Revealing the CO Coverage-Driven C–C Coupling Mechanism for Electrochemical CO2 Reduction on Cu2O Nanocubes via Operando Raman Spectroscopy

Electrochemical reduction of carbon dioxide (CO2RR) is an attractive route to close the carbon cycle and potentially turn CO2 into valuable chemicals and fuels. However, the highly selective generation of multicarbon products remains a challenge, suffering from poor mechanistic understanding. Herein, we used operando Raman spectroscopy to track the potential-dependent reduction of Cu2O nanocubes and the surface coverage of reaction intermediates. In particular, we discovered that the potential-dependent intensity ratio of the Cu–CO stretching band to the CO rotation band follows a volcano trend similar to the CO2RR Faradaic efficiency for multicarbon products. By combining operando spectroscopic insights with Density Functional Theory, we proved that this ratio is determined by the CO coverage and that a direct correlation exists between the potential-dependent CO coverage, the preferred C–C coupling configuration, and the selectivity to C2+ products. Thus, operando Raman spectroscopy can serve as an effective method to quantify the coverage of surface intermediates during an electrocatalytic reaction.

the ultrahigh vacuum (UHV) system where the XPS chamber is located to allow the sample transfer without air exposure ( Figure S7). The electrochemical measurements were carried out using a potentiostat (Autolab PGSTAT 302N).

Operando Raman experiments
The operando Raman spectra were obtained by means of a Renishaw (InVia Reflex) confocal Raman microscope with a 785 nm laser. To perform the operando experiments in an electrolyte, a water immersion objective with a long working distance (Leica microsystems, 63x, numerical aperture of 0.9) was chosen. The objective with a long working distance is needed to avoid diffusion hindrance problems during the Raman measurements. The laser power was about 0.36 mW. The acquisition time was 10 s for the steady-state experiments at different potentials and 5 s for the time-dependent experiments. For the operando measurements, the objective is protected from the electrolyte by a Teflon film (DuPont, film thickness of 0.013 mm). A drop of water is used to drive away the air between the film and the objective to match the refractive index, which ensures efficient excitation and collection of the Raman signal. The electrochemical measurements were performed in a home-built spectroelectrochemical cell made of Teflon and controlled by a Biologic SP-240 potentiostat ( Figure S10). The cell was equipped with a reference electrode (leak-free Ag/AgCl, Alvatek), a counter electrode (Pt ring), and a working electrode with the catalyst dropcasted on glassy carbon. Typically, a 15 ml CO2-saturated 0.1 M KHCO3 solution was used as electrolyte, and CO2 was continuously injected into the solution during the experiment. Ar-saturated 0.1 M KHCO3, Ar-saturated 0.1 M NaClO4, CO-rich 0.1 M KHCO3 were used in the experiments as well. We mixed the CO-saturated 0.1 M KHCO3 solution with the Ar-saturated 0.1 M KHCO3 solution to prepare the 33% COrich 0.1 M KHCO3 and 66% CO-rich 0.1 M KHCO3. The percentage represents the volume fraction of CO-saturated KHCO3 in the electrolyte. For the steady-state experiment, each potential was applied for at least 10 min before collecting the spectra to ensure steady-state conditions at the surface of the catalyst. For the time-dependent experiment, we acquired three spectra at open circuit potential, then applied the potential and continuously recorded the Raman spectra every 5 s.

CO2RR testing
Electrocatalytic measurements were performed with an Autolab (Metrohm) potentiostat in a H-type cell equipped with an anion exchange membrane (Selemion AMV, AGC Inc). The three-electrode system consisted of the catalyst deposited on carbon paper as working electrode, a platinum gauze (MaTecK, 3600 mesh cm −2 ) as counter electrode, and a leak-free Ag/AgCl reference electrode (LF-1, Alvatek). A 0.1 M KHCO3 aqueous solution was used as electrolyte and saturated with CO2 (99.995%) for at least 20 min prior to the measurements. The CO2 flow was 20 ml/min. The electrochemical protocol consisted of a linear sweep voltammogram from the open circuit potential to the cathodic potential followed by chrono-amperometry at this potential for 3600 s. All potentials are given versus the RHE scale and were corrected for the iR drop. Each presented data point corresponds to an identical freshly prepared sample following this protocol at different potentials. The electrochemical surface roughness factor was estimated from double layer capacitance measurements. 2 Gas products were detected and quantified every 15 min by online Gas Chromatography (GC, Agilent 7890B), equipped with a Thermal Conductivity Detector (TCD) and a Flame Ionization Detector (FID). Liquid products were analyzed after each measurement with a high-performance liquid chromatograph (HPLC, Shimadzu Prominence), equipped with a NUCLEOGEL SUGAR 810 column and a refractive index detector (RID), and a liquid GC (L-GC, Shimadzu 2010 plus), equipped with a fused silica capillary column and a FID detector. All catalytic results in this study are shown in terms of Faradaic efficiency (F.E.). The Faradaic efficiency of the gas product x was calculated as: Eq. (1) and for the liquid product x was calculated as Here . : Faradaic Efficiency of product . ̇: CO2 gas flow rate / l s -1 . : Volume-fraction of the product detected by GC. : Electrons transferred for reduction to product . : Faradaic constant / C mol −1 . : Geometric area of the electrode / cm −2 .
total : Total current density during CO2 bulk electrolysis / A cm −2 . Δ : Final concentration of product detected by HPLC and liquid GC / mol l −1 . Δ : Total charge transferred during electrolysis at const. potential or current/C. : Volume of the electrolyte / l.

Computational Details
The Density Functional Theory (DFT) calculations were conducted using the PBE density functional 3 within VASP 4,5 . To account for interactions between catalyst surface and adsorbates, we included dispersion through the DFT-D2 method, 6,7 with our reparametrization of the C6 coefficients for metals. 8 CO adsorption energies were corrected for implicit solvation contributions within the VASP-MGCM framework. 9,10 Inner electrons were represented by PAW pseudopotentials 11,12 and we expanded the monoelectronic states for the valence electrons as plane waves with a kinetic energy cutoff of 450 eV. We sampled the Brillouin zone by a Γ-centered k-point mesh from the Monkhorst-Pack method, 13 with a reciprocal grid size smaller than 0.03 Å −1 . Hubbard corrections were applied via the Dudarev approach 14 to the 2p orbitals of C and O to tune the HOMO-LUMO gap of the CO molecule. The U parameter was varied between 1.25 and 2.00 eV, whilst J was always kept equal to 1 eV. The resulting Ueff = U -J parameter was then in line with DFT studies on site preferences on Cu(100) ( Figure  S31). 15 To assess electric field effects ( Figure S32), we employed the corresponding tag in VASP which introduces it via a dipole correction. 8,16 We modeled the oxide-derived Cu catalysts as Cu(100) slabs at least four layers thick. The two outermost layers were fully relaxed to allow surface reconstruction, whilst the rest fixed to mimic the bulk. The vacuum between the slabs was always larger than 10 Å. Different surface structures were employed to study CO surface coverage, θCO, between 0.11 and 0.88 ML, 17 Table S2, and the initial CO configurations were retrieved from a theoretical study on high CO coverage adsorption configurations on Pt(100). 18 CO molecules were placed only on one side of the slab, thus, we applied a dipole correction to remove spurious artifacts from the asymmetric slab model. 19 To benchmark spectroscopic experimental evidences, we calculated the vibrational modes for the lowest energy CO adsorption configuration at different surface coverages, Table  S2. We then determined the intensity ratio of the Cu-CO stretching band and CO restricted rotation from the superposition of these vibrational modes associated to different CO adsorption sites ( Figure S33, Table S4-S6). To calculate CO-CO activation barrier, we employed a simplified computational setup with low CO coverage, implicit solvation, 9 and no electric field applied. In general, coverage and electric field effects are reported to stabilize CO-CO dimerization on Cu(100) by 0.2 eV and 1.0 eV, respectively, [20][21][22] and both factors affect CObridge and COatop configurations equally ( Figure S32). Thus, our simplified approach does not alter the overall trend among different adsorption sites. Transition states for C-C coupling were located through the Climbing Image Nudged Elastic Band (CI-NEB) method and all of them exhibit a single imaginary vibrational frequency. 23 CO adsorption energies and kinetic barriers at θCO = 0.11 ML were reported using as references: CO2(g), H2(g), and the pristine Cu surfaces, in line with the Computational Hydrogen Electrode formalism. 24,25 Gibbs free energies (G) were calculated at 298.15 K by correcting DFT energies (E) for entropic effects.         . Current densities of Cu2O nanocubes on carbon paper normalized to the electrochemical surface area as a function of the applied potential obtained after 1 h of CO2RR. It indicates that the system is potential-controlled in the potential range presented. The catalyst surface roughness factor, estimated from double layer capacitance measurements, is 4.6.   It can be found that the peak at about 360 cm -1 , which was at ~390 cm -1 at 0.4 VRHE, co-exists with the peaks at 1077 cm -1 and 706 cm -1 , which are assigned to the carbonate species and surface hydroxyl species, respectively (Table S1).        Individual error bars of P1 and P2 were obtained as the standard deviation and correspond to the average of three repeated measurements conducted on three identically prepared samples. The P2/P1 ratio was obtained as an average of the (P2/P1)i ratios extracted for i=1,2,3 different measurements. The uncertainty of the P2/P1 ratio was also defined as the standard deviation. The error bars of the average peak intensities (P1 or P2), Figure S21, coming from the intensity differences observed for each of these peaks in three different Raman spectra, are much larger than those of their average ratio (P2/P1), which was calculated by normalizing every (P2)i by the (P1)i from the same Raman spectrum (i). The error bar of the averaged P2/P1 was also obtained as the standard deviation and is shown in Figure 2c. Using the P2/P1 ratio instead of specific P2 and P1 intensities for comparison minimized the influence of the nanostructuresensitive surface enhancement effect, which generally hampers the quantification of adsorbate surface concentrations (e.g. CO) directly from the spectral Raman intensity.
The formulas used to calculate error margins (standard deviation, SD) for the individual P1 and P2 and for the ratio R = P2/P1 are shown below, with i=1,2,3 indicating the three individual measurements conducted:          10-layer Cu slab thickness. A thorough theoretical assessment of site specific CO adsorption through DFT is partially hindered by the underestimation of the HOMO-LUMO gap for CO, which leads to an overestimation of metal/adsorbate interactions. 15 In the literature, a Hubbard correction Ueff (U -J) for the C or O 2p levels between 0.25 and 1.0 eV has been employed to improve DFT accuracy, leading to the reversal of the adsorption site preference between bridge (no U) and atop (Ueff > 0.25) for 0.25 ML CO coverage on Cu(100). 15 We repeated the same benchmark here and for any value of Ueff CO adsorption on a bridge site was more favorable than atop. Figure S32. Electric field within the electrical double layer stabilizes CO adsorption on a bridge site via electric dipole / field interaction. (a)-(d) For a CO surface coverage lower than 0.67 ML, DFT predicts that CO adsorption is favored on bridge rather than atop sites, in disagreement with experimental results by thermal desorption spectroscopy (TDS) under ultra-high vacuum (UHV) conditions. 15,27 However, CObridge adsorption is further stabilized by the electric fields applied, which may revert the preferential adsorption site at low CO coverage under electrochemical conditions. Electric field stabilization is due to the higher normal electric dipole moment for CObridge (0.11 |e -| Å vs 0.03 |e -| Å for COtop). 28 . (e)-(i) At high coverage local CO-CO repulsion lowers CO binding energy and the stabilization effect of the electric field becomes independent of CO adsorption configuration. Regression parameters are reported in Table S3. Figure S33. Theoretical vibrational spectra for different CO coverages, θCO, on Cu(100). A Gaussian function was centered in the Cu-CO stretching band (P2) and CO restricted rotation (P1) vibrational modes and the resulting Gaussian peaks from different adsorption configurations (atop, bridge, etc.) added up to define the spectral lines. For CO surface coverage higher than 0.60 ML, the most stable configuration is a mix of COatop-CObridge population, whose adsorption is independent of applied electric field and COatop/CObridge ratio ( Figure S32, Table S2). Thus, multiple spectra lines are represented. *COatop represents the percentage of CO adsorbed on an atop Cu site. Tables: Table S1. DFT vibrational frequencies, ν (cm -1 ), in the region between 200 cm -1 and 4000 cm -1 for relevant adsorbates on Cu(100). For each adsorbate, we report the surface coverage, θ (ML), and the adsorption configuration. The initial configurations were retrieved from a previous study on Pt(100). 18 The systems are defined as a surface supercell, number of adsorbed CO, and the adsorption sites (T: top, B: bridge). *COatop represents the percentage of CO adsorbed on top Cu sites, vs. CO adsorbed on bridge sites. ∆ ̅ * CO (eV) is the average CO binding energy calculated with CO2, H2, and H2O as reference energies. The lowest energy configurations for each surface coverage are highlighted in grey, although at low CO coverage DFT has shown limitations in correctly predicting the preferential adsorption site. 15 For high coverage, different configurations of mixed COatop-CObridge populations are isoenergetic within DFT-D2 typical error bars (±0.1 eV). The vibrational spectra associated with Cu-CO stretching band and CO restricted rotation modes are shown in Figure S33.  3.53 ± 0.08 Table S7. Structural parameters and Bader charges for the initial state CO(1)-CO(2) configuration toward C-C coupling from different CO-CO precursors, shown in Figure  S34. COb: CO on bridge site. COt: CO on atop site.