Engineering d-orbital of copper single-atom sites toward industrial-level electrocatalytic methanation

Abstract

Copper (Cu)-based single-atom catalysts (SACs) enable electrocatalytic CO₂ reduction into methane (CH₄) fuel for thermal power plant decarbonization, yet conventional Cu SACs face industrial deployment barriers like instability and sluggish kinetics caused by d − p orbital coupling. Herein, we develop a Cu–Ti₁O₃ catalyst with localized Cu single-atom sites by oxygen vacancy (O_v)-involved orbital engineering, achieving industrial-level CH₄ production. Theoretical and in-situ studies reveal the intensification of the d − d coupling at Cu sites triggered by [Cu−O_v − Ti] motifs, which enhances d-π* polar interactions upon *CO₂ and accelerates C − O bond cleavage in *OCH₃ intermediate. As a result, Cu–Ti₁O₃ achieves a competitive performance, i.e., the highest Faradaic efficiency of 76% and a peak partial current density of 670 mA cm⁻² toward CH₄ (corresponding turnover frequency = 24,930 h⁻¹), ~3.5-fold promotion over conventional Cu SACs. Furthermore, it demonstrates high durability (>1,230 hours) at an industrial-level current density, exceeding the longevity of conventional Cu SACs by over 20 times. Our findings highlight the prospect of d-orbital engineering in enabling industrial-level electrocatalytic methanation, offering promising implications for decarbonizing traditional power plants.

Introduction

Carbon emissions from traditional thermal power plants significantly hinder carbon neutrality efforts, e.g., overseas coal-fired power plants currently emit 0.53 gigatons of CO₂ per year¹. To address this, abundant decarbonization technologies are being proposed, among which the electrocatalytic CO₂ reduction reaction (CO₂RR) emerges as a critical pathway for CO₂ valorization and sustainable fuel production^2,3. Specifically, electro-catalytic methanation (i.e., CO₂-to-CH₄ conversion) using flue gas provides a promising decarbonization strategy by coupling CO₂ fixation with high-energy CH₄ (55.5 MJ kg⁻¹) production for thermal power plants^4,5. Regarding catalyst candidates, copper (Cu)-based catalysts facilitate pathways for deeply reduced products, including CH₄ (>2e⁻ transfer)^6,7, but face challenges of sluggish kinetics (| J_total| <500 mA cm⁻²) and poor selectivity (<50%).

From a mechanistic perspective^8,9, spatially coupled Cu–Cu sites promote C–C coupling over the CH₄ pathway (e.g., 0.53 eV higher barrier for CH₄ on bulk Cu¹⁰). Therefore, enhancing CH₄ selectivity requires spatially isolated Cu single-atom catalysts (SACs). Current synthesis strategies for Cu SACs rely on embedding Cu into non-metallic matrices like carbon, C₃N₄, and so on (Supplementary Fig. 1a)⁶. For instance, N-coordinated Cu SACs have been confirmed to own a favored C₁ selectivity (e.g., CO¹¹, HCOOH¹², and CH₄^13,14). Nevertheless, the modern hard-soft acid-base (HSAB) theory^15,16 illustrates that the inherently delocalized p-orbitals¹⁷ may cause a d-electronic delocalization via d–p orbital coupling, thus weakening hard-basic intermediate adsorption like *OCH₃ and facilitating hydrogen evolution side reaction¹⁸ at Cu sites (Details in Supplementary Note 1)¹⁵. Moreover, Cu−X (X=C, N, O) coordinates are thermodynamically unstable¹⁹, causing severe reconstruction such as agglomeration of Cu atoms. This may facilitate C−C coupling²⁰ and cause performance degradation under industrial-level current density (> 500 mA cm⁻²)²¹.

To these issues, constructing localized Cu single-atom sites (i.e., harder acid¹⁵) by introducing matrices with localized d/f valence orbitals may be essential²², and proposed solutions include constructing Cu₁M single-atomic alloys (M = f/d-block elements) as well as Cu₁/MO_x SACs (Supplementary Fig. 1b, c). Although Cu₁M systems feature localized d-orbitals, free-atom-like Cu sites (i.e., resembling d−p coupled molecular metal complexes) will cause an unstable condition^23,24, thus favoring a tight *CO binding and predominant CO selectivity²⁵. For Cu₁/MO_x SACs, the intrinsic charge polarization of M−O bonds²⁶ and spin-orbit interaction triggered insurmountable energy barriers for direct d−d/f coupling^27,28, finally causing dominative [Cu−O−M] motifs that facilitate indirect d−p−d/f coupling during catalyst synthesis^29,30. In this indirect d−p−d/f coupling structure, the continuous electron-polaron accumulation at d-orbitals during CO₂RR may destroy the Cu−O coordinates³¹, significantly threatening the robustness (Fig. 1a). For instance, Zhou et al. synthesized a Cu-Ce-O_x catalyst (i.e., d−p−f coupling) with isolated Cu sites³², achieving an efficient CO₂-to-CH₄ conversion (FE of CH₄ (FE_CH₄) ≈ 65%), but durability remains poor (<10 h). Unlike typical electron-polaron trapping at metallic sites, electron in anatase titanium oxide (TiO₂) prefers a free-carrier state, and can only be trapped at oxygen vacancy (O_v)^31,33, providing a vital opportunity for O_v-induced metal-metal bond formation (strong d−d coupling, Fig. 1b), thereby constructing localized Cu single-atom sites (l-Cu sites) and stabilizing the active sites^34,35. Inspired by these, employing O_v-rich TiO₂ as matrices may be a promising strategy to realize strong d−d coupling and facilitate selective and stable methanation.

Fig. 1: Schematic illustration of orbital engineering on Cu single-atom catalysts.

a Conventional Cu₁/MO_x SACs display a delocalized d−p−d orbital coupling involving lattice oxygen, causing two major challenges: sluggish kinetics and instability. b Electron-polaron accumulation at O_v sites induces strong d−d orbital coupling in a valent-variable system like Cu₁/TiO_x. c Synthesis of Cu–Ti₁O₃ via lattice O removal in Cu-embedded TiO₂, obtaining a localized [Cu−O_v − Ti] motif with strong d−d orbital coupling. d Cu–Ti₁O₃ achieves enhanced d-π* polar interactions on *CO₂ and accelerated C−O bond cleavage of *OCH₃, thereby leading to highly efficient and stable CH₄ production.

Herein, we have developed an oxygen-vacancy-involved orbital engineering method to construct [Cu−O_v−Ti] motifs through O removal in Cu-embedded TiO₂ SAC (Fig. 1c). As-synthesized Cu-based SAC (marked as Cu–Ti₁O₃) achieves an efficient CO₂-to-CH₄ conversion in flow cells, including an excellent FECH₄ of 76% at −0.88 V (potentials converting into the reversible hydrogen electrode (RHE), see “Methods”), and a record-high turnover frequency of CH₄ to date (24,390 h⁻¹ at −1.0 V). Furthermore, Cu–Ti₁O₃ also exhibits exceptional stability of >1230 h at an industrial-level current density (600 mA cm⁻²) in a zero-gap electrolyzer, over 12 times longer than a conventional Cu₁/TiO₂ catalyst with major [Cu−O−Ti] motifs (marked as Cu−O₄(Ti)). Synergistic experimental/theoretical analyses reveal that Oᵥ-intensified d−d orbital coupling between Cu and Ti atoms enhances electronic localization of Cu^δ+ sites (i.e., harder acidity than pristine Cu^δ+ sites in Cu-O₄(Ti)), strengthening d-π* polar interactions with adsorbed *CO₂ (Fig. 1d) and stabilizing the Cu−O binding with hard-basic *OCH₃. Simultaneously, in-situ time-field electrochemical electron magnetic resonance (E-EMR) reveals a new mechanism involving pseudo-lattice oxygen formation through *OCH₃ insertion into O_v. This reversible pseudo-lattice oxygen formation route supplements a localized ultra-stable adsorption for *OCH₃ surpassing the typical physico/chemical adsorption at conventional Cu single-atom sites (c-Cu sites) with majorly Cu−O coordination (i.e., bond dissociation enthalpy (BDE) of Cu−O = 263.1 vs. 41.9 kJ mol⁻¹). Therefore, an accelerated C−O bond cleavage (~16.8% lower BDE) and rapid methanation generation on Cu–Ti₁O₃ are achieved compared with Cu-O₄(Ti) (Supplementary Fig. 2). Overall, this work establishes an orbital engineering method for constructing localized Cu single-atom sites and provides a universal strategy to design durable SACs toward industrial-level CO₂-to-fuel conversion.

Results

Synthesis and characterization

Density functional theory (DFT) calculations were first employed to evaluate the influence of d−d orbital coupling at Cu single-atom sites. The local density of states (LDOS) for Cu atom at [Cu−O_v−Ti] motifs exhibits intensified peaks relative to the conventional [Cu−O−Ti] structures, confirming localized electronic states of l-Cu sites (Fig. 2a, Supplementary Fig. 3). Integrated projected density of states (IPDOS) analysis identifies the d_x²-_y² orbital as the frontier orbital of the Cu sites (Supplementary Fig. 4). The higher electron density within the d_x²-_y² orbital (e.g., IPDOS = 0.980 for up-spin states, Fig. 2b and Supplementary Table 1) of l-Cu site likely facilitates strong d-π* polar-interactions on adsorbed *CO₂, corroborating accelerated CO₂ activation^36,37, and leading to an enhanced CO₂-to-CH₄ conversion trend as reported by Huang’s group³⁷. Furthermore, crystal orbital Hamilton population (COHP) analyses reveal a significantly larger negative integrated COHP (−ICOHP) value for the Cu−Ti bond on l-Cu sites (2.422 a.u.) versus c-Cu sites (0.002 a.u., Fig. 2c), indicating robust d−d coupling induced by oxygen vacancies. BDE assessments of *OCH₃ confirm tighter Cu–O bonding post orbital engineering (BDE_Cu–O = 263.1 kJ mol⁻¹ on l-Cu sites, Fig. 2d) compared with unstable adsorption on pristine c-Cu sites (BDE_Cu–O = 41.9 kJ mol⁻¹, Supplementary Table 2). Meanwhile, a 16.8% reduction in BDE_C−O at l-Cu sites suggests potential acceleration of C−O bond cleavage during CO₂RR (inset slab model in Fig. 2d). Hence, d−d orbital coupling at l-Cu sites confers a kinetically enhanced methanation pathway.

Fig. 2: Assessments on the electronic states of two distinct Cu single-atom site configurations.

a LDOS of Cu atom at l-Cu and c-Cu sites. b IPDOS of Cu−O−Ti and l-Cu sites around the Fermi level. c COHP analysis for d−d orbital coupling at c-Cu and l-Cu sites. d BDE evaluations illustrate that strong d−d coupling promotes C−O bond cleavage of *OCH₃ on l-Cu sites compared with c-Cu sites. Source data for Fig. 2 are provided as a Source Data file.

Guided by computational insights, we designed Cu–Ti₁O₃ SAC with partial Cu–Ti bonding toward CO₂RR. In detail, Cu single-atom precursor was first fabricated through a hydrothermal method (Details in “Methods”), where Cu substitutes Ti in [TiO₆] octahedra and forms Cu–O–Ti motifs³⁸. Given that lattice oxygen removal facilitates adjacent metallic bonding^39,40, Cu single-atom precursor was then calcined in 10% H₂/Ar for 3 h (surface Cu: 2.67 wt%, Supplementary Table 3) to synthesize Cu–Ti₁O₃. As a control group, Cu–O₄(Ti) catalyst was prepared by an air-calcination of Cu single-atom precursors. X-ray diffraction (XRD) patterns confirmed the major anatase phase (PDF#71-1167, Supplementary Fig. 5a) in Cu–Ti₁O₃, while Cu-related peaks were absent, indicating the lattice incorporation of Cu atoms. Compared to pristine TiO₂, the (101) diffraction peak of Cu–Ti₁O₃ shifted to a smaller angle, which might result from the lattice expansion³⁸. Meanwhile, Cu–Ti₁O₃ exhibited a higher 2θ angle for the (101) plane than Cu–O₄(Ti), suggesting the lattice shrinkage caused by O_v accumulation (Supplementary Fig. 5b)⁴¹.

The morphologies and atomic structures of Cu–Ti₁O₃ and Cu–O₄(Ti) were then examined by transmission electron microscopy (TEM) and aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM). The TEM images illustrated a retained TiO₂ particle size of ~20 nm after Cu embedding (Supplementary Fig. 6). High-resolution transmission electron microscopy (HRTEM) of the Cu–Ti₁O₃ revealed an ~0.194 nm interplanar spacing for the (200) plane (d₂₀₀, Supplementary Fig. 6), indicating a lattice expansion compared with TiO₂ (d₂₀₀ = 0.189 nm). The energy-dispersive spectroscopy (EDS) mapping confirmed the uniform distribution of Cu, O, and Ti across the nanoparticles (Supplementary Fig. 7). In addition, HAADF-STEM revealed the dumbbell-shaped periodic arrays consisting of adjacent Ti atom pairs along the [100] direction in Cu–O₄(Ti) and Cu–Ti₁O₃ (Fig. 3a, b). Bright spots that refer to Cu atoms were exclusively located at the original Ti sites, indicating the on-site substitution and formation of Cu single-atom sites in Cu–Ti₁O₃ and Cu–O₄(Ti). Moreover, in the [100]-directed images (Fig. 3c), the distance between one bright (Cu) and one dark spot (Ti) in Cu–Ti₁O₃ was about 2.83 Å, shorter than that in Cu–O₄(Ti) (2.97 Å), suggesting an atom approaching and a possible direct Cu−Ti bond formation triggered by O_v.

Fig. 3: Structural characterization of Cu single-atom catalysts.

Aberration-corrected HAADF-STEM images of a Cu–Ti₁O₃ and b Cu–O₄(Ti) along the [100] direction. c Left panel: Front and top views of Cu–Ti₁O₃, where the purple, blue, and red spheres represent the Cu, Ti, and O atoms, respectively. Right panel: the measured distance between the adjacent Cu and Ti atoms on Cu–Ti₁O₃ and Cu–O₄(Ti). d Normalized XANES spectra at Cu K-edge. e Fourier-transformed k²-weighted EXAFS spectra at Cu K-edge for Cu foil, Cu₂O, CuO, Cu–Ti₁O₃ and Cu–O₄(Ti) samples. Source data for Fig. 3c–e are provided as a Source Data file.

Interfacial electronic states were then assessed by X-ray photoelectron spectroscopy (XPS). The high-resolution spectra of O 1s showed a higher O_v peak at 531.2 eV⁴² for Cu–Ti₁O₃ than Cu–O₄(Ti) (ca. 18.5% vs. 10.8%, Supplementary Fig. 8). As for Cu 2p spectra (Supplementary Fig. 9a), Cu–O₄(Ti) catalyst showed two main deconvoluted peaks at around 951.7 eV and 932.0 eV, respectively, attributed to Cu⁰/Cu⁺ species¹⁵. These two peaks exhibited a ~0.40 eV red-shift in Cu–Ti₁O₃, demonstrating an intensified electron density at interfacial Cu sites⁴³. Cu LMM Auger spectra unveiled that both catalysts exhibited two fitted peaks at 917.7 and 916.7 eV (Supplementary Fig. 10), ascribed to the Cu²⁺ and Cu⁺ species, respectively⁴⁴. In detail, Cu–Ti₁O₃ possessed a higher ratio of Cu⁺ with 86.1% compared with that of Cu–O₄(Ti) with 67.9%, signifying the low valence state of surface Cu. In contrast, Ti 2p_3/2 and 2p_1/2 peaks showed a slight blue-shift (~0.13 eV) post orbital engineering, suggesting an electron outflow at interfacial Ti sites (Supplementary Fig. 9b)⁴⁵. Taken together, a strong d−d coupling may appear in Cu–Ti₁O₃, which modulates the local electron density distribution and results in l-Cu sites⁴⁶. Moreover, the structures of these two catalysts were also studied by Raman spectroscopy (Supplementary Fig. 11). The bands at 637, 517, 395, 200, and 146 cm⁻¹ were observed for all of Cu SACs and TiO₂, which could be ascribed to the vibrational mode of E_g(3), A_1g & B_1g(2), B_1g(1), E_g(2), and E_g(1), respectively⁴⁵. The E_g(1) band in Cu–Ti₁O₃ showed the highest wavenumber among all samples, indicating the highest O_v concentration⁴⁵.

To further investigate the electronic states and coordination environment at Cu sites, X-ray absorption spectroscopy (XAS) was conducted at the Cu K-edge. X-ray absorption near-edge fine structure exhibited that the adsorption edges and corresponding first derivative peaks of two Cu SACs lay between those of Cu₂O and CuO references (Fig. 3d and Supplementary Fig. 12), suggesting the existence of major Cu^δ+ species in Cu SACs. The peak position of Cu–Ti₁O₃ catalyst (8982.3 eV) was lower than Cu–O₄(Ti) catalyst (8982.7 eV, Supplementary Fig. 12), further indicating localized electronic states of l-Cu sites when O_v accumulation, consistent with the results of XPS. As for extended X-ray absorption fine structure (EXAFS) spectra, the Fourier-transformed EXAFS exhibited an intense peak at R ≈ 1.92 Å for Cu–O₄(Ti) (Fig. 3e and Supplementary Fig. 13; Supplementary Table 4), which may be ascribed to Cu−O bonds¹⁵. While the wavelet-transformation analyses revealed the absence of Cu−Cu bonds in Cu–O₄(Ti), confirming the formation of Cu−O−Ti motifs where an average Cu−Ti interatomic distance of 2.97 Å was observed (Supplementary Fig. 14)¹⁶. While in Cu–Ti₁O₃, a peak at R = 2.83 Å appeared, and wavelet-transformed EXAFS revealed a high-energy region at ~6.53 Å⁻¹ (Supplementary Fig. 14), which together confirms the formation of Cu−Ti bonds⁴⁶. The generation of Cu−Ti bonds, originating from O_v accumulation, implies a strong d−d orbital coupling in Cu–Ti₁O₃.

Electrochemical measurements

The catalytic performance of CO₂RR was then tested in a flow cell with 1 M KOH electrolyte (Details in “Methods”). The linear sweep voltammetry curves with the CO₂ flow revealed a higher reaction current for Cu–Ti₁O₃ than Cu–O₄(Ti) (Supplementary Fig. 15). in situ electrochemical impedance spectroscopies (EIS) at –0.88 V were executed, which showed that Cu–Ti₁O₃ obtained lower charge-transfer resistance compared with Cu–O₄(Ti) (Supplementary Fig. 16 and Table 5). Corresponding advanced distributed relaxation times analyses exhibited two characteristic peaks at –0.88 V over two catalysts, i.e., P2 and P3, which refer to the charge-transfer process of CO₂RR and HER, respectively⁴⁷. The interfacial charge transfer impedances of CO₂RR on Cu–Ti₁O₃ were calculated as 48.62 Ω, lower than that over Cu–O₄(Ti) (123.7 Ω), while the trend was opposite toward HER (Supplementary Table 6), indicating a more favorable CO₂RR pathway on Cu–Ti₁O₃.

In electrolysis experiments, Cu–Ti₁O₃ achieved a peak FE_CH₄ of 76% at −0.88 V, about 2.2-fold higher than Cu–O₄(Ti) (Fig. 4a and Supplementary Fig. 17). The FE ratios of CH₄ to 2e⁻-product (CO + HCOOH) were calculated as 14.2 on Cu–Ti₁O₃, much larger than that of Cu–O₄(Ti) (Fig. 4b and Supplementary Fig. 18), indicating a favorable multi-electron (>2e⁻) pathway on Cu–Ti₁O₃. Furthermore, partial current density of CH₄ (J_CH₄) on the Cu–Ti₁O₃ reached a maximal value of 670 mA cm⁻² at −1.00 V (i.e., turnover frequency of CH₄ (TOF_CH₄) = 24,930 h⁻¹, Fig. 4c), surpassing Cu–O₄(Ti) (J_CH₄ = 205 mA cm⁻², TOF_CH₄ = 7100 h⁻¹). The corresponding J_CH₄ normalized by electrochemically active surface area (ECSA) was also calculated to eliminate the influence of the reaction surface, and Cu–Ti₁O₃ delivered a 2.5 times greater current density than Cu–O₄(Ti) (Supplementary Fig. 19). Furthermore, the energy efficiency (EE) of Cu–Ti₁O₃ attained a peak value of 38.7% at −0.88 V, exceeding that on Cu–O₄(Ti) (19.2%, Fig. 4d). By tuning the CO₂ flow rate to 50 sccm, the Cu–Ti₁O₃ achieved the largest FE_CH₄ of 79% at around −0.88 V, while the peak FE_CH₄ of the Cu–O₄(Ti) was 38% (Supplementary Fig. 20). The pristine TiO₂ showed that H₂ was dominant product (FE > 95%) with no detection of carbonous product, confirming that Cu sites on the support facilitate the CO₂RR (Supplementary Fig. 21). For mechanistic insight, O_v-free Cu₁/MgO(Al) without other d-block elements was then synthesized and tested (Details in Supplementary Note 2, XRD/XPS/TEM/EDS in Supplementary Figs. 22 and 23), exhibiting a FE_CH₄ of less than 30% (Supplementary Fig. 24), further suggesting the essential role of O_v-intensified d−d orbital coupling. We then study the influence of impurities such as sulfur dioxide (SO₂) and oxygen (O₂) on the CO₂RR performance of Cu–Ti₁O₃ at −0.88 V. Under pure CO₂ flow, the Cu–Ti₁O₃ realized a high FE_CH₄ of ~76%. However, upon introducing 1% SO₂ (~10⁴ ppm) at t = 0.5 h, the total FE, along with FE_CH₄, decreased markedly (Supplementary Fig. 25a), which can be attributed to the preferential reduction of SO₂⁴⁸. After 1 h of electrolysis (including a feed of 1% SO₂ for 30 min), the FE_CH₄ recovered to >66%, suggesting that the influence of SO₂ on the catalyst structure is largely reversible. A similar reversible trend was observed when feeding 10% O₂ (~10⁵ ppm, Supplementary Fig. 25b).

Fig. 4: Electrocatalytic performance over different catalysts.

a Current densities (top) and FEs (bottom) of CO₂RR product and H₂ on Cu–Ti₁O₃ catalyst at different applied potentials in 1 M KOH (with 90% iR-correction). b FEs and corresponding ratios of CH₄: (HCOOH + CO) on Cu–Ti₁O₃ and Cu–O₄(Ti) at around −0.88 V. c TOFs of CH₄ on Cu–Ti₁O₃ and Cu–O₄(Ti) catalysts at different applied potentials (with 90% iR-correction). d Energy efficiency (EE) of CH₄ on Cu–Ti₁O₃ and Cu–O₄(Ti) catalysts (with 90% iR-correction). Cell resistance and pH of 1 M KOH electrolyte were measured as 0.98 ± 0.06 Ω and 14.00 ± 0.02, respectively, with an electrode surface area of 1 cm². e Stability tests in a 5-cm² zero-gap electrolyzer with 0.1 M KHCO₃. f Comparison of this work with state-of-the-art CO₂RR electrocatalysts, including J_CH₄, CH₄ selectivity, EE_CH₄, stability and TOF_CH₄. Data are presented as mean ± s.d. (n  =  3), with the error bars representing the s.d. Source data for Fig. 4 are provided as a Source Data file.

Both two catalysts were further integrated into a 5-cm² zero-gap electrolyzer to evaluate the potential of our strategies for practical applications towards the CH₄ formation (Supplementary Fig. 26). The Cu–Ti₁O₃ exhibited a higher FE_CH₄ of 75% than that of Cu–O₄(Ti) (37%) at a constant current density of –0.6 A cm^–2 (Supplementary Fig. 27). Catalyst stability was then evaluated at –0.6 A cm⁻². Cu–Ti₁O₃ maintained continuous operation for >1230 h at an industrial current density of –0.6 A cm⁻², exhibiting a cell voltage of −4.05 ± 0.16 V and an average FE_CH₄ of ~71% (Fig. 4e). Post-test XRD, HRTEM, aberration-corrected HAADF-STEM, and XPS confirmed structural integrity via persistent d−d coupling (Supplementary Figs. 28–32). In contrast, Cu–O₄(Ti) showed rapid degradation: FE_CH₄ dropped to 23% within 64 h under identical conditions, with observed Cu nanoparticle formation that suggests Cu−O bond cleavage and metallic atomic aggregation. In addition, the Cu content of Cu–Ti₁O₃ after reaction was almost unchanged, yielding less Cu dissolution (5 ppb vs. 65 ppb for Cu–O₄(Ti), Supplementary Tables 7 and 8), implying screened Cu corrosion. From the perspective of reaction activity and stability, Cu–Ti₁O₃ surpassed most of the advanced Cu SACs in CH₄ production metrics (Fig. 4f and Supplementary Table 9). Moreover, techno-economic assessments based on a power plant model (Details in Supplementary Note 3 and Supplementary Fig. 33), utilizing established literature parameters^21,49 and the durability data presented herein (Supplementary Table 10), demonstrate that a profitable pathway is feasible for this CO₂-to-CH₄ conversion system under a given carbon price (Supplementary Fig. 34).

Mechanism investigations

To gain insight into the role of orbital engineering in modulating the reaction mechanism, a series of in situ electrochemical spectroscopies was employed. In situ time-field E-EMR tracked O_v dynamics and intermediates during CO₂RR. For Cu–Ti₁O₃, the time-field scans suggested that a signal related to O_v (centered at g-tensor of 1.997) maintained well after 10-min CO₂ injection, excluding the influence of CO₂ adsorption on the O_v concentrations (Fig. 5a and Supplementary Fig. 35). However, O_v gradually faded out under applied potential, and recovered when the potential was revoked (Fig. 5a). On the contrary, Cu–O₄(Ti) catalyst exhibited a maintained O_v intensity throughout the catalytic process (Supplementary Fig. 36). Further quantitative spin counting (Fig. 5b) showed an initial O_v concentration of 1.08 × 10¹⁷ spins in Cu–Ti₁O₃, approximately 3 times higher than that in Cu–O₄(Ti) (3.52 × 10¹⁶ spins). Under applied potential, O_v concentration of Cu–Ti₁O₃ dropped to 3.78 × 10¹⁶ spins, while recovering to 1.08 × 10¹⁷ spins at open-circuit potential (OCP), suggesting an intermediate-mediated (e.g., *OCH₃) reversible O_v occupancy process. We then used heavier deuterium (D) from D₂O solvent as the source of H to investigate the O_v concentration variation during CO₂RR. Under applied potential, O_v concentration decreased to 5.81 × 10¹⁶ spins, higher than that using H₂O as solvent under the same condition (Supplementary Fig. 37). As hydrogenation activity is weakened in D₂O⁵⁰, lower *OCD₃ was produced during CO₂RR, hence causing a lower occupancy rate of O_v. To reveal these short-lived intermediates, the spin traps were spiked under flow employing 5,5-dimethyl-1-pyrroline N-oxide (DMPO)⁵¹ (Supplementary Fig. 38). A typical six-finger peak (g = 2.0055, A_N = 1.35 mT, A_Hβ = 0.74 mT) appeared (Fig. 5c), exhibiting a negative correlation with O_v, which demonstrated the replacement of O_v by alkoxy (i.e., *OCH₃). Subsequently, this signal disappeared during the power outage, and signals associated with O_v gradually reverted to their initial state. The O_v concentration of Cu–Ti₁O₃ was also probed during water electrolysis, which retained well, indicating that H₂O and *OH adsorption have only a minor influence on O_v stability (Supplementary Fig. 39). Signal reversibility during potential interruption confirmed O_v’s dynamic participation in CO₂RR on Cu–Ti₁O₃, revealing a new *OCH₃ insertion mechanism (Supplementary Fig. 40). Notably, as methoxyl radicals (*OCH₃) are typically poorly detectable in bulk water solution, this result is more indicative of an association with the electrochemical interface. By regulating the calcination condition, the O_v concentration could be controlled rationally (Supplementary Fig. 41). EMR spectra showed that the O_v concentration of catalyst increased to higher values with 1.58 × 10¹⁷ spins when calcined in 10% H₂/Ar for 4 h, which could further improve the FE_CH₄ to 78%. However, the TOF_CH₄ per O_v of Cu–Ti₁O₃ showed the maximum value with 1.62 × 10⁻¹³ h⁻¹ spin⁻¹ (Supplementary Fig. 41e), indicating the most efficient CH₄ formation on l-Cu sites.

Fig. 5: In-situ electrochemical spectra measurements.

a In-situ time-field E-EMR spectra in CO₂-saturated 0.1 M KHCO₃ at −0.88 V over Cu–Ti₁O₃ (with 90% iR-correction), referring to in-situ O_v evaluation. b Quantitative spin counts of signals related to O_v versus time over Cu–Ti₁O₃ and Cu–O₄(Ti). c In-situ time-field E-EMR spectra with spin-trap agent (i.e., DMPO) in CO₂-saturated 0.1 M KHCO₃ at −0.88 V over Cu–Ti₁O₃ (with 90% iR-correction), corresponding to the in-situ concentration assessments for both O_v and *OCH₃ intermediate during CO₂RR. d In-situ Raman spectra of Cu–Ti₁O₃ and e Cu–O₄(Ti) during a 10-min CO₂RR in CO₂-saturated 0.1 M KHCO₃ at −0.88 V (with 90% iR-correction). Cell resistance and pH of electrolyte were measured as 0.98 ± 0.06 Ω and 6.83 ± 0.03, respectively, with an electrode surface area of 1 cm². Source data for Fig. 5 are provided as a Source Data file.

The quasi-operando XPS also revealed the dynamic switch of O_v during CO₂RR from a semi-quantitative perspective. For Cu–Ti₁O₃, O_v content decreased to 15.9% after 10-min electrolysis (intermediate-retained state), while recovering upon degassing (Supplementary Fig. 42a). This reversible change indicates pseudo-lattice oxygen formation by adsorbed intermediates. Concurrent Ti 2p binding energy upshift correlated with O_v depletion (Supplementary Fig. 42b), further evidencing intermediate-induced electronic perturbation. In stark contrast, Cu–O₄(Ti) showed negligible O_v/Ti 2p changes (Supplementary Fig. 42c, d), confirming Cu−O−Ti motifs’ minimal reactivity impact. Complementary in-situ Raman (Supplementary Fig. 43) also revealed a reversible downshift of the E_g(1) band (147→143 cm⁻¹) in Cu–Ti₁O₃ during reaction (Fig. 5d), while remaining invariant in Cu–O₄(Ti) (Fig. 5e), consistent with the *OCH₃ insertion mechanism mentioned above. These results elucidate the unique mechanistic pathway governing CH₄ formation on Cu–Ti₁O₃, mediated through d−d orbital coupling at l-Cu sites.

CO₂ temperature-programmed desorption (CO₂-TPD) was then carried out to assess the CO₂ adsorption capacity of the Cu SACs (Supplementary Fig. 44). Cu–Ti₁O₃ displayed a large chemisorption peak at a higher temperature (455 °C), compared with the weak chemisorption peak of Cu–O₄(Ti) at 430 °C, manifesting the intensive CO₂ adsorption ability⁵². In situ attenuated total reflection surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS, Supplementary Fig. 45) was then employed to probe intermediates. On Cu–Ti₁O₃, the 1355 cm⁻¹ peak (i.e., *COOH)¹² at −0.1 V disappeared distinctly at −0.5 V, indicating rapid conversion to *CO (Fig. 6a). Both catalysts exhibited linear-bonded *CO (*CO_L) at ~2075 cm⁻¹ (Fig. 6a, b)^53,54, with Cu–Ti₁O₃ showing stronger intensity over wider potentials, reflecting higher *CO coverages. CO_L signal on Cu–O₄(Ti) red-shifted with increasing applied potential, ascribing to the Stark effect⁵⁵. In contrast, CO_L peak on Cu–Ti₁O₃ intensified and blue-shifted with the potential increasing to –1.1 V. This phenomenon is ascribed to dipole–dipole coupling resulting from the increased *CO coverage at localized Cu^δ+ sites and adjacent Ti atom^55,56. The in-situ ATR-SEIRAS spectra of pristine Au film, which served as the SEIRAS-active template, were also measured during CO₂RR (Supplementary Fig. 46). A peak observed at ~2100 cm⁻¹ was assigned to adsorbed *CO⁵⁷, confirming that the detected *CO_L peak on the Cu–Ti₁O₃ surface indeed stems from Cu–*CO interaction. The in-situ Raman spectra showed a band at around 358 cm⁻¹ (Supplementary Fig. 47), which was attributed to the frustrated rotation of adsorbed *CO on Cu⁵⁸. A distinct *CO signal at low cathodic potential further highlights the higher *CO coverage on Cu–Ti₁O₃ surface. Moreover, C–H bending modes of *OCH₃ appeared⁵³, yet notably red-shifted on Cu–Ti₁O₃ (1380 cm⁻¹) versus Cu–O₄(Ti) (1398 cm⁻¹). This 18 cm⁻¹ shift signifies reduced C–H vibrational frequency, consistent with stabilized *OCH₃ insertion at O_v mentioned above^35,59. Meanwhile, earlier emergence and intensified *OCH₃ signal on Cu–Ti₁O₃ further evidence facilitated *CO hydrogenation toward CH₄.

Fig. 6: Mechanism research of CO₂RR on Cu single-atom catalysts.

In-situ ATR-SEIRAS spectra on a Cu–Ti₁O₃ and b Cu–O₄(Ti) in a potential window from OCP to −1.1 V in CO₂-saturated 0.1 M KHCO₃ (with 90% iR-correction). Cell resistance and pH of electrolyte were measured as 0.98 ± 0.06 Ω and 6.83 ± 0.03, respectively, with an electrode surface area of 1 cm². c The charge density difference analyses of *CO₂ on Cu–Ti₁O₃ and Cu–O₄(Ti). Green isosurfaces indicate negative-charged regions, and the yellow isosurfaces refer to positive-charged regions when CO₂ is adsorbed on catalysts. d Free energy diagram of CO₂-to-CH₄ conversion on Cu–Ti₁O₃ and Cu–O₄(Ti) at 298.15 K. Source data for Fig. 6a, b, d are provided as a Source Data file.

DFT calculations were then carried out to assess the effect of orbital engineering on catalytic performance (Model details in Supplementary Data 1). The CO₂ adsorption features were first evaluated (Supplementary Table 11), and detailed charge density difference (CDD) analyses revealed an enhanced electron donation from Cu sites to *CO₂ on Cu–Ti₁O₃ (Fig. 6c). The highest CDD value on Cu–Ti₁O₃ was calculated as −1.8× 10⁻³ e Bohr⁻³ (63.6% larger than that on Cu–O₄(Ti)), showing a negative charge cloud (blue color in Fig. 6c) surrounded *CO₂, indicating a stronger d-π* polar-interaction. Moreover, the free energy diagrams illustrate how the CO₂RR pathways proceed on Cu single-atom sites (Fig. 6d). The overall free energy change (ΔG) for methanation (i.e., *CO₂→* + H₂O_(l)+CH_4(g)) on both Cu SACs shows energy downhill, indicating a thermodynamically favorable methanation. With the Brønsted−Evans−Polanyi relation, we further assessed the reaction kinetics in CO₂-to-CH₄ conversion. For *CO₂→*COOH conversion, Cu–Ti₁O₃ displayed a 0.52 eV smaller energy rise relative to Cu–O₄(Ti), suggesting a promoted CO₂ activation resulting from stronger d-π* polar-interactions^36,37, which has been mentioned in Fig. 2b. For the vital *OCH₃ intermediate, ΔG for *OCH₂→*OCH₃ step was only 0.01 eV on Cu–Ti₁O₃, ~100% reduction from that on Cu–O₄(Ti), demonstrating the accelerated kinetics induced by the new *OCH₃ insertion mechanism at l-Cu sites. Meanwhile, ΔG for hydrogenation of *OCH₃ was 1.57 eV and 0.97 eV on Cu–O₄(Ti) and Cu–Ti₁O₃, respectively (Fig. 6d), revealing faster *OCH₃-to-*CH₄ converting kinetics at l-Cu sites. Taken together, the DFT calculations indicate a more favorable CO₂-to-CH₄ conversion on Cu–Ti₁O₃, which is attributed to the localization of Cu electronic states and subsequently enhanced d-π* polar-interaction and Cu–O bonding.

Discussion

In summary, we employed oxygen-vacancy-involved orbital engineering to construct localized Cu single-atom sites on O_v-rich TiO₂ toward efficient methanation. Oxygen vacancies intensify d−d orbital coupling between Cu and Ti, heightening the electronic localization at l-Cu sites. This coupling strengthens the d-π* polar-interaction on *CO₂, and helps switch the O in *OCH₃ to pseudo-lattice oxygen. These synergistic effects promote the CO₂ activation and targeted cleavage of the C−O bond in *OCH₃, thus driving efficient CH₄ production. In electrocatalytic experiments, Cu–Ti₁O₃ catalyst achieves a FE_CH₄ of 76%, a peak TOF_CH₄ of 24,390 h⁻¹, and record-long durability of >1230 h at industrial-level current densities. This work highlights the prospect of d-orbital engineering in enabling industrial-level electrocatalytic methanation on Cu SACs, and provides a promising catalyst design strategy toward sustainable CH₄ production for decarbonizing traditional thermal power plants.

Methods

Chemicals and materials

Potassium hydroxide (KOH, ≥85.0%), Sodium hydroxide (NaOH, ≥85.0%), acetone (99.7%), ethanol (99.7%), glycerol (99.7%), Copper chloride dihydrate (CuCl₂·2H₂O, ≥99.0%), Titanium oxysulfate dihydrate (TiOSO₄·2H₂O, ≥99.0%), Magnesium nitrate hexahydrate (Mg(NO₃)₂·2H₂O, ≥99.0%) and Aluminum nitrate nonahydrate (Al(NO₃)₃·9H₂O, ≥99.0%) were purchased from the Sinopharm Chemical Reagent Company. Ni foam was purchased from the Fangdou Technology Co., LTD (99% metal basis) with 0.3 mm thickness and 0.1 mm pore size. The Nafion solution (Dupont, D-520 dispersion, 5% w/w in water and 1-propanol) and triethylene glycol were purchased from Alfa Aesar. Gas diffusion layer (SGL Carbon, 28BC) was purchased from Fuel Cell Store. The hydroxide exchange membrane (Fumasep FAB-PK-130, 3 × 2 cm) for the flow cell was purchased from Fumatech, which was activated by soaking in 1 M KOH for 24 h for further test. The hydroxide exchange membrane (Grade 60, 3 × 3 cm) for the zero-gap electrolyzer was purchased from Dioxide Materials, which was activated by soaking in 1 M KOH for 24 h before the test. CO₂ gas (99.99%), Ar gas (99.99%), N₂ gas (99.99%), and all the standard gases were purchased from Shanghai Tomoe Gas Company.

Fabrication of Cu–Ti1O3

In a typical procedure, 2.2 g TiOSO₄·2H₂O, 0.15 g CuCl₂·2H₂O, 12 mL deionized water, 6 mL glycerol, and 20 mL ethanol were mixed and stirred for 1 h. The mixed solution was then moved into a 100 mL autoclave and kept at 200 °C for 8 h, followed by natural cooling to room temperature. Finally, the precipitate was washed thoroughly with alcohol and dried at 60 °C in a vacuum oven overnight. The sample was then calcined at 300 °C in 10% H₂/Ar for 3 h to obtain the Cu–Ti₁O₃ catalyst.

Fabrication of Cu–O4(Ti)

Cu–O₄(Ti) were prepared in a similar way as Cu–Ti₁O₃ nanoparticles, except that the final sample was calcined at 300 °C in the air for 3 h.

Preparation of electrodes

For the CO₂RR in a flow cell, a well-dispersed catalyst slurry containing 18 mg of catalysts, 3 mL of acetone, and 120 μL of Nafion (Dupont, D-520 dispersion, 5% w/w in water and 1-propanol) ionomer solution was slowly dropped onto a gas diffusion layer (SGL Carbon, 28BC), attaining a catalyst loading of ~0.3 mg cm⁻² measured by 0.01 mg analytical balance.

Materials characterizations

XRD was tested on a Bruker SMART APEX (II)-CCD. XPS data were characterized by a Thermo Scientific K-Alpha XPS. Scanning electron microscopy measurements were carried out by a Hitachi S4800 SEM. HRTEM measurements were performed by a JEM 2100 F (JEOL, Japan). ¹H-nuclear magnetic resonance (¹H-NMR) measurements were performed on a Bruker AVANCE AV III HD 500. Moreover, Athena and Artemis software with a standard IFEFFIT package were used to process XAS data.

Electrocatalytic CO2 reduction measurements

For CO₂RR in flow-cells, the electrochemical tests were conducted using a 1-cm² cell with a Gamry (Model: Reference 3000) electrochemical workstation at room temperature (25 ± 1 °C). The flow-cell configuration consists of a gas diffusion layer loaded with catalysts as the working electrode, a Ni foam as the anode, an Ag/AgCl electrode as the reference, and an anion exchange membrane to separate the cathode and anode chambers (Supplementary Fig. 48). The chamber volume of the cathode and anode is 0.5 cm³. All applied potentials were converted to the RHE scale according to the Nernst equation:

$${E}_{{{\rm{RHE}}}}={E}_{{{\rm{Ag}}}/{{{\rm{AgCl}}}}^{0}}+0.197+0.059\times {{\rm{pH}}}$$

(1)

The electrode potential was iR-corrected using the following equation:

$${E}_{{{\rm{sample}}}}={E}_{{{\rm{applied}}}}-0.9\times {I}_{{{\rm{total}}}}\times {R}_{{{\rm{cell}}}}$$

(2)

where E_sample was the corrected potential at the cathode, E_applied was the applied potential, and I_total was the total current. R_cell was the cell resistance of 0.98 ± 0.06 Ω, which was measured via potentiostatic EIS.

1 M KOH solution (pH = 14.00 ± 0.02) was utilized as the electrolyte that was prepared before electrochemical tests to ensure freshness. The reaction was performed with a steady stream of CO₂ (30 mL min⁻¹) measured by a soap bubble flowmeter. The gas product was collected and analyzed by a gas chromatograph (Shanghai Ramiin GC 2060) while the liquid products were quantified by ¹H-NMR spectroscopy (Bruker AVANCE AV III HD 500) using a method previously reported⁶⁰.

The gas product FE was calculated by the following equation:

$${{{\rm{FE}}}}(\%)=N\times F\times \frac{{S}_{{{\rm{x}}}}}{S}\times C\times v\,/\,I\times 100\%$$

(3)

where N was the number of charge transfers (H₂: 2; CH₄: 6; C₂H₄: 8); S_x was the peak area of the gas product and S was the peak area of corresponding standard gas; C was the concentration of standard gas; v was the flow rate of CO₂; I was the real-time current during CORR; F was the Faraday constant (96485 C mol^–1).

The liquid product FE was calculated by the following equation:

$${{\rm{FE}}}\left(\%\right)={{\rm{N}}}\times {{\rm{V}}}\times {{\rm{C}}}\times {{\rm{F}}}/{{\rm{Q}}}\times 100\%$$

(4)

where N was the number of charge transfers (Acetate: 4; EtOH: 8; n-PrOH: 12); V was the electrolyte volume (20 mL); C was the concentration of products; Q was the reaction charge.

The ECSA was calculated by assuming a monolayer of Pb adatoms over a Cu surface and 2e⁻ Pb oxidation with a conversion factor of 310 μC cm^{−2 61}. The EIS was performed at an open-circuit potential state in the frequency range from 100 kHz to 100 MHz. The Pb underpotential deposition was conducted in an Ar-saturated 0.1 M HClO₄ aqueous solution with 10 mM Pb(ClO₄)₂ by using cyclic voltammetry. This is the deposition of an adlayer of a foreign metal (i.e., Pb²⁺) on a substrate (i.e., surface Cu site) at potentials higher than the thermodynamic reduction potential of the electrodeposited metal. The foreign metal tends to deposit on the Cu rather than TiO₂ due to higher electrical conductivity, leading to accurate deposition on the Cu surface⁶². Before the experiment, the potential was first set at −0.20 V vs. RHE for 150 s, then swept from −0.0840 to 0.161 V vs. RHE with a sweep rate of 10 mV s⁻¹.

Stability tests in a zero-gap electrolyzer

The zero-gap electrolyzer (Shanghai Keqi, active area 5 cm², Supplementary Fig. 26) was utilized, which was established with a well-established membrane electrode assembly containing a cathode electrode, an anion membrane (Dioxide Materials, Grade 60), and an anode electrode (Ti mesh loaded with 2 mg cm⁻² IrO_x). 0.05 M HCl-humidified CO₂ was supplied with a flow rate of 30 sccm through the flow channels in the cathode flow field to reduce (bi)carbonate salt accumulation according to previous work⁶³, and 0.1 M KHCO₃ (pH = 6.83 ± 0.03) was fed into the anode flow channels with a flow rate of 15 sccm. The stability tests were performed at a constant current of −3 A. During operation, the anolyte was replaced regularly, and the precipitated salt was further removed by separating the device to inhibit channel block and active site shield.

Calculation for EE

For flow-cell systems, the ${{{\rm{EE}}}}_{{{{\rm{CH}}}}_{4}}$ was calculated based on the assumption that the overpotential of oxygen evolution is zero. The methane ${{{\rm{EE}}}}_{{{{\rm{CH}}}}_{4}}$ can be calculated as follows:

$${{{\rm{EE}}}}_{{{{\rm{CH}}}}_{4}}=\frac{(1.23+(-{E}_{{{{\rm{CH}}}}_{4}}))\times {{{\rm{FE}}}}_{{{{\rm{CH}}}}_{4}}}{(1.23+(-{E}_{{{\rm{applied}}}}))}$$

(5)

where E_applied is the potential applied in the experiment, ${{{\rm{FE}}}}_{{{{\rm{CH}}}}_{4}}$ is the measured FE of CH₄ in percentage, and ${E}_{{{{\rm{CH}}}}_{4}}$ = 0.17 V vs. RHE for CO₂RR⁶⁴.

For the zero-gap electrolyzer, ${{{\rm{EE}}}}_{{{{\rm{CH}}}}_{4}}$ is calculated as follows:

$${{{\rm{EE}}}}_{{{{\rm{CH}}}}_{4}}=\frac{(1.23+(-{E}_{{{{\rm{CH}}}}_{4}}))\times {{{\rm{FE}}}}_{{{{\rm{CH}}}}_{4}}}{-{E}_{{{\rm{full}}}-{{\rm{cellapplied}}}}}$$

(6)

where E_{full-cell applied} is the full-cell voltage applied in the zero-gap electrolyzer.

Calculation for TOF

The TOF of CH₄ is calculated according to the following equation:

$${{{\rm{TOF}}}}({{{\rm{h}}}}^{-1})=\frac{{I}_{{{{\rm{CH}}}}_{4}}/{{{\rm{n}}}}F}{\alpha {m}_{{{\rm{cat.}}}}/{M}_{{{\rm{Cu}}}}}\times 3600$$

(7)

Where n is the number of electrons transferred to CH₄ with 8, F is the Faradaic constant with 96485 C mol⁻¹, m_cat. is the catalyst mass in the electrode, ${{{\rm{\alpha }}}}$ is the mass ratio of surface Cu active sites in the catalyst measured by XPS, and M_Cu is the atomic mass of Cu.

In-situ electrochemical electron magnetic resonance

The in-situ E-EMR was carried out on a self-made in-situ cell according to our previous report with the employment of a magnetic resonance spectrometer (CIQTEK, EPR200, China)⁶⁵. The 0.17 M of DMPO (Dojindo Laboratories, Japan) was selected as the trap agent without additional purification. In the Schlenk line, the just-prepared electrode was cleaned in argon and then pumped with CO₂. The electrochemical analyzer (CHI852D, CH Instrument, USA) applied multi-step potentials and recorded current response with sequential 10 min of 0, −0.88, and 0 V vs. RHE. The collaboration between magnetic resonance and electro-analysis was ensured through a serial communication port. The obtained raw data were processed and analyzed based on Easyspin 6.0.6^66,67.

In-situ electrochemical infrared spectroscopy experiments

The in-situ attenuated total reflection surface enhanced infrared absorption (ATR-SEIRAS) measurements were conducted on a home-made spectro-electrochemical cell (Supplementary Fig. 31). A ~60-nm-thick SEIRA-active Au film was pre-deposited on the reflection plane of a semicircular cylinder Si prism. Then, the catalyst was drop-cast on the above film. The operando ATR-SEIRAS measurements were performed in an optical system embedded in the spectroscopic chamber. To reach a high signal/noise ratio, the spectrum was obtained by averaging 256 interferograms. The incident angle of the IR beam was set to 55°. The spectra were shown as absorbance defined as A = ‒logI₁/I₂, where I₁ and I₂ represent the irradiation intensity of incident and reflective beams, respectively.

For nano Au film preparation, a hemicylindrical Si prism was polished and then cleaned with the RCA method: soaking it in a 1:1:5 (v/v) solution of NH₃, H₂O₂, and H₂O, and afterward in a 1:1:5 (v/v) solution of HCl, H₂O₂, and H₂O at about 80 °C, followed by rinsing with water. Then the total reflecting plane was immersed in 40% NH₄F solution for 90 s to terminate the Si surface. Deposition of Au was performed at 60 °C by dropping a mixture of a plating solution and 2% HF (2:1 v/v) onto the Si surface, and the deposition time was 60 s. The composition of the plating solution was 0.015 M NaAuCl₄·2H₂O + 0.15 M Na₂SO₃ + 0.05 M Na₂S₂O₃·5H₂O + 0.05 M NH₄Cl. After deposition, the prism was rinsed with deionized water to finish the deposition.

For catalyst deposition, a well-dispersed catalyst slurry containing 18 mg of catalysts, 3 mL of acetone, and 120 μL of Nafion ionomer solution was slowly dropped onto Au film, attaining a catalyst loading of ~0.3 mg cm⁻² that can cover the Au film.

In-situ Raman experiments

A Raman spectrometer (HORIBA Jobin Yvon Inc., Model: HR Evolution) coupled with an optical microscope was used for in-situ Raman spectroscopy. A near-infrared laser (λ  =  532 nm) was used as the excitation source. The types of objective and grating were 50XL and 500 nm, respectively. A homemade spectro-electrochemical cell equipped with a Pt wire counter electrode and an Ag/AgCl reference electrode was used, with the working electrode prepared by coating the catalyst onto glassy carbon. Each potential was maintained for at least 10 min before collecting the spectra.

Quasi-in-situ XPS measurement

All samples were prepared in the glovebox with an oxygen concentration of less than 4 ppm. The samples were prepared through 10-min CO₂RR in Ar-saturated 1 M KOH at around −0.88 V, and were then packaged with liquid nitrogen protection to stabilize the surface reaction intermediate for studying its influence on O_v. For the exploration of the O_v structure after reaction over catalysts, the samples were prepared in the same reaction conditions and were then dried under vacuum at 120 °C for 24 h to remove the surface species.

Theoretical computational

DFT calculations were performed with the Vienna ab initio simulation package (VASP), and the detailed calculation methods and parameters are presented in Supplementary Note 4 and Supplementary Data 1. Generally, all structures were relaxed by a conjugate gradient method until the residual force component was <0.04 eV Å⁻¹, and the convergence criterion of total energy in the self-consistent field method was set to 10⁻⁶ eV. The original structure of anatase was employed for calculations with the parameters that a = b = 3.7825 Å, c = 9.6150 Å, and α = b = γ = 90.0000°. With the relaxed cell structure, the (100) plane was utilized for further assessments according to XRD and TEM results. DFT + U calculations on the density of states were performed for both Cu−Ti₁O₃ and Cu−O₄(Ti), and Lobster was employed for COHP analyses⁶⁸.

Furthermore, (100) surface was also employed for CO₂RR evaluations, using a slab model consisting of three atomic layers and 2 × 1-unit cells for the lateral dimensions, where the thickness of the vacuum was set at 12 Å to ensure that there was no superficial interaction between different layers. The total size for the established slab model was 7.6134 × 9.5168 × 21.3227 ų, and α = β = γ = 90.0000°. The Brillouin zone of the supercell was sampled by a 3 × 3 × 1 uniform k-point mesh. All the energy calculations were based on a two-step process involving the structure relaxation in vacuum (i.e., electron energy, E_DFT) and a subsequent single-point calculation using an implicit solvation model.

The bond dissociation enthalpies of Cu–O and O–C in Cu–*O–CH₃ were calculated based on a similar method reported previously¹⁵:

$${\Delta H}_{{{\rm{Cu}}}-{{\rm{O}}}}={H}_{\ast+{{{\rm{OCH}}}}_{3}}-{H}_{\ast {{{\rm{OCH}}}}_{3}}$$

(8)

$${\Delta H}_{{{\rm{O}}}-{{\rm{C}}}}={H}_{\ast {{\rm{O}}}+{{{\rm{CH}}}}_{3}}-{H}_{\ast {{{\rm{OCH}}}}_{3}}$$

(9)

where ${H}_{\ast {{{\rm{OCH}}}}_{3}}$ and H_*O are the enthalpies of the catalyst with adsorption of the corresponding molecules, ${H}_{\ast {{{\rm{O+CH}}}}_{3}}$ and ${H}_{\ast {{{\rm{O+CH}}}}_{3}}$ are the enthalpies of the corresponding bond-breaking models.

The Gibbs free energy change of formation (at T K) for each elementary step was calculated as:

$$\Delta G\left(T\right)={\Delta E}_{{{\rm{DFT}}}}+{\Delta E}_{{{\rm{sol}}}\,}+{\Delta G}_{{{\rm{cor}}}}(T)$$

(10)

where ∆E_sol was the solvation correction, ${\Delta G}_{{{\rm{cor}}}}\left(T\right)={\Delta E}_{{{\rm{ZPE}}}}-T\Delta S+\Delta U(T)$ was obtained by processing frequency calculation results, and ∆G_298.15K was used to construct the free energy profiles. The computational hydrogen electrode model was also employed⁶⁹, in which the chemical potential of (H⁺ + e^–) is equal to that of half a H₂ molecule with a pH correction of k_BTln([H⁺]).