Photocatalytic asymmetric C-C coupling for CO₂ reduction on dynamically reconstructed Ru^δ+-O/Ru⁰-O sites

Abstract

Solar-driven CO₂ reduction to value-added C₂ chemicals is thermodynamically challenging due to multiple complicated steps. The design of active sites and structures for photocatalysts is necessary to improve solar energy efficiency. In this work, atomically dispersed Ru-O sites in Ru_xIn_2-xO₃ are constructed by interior lattice anchoring of Ru. This results in the dynamic reconstruction of Ru^δ+-O/Ru⁰-O sites upon photoexcitation, which facilitates the CO₂ activation, *CO intermediates adsorption, and C-C coupling as demonstrated by varied in situ techniques. A SiO₂ core in Ru_xIn_2-xO₃/SiO₂ construction further enhances the solar energy utilization and individual Ru_xIn_2-xO₃ nanocrystals dispersion for photocatalytic CO₂ reduction reaction. It results in the maximum ethanol production rate up to 31.6 μmol/g/h with over 90% selectivity. DFT simulation reveals that the C₂ dimer formation primarily underwent an asymmetric *CO-*CHO coupling route via a low-energy precedence ladder of *CHO. This work provides an insightful understanding of active sites with dynamic reconstruction towards asymmetric C-C coupling for CO₂RR at the atomic scale.

Introduction

Solar-driven carbon dioxide (CO₂) reduction to value-added chemicals is a desirable approach towards carbon neutralization^1,2,3. For comparison, C₂₊ chemicals produced from CO₂ reduction reaction (CO₂RR), such as ethanol and ethylene, are more desirable than C₁ chemicals like methane, carbon monoxide, and methanol, due to the merit of higher-energy density^4,5. During the CO₂RR process to produce C₂₊ chemicals, beyond the energy required for the dissociation of the inert C=O bond (~750 kJ/mol) and activation of intermediates, additional energy is needed to drive the C–C coupling step, leading to the formation of carbonous dimers (such as *CO-*CO and *CO-*CHO)^6,7. Additionally, to facilitate C-C coupling, it is essential to increase the coverage of *C1 species (such as *CO and *CHO) and reduce the adsorption energy of CO on the active sites to enhance carbon-based dimer formation^8,9,10. These thermodynamic and kinetic challenges make a photocatalytic reduction of CO₂ into C₂₊ chemicals an exceptionally difficult process.

In a photocatalysis system, only a fraction of solar energy is absorbed to generate charge carriers for photocatalysts, leading to inefficient utilization of solar energy (Supplementary Fig. S1)¹¹. Core-shell is a distinctive catalyst structure designed to enhance solar energy efficiency, where either the outer shell or the inner core of the photocatalysts helps reduce light energy loss via secondary irradiation^12,13. Hence, a silica-based core-shell system has been applied in the CO₂ hydrogenation reaction^14,15,16.

The actual physical and chemical states of the metal active center play a crucial role in multi-step CO₂RR towards C₂₊ chemicals production. For CO₂RR on copper-based catalysts, the metallic species (Cu⁰) exhibit thermodynamically favorable activation of CO₂ compared with oxidized species, while Cu^δ+ species in a higher oxidation state serve as indispensable sites for C-C coupling step^17,18,19. Moreover, a suppressed generation of Cu⁰ led to the insufficiency of Cu⁰-Cu^δ+ pairs for C₂₊ production^18,20. Notably, the synergistic reconstruction of Cu⁰ and Cu^δ+ species effectively completes the complicated CO₂ conversion to C₂₊ chemicals. However, the dynamic reconstruction nature of catalysts beyond Cu has rarely been unveiled for photocatalytic CO₂RR. With an abundant supply of electrons in its d orbital that can participate in the formation of d-π* antibonding with CO₂, Ruthenium (Ru) has been recognized as a highly active metal center for CO₂RR^21,22. Nevertheless, CO₂RR into C₂₊ chemicals has rarely been achieved on Ru-based photocatalysts, likely due to the limited understanding of the dynamic construction of Ru catalysts. Recently, single-atom catalysts (SAC) with unsaturated coordination or hemilability have been extensively adopted for CO₂RR because of the merit of high atomic efficiency, high catalytic efficiency, and unique reaction selectivity^23,24,25. More importantly, the nature of the isolated catalytic center for SAC enables the investigation of the dynamic reconstruction of photocatalysts during the CO₂RR process.

In this work, we synthesized Ru_xIn_2-xO₃ nanocrystals with Ru atomically dispersed within In₂O₃ lattice, serving as a model photocatalyst to investigate the dynamic reconstruction of active sites and the resultant reaction pathway for photocatalytic CO₂RR. Multiple techniques (STEM, XANES, Raman et al.) revealed that Ru atoms anchored within the lattice exist in a single-atom state, exhibiting unsaturated coordination with adjacent O to form Ru^δ+-O sites. The improved productivity of C₂ chemical (ethanol) from CO₂RR confirms the synergy between enhanced solar energy utilization and effective dispersion of individual active nanocrystals in such Ru_xIn_2-xO₃/SiO₂ core-shell structure. In situ XPS, DRIFTS and Raman spectroscopy unveiled a dynamic reconstruction between Ru⁰-O and Ru^δ+-O: the metallic state Ru⁰-O site exhibited thermodynamically favorable activation of CO₂, providing high coverage and low adsorption energy of *CO species on the active sites, while the oxidized state of Ru^δ+-O facilitated the intermediates stabilization and C-C coupling. Under solar irradiation, the local electron redistribution in the Ru-O-In sites resulted in the transient acid-base pairs generation, which promoted CO₂ activation, intermediates regulation and C-C coupling. Consequently, this dynamic reconstruction enabled the solar-driven Ru_xIn_2-xO₃/SiO₂ to convert CO₂ into ethanol with a maximum productivity of 31.6 μmol/g/h and a selectivity greater than 90%. Density functional theory (DFT) simulation further revealed the dominant *CO-*CHO dimerization pathway for ethanol production.

Results

Morphology and Structure analysis of catalysts

A series of Ru_xIn_2-xO₃ (x = 0%, 0.5%, 1.0%, 3.0%, and 5.0%) nanospheres were synthesized by varying the amount of Ru³⁺ precursor (Supplementary Fig. S2). Typically, the average size of pure In₂O₃ was ~58 nm (Supplementary Fig. S3), whose main exposed lattice fringes were (222) and (400) facets with space of 0.292 nm and 0.253 nm, respectively (Supplementary Fig. S4). Because of the smaller radius of Ru³⁺ (0.68 Å) compared to In³⁺ (0.81 Å), the incorporation of Ru³⁺ into In₂O₃ resulted in a smaller size of Ru_xIn_2-xO₃ nanosphere, with the 3.0%Ru_xIn_2-xO₃ nanosphere, for example, measuring around 52 nm (Supplementary Fig. S5). The corresponding (222) facet in 3.0%Ru_xIn_2-xO₃ exhibited a slightly smaller interplanar spacing of 0.286 nm compared with bcc In₂O₃ (0.292 nm) (Supplementary Fig. S6). Moreover, Ru atoms presented a homogeneous dispersion in the 3.0%Ru_xIn_2-xO₃ nanocrystal with no obvious evidence of metallic Ru and RuO₂ characteristic fringes (Supplementary Fig. S6). Additionally, the incorporation of smaller Ru ions within Ru_xIn_2-xO₃ nanocrystals resulted in a larger-surface area compared with pure In₂O₃ (Supplementary Fig. S7), thereby providing more adsorption sites for CO₂ photoreduction. To enhance the photothermal effect and ensure the dispersion of individual nanocrystal, a Ru_xIn_2-xO₃/SiO₂ core-shell structure was fabricated, aiming to reduce solar energy loss and improve the interaction of active site with adsorbates. Herein, amorphous silica acted as the core, and its larger surface (size of ~0.64 μm) (Supplementary Fig. S8) provided more planting-bed sites for the agglomeration of 3.0%Ru_xIn_2-xO₃ nanocrystals (Supplementary Fig. S2). This allowed individual Ru_xIn_2-xO₃ nanocrystal to be efficiently dispersed onto the larger surface support, further boosting the interaction with adsorbates (CO₂). As shown in Fig. 1a–d and Supplementary Fig. S9, these images confirm a core-shell structure of 3.0%Ru_xIn_2-xO₃ shell with SiO₂ core. The 3.0%Ru_xIn_2-xO₃ nanocrystals were closely attached to the surface of SiO₂. Energy-dispersive X-ray spectroscopy line scanning (Fig. 1f, g) and elemental mapping (Fig. 1h) images further proved that the 3.0%Ru_xIn_2-xO₃ nanocrystals were uniformly dispersed on the SiO₂ support. Furthermore, atomic force microscopy demonstrated the arc shape of the core-shell 3.0%Ru_xIn_2-xO₃ /SiO₂ surface (Fig. 1e and Supplementary Fig. S10). The Ru mole proportion in each Ru_xIn_2-xO₃ series nanocrystals were measured by the inductively coupled plasma mass spectrometry (ICP-MS) (Supplementary Table S1).

Fig. 1: Morphology and structure of photocatalyst.

a, b TEM and d HRTEM images of 3.0%Ru_xIn_2-xO₃/SiO₂ nanocrystal (inset: relevant FFT pattern of (222) facet), and c corresponding lattice space of (222) facet; e Line-scanning profile of 3.0%Ru_xIn_2-xO₃/SiO₂ surface from 2D AFM image in Supplementary Fig. S10 and corresponding 3D AFM image (bottom); f HAADF-STEM image, and corresponding g line-scanning elemental distribution profile along red line in (f), and h relevant elemental mappings of 3.0%Ru_xIn_2-xO₃/SiO₂ nanocrystal.

The crystal phase structure of the as-synthesized catalysts was investigated by X-ray diffraction (XRD) (Supplementary Fig. S11). All the characterized peaks of Ru_xIn_2-xO₃ series samples matched with the identical XRD patterns of cubic In₂O₃, with the primary (222) facet. Due to its smaller radius in contrast with In inos mentioned above, the incorporation of Ru ions into Ru_xIn_2-xO₃ nanocrystals led to a larger 2 theta shift of (222) facet compared to pure In₂O₃, which contributed to internal lattice disrtortion. All the Ru_xIn_2-xO₃ nanocrystals’ XRD patterns showed no obvious characteristic peaks related to metallic Ru and RuO₂. The extended shoulder of the characterized peak around 2 theta of 21° (Supplementary Fig. S12) attributed to amorphous silica (Supplementary Fig. S13) confirmed the formation of 3.0%Ru_xIn_2-xO₃/SiO₂ composite. The crystal structure of catalysts was further supported by Raman spectra. As displayed in Supplementary Fig. S14, all the peaks located in 130.6, 307.2, 367.1, 494.8, and 627.9 cm⁻¹ are identically indexed to the E_2g, E_1g, and A_1g vibration models of In₂O₃, respectively²⁶. Significantly, an obvious B_2g vibration peak at 740.5 cm⁻¹ assigned to the in-plane Ru-O mode indicates that Ru atoms are bonded with adjacent O in Ru_xIn_2-xO₃²⁷, moreover, the intensity of B_2g peak increased with higher Ru contents. Interestingly, the intensity of B_2g peak decreased at Ru mole proportion above 3.0%, likely due to excessive Ru³⁺ ions aggregating into metallic Ru-Ru bonds instead of forming Ru-O bonds (Supplementary Fig. S15). Taking into consideration of above XRD and Raman discussion, the Ru configuration in Ru_xIn_2-xO₃ probably was interior lattice anchoring. To deepen insights into the chemical state of Ru and electron redistribution of Ru-(O)-In sites, high-resolution spectra of Ru 3 d, In 3 d, and O 1 s were investigated by X-ray photoelectron spectroscopy (XPS). As shown in Fig. 2a, Ru predominantly existed in the oxide state of RuO_x, and with a minor presence of metallic state Ru⁰ in 3.0%Ru_xIn_2-xO₃ sample^28,29. Particularly, the Ru 3d_5/2 peak corresponding to the oxide state in 3.0%Ru_xIn_2-xO₃ exhibted a slight ~0.41 eV shift to lower binding energy compared with that in RuO₂, indicating a lower oxide state of Ru^δ+ realtive to Ru⁴⁺³⁰. In other words, Ru^δ+ probably coordinated with O in an unsaturated state in contrast to Ru-O bonding in RuO₂. Additionally, oxide states in O 1 s and In 3 d spectra displayed a slight ~0.55 eV and ~0.61 eV shift to higher binding energy (Fig. 2b and Supplementary Fig. S16) compared to those in In₂O₃. This further proved that smaller Ru ions could attract more electrons from neighboring In and O atoms (Fig. 2c), ultimately leading to electron redistribution and accumulation at the local Ru-(O)-In sites.

Fig. 2: Chemical state and coordination structure of photocatalyst.

High-resolution XPS spectra of a Ru 3 d + C 1 s and b O 1 s; c Schematic illustration of electrons redistribution in local In-O site of pure In₂O₃ (bottom) and Ru-O-In site of 3.0%Ru_xIn_2-xO₃ (top); d HAADF-STEM images of 3.0%Ru_xIn_2-xO₃ nanocrystal, and e corresponding 3D surface plot of yellow-dotted area (insert: atomic profile of white-line scanning). f Normalized Ru K-edge XANES spectra (insert: enlarged pink area) and g corresponding k²-Weighted Ru K-edge Fourier-transform EXAFS spectra; h Normalized In K-edge XANES spectra and i corresponding k²-Weighted In K-edge Fourier-transform EXAFS spectra of In₂O₃, 0.5%Ru_xIn_2-xO₃, 3.0%Ru_xIn_2-xO₃ and 3.0%Ru_xIn_2-xO₃/SiO₂ nanocrystals respectively, as well as Ru foil, RuO₂ and In foil references. j Schematic illustration of In-O bond length (R_In-O) in pure In₂O₃ (top), and Ru-O bond length (R_Ru-O) in 3.0%Ru_xIn_2-xO₃ (middle) and RuO₂ (bottom) (purple: In; red: O; cyan: Ru; black: oxygen vacancy).

To deepen insights into the Ru configuration in Ru_xIn_2-xO₃, spherical aberration-corrected scanning transmission electron microscopy (STEM) images further unveiled homogenously atomical dispersion of Ru atoms with a lower contrast compared to In atoms (Fig. 2d and Supplementary Fig. S17), wherein atomic Ru anchored into the interior lattice site to form Ru-(O)-In configuration (Fig. 2e). Furthermore, the brightness-faded area observed after Ru incorporation suggested unsaturated coordination, probably deriving from O vacancies. The electron paramagnetic resonance (EPR) spectra (Supplementary Fig. S18) revealed the existence of paramagnetic oxygen vacancy species (g = 2.003), thereby further providing insightful evidence for unsaturated Ru-O coordination³⁰. To further investigate the electronic structure and coordination environment of Ru and In atoms in Ru_xIn_2-xO₃ nanocrystals, the X-ray absorption near edge structure (XANES) and extended X-ray absorption fine structure (EXAFS) spectra of Ru and In were obtained in Fig. 2f–i, respectively. From the Ru K-edge XANES spectra of Ru_xIn_2-xO₃, the near absorption edge (pink area) appeared at a slightly lower energy than that of RuO₂ (Fig. 2f), indicating that the chemical state of Ru in Ru_xIn_2-xO₃ is lower than +4. Additionally, the white-line peak of Ru-O in Ru_xIn_2-xO₃ exhibited both a lower energy position and reduced intensity compared to RuO₂. This suggests more electrons geometrically deviated from neighboring In and O to Ru 3 d orbitals, enhancing the electron donation for dynamic Ru^δ+ reduction and CO₂ activation. This intriguing electron redistribution around Ru-(O)-In sites was further proved by the higher-energy shift of In-O peak in the In K-edge XANES spectra of Ru_xIn_2-xO₃ (Fig. 2h) compared with that of In₂O₃ as well. The above discussion is also in accordance with the aforementioned XPS analysis. The Fourier-transformed k²-weighted EXAFS analysis unveiled that a dominant peak at 1.67 Å (Fig. 2g) was assigned to the near-shell coordination of Ru^δ+-O bonding in 3.0%Ru_xIn_2-xO₃ nanocrystal. Further EXAFS spectra fitting (Supplementary Figs. S19–S26 and Supplementary Tables S2, S3) determined the accurate Ru^δ+-O bond length to be 2.06 Å, which is longer than Ru⁴⁺-O bond (1.98 Å) in RuO₂ and shorter than In³⁺-O bond (2.18 Å) in In₂O₃ (Fig. 2j). This is attributed to the radius of Ru^δ+ was longer than Ru⁴⁺ and shorter than In³⁺, respectively. The fact of the shortened Ru-O bond length in Ru_xIn_2-xO₃ compared to In-O bond in In₂O₃ was in good agreement with the simulated geometry structure (Supplementary Fig. S27). This change would contribute to the lattice distortion of (222) facet in Ru_xIn_2-xO₃ (Supplementary Fig. S11). From the EXAFS data fitting table (Supplementary Table S2), the Ru atoms are coordinated with neighber O (Ru-O) and In (Ru-In) due to the interior lattice anchoring (Fig. 2d). Moreover, the Ru-O coordination in 3.0%Ru_xIn_2-xO₃ presented a lower coordination number of 2 compared to saturated coordination of Ru⁽⁴⁺⁾-O (6) in RuO₂ as well as In⁽³⁺⁾-O (6) in In₂O₃ standard^4,23. This unsaturated coordination state may derive from the formation of adjacent oxygen vacancies (Supplementary Fig. S18). In addition, due to the Ru incoporation, the related In-O and In-In coordination in Ru_xIn_2-xO₃ framework appeard a slight decline (Supplementary Table S3), which can be attributed to electron delocalization from O and In⁴. It is worth noting that the SiO₂ core combining with 3.0%Ru_xIn_2-xO₃ basically had a negligible influence on the Ru and In atoms coordination environment (Supplementary Tables S2 and S3). Interestingly, because of the electron redistribution around Ru-(O)-In sites, the CO₂ and NH₃ temperature-programmed desorption (CO₂/NH₃-TPD) spectra revealed a strengthened acid-base pair formation in Ru_xIn_2-xO₃ nanocrystals in contrast with pure In₂O₃ (Supplementary Fig. S28), which could provide abundant adsorption sites for the CO₂ activation.

The density of state (DOS) was analyzed to gain insights into the electronic structure of Ru_xIn_2-xO₃ nanocrystals. Because of the interior lattice anchoring of Ru into Ru_xIn_2-xO₃ nanocrystals, the d orbitals of Ru would hybridize with the O p orbitals, leading to a defective structure with a visible signature below the conduction band (Fig. 3a, b). This defective state formation led to the darkened color of the Ru_xIn_2-xO₃ catalysts (Supplementary Fig. S29), enhancing the absorption of photons (Supplementary Fig. S30). On the other hand, due to the strong interaction of Ru d orbitals with the 2π orbitals of CO₂ (Fig. 3c), more electrons infused into the bonding state (d-π) below Fermi level (Fig. 3d, e, and Supplementary Fig. S31), which resulted in a strong interaction between CO₂ and Ru-O site, further boosting the C = O bond dissociation.

Fig. 3: Electronic hybridization property of photocatalysts.

a Simulated DOS plots of In₂O₃ (left) and Ru_xIn_2-xO₃ (right), and b relevant schematic illustration of defective energy level formation from hybridization of Ru d orbital and O p orbital; c pDOS simulation plot of Ru orbitals (including s, p, and d) in Ru_xIn_2-xO₃ hybridizing with CO₂ orbitals (including s and p), and d corresponding schematic illustration d_xz/yz-π bonding orbital deriving from Ru-CO_2; e Charge difference plots for side-view CO₂ adsorbed on Ru_xIn_2-xO₃ (right) and In₂O₃ (left) models (The yellow and blue contours represent electron density accumulation and depression, respectively).

To decipher the mechanism of photogenerated charge carriers transfer dynamics induced by the defective state after Ru incorporation, the femtosecond transient absorption (fs-TA) spectroscopy was carried out to track electrons transfer from the excited state to trap or interfacial state³¹. The fs-TA spectra of In₂O₃ and 3.0%Ru_xIn_2-xO₃ with a range of 425–750 nm are demonstrated in Fig. 4a, b, d, e, wherein both a minor ground-state bleach and a prominent photoinduced absorption (PIA) process could be observed at different time delays. The PIA is related to the absorption of excited-state electrons generated in conduction band³². Due to the Ru incorporation, the PIA of 3.0%Ru_xIn_2-xO₃ (on set at 465 nm) appeared a slight blue-shift compared with that of In₂O₃ (on set at 500 nm), which suggests the Ru incorporation is favarable for the PIA process under illumination. Furthermore, a more comparable absorbance of 3.0%Ru_xIn_2-xO₃ revealed more charge carriers generated upon light absorption in contrast with In₂O₃, due to the defect level formation. To give a quantitative description of charge carriers transfer dynamics, we focused on the major PIA relaxation dynamics in Fig. 4a, d. The triexponential fitting based on the dynamics curve probed at 630 nm yields the relaxation time constants (and weighted coefficients) (Fig. 4c, f): τ₁ = 2.9 ± 0.2 ps (88.0%), τ₂ = 33.6 ± 3.3 ps (10.0%), and τ₃ = 837.1 ± 187.2 ps (2.0%) for In₂O₃; and τ₁ = 1.9 ± 0.2 ps (97.4%), τ₂ = 21.7 ± 1.7 ps (2.3%), and τ₃ = 412.7 ± 65.2 ps (0.3%) for 3.0%Ru_xIn_2-xO₃ (Supplementary Table S4). Because the charge trapping and radiative recombination usually occurs at a picosecond and nanosecond time scale, respectively, we attributed the relaxation time constants to the event of electron trapping at different depth (τ₁ and τ₂) and non-radiative recombination (τ₃)³³. Notably, faster PIA relaxation time constants associated with electron trapping could be affirmed in 3.0%Ru_xIn_2-xO₃. This suggets that the defective energy level arisen from Ru incorporation benefits photoinduced electron trapping, faciliating dissociation into free electrons (Supplementary Fig. S32). Consequently, the accelerated charge separation contributed to enhanced photocurrent density (Supplementary Fig. S33). Although photogenerated charge underwent a relative fast non-radiative recombination in 3.0%Ru_xIn_2-xO₃ compared with In₂O₃, the weight ratio merely accounts for 0.3%. Taken together, the aforementioned discussion demonstrates that a unique defective level continuation attributed to Ru incorporation accelerates electron trapping and charge separation, thereby endowing photogenerated electrons with more opportunity to reduce Ru^δ+ species and activate CO₂.

Fig. 4: Photogenerated charges performance.

Normalized transient adsorption kinetic decay curves, contour maps fs-TA spectra, and relevant PIA relaxation dynamics curve of (a–c) In₂O₃ and (d–f) 3.0%Ru_xIn_2-xO₃, respectively (excitation at 400 nm and probe at 630 nm); 3D KPFM images of 3.0%Ru_xIn_2-xO₃/SiO₂ catalyst under dark g and illumination (h, i) with different monochromatic wavelength (h: λ = 420 nm, i: λ = 532 nm).

To verify the efficient photogenerated charges carriers transfer onto surface, Kelvin probe force microscopy (KPFM) was performed under dark and illumination with different monochromatic wavelengths, respectively (Fig. 4g–i and Supplementary Fig. S34). As shown in Supplementary Fig. S35, the line-scanning plots of charge distribution demonstrated that the average surface charge of 3.0%Ru_xIn_2-xO₃/SiO₂ is ~−122 mV under dark. While, that shifts to −379 mV (λ = 420 nm) and −248 mV (λ = 532 nm) under illumination, respectively. This directly verified that abundant active electrons were induced onto the catalyst surface under light, which provided reliable evidence of a negative charge microenvironment for 12e-required CO₂ photoreduction to ethanol ($2{{CO}}_{2}+12{H}^{+}+12{e}^{-}\to {C}_{2}{H}_{5}{OH}+3{H}_{2}O$).

Photocatalytic CO₂ reduction performance

The photocatalytic CO₂RR performance over Ru_xIn_2-xO₃ series catalysts was evaluated in a gas–liquid–solid three phases system. From Supplementary Figs. S36–S38, the possible products include C₁ (CO, CH₄ and HCOOH) and C₂ (C₂H₄, C₂H₆, and CH₃CH₂OH) chemicals. Notably, because of the absence of active Ru^δ+(-O) sites for C-C coupling in pure In₂O₃, there is no C₂ chemicals production. While, a good productivity of ethanol was obtained in the Ru_xIn_2-xO₃ series catalysts (Supplementary Fig. S37b). The ethanol productivity increased with progressively higher Ru content, following a slight decrease over 3.0% loading, which resulted from a decline of Ru^δ+-O sites on the Ru nanoparticles (Supplementary Fig. S15). This result revealed the Ru^δ+-O sites dominantly acted as the C–C coupling center. Interestingly, the ethanol productivity over 3.0%Ru_xIn_2-xO₃/SiO₂ (18.1 μmol/g/h) improved ~12.3% compared with that of single 3.0%Ru_xIn_2-xO₃ (16.1 μmol/g). The enhanced activity was attributed to extra thermal energy provided from a secondary irradiation by silica core (Supplementary Fig. S39), which stabilized more excited hot electrons and boosted endothermic C-C coupling step. Furthermore, sufficient dispersion of individual Ru_xIn_2-xO₃ nanocrystals improved more interaction between active sites and CO₂. To further improve the productivity of ethanol by enhancing mass transfer and driving force for product direction⁶, a series of pressurized experiments were conducted using single 3.0%Ru_xIn_2-xO₃ and 3.0%Ru_xIn_2-xO₃/SiO₂ catalysts under similar condition otherwise varied CO₂ pressure. As shown in Fig. 5a, the ethanol productivity of both 3.0%Ru_xIn_2-xO₃ and 3.0%Ru_xIn_2-xO₃/SiO₂ catalysts finally improved up to 28.3 and 31.6 μmol/g/h, respectively, with over 90% selectivity. The photocatalytic performance of 3.0%Ru_xIn_2-xO₃ and 3.0%Ru_xIn_2-xO₃/SiO₂ catalysts under visible light was investigated at maximum pressure of CO₂ as well. The corresponding ethanol productivity was 14.2 μmol/g/h (86.9% Sel.) and 16.4 μmol/g/h (85.9% Sel.), respectively (Fig. 5b). A peer performance summary of photocatalytic CO₂RR to C₂ products was discussed in Supplementary Table S5. It is known that *CO is a key intermediate in the CO₂RR to C₂ products. In particular, the formation of *COCO* or *COCHO* species through C-C coupling of dual *CO or *CO with *CHO further promoted the formation of C₂ products^8,34,35. Therefore, we tried to aerate extra high-purity CO gas into the reaction system at a similar condition, aiming to enhance the adsorption of CO species onto the surfaces and boost the formation of *CO species directly. As a result, from Fig. 5c, the ethanol productivity was remarkably improved at a CO₂ + CO combination compared with single CO₂ at the same pressure. This result unveiled that the *CO is the key precursor of the C₂ dimer during the C-C coupling process. A series of control and isotopic trace experiments further confirmed the photocatalytic CO₂ conversion into ethanol (Supplementary Figs. S40, 41). The recycle experiment revealed the 3.0%Ru_xIn_2-xO₃/SiO₂ could remain a well-stable activity for photocatalytic CO₂RR (Supplementary Fig. S42). Moreover, 3.0%Ru_xIn_2-xO₃/SiO₂ could keep a good crystal and electronic structure during the course of the reaction (Supplementary Figs. S43–S45). Besides, the interior anchored Ru atoms still remained a homogeneously atomic state without serious aggregation after reaction (Supplementary Fig. S46). The relevant band structure worked for the photocatalytic CO₂RR was illustrated in Supplementary Fig. S47.

Fig. 5: Photocatalytic performance of CO₂RR.

a Photocatalytic CO₂RR performance and related ethanol selectivity under full-spectrum illumination for 3 h at varied pressure of CO₂ (P_CO2 = 1, 2, 4, and 6 bar) over 3.0%Ru_xIn_2-xO₃ (top) and 3.0%Ru_xIn_2-xO₃/SiO₂ (bottom); b Photocatalytic CO₂RR performance and related ethanol selectivity under visible light (λ > 420 nm) for 3 h at CO₂ = 6 bar over 3.0%Ru_xIn_2-xO₃ (top) and 3.0%Ru_xIn_2-xO₃/SiO₂ (bottom); c Ethanol productivity of CO as intermediate exploration in CO₂ photoreduction process under full-spectrum illumination for 3 h at different pressure of CO₂ and CO combination over 3.0%Ru_xIn_2-xO₃/SiO₂.

In situ dynamic reconstruction of Ru^δ+-O/Ru⁰-O sites analysis

From the above discussion, the unsaturated coordination Ru^δ+-O sites in Ru_xIn_2-xO₃ nanocrystal played a decisive role in photocatalytic CO₂RR. To monitor the dynamic change of Ru^δ+-O sites during the CO₂RR process, in situ XPS spectra were carried out over 3.0%Ru_xIn_2-xO₃/SiO₂ under initial dark, time-resolved illumination, and following a “light-off” again. As illustrated in Ru 3 d spectra (Fig. 6a), the Ru⁰/Ru^δ+ 3d_5/2 species showed a remarkable dynamic change along with illumination continuation. Specifically, the peaks of both Ru⁰ 3d_5/2 and Ru^δ+ 3d_5/2 species synchronously shifted to a lower energy range over the first 30 min, followed by a sluggish shift, and then a slightly additional shift to lower energy after 45 min. Furthermore, after removed light source, the peaks of Ru⁰/Ru^δ+ 3d_5/2 slightly shifted to a higher-energy range as well as Ru⁰/Ru^δ+ 3d_3/2, which suggests the dynamic transformation of Ru⁰/Ru^δ+ 3d_5/2 would recover to a thermodynamics equilibrium tendency in the absence of light excitation. However, it was worth noting that the equilibrium would undergo a sluggish process, and an ideal recovery to the initial state would be hard to observe immediately. This potential reducibility of Ru species could be confirmed by H₂-TPR spectra (Supplementary Fig. S48). On the contrary, the oxides peak in O 1 s spectra presented an opposite shift trend along with the Ru 3 d_5/2 peak shift (Fig. 6b). Interestingly, because of the stronger electron attraction of Ru^δ+ at Ru-(O)-In site as above discussion, a fraction of electrons from In atoms would contribute to Ru⁰ species formation during illumination (slight higher-energy shift of In 3d in Supplementary Fig. S49), confirming that no obvious In⁰ species would be observed. The above results unveiled that unsaturated coordination of Ru^δ+ exhibited an intimate electron transfer interaction with neighboring O and In.

Fig. 6: Dynamically reconstructed analysis of Ru^δ+-O/Ru⁰-O sites.

In situ XPS spectra of high-resolution a C 1 s + Ru 3 d and b O 1 s of 3.0%Ru_xIn_2-xO₃/SiO₂ catalyst under dark and illumination (full-spectrum light, 300 W Xe lamp) at time scale, c corresponding dynamic change of proportion of Ru⁰ 3 d_5/2 (yellow) and Ru^δ+ 3 d_5/2 (purple) in RuO_x peaks in Ru 3d XPS spectra, and d relevant plot for the ratio of Ru⁰ 3d_5/2 peak to Ru^δ+ 3d_5/2 in RuO_x peak proportion (I_R0/I_Ruδ+) in Ru 3d XPS spectra, ratio of oxides peak to Si-O peak proportion (I_Oxides/I_Si-O) in O 1 s XPS spectra, and ratio of O-C = O peak to C-C peak proportion (I_O-C=O/I_C-C) in C 1 s XPS spectra; e In situ Raman spectra of B_2g (in-plane Ru-O mode) of 3.0%Ru_xIn_2-xO₃/SiO₂ (i) and In₂O₃ (ii) under dark and illumination (full-spectrum) at time scale at a range of 500–1000 cm⁻¹, and f corresponding time-resolved relative intensity change of B_2g mode; g schematic illustration of dynamic Ru⁰-O/Ru^δ+-O sites reconstruction during photocatalytic CO₂ reduction process (P represents product, I represents intermediates).

To further validate the above hypothesis, the dynamic reconstruction process of Ru^δ+-O/Ru⁰-O will be investigated by quantifying the changes in proportions of characteristic peaks in Ru 3 d, C 1 s, and O 1 s spectra over time. As shown in Fig. 6c, the proportion of Ru⁰ and Ru^δ+ in Ru 3 d spectra exhibited a complementary trend. Moreover, the dynamic change in ratio of I_Ru0 to I_Ruδ+ showed a similar wave-like trend (Fig. 6d), as well as did in ratio of I_Oxides to I_Si-O (Fig. 6d and Supplementary Fig. S50). This result revealed the metallic Ru⁰ (3d_5/2) species were produced from Ru^δ+ (3d_5/2) reduction through electron transfer from Oxides species of O and In during illumination. And aforementioned thermodynamics equilibrium was further affirmed by dynamic changes in the proportions among different chemical species after “off” light (Fig. 6c, d and Supplementary Fig. S50). In addition, the above trend unveiled that the kinetic rate of electrons filling into Ru^δ+ orbital to form metallic Ru⁰ is much faster than the electron consumption rate of Ru⁰ for CO₂ and intermediates activation. This resulted in more Ru⁰(-O) species formation (ratio of I_Ru0 to I_Ruδ+ >1) (S-I in Fig. 6d) with excess electrons accumulating around Ru-O sites. Subsequently, as more CO₂ molecule orbitals hybridized with Ru⁰(-O) orbitals, the abundant electrons in Ru⁰-O sites were consumed, resulting in Ru⁰(-O) oxidization to Ru^δ+(-O). At this stage, the kinetic rate of electrons refilling into Ru^δ+(-O) to form Ru⁰(-O) could not keep up with the consumption rate of Ru^δ+(-O), causing the ratio of I_Ru0 to I_Ruδ+ back to ~1:1 (S-II in Fig. 6d). This is because Ru^δ+-O served as the active sites for C-C coupling process, and other intermediates undergo a slow kinetic protonation rate³⁶, as a result, many Ru^δ+-O sites would not be released timely and not further be reduced to Ru⁰(-O) species by nearby electrons, causing a significant accumulation of electrons around these sites again. Once these electrons refilled into the Ru^δ+(-O) orbital after releasing product and intermediates, more Ru⁰(-O) species were newly formed (ratio of I_Ru0 to I_Ruδ+ >1) (S-III in Fig. 6d). Interestingly, this dynamic transformation between Ru⁰(-O) and Ru^δ+(-O) pairs could be indirectly confirmed by dynamic changes of in-plane Ru-O (B_2g) mode in in situ Raman spectra (Fig. 6e, f). The time-resolved fluctuations of B_2g intensity suggests a dynamic regeneration of Ru^δ+-O and Ru⁰-O species. Considering such intimate dependence on active Ru^δ+-O/Ru⁰-O pairs during photocatalytic CO₂RR, the ratio of I_O-C=O and I_C-C initially decreased at S-I stage, subsequently plateaued at S-II stage due to a deficiency of Ru⁰-O species, and then continuously declined as more Ru⁰-O species were newly formed at S-III stage (Fig. 6d and Supplementary Fig. S51). To sum up, above results verified an intimately dynamic reconstruction between Ru^δ+-O and Ru⁰-O species during photocatalytic CO₂ activation, intermediates regulating, and even C-C coupling process.

Taken together, dynamic reconstruction of Ru^δ+-O and Ru⁰-O pairs was schematically illustrated in Fig. 6g. A majority of electrons from O transfer to Ru^δ+ after absorbing photons’ energy, gradually reducing it to the metallic state of Ru⁰(-O) (opening Ru-O bond). Later, the electrons in metallic Ru⁰(-O) species further interact with the absorbed CO₂, and then Ru⁰(-O) returned to Ru^δ+(-O), resulting in CO₂ activation and intermediates regulating. The associated Ru^δ+(-O) species further worked for the following C-C coupling steps. At last, the Ru^δ+ re-bonded with O by releasing products (close Ru-O bond). It is worth noting that the above discussion merely demonstrated a dynamic shuttle process of Ru⁰-O/Ru^δ+-O pairs during photocatalytic CO₂RR. That did not mean that the whole photocatalysis process would take 45 min. Because the electrons excitation and transfer among these species are extremely rapid³⁷, while, the reduction process of oxides species for intermediates activation and protonation would undergo a slow kinetics, resulting in electrons accumulating around the active sites^18,36,38.

This alternant redox between Ru^δ+ and Ru⁰ was also confirmed by cyclic voltammetry (CV). From Supplementary Fig. S52, an obvious broad peak of Ru⁰ (pink area) derived from Ru^δ+ could be observed towards the negative scanning under both dark and illumination without CO₂, as well as Ru⁰ oxidizing to Ru^δ+ (cyan area) towards the positive scanning. Moreover, a new reduction peak came from CO₂ activation ahead of Ru^δ+→Ru⁰ peak within the negative scanning range in CO₂-saturated electrolyte. Due to the Ru^δ+-O/Ru⁰-O pairs sites in Ru_xIn_2-xO₃ nanocrystal, an obvious peak of CO₂ activation was observed in the Bode phase plot (Supplementary Fig. S52d) compared with pure In₂O₃.

Photocatalytic reaction pathway

CO₂ photoreduction to ethanol is a multiple-step process with various activations and coupling of intermediates. To understand the interaction of intermediates on the catalyst surface, in situ diffuse reflectance infrared Fourier-transform spectroscopy (DRIFTS) was carried out under simulated photocatalytic CO₂RR conditions over a time scale. Overall, due to the active Ru-O site existence, the 3.0%Ru_xIn_2-xO₃/SiO₂ showed enhanced adsorption for CO₂ (Supplementary Fig. S53). Moreover, the DRIFTS spectra of 3.0%Ru_xIn_2-xO₃/SiO₂ after illumination demonstrated more abundant characteristic fingerprints compared to pure In₂O₃ (Fig. 7a, b). The bands at 1507 and 1374 cm⁻¹ were assigned to the asymmetric and symmetric OCO stretching modes of monodentate carbonates (m-CO₃²⁻) respectively, while the OCO stretching modes of bidentate carbonates (b-CO₃²⁻) appeared at 1541 and 1338 cm^{−1 22,23}. And the evidence of bands at 1222, 1433, and 1622 cm⁻¹ indicated the bicarbonate (HCO₃⁻) species formation^23,39. Moreover, the stretching modes of H₂O providing H source were detected at 1650 cm⁻¹ and a range of 3580 to 3760 cm⁻¹ (Supplementary Fig. S54), corresponding to the -OH group^22,40. The appearance of broad bands centered at 2968 and 2738 cm⁻¹ were respectively linked to the H-C and OCO stretching modes of formate species (*OCHO) as well as at 1472 cm^{−1 23,41,42}, proving the formation of formic acid precursor. As a precursor of *CO, the relevant stretching mode of *COOH could be observed at 1258, 1561, and 1716 cm^{−1 43,44}. Most importantly, much remarkable stretching modes of *CO-Ru^δ+ were detected at 2090 and 2047 cm⁻¹, corresponding to the linear *CO species in low-frequency and high-frequency range, respectively, as well as bridged *CO at 1945 and 1843 cm^{−1 9,34,45}. Due to the high coverage of surface-absorbed *CO species at the Ru^δ+-O sites, it provided a high probability for the formation of *COCO/*COCHO dimer and *CHO precursor of *COCHO. The high coverage of *CO at the Ru^δ+-O sites and further hydrogenation to be *CHO were also proven by the direct activation of CO gas (Supplementary Fig. S55). The related *CHO and *COCHO stretching modes were observed at 1748 cm⁻¹ and 1574 cm⁻¹ as well, respectively^{34,40,46,47,48}. The efficient activation for CO resulted from a strong interaction of the atomically dispersed Ru sites with CO by molecular orbital hybridization (Supplementary Figs. S56 and 57). Because of abundant monomer and dimer formation upon active sites, relevant stretching modes of ethanol product⁴⁴ were also observed in Supplementary Fig. S54.

Fig. 7: Hypothesized mechanism of photocatalytic CO₂RR.

In situ DRIFTS spectra of a 3.0%Ru_xIn_2-xO₃/SiO₂ and b In₂O₃ catalysts at CO₂ + H₂O_(g) atmosphere under dark and time-resolved illumination for 105 min (with interval of 15 min); c corresponding time-resolved dynamic absorbance profile of *CO_linear at 2047 cm⁻¹, *CO_bridged at 1843 cm⁻¹, *CHO at 1748 cm⁻¹ and *COCHO species in In situ DRIFTS spectra over 3.0%Ru_xIn_2-xO₃/SiO_2; d In situ Raman spectra of 3.0%Ru_xIn_2-xO₃/SiO₂ (i) and In₂O₃ (ii) catalysts at CO₂ + H₂O_(g) atmosphere under dark and time-resolved illumination for 105 min (with interval of 15 min); e free energy diagram of CO₂RR to ethanol by *COOH pathway on the (222) surface of Ru SAs sites of Ru_xIn_2-xO₃ and In₂O₃ (insert: illustration of energy barrier of *CO-*CO, *CO-*CHO and *CHO-*CHO coupling); f hypothesized mechanism illustration of photocatalytic CO₂RR towards ethanol production on the surface of Ru SAs sites.

To verify the hypothesized relationship between intermediates and dynamic reconstruction of Ru^δ+-O/Ru⁰-O pairs in the in situ XPS section, the time-resolved dynamic absorbance of several key intermediates (*CO_linear, *CO_bridged, *CHO and *COCHO) during photocatalytic CO₂RR was collected in Fig. 7c. As abundant Ru⁰(-O) species were formed from Ru^δ+(-O) by photoinduced electrons, consequently, metallic Ru⁰(-O) boosted the CO₂ activation as well as efficient intermediates regulating by Ru^δ+(-O) (Ru⁰→Ru^δ+), all above species increased obviously at the first stage (S-I). Because of a low reduction kinetics of Ru^δ+(-O), insufficient regeneration of Ru⁰(-O) hampered efficient activation of adequate *CO₂ species, leading to a declined production rate of *CO and *CHO species at the second stage (S-II). However, *COCHO species kept a good increase as more Ru^δ+-O species were generated and occupied. This result indicated that the metallic state Ru⁰(-O) was favorable for CO₂ activation, while oxide state Ru^δ+(-O) could boot intermediates regulating and C-C coupling, which is consistent with the case of Cu⁰-Cu^I pairs^18,24. Subsequently, as more Ru⁰-O species were regenerated in the S-III stage, the formation rate of all C₁ intermediates increased again, while the *COCHO species decreased due to the decline of Ru^δ+-O species.

Above powerful adsorption of *CO species (including *CO_linear at 2091 cm⁻¹ and CO_bridged at 1845 cm⁻¹)⁴⁵ was also confirmed by in situ Raman spectra (Fig. 7d). Similar time-resolved dynamic evolution of *CO_linear species was observed as well (Supplementary Fig. S58). Interestingly, this strong adsorption of *CO species would diminish after removing the light source (Fig. 7d and Supplementary Fig. S58), which suggests that the primary contribution to *CO species regulating comes from the Ru⁰(-O) species driven by illumination. To highlight the beauty of Ru incorporation in enhancing adsorption and achieving high coverage of *CO for C₂ dimer formation, other metals (such as Co and Ni) were investigated in photocatalytic CO₂RR (Supplementary Fig. S59).

Taking into consideration of both in situ DRIFTs and Raman spectra analysis, key intermediates’ dynamic evolution was in good agreement with the dynamic reconstruction of Ru^δ+-O/Ru⁰-O pairs in in situ XPS results (Fig. 6c), further proving the dynamic reconstruction of Ru^δ+(-O)/Ru⁰(-O) pairs played a paramount role in CO₂ activation, intermediates regulating, and C-C coupling.

To further understand the pathway to C₁ and C₂ chemicals during photocatalytic CO₂RR process, various possible mechanisms were studied using DFT calculations on (222) facet with Ru SAs exposure. The results displayed in Fig. 7e and Supplementary Tables S6 and S7 revealed that the Ru SAs site (0.15 eV) exhibited 0.06 eV lower free energy (∆G) of adsorbed *CO₂ compared to In₂O₃ (0.21 eV) as well as *OCOH (Supplementary Fig. S60), which demonstrates a thermodynamically favorable CO₂ activation and intermediates regulating by Ru SAs. As the key precursor of *CO, the Ru SAs site shows exothermic adsorption (-0.63 eV) for *COOH in contrast with pure In₂O₃ (1.04 eV). Because of the strong interaction between CO and the Ru SAs site, further reduction of *COOH to form *CO is favored with −0.22 eV (Supplementary Fig. S57). This results in favorable coverage of *CO and intermediates (e.g., *CHO) production, which further facilitates C₂ dimer production. This is in line with the strengthened stretching modes of *CO and *CHO species in the DRIFTS spectra (Fig. 7a). To investigate the lowest free energy pathway for C-C coupling, three possible C-C coupled intermediates were investigated using DFT calculations, i.e., *CO-*CO, *CO-*CHO and *CHO-*CHO coupling. We found the ∆G of *CO-*CO (1.35 eV) and *CHO-*CHO (1.55 eV) is higher than that of *CO-*CHO (1.30 eV) (Supplementary Table S6), indicating that the *CHO precedence formation provides a favorable path for C-C coupling of *CO and *CHO intermediates. A peer summary of possible C-C couping mechanism during photocatalytic CO₂RR was discussed in Supplementary Table S5. Due to C-C coupling step involving a high energy barrier during C₂ dimer production, more energy from the solar in the Ru_xIn_2-xO₃/SiO₂ core-shell structure could be utilized to overcome this barrier. That was the reason why Ru_xIn_2-xO₃/SiO₂ photocatalyst showed enhanced ethanol productivity under illumination. Based on the above discussion, Supplementary Fig. S61 displayed the full mechanistic study for the reduction of CO₂ to ethanol over Ru SAs system.

Discussion

In this work, a series of active Ru_xIn_2-xO₃ photocatalysts with atomically dispersed Ru atoms were synthesized. Diverse techniques unveiled that unsaturated coordination Ru^δ+-O sites played a crucial role in the photocatalytic CO₂RR process. The Ru_xIn_2-xO₃/SiO₂ core-shell composites showed a higher reaction rate and selectivity for photothermal catalytic CO₂RR into ethanol compared to the single Ru_xIn_2-xO₃. In situ XPS, DRIFTS, and Raman revealed that the metallic state of Ru⁰(-O) showed thermodynamically favorable activation of CO₂, resulting in high coverage and strong interaction of *CO species on the Ru SAs site, while the oxide state of Ru^δ+(-O) boosted intermediates regulating and asymmetric C-C coupling process. The intimately dynamic reconstruction of Ru^δ+(-O)/Ru⁰(-O) pairs jointly completed the multi-step C₂ chemical production from CO₂. The DFT calculations further confirmed the *COCHO intermediate was the key dimer for ethanol production, and it would be formed through a precedence ladder of *CHO coupling with *CO instead of dual *CO or dual *CHO. This work provides a comprehensive insight into the dynamic construction of active sites towards intermediates regulating during the CO₂RR process at the atomic level.

Methods

Synthesis of materials

Synthesis of SiO₂

The silica nanosphere was synthesized by sol-gel method. In brief, firstly, a volume of 81 mL absolute alcohol was mixed with 24 mL NH₃·H₂O with stirring. And then, a volume of 4.2 mL TEOS solution was added into the above solution drop by drop, and stirring for 1 h to get white suspension. Finally, the prepared suspension was filtered several times, washed with deionized water, and dried in a vacuum oven overnight. The prepared white silica nanosphere was further annealed at 300 °C for 2 h to remove the possible impurity.

Synthesis of photocatalysts

A series of Ru_xIn_2-xO₃ photocatalysts were prepared through a simple solvothermal method. Typically, 0.6 g of In(NO₃)₃·xH₂O was dissolved in a volume of 34 mL anhydrous DMF solution with vigorous stirring for 1 h. And then, different masses of RuCl₃·xH₂O (3.0 mg, 6.0 mg, 18.5 mg, 31.5 mg) were added into the above solution for another 1 h stirring and 30 min ultrasonic treatment to synthesize a series of different Ru single atoms loading catalysts (Ru_xIn_2-xO₃, where x means weight proportion of Ru, x = 0%, 0.5%, 1.0%, 3.0%, and 5.0%). After that, the solution was transferred into the Teflon-lined stainless autoclave and heated at 150 °C for 24 h. After cooling down to room temperature, the catalyst samples were collected by filtration and washed with absolute alcohol and warm deionized water several times and then dried in a vacuum overnight. Finally, the dried catalyst samples were further annealed at 250 °C for 1 h with argon flow of 40 sccm, aiming to remove possible impurity. Other metals modification is similar as above synthesis procedure otherwise different metal precursor (Ni(NO₃)₂·6H₂O and Co(NO₃)₂·6H₂O). The pure In₂O₃ (x = 0%) catalyst was synthesized followed by the above procedure without any RuCl₃·xH₂O adding.

The Ru_xIn_2-xO₃/SiO₂ composite catalyst was synthesized followed by the above procedure. After got the homogeneous mixture of In³⁺ and Ru³⁺ solution, 90 mg SiO₂ nanosphere was added into above solution for 30 min stirring and another 30 min ultrasonic treatment, respectively. And then, the mixture solution would undergo the solvothermal synthesis method and annealed treatment as above.

Characterization

The crystallographic phases were analyzed using a D8 Advanced XRD (USA) with Cu Kα X-ray radiation (λ = 1.54056 Å). The morphology was measured by a field-emission scanning electron microscope (SEM, Hitachi S-4800, Japan) and transmission electron microscope (TEM, JEOL JEM-2100F, Japan). High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were performed on a JEOL JEM-2100F field-emission electron microscope working at 200 kV. The ex-situ Raman spectra were obtained using a DXR3xi confocal Raman spectrometer (USA) with a 532 nm excitation laser. EPR spectra were recorded on an EPR spectrometer (Bruker EMX plus, USA). Surface morphology and potential were measured by the Kelvin probe force microscopy (KPFM, SPA-400, Japan), the surface potential was mapped by sample routing at zero tip bias, and dark and illumination conditions were maintained for 5 min before performing to achieve the equilibrium of carriers transport. XPS measurements of specific chemical states and elemental components were conducted using an ESCALAB 250Xi spectrometer system (Thermo Fisher, USA) with a monochromatic Al Kα source. Binding energies were calibrated to the C 1 s peak at 284.8 eV. Nitrogen adsorption-desorption isotherm was tested by Micromeritics (ASAP 2460, USA), surface area, and porosity analyzer at 77 K. The specific surface areas were calculated by the Brunauer-Emmett-Teller method. UV–vis diffuse reflectance spectra were recorded by SHIMADZU UV–vis spectrophotometer in the range of 300~900 nm. Time-resolved photoluminescence spectrum was measured on a transient fluorescence spectrometer (FLS-1000, Edinburgh) with an excitation and emission wavelength at 360 nm and 450 nm, respectively. The content of Ru in Ru_xIn_2-xO₃ catalysts was measured by an ICP-MS instrument (Agilent 7700 series, USA).

CO₂/NH₃/H₂ temperature-programmed desorption

CO₂ and NH₃ temperature-programmed desorption (CO₂/NH₃-TPD) measurements were performed using an AutoChem II 2920 chemical gas adsorption analyzer (USA). Prior to analysis, the catalyst was pretreated with helium (He) at 300 °C for 1 h (ramp: 10 °C/min) to remove adsorbed CO₂, followed by cooling to 50 °C. For CO₂-TPD, a 10% CO₂/He gas mixture (or NH₃/He for NH₃-TPD) was introduced at 30–50 mL/min for 1 h to ensure saturation. The system was then purged with pure He and heated to 600 °C (ramp: 10 °C/min), with desorbed gases detected using a TCD-equipped GC.

The H₂-TPR measurement is similar to the CO₂/NH₃-TPD measurements; otherwise, H₂ flow replacement.

XAFS measurement

The X-ray absorption fine structure (XAFS) spectra were measured with beamline BL08U1A in Shanghai Synchrotron Radiation Facility (BSRF 1W1B, Shanghai, China) by using Si (111) double-crystal monochromator, in which storage rings of BSRF were operated at 2.5 GeV with a maximum current of 250 mA. The data collection was performed in transmission mode using an ionization chamber for metal foil and metal oxides as references, and in fluorescence excitation mode using a Lytle detector for SACs, respectively. All spectra were collected under ambient conditions.

In situ XPS measurement

In situ XPS spectra of Ru_xIn_2-xO₃/SiO₂ catalyst were measured by an X-ray photoelectron spectrometer (Thermo Fisher, ESCALAB 250Xi, USA, Alpha equipped with a monochromatic Al Kα source), and all the raw data were calibrated by C 1 s at 284.8 eV. To investigate the dynamic change of Ru 3 d, O 1 s, and C 1 s, this experiment would be conducted under dark and time-resolved illumination (300 W Xe lamp: full spectrum) with CO₂/H₂ atmosphere. Typically, prior to performing the test, the catalyst would be pretreated with He gas flow at 100 °C for 1 h to remove the impurity absorbed on the surface, and then changed the atmosphere to be CO₂/H₂ till at 100 °C for another 1 h to achieve saturated adsorption. After that, firstly, the XPS spectra of different elements would be recorded in the dark at room temperature. And then, time-resolved XPS spectra would be recorded under illumination for 60 min with an interval of 15 min. At last, the spectra after removing the light source were recorded immediately again.

In situ DRIFTS measurement

The in situ DRIFT spectra were recorded by a Fourier-transform infrared spectroscopy (Thermo Fisher: Nicolet iS50, USA) with PerkinElmer Frontier in the range of 4000~600 cm⁻¹, aiming to investigate the key intermediates during CO₂ photoreduction process over pure In₂O₃ and Ru_xIn_2-xO₃/SiO₂ catalysts under dark and time-resolved illumination (300 W Xe lamp: full spectrum) reaction condition. Typically, before carrying out the measurement, the catalyst was pretreated with He gas flow at 100 °C for 1 h to remove absorbed impurities. Subsequently, the background spectrum was recorded with a resolution of 4 cm⁻¹ at 100 °C in He flow. And then, a gas mixture of CO₂/H₂O_(gas)/He with a flow rate of 40 sccm (30 sccm of CO₂/H₂O (g), 10 sccm He, respectively) was aerated into the reaction system for 40 min at 100 °C under dark. First, the in situ DRIFT spectrum of catalyst was recorded under dark, and then, the time-resolved DRIFT spectra was recorded under illumination for 105 min with an interval of 15 min in a resolution of 4 cm⁻¹. The in situ DRIFT spectra of CO adsorption were measured following the CO₂ measurement procedure.

In situ Raman measurement

The in situ Raman spectra were measured by a microscope confocal Raman spectrometer (Lab RAM HR Evolution Raman microscope, France) with an excitation of 532 nm laser. Simulated gas-liquid-solid phases condition as photocatalytic CO₂RR was carried out. Prior to test, the catalysts samples were pretreated by He gas flow at 80 °C for 1 h to remove potential absorbed impurity. Keep at this condition, the background was recorded at a step rate of 4 cm⁻¹. Next, the high-purity CO₂ was aerated into system for 40 min to be saturated under dark. The spectra after absorbed CO₂ would be record under dark first. And then, time-resolved Raman spectra would be recorded under illumination (300 W Xe lamp: full spectrum) for 105 min with an interval of 15 min. At last, the spectra after removed light source for 15 min later were recorded again.

Ultrafast transient fs-TA spectroscopy measurement

The femtosecond transient absorption (fs-TA) spectra were measured using optical pump-probe spectroscopy with Femto-TA100 spectrometer (Time-Tech Spectra). The fs radiation source was an 800 nm output pulse from a regenerative amplified Ti: sapphire laser system, delivering 35 fs pulse at 6 mJ and 1 KHz. The 800 nm pump pulse was split into two beams via a 50% beam splitter. The transmitted beam was directed to a TOPAS Optical Parametric Amplifier to generate tunable pump pulses (250 nm to 2.5 μm), while 10% of the reflected beam was attenuated by a neutral-density filter and focused through a 2 mm CaF₂ window to generate a white light continuum (340–800 nm) for the probe. This probe beam was focused onto the catalysts using an aluminum parabolic reflector, collimated after interaction, and passed into a fiber-coupled spectrometer with CMOS sensors operating at 1 KHz. Pump pulse intensity was controlled by a variable neutral-density filter wheel, and the pump-probe temporal delay was precisely regulated by a motorized delay stage. A 500 Hz mechanical chopper modulated the pump pulse, and changes in absorbance were calculated by comparing the probe pulses with and without the pump pulse. All measurements were carried out at ambient condition, and the catalysts were excited by a 400 nm pump pulse. The induced absorption change (∆A), as a function of both wavelength and temporal profile, was recorded to analyze the dynamic of the exited states.

Photocatalytic performance evaluation

CO₂ Photoreduction experiment was carried out in a 70 mL high-pressure reactor with a quartz window in the center area on the top (Supplementary Fig. S62). The light source was provided by a 300 W Xe lamp (Excelitas Tech., United States, Light intensity: 1 W/cm²). The lamp was placed on the top of the high-pressure reactor with a distance of 15 cm between them. In a typical experimental procedure, 20 mg catalyst was dispersed into 10 mL deionized H₂O. The reactor containing catalyst suspension was treated under ultrasound for 10 min. Prior to sealing, high-purity CO₂ (99.999%, Air liquid) was aerated into suspension for 40 min under stirring. And then, the reactor was sealed and high-purity CO₂ was purged into the reactor up to expected reaction pressure (1 bar, 2 bar, 4 bar, and 6 bar) and evacuated 5 times by turns to remove air, at last maintaining the high pressure for 20 min to stabilize. Subsequently, the reactor was conducted under illumination for 3 h. Before analyzing the products, the reactor was cooled down at a low temperature (4 °C). The gas products were quantified by a Gas Chromatograph (GC, Clarus 590, PerkinElmer) equipped with a TCD (5 A molecular sieve and a Carboxen 1000 packed column) and an FID detector. And the liquid products were quantified by high-performance liquid chromatography (HPLC, Thermo Fisher ICS-6000 system) with a refractive index detector and an Aminex HPX-87 H column (300 × 7.8 mm, Bio-Rad). The sulfuric acid solution (5 mM) was used as the mobile phase at a flow rate of 0.6 mL/min. The liquid products were detected by the ¹H-NMR (Bruker Advance III 400 MHz spectrometer with a BBFO probe, USA) and GC-MS (Agilent 5973, USA). In the quantification of gas products, it was noted that traces C₂H₄ and C₂H₆ were ignored. Because of the manual injection into the GC, the O₂ in the reactor was just qualified and not quantified due to the inevitable exposure to air when the gas products were taken by the syringe. The ethanol selectivity was calculated by the equation as follows:

$$ {Selectivity}\;\left(\%\right)=\\ \frac{{mole}\;C{H}_{3}C{H}_{2}{OH}*12({e}^{-})}{\Sigma ({mole\; prpducts}\;\left({H}_{2},\,C{H}_{4},{CO},{HCOOH},\,C{H}_{3}C{H}_{2}{OH}\right)*{numbers\; of\; electrons})}\times 100\%$$

(1)

*CO as key intermediate exploration experiment was similarly followed as CO₂ photoreduction procedure. Differently, after sealing the reactor, high-purity CO₂ was purged into the reactor and kept at 2 bar pressure, following extra high-purity CO compensation (99.99%, Air liquid) was added up to the expected pressure (for example total pressure (4 bar) = CO₂(2 bar) + CO(2 bar)). All of the error bars were calculated based on the duplicated experiments.

A 3-round recycle experiment was performed over 3.0%Ru_xIn_2-xO₃/SiO₂ catalyst under full -spectrum light illumination for 3 h at a CO₂ pressure of 6 bar condition.

The isotopic tracing experiment was carried out by using ¹³CO₂ (99% atom% ¹³C, Sigma) as the carbon source. The experimental procedure and condition were followed as CO₂ photoreduction reaction by replacing normal CO₂. The mass spectra of ethanol were measured by GC-MS (Agilent 5975 GC-MS, USA, source type: electron ionization (EI)).

Computational details

DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP) with spin polarization⁴⁹, in conjunction with the Atomic Simulation Environment⁵⁰, which was used to set up the VASP calculations. The generalized gradient approximation exchange-correlation functional parametrized by Perdew, Burke, and Ernzerhof for the electronic interactions⁵¹. The cutoff energy for the plane-wave-basis set was 500 eV, and the projector-augmented wave method for the core electrons was used⁵². The convergence criterion for the electronic self-consistent iteration was set at 10⁻⁶ eV. During the relaxation, we assumed the relaxation was achieved when the atomic forces were lower than 0.05 eV/Å. We used a Monkhorst−Pack grid with dimensions of 4 × 4 × 1 for sampling the first Brillouin zones⁵³. The DOS calculations were done using a 12 × 12 × 12 k-points grid, and the tetrahedron method with Blöchl corrections⁵⁴.

Because of the same miller indices of (111) and (222) facets, the In₂O₃ surface was modeled using 2 × 2 supercells of the (111) facets with four layers, for which only the bottom two layers were frozen, while the rest of the system was allowed to relax. The In₂O₃(111) is doped with Ru by substituting one In atom with Ru, which corresponds to a 3% Ru doping concentration in Ru-doped In₂O₃ surface. To ensure that the interactions between neighboring periodic images are negligible, a vacuum region along the z-direction has been added so that the distance between the two nearest surface atoms in neighboring images is at least 18 Å. We employed the DFT-D3 Grimme method for long-range dispersion interaction correction⁵⁵.

The computational hydrogen electrode model proposed by Nørskov et al. was used to calculate the free energies of CO₂ reduction intermediates, based on which the free energy of an adsorbed species is defined as: $\Delta {G}_{{ads}}=\Delta {E}_{{ads}}+\Delta {E}_{{ZPE}}-T\Delta {S}_{{ads}}+\int {C}_{p}{dT}$ (2), where $\Delta {E}_{{ads}}$ is the electronic adsorption energy, $\Delta {E}_{{ZPE}}$ and $T\Delta {S}_{{ads}}$ represent zero-point energy and entropy (difference between adsorbed and gaseous species), respectively, $\int {C}_{p}{dT}$ is the enthalpy correction and T is at room temperature⁵⁶.

Bader charge population analysis was used to analyze the charge population and electron transfer⁵⁷. The Bader charge density difference was analyzed with the VESTA program⁵⁸.