Subsurface oxide plays a critical role in CO₂ activation by Cu(111) surfaces to form chemisorbed CO₂, the first step in reduction of CO₂

Significance

Combining ambient pressure X-ray photoelectron spectroscopy experiments and quantum mechanical density functional theory calculations, this work reveals the essential first step for activating CO₂ on a Cu surface, in particular, highlighting the importance of copper suboxide and the critical role of water. These findings provide the quintessential information needed to guide the future design of improved catalysts.

Abstract

A national priority is to convert CO₂ into high-value chemical products such as liquid fuels. Because current electrocatalysts are not adequate, we aim to discover new catalysts by obtaining a detailed understanding of the initial steps of CO₂ electroreduction on copper surfaces, the best current catalysts. Using ambient pressure X-ray photoelectron spectroscopy interpreted with quantum mechanical prediction of the structures and free energies, we show that the presence of a thin suboxide structure below the copper surface is essential to bind the CO₂ in the physisorbed configuration at 298 K, and we show that this suboxide is essential for converting to the chemisorbed CO₂ in the presence of water as the first step toward CO₂ reduction products such as formate and CO. This optimum suboxide leads to both neutral and charged Cu surface sites, providing fresh insights into how to design improved carbon dioxide reduction catalysts.

The discovery of new electrocatalysts that can efficiently convert carbon dioxide (CO₂) into liquid fuels and feedstock chemicals would provide a clear path to creating a sustainable hydrocarbon-based energy cycle (1). However, because CO₂ is highly inert, the CO₂ reduction reaction (CO₂RR) is quite unfavorable thermodynamically. This makes identification of a suitable and scalable catalyst an important challenge for sustainable production of hydrocarbons. We consider that discovering such a catalyst will require the development of a complete atomistic understanding of the adsorption and activation mechanisms involved. Here the first step is to promote initiation of reaction steps.

Copper (Cu) is the most promising CO₂RR candidate among pure metals, with the unique ability to catalyze formation of valuable hydrocarbons (e.g., methane, ethylene, and ethanol) (2). However, Cu also produces hydrogen, requires too high an overpotential ($>$1 V) to reduce CO₂, and is not selective for desirable hydrocarbon and alcohol CO₂RR products (2). Despite numerous experimental and theoretical studies, there remain considerable uncertainties in understanding the role of Cu surface structure and chemistry on the initial steps of CO₂RR activity and selectivity (3, 4). To reduce CO₂ to valuable hydrocarbons, a source of protons is needed in the same reaction environment (2), with water (H₂O) the favorite choice. Thus, H₂O is often the solvent for CO₂RR, representing a sustainable pathway toward solar energy storage (1). However, we lack a comprehensive understanding of how CO₂ and H₂O molecules adsorb on the Cu surface and interact to first dissociate the CO₂ (5, 6). An overview of the various surface reactions of CO₂ on Cu(111) is reported in Fig. 1, illustrating the transient carbon-based intermediate species that may initiate reactions.

Fig. 1.

Overview of surface reactions of CO₂ on Cu(111) under various in situ conditions. Here the $g$-CO₂ indicates gas-phase CO₂, $l$-CO₂ indicates linear (physisorbed) CO₂, and $b$-CO₂ indicates bent (chemisorbed) CO₂. (A and B) The forms of absorbed CO₂ on pristine Cu(111). (A) Both physisorbed $l$-CO₂ and ${\text{CO}}_2^{\delta -}$ are observed at 75 K for pressures up to 10⁻⁶ Torr. (B) Only ${\text{CO}}_2^{\delta -}$ is observed at 298 K for pressures ranging from 10⁻⁶ Torr to 0.1 Torr. (C) The adsorption of CO₂ when a subsurface oxide structure is deliberately incorporated into Cu(111) but without additional H₂O. In this work we observe that a subsurface oxide coverage of about 0.08 ML is responsible for stabilizing $l$-CO₂ at 298 K and 0.7 Torr. Here O_sub indicates subsurface oxygen between the top two layers of Cu. (D) The cooperative interaction of codosed CO₂ and H₂O on Cu(111) composed of 0.08 ML of subsurface oxide, leading to the first reduction step of CO₂ by adsorbed H₂O_ads; HCOO$-$ indcates adsorbed formate.

Previous studies using electron-based spectroscopies observed physisorption of gas-phase $g$-CO₂ at 75 K, whereas a chemisorbed form of CO₂ was stabilized by a partial negative charge induced by electron capture (${\text{CO}}_2^{\delta -}$) (Fig. 1A) (7, 8). The same experiments showed that no physisorption is observed upon increasing the temperature of the Cu substrate to room temperature (r.t.) (298 K) (Fig. 1B). Previous ex situ studies performed in ultrahigh vacuum (UHV) (about 10⁻⁹ Torr) after relatively low CO₂ exposures [from a few to hundreds of Langmuir (L)] at temperatures between 100 K and 250 K did not reveal CO₂ adsorption or dissociation on clean Cu(100) (9), Cu(110) (10), and Cu(111) (11). However, Nakamura et al. (12) showed that when the exposure is increased to sensibly higher values (pressures ranging between 65 Torr and 1,300 Torr for hundreds of seconds), a nearly first-order dissociative adsorption of CO₂ on clean Cu(110) can be detected between 400 K and 600 K (with an activation energy of about 67 kJ$\cdot$mol⁻¹), according to the reaction CO_2,g $\to$ CO_g + O_ads (where O_ads stands for surface adsorbed oxygen). A similar phenomenology was also observed by Rasmussen et al. (9) on clean Cu(100) for CO₂ pressures of about 740 Torr and temperatures in the range of 475–550 K (finding an activation energy of about 93 kJ$\cdot$mol⁻¹). On the other hand, a recent study by Eren et al. (13) performed at much lower CO₂ partial pressures (between 0.05 Torr and 10 Torr) revealed that CO₂ can dissociatively adsorb on Cu(100) and Cu(111) with the consequent formation of surface oxygen as well. Indeed it has been suggested that the CO₂ might dissociate more easily on preoxidized Cu surfaces (3), but there is little evidence to support this important concept. Activation of CO₂ via assumed chemisorbed CO₂ species was reported also on Cu stepped surfaces (11, 14), but direct in situ proof of the existence of such species on Cu(111) is lacking. These uncertainties and discrepancies indicate the importance of determining the initial species formed while exposed to realistic gas pressures of CO₂ and H₂O (13, 15).

To advance this understanding, we investigated in detail the initial steps of CO₂ adsorption both alone and in the presence of H₂O on Cu(111) and suboxide surfaces (Cu_x=1.5,25O) via in situ probing of the electronic structure of the surface and reaction products, using ambient pressure X-ray photoelectron spectroscopy (APXPS) performed with soft X-rays (200–1,200 eV) at the solid/gas interface. These studies are complemented with molecular structures and binding free energies of the reaction products at the M06L level (16) of density functional theory (DFT) that was optimized for molecular clusters and reaction barriers. This combination of experiments and calculations allows us to conclude that the presence of suboxide species below the Cu surface and the presence of H₂O play a crucial role in the adsorption and activation of CO₂ on Cu (Fig. 1). Specifically, the presence of subsurface oxygen leads to a specific interaction with gas-phase CO₂ that stabilizes a physisorbed linear CO₂ configuration ($l$-CO₂, Fig. 1C). In addition, H₂O in the gas phase ($g$-H₂O), aided by small amounts of suboxide, drives CO₂ adsorption through the transition from the linear physisorbed state to a bent chemisorbed species ($b$-CO₂), which with the aid of H₂O promotes the initial reduction of CO₂ to formate (HCOO$-$, Fig. 1D).

The Cu surface exposing mainly the Cu(111) orientation was prepared in situ from a polycrystalline sample, by repeated argon (Ar) sputtering (normal incidence, 2 keV, 45 min) and annealing cycles in hydrogen (0.15 Torr) at 1,100 K (for 60 min), to obtain a typical 1 × 1 reconstruction as shown by the low-energy electron diffraction (LEED) pattern in Fig. 2A (17). Scanning electron microscopy (SEM) measurements Fig. 2A determine that this sputtering and annealing procedure leads to crystalline regions with tens of micrometers mean sizes. The characterized sample surface location remained unchanged throughout the APXPS experiments. The collected spectra were averaged over a beam spot size of ∼0.8 mm in diameter. Although we cannot exclude possible contributions from the presence of grain boundaries, averaging the data over the large probed area led to an eventual grain boundary contribution less than 1% of the overall measured signal, which is below the detection limit. Therefore, their physical/chemical features were not captured in the spectra and do not constitute the focus of this study.

Fig. 2.

Investigation of various Cu surfaces using APXPS. (A) LEED pattern obtained at an electron kinetic energy of 110 eV and SEM micrograph of the Cu surface after sputtering and annealing cycles obtained by detecting the secondary electrons (SE) with a kinetic energy of the primary beam of 5 keV. (B) Schematic of the APXPS measurements with the highlighted probed volume (3$\lambda$) along the (111) direction. (C and D) C 1s and O 1s photoelectron peaks and multipeak fitting results obtained for the various experimental conditions and investigated surfaces (at r.t., 298 K): (experimental condition A) pure CO₂ 0.7 Torr on metallic Cu(111); (experimental condition B) CO₂ + H₂O 0.7 Torr on metallic Cu(111); (experimental condition C) CO₂ + H₂O 0.7 Torr on Cu_x=2.5O; (experimental condition D) CO₂ + H₂O 0.7 Torr on Cu_x=1.5O); and (experimental condition E) pure CO₂ 0.7 Torr on Cu_x=1.5O. The experimental conditions are summarized in Table 1.

During the APXPS experiments (Fig. 2B) performed at r.t. (298 K), CO₂ was first introduced at 0.7 Torr on the pristine metallic Cu(111) surface. For the other experimental conditions and investigated surfaces (see Table 1 and Supporting Information for further details), the CO₂ partial pressure ($p$(CO₂)) was kept at 0.35 Torr whereas the total pressure ($p_{\text{tot}}$) was kept constant at 0.7 Torr by codosing H₂O. The APXPS measurements were performed while dosing CO₂ on both metallic Cu(111) and Cu_x=1.5O surfaces, whereas CO₂ and H₂O were codosed on metallic Cu(111), Cu_x=1.5O, and Cu_x=2.5O suboxide surfaces (18). The sample surface was clean and no evident C- or O-based contaminations were observed after the cleaning procedure, as shown in Fig. S1. In addition, the in situ mass analysis of the reactants (O₂, CO₂, and H₂O), using a conventional quadrupole mass spectrometer (QMS) mounted on the analysis chamber (and operating at a partial pressure of about 10⁻⁶ Torr), did not reveal CO cross-contaminations of the gases. However, Fig. S2 shows that, concomitantly with the gas dosing (for pressures exceeding 10⁻⁶ Torr), uptake of carbon contaminations readily occurred [the corresponding binding energy (BE) being centered at 285.1 eV]. Therefore, we cannot completely exclude eventual side reactions and interplay between carbon contaminations and the copper surface.

Table 1.

Various Cu surface structures and experimental conditions explored with APXPS

Experimental condition	Surface structure	Gas environment	Total pressure, Torr	Temperature, K
A	Metallic Cu(111)	CO2	0.7	298
B	Metallic Cu(111)	CO2 + H2O (1:1)	0.7	298
C	${\text{Cu}}_{x=2.5}$O	CO2 + H2O (1:1)	0.7	298
D	Cux=1.5O	CO2 + H2O (1:1)	0.7	298
E	Cux=1.5O	CO2	0.7	298

Fig. S1.

Survey scan acquired in UHV at 650 eV at r.t. (298 K) on the Cu(111) surface after the cleaning cycle. Insets report the scans over the O 1$s$ and C 1$s$ spectral region (acquired in UHV at r.t. with a photon energy of 632 eV and 387 eV, respectively). It is possible to see that the surface did not present traces of O- or C-based contaminations after the cleaning and annealing procedure (surface preparation).

Fig. S2.

C 1$s$ photoelectron peak acquired at r.t. (298 K) as a function of CO₂ and H₂O partial pressures in the analysis chamber [on metallic Cu(111)]. It can be seen that upon H₂O codosing, a downward shift of the $l$-CO₂ BE takes place with the formation of $b$-CO₂. $b$-CO₂ remains stable upon H₂O removal and final pumping of the chamber to HV conditions. Note that the increase in CO₂ partial pressure leads to the emergence and subsequent growth of a gas-phase C 1$s$ core-level signal (the onset occurs between 0.05 Torr and 0.10 Torr). Therefore, the spectral intensity shown here is rescaled by the intensity of the CO₂ gas phase above the onset pressure. Furthermore, the relatively high gas pressure used in this study attenuates the photoelectron intensity coming from the sample surface, leading to an overall apparent decrease of the surface species intensity.

To understand how interactions between the catalyst surface and CO₂ determine the mechanisms of the initial CO₂ reduction steps, we established the experimental conditions under which a chemisorbed CO₂ state can be stabilized. This provides the basis for tailoring novel catalysts with improved electrochemical performance toward the CO₂RR.

Previously it was difficult to probe these early steps experimentally because r.t. studies require pressures of CO₂ high enough to stabilize a physisorbed configuration sufficiently to allow detailed investigations of various adsorption dynamics, but this high-pressure gas makes it difficult to use electron-based spectroscopies. Our use of APXPS overcomes this difficulty (19–21). To discriminate between physisorbed and chemisorbed CO₂, we monitor the spectral BE shifts of the corresponding C 1s and O 1s photoelectron peaks as a function of the different surfaces and experimental conditions. Physisorption mediated by weak van der Waals (vdW) interactions [surface binding energies of a few millielectronvolts, comparable to $k_{B}T$ = 25.7 meV at 298 K (7)] generally leaves the adsorbate electronic structure unchanged compared with its gas-phase configuration (7, 22–24). In contrast, the chemical bonding needed to form chemisorbed CO₂ on the Cu surface redistributes the electronic density in the adsorbate, leading to appreciable BE shifts compared with the physisorbed state (8).

The adsorption state of CO₂ and the overall surface chemistry of the various systems were monitored by multipeak deconvolution on both the C 1s and O 1s photoelectron spectra (Fig. 2 C and D), using chemically shifted components sensitive to the initial state effects. Fig. S3 A and B reports the integrated peak areas of the chemically shifted components for C 1s and O 1s deconvolution, respectively, normalized by the total area underneath the spectra. C 1s and O 1s photoelectron spectra were acquired under APXPS conditions at photon energies of 387 eV and 632 eV, respectively. Because the kinetic energy of the escaping C 1s and O 1s photoelectrons is about 100 eV, the probed depth, 3$\lambda$ ($\lambda$ is the electron mean free path) is about 1.2 nm, from the topmost layer (Fig. 2B).

Fig. S3.

Multipeak fitting procedure and normalized integrated peak areas for the different chemically shifted components on C 1$s$ (A) and O 1$s$ (B) spectral regions and multipeak fitting procedure results obtained for the different experimental conditions and investigated surfaces (at r.t., 298 K). (Experimental condition A, summarized in Table 1) Pure CO₂ 0.7 Torr on metallic Cu(111); (experimental condition B) CO₂ + H₂O 0.7 Torr on metallic Cu(111); (experimental condition C) CO₂ + H₂O 0.7 Torr on ${\text{Cu}}_{x=2.5}$O; (experimental condition D) CO₂ + H₂O 0.7 Torr on Cu_x=1.5O; (experimental condition E) pure CO₂ 0.7 Torr on Cu_x=1.5O.

The deconvoluted C 1s spectra (see Supporting Information for further details) exhibit two main spectral regions: (i) At low BE we see chemical species that can be assigned as graphitic carbon (284.5 eV), $s{p}^3$ (C-C) carbon (285.2 eV), and C-O(H) bonds (286.3 eV), based on the literature values (15). (ii) At higher BE we see spectral fingerprints of higher oxidized carbon structures and adsorbed CO₂, where deconvolution of the spectra indicates the presence of formate (HCOO$-$) (287.3 eV), chemisorbed (denoted $b$-CO₂ for bent), and physisorbed CO₂ (denoted $l$-CO₂ for linear) (287.9 eV and 288.4 eV, respectively) and carbonate ($-$CO₃) (289.4 eV) (15). Finally, a sharp peak centered at about 293.3 eV corresponds to the photoelectron emission of $g$-CO₂ (Fig. S4).

Fig. S4.

Example of gas phases of CO₂ and H₂O detected with APXPS on C 1$s$ (A) and O 1$s$ (B) spectral regions [B: CO₂ + H₂O 0.7 Torr on metallic Cu(111)] (r.t., 298 K).

To disentangle the role of oxygen on the surface and subsurface regions, we carried out a similar analysis on O 1s core-level spectra (Fig. 2D). The analysis performed on C 1s was used to help the interpretation of the O 1s spectral envelope while also accounting for the different relative abundances. As with C 1s, we partition the O 1s spectral window into three regions. At low BEs we identify the states of O bonded as follows: (i) surface adsorbed O (Cu-O_ads)on metallic Cu and on suboxidic Cu_xO structures (Cu_xO-O_ads) at 531.0 eV and 529.6 eV, respectively (15, 25–27); (ii) subsurface adsorbed O (O_sub) on metal Cu (Cu-O_sub) at 529.8 eV (27) (as we discuss in a later section, such a presence of suboxide plays an important role in stabilizing the $l$-CO₂); and (iii) for ${\text{Cu}}_{x>1}$O the O 1s is centered at 530.3 eV (15, 18, 25). It is noteworthy that O_ads groups on the Cu surface can serve as nucleation sites for hydroxylation when in the presence of H₂O. However, the detection of eventual Cu-OH groups via photoelectron chemical shift identification is complicated by the fact that in the same spectral range (530.6–530.8 eV) several oxygen-based species overlap (such as formate, C-(OH), and O-R species with R = $-$CH₃, $-$CH₂CH₃). On the other hand, the presence of the C 1s spectral counterpart of formate and C-(OH) (well discriminated in BE) allowed us to build up a consistent O 1s fitting. Therefore, although we cannot completely exclude the presence of surface Cu-OH, its concentration is most likely below the detection limit of the technique (about 0.02 ML). C-O bonds fall instead in the middle region, namely between 530.8 eV and 532.0 eV. Within this range, from lower to higher BE, we identify chemisorbed CO₂, C-O(H), and formate (HCOO$-$) overlapping at 530.8 eV; $l$-CO₂ at 531.4 eV; and carbonates at 531.8 eV (15, 25). Finally, at high BE we observed adsorbed H₂O (H₂O_ads) at 532.4 eV (15, 25).

The difficulty in discriminating between Cu⁰ and Cu⁺ using Cu core levels has been well established and is clearly evidenced from Fig. S5A, reporting the Cu 3p photoelectron spectra. To overcome this limitation, the various Cu surfaces were characterized by means of the Cu Auger L₃M_4,5M_4,5 transition and the valence band (VB) as described in Discussion and as reported in Figs. S5B and S6.

Fig. S5.

Cu 3$p$ (A) and VB (B) reported for the different experimental conditions and investigated surfaces (at r.t., 298 K). (Experimental condition A) Pure CO₂ 0.7 Torr on metallic Cu(111); (experimental condition B) CO₂ + H₂O 0.7 Torr on metallic Cu(111); (experimental condition C) CO₂ + H₂O 0.7 Torr on ${\text{Cu}}_{x=2.5}$O; (experimental condition D) CO₂ + H₂O 0.7 Torr on Cu_x=1.5O; (experimental condition E) pure CO₂ 0.7 Torr on Cu_x=1.5O. Metallic Cu, ${\text{Cu}}_{x=2.5}$O, and Cu_x=1.5O were collected in UHV as reference.

Fig. S6.

APXPS investigation of the surface chemistry for various surfaces and conditions (performed at r.t., 298 K). (A) Cu L₃M_4,5M_4,5 Auger transition for the investigated systems and the reference spectra for metallic Cu, ${\text{Cu}}_{x=2.5}$O, and Cu_x=1.5O acquired under UHV conditions. The deconvolution of the spectra was performed accordingly with the final state simulation of the primary excitation spectrum F(E). (B) VB spectra at the Fermi edge for the investigated systems and the reference spectra for metallic Cu, ${\text{Cu}}_{x=2.5}$O, and Cu_x=1.5O acquired under UHV conditions. (Experimental condition A) Pure CO₂ 0.7 Torr on metallic Cu(111); (experimental condition B) CO₂ + H₂O 0.7 Torr on metallic Cu(111); (experimental condition C) CO₂ + H₂O 0.7 Torr on ${\text{Cu}}_{x=2.5}$O; (experimental condition D) CO₂ + H₂O 0.7 Torr on Cu_x=1.5O; (experimental condition E) pure CO₂ 0.7 Torr on Cu_x=1.5O.

It is important to note that the BE of the aforementioned chemically shifted components for C 1s and O 1s do not change with the experimental conditions (within the spectral resolution, ∼0.15 eV), with an exception only for the adsorbed CO₂, where the adsorption configuration ($b$-CO₂ vs. $l$-CO₂) depends on the experimental conditions. In particular, we observe an important decrease by ∼0.50 eV (Fig. 2B) of the C 1s BE when CO₂ is codosed with H₂O on the metallic Cu(111) surface (Fig. S2). This work was inspired by similar experiments previously reported by Deng et al. (15), where they dosed CO₂ and H₂O separately and together on a polycrystalline (nonoriented) Cu sample. The authors observed the presence of an adsorbed CO₂ species at r.t. (with the corresponding C 1s centered at BE = 288.4 eV), which they labeled as a negatively charged adsorbed “${\text{CO}}_2^{\delta -}$.” We believe their adsorbed CO₂ species could actually be attributed to the $l$-CO₂ configuration observed and computed in this work. Interestingly, however, the authors did not observe a new component in the adsorbed CO₂ spectral region (287.9–288.5 eV, i.e., the $b$-CO₂), passing from the exposure to pure CO₂ to CO₂+ H₂O. In addition, we observe only a weak presence of reaction products between CO₂ and H₂O codosed at r.t. (Fig. 2 C and D), whereas Deng et al. (15) observed the significant development of the methoxy group spectral component ($-$OCH₃, BE = 285.2 eV) when codosing CO₂ and H₂O. These differences might be addressed by the higher experimental gas pressures used in this study, as well as potentially different investigated surface structures formed by different surface cleaning and annealing procedures. Overall, these differences can potentially lead to a different surface reactivity. The results reported by Deng et al. (15) have been obtained on a polycrystalline surface likely exposing extended grain boundaries and coexistence of different surface orientations, whereas the present study was performed on an oriented surface. Our experimental results can be explained in terms of two different adsorption configurations of CO₂: (i) physisorbed linear CO₂ ($l$-CO₂) above 0.150 Torr (Fig. 3) stabilized by small amounts of residual O_sub and (ii) chemisorbed CO₂ ($b$-CO₂) that is formed only after adding H₂O, but also requires O_sub.

Fig. 3.

Predicted structures for 1/4 ML of physisorbed $l$-CO₂ on various Cu surfaces (Cu, light blue; C, brown; O, red, but O_sub is marked in orange). (A–D) Top and side views of (A) pristine Cu(111), $\mathrm{\Delta }$G = +0.27 eV, $p_{\text{thresh}}$ = 33 atm; (B) Cu(111) with 1/4 ML O_ads (row 1 of Table 2) $\mathrm{\Delta }$G = +0.21 eV, $p_{\text{thresh}}$ = 3 atm; (C) Cu(111) with 1/4 ML O_sub, (row 2 of Table 2) $\mathrm{\Delta }$G = $-$0.39 eV, $p_{\text{thresh}}$ = 2 × 10⁻⁷ Torr; and (D) Cu(111) with 1/4 ML of both O_ads and O_sub (row 3 of Table 2), $\mathrm{\Delta }$G = $-$0.13 eV, $p_{\text{thresh}}$ = 7 Torr. Both C and case D are consistent with experiment.

For pure CO₂ on pristine metallic Cu(111) (Fig. 2 C and D, experimental condition A), we observe experimentally a weakly adsorbed $l$-CO₂ at 298 K with a pressure of 0.7 Torr CO₂. This does not agree with our DFT calculations, performed at the M06L level, including the electron correlation required for London dispersion (vdW attraction) (16). We find an electronic binding energy of $\mathrm{\Delta }$E = $-$0.36 eV and an enthalpy of binding of $\mathrm{\Delta }$H(298 K) = $-$0.30 eV [after including zero-point energy (ZPE) and specific heat]; however, due to the large decrease in entropy from the free CO₂ molecule, the free energy for $l$-CO₂ is uphill by $\mathrm{\Delta }$G(298 K, 0.7 Torr) = $+$0.27 eV. These energetics would require pressures of 33 atm (∼25 M Torr) for the adsorbed $l$-CO₂ to be observed at 298 K on pure metallic Cu(111). This is in line with previous experimental observations reported in the literature (also Fig. 1), where only $l$-CO₂ was observed on metallic Cu(111) surface at 298 K (7).

On the other hand, our DFT calculations show that very small amounts of suboxide (one suboxide O per every four surface Cu in our calculations, but likely much smaller levels are sufficient) lead to a negative free energy of $\mathrm{\Delta }$G(298 K, 0.7 Torr) = $-$0.12 eV, which would stabilize physisorbed $l$-CO₂ at our experimental conditions. Indeed, our experiments find evidence for small amounts (∼0.08 ML) of surface suboxide on our freshly prepared Cu(111) (Fig. 2D, experimental condition A). Such subsurface adsorbed O (denoted Cu-O_sub) has been observed often near the Cu surface, most likely resulting from oxygen impurities in the chamber (28) or partial dissociative adsorption of CO₂ (13). Interestingly, even if CO₂ is still in a linear configuration (similar to the gas phase), we observe experimentally that the O 1s and C 1s core-level BEs of $l$-CO₂ shift downward by ∼4.9 eV compared with $g$-CO₂ (Fig. S4). This important shift means that an actual interaction is taking place between the adsorbate and the surface (7, 22), although the adsorption state still resembles physisorption.

To interpret these findings, we investigated in detail the influence of O_sub on the formation of $l$-CO₂, using various levels of DFT calculations. These calculations are discussed in detail in Supporting Information. It is well known that standard DFT methods [e.g., generalized gradient approximation (GGA) and local-density approximation (LDA)] do not account for London dispersion, which is usually included with empirical corrections (29). However, there is no rigorous basis for the empirical vdW correction for Cu. Instead we use the M06L version of DFT that includes both kinetic energy and exchange correlation functions optimized by comparing to a large benchmark of known vdW clusters with accurately known bonding energies (16). Further details are presented in Computational Details of DFT Calculations, Dataset S1, and Tables S1 and S2.

Table S1.

The performance of M06L and PBE-D3 on Cu fcc metal

Method	Ecohesive, kJ/mol	$a$, Å
Experiment (44)	340	3.595
M06L	411	3.603
PBE-D3	394	3.569

Note that the zero-point phonon effects were removed from the reference (Experiment).

Table S2.

The performance of M06L and PBE-D3 on CO₂ molecular crystal (ground-state $Pa\overline{3}$ structure)

Method	Hsublimation, kJ/mol	$a$, Å	$R_{\text{C}-\text{O}}$, Å
CCSD(T)/CBS (45)	27	5.475	1.162
M06L	28	5.426	1.162
PBE-D3	24	5.673	1.168

Note that the zero-point phonon effects were removed from the reference (coupled cluster singles, doubles, and perturbative triples [CCSD(T)] calculation extrapolated to the complete basis set (CBS) limit).

Physisorbed CO₂ on Cu(111)

Fig. 3A shows the predicted surface structure for 1/4 monolayer (ML) equivalents (MLE) of CO₂ on metallic (O_sub-free) Cu(111). The physisorbed $l$-CO₂ molecules have a C-O bond distance of 1.164 Å compared with 1.163 Å in gas phase and O-C-O angles of 179°, with an equilibrium distance of 3.11 Å from the C atom of CO₂ to the Cu surface, characteristic of weak vdW interactions. The quantum mechanical (QM) electronic bond energy to the surface is $\mathrm{\Delta }$E = $-$0.36 eV with $\mathrm{\Delta }$H(298 K) = $-$0.30 eV enthalpy of bonding (after including ZPE and specific heat). However, the large decrease in entropy from the free CO₂ molecule leads to a free energy for physisorbed CO₂ that is unfavorable by $\mathrm{\Delta }$G(298 K, 0.7 Torr) = +0.27 eV, which would require a pressure of 33 atm to observe at 298 K.

Physisorbed CO₂ with O on Cu(111)

The experimentally observed O 1s shifts indicate a small amount of surface and/or subsurface adsorbed O (denoted O_ads and O_sub, respectively) is present in our pristine Cu(111). Compared with the O 1s BE of O in the $l$-CO₂ configuration, we observe an experimental shift ($\delta$O_ads) of $-$0.4 eV for O_ads and an experimental shift ($\delta$O_sub) for O_sub of $-$1.6 eV. To deduce the nature of this O_ads, we consider the three cases reported in Table 2.

Table 2.

DFT models of Cu(111) with various distributions of surface O atoms and of calculated O 1s BE with experimental APXPS results

Method	Structure	Predicted $\delta$Oads and $\delta$Osub
DFT	1/4 ML Oads	$\delta$Oads = $-$2.2 eV
DFT	1/4 ML Osub	$\delta$Osub = $-$1.3 eV
DFT	1/4 ML Oads + 1/4 ML Osub	$\delta$Oads = $-$0.3 eV; $\delta$Osub = $-$1.5 eV
APXPS	0.06 ML Oads + 0.08 ML Osub	$\delta$Oads = $-$0.4 eV; $\delta$Osub = $-$1.6 eV

For computational convenience we assumed a 2 $\times$ 2 surface cell, but the experimental O_sub coverage is about 0.08 MLE. For the 2 $\times$ 2 unit cell, our DFT calculations find two cases with O 1s BE consistent with experiment. Fig. 3C with one O_sub per cell leads to a BE = $-$1.35 eV whereas Fig. 3D with one O_sub and one O_ads leads to BE = 0.31 and 1.54 eV. Referencing to gas-phase O₂ (standard conditions), Fig. 3D is $\mathrm{\Delta }$G = $-$2.34 eV more stable than Fig. 3C. For case Fig. 3C we predict $\mathrm{\Delta }$G = $-$0.39 eV bonding for $l$-CO₂ (a pressure threshold of 2 $\times$ 10⁻⁷ Torr), whereas Fig. 3D leads to $\mathrm{\Delta }$G = $-$0.13 eV with a pressure threshold of 7 Torr, both consistent with experiment.

Simultaneous dosing of CO₂ in the presence of H₂O leads to a dramatic change in the character of the surface CO₂, showing clearly the adsorption characteristics for chemisorbed $b$-CO₂. For a Cu(111) surface that includes some surface suboxide, the DFT calculations lead to several local minima (Fig. 4): (i) physisorbed $l$-CO₂ plus H₂O_ad, (ii) chemisorbed $b$-CO₂ plus H₂O_ads (Fig. 4 A–C), (iii) reacted COOH_ads plus OH_ads (Fig. S7A), and (iv) HCOOH plus surface O_ads (Fig. S7B).

Fig. 4.

M06L predicted structures for chemisorbed $b$-CO₂ with H₂O on Cu(111) with different levels of O_sub. $\mathrm{\Delta }$G is reported for 298 K, and $p$ = 0.35 Torr for H₂O and CO₂. (A–C) Top and side views with chemical illustration of predicted structures (A) on pristine Cu(111), $\mathrm{\Delta }$G = +1.07 eV; (B) on Cu(111) with 1/4 ML O_sub, $\mathrm{\Delta }$G = $-$0.06 eV; and (C) on Cu(111) with 1/2 ML O_sub, $\mathrm{\Delta }$G = +0.28 eV.

Fig. S7.

Illustration of structures for (A) reacted COOH_ad plus OH_ad and (B) HCOOH plus surface O_ad. This complements the structures shown in Fig. 4. On a pristine Cu(111) surface, $\mathrm{\Delta }$G(A) = +1.36 eV, $\mathrm{\Delta }$G(B) = +1.26 eV; on Cu(111) with 1/4 ML O_sub, $\mathrm{\Delta }$G(A) = +0.20 eV, $\mathrm{\Delta }$G(B) = $-$0.05 eV; on Cu(111) with 1/2 ML O_sub, $\mathrm{\Delta }$G(A) = +0.71 eV, $\mathrm{\Delta }$G(B) = +0.48 eV. All free energies are referenced to $p_{\text{CO}2}$ = $p_{\text{H}2\text{O}}$ = 0.35 Torr, and T = 298 K.

In the case of Cu(111) without O_sub (Fig. 4A), the C atom of $b$-CO₂ is chemically bonded to a surface Cu⁰, whereas the two O atoms accommodate the partial negative charge transferred from the Cu surface, with one stabilized by hydrogen bonding to H₂O_ad. However, this $b$-CO₂ leads to a QM binding energy of $\mathrm{\Delta }$E = $-$0.23 eV, but including vibrational and entropy contributions we find $b$-CO₂ is unstable, with $\mathrm{\Delta }$G(298 K, 0.7 Torr) = 1.07 eV, which agrees with our experiments.

When the O_sub is increased to 1/4 ML (Fig. 4B), we find that the C atom is chemically bonded to two surface Cu⁰, one O atom is chemically bonded to one Cu⁰ center, and the other O atom is stabilized by the surface Cu⁺ pulled up by H₂O_ad. This $b$-CO₂ leads to $\mathrm{\Delta }$G(298 K, 0.7 Torr) = $-$0.06 eV, which is stable in agreement with our experiments.

However, increasing the O_sub to 1/2 ML, we predict that $\mathrm{\Delta }$G(298 K, 0.7 Torr) = $+$0.28 eV, which is unstable. Here the C atom is chemically bonded to a surface Cu⁺ that shares an O atom bearing a partial charge (stabilized by a hydrogen bonding to H₂O_ad on surface Cu⁺). Our experiments also show that increased levels of O_sub decrease the binding of $b$-CO₂. Thus, we find that chemisorbed $b$-CO₂ is stable only for the case in Fig. 4B with 1/4 ML O_sub. Having more O_sub or none at all destabilizes $b$-CO₂. We explain this in terms of the distinct interactions of Cu⁰ and Cu⁺ induced by the Cu(111)${\text{O}}_{\text{sub},x=0.25}$.

This result of an optimum O_sub for $b$-CO₂ is in agreement with our experiments for CO₂ and H₂O codosing on the ${\text{Cu}}_{x=2.5}$O and Cu_x=1.5O suboxide structures, which shows both $b$-CO₂ and $l$-CO₂, but with a $l$-CO₂/$b$-CO₂ ratio of 3.8 and 5.3 for ${\text{Cu}}_{x=2.5}$O to the Cu_x=1.5O structure, respectively (the ratio was determined from both C 1s and O 1s spectra) (Fig. S3 A and B). In addition, Fig. S8 reports the experimental results of exposing the Cu_x=1.5O structure to 0.7 Torr of 1:1 CO₂ and O₂. In this case we do not observe chemisorbed $b$-CO₂, but only physisorbed $l$-CO₂ and its conversion to surface $-$CO₃ (carbonate).

Fig. S8.

O 1$s$ and C 1$s$ photoelectron peaks acquired under APXPS conditions at r.t. (298 K) on Cu_x=1.5O and at r.t. and at a total pressure of CO₂ and O₂ (1:1) of 0.7 Torr. The relative abundance of the different chemically shifted components is expressed in percentage of the total integrated area of each photoelectron spectrum.

The DFT calculations predict that on Cu(111)O_sub,x=0.25, $b$-CO₂ can react with H₂O_ad to form formate plus OH_ads, but the product is unstable in our conditions, with $\mathrm{\Delta }$G(298 K, 0.7 Torr) = +0.20 eV, making it endothermic from $b$-CO₂ in Fig. 4B by $\mathrm{\Delta }$G(298 K, 0.7 Torr) = 0.26 eV [it is 0.43 eV endothermic for Cu(111)O_sub,x=0.5]. On the other hand, our DFT calculations predict that this formate can extract an H from the $-$OH to form formic acid plus O_ads, which is stable with $\mathrm{\Delta }$G(298 K, 0.7 Torr) = $-$0.05.

We expect that learning how to tune the character of the surface atoms (Cu⁰ vs. Cu⁺ in this case) to manipulate these relative energetics of $l$-CO₂ plus H₂O, $b$-CO₂ plus H₂O, formate plus OH, and formic acid plus O_ads may allow us to design modified systems aimed at accelerating these reaction steps. For example, we hypothesize that other subsurface anions such as S or Cl might favorably modify the energetics by changing the charges and character of the surface atoms and/or replacing some Cu with Ag, Au, or Ni with different redox properties.

Activation of the inert linear $l$-CO₂ molecule requires enforcing a bent $b$-CO₂ configuration (30) with great chemical stabilization, but pristine Cu(111) and corresponding derivatives with O_sub and/or O_ads do not deliver sufficient stabilization, as shown in our calculations. Thus, forcing CO₂ to have the necessary angle (120°∼140°) and appropriate distance (∼2 Å) to the pristine Cu surface, we find no stable local minimum; all of the initial bent CO₂ structures relax into the stable $l$-CO₂ physisorption state.

However, the presence of modest amounts of O_sub generates a mixture of surface Cu⁺ and Cu⁰ atoms that combines with H₂O_ad to stabilize the $b$-CO₂ structure reported in Fig. 4B. We conclude that this configuration of surface atoms and H₂O is responsible for stabilizing $b$-CO₂ and opening up the possibility of forming formate, formic acid, etc. This elucidates the first reduction step of CO₂.

This combination of APXPS experiments and DFT calculations enabled us to obtain a detailed understanding of the initial steps of CO₂ activation by H₂O on a Cu surface. We find that a modest level of O_sub between the top two Cu layers is essential for stabilizing physisorbed $l$-CO₂.

This unexpected finding may explain a general observation empirically derived in the literature from the catalytic performance of Cu oxides for CO₂RR: It is known that Cu catalysts previously treated to generate surface oxides generally show improved activity compared with the pristine metallic surface (3, 6, 31). From our experimental results and theoretical predictions, we conclude that the topmost layer needs to expose metallic centers, because CO₂ can efficiently chemisorb only on such centers (Fig. 4B), to form the activated molecular substrate for subsequent reduction to formate and other products. However, we find that the presence of a subsurface oxide structure is also needed to promote H₂O chemisorption onto a Cu⁺ center. This enables the electronic communication between adsorbed CO₂ and H₂O, favoring the transition from a linearly physisorbed $l$-CO₂ to a bent chemisorbed $b$-CO₂. From Fig. 4B, reactions to form formate and formic acid are possible but not favored under our conditions.

These results provide the insight that subsurface oxide plays a critical role in the initial steps for activating CO₂, providing a foundation for the rational development of unique active electrocatalysts.

Beamline 9.3.2 and APXPS Measurements.

Beamline (BL) 9.3.2 at the Advanced Light Source (ALS, Lawrence Berkeley National Laboratory) is equipped with a bending magnet and two grating monochromators [100 lines (l)/mm and 600 l/mm] having a total energy range between 200 eV and 800 eV (soft X-ray range) (21). The analyzer (Scienta R4000 HiPP) (21) pass energy was set to 100 eV, using a step of 100 meV and a dwell time of 200 ms. Under these conditions, the total resolution (source and analyzer) was equal to about 220 meV at r.t., for photon energies ranging between 280 eV and 660 eV (the maximum flux, around 4.5 $\times$ 10¹⁰ photons$\cdot$ s⁻¹, is reached at an energy of 490 eV). The incidence angle between the incoming photons and the sample surface was equal to 15°. The PES measurements were taken in normal emission (NE), at a pressure in the experimental (high pressure) chamber ranging from UHV (10⁻⁹ torr) to 0.7 Torr (in backfilling configuration), whereas the detection stage in the analyzer was under UHV conditions (∼10⁻⁹–10⁻⁷ torr). The calibration of the BE scale was carried out using the Au 4f photoelectron peak as reference (4$f_{7/2}$ BE = 84.0 eV), from a clean gold polycrystalline surface. The calibration of the photon energy was performed acquiring the Au 4$f_{7/2}$ core level at the desired photon energy and its corresponding II harmonic reflection. The BL, maintained under UHV to avoid contamination of optical elements, is separated from the high-pressure chamber by a 2-mm $\times$ 2-mm $\times$ 100-nm thick Si₃N₄ window placed at a distance of about 3.0 cm from the sample.

All of the fits reported in this work have been carried out using a Doniach–Šunjić shape for the Au 4f photoelectron peak, whereas a symmetrical Voigt function (G/L ratio ranging from 85/15 to 75/25) was used to fit C 1s and O 1s photoelectron peaks (after Shirley background subtraction). During the fitting procedure, the Shirley background was optimized as well together with the spectral components, increasing in this manner the precision and reliability of the fitting procedure (32–34). The ${\chi }^2$ minimization was ensured by the use of a nonlinear least-squares routine, with increased stability over simplex minimization (33).

The Cu L₃M_4,5M_4,5 Auger transition was acquired at a photon energy ($h\nu$) of 1,150 eV, using the second harmonic reflection of the 575-eV primary photon energy.

Simulations of the Cu L₃M_4,5M_4,5 Auger Spectra.

The main Cu Auger transition (L₃M_4,5M_4,5) arises from a single L₃ (2$p_{3/2}$) core-hole decay via the Auger process involving two M_4,5 (3$d$) electrons resulting in a final 3$d^8$ configuration. This is expected to lead to a final state term splitting from L-S coupling, namely ³F, ¹D, ³P, ¹G, and ¹S corresponding to two $d$ holes (35, 36) that constitute the primary excitation spectrum (F(E)) (the final state terms labeled in Fig. S6A with * are indicative of Auger satellite transitions). The final state labeled “sum” denotes the combination of five different final states (⁴P, ²G, ²P, ²H, and ²D) that cannot be separated into individual spectral components (36). The Auger transition spectrum can be seen then as the addition of the F(E) and the contribution from electrons that have undergone an increasing number of inelastic scattering (energy loss) events (36), which constitutes the background signal and has a crucial importance for the quantification of the different final states present in the experimental mixed spectrum. The Cu Auger L₃M_4,5M_4,5 spectrum was simulated, for all of the investigated systems, by convolving the final state terms with the multiple inelastic scattering background determined via a Drude–Lindhard continuum medium approach, as reported in previous works by Pauly et al. (36), Tougaard and Yubero (37), Simonsen et al. (38), and Yubero and Tougaard (39). These simulations were then used to fit the experimental signals to obtain the final state term population from the experimental data.

As can be seen from Fig. S6A, Cu⁰ and Cu⁺ are characterized by similar final state composition, which leads to the observed limited chemical shift in the core-level spectroscopy. Passing from metallic Cu to Cu_x=1.5−2.5O, a downward kinetic energy shift can be observed for the poorly shielded ³F and ¹D final state terms, as results from the loss of repulsive interaction between the emitted Auger electron and the 4$s$ electron at the Fermi level, as a consequence of the partial ionization of the Cu 4s band passing from metallic Cu to Cu_x=1.5−2.5O. In addition, an important increase of the satellite ⁴F* term and decrease of the ¹G term can be observed, as a function of the surface oxidation. The ratio between these two final state terms can be used to estimate the stoichiometry of the system. Fig. S6A reports the Auger L₃M_4,5M_4,5 spectra for metallic Cu, Cu_x=2.5O, and Cu_x=1.5O, as references, acquired in UHV conditions (∼10⁻⁹ torr).

We observe that despite the adsorption of CO₂ and CO₂ plus H₂O with partial oxidation of the Cu surface, the Cu surface remains metallic (see Fig. S6A, comparison between the metallic Cu reference and spectra A and B). On the other hand, the same coadsorption on the nominal Cu_x=2.5O and Cu_x=1.5O structures leads to slightly different surface stoichiometry under APXPS conditions (namely to ${\text{Cu}}_{x=2.8}$O and Cu_x=1.7O, respectively). With the current signal-to-noise ratio, a direct attribution of these observations to a specific surface interaction between CO₂, H₂O, and the investigated oxidic structures is not yet possible.

Sample Preparation.

The experiments were performed starting from a Cu polycrystalline surface that was in situ oriented to the (111) plane by repeated Ar sputtering (2 keV, 45 min) and annealing cycles in hydrogen (0.15 Torr) at 1,100 K (for 60 min) (17). The sample preparation as well as the LEED investigation was performed in the preparation chamber connected to the high-pressure chamber available at BL 9.3.2. The sample was then transferred to the high-pressure chamber, using a transfer arm without breaking the UHV conditions (21).

The preparation of different Cu suboxide surfaces (performed in the high-pressure chamber by BL 9.3.2) was adapted from the results obtained by Schedel-Niedrig et al. (18). A first oxidation cycle consisted of a Cu(111) treatment at 625 K for 30 min with 0.3 Torr of molecular oxygen. The sample was then slowly (∼10 K$\cdot$min⁻¹) cooled down in the same pressure of oxygen to r. t. (∼298 K). This procedure leads to the formal stoichiometry [determined via the Cu L₃M_4,5M_4,5 Auger analysis (36)] of Cu_x=2.5O. A second cycle performed as described above leads instead to a further surface oxidation, with a formal stoichiometry of Cu_x=1.5O.

Computational Details of DFT Calculations.

The M06L flavor of DFT (16) was chosen for this study because we found that it describes well both the Cu metal and CO₂ molecular crystal, even better than the dispersion-corrected PBE-D3 (29) (see Tables S1 and S2 for details). The calculations were performed with the CRYSTAL14 package (40), which uses local atomic Gaussian-type basis sets. We used all-electron 6-311G(d) basis sets of triple-$\zeta$ quality for H, C, and O, but we added one extra sp shell for the O basis set and reoptimized it. For Cu, we used the SBKJC relativistic effective core potentials (41) and modified associated basis sets of triple-$\zeta$ quality. We used an extra-large grid (consisting of 75 radial points and 974 angular points) for accurate integration. The reciprocal space was sampled by a $\mathrm{\Gamma }$-centered Monkhorst–Pack scheme with 3 $\times$ 3 $\times$ 1 for all slab calculations.

Discussion.

The difficulty in discriminating between Cu⁰ and Cu⁺ using Cu core levels has been well established and is clearly evidenced from Fig. S5A, reporting the Cu 3$p$ photoelectron spectra. To overcome this limitation, the various Cu surfaces were characterized by means of the Cu Auger L₃M_4,5M_4,5 transition (using a photon energy of 1,150 eV, Fig. S6A) and the VB (acquired under resonance conditions with the O K edge, at a photon energy of 537 eV, Fig. S5B). Upon passing from metallic Cu to Cu_x=1.5,25O, we observed a downward kinetic energy shift for the poorly shielded ³F and ¹D final state terms (Fig. S6A), due to decreased repulsive interaction between the emitted Auger electron and the 4$s$ electron at the Fermi level, as a consequence of the partial ionization of the Cu 4$s$ band passing from metallic Cu to Cu_x=1.5,25O (35, 36). In addition, we observe an important increase of the satellite ⁴F* term and decrease of the ¹G term as a function of the surface oxidation.

From the VB spectra (as well as the difference VB spectra) reported in Fig. S5B, no unambiguous changes can be detected in the electronic structure of the metallic or oxide surfaces upon CO₂ or CO₂-H₂O coadsorption. With respect to the suboxide structures (Cu_x=1.5,25O), the VB spectra exhibit the typical intensity increase of the <O 2p><Cu 3d> hybridization bands below the Fermi edge (Fig. S5B), as well documented by the difference VB spectra obtained taking the metallic Cu VB as reference (18). Furthermore, as reported in Fig. S6B, the region below the Fermi level shows the typical increase of the density of states (DOS) at around 0.7 eV expected for Cu⁺-based suboxidic structures (18) [also confirmed by the increase of the positive slope of the DOS (dDOE/dBE)], obtained by interpolating a straight line where the DOS is centered at 0.7 eV within a range of BEs given by taking the double of the spectral resolution (42, 43). The coupling of Auger and VB analyses provides then the necessary information to establish in situ copper metallic to subsurface suboxide transitions that occur while changing the experimental conditions.