Operando time-resolved X-ray absorption spectroscopy reveals the chemical nature enabling highly selective CO2 reduction

Copper electrocatalysts have been shown to selectively reduce carbon dioxide to hydrocarbons. Nevertheless, the absence of a systematic study based on time-resolved spectroscopy renders the functional agent—either metallic or oxidative Copper—for the selectivity still undecidable. Herein, we develop an operando seconds-resolved X-ray absorption spectroscopy to uncover the chemical state evolution of working catalysts. An oxide-derived Copper electrocatalyst is employed as a model catalyst to offer scientific insights into the roles metal states serve in carbon dioxide reduction reaction (CO2RR). Using a potential switching approach, the model catalyst can achieve a steady chemical state of half-Cu(0)-and-half-Cu(I) and selectively produce asymmetric C2 products - C2H5OH. Furthermore, a theoretical analysis reveals that a surface composed of Cu-Cu(I) ensembles can have dual carbon monoxide molecules coupled asymmetrically, which potentially enhances the catalyst’s CO2RR product selectivity toward C2 products. Our results offer understandings of the fundamental chemical states and insights to the establishment of selective CO2RR.

O f versatility, energy efficiency, and cost effectiveness, electrochemical CO 2 reduction (eCO 2 RR) has been deemed a potential strategy to mitigate excessive anthropogenic CO 2 emissions [1][2][3][4][5][6][7][8] . From a variety of catalysts for eCO 2 RR, those copper-based have drawn much attention 1,6 , since Cu is the one which can electrochemically reduce CO 2 molecules to economically favorable and energy-dense C 2 products, such as C 2 H 4 and C 2 H 5 OH 5,9 . However, the typically low product selectivity hinders the catalysts from industry applications. There, accordingly, have been numerous studies on improving the selectivity by means of nanostructuring 10,11 , bimetallic systems [12][13][14] , and oxidative modifications [15][16][17][18] . Among them oxidative modifications have received considerable attention, since they could significantly improve the selectivity of Cu-based catalysts toward valuable C 2 products.
Nevertheless, despite showing improvements, these oxidative modifications still cannot meet the decent selectivity for industry applications, 90% as Jouny et al. pointed out 5 . It has been noticed that these studies, whether knowingly or otherwise, suggested that Cu(I) may play a critical role in improving the selectivity toward C 2 products, but the chemical state of Cu in the catalysts were varied drastically during electrochemical reduction by using a chronoamperometry 15,18 . This ineffective control of the chemical compositions and/or chemical states leads to the resulting low selectivity toward C 2 products, and a fair speculation could be that a stable chemical composition of Cu species during electrolysis is indispensable to enhance the related C 2 product selectivity. For instance, Nogami et al. showed that an oxidation process stimulated by switching applied potentials in corresponding time intervals could bring an improvement in the product selectivity in their eCO 2 RR system 19 . Recent studies also agreed with the integrated oxidation process for improving CO 2 RR product selectivity [20][21][22][23] . For example, the employment of pulses voltage has been revealed to enhance the selectivity toward the C 2 products on polycrystalline Cu 21,23,24 . Recently, we have justified the existence of another competing reaction (i.e., the spontaneous oxidation of Cu(0) in aqueous electrolyte) that significantly governs the chemical state of active centers of Cu, establishing a strong correlation between the chemical state under reaction conditions and the CO 2 RR product profile 25 . A similar effect has been also revealed in the case of Fe single atom catalyst, in which the reduction activity toward CO is evidently dominated by the chemical state of metal centers 26 .
However, some studies brought a different perspective that the active catalyst is metallic copper because no significant concentration of residual oxide was detected in the oxidederived samples; ambient pressure X-ray photoelectron spectroscopy (XPS) and electron energy loss spectroscopy (EELS) proved the absence of residual copper oxide in the reduced electrocatalyst 27,28 . Regarding this controversial issues, we have to point out a paramount subject concerning a significant mismatch in time-scale of acquiring spectrum. For conventional X-ray absorption spectroscopy (XAS), the spectroscopic acquiring time is approximately several 10 min (normally 20-30 min). In the cases of XAS and EELS, both of them would require a time-scale in minutes at least for spectrum acquisition. Notably, the chemical state of metal centers could reach a steady state in a few minutes under the eCO 2 RR; as revealed by in situ Raman, the CuO x precursor would achieve a steady-state condition due to the co-occurrence of electrochemical reduction and of spontaneous oxidation caused by the trace oxidant in the electrolyte 25 . Consequently, a time-resolved identification of the characteristics in chemical state of metal centers should be highly desirable, especially for a time-scale of few seconds for revealing the real dominated factor in CO 2 reduction electrocatalysis.
Thus far, there have been various types of time-resolved XAS developed. For example, the fixed energy X-ray absorption voltammetry (FEXRAV) has demonstrated its capability to provide valency information of a material under time-dependent electrochemical working conditions [29][30][31] . Another technique in the time-resolved XAS family is energy-dispersive XAS (EDXAS). The parallel acquisition of all data points in a XAS spectrum using EDXAS allows its time resolution down to the level of ms and, if possible, to ps 32 . While the FEXRAV and the EDXAS are of time resolution, certain drawbacks make both of them possibly not suitable for the present study. In the case of the FEXRAV, it fails to offer the information regarding chemical surroundings, which has been reported to influence the selectivity of a catalyst 33,34 and normally can be derived using EXAFS spectroscopy. These technical limitations can be well removed by using quick-XAS. The basic setup for quick-XAS is shown in Fig. 1a. Via a quick-scanning monochromator, an acquisition time of a single XAS spectrum, including X-ray absorption near edge structure (XANES) and extended X-ray absorption structure (EXAFS), can be cut down to a few seconds, reaching secondsresolution. Moreover, its sequential detection mode also renders the operando detection achievable. Overall, the advantage of data collection and interpretation makes quick-XAS (TR-XAS hereafter) the time-resolved technique of choice in present study.
Herein, utilizing TR-XAS with a small X-ray incident angle, we develop an operando methodology to realize the seconds-resolved near-surface investigations of materials under CO 2 RR working conditions (Fig. 1a, b). This allows us to in situ characterize the chemical nature of catalytic surfaces. An oxide-derived Cu electrocatalyst with great potentials to produce hydrocarbons is employed as a model to provide scientific insights to the roles of metal states in CO 2 RR. We can conclude that chemical state of copper ions can reach a steady-state in~5 min toward the metallic Cu(0) under a conventional chronoamperometry. Most interestingly, by utilizing a potential switching approach, the Cu (I)/Cu(0) redox could achieve a steady state of half-and-half during a cathodic CO 2 RR electrolysis, producing an asymmetric C 2 product -C 2 H 5 OHwith fairly high selectivity in a wide potential range. Furthermore, a theoretical computation revealed that a surface composed of Cu-Cu(I) ensembles could have dual CO molecules coupled asymmetrically, which potentially enhanced the catalyst's CO 2 RR product selectivity toward C 2 products 35 . Operando TR-XAS gives conclusive evidence that chemical state of copper significantly dominates the catalytic behavior and the product profile of CO 2 RR. Our results can offer understandings of the fundamental chemical state and insights into the establishment of selective eCO 2 RR.

Results
Catalyst synthesis and characterization. A two-stage wet chemical synthesis was employed to realize our model catalyst. Uniform Cu nanocubes were first synthesized with an average size of 38.0 ± 4.2 nm (Fig. 1c, and Supplementary Figs. 1 and 2). To investigate the individual role of Cu(0) and Cu(1+), 1hexadecylamine (HDA) was utilized to mediate the oxidation rate of targeted metallic nanomaterials and to gently oxidize the asprepared Cu nanocubes 36 , which protected the morphologies of the as-prepared nanomaterials from being crushed and further suppressed the effects caused by morphologies of materials 37 . TEM image of the oxidation modified Cu nanocubes (CuO x hereafter) clearly demonstrated that the morphology of CuO x stayed a cube-like after the oxidation modification (Fig. 1d). Regarding the chemical composition of the as-prepared CuO x , selected area electron diffraction (SAED) was characteristic of two domains, metallic Cu and oxide Cu 2 O species, in the CuO x (Fig. 1e). These observations confirm the formation of our desired material made of Cu-Cu(I) ensemble. Electrochemical analyses also agreed with the electron microscopic analysis. Shown in Supplementary Fig. 3a, the LSV indicated at −0.45 V (vs RHE), the system had a strong cathodic response, suggesting the existence of oxide species in CuO x . To identify species of such oxides, as depicted in Supplementary Fig. 3b, the CV scanning in a cathodic direction showed that in the first cycle (green line), a noticeable reduction peak which was preceded by a slight oxidation peak could be corresponding to a reduction from Cu + to Cu, suggesting the oxidized copper atoms in our catalyst were nearly cuprous 38 . In the second cycle of the CV measurement (indicated in red), the original slight oxidation peak grew clearer as shown in the inset of Supplementary Fig. 3b. This growth supported the idea that our catalysts contained Cu + ions, since the following reduction phenomenon also occurred and represented the reduction of Cu + to Cu.
To further confirm the chemical composition, X-ray absorption spectroscopy (XAS) of Cu K-edge for the CuO x was conducted ( Fig. 1f and Supplementary Fig. 4). As depicted in Fig. 1f, the normalized 1st derivative of X-ray absorption near edge structure (XANES) of CuO x and references suggested that no detectable Cu(II) species existed in the as-prepared CuO x 39 . On the other hand, a strong peak overlapping that of Cu 2 O at 8980 eV indicated the presence of Cu(I) species in the CuO x catalyst. Notably, a slight difference between the normalized absorbance at 8981 eV of CuO x and that of Cu 2 O might be attributed to a couple of reasons, such as a dilution effect from metallic Cu species and the catalysts' nanoscale 25 . Wavelet transform was employed to clarify the coordination environment near by the Cu atom and demonstrate the atomic dispersion of Cu in the CuO x sample. As shown in Fig. 1g, a strong wavelet transform signal focused at 7.4 Å −1 derived from Cu-O contribution can be observed, while another stronger signal focused at 8.8 Å −1 can be attributed to a contribution of Cu-Cu. The Fourier-transformed k 3 -weighted spectra of CuO x and corresponding references reveal the characteristic peaks at 2.3 Å, which correspond to the Cu-Cu scattering, implying the existence of metallic Cu (Fig. 1h). Note that a significant Fouriertransformed peak at 1.6 Å can be referred to the scattering path of Cu-O, which implied the presence of the Cu oxide with a similar nature to that of Cu 2 O. The wavelet transform contour plot of CuO x also further clarified the neighboring environment around Cu, which is consistent with all observations above and indicated the fact that the copper species in the as-prepared CuO x were nearly of Cu and Cu(I).
CO2RR performance evaluation. To realize the steady chemical state of the as-prepared CuO x and clarify the real dominated factor in CO 2 reduction electrocatalysis, we adopted the method of switching potential (redox shuttle; R.S. hereafter). Specifically, we applied a constant anodic potential (0.5 V), which was in light of the above discussion regarding the oxidation of CuO x , and various cathodic potentials (from −0.7 to −1.15 V). The time interval of each anodic and cathodic potential cycle was set for 10 s (see Supplementary Fig. 6 for the potential map of R.S.). As stated above, the technique provides the electrocatalytic materials with a cycle of oxidation and reduction, thereby having the potential to stabilize the chemical nature of materials. Figure 2a exhibited the overall CO 2 RR performance of the as-prepared CuO x catalyst. Surprisingly, until −0.9 V, the C 2 H 5 OH molecule was the only identified CO 2 RR product in our system while the rest product was hydrogen gas generating from proton reduction, and the corresponding F.E. (Faradaic Efficiency) reach a maximum value of~12.9 % @ −0.75 V. With increasing the cathodic potential through the redox shuttle, the F.E. of ethanol gently declined. Note that the ethanol was still the only product from CO 2 RR. In contrast to the case of redox shuttle, the CO 2 reduction product through a conventional chronoamperometry (CA hereafter) at the same potential of −0.75 V was carbon monoxide ( Supplementary Fig. 5), which was characteristic of a copper-like behavior with the only product of CO from CO 2 RR at such potential 1,37 . The CuO x catalyst in present study gave a   of Cu species in CuO x and the corresponding electrochemical responses during CO 2 RR at −0.75 V and e the corresponding time-resolved EXAFS spectra without phase-correction of CuO x under CO 2 RR. The spectra were depicted in a 2D contour plot with interatomic distance (Å) on x-axis and with time (sec) on y-axis, while normalized Fourier transform magnitude served as corresponding intensity. Quantification of time-resolved chemical composition information extracted from operando Quick Cu XAS for CuO x under eCO 2 RR using redox shuttle (f) and conventional chronoamperometry (g). Each error bar indicated an estimated standard deviation based on the curve fitting of EXAFS spectra. Both cathodic potentials were −0.75 V and electrolyte was 0.5 M CO 2 -saturated KHCO 3 . copper-like CO 2 RR performance once applying a higher cathodic potentials (higher than −0.95 V) where the dominant CO 2 RR product was CH 4 . We could say that even though the increasing potential would bring a decline in F.E. toward ethanol, the single product from CO 2 RR still could be realized without presence of other common CO 2 RR products, such as C 2 H 4 , CO, and CH 4 . This finding further validates the fact that the redox shuttle can achieve the desired single selectivity of CO 2 RR. Note that we kept anodic potential at 0.5 V (vs RHE), referred to the oxidation potential of as-prepared CuO x (as revealed in Supplementary Fig.  3), and varied the cathodic potentials from −0.7 to −1.15 V (vs RHE). Anodic and cathodic potentials were switched in an interval of 10 s ( Supplementary Fig. 6). Apparently, it seems that a large cathodic potential was able to result in significant increasing the current density of CO 2 RR. Nevertheless, the large cathodic potential was failed to achieve a highly selective CO 2 RR production since that was going to cause a remarkably generation of metallic Cu 0 (vide infra). As indicated by the CV results (Supplementary Fig. 7 and Supplementary Note 1), the oxidation variations of Cu species between applied cathodic potentials were relatively subtle through using R.S. and therefore highly suggested the technique enabled the composition of Cu species in the CuO x to be well preserved. Interestingly, it is well known that the formation of C 2 H 4 and C 2 H 5 OH may share the same path way of CO 2 RR through a hydroxyl containing intermediates [40][41][42] . As a result, more detail analyses in following sections are essential to address why C 2 H 5 OH is the favorable product rather than C 2 H 4 in our system.
Operando XAS characterization. To assess the chemical composition of the CuO x under CO 2 RR, operando XAS was performed, due to its capability to probe the chemical composition of the catalysts under realistic reaction conditions 26,43-45 . As we mentioned above, conventional XAS only offers the information which is referred to a steady condition because of time requirement in spectrum acquirement (typically several tens mins). Thus, it inaccessibly provides a convincing time-resolved information for an electrochemical system where the reactions are typically being steady on a timescale of seconds or few minutes 46 . With the aim of addressing this timescale mismatch, the operando time-resolved X-ray absorption spectroscopy (operando TR-XAS) with a sub-second time-resolution can be reached by equipping a continuously oscillating crystal on a monochromator 47 . Operando time-resolved-XANES spectra for CuO x during CO 2 RR at −0.75 V (vs RHE) using R.S. and those of using CA were displayed in Supplementary Figs. 8 and 9, respectively. A linear model developed by Vitova et al. 48 was utilized to extract the pristine and time-resolved compositional information from acquired XANES spectra. As shown in Supplementary Fig. 10, the result suggested that the composition of the pristine CuO x was of equal Cu(I) and Cu. This not only supported again our previous characterization that Cu(I) and Cu coexisted in the CuO x (Fig. 1c-h), but also proved the feasibility of the linear model to quantitatively analyze the composition of Cu species, as Vitova et al. suggested. In this regard, we used the linear model to extract the chemical nature information of CuO x from the operando time-resolved-XANES spectra. Extracted time-resolved chemical composition information combined with the electrochemical responses was shown in Fig. 2 (2b for R.S. and 2d for CA) and in Supplementary Fig. 11 (the magnified portion of Fig. 2b by which one can track the footprint of XAS acquisition under the R.S. working condition). During the CO 2 RR, R.S. restricted the chemical composition of CuO x in around a composition of half Cu and half Cu(I), while the CA transformed the chemical composition of CuO x into that Cu-dominated. Following the 1hour reaction time, XANES spectrum of the CuO x after the CO 2 RR using R.S. remained nearly unchanged, further indicative of R.S. having the ability to preserve the chemical composition. On the other hand, that of the CuO x after the electrolysis using CA clearly showed an electrochemical reduction toward metallic Cu ( Supplementary Fig. 10).
As shown in Fig. 2c, e, operando time-resolved EXAFS verified that the intensity of bond pairs regarding Cu-O at 1.6 Å and Cu-Cu at 2.3 Å stayed highly stable under the employment of R.S., whereas the intensity under the case of using CA showed a decline at 1.6 Å and an increase at 2.3 Å that could be corresponded to a transformation from Cu 2 O to metallic Cu 0 . To further quantify such chemical composition information from operando EXAFS, we obtained the Cu-Cu path from both Cu 2 O (Cu-Cu (Cu2O) ) and metallic Cu (Cu-Cu (Cu) ) extracted by a standard fitting procedure (detailed results shown in Tables 1-2 and Supplementary Figs. [12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31]. We note here that in order to have quality spectra for EXAFAS fitting, the related XAS acquisition time was extended from 5 to 60 s. While the time resolution could not match that of seconds (Fig. 2b-e), the minutes-resolved EXAFS fitting results were also in line with those seconds-resolved (see below). The quantification was displayed in terms of coordination numbers and, more straightforward, in terms of the percentage of both Cu(0) and  (2) Uncertainties in the last digit are given in parentheses.  Fig. 2f, g, time-resolved fitting results fairly agreed with both analyses of XANES and EXAFS. Using R. S., the CNs of Cu(I) and Cu(0) in the as-prepared CuO x stayed relatively steady during CO 2 RR (Fig. 2f). By contrast, under the employment of CA, the CNs of Cu(I) and Cu(0) in the CuO x gradually reduced and increased, respectively, suggesting the electrochemical reduction of the CuO x (Fig. 2g). Consequently, the results of time-resolved operando XAS proved that under CO 2 RR, the chemical composition of CuO x could remain in a steady state of half-and-half Cu and Cu(I) through using R.S.
By contrast, under the employment of CA, the chemical composition of CuO x continued a downward trend toward Cu(0). A similar result as illustrated in Fig. 3 was also revealed by the conventional XAS, showing a significant reduction of Cu + to generate metallic Cu 0 . XANES data elucidated as a cathodic potential was applied from −0.05 to −1.25 V (Fig. 3a, b). The XANES features as illustrated in Fig. 3b were denoted as three distinct peak-like structures, while no remarkable pre-edge feature that caused by a dipole transition from 1s to the mixing states of 4p and 3d could be visible and suggested a centrosymmetry nature of copper 37,49 . An additional small feature of the excitation in the 3d 10 L hole state at a lower energy region (as indicated by green arrow) gradually intensifies with increasing cathodic voltage, this indicates the formation of metallic Cu (Fig. 3b). Furthermore, the shifts in the edge position could conclude the chemical states in CuO x were reduced into copperlike and the content of metallic Cu(0) species in the CuO x evidently increased. Consistent with XANES spectra, the analysis of extended X-ray absorption fine structure (EXAFS) revealed that as an applied cathodic potential increased, the metallic Cu-Cu bond pair at 2.3 Å, matching a standard metallic Cu, intensified significantly (Fig. 3c). Worth noting is that like cases in literature 39 , the as-prepared CuO x exhibited a resistance to electrochemical reduction, as the intensity of a Cu-O bond pair at 1.6 Å, referring to a Cu 2 O standard, did not entirely disappear even under a cathodic environment. Despite having this resistance to the electrochemical reduction, however, the asprepared CuO x could not remain its initial chemical nature under CO 2 RR when the CA was adopted. This observation can refer to the fact that the catalytic sites and the chemical states can fail to reach a steady situation under the conventional chronoamperometry. These results also can explicate the finding from numerous reports that the products profile of CO 2 RR are commonly characteristic of a mixture nature.
To extend the investigation further, we also performed conventional operando XAS analyses for the CuO x under CO 2 RR using R.S. in various potentials (Fig. 3d-f). Only slight variations on XAS results across the various applied potentials, the chemical composition of CuO x could remain in a steady condition of the composition of Cu and Cu(I) through using R.S. as illustrated in Fig. 3d, e. Note that, as compared with those of Fig. 3b, no additional feature of the excitation in the 3d 10 L hole state at a lower energy region can be found with increasing cathodic voltage, this further indicates the chemical nature of CuO x can reach steadystate under CO 2 RR using R.S. (Fig. 3e). In the case of operando EXAFS spectra, the final condition from various applied potentials were similar except for the situations in high cathodic potentials (<−0.95 V vs RHE). This finding corroborated the observation of potential-dependent product profiles in Fig. 2a, in which the CO 2 RR products have become a metallic copper-like profile with a mixed feature. Therefore, these results became solid evidence to conclude the existence of the steady composition of Cu species under CO 2 RR using R.S. Aside from spectroscopic analysis, the post electron microscopy analysis also supported the validation of R.S. As shown in Supplementary Fig. 32, the TEM image of CuO x after a 6-h CO 2 RR electrolysis at −0.75 V using R. S. demonstrated that the morphology of the CuO x was similar to those before the electrolysis (Fig. 1d). This consistence fairly helped rule out the possibility that the observed CO 2 RR performance in the present study was brought by morphology change 37 . Meanwhile, the HR-TEM analysis, as well as related SAED, of the post-catalysts in Supplementary Fig. 33 also confirmed the chemical nature of the CuO x was the same as the as-prepared catalyst (Fig. 1e). Furthermore, in terms of electrochemical stability, we also evaluated the electrochemical active surface area (ECSA) of CuO x before and after the CO 2 RR electrolysis. The double layer capacitance (C dl ) was used for the assessment, since it is directly proportional to the ECSA of materials 50,51 . Using a cyclic voltammetry (CV) method, the C dl values of CuO x before and after the electrolysis were 77.27 and 76.60 μF cm −2 , respectively (Supplementary Figs. 34 and 35). The result of nearly the same C dl values reflected the fairly stable ECSA of CuO x after the electrolysis and highly suggested the stability of using R.S. Altogether, results of the operando XAS analyses, the electron microscopy analyses, and electrochemical studies convincingly proved the validation of R.S. as a steady-state approach.
Through operando TR-XAS analyses, we correlated the chemical nature of CuO x with its unique selectivity toward asymmetric C 2 H 5 OH molecules, which has evidently concluded a paramount role played by the composition of the Cu-Cu(I) ensemble. Moreover, the preceding analyses also revealed that different CO 2 RR products may result from catalytic surfaces with different compositions of Cu and Cu(I); a Cu-dominated surface synthesized CO, while the surface with halves of Cu and Cu(I) gave rise to ethanol (Supplementary Fig. 10). This rather suggests a delicate Cu(I)/Cu(0) interface is the likely active center for CO 2 to asymmetric products and motivates future work regarding such interfacial engineering. Analogous interfacial engineering strategies have been applied to enhance product selectivity in electrochemical reaction systems 52,53 . The study by Feng et al., for example, has demonstrated a linear correlation between the boundary densities within their as-prepared Cu nanoparticles and the corresponding catalytic behavior of reducing CO to multicarbon products, i.e., ethanol and acetate 54 . Furthermore, using scanning electrochemical cell microscopy (SECCM), Mariano and coworkers provided a submicrometer-resolved study on gold films 55 . The study with such spatial resolution clearly demonstrated that the electrocatalytic activity of a gold film for CO 2 RR depends on the grain boundary types and densities on the film.
Theoretical understanding of the CO2RR selectivity. As follows, we make an attempt to elucidate the unique selectivity in our reaction system thorough theoretical computation. By using density functional theory (DFT) calculations, we further evaluated the stability of oxygen-vacancy generation/removal with various compositions of Cu and Cu(I) at the top layer of Cu 2 O. All details regarding the structural simulation are addressed in Methods, while the model coordinates are listed in Supplementary Table 1. Such a geometric stability during the oxygen removal process can potentially facilitate the regeneration of the acting-catalytic environment by the redox-shuttle approach. These structural simulation results demonstrated that the equal numbers of Cu and Cu(I) on the top layer could provide an ideal environment to host the C-C bond formation intermediate (OCCO) (Supplementary Figs. 36 and 37). All studies in following parts, therefore, will focus on the surface composed of the equal numbers of Cu and Cu(I) for subsequent modeling (defined it as a Cu-Cu(I) ensemble). Afterwards, we characterized C-C coupling behavior on the Cu-Cu(I) ensemble, and the configurational results-calculated absorption energy and geometrically optimized structures of absorbed CO molecules at the gas-solid interfacewere shown in Supplementary Fig. 37. The ensemble configured dimerized CO molecules to an asymmetric "end-side" configuration; a CO molecule was absorbed in the side-on direction and coupled with an end-on CO molecule, which was consistent with previous work 56 .
On the other hand, the explicit water solvation effect has been demonstrated to well achieve a realistic modeling in free energy calculations. As depicted in Fig. 4a, the free energy barriers (ΔG ‡ ) for the C-C bond formation under the explicit water solvation indicated the overall OCCO formation on the ensemble was thermodynamically accessible at 0.73 eV, as the corresponding reaction free energy ΔG was −1.22 eV (for computation details, refer to Supporting Information). Furthermore, Bader charge analysis also suggests that the mixed-valence ensemble constructively stabilize the asymmetrical OCCO intermediate though a favorable electrostatic interaction, in which the highly polar OCCO (+0.50 and −1.68 for the positive and negative fragments of (O (1) C (1) ) δ+ (C (2) O (2) ) δ− , respectively) is stabilized by the presence of Cu(0) and Cu(I) on the surface, respectively (Fig. 4b).
All computational results demonstrated the potential introduced by the mixed-valence ensemble could produce an asymmetric OCCO intermediate in CO 2 RR, which might considerably facilitate the formation of C 2 products from CO 2 RR.
The plausible mechanistic steps of ethanol formation are computationally explored with considering the various hydrogenated intermediates on the Cu-Cu(I) ensemble, being simulated by the mixed-valence Cu 2 O surface model, as highlighted as 1H to 6H in Fig. 4c. The calculated electrochemical step started from the hydrogenation step of CCO (1H), being generated from the dehydration step of the experimental observed OCCOH intermediate 54 . The relative free energies in respect to the minimum geometries (all marked as 0 eV) at each hydrogenation stage were labeled in Fig. 4c. It should be noted that all of the hydroxyl containing intermediates (CCOH, HCCOH_OH, HCCHOH_OH, H 2 CCOH_OH, and H 2 CCHOH_OH) were found to be less stable by 0.82, 0.43, 0.65, 1.08, and 0.79 eV than the corresponding minima, respectively, where the presence of hydroxyl groups within the hydrocarbon species could lead to the pathway of ethylene formation via a proton-coupled-electron dehydration process. Although HCCOH (crossed out in grayscale) was proposed as a necessary intermediate for C 2 H 4 formation 55 , it became inaccessible due to its unfavorable precursor -HCCHO.
By contrast, OH species at the boundary of Cu-Cu(I) ensemble could play a critical role in facilitating the ethanol formation. Such a mixed-valance region favored the formation of surface hydroxyl given the electron-rich Cu region enhances the basicity of the boundary oxygen sites of Cu(I) region. Theoretically, OH species at the Cu-Cu(I) boundary was commonly identified along the hydrogenation pathway of CO 2 reduction, and that resulted in the attractive electrostatic interactions between surface-OH dipole and the dipole of carbonyl group of hydrocarbon intermediates. This carbonyl stabilization effect could protect the oxygen end from being protonated as well as occupy the valence electron of carbon to avoid the CC double bond formation. Consequently, managing the boundary oxygen sites during the electrochemical CO 2 reduction could be the determined factor of ethanol generation. The unique selectivity toward C 2 H 5 OH could be attributed to the fact that the CuO x could maintain well balance of Cu and Cu(I) during eCO 2 RR using R.S. and offer an asymmetrically OCCO adsorbing site with the subsequent carbonyl group stabilization effect by the OH groups at the boundary of Cu-Cu(I) ensemble, eventually synthesizing an asymmetric oxygenated product (Fig. 4d).

Discussion
In summary, using redox shuttle manner, the results of product characterization demonstrated that in a wide potential range, the as-prepared CuO x exclusively produced C 2 H 5 OH over other CO 2 RR products. Using the operando time-resolved XAS, we tracked the evolution of its chemical nature under CO 2 RR. Quantitative XANES and EXAFS analyses evidently revealed that the steady chemical composition of the material could be achieved via the usage of the redox shuttle approach, while the employment of conventional chronoamperometry drastically altered the chemical nature of the material. Collectively, a strong correlation between the stabilized ensembles and the corresponding unique selectivity of CO 2 RR products has been established. More importantly, the carbonyl stabilization mechanism by the boundary OH species was proposed to prevent the protonation of the terminal oxygen sites of hydrocarbons, and such a hedge device produced an oxygenated CO 2 RR product (ethanol in our case) with the unique selectivity. Our studies provided the first empirical real time information, correlating the chemical nature of Cu-based catalysts with their corresponding selectivity toward CO 2 RR products.
Synthesis for copper nanocubes. The nanocubes were prepared through a polyamine process under an inert atmosphere. CuBr (1 mmol) and TOPO (3 mmol) were firstly dissolved into oleylamine (15 mL) in a three-neck round-bottom flask. After well mixed, the solution was flushed with N 2 for 30 mins to remove the oxygen in the apparatus. Under the inter atmosphere, the solution was heated at 80°C for 5 mins and then up to 230°C for another 3 h, turning into a reddishbrown clouded solution. Afterwards, the solution was air-cooled to room temperature for purification. The solution was washed with n-hexane (40 mL) and centrifuged at 8000 r.p.m. for 10 mins three times. The purified nanocubes were stored in n-hexane for the later use.
Synthesis for mixed-valence copper nanomaterials. In all, 20 mg HDA were added into 9 mL 1, 5-pentanediol. Then, the solution was preheated at 170°C for 10 mins. Afterwards, as-prepared copper nanocube solution (1 mL) was transferred into the solution via one-shot injection. The solution was kept at the temperature for a half hour. Eventually, the solution turned yellow, suggesting the formation of mixed-valence copper nanomaterials. The materials were washed with an ethanolwater mixture in ratio 7:3 and centrifuged at 10,000 r.p.m. for 10 min three times. The purified materials were kept in absolute alcohol for the later use.
Catalyst characterization. The morphology of as synthesized materials was characterized using the transmission electron microscopy (TEM, Hitachi H-760 operated at 80 kV) and the scanning electron microscopy (SEM, JSM-6700F). The materials' crystal structure was studied by using selected area electron diffraction (SAED) in a high-resolution TEM (HRTEM, JEOL JEM-2100F operated at 200 kV). The elemental analysis of the materials was conducted using energy-dispersive X-ray spectroscopy (EDS) inside the HRTEM.
Electrochemical measurements. All electrochemical measurements were carried out using Biologic-VSP instrument and in a three-electrode setup. The working electrode was a glassy carbon electrode (0.782 cm 2 ); the counter electrode, a Pt wire; the reference electrode, an Ag/AgCl electrode (in 3 M KCl). The test cell was a home-made electrochemical cell. The cell consisted of two parts, cathode and anode compartments. The working electrode and reference electrode were placed into the cathode compartment, and the counter was in the anode compartment. Both compartments were separated by an anion exchange membrane (Fumasep FAA-3-PK-130, FUMATECH) to prevent potential contamination resulting from the dissolution of Pt counter. In electrochemical experiments, 10 μL of the solution of as-prepared materials (2.5 M) were loaded on the glassy carbon. Each compartment was filled with 30 mL pre-bubbled CO 2 -saturated 0.5 M KHCO 3 , whose pH changed from 8.6 to 7.2, and the CO 2 flow was kept at 100 sccm for 10 more mins in the cathode compartment to build an air-free atmosphere for the later detection of as-prepared materials' electrocatalytic activity for CO 2 reduction. Several electrochemical techniques were employed in the study, including cyclic voltammograms (CV), chronoamperometry (CA), and electrochemical redox shuttle (R.S.). CVs were conducted in the potential interval between −0.35 V and −1.05 V at a scan rate 100 mV/s. CA was performed at stated potentials. R.S. was adopted at stated potentials and at stated time intervals (one may refer to Supplementary Fig. 6 for the potential map). Potentials listed in the article were iR-corrected and referenced to the RHE (short for reversible hydrogen electrode) scale by Eq. (3). The pH of 7.2 was used for the conversion, since the value before and after CO 2 RR treatment stayed constant at 7.2. The iR drop was determined using impedance measurement technique (ZIR) available in the potentiostat.
Double layer capacitance (C dl ) Measurement. A cyclic voltammetry (CV) method was used to study the C dl of as-prepared catalysts. CVs were conducted under N 2saturated 0.5 M KHCO 3 at various scan rates (10-300 mV/s). A potential window from 0.1 to 0.2 V was adopted due to the absence of redox peaks in this region ( Supplementary Fig. 3). Later, the anodic and cathodic current density difference of each CV at the center of the potential window was plotted as a function of scan rate, whose slope will automatically give the C dl .
CO2RR product analysis. Analytical methods for product characterization here have been outlined previously 37 . Gaseous products were evaluated using gas chromatography (GC, Agilent 7890B). The GC was equipped with a thermal conductivity detector (TCD) and a flame ionization detector (FID) in series. The TCD was for H 2 and CO detection, and the FID was for CH 4 and C 2 H 4 detection. As for those products in liquid phase, gas chromatography-mass spectrometry (GC-MS, 5977A) was adopted to detect ethanol and propanol, and nuclear magnetic resonance (NMR, Bruker Advance III 500 MHz) was used for the detection of formic acid. The faradaic efficiency (F.E.) of each product was calculated by Eq. (4): where n is the number of electrons transferred, F stands for Faraday's constant (96485 C mol −1 ), and C represents the total amount of charge passed through a working electrode.
Operando X-ray absorption spectroscopy. Operando X-ray absorption experiments for Cu K-edge measurement were conducted at beamline BL 17C of National Synchrotron Radiation Research Center (NSRRC). A three-electrode arrangement (the same setup as those described in electrochemical measurements) was used for the operando measurements; the measurements were performed in customized Teflon reactors with a Kapton tape window, allowing X-ray transmission.
Operando quick-scanning X-ray absorption spectroscopy. The measurements for Cu K-edge absorption were taken at TPS 44A, Hsinchu, Taiwan. The corresponding data were recorded in total-fluorescence-yield mode. The setup of the experiments was stated in the article. To have a quality spectrum, we applied the Quick-XAS mode in a time-resolution of 5 s for XANES and EXAFS analyses, while for EXAFS fitting the time-resolution was chosen to be 60 s. The X-ray absorption experimental data were collected in total-fluorescence-yield mode, in which the metallic Cu foil was taken as reference to calibrate the energy scale. Using Demeter, all X-ray absorption spectra were processed by subtracting the baseline of pre-edge and normalizing that of post-edge; such data processing was for both X-ray absorption near edge spectra (XANES) and extended X-ray absorption fine structure (EXAFS) analyses. EXAFS analysis was carried out using Fourier transform on k 3 -weighted EXAFS oscillations to assess the contribution of each bond pair to Fourier transform peak. The curve fitting of EXAFS spectra was conducted using the software, REX2000, with FEFF8 program.
Theoretical analysis. All calculations were performed with the DFT plane-wave method utilizing the Vienna ab-initio simulation package (VASP) [57][58][59][60] . We used the projector-augmented-wave method (PAW) 61,62 in conjunction with generalized gradient-approximation (GGA) calculation, and Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional was applied. For the design of slab models, we used 8, 6, and 12 atomic layers for Cu 2 O(111), Cu 2 O(110), and Cu 2 O(100), respectively, as shown in Supplementary Fig. 36. For the DFT + U calculations, we chose the cutoff energy at 500 eV with 8 × 8 × 8 and 4 × 4 × 1 k-point mesh for bulk and surfaces models, respectively. On-site Columbic correction (U = 5.2 eV) was used to prevent delocalization of Cu d-orbitals, that has been demonstrated for modeling the previous Cu 2 O system 63,25 . The bottom two atomic layers were kept frozen with the remaining layers fully relaxed for these Cu 2 O slabs. In all cases, the slab calculations included a vacuum region of thickness greater than 15 Å, large enough to ensure no interaction between the two slabs. The adsorption energies were calculated based on the following equation, were the calculated electronic energies of adsorbed species on the above surfaces, clean surfaces, and free molecules, respectively. The climbing-image-nudged-elastic-band (CI-NEB) method 25,64-66 was applied to locate transition structures; the profile of the potential-energy surface (PES) was constructed accordingly. Frequency calculations were applied to verify the adsorbed intermediates and the transition states (with only one imaginary frequency). Based on the stepwise oxygen-vacancy formation energy, the Cu 2 O(110) surfaces was found to contain the most subtle surface reconstruction among the three common Cu 2 O cases (i.e., (111), (110), and (100)) as illustrated in Supplementary Fig. 36.
Solvent model. To provide an appropriate solvent effect of CO dimerization reaction, we constructed an explicit water layers (64 water molecules) to represent the hydrogen bond network of the aqueous environment. Water molecules were described by SPC-3f(α) models 67 using LAMMPS molecular dynamics simulator 25 . The liquid-solid interfaces were brought to equilibrium with 2 ns pre-equilibrium simulations 68 at 350 K while the solid geometry (with adsorbate) remained frozen during the molecular dynamics simulations, being represented by atomic charges (from Bader charges of DFT). The adsorbate and solid structures were also taken from our DFT calculations. After 2 ns of pre-equilibrium, we sampled 10 solvation structures (0.2 ns apart per sample) for the initial state 2CO (IS), C-C bond coupling transition state (TS), and OCCO final state (FS), followed by solvent structure optimization at DFT-level. Finally, we took the average energy of these 10 structures to represent E(IS), E(TS), and E(FS) for CO-CO coupling process, respectively.
Thermodynamic correction. The relative Gibbs energy in solution (ΔG) for the free energy difference between any adsorbate pair on the surfaces was defined as ΔG = ΔH -TΔS. ΔH was estimated as Δ(E + ZPE + E vib ) with E representing the average electronic energy of DFT under explicit solvation treatment, ZPE representing zero-point energy, and E vib representing the internal energy correction of vibrational models (the relative translational and rotational contribution for any adsorbate pair was assumed canceled out) 69,70,25 . ΔS was predicted by the vibrational partition function using harmonic vibrations. For the CO-CO coupling process, the vibrational modes of two surface Cu atoms were taken into account. For example, ΔG z ¼ ΔGðTSÞ À ΔGð2COÞ ¼ ½ΔHðTSÞ À TΔSðTSÞ À ½ΔHð2COÞÀ TΔSð2COÞ was estimated by each 10 sampled configurations. For the ethanol formation pathway, we did not take into account the explicit solvation effect due to its trivial difference.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Received: 10 February 2020; Accepted: 19 June 2020; Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/.