, sealed and subjected to solvothermal treatment at 160 °C for 4 h. After the solvothermal treatment, the solid product was collected and washed with ethanol and water and then dried in a freeze dryer to obtain the CuO_x precursor. Finally, the GDE with CuO_x precursor was prepared according to the procedure described above and the Cu₂O−Cu⁰ electrocatalyst was formed by the subsequent electroreduction.

Fabrication of the Cu⁰ catalyst

The CuO_x precursor was transferred to a crucible placed in an atmosphere-controllable furnace and was heated at 1 °C min⁻¹ to 350 °C in a 5% H₂−Ar gas atmosphere for 2 h. The metallic Cu⁰ powders were then obtained after naturally cooling to room temperature. The GDE with Cu⁰ powders was prepared according to the procedure described for the preparation of electrode and the Cu⁰ electrocatalyst was formed after the subsequent electroreduction.

Preparation of electrodes

Typically, the Cu₂O−Cu⁰ or Cu⁰ catalyst (6 mg), Nafion (3 μL) and aQAPS-S₁₄ solution (35 μL) were dispersed in isopropanol-water solution (the volume ratio of isopropanol to water: 3:1, 200 μL) by sonication for 1 h to form a homogeneous ink. The GDE was prepared by dispensing the catalyst ink (200 μL) onto the GDL in four aliquots using a pipette and was dried at 60 °C for 0.5 h. The same procedure was adopted for the preparation of the control electrode. Unless otherwise specified, the catalyst loading was 1.5 mg cm⁻².

MEA electrolyser and electrocatalytic reactions

The full-cell electrocatalytic reaction was carried out in a MEA electrolyser. Two polytetrafluoroethylene (PTFE) gaskets were placed between the cathodic and anodic flow plates to avoid short-circuiting. A fixed tightening force (3 N m) was applied on the plates to hold cathodic GDE (2 cm × 2 cm), AEM (4 cm × 4 cm), and anodic IrO_x/Ti mesh (3 cm × 3 cm) together. A CO₂ − H₂O gas flow (typically 50 mL min⁻¹) was fed to the cathode side and the partial pressure of H₂O was controlled by a humidity generator (Suzhou friends experimental equipment co., LTD, Supplementary Fig. 3). An aqueous solution of KHCO₃ (0.1 M) was employed as the anolyte and was circulated at a flow rate of 2 mL min⁻¹, unless otherwise stated. Typical electrolytes, 1.0 M KOH and 0.1 M KHCO₃, were prepared by dissolving specific amounts of KOH or KHCO₃ into ultrapure water. The electrolytes were prepared before testing and used immediately. The pH of the 0.1 M KHCO₃ and 1.0 M KOH solutions were 8.30 ± 0.02 and 13.80 ± 0.03, respectively, as measured with a pH meter (Shanghai Lei-ci Instrument Co., Ltd, PHS-3C). The AEM was activated with OH⁻ conductor in 1.0 M KOH solution at 60 °C for 24 h, and then was washed with pure water. The electrolysis was controlled by Ivium-4-Stat electrochemical workstation (Ivium Technologies BV), operated with Iviumsoft software (4_960). All electrochemical data were exported in ASCII format through Iviumsoft for further analysis and plotting. For the CO₂RR performance evaluation, a constant voltage of 2.3 V was first applied for 10 min to build a stable gas-liquid transport channel, and then the voltage was gradually increased by 0.1 V every 5 min until the target value. All electrochemical experiments were conducted at ambient pressure and temperature, maintained between 25 and 30 °C. All electrocatalytic CO₂RR measurements conducted in the MEA electrolyser were not corrected for IR drop. The recorded voltages are the settings from the electrochemical workstation, while the currents are the average values of the reactions over the specified time periods. For the stability test, a 0.1 M KHCO₃ electrolyte solution of 200 mL was circulated, with the electrolyte being replaced every 50 h.

Product analysis and performance evaluation for CO₂RR

The gas-phase products formed during the CO₂RR were analysed and quantified by an online gas chromatograph (GC 2060) equipped with packed columns connected to a thermal conductivity detector (TCD) and a flame ionization detector (FID). Argon (Messer gas, 99.99%) was employed as the carrier gas. The gas-phase products were collected after an initial reaction time of at least 5 min and at least twice for each reaction. To calculate the Faradaic efficiency of the gaseous products and other performance indicators, a soap film flowmeter was connected to the cathode outlet of the MEA electrolyser to measure the real-time flow rate during the reaction. At each potential, measurements were taken twice and averaged. The liquid-phase products were analysed by ¹H nuclear magnetic resonance (¹HNMR) spectroscopy and the NMR spectrum was recorded on an Advance III 500-MHz Unity plus spectrometer (Bruker). Dimethyl sulfoxide (DMSO) was used as the internal standard. Typically, the mixture of the electrolyte (0.50 mL) and DMSO (0.10 mL) diluted to 100 ppm (v/v) by deuterated water was used for measurement. The details of the calculation methods for CO₂RR performance parameters were displayed in the Supplementary Information.

Electrochemical CO stripping measurement

The electrochemical CO stripping measurement was conducted in an H-cell. In brief, the working electrode was prepared following the same method used for performance evaluation in the MEA. The effective electrode area was 4 cm², and 0.1 M KHCO₃ aqueous solution was employed as the electrolyte. The electrode was pre-reduced at a potential of −1.8 V versus Ag/AgCl for 10 min under a CO₂ atmosphere. Subsequently, the electrode was held at −1.1 V versus Ag/AgCl for 15 min to allow CO to be adsorbed on the electrode surface under CO atmosphere, and then excess non-adsorbed CO was removed by purging Ar gas for 10 min. Finally, CO was stripped by conducting CV from 0 to 1.0 V versus Ag/AgCl at 50 mV s⁻¹. The CO stripping curve was obtained by subtracting the background from the CO oxidation currents.

Characterizations

Powder X-ray diffraction (XRD) patterns for the fresh and used catalysts were recorded on a Rigaku Ultima IV diffractometer using Cu K_α radiation (40 kV, 30 mA). The operando XRD patterns were recorded on a Bruker D8 Discover diffractometer using Cu K_α (40 kV, 40 mA) radiation and the catalyst sample was placed in a modified MEA cell with 0.10 M KHCO₃ as the anolyte (Supplementary Fig. 18). The XRD pattern was collected after 30 min of reaction at each investigated voltage. The diffraction pattern was analysed by Rietveld refinement using the TOPAS software. SEM measurements were performed on a Hitachi S-4800 operated at 15 kV. TEM and HRTEM measurements were carried out on a Phillips Analytical FEI Tecnai 20 electron microscope operated at an acceleration voltage of 200 kV. The XANES and EXAFS measurements were performed at the BL14W1 beamline of the Shanghai Synchrotron Radiation Facility using Si(111) mode. This crystal orientation was chosen for its high signal intensity and stability, which are particularly favourable for Cu elements, resulting in efficient and reliable XAS measurements. The operando measurements were recorded in a modified MEA cell for operando XAS studies (Supplementary Fig. 21), and the data were collected after 10 min of reaction at each investigated voltage. The data were analysed using the software package Athena.

Raman spectroscopic measurements were conducted on a LabRAM HR Evolution Raman Microscope (Horiba Jobin Yvon) with a laser wavelength of 632.8 nm. For in situ Raman measurements in an H-cell, a water immersion objective was selected as the Raman lens. The catalyst ink, which was the same as that used for the preparation of the gas diffusion electrode (GDE) in our work, was deposited onto a titanium (Ti) substrate (diameter 3 mm, GaossUnion Optoelectronic Technology Co., Ltd) by pipette to serve as the working electrode. A saturated Ag/AgCl electrode (3 M KCl, RE-AgCl-1038, GaossUnion) and a Pt wire (diameter 0.5 mm, GaossUnion) electrode were used as the reference and the counter electrodes, respectively. A CO₂-saturated 0.10 M KHCO₃ aqueous solution was employed as the electrolyte. The pH of the CO₂-bubbled 0.1 M KHCO₃ was 6.82 ± 0.02, which was measured with a pH meter. After applying potential for 2 min, the Raman spectrum was recorded with an accumulation time of 30 s. For all measurements conducted in a three-electrode H-cell, the applied potential was converted to RHE using the following equation (Eq. 1):

$$E_{\mathrm{RHE}}\left(\mathrm{V}\right)=E_{\mathrm{Ag}/\mathrm{AgCl}}+0.197+0.0592\times \mathrm{pH}$$ 1

The operando Raman measurements were performed using a modified MEA cell (Supplementary Fig. 25). An air objective was selected as the Raman lens. The catalyst ink, which was the same with that for preparation of electrodes for the evaluation of CO₂RR performance, was then sprayed on the AEM (3.5 × 3.5 cm²) instead of GDL. The spray pen used air as the carrier gas and the AEM was adsorbed onto a vacuum adsorption table heated to 70 °C. The spray flow rate was approximately 0.1 mL min ^− 1, and the effective area of sprayed catalyst was 1 cm². The final catalyst loading was about 1.5 mg cm ^− 2. The AEM sprayed with catalyst was activated by immersing it in a 1.0 M KOH solution at 60 °C for 24 h, followed by rinsing with pure water. The modified MEA cell was assembled using the same method as for the evaluation of CO₂RR performance, except that the GDL used had a circular hole with a diameter of about 1 mm for collecting Raman signals. After applying an initial voltage for 10 min, the Raman spectra were recorded with an accumulation time of 30 s.

DFT calculations

The DFT calculation was performed by the Vienna ab initio simulation package (VASP 5.4.4), a plane-wave DFT software package^46,47. The projector-augmented wave potentials were used to describe the core-valence interaction⁴⁸ with a plane-wave cutoff energy of 400 eV. The Perdew–Burke–Ernzerhof functional within the generalized gradient approximation was used to calculate the electron exchange and correlation energies^49,50. All the DFT calculations used the (3 × 3 × 1) Monkhorst-Pack k-point meshes to sample the Brillouin zone in reciprocal space⁵¹. The convergence criterion of energy and force was set as 0.05 eV Å^–1 and 10 ^− 5 eV, respectively. The Cu₂O−Cu⁰ model was optimized by relaxing lattice constants from a (2 × 2) three-layer Cu₂O(111) slab by removing half of the oxygen atoms. The Cu(111) surface was modeled by a (4 × 4) three-layer slab. Their bottom two layers were fixed and the vacuum space was 20 Å to avoid interactions with their periodic images. The software package VASPsol was used during all calculations to describe implicitly the effects of electrolytes and solvents^52,53.

The Gibbs free energy of each reaction state was calculated by Eq. 2 as follows:

$$G=E_{\mathrm{total}}+\mathrm{ZPE}-T\times S$$ 2

The total energy (E_total) was gained directly by DFT calculations, and the zero point energy (ZPE) correction and entropy (TS) correction are calculated via DFT calculations of the vibrational frequencies.

The calculation of Gibbs reaction free energy of proton-coupled electron transfer (PCET) processes was simplified by the computational hydrogen electrode (CHE) model proposed by Nørskov et al. ⁵⁴. In these schemes, the concerted proton and electron transfer was assumed and the corresponding free energy was referred to the chemical potential of gas-phase H₂ at 0 V versus RHE. As shown in Eq. 3:

$$\mu \left[{\mathrm{H}}^{+}+{\mathrm{e}}^{-}\right]=0.5\mu \left[\mathrm{H}\left(\mathrm{g}\right)\right]$$ 3

Author contributions

X.H. and L.L. conducted most of the experiments and analysed the data. X.L. performed the DFT calculations and analysed the data. M.Z. conducted Raman spectroscopic measurements and analysed the data. Q.Z. analysed the data and revised the paper. S.X. designed the study, guided the electrochemical measurements and co-wrote the paper. B.M. and F.S. performed the XAS measurements at the BL14W1 beamline of SSRF. Z.J. guided the XAS studies and analysed the data. J.C. guided the DFT calculations and analysed the data. Y.W. designed the study, supervised the project and co-wrote the paper.

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.

Data availability

The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files. Source data are provided as a Source data file. 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/s41467-024-54282-2.