Towards understanding of CO2 electroreduction to C2+ products on copper‐based catalysts

The electrochemical CO2 reduction reaction (CO2RR) has been regarded as a promising technique for converting CO2 into high‐value fuels and chemicals. Powered by renewable electricity, the CO2RR provides a viable strategy to mitigate the CO2 concentration in the atmosphere and close the anthropogenic carbon cycle. Recent studies exhibit that copper‐based catalysts are capable of reducing CO2 to C2+ products, such as ethylene and ethanol, which are of higher value compared with C1 products. The reaction process toward C2+ products involves the formation of key intermediate *CO, the C–C bonding, and the post‐C–C bonding to final products. This perspective is focusing on the mechanism leading to C2+ products, examining the evidence from in situ/operando spectroscopy and density functional theory calculations. The effects of Cu facet and electrolyte on catalytic performance are reviewed. An in depth discussion of mechanistic aspects of Cu catalyst is presented, shedding light on the intrinsic features of catalyst and electrode‐electrolyte interface, therefore moving towards an understanding of CO2RR at the atomic level.

sources such as solar and wind energy. 4,5 On the other side, the astonishing progress of renewable energy also requires efficient techniques for energy storage. The energy stored in the form of chemical bonds is transportable, easy to be merged with the current industry, and safe. 6,7 The C 1 products of CO 2 RR are CO and HCOOH, which are relatively easy to be generated in CO 2 RR, while the C 2+ products, such as oxygenates and hydrocarbons with high energy density, are more ideal for the practical mass application of CO 2 RR. [8][9][10] The C 2+ products possess merits such as higher volumetric energy densities and the potential to build complex long-chain hydrocarbon chemicals. However, the production of C 2+ oxygenates and hydrocarbons (C 2 H 4 , C 2 H 5 OH, CH 3 COOH, etc.) involves C-C bonding with a high energy barrier to overcome. 11 Compared with the C 1 products involving only oxygenation and hydrogenation, the inert C-C bonding is much more difficult to achieve. [12][13][14] Therefore, C-C coupling requires high overpotential on the cathode catalyst. In the literature, the reported Faradaic efficiency (FE) for CO and HCOOH could reach ∼100%, but the highest FE for C 2 H 4 is only around 60%-70%. 15 The economic analysis indicates that the high reaction rate is crucial for the practical application of CO 2 RR, which requires a high current density of C 2+ products with satisfying stability. Moreover, the extra costs could be incurred from the separation of different products, liquid electrolytes, and unreacted CO 2 . 16,17 Currently, the most commonly used catalysts for C 2+ products in CO 2 RR are copper-based materials. [18][19][20] The performance of C-C coupling is especially sensitive to the interface of copper catalysts and electrolytes. 21,22 Therefore, the improvement of reaction rate and energy efficiency for C 2+ products depends heavily on a deeper understanding of the C 2+ production mechanism.
In the literature, the mechanism of C 2+ product formation is widely investigated with density functional theoretical (DFT) calculations and in situ/operando spectroscopy characterizations. In situ/operando spectroscopy is utilized for the inspection of the presence and binding configurations of related intermediates, 23,24 while DFT calculations are widely applied in investigating the binding energy of key intermediates, depicting the free energy profile, predicting, and explaining the activity and selectivity. 6,25 Moreover, DFT calculations have evolved to understand the complexity of the operating conditions, such as solvent molecules and surface charge distributions. [26][27][28][29] In this perspective, as shown in Figure 1, we focus on the studies with DFT calculations and in situ/operando spectroscopy characterizations in the C-C bonding mechanism, facet, and electrolyte effect, aiming at providing a brief account of C 2+ products formation on copper catalysts.
2 | REACTION MECHANISM TOWARDS C 2 + PRODUCTS 2.1 | The production of *CO On a Cu-based catalyst, the CO 2 molecule is first reduced to *CO intermediate, and then *CO is reduced to *CHO, *COH, *COCO, *OCCHO, or *OCCOH, of which the binding strength determines the final products. 30,31 The suitable Cu-based catalysts are capable of lowering the energy barrier by stabilizing the intermediate in the process of multistep proton-electron transfer. 32 *CO is a key intermediate in C-C coupling reaction because *CO dimerization is generally believed to be the rate-determining step in CO 2 electroreduction to C 2+ products. Apparently, a stable and high coverage of *CO binding on catalyst surface is crucial for C 2+ products, that *CO is a prerequisite for the formation of *COCO, *OCCHO, or *OCCOH. The surface morphology of the catalyst could be altered to enhance the coverage and binding strength of *CO, as well as electrolyte effects to improve the proton-electron transfer. 13,33 The reaction pathway of C-C coupling starts with *CO intermediate which is generated from the CO 2 RR. It has F I G U R E 1 The outline of the review. There are three parts: The mechanism of C 2+ production; structure-activity relationships; and the design of catalysts assisted by theoretical calculations. We focus on the evidence and insights obtained from DFT calculations and in situ/operando spectroscopy. CO 2 RR, CO 2 reduction reaction; DFT, density functional theory. been recognized that there are two possible C 1 products, that is, CO and HCOOH from two-electron transfer of CO 2 reduction. 2,34 Concomitantly, hydrogen evolution reaction is competing with the CO 2 RR. 35 The initial activation of the CO 2 molecule could occur in two ways: proton-electroncoupled transfer (Equation 1) or electron-transfer-mediated CO 2 binding step (Equation 2):  * + CO + e + H + *COOH, On copper catalysts, the activation of CO 2 is more likely to occur via Equation (1), while on the molecular catalyst, CO 2 − anionic adduct is adsorbed on the metal center. It should be noted that for some posttransition metals such as Pb and Sn, *OCOH formation is more favorable than *COOH, which indicates that Pb and Sn are more selective for formate. 36 In the next step, *COOH intermediate is further reduced to *CO (Equation 3). The binding strength of *CO determines the subsequent pathway: *CO either is desorbed to form CO product or undergoes hydrogenation to *CHO or dimerization to *COCO, to form C 2 or C 3 products. Moreover, one of the crucial features of highperformance copper catalysts is the high coverage of *CO. 37,38 High concentration of CO facilitates C-C coupling reaction, thus enhancing C 2+ production. An intensity ratio of Cu-CO stretching mode was reported to the restricted rotation of adsorbed CO in operando Raman experiment, which not only increased with the higher CO coverage but also followed the trend of C 2+ production. 39 The adsorption configurations of *CO is also a key point in C-C bond formation, which will be discussed in the part on facet effect.

| The formation of the C-C bond
The most important step for C 2 products is the C-C bond formation. As shown in Figure 2, there are several possible pathways: the *CO dimer could form directly, or *CO reacts simultaneously with H + and e − , forming *HOCCO intermediate. Another competing pathway is *CO protonation to *CHO, which is selective to methane and glyoxal products. It is reported that on Cu(111) surface, CH 4 is more favorable at high overpotential, while on Cu(100) surface, C 2 products dominate over methane at low overpotential. 14 The different selectivity of Cu(111) and Cu (100) is also related to the pH effects: Cu(111) pathway is pH dependent, while Cu(100) pathway is not, which indicates that CO dimerization is more dominant on Cu (100). However, at a very negative overpotential, CO dimerization is reportedly difficult. 40 Instead, *CO is prone to (1) first reduce to *COH, and then form *HOCCO intermediate or (2) directly form *HOCCO with *CO and proton-electron-coupled transfer. The proton-coupled electron transfer is the ratelimiting step in this pathway, which indicates that it is pH sensitive.
The C-C bonding mechanism is one of the hottest debate fields in CO 2 RR research. Several alternative pathways were proposed to understand this demanding step. Earlier studies by Hori et al. 41 proposed that *CH 2 intermediate is generated from protonation of *COH.
The pathways for C 2+ products in CO 2 reduction reaction. LIU ET AL.
For C-C bonding formation, two *CH 2 are dimerized or CO is inserted into *CH 2 . 41 However, this pathway is challenged by the fact that C 2 H 6 is rarely observed on the copper catalyst. It is found that the charged water layer could stabilize the CO dimer, and without charged water layer, the CO dimer formation is energetically prohibited. 42 Another possible C-C coupling pathway is proposed that negatively charged CO-CO − species is formed during the dimerization process, and then protonated to form CO-COH. 43 On the other hand, the C 3 products are reported to be generated from the insertion of CO to *C 2 H 4 , but details are still obscure. 7,44 Operando Fourier transform infrared spectroscopy (FTIR) has been used to provide useful information for C-C bond formation. FTIR reveals that two vibrational bands detected at 1191 and 1584 cm −1 were ascribed to C-O-H and C-O stretching vibrations of a hydrogenated dimer intermediate *OCCOH. 45 An emerged peak at 2053 cm −1 was observed in operando Raman spectra on Cu 2 O catalyst, corresponding to confined *CO intermediate ( Figure 3A). 46 Cu 2−x S catalyst showed higher selectivity of C 2+ products during CO 2 RR, and C-H stretching mode of aldehydes (*CHO) was observed instead of C-O stretching mode of *CO with the F I G U R E 3 (A) Raman spectra of the multihollow Cu 2 O catalyst obtained with and without an applied current density. Reproduced from Yang et al. 46 Copyright © 2020, American Chemical Society. (B) Operando Raman spectra were obtained during the CO 2 RR using Cu 2−x S catalyst at different applied potentials. Reproduced from He et al. 47 Copyright © 2021, Tsinghua University Press and Springer-Verlag GmbH Germany, part of Springer Nature. (C) Operando Raman spectroscopy over Cu 2 O catalyst during CO 2 RR, P1 refers to the restricted rotation of adsorbed CO and P2 refers to the Cu-CO stretching. Reproduced from Zhan et al. 39 Copyright © 2021 The Authors. Published by American Chemical Society. (D) The -CH X stretching region in the in situ Raman spectra in CO 2 -saturated 0.2 M NaHCO 3 for ID-Cu and OD-Cu catalysts. Reproduced from Vasileff et al. 48 Copyright © 2020 Wiley-VCH Verlag GmbH & Co. KGaA. CO 2 RR, CO 2 reduction reaction. operando Raman spectroscopy ( Figure 3B). *CHO presumably coupled with *CO to form *COCHO, following the formation of C 2+ products. 47 The in situ and operando spectroscopy studies overwhelmingly detect *CO and other protonated C-C intermediates such as *OCCOH, but *OCCO, if any, is much more difficult to be directly observed.

| The postcoupling reactions
For the post-C-C-coupling step, it is reported that protonation of *HOCCO intermediate leads to *CCO or *HOCCOH, which is further reduced to CH 3 COOH and C 2 H 4 /CH 3 CH 2 OH, respectively. 49 It could be seen that *HOCCO is the key intermediate of which protonation determines the selectivity towards ethylene/ethanol versus acetate. However, it is noted that the *CCO could also be reduced to ethylene via *HOCCH 2 intermediate. 50 On the other hand, the bonding between *CHO and *CO generates *OCCHO, which is the precursor of glyoxal product. It should be pointed out that an alternative pathway proposed by the Sargent group argues that *HOCCH is the key intermediate instead of *HOCCO. *HOCCH is reduced to *CCH or *CHCHOH, which are the precursor of ethylene or alcohol, respectively. 38 Meanwhile, with millisecond-resolved differential electrochemical mass spectrometry, the Strasser group detected and proposed that *CH 3 could serve as an unexpected intermediate shared for CH 4 , CH 3 CH 2 OH, CH 3 CHO, and CH 3 COOH formation on (110) step sites adjacent to (100) terraces at higher overpotential, which shows that a much closer link between CH 4 and ethanol exists than previous thought. 51 3 | STRUCTURE-ACTIVITY RELATIONSHIP

| The facet effects
The surface facet of copper catalyst exerts a significant influence on selectivity and activity. As mentioned above, Cu(100), Cu(111), and Cu(110) are all reported to be effective for CO 2 RR, while Cu(100) is more commonly recognized as the active site for ethylene, and Cu(111) is inclined to produce methane, while acetate, ethanol, and acetaldehyde production are reported produced on Cu(110). 14 On Cu( 100), the CO-CO dimerization is more likely to occur, leading to ethylene production, while on Cu(111) and Cu(110), *COH intermediate was observed. However, it should be noted that in the CO 2 RR experimental conditions, the copper catalyst exhibits complicated surface structures, where the facet index evolves and reconstructs all the time. For example, with the operando scanning tunneling microscopy, the polycrystalline Cu was observed to reconstruct from Cu (111) to Cu(100) in CO 2 RR conditions. 52 Moreover, the surface stepped sites (high indexed surfaces) are suggested to play an important role in the CO 2 RR. Stepped sites have long been observed on the copper roughened surface. 53,54 Cu(511) was shown to be generated in the electrolyte with surface *H species inducing the reconstruction and stabilizing stepped sites. High indexed surface/stepped sites are effective in adsorbing CO molecules and decreasing the energy barrier of the *CO + HCO* → OCCHO* step. 55 With microkinetic modeling, it is found that CO 2 RR exhibits a significantly higher current density on Cu(211) surface than Cu(111) and Cu(100). 56 As shown in Figure 4A-C, Wang et al. found that the Cu(211), Cu(611) and pocket shape surfaces which represent the high-index facet are not sensitive in the formation of *HOCCOH. On the contrary, the C-O bond dissociation barrier is reduced on all high index surfaces, which indicates that *CCO formation is more favorable on high index surfaces. Considering that *HOCCOH is the key intermediate for C 2 H 4 and *CCO for CH 3 COOH, the high index surface is more selective for acetate, which agrees well with the experimental observation that Cu nanoparticles with a large edge-to-(100) ratio exhibit the highest acetic acid selectivity. 49

| Grain boundaries
On the other hand, grain boundaries are also crucial active sites in CO 2 RR. Kanan group first proposed that the high-density grain boundaries in the interconnected nanocrystallites derived from the constrained environment of Cu 2 O lattice significantly enhance the CO electroreduction to C 2+ product. 58 In a later study, using high-resolution transmission electron microscopy (TEM), Kanan and colleagues calculated the termination lengths on 200 copper nanoparticles by a conversion factor derived from simple geometrical models from twodimensional (2D) images of TEM to 3D particle surface areas, and thus correlated CO electroreduction activity with the density of grain-boundary on the copper nanoparticle. 59 Cu'(101) formed from lattice mismatch is reportedly favorable to adsorbed *CO with high coverage, which promotes oxygenate production. The complex lattice planes of Cu(111), ( 100), and (220) observed with spherical aberration-corrected TEM are stacked upon each other as premature crystalline with rich grain boundaries, which are attributed to the kinetic-driven nucleation instead of Ostwald ripening. 60 DFT calculations indicate that the boundary between (100) and (111) surface is selective for C 2 H 4 production, via the route of proton-electron coupled *CO dimerization to *CO-COH. 61 In a recent study, the oxide-derived copper with a well-controlled (100)/(111) interface shows higher activity for *CO adsorption and subsequent C-C bond formation compared to individual Cu(100) and Cu(111). It is noted that the highly esthetic copper structure was carefully dissected with 14 facets, and the C 2+ selectivity is correlated linearly with the length of the Cu(100)/Cu(111) interface, which is attributed to the lowered free energy barrier of *OCCO intermediate formation. 62

| CO adsorption configurations
The CO adsorption configuration is also one emerging and interesting topic for C 2+ formation. With multiscale simulations, Cheng et al. 57 found that the surface undercoordinated square site substantially decreases the formation energy of *OCCOH. However, not all of the strong CO binding sites are active for C-C coupling reaction, and the best combination for C-C bond formation is one strong CO binding site and at least one under-coordinated neighboring square site ( Figure 4D). 57 Sargent group reported a volcano-curve relationship between the ethylene selectivity and the ratio of *CO with two adsorption configurations on Cu catalyst functionalized with organic molecules. 37 In this study, they summarized the ratios between the intensities of atop bounded CO and bridge bounded CO using in situ Raman spectroscopy, they found that an appropriate amount of CO atop and CO bridge favored the ethylene production. In combination with DFT calculations, they calculated that a CO dimerization with atop:bridge site exhibits the lowest energy barrier, resulting in high C 2+ selectivity. Cuenya and colleagues demonstrated a potential-dependent intensity ratio of Cu-CO stretching mode to the restricted rotation of adsorbed CO in operando Raman experiment, which not only increased with the higher CO coverage but also followed the trend of C 2+ production. 39 DFT calculations revealed that the participation of CO atop played a key role in facilitating the CO-CO dimerization step. Our group introduced phosphorus into the lattice of cupric oxide (CuO) to synthesize Cu 2 P 2 O 7 with a high density of defects. The catalyst was electrochemically reconstructed under CO 2 RR conditions, thus obtaining a highly porous Cu with a high electrochemically active surface area and abundant defects with low-coordinated sites. In situ Raman spectroscopy and DFT calculations suggest that the defects and low-coordinated sites from the reconstructed Cu 2 P 2 O 7 catalyst provide a suitable condition for bridge and atop adsorbed *CO, which was more favorable for C-C coupling. 63 F I G U R E 4 (A) The proposed reaction mechanism of CO electroreduction to C 2+ products and the reaction energies on Cu(611) and related facets: (B, C) Proposed reaction mechanism for CO electroreduction to C 2+ products. The blue path produces ethanol and ethylene, while the green path can also generate acetic acid. (B) to form *HOCCOH through PCET at 0 V versus RHE and −1 V versus RHE and (C) to dissociate *OCCOH to *CCO and *OH on Cu(100) in black, Cu pocket-shape in orange, Cu(611) in red, and Cu(211) in blue. Reproduced from Zhu et al. 49 Copyright © 2021, National Academy of Sciences. (D) The atomic structure of Cu surface site proposed to favor binding to *OCCOH and hence ethanol product. This has a strong CO binding site next to a weak binding site around a square surface configuration. Reproduced from Cheng et al. 57 Copyright © 2017, American Chemical Society. PCET, proton-coupled electron transfer; RHE, reversible hydrogen electrode.

| Anion effects
Ions in electrolytes, especially halogen, have been recognized as effective in improving the CO 2 RR performance of the copper catalyst. It is reported that the ethane pathway is dominant over iodide-derived Cu foams. 48 The Fourier transformed extended X-ray fine structure spectra (EXAFS) presented a Cu-O bond length of 2.15 Å, which matches the bond length of adsorbed ethoxy intermediate (Cu-OCH 2 CH 3 ). Further potentialdependent Raman spectra identified the symmetric -CH 2 and -CH 3 stretching located at 2890 and 2920 cm −1 . Such intermediate was involved in a selectivity-determining step that occurred later in ethane or ethanol production. Our group reported a CuI/Cu composite catalyst substantially improves the partial current density of C 2+ products, which was attributed to the residue iodine inducing a Cu 0 /Cu + interface and improving the CO adsorption. 64 In addition, fluorine on the surface of copper catalyst was effective to assist the water activation and enhance the CO adsorption, and the authors proposed that the fluorine lowered the energy barrier of C-C coupling via two *CHO dimerization, and *CO hydrogenation to *CHO is the rate-determining step, which is arguably uncommon, and the in situ FTIR reveals that the band at 1754 cm −1 is attributed to the *CHO species. 65 For the cations, the majority of studies are dealing with alkali metals, with the conclusion that Cs + > K + > Na + > Li + . [66][67][68][69][70][71][72][73][74] The researchers also checked the effect of cations and proposed that there are three different theories concerning the mechanism as reviewed by Waegele et al., 75 even though the general trend is similar. The first one proposed by Hori et al. 76,77 argued that the potential was charged by cations in the electric double layer through specific adsorption. The cations are adsorbed near the surface via a noncovalent interaction to induce a high electric field in the outer Helmholtz plane. A continuum electrolyte model was presented by Ringe et al. 78 to show that weakly hydrated cations are more likely to concentrate at the outer Helmholtz plane, which lowers the activation energy of the overall reactions, leading to favorable interfacial energetics of the solution layer and stronger CO 2 adsorption. The second theory is that the cations buffer the interfacial pH, by stronger hydrolysis induced by the electrostatic field. 79 The surface-enhanced infrared (IR) spectroscopy was used to probe the pH at the gold electrode surface, and the ratio between the intensity of CO 2 and HCO 3 − band indicates that the larger cations could buffer the pH better than smaller ones. 80 The third theory suggests that the metal cations stabilize the intermediates by local electrostatic interactions within the electrical double layer. 81 The stabilization effects are rationalized by DFT calculations, which indicates that the constrained minima hopping molecular dynamics in the interface between cation and electrode surface were improved by the solvated cations. 82 Recently, Koper and colleagues presented a report showing that CO is only produced on gold, silver, or copper if a metal cation is added to the electrolyte. 83 With well-designed electrochemical experiments and ab-initio molecular dynamics simulations, the partially desolvated cations have arguably three promotional effects for CO 2 electroreduction: thermodynamics which stabilizes CO 2 adsorption, the changing of O-C-O bond angle, and the enhanced electron transfer from the surface to activated CO 2 . An accurate understanding of the role of anion and water is still on the way, but more and more evidence indicates the crucial stabilizing effect of anion on CO 2 .

| Copper chemical state
Most copper catalysts are prone to reconstruct under electroreduction conditions, leading to a significant change in valence state or morphology, especially Cu oxide or other derivatives. 12,84 Electronic structure (oxidation state) of Cu is usually analyzed by in situ X-ray absorption spectroscopy (XAS) and quasi in situ X-ray photoelectron spectroscopy (XPS). [85][86][87][88] Oxidized copper species (CuO x ) have been believed to facilitate the C-C coupling step due to a more stable binding and higher coverage of *CO. [89][90][91] In situ XAS at the Cu-L2,3 edge and Cu-K edge by Cuenya and colleagues indicated that a full or partial reduced Cu + electrode makes dissociative CO 2 adsorption occur, 90,91 which is inhibited over the reduced Cu 2+ because the formation of copper carbonates or the lack of conductivity at negative potentials. Cu δ+ species remaining on the surface have been suggested to be an active site in improving the selectivity towards C 2+ products 92 ; however, the chemical state evolved drastically during CO 2 RR as Cu 2 O reduction is highly favored compared to CO 2 electroreduction or water splitting. 93 Therefore, effective control of the chemical state and electronic structure is extensively studied, among them, devising the interfaces of copper and other oxides should be efficient for CO 2 RR due to the modification of electronic structure or stabilization of Cu δ+ . [94][95][96][97] Wen and colleagues demonstrated that the Cu δ+ species was stabilized at the Cu/Cu x S y interfaces obtained from the structural evolution of the S-HKUST-1. 95 They confirmed the formation of Cu 0 species under CO 2 RR conditions by observing the Cu-Cu peak in the operando EXAFS spectra. Furthermore, they suggested an average oxidation state of Cu which is between 0 and +1 after 15 min reactions by comparing the Cu K-edge adsorption of S-HKUST-1 with Cu foil and Cu 2 O. The energy barrier of *CO dimerization step at the Cu/Cu 2 S interface was then proved to be lower than that over Cu(111).
By calculating the ratios of Cu states based on Cu/ Cu 3 P 2 O 8 catalyst using linear combination fitting from in situ X-ray absorption near edge structure (XANES) spectra, Zheng and colleagues found a coexistence of metallic Cu and Cu 3 P 2 O 8 during CO 2 RR, and the proportion of Cu 2+ species maintains at 38%, leading to high C 2 H 4 production, 97 The analysis of wavelettransform EXAFS further indicated the coexistence of Cu-O and Cu-Cu, and the experiment result was confirmed by DFT calculations that Cu 3 P 2 O 8 /Cu(111) interface to stabilize the *OCCO intermediate and shifted the reaction towards C 2 products.
Wang and colleagues found an increase of Cu δ+ species with the electronegativity of halogen in X-Cu catalysts (X═I, Br, Cl, F) by comparing the Cu K-edge adsorption with Cu and Cu 2 O references ( Figure 5A). 65 DFT calculations further indicate that fluorine modification accelerated the *CO → *CHO step and then enhanced the coupling of *CHO, furthermore, an improved CO adsorption occurred on the increased Cu δ+ sites, giving a highly active and selective catalyst for C 2+ formation in CO 2 RR. Sargent and colleagues introduced Cu-SiO x step sites with enhanced CO coverage and lower formation energy of OCCOH* to promote C-C coupling toward ethylene production. 94 The band at 530 cm −1 from in situ Raman investigation confirmed a stable Cu-O-Si interaction. Further in situ XANES at the Cu K-edge revealed metallic Cu mainly existed in the Cu-SiO x catalyst.
The synergistic effect between Cu 0 and Cu + boosts the selectivity and efficiency of CO 2 RR towards C 2+ products, which is proved by DFT calculations. 46,85,[98][99][100] Yu and colleagues performed linear-combination fitting of Cu K-edge XAS spectra to quantitative determined the retained Cu + in Cu 2 O cavities during CO 2 RR ( Figure 5B,C), 46 The stabilized Cu + species retained up to ∼32%, with ∼68% of Cu 0 , which led to the marked C 2+ selectivity. Strasser and colleagues found that the oxides in the surface layer stabilized on bulk Cu against reduction during CO 2 RR, 100 which resulted in a high ethylene selectivity. Based on the operando XAFS, which showed a combination of Cu and Cu 2 O features during 15 min of reaction. Yu and colleagues observed the Cu-Cu and Cu-O coordination peak in the FT-EXAFS spectra, 88 which suggests that oxygen-bearing Cu clusters were formed in the coordinatively unsaturated Cu paddle wheel (CU-CPW) catalyst. Further analysis by in situ XAS data was recorded in CO 2 and Ar saturated KHCO 3 electrolyte, respectively ( Figure 5D,E). A lower 1s → 4p z transition intensity in CO 2 saturated electrolyte was presented in the Cu K-edge of CU-CPW, indicative of the charge transfer from Cu to the CO 2 for CO 2 − species generation. The DFT calculations further proved that the CU-CPW model facilitated the adsorption and desorption of intermediates. CU-CPW showed a lower energy barrier of the formation towards *CO, *CH 3 O, and *CHOOHC*, which prefers the PCMET process to a deep CO 2 RR.

| Modeling the spectroscopy with DFT calculations
The band of spectroscopy is effectively simulated and identified with DFT calculations. Katayama and colleagues reported that with the in situ surface-enhanced IR absorption spectroscopy, spectra at ca. 2100 cm −1 could be assigned to C≡C corroborated by the calculated vibration frequency of linearly bonded CO in the implicit water layer. 101 The calculated vibration frequency of C═O is modified considering the effect of the water layer: the implicit level and the hydrogen bonding effect, which shift down the C═O frequency in *COOH but shift up in HCO 3

−
. The assignment of C═O stretching mode on Cu(111) and Cu (100) is also supported by DFT calculations. Qiao and colleagues reported that with in situ XAS, Sn in Cu-Sn 3 alloys exist as an Sn δ+ state under reduction potentials, which is in accordance with the XPS indicating that Cu 2p 3/2 peak shifts negatively with increasing tin content. 102 The spectroscopic character was linked with the Bader analysis, showing that Sn donates charge to Cu and is positively charged in the Cu-Sn alloy, which significantly influences the intermediate adsorption and reaction selectivity. Koper reported that using FTIR spectroscopy to detect the copper surface in LiOH solutions, they found that the hydrogenated dimer intermediate *OCCOH was detected at frequencies of 1313, 1200, and 1071 cm −1 , which are corroborated by DFT calculations. 45 It is noted that the influence of water solvent in the simulation of IR absorption spectroscopy was studied by Katayama et al., 101 and the researchers pointed out that an implicit and explicit model of water could be very different in the shift of vibration frequency C-O bond. An implicit solvent model through a quantum-surface-dependent contribution and an explicit model with water molecules are included in the DFT calculations. It is shown that the C═O bond stretching frequency in HCO 3 − is 1355 cm −1 in the implicit model, and the frequency shifts to 1376 cm −1 , which could be attributed to the water molecule in the first coordination shell. However, the C═O stretching frequency in the *COOH is 1649 cm −1 , but shifts down to 1620 cm −1 . It could be seen that the implicit and explicit water model could produce positive or negative trends in a shift of frequency, which should be considered carefully in simulations.

| Machine-learning assisted design of catalysts
Recently, the high-throughput computational screening of catalysts is proved to be an effective way to predict the performance of CO 2 RR catalysts. However, the cost of time and money is huge for screening with DFT calculations, which involves potentially hundreds or even thousands of structures. The machine-learning assisted materials design uses DFT calculation results as training data and produces learned models for much larger data space. Chen et al. 103 applied an extreme gradient boosting regression algorithm to predict CO adsorption energy on 1060 metal − graphene systems and predicted that Co-, Fe-, and Sc-supported on nitrogen or sulfur-doped graphene are active for CO 2 electroreduction. Zhong et al. 104 uses a machine-learning accelerated workflow for screening copper−based intermetallic crystals with CO adsorption energy, and predicted that Cu-Al is a promising candidate material for CO 2 reduction to produce C 2 H 4 , which is successfully validated with experiments. A cyclical neural network framework was trained and refined by around 4000 DFT single-point simulations of the adsorption energy to correlate the E CO and E HOCO , and the (211) surface of nickel gallium bimetallic material was found to be active, while the crystal facets follow the step scaling relation. 105 A regime combining multiple levels of theoretical computations with machine learning was developed to study the copper nanoparticle with large size of 10 nm, and *OCCOH intermediate was shown to be a more accurate descriptor for C 2+ selectivity than *CO. 106 Guo and colleagues reported that the gradient boosting regression algorithm is verified as the most desirable model for investigating single-atom and dual-metal-sites catalysts for CO 2 RR, and LIU ET AL.
Ag-MoPc is identified as the most active with a limiting potential of −0.33 V. 107 Wang and colleagues collected experimental data to build an additive library for electrochemical deposition of Cu catalysts and analyzed with random intersection tree machine-learning method, and then extracted recipe for the best copper catalyst for different products, which is verified by experiments. 108 The combination of machine learning and advanced experimental techniques is promising in revealing the CO 2 RR mechanism and designing novel catalysts.

| SUMMARY
The reaction roadmap of CO 2 RR towards C 2+ products is reviewed and summarized above. In recent years, rapid progress has been witnessed in mechanistic studies with DFT calculations and in situ/operando spectroscopic characterization, which have already led to breakthroughs in the synthesis of high-performance copper catalysts. The energetic profile of the overall reaction pathway and evidence in FTIR spectroscopy of the key intermediates are laying a substantial foundation for development and research in reducing CO 2 to C 2+ products. However, it should be noted that the post-C-C-bonding steps are less investigated compared to the C-C bonding step. Considering that post-C-C bonding determines the selectivity among various C 2+ products, more efforts need to be devoted to understanding that. The crystalline facet of catalysts and electrolytes are imposing a significant influence on the catalytic performance through tuning *CO adsorption configurations and therefore improving C-C coupling reaction barriers. The specifics in the in situ electrochemical environment, especially the copper chemical state and charge distribution in the electrolyte, have been discussed in detail in this perspective. The combination of theoretical calculations with experiments in the simulation of spectroscopy and machine-learning assisted design of catalysts is also included, which shed light on the mechanism and efficient catalysts. We hope the presented perspective help readers to better understand the CO 2 RR at the atomic level, which benefits the development of theory for describing the CO 2 RR towards C 2+ products, and more intrinsically active catalysts for a profitable practical system for CO 2 electrolysis in the future.