Introduction

The extensive consumption of coal, oil and other fossil fuels has resulted in the emission of enormous amounts of carbon dioxide (CO2) into the surrounding atmosphere, which is the culprit of the global greenhouse effect1,2,3,4. Therefore, the management of CO2 concentrations in the atmosphere has become an important topic5. The use of clean carbon-neutral fuel systems, CO2 sequestration, and CO2 transformation into high value−added products are several strategies for inhibiting continuous increases in or even reducing the CO2 concentration in the atmosphere6. Among these techniques, electrochemical CO2 reduction (ECR) driven by clean and renewable electricity sources (e.g. solar energy) is particularly promising7,8, and can synthesize a wide variety of chemicals, such as formic acid, carbon monoxide (CO), alcohol and methane, along with the elimination of CO29,10,11,12,13. Among the synthesized chemicals, CO is regarded as having the most potential to achieve positive gross margins in terms of techno-economic assessments6,14,15. However, due to the small difference in standard reaction potentials for different products, it is a substantial challenge to achieve high selectivity towards one specific chemical, which requires the adoption of highly selective and efficient electrocatalysts16.

Among various electrocatalysts for ECR to CO investigated over past decades, metal-based materials11, especially metallic Ag17,18, may be the most promising candidate for industrial development in view of their remarkable performance and economic viability compared to noble Au-based materials19. It has been found that the performance (selectivity, activity and stability) of Ag-based catalysts is strongly related to their morphologies, shapes, particle sizes, crystal facets, electronic structures and so on17. Unmodified polycrystalline Ag metal, such as traditional Ag foil, shows an unsatisfactory CO faradaic efficiency (FE) of only 70–80%. Various design strategies for novel Ag catalysts17, such as shape controlling20, alloying21 and the construction of nanostructures22,23, have therefore been proposed to optimize the performance of ECR to CO. In particular, defect engineering has been identified as a significant method to enhance reactivity3,24. For instance, it has been proposed that plasma oxidation pretreatment of Ag foil could introduce an enhanced content of low-coordinated active sites, leading to an improved FE and a remarkably reduced overpotential (η) for ECR to CO25. Electrochemical reduction of metal oxides, such as Au oxides and Cu oxides, could also introduce vacancies or other defects to result in better performance26,27; this approach is highly attractive because of its simplicity compared to plasma treatment. However, how electrochemical reduction facilitates defect generation is still unclear.

Advanced operando X-ray, optical and electron-based characterizations may provide useful information about defect generation under real reaction conditions28,29,30,17. Although the DFT calculations demonstrated the strengthened binding of intermediate COOH by defect sites, which accounts for the enhanced FE for ECR to CO, the mechanism for the dramatically inhibited HER is still ambiguous. Examining the possible reaction pathways of HER and ECR to CO may provide some useful clues. As shown in Fig. 5c, it is believed that HER in acidic electrolytes begins from a Volmer reaction, where an electron transfer is coupled with a proton adsorption on an empty active site of the electrode. After the formation of an absorbed hydrogen atom, two different reaction processes, the Tafel reaction or Heyrovsky reaction, may occur to yield the final H2 product77. In contrast, the reaction pathways for ECR are quite complex due to the multi-electron and multi-proton transfer steps. For ECR to CO, the COOH is generally regarded as a key intermediate74,78, which is also reported in our present work. According to Kortlever et al., the formation of *COOH may have two different mechanisms78, as shown in Fig. 5d. In the two-step mechanism (purple arrows), the electron and proton transfer processes occur in two separate steps, while in another mechanism (light green arrows), the proton transfer is coupled with the electron transfer (PCET). Regardless of the initial step, the formed *COOH will subsequently react with another electron and proton to yield H2O and *CO. Then *CO desorbs from the electrode to release CO and leaves an empty active site39. As shown in Fig. 5a, on the defect Ag surface, the COOH and CO stably bind at the vacancy edge sites where two Ag atoms are occupied, which likely blocks the adsorption of protons to Ag, leading to the poor selectivity of the HER.

Discussion

In summary, the nanostructured Ag-D catalyst exhibits outstanding performance for ECR to CO, both in an H-cell reactor and a flow-cell reactor. Fast operando XAS measurements and ex situ SEM and TEM images revealed that massive defects are efficiently created at the initial stage of ECR. Carbon dioxide in the electrolyte is proved to undertake a role in the protection of these defect sites since their formation. We proposed the concept of the RCd based on the coordination numbers to quantify the defects, which may give insights to the future work on the quantitative discussion of the defects, especially for the metal and metallic oxides. DFT calculations demonstrated that the adsorption of intermediate COOH is strengthened on Ag atoms surrounding the defect sites, benefiting the activation of CO2. Therefore, combining the experimental and theoretical results, the origin of the high performance of Ag-D catalyst was attributed to the abundant defect sites. Our work not only highlights the importance of operando characterization techniques with high time resolution, but also provides new insights into Ag-based catalysts for ECR.

Methods

Sample preparation

The Ag2O nanoparticles were synthesized by a precipitation reaction. In detail, 0.02 mol AgNO3 was first dissolved in a volume of 200 mL deionized water. Then, a volume of 100 mL sodium hydroxide solution (0.2 M) was dropped in the AgNO3 solution under stirring to enable a homogenous mixture. The final product of Ag2O was collected after filtration and drying at ~60 °C. Ag-D was prepared by operando electrochemical reduction of Ag2O, which endowed massive defects in Ag-D. Polycrystalline Ag foil (0.127-mm thick, 99.9%, annealed) was purchased from Alfa Aesar. Ag powder was purchased from Aladdin.

Characterization

XRD was conducted on a Rigaku Smartlab with filtered Cu Kα radiation (λ = 1.5418 Å) to determine phase constitutions and crystal structures. XPS was carried out on a PHI5000 VersaProbe spectrometer using an Al-Kα X-ray source to analyse element states of the electrocatalysts. SEM images were obtained from a scanning electron micro-analyzer (HITACHIS4800). TEM images (JEOL JEM 2100, 200 kV) were collected to uncover catalysts’ morphology and structure.

Electrode preparation

The electrode ink for Ag-D was prepared as follows. Pristine Ag2O powder (10 mg) and Super P Li (10 mg) were suspended in a mixture containing 5 wt% Nafion solution (100 μL) and ethanol (1 mL) via ultrasound. Then, 30 μL of the above homogeneous ink was spread on a prepolished L-type GCE with a diameter of 8 mm. The loading of Ag2O in this electrode was ~0.543 mg cm−2. Polycrystalline Ag foil served as the working electrode for comparison after thorough ultrasonic cleaning in acetone and ethanol.

For the fabrication of the cathodic GDE, the same ink was dropped onto the gas diffusion layers (29BC, SGL) with an Ag2O loading of ~2 mg cm−2. To avoid the possible ‘flooding’ of GDE, another commercial PTFE film (~0.22 μm pore size) was fitted on top of the GDE.

ECR performance measurements and product analysis

Electrochemical tests were conducted on a workstation (CHI 760E) at 25 °C. A gas-tight H-cell reactor segregated with a Nafion-117 membrane was used for the electrolysis experiments. In the cathodic chamber (100 mL), an L-type GCE and Ag/AgCl served as the working and reference electrode, respectively. A Pt foil in the anodic chamber (100 mL) served as a counter electrode. Each chamber contained 50 mL KHCO3 (99.99%, Macklin) electrolyte (0.1 M, pH = 6.8). Carbon dioxide was continuously purged (30 mL min−1, 99.999%, Shangyuan) during the ECR process. All potentials were recalculated into RHE by ERHE = EAg/AgCl + 0.1989 + 0.0591 × pH according to Nernst equation unless otherwise stated54,65, and iR compensation was also taken into consideration.

For the measurements in the flow cell, the as-prepared cathodic GDE (geometric active area of 1 cm2) was separated from the anode with a commercial anion exchange membrane (PK130, Fumatech). A Pt plate acted as the anode. Since 1 M KOH electrolyte was used, the Hg/HgO electrode was mounted in the catholyte stream at a fixed location, serving as the reference electrode. All measured potentials vs. Hg/HgO were recalculated into RHE through ERHE = EHg/HgO + 0.098 + 0.0591 × pH according to the Nernst equation unless otherwise stated. The catholyte (~30 mL) was circulated by a peristaltic pump (BT100-2J, Longer).

Gaseous products were analyzed by a gas chromatograph (GC, HOPE 9860) with a thermal conductivity detector (TCD) and flame ionization detectors (FID). Nuclear magnetic resonance spectroscopy (1H NMR, Bruker ACF-400) was used to analyze the liquid-phase products.

The cathodic energy efficiency (EE) was calculated as follows79,80:

$${\mathrm{EE}}\left( \% \right) = \frac{{1.23 - {\mathrm{E}}^0}}{{1.23 - {\mathrm{E}}}} \times {\mathrm{FE,}}$$
(1)

where E0, FE and E represented standard potential, faradaic efficiency and applied potential, respectively.

Fast operando XAS measurements

For the conventional XAS, it usually takes 15–30 min to obtain a single spectrum, and its quality may be unsatisfactory due to some accidental disturbances. Therefore, employing conventional XAS to study the chemical information of the catalyst may give confusing results, since the real state of the catalyst may change quickly with time. Moreover, the tracking of the changing process has been pursued and received growing attention in order to gain a comprehensive understanding on the reconstruction of catalysts28. Obviously, advanced characterization techniques with high time resolutions are essential for obtaining the desired information. The fast XAS technique introduced here can provide not only a clear evolution information on the catalyst, but also a reliable data quality, both of which are crucial for the further analysis on the structure of catalysts. Fast operando XAS was collected at the 44 A beam line of TPS. The 44 A beam line equipped with a quick-scanning monochromator (Q-Mono) can provide 120 sheets of spectra per minute. By averaging these spectra, a good quality of the final spectra can be ensured. A PMMA reactor was prepared for fast operando experiments, as shown in Supplementary Fig. 5. Catalysts were hand-brushed onto carbon-fibre paper serving as a working electrode. Carbon dioxide was constantly bubbled into the electrolyte (0.1 M KHCO3). A potential of −1.5 V vs. Ag/AgCl was applied to the working electrode to study the evolution of catalysts. All spectra were obtained in transmission mode and analyzed using Athena and Artemis81. The k range for Fourier transforms (FT) was from 2.5 to 12.2 Å-1. Fitting of the EXAFS spectra χ(k)k2 for Ag–O and Ag–Ag bond was conducted in R space in the range from Rmin = 1.2 Å to Rmax = 3.5 Å. When fitting the single peak for Ag–Ag bond, the Rmin was set to 2.16 Å. For Ag-D after 6, 8 and 10 min, the high shells were fitted, respectively. The Rmax was accordingly set to 6.0 Å. Phase corrections were taken into consideration during the fitting process.

DFT calculations

To obtain a deeper insight into the effects of Ag defects on electrocatalytic reactions, DFT calculations were performed. In the current study, we built three vacancy models with point defects as representatives of defect structures to investigate their influence on ECR. The Ag (111) crystal surface is selected on the basis of the experimental observation and theoretical consideration. The Ag-D catalyst exhibited the strongest peak for (111) facet in XRD pattern, which was also frequently observed in HRTEM images. DFT calculations based on the selected Ag (111) plane was also reported in literatures for the mechanism studies20,25,82. Thus, it is reasonable to choose the representative Ag (111) facet to perform the DFT calculations in the present work. A three layer p(3 × 3) (111) slab with 20 Å of vacuum avoiding imaging interactions in the z-axis was constructed to represent the Ag pristine model. Meanwhile, the Ag (111)-vacancy system was built by removing one Ag atom from the optimized pristine Ag (111) system, and then further geometry optimization was performed to relax the system. Since the CN and DW factors reflect the coordination environment and structural disorder of the catalysts, accordingly, in the DFT studies, the influence of defects on the Ag (111) surface structures was also taken into consideration by optimizing lattice parameters and atomic positions of the slab. As shown in Supplementary Table 4, the existence of the vacancy does not change the cell shape of the slab, but leads to slight variation of the cell volume. This indicates that the vacancy gives rise to structural changes to some extent. For the clarity, the vacancy concentration is defined as \({\mathrm{C}}_{\mathrm{v}} = \frac{{{\mathrm{N}}_{\mathrm{v}}}}{{{\mathrm{N}}_{{\mathrm{total}}}}}\), where Nv is the number of vacancy, and Ntotal is the number of atoms of the first atomic layer in Ag (111)-pristine slabs. In this case, Cv = 1/27 ≈ 4%, and the model is referred to as Ag (111)-4% vacancy. The p(2 × 2) (111) slab models were constructed to simulate higher vacancy concentrations. By removing one and two Ag atoms, we obtained the models with Cv of ca. 8% and 17%, respectively (Supplementary Fig. 19). From Supplementary Table 4, we can see that when Cv is ca. 17%, the vacancies lead to slight variation of the cell shape compared with lower vacancy concentration carrying models, which indicates larger structural fluctuations with respect to the pristine structure. Note that, the Cv is not exactly equivalent to the RCd. The aim to introduce the Cv into DFT calculations is for the quantitative discussion on the effects of defect structures. For all models, the bottom layer was immobilized at the optimized lattice constant (2.03 Å), agreeing with the value of 2.04 Å obtained in experiments83. Initial adsorption configurations of COOH and CO intermediates at four sites (top site, bridge site, fcc hollow site and hcp hollow site, Supplementary note 1 and Supplementary Fig. 21) were constructed to study the adsorption properties.

The Vienna Ab-initio Simulation Package (VASP) was used to conduct periodic DFT calculations84. Electron-core interactions were described by the projector augmented wavefunction (PAW) method85. The Perdew-Burke-Ernzerhof (PBE) functional in a plane wave pseudopotential implementation within the generalized gradient approximation (GGA) was applied for describing the electron exchange-correlation86. Grimme’s semi-empirical correction for the dispersion potential (DFT-D3) was applied to include the long-range dispersion effect87. Plane wave cut-off energies of 700 eV and 450 eV were chosen for the bulk primitive crystal cell and slab models, respectively. The Monkhorst–Pack technique was used to automatically generate the gamma centred 3 × 3 × 1 k-points grid in the Brillouin zone88. The width of 0.1 eV was employed for Gaussian smearing. The electronic self-consistent iteration convergence criterion was 1.0 × 10−6 eV. The convergence criteria for the force was 0.01 eV/Å for each free atom.