Single Ce atom-assisted oxygen vacancies as active and stable sites for CO₂ conversion

Abstract

Oxygen vacancies are often considered the active sites for CO₂ conversion, but their engineering to enhance activity and stability has been scarcely investigated, especially at high reaction tempeture (≥600 °C) or under severe stressing conditions (i.e., in presence of H₂S). We demonstrate that oxygen vacancies in defective MgO nanocrystals (less than 20 nanometres) can be significantly promoted by the addition of atomically dispersed Ce atoms anchored at these defective MgO sites. These catalysts combine high performance in CO₂ conversion with exceptional stability even in the presence of 300 ppm H₂S. For a loading <2 wt%, the Ce atoms are exclusively present as isolated species coordinated to oxygen-defective sites of MgO nanocrystals. These Ce single atoms promote the formation of these defects and facilitate the adsorption of CO₂, enhancing the activation of H₂ molecules, but do not play a role alone as single-atom catalysts. Extensive mechanistic studies, chemical kinetics and theoretical modelling prove that the single-atom-assisted oxygen vacancies (SA-O_v) enables the high catalytic performance and stability, thereby opening new avenues for catalyst design.

Oxygen vacancies are key active sites for CO₂ conversion, but enhancing their activity and stability under harsh conditions is rarely explored. Here, the authors show that adding atomically dispersed Ce to defective MgO greatly increases vacancy density, leading to superior performance in reverse-water-gas-shift reaction, even with 300 ppm H₂S present.

Introduction

The development of novel catalysts for CO₂ conversion that exhibit stable behaviour at high temperatures and in the presence of typical poisoning elements, such as S-compounds present in real emissions^1,2, remains a challenge to enable carbon circularity^3–5 and foster the recirculation of industrial CO₂ emissions^6–8.

Defects such as oxygen vacancies and step edges on metal oxide surfaces are often described as the most active sites in CO₂ conversion^9–12. Understanding the number and distribution of oxygen vacancies, particularly their potential modification by doping with transition metal single atoms, is crucial for enhancing catalyst design and optimising structure-reactivity relationships. Previous studies reported that atomically dispersed Cu on In₂O₃ seeds abundant oxygen vacancies to accelerate photothermal CO₂ conversion¹³, while Cr³⁺ doping into CeO₂ likewise enriches vacancy concentration and delivers exceptional low-temperature methanation activity¹⁴. These findings clearly demonstrated that the single atoms increase the number of vacancies, thereby promoting catalytic performance. However, the reaction usually needs to be carried out at high temperature to achieve high CO₂ catalytic efficiency, and industrially relevant reactions also contain a large amount of poisoning elements, which can decrease the stability of single atoms and erode the crystal structure rich in oxygen vacancies, and lead to a decrease in durability for the RWGS reaction^15,16. Therefore, it is essential to strike a well-balance between activating more active vacancies and maintaining the stable crystal structure.

To address this issue, we report here on a concept of catalysis using Ce single-atom-assisted oxygen vacancies, supplanting the CeO₂ nanoparticles. These rare-earth Ce atoms, as a promoter, can precisely tailor both morphology and electronic structure^17,18, markedly expanding the number of active sites^19,20. Therefore, cerium, acting as a dopant, holds great potential for regulating and activating the activity and stability of catalysts. We demonstrate, using the reverse-water-gas-shift (RWGS) reaction, that it serves as a model example while also being industrially relevant and applicable. The starting material is MgO, being a low-cost and not critical (earth-abundant) raw material. It is a typical CO₂ adsorbent and typically forms stable carbonate species^21,22. When used in catalysts, it acts only as a support or as a promoter for reactive catalytic nanoparticles such as Ni^23,24 or others^25,26.

Defective MgO nanocrystals can be prepared with enhanced activity in CO₂ conversion; furthermore, by inserting Ce-isolated atoms at defective MgO sites, it is possible to promote their behaviour significantly. These Ce single atoms act as modifiers and stabilisers to enhance the number of coordinatively unsaturated sites (corners and steps) on the MgO surface, favouring the formation of oxygen vacancies on MgO. These catalysts exhibit superior performance, demonstrating stability for over 1000 h of time on stream, with no deactivation even under a gas environment containing an H₂S impurity of up to 300 ppm. Extensive physicochemical characterisation, combined with reactivity and kinetic studies, proved these indications and the nature of the single-atom-assisted oxygen vacancies (SA-O_v). The study demonstrates that SA-O_v represents a novel concept in catalysis, offering new perspectives with industrial relevance.

Results

Decoration of MgO defect sites with Ce single atoms

For this study, we prepared a highly defective MgO (designated as MgO-D), which was further modified by the addition of Ce with different mass loadings (Supplementary Material, Table S1; the number within the brackets indicates the Ce amount). These samples were compared to a commercial MgO sample labelled MgO-C. Powder X-ray diffraction (XRD) data indicate a face-centred cubic (FCC) structure for MgO without additional diffraction peaks (Fig. S1, Supplementary Note 1). No crystalline phases were detected for Ce, indicating a high degree of dispersion.

Ce-MgO-D(1.6) shows by transmission electron microscopy (TEM) a roughened surface rich in steps and corners (Fig. S2), remarking the presence of highly coordinatively unsaturated sites compared to MgO-C (Fig. S3) and MgO-D without Ce addition, e.g., prepared with the same citrate methods without adding the Ce precursor (Fig. S4). The average MgO particle sizes of the Ce-MgO-D with variable Ce contents (0.4 wt%–4 wt%) were maintained in a range of 9.3–16.4 nm (Figs. S5–S7). The atomic resolution scanning transmission electron microscope-second electron (STEM-SE) image of Ce-MgO-D(1.6) clearly shows numerous surface steps and corners on the surface of MgO nanoparticles (Figs. 1a, S8).

Fig. 1. Identification of the chemical structure for Ce-MgO-D(1.6).

a Atomic resolution STEM-SE image; b Atomic resolution STEM-ADF image and the line profiles of Ce-MgO-D(1.6) sample. The blue dotted line represents the Ce atom, which is mainly anchored at the MgO atomic step. c, d Atomic resolution STEM-ADF images of a (100) plane on Ce-MgO-D(1.6). Terraced atomic layers represent a stepped surface along the edge. e Schematically anchored Ce at the MgO atomic step. In the top view, the single-atom site is directly above the Mg column of the (100) plane. f Ce 3d XPS spectrum of Ce-MgO-D(1.6). g Ce L_III-edge XANES spectrum and h Fourier transform of k²-weighted EXAFS spectrum of Ce-MgO-D(1.6). The XANES and EXAFS spectra of Ce(NO₃)₃ and CeO₂ were obtained as references.

The atomic resolution STEM-annular dark field (STEM-ADF) image was used to confirm the dispersion of Ce on MgO. As shown in Figs. 1b, S9, the Ce species are predominantly present as single atoms on the MgO (100) surface. Additional STEM-HAADF images with different magnifications, reported in Fig. S10, indicate the absence of detectable clusters or nanoparticles. The energy-dispersive X-ray spectroscopy (EDS) mapping provides further evidence for the even distribution of elemental Ce in these Ce-MgO-D nanocrystals (Fig. S11).

Moreover, the isolated Ce single atoms on the Ce-MgO-D(1.6) catalyst fit well into the Mg columns on the MgO (100) plane. Considering that the STEM-ADF images are only projection images, the Ce single atoms on the MgO surface may be located at different substitution sites, which leaves Ce atoms in different statuses. Through line intensity analysis of the STEM-ADF image in Fig. 1b, we found the different locations of Ce single atoms on Ce-MgO-D(1.6). For Ce-MgO-D(1.6), the Ce atoms were mainly located at the step sites on the surface of MgO, while only a small amount of Ce atoms were located at the MgO terrace. The increase in the number of steps in the MgO structure of the Ce-MgO-D(1.6) catalyst is due to the substitution of the Ce atom inside the structure, which lowers the thermodynamic surface energy. These results clearly show that the Ce-MgO-D(1.6) sample has more abundant low-coordinated sites (edges, steps and corners) (Figs. 1c, d, S12 and S13), consistent with previous literature that the stepped edges or kinks of the terrace constitute the most common defects on the surface of MgO nanoparticle (modelling in Fig. 1e)²⁷.

X-ray photoelectron spectroscopy (XPS) characterisation was used to confirm the valence state of Ce species at MgO steps. The prominent peaks appearing at 885.0 eV (Ce_3d5/2) and 903.3 eV (Ce_3d3/2) can be attributed to the Ce³⁺ doublet, while the other peaks are due to the Ce⁴⁺ species (Figs. 1f, S14). The concentrations of Ce³⁺ species in Ce-MgO-D(1.6) and pure CeO₂ nanoparticles were 46% and 33% (the value of Ce³⁺/(Ce³⁺ + Ce⁴⁺) was determined by taking the ratio of the peak areas of Ce³⁺ and Ce⁴⁺ reported previously)^28,29. The majority of Ce atoms on the MgO surface exhibit a 3+ valence state (Table S2). Furthermore, we obtained the EELS spectrum of the Ce single atoms. By calculating the M₅/M₄ ratio and comparing it with the CeO₂ reference, we found that the valence state of the Ce single atoms is close to 3+ (Fig. S15). X-ray absorption spectra provided further insights into the nature of the Ce species. The Ce L_III-edge X-ray absorption near-edge structure (XANES) spectrum (Fig. 1g) of the reduced Ce-MgO-D(1.6) shows that the Ce was cationic, the white line intensity in the spectra was between that of Ce(NO₃)₃ and CeO₂, suggesting that Ce³⁺ and Ce⁴⁺ species coexist in the Ce-MgO-D(1.6) catalyst. Extended X-ray absorption fine structure (EXAFS) spectra (Fig. 1h) indicate atomically dispersed Ce without evidence of a Ce-Ce scattering path in the spectra of the Ce-MgO-D(1.6)^30,31. A Ce-O shell indicating the presence of Ce-MgO bonds was found, consistent with STEM results for the site-isolated cerium atom on the magnesium oxide surface. Above a 2 wt% Ce loading, CeO₂ aggregates were found (Fig. S7). The structural characterisation (TEM and STEM) and valence state analyses (XPS, EELS and EXAFS) thus indicate that below about 2 wt% Ce, individual oxygen-coordinated cerium atoms replace Mg atoms on the MgO atomic steps, as modelled in Fig. 1e.

Presence and nature of MgO defect sites

MgO adsorption reactivity to CO₂ is directly or indirectly related to the presence of defects, particularly of the coordinatively unsaturated oxygen sites (O_nc, n = 3, 4, 5) and oxygen vacancies. Defective sites bind CO₂ molecules more strongly than normal oxide sites and assist in their dissociation^22,27,32. A recent study by Du et al.³³ showed that the two 3-coordinated bare surface oxygen sites are the specific active sites for CO₂ adsorption. However, the role of these defects as catalytic sites was not investigated.

The oxygen defects, including unsaturated oxygen species and oxygen vacancies in MgO, have been proven to influence the reactivity abilities in CO₂ capture^27,32. Therefore, UV-visible absorption measurements were carried out to analyse these surface defect sites. Unlike the bulk excitation of MgO, nanosized MgO with a high surface area typically exhibits optical absorptions at significantly lower energies due to defects^34,35. The presence of defects, including coordinatively unsaturated surface sites, oxygen vacancies or impurities/dopants, etc., would give rise to electronic transitions upon light absorption. In our system, all of the UV-vis absorption spectra showed a broad peak at approximately 250–330 nm(Fig. 2a), and their intensity increased with an increase in the Ce content of the samples. However, the diffuse reflectance UV-vis spectra of CeO₂ show two distinct bands: a small absorption at 270 nm and a large absorption at 350 nm, which may be assigned to the charge transfer of the p^π-dπ transition between oxygen and low-coordination Ce⁴⁺ ions and hexacoordinated Ce⁴⁺ species, respectively (Fig. S16)^36,37. Therefore, it may be inferred that the absorption band centred at 270 nm for Ce-MgO-D catalysts is attributed to a combination of the surface defects (O-vacancies, low-coordinated sites) in MgO and one type of well-dispersed Ce⁴⁺ species (presumably in a tetra-coordinated environment), which can act as activation sites and donate electrons to the empty orbitals of incoming molecules such as CO₂²². Additionally, the absence of absorption peaks at ~350 nm suggests that the CeO₂ species do not exist on the MgO nanocrystal. This observation indicates that the number of defect sites from MgO increased with the increase in Ce dosage (Fig. 2b).

Fig. 2. Examination of the surface oxygen species in the Ce-MgO-D catalysts.

a UV-vis absorption spectra and b corresponding quantitative surface defective sites (corners, steps and Ce dopants) concentration. c EPR spectra of MgO-C, MgO-D and Ce-MgO-D(1.6) catalysts at room temperature and ambient conditions. Quasi in-situ d O1 s and e Mg 2p XPS spectra of MgO-C, MgO-D and Ce-MgO-D(1.6) catalysts obtained in a vacuum after reduction for 2 h in H₂ (X-ray energy, hν = 1486 eV). f Perspective view of the bond configuration of the defect sites. Yellow: Mg, Blue: Ce, red: O, black dashed circle: oxygen vacancy.

To confirm the Ce substitution in the MgO lattice and the simultaneous generation of oxygen vacancies (F_s⁺ centres), electron paramagnetic resonance (EPR) experiments were further conducted. As shown in Fig. 2c, a characteristic symmetrical EPR signal with a g value of 2.004 was observed in MgO-D and Ce-MgO-D(1.6) catalysts due to the free electron trapped in oxygen vacancies (O_v)^34,38. More than one signal can be observed at the g value of 1.979, indicating some different types of defects depending on the surrounding environment of the paramagnetic centre (Fig. 2c). Similarly, CeO₂ also exhibits weak EPR signals at 2.003 and 1.967 (Fig. S17), which are attributed to oxygen vacancies and Ce³⁺, respectively³⁹. The Ce-MgO-D(1.6) catalyst exhibited higher EPR signal intensity than that of the MgO-D due to the increased amount of oxygen vacancies associated with the Ce atom. Therefore, the Ce-dopant facilitates the formation of oxygen vacancies. However, a similar signal did not occur on MgO-C, as indicated in Fig. 2c.

Quasi-in-situ XPS after H₂ pretreatment was further used to characterise the presence and nature of defect sites. The fitted O1 s spectra of MgO-C, MgO-D and Ce-MgO-D(1.6) show three peaks, attributed to lattice oxygen (O_L, 529.0–529.6 eV), oxygen anions near the oxygen vacancies (O_v, 530.7–531.1 eV), and adsorbed oxygen species^40,41 (O_ads, such as OH and carbonate, 532.0–533.0 eV), respectively (Figs. 2d, S19a)^35,42. It should be noted that the XPS spectra of C1 s at 288.3 eV confirm the existence of carbonate (CO₃²⁻, Fig. S18)⁴³. After H₂ treatment at 500 °C, the fraction of O_v follows the order of Ce-MgO-D(1.6) (30.2%) > Ce-MgO-D(0.8) (28.6%) > MgO-D (24.1%) > MgO-C (15.6%; Figs. 2d, S19a), indicating that MgO-D has more oxygen vacancies than MgO-C and Ce introduction further enhances the number of oxygen vacancies, which further supports the EPR results and is in agreement with previous indications²³. The XPS peak of O 1 s (Fig. 2d) shifts to lower binding energies with the addition of Ce species. The peak of Mg 2p (Fig. 2e) shifts in the inverse direction. This result demonstrates efficient charge transfer from Ce to MgO as the number of Ce atoms added to MgO increases.

The coordination number of O atoms can identify the difference of defect sites in stepped Ce-MgO-D(1.6). The surface oxygen defect species on the MgO surface are further elucidated through density functional theory (DFT) calculations. All DFT calculations are performed on the close-packed MgO (100) surface, as it has a lower surface energy and is more stable than other low-index surfaces. The MgO(100) surface consists of three types of defect sites, e.g., defective stepped edges (MgO-D), Ce atom-doped defective stepped edges (Ce-MgO-D), and Ce atom-doped defective stepped edges with enriched oxygen vacancies (Ce-O_v-MgO-D). They are presented in Fig. 2f. The atoms at the defect sites exhibit low coordination, and oxygen vacancies are likely those exhibiting a higher catalytic activity.

Reactivity and stability of MgO SA-O_v

The addition of a Ce single-atom at MgO defect sites (SA-O_v) induces a change in the intrinsic reactivity of these oxygen atoms, as proven using the RWGS model test reaction. Ce-MgO-D(1.6) achieves an impressive CO₂ conversion of 52.1% and RWGS rate of 37.2 μmol·g_cat.⁻¹·s⁻¹ at 600 °C, a result significantly higher activity than MgO-C, CeO₂ nanoparticles (NPs), Ce atoms supported on the oxides or other MgO-based RWGS catalysts (Fig. 3a, b).

Fig. 3. Catalytic performance of Ce-MgO-D catalysts in CO₂ conversion.

a Catalytic performance of Ce-MgO-D catalysts. Reaction conditions: CO₂/H₂ = 1/3 (24%CO₂/72%H₂/4%N₂), GHSV = 24,000 mL·g_cat⁻¹·h⁻¹, 1 atm. The letter ‘α‘ indicates the activity evaluation of the Ce-MgO-D(1.6) catalyst under 4800 mL·g_cat⁻¹·h⁻¹. b RWGS rate of Ce-MgO-D(1.6) at 600 °C with respect to a series of reference catalysts without Ce or with Ce at similar loading. c Ce-normalised rate at 600 °C as a function of the Ce content in Ce-MgO-D samples, with an indicative transition from when only Ce single atoms are detected to when CeO₂ nanoparticles (NPs) started to be present. d Stability evaluation of the Ce-MgO-D(1.6) catalyst for the RWGS reaction. Reaction conditions: T = 600 °C, CO₂/H₂ = 1/3 (24%CO₂/72%H₂/4%N₂), GHSV = 4800 mL·g_cat⁻¹·h⁻¹, 1 atm. e Catalytic performance of Ce-MgO-D(1.6) catalyst under different H₂S concentrations (5, 50 and 300 ppm, respectively) in the reactant stream, where the CO₂ conversion is kept below 30%. f Comparison of anti-deactivation ability of Ce-MgO-D(1.6) and conventional CuZnAl catalysts in the presence of different H₂S concentrations. Reaction conditions: T = 600 °C, CO₂/H₂ = 1/3 (17.7%CO₂/53.1%H₂/29.2%N₂), [H₂S] = 5, 50, 300 ppm in reactant stream, GHSV = 60,000 mL·g_cat⁻¹·h⁻¹, 1 atm.

Additionally, the CO₂-TPSR results indicate a higher CO formation rate for Ce-MgO-D(1.6), as shown in Fig. S20, Supplementary Note 3. The Ce addition improves the activity up to a maximum Ce loading corresponding to the transition when CeO₂ NPs are also detected (Fig. 3c). When normalised to the molar amount of Ce, the catalytic activity of the Ce-MgO-D(1.6) catalyst reaches 95.0 ⅹ 10³ μmol·mol_Ce⁻¹·s⁻¹ (Fig. 3c). When normalised to the amount of catalyst mass, Ce-MgO-D(1.6) catalytic activity to CO is 84.8 μmol·g_cat.⁻¹·s⁻¹, which can be further enhanced to 102.9 μmol·g_cat.⁻¹·s⁻¹ when diluted with a 10% inert SiO₂ support (Ce-MgO-D(1.6)-I; Fig. S21, Table S3 and Supplementary Note 4). These catalytic performances rank among the highest values reported in the literature (Table S3).

Ce-MgO-D sample with an optimal Ce loading shows stable performances over 1000 h on stream near the thermodynamic equilibrium (Fig. 3d, 4800 mL·g_cat⁻¹·h⁻¹) and over 300 h (Fig. S22, 300,000 mL·g_cat⁻¹·h⁻¹) on stream away from the thermodynamic equilibrium with ~100% selectivity at 600 °C, respectively, better than a reference catalyst (i.e., conventional copper-zinc-alumina catalyst, Fig. S23). The dramatic high CO selectivity can reach as high as 98.5% when the reaction pressure increases to 20 bar, whereas the MgO-C materials decrease to 85.4% (Fig. S24, Supplementary Note 5). Moreover, significantly lower performances were detected (Fig. 3b) by supporting CeO₂ NPs (Ce_NPs-MgO-D) or other elements (such as La or Zr) on the MgO-D.

More importantly, the Ce-MgO-D(1.6) catalyst exhibited exceptional S-resistance in CO₂ conversion during accelerated ageing tests. The Ce-MgO-D(1.6) catalyst demonstrates a relatively stable activity, exhibiting an RWGS rate of 36.9 μmol·g_cat⁻¹·s⁻¹ during the initial 16 h in the absence of H₂S (Step I, Fig. 3e). Following the introduction of 5 ppm (Step II, Fig. 3e), 50 ppm (Step III) and 300 ppm (Step IV) H₂S in sequence, the RWGS rate for the Ce-MgO-D(1.6) catalyst remained unaffected. Following the removal of H₂S from the reactant flow, the rate of the Ce-MgO-D(1.6) catalyst exhibited only a slight decrease (Fig. 3f). The metal-based (i.e., CuZnAl) and Mo-carbide (MoC_x) show instead a dramatic deactivation during Step II (5 ppm H₂S was introduced) and almost complete inactivation in the presence of 300 ppm H₂S (Figs. 3f, S25 and S26). While Mo-carbide catalysts, i.e. the most active RWGS catalyst reported in the literature up to now^44,45, show stable behaviour in the absence of H₂S, it fast deactivates in the presence of traces of H₂S, typically present in the industrial emissions. In industrially relevant conditions, where the space velocities are limited by factors such as pressure drop, the activity of Ce-MgO-D(1.6) is enough to reach equilibrium at 600 °C. Thus, a higher activity is not relevant. Differently, achieving a selectivity close to 100% and especially a high stability in the presence of S-compounds is crucial from the application side.

The superior RWGS performance of the Ce-MgO-D(1.6) catalyst is related to the peculiar reaction mechanism in CO₂ conversion, particularly the role of Ce-SA-O_v in MgO nanocrystals. A relationship between the ratio of these oxygen vacancies, which depends on the amount of Ce single atoms localised at MgO defect sites (see Figs. 2, S28a), is observed. This result is confirmed by the analysis of the reaction kinetics (Fig. S27, Table S4 and Supplementary Note 6) and the calculated numbers of surface defective oxygen anions (Fig. S29, Table S5 and Supplementary Note 7), showing a dependence on the pre-exponential factor (Fig. S28b) and the numbers of surface defective oxygen anions (Fig. S29), e.g., the number of active sites. Besides, quasi-in-situ XPS results show that the oxygen vacancy content of Ce-MgO-D(1.6) catalyst after reaction was remained at a similar level with the H₂-treated sample (c.a. 30.3%, Fig. S30), thus ensuring stable performance.

Role of Ce single atoms in the reactivity of Ce-MgO-D materials

The specific surface area, calculated by the Brunauer-Emmett-Teller (BET) method, is 13 m²·g⁻¹ for MgO-C and 65 m²·g⁻¹ for MgO-D (Table S6), respectively, indicating an increase in surface area when the MgO oxide is nanosized. Further increments in the specific surface area for the Ce-MgO-D catalysts (in the range of 98 m²·g⁻¹ and 135 m²·g⁻¹) could favour the exposure of the active site and CO₂ adsorption and activation. CO₂ temperature programmed desorption (CO₂-TPD) experiments show that after CO₂ adsorption at 50 °C for 1 h, three desorption peaks (Ⅰ: weak base sites at <200 °C, Ⅱ: medium base sites at 200–400 °C and Ⅲ: strong base sites above 400 °C) are evidenced (Fig. 4a)^46,47. The weak base sites are generated by reaction with the hydroxyl groups on the MgO surface, forming bicarbonate species. The medium-bonded species are derived by a reaction of CO₂ with Mg²⁺ and/or O²⁻, forming a bidentate carbonate. Finally, the desorption peak at higher temperatures is due to a strong interaction and can be assigned to the decomposition of monodentate carbonate species. Ce-MgO-D(1.6) shows a 16.7-fold increase in the CO₂ adsorption capacity compared with MgO-C. The intensity of all three peaks increased in the Ce-MgO-D(1.6) catalyst in accordance with the increase in defect sites.

Fig. 4. CO₂ adsorption and hydrogenation.

a CO₂-TPD profiles of the MgO-C, MgO-D and Ce-MgO-D(1.6) catalysts. b In-situ DRIFTS spectra for the adsorption of CO₂ on MgO-C, MgO-D and Ce-MgO-D(1.6) catalysts. c Adsorption energy of CO₂ and corresponding adsorption geometries over MgO, MgO-D, Ce-MgO-D, O_v-MgO-D and Ce-O_v-MgO-D models of MgO (100) surface. d Energy profiles for the hydrogenation of CO₂ to CO on the MgO, MgO-D, and Ce-O_v-MgO-D models of MgO (100) surface. (1) and (1)’, CO₂ + 2H*; (2) and (2)’, HCOO* + H*-Ⅱ; (3) and (3)’, CO* + H₂O*; (4) and (4)’, CO + H₂O. The capital letters A, B and C represent the rate-determining steps (RDS) of MgO, MgO-D and Ce-O_v-MgO-D, respectively.

In-situ Diffuse Reflectance Infrared Fourier Transform Spectroscopy (in-situ DRIFTS) experiments indicate the stability of chemisorbed species. Upon exposing catalysts to 5% CO₂ at 450 °C, the transient evolution of the adsorbed carbonate species was monitored. The results are presented in Fig. 4b, with additional details provided in the Supporting Information (Figs. S31–S33). All the formed carbonate species were observed on the MgO-C, MgO-D, and Ce-MgO-D(1.6) surfaces, with peaks in the range of 1800–1100 cm⁻¹. The peak at 1670 cm⁻¹ confirmed the presence of HCO₃⁻ species⁴⁸. The infrared (IR) peaks at 1299 and 1601 cm⁻¹ are assigned to the symmetric and asymmetric C–O–C stretching modes of bidentate carbonates (b-CO₃²⁻), while the peaks at 1437 and 1519 cm⁻¹ are attributed to the symmetric and asymmetric C–O–C stretch of monodentate carbonates (m-CO₃²⁻)^49,50. Thus, all the observed HCO₃⁻ and CO₃²⁻ species via in-situ DRIFTS were well consistent with the CO₂-TPD results, indicating different strengths of surface basic oxygen sites on MgO-C, MgO-D and Ce-MgO-D(1.6) nanocrystals. Noticeably, the stronger peak intensity over MgO-D and Ce-MgO-D(1.6) indicated the enhanced CO₂ adsorption-bonding-activation ability with respect to MgO-C due to the increased defect stepped sites on the MgO-D and Ce-MgO-D(1.6) nanocrystal.

DFT calculations provide further insights. Based on the mainly exposed (100) facet of MgO (Fig. S34, Supplementary Note 2) and calculations of the oxygen vacancy formation energies (E_ov) on MgO-D and Ce-MgO-D (Fig. S35, Table S7 and Supplementary Note 8), five types of models (Fig. S36) are constructed to simulate the pristine MgO(100) surface, defective MgO(100) surface, Ce atom-doped defective MgO(100) surface, defective MgO(100) surface with enriched oxygen-vacancy, and Ce atom-doped defective MgO(100) surface with enriched oxygen-vacancy, which are named as MgO, MgO-D, Ce-MgO-D, O_v-MgO-D, and Ce-O_v-MgO-D, respectively. The electronic properties of the oxygen vacancy were studied on the Ce-O_v-MgO-D surface. As shown in Fig. S37, the degree of electron localisation at the oxygen vacancy was higher (red region), indicating the localised electrons at the oxygen vacancy may have an attractive and activating effect on CO₂ adsorption and activation (Supplementary Note 9). We therefore examined the adsorption and activation of CO₂ over five models, and the corresponding adsorption energies, adsorption configurations, and differential charge density of CO₂ are shown in Table S8, Figs. S38, S39. Over MgO, MgO-D, and Ce-MgO-D catalysts, CO₂ adsorbed in a bent configuration to form carbonates, with the corresponding adsorption energies of −0.67, −2.46, and −1.68 eV (Fig. 4c), respectively. The adsorption energies of CO₂ over O_v-MgO-D and Ce-O_v-MgO-D catalysts are −3.47 and −2.89 eV, respectively, indicating that the presence of oxygen vacancies significantly enhanced the adsorption of CO₂. The density of states (DOS) plot (Fig. S40a) shows that the introduction of oxygen vacancies makes the Fermi level shift from the top of the valence band to near the bottom of the conduction band, and a defect level corresponding to the oxygen vacancy is formed between the valence band and the conduction band. O_v-MgO-D and Ce-O_v-MgO-D have greater overlap with CO₂ below the Fermi level (Fig. S40b), indicating a stronger interaction between the O_v-MgO-D/Ce-O_v-MgO-D and CO₂.

Theoretical simulations showed that the CO₂ molecule exhibited significant activation upon adsorption on the O_v-MgO-D and Ce-O_v-MgO-D surface models, as reflected in the bending of its molecular configuration and the acquisition of an electric charge. From the geometric perspective, the CO₂ molecule in O_v-MgO-D and Ce-O_v-MgO-D had the largest angular bending (from 180.0° to 117.2° and 119.6° vs to 127.8°–132.9°). The bond lengths were extended to 1.299 Å in O_v-MgO-D and to 1.288 Å–1.294 Å in Ce-O_v-MgO-D, indicating that CO₂ was mostly activated in the presence of oxygen vacancies compared to MgO, MgO-D and Ce-MgO-D catalysts (Fig. S38). On the other hand, from an electronic perspective, CO₂ interacts with the energy levels at oxygen vacancies (Fig. S40), indicating that electrons at the defect level readily transfer to the antibonding orbitals of the CO₂ molecule. The results of differential charge density and Bader charge diagrams (Fig. S39) further confirm that the C atoms in CO₂ adsorbed at oxygen vacancies lose very few electrons (−0.84 e on O_v-MgO-D and −0.49 e on Ce-O_v-MgO-D). The total amount of electrons gained for CO₂ adsorbed at different sites follows the order of MgO-D (0.36 e) ≈ Ce-MgO-D (0.38 e) ≈ MgO (0.42 e) < Ce-O_v-MgO-D (1.32 e) < O_v-MgO-D (1.64 e). This indicates that oxygen vacancies serve as effective electron donors, transferring electrons to CO₂ and thus promoting its activation.

H₂ adsorption and dissociation were also studied theoretically. Specifically, H₂ molecules are weakly physically adsorbed on the MgO, MgO-D, and Ce-MgO-D catalysts and heterolytically dissociate at the Mg site and O site (Figs. S41, S42, Supplementary Notes 10, 11). Interestingly, H₂ molecules are spontaneously dissociated over O_v-MgO-D and Ce-O_v-MgO-D catalysts (Fig. S43). The co-adsorption energy of two H atoms is more negative on O_v-MgO-D and Ce-O_v-MgO-D (Fig. S44b). The results indicate that the presence of unsaturated, coordinatively bonded oxygen sites (O_4C) and oxygen vacancies is crucial for enhancing H₂ dissociation ability, which is confirmed by the H₂-D₂ exchange reaction (Fig. S45) and H-D isotope exchange DRIFT experiments (Fig. S46, Supplementary Note 12).

To probe the reaction mechanism, the evolution of adsorbates and surface intermediate species during CO₂ hydrogenation to CO was evaluated by in-situ DRIFTS (Figs. S47–S49). For the Ce-MgO-D(1.6) catalyst, when the CO₂/H₂ mixture was introduced into the system, various surface species, including b-CO₃²⁻ (1299 and 1601 cm⁻¹), m-CO₃²⁻ (1437 and 1519 cm⁻¹) and formate species (HCOO*, 2833, 1626, 1606, 1378 and 1340 cm⁻¹)^51–53 appear and gradually intensify with the reaction time (Figs. S49a, 50c). The peaks for gaseous CO are clearly visible at 2181 and 2107 cm⁻¹. Then, the peak intensities of carbonates and formate species decreased simultaneously when the CO₂/H₂ mixture was switched to H₂ (Fig. S50c). The simultaneous evolution of these surface species demonstrates that CO₂ adsorbed on MgO can rapidly react with active H⁺ to produce formate intermediates, which are subsequently hydrogenated to CO. The abovementioned surface species and their evolution could also be observed in the DRIFT spectra of MgO-C (Fig. S50a) and MgO-D (Fig. S50b), and the intensity of the formate peaks decreases in the order of Ce-MgO-D(1.6) > MgO-D > MgO-C (Fig. S50d), which further supports the higher reactivity of Ce-MgO-D(1.6) catalyst.

The calculated potential energy profiles offered a more comprehensive understanding of the RWGS reaction on Ce-MgO-D catalysts. Firstly, we compared the formate pathway and the direct redox pathway on the Ce-O_v-MgO-D surface and found that the formed HCOO* species from CO₂ hydrogenation has a significantly lower activation energy (0.29 vs 1.76 eV) and reaction energy (−1.32 vs 0.06 eV) than the direct redox pathway to form CO* species (Fig. S51). This indicates that the formate pathway is more favourable in terms of both kinetics and thermodynamics in the Ce-O_v-MgO-D catalyst, consistent with the DRIFTS results (Fig. S49). Furthermore, the formate pathway on different catalysts was analyzed, as shown in Fig. 4d. Compared to those over MgO and MgO-D catalysts, both the dissociation of H₂ and the formation of the HCOO* intermediate over the Ce-O_v-MgO-D catalyst are thermodynamically favourable, which facilitates the subsequent reactions over the Ce-O_v-MgO-D catalyst. However, the H* reduction of HCOO* into CO*, OH* and H* in Ce-O_v-MgO-D has to overcome a transition state barrier of 2.29 eV (TS3” in red box area C in Fig. 4d), and this step is identified as the rate-determining step (RDS) of the entire reaction. For MgO and MgO-D, the hydrogen dissociation to 2H* (TS1 in grey box area A in Fig. 4d) and the HCO* reacted with OH* to form CO* and H₂O* (TS4’ blue box area B in Fig. 4d) are the rate-determining steps, which must overcome a transition state barrier of 2.50 eV and 2.15 eV, respectively. Notably, the DFT-calculated activation barriers of the rate-determining steps are similar (see the energy profiles of the rate-determining step for (A) MgO, (B) MgO-D, and (C) Ce-O_v-MgO-D). These results strongly support the kinetic results (Table S4). The corresponding configurations of reaction pathways are displayed in Figs. S52–S54, Table S9 and Supplementary Note 13.

These theoretical results indicate that (i) in defective MgO samples, there is a significantly lower activation energy than in non-defective MgO, confirming that oxygen vacancies are crucial to obtain a high activity in MgO-based samples, and (ii) the rate-determining steps in MgO-D and Ce-O_v-MgO-D are different (TS4’ and TS3”, respectively), but the activation barrier of the rate-determining step is very similar. Thus, the difference in the reactivity should be related to the pre-exponential factor rather than to a change in the activation energy.

The extensive kinetic results (Table S4) reported in SM for conciseness proved well that the main difference between these two catalysts is related to the pre-exponential factor, which is related to the number of active sites. The data in SM also demonstrate the superior performance of these RWGS catalysts compared to those reported in the literature. The structure of the Ce-MgO-D(1.6) catalyst remains very stable (Figs. S55–57, Supplementary Note 14) after a stability test.

Discussion

The extensive characterisation data of the Ce-MgO-D samples evidence that for a loading <2 wt%, the Ce atoms are exclusively present as isolated species coordinated to oxygen-defective sites (SA-O_v) of MgO. These Ce single atoms promote the formation of these defects in Ce-MgO-D with respect to MgO-D and especially with respect to MgO-C. In-situ characterisations and theoretical calculations demonstrate that the Ce atoms facilitate the increase of unsaturated step sites and weaken the adsorption of CO₂ while simultaneously enhancing the activation of H₂ molecules. This concept enables us to achieve the superior S-resistance of these catalysts, coupled with high activity and selectivity, making them superior in RWGS under industrial-relevant conditions. The Ce-MgO-D catalysts exhibited an 8.1-fold increase in CO₂ conversion ability compared to that of commercial MgO and superior activity compared to other catalysts, with a CO selectivity of nearly 100% at 600 °C, where equilibrium conversion is reached. These superior performances are maintained for more than 1000 h on stream, even surviving in highly concentrated H₂S (up to 300 ppm) containing reactants.

Methods

Preparation of Ce-MgO-D(x) catalysts

A series of Ce-MgO-D catalysts with different Ce dosages were prepared using the citrate complex method. Taking the synthesis procedure of Ce-MgO-D(1.6) catalyst as an example: 0.12 mmol cerium nitrate hexahydrate, 24.5 mmol magnesium nitrate hexahydrate precursors and 49.3 mmol citric acid were dissolved in 20 mL deionised water and stirred at room temperature for 4 h. The mixed solution of precursors was then transferred to a rotary evaporator for rotary evaporation. After 2 h of evaporation, the resultant solution was dried overnight in an oven at 120 °C. The obtained powders were ground and calcined in a muffle furnace at 650 °C for 2 h with a ramping rate of 5 °C·min⁻¹. All of the as-prepared catalysts were denoted as Ce-MgO-D(x), where x represents the weight fraction of Ce. The molar ratio of citric acid to metals (Mg and Ce) was fixed at 2. Additionally, pure MgO and pure CeO₂ were prepared according to the procedures mentioned above.

Structure and composition of catalyst

The PXRD results were collected on an X’pert-Pro diffractometer (PAN Analytical, Netherlands) using Cu Kα radiation (λ = 1.54 Å) at 40 kV and 40 mA in the range of 10–90°. Elemental contents were measured with ICP-OES using a PerkinElmer 7300 DV analyser. The specific surface area of the materials was determined by N₂ physisorption on an ASAP 2020 plus Micromeritics instrument working at −196 °C. Before measurement, the samples were outgassed at 250 °C for 2.5 h. TEM measurements were performed on a JEM-2100 operating at an accelerating voltage of 200 kV. Atomic resolution STEM with ADF detection and field-emission scanning electron microscopy (FE-SEM) was performed on an HF5000 microscope from Hitachi, equipped with a 200 kV cold-field emission gun and a Cs-corrector for the electron probe. The EELS spectra were acquired with a Gatan GIF camera. An energy dispersion of 0.25 eV was used to acquire the Ce L-edge, with an energy resolution of approximately 1 eV. For core-loss acquisition, the pixel size was set to 2 Å × 2 Å, and the dwell time was set to 0.5 s. The EELS spectrum of CeO₂ particles was used as a reference. As a complementary technique to STEM microscopy, STEM images and element mapping images were collected using a JEM ARM200F thermal-field emission microscope with a probe spherical aberration (Cs) corrector operating at 200 kV. The samples were ultrasonicated in an ethanol solution, dropped onto the copper grid covered with a holey carbon membrane and dried for observation. Due to the weak signal of Ce single atoms, we manually selected individual spectra with a significant Ce signal and summed them to enhance the signal-to-noise ratio (SNR). The valence state of Ce was determined by calculating the M5/M4 ratio. Firstly, a power-law model was fitted to remove the background of the raw spectra. Then, a double arctan function was used to remove the continuum background in the EELS spectra. The areas of M5 and M4 were integrated within an 8 eV-wide window, and the M5/M4 ratio was calculated. The M5/M4 ratio of the Ce single atom was compared with that of the CeO₂ (IV) reference to determine the valence state. The EPR spectra were obtained using a Bruker A200 spectrometer at room temperature. The catalysts’ UV-vis absorption spectra were measured at room temperature using a PerkinElmer Lambda950 UV-vis near-infra-red spectrometer (200–800 nm, 10 nm/s scan rate).

Quasi in-situ X-ray photoelectron spectroscopy (XPS)

Quasi in-situ XPS analyses were performed on a Thermofisher Escalab 250 Xi+ photoelectron spectroscope equipped with a monochromatic Al Kα radiation (hν = 1486.6 eV). The calcined Ce-MgO-D catalysts were pressed into a small pellet (3 mm diameter) and held on the sample holder. The pellet was then reduced in the quasi in-situ XPS pretreatment chamber using an H₂ flow (40 mL·min⁻¹) at 500 °C for 2 h. After cooling to room temperature, the sample was transferred to the ultrahigh-vacuum chamber for XPS measurement without exposure to air. The data analysis was performed using CasaXPS software, and a Shirley background was then subtracted from the raw data to obtain the areas of the elements’ peaks. All binding energies were calibrated to the carbon C 1 s peak (284.8 eV) in the containment.

Temperature-programmed experiments

The CO₂-TPD experiments were performed in a fixed-bed reactor with a U-shaped quartz tube (inner diameter, 6 mm) at atmospheric pressure, with signals recorded by a QMD mass spectrometer. Firstly, a 100 mg catalyst was pretreated at 500 °C for 1 h under a flow of 99.99% He (40 mL·min⁻¹) and then cooled to 50 °C. After that, 100% CO₂ (40 mL·min⁻¹) was passed through the catalyst for 1 h during adsorption, followed by flushing with He at 30 mL·min⁻¹ for 1 h to remove physically adsorbed CO₂. Finally, the catalyst was heated to 600 °C at 10 °C·min⁻¹ under a He flow of 30 mL·min⁻¹, and the desorbed CO₂ was monitored using mass spectroscopy. The Temperature-Programmed Reaction characterisation of CO₂/H₂ was carried out on a fixed-bed reactor, and then the reactor was inspected online using a QMD mass spectrometer. 100 mg of catalyst was loaded in a U-shaped quartz tube (inner diameter 6 mm) reactor. It was pretreated in H₂ for 1 h and in He for 1 h at 500 °C, respectively. After cooling to 50 °C in He, the reactants, with a molar ratio of CO₂/H₂/N₂ of 5:15:80, were introduced at a flow rate of 40 mL·min⁻¹. The temperature of the reactor was increased from 50 °C to 800 °C at a rate of 10 °C·min⁻¹. H₂-D₂ exchange reactions were performed on a temperature-programmed micro-fixed bed reactor equipped with a mass spectrometer. Typically, 100 mg of the sample was reduced in H₂ (30 mL·min⁻¹) at 500 °C for 1 h, then the gas was switched to Ar (30 mL·min⁻¹) and treated at 500 °C for 1 h, followed by cooling down in Ar until the baseline signal in MS spectra was steady at 30 °C. At that time, a mixture of D₂ (5 mL·min⁻¹), H₂ (5 mL·min⁻¹), and Ar (20 mL·min⁻¹) was introduced into the catalyst sample and maintained for about 30 min at 30 °C. Next, the catalyst was heated to 300 °C at a rate of 10 °C·min⁻¹. A mass detector (QMD) was used to analyse the signals of m/z = 2 (H₂), 3 (HD), and 4 (D₂).

In-situ Diffuse Reflectance Infrared Fourier Transforms Spectroscopy (DRIFTS)

The in-situ DRIFTS investigations were recorded using a Bruker Vertex 80 V spectrometer equipped with a Mercury-Cadmium-Tellurium (MCT) detector in the range 850–4000 cm⁻¹ at a 32-scan integral at a resolution of 4 cm⁻¹. For CO₂ adsorption, the catalyst was sequentially pretreated with H₂ at 500 °C for 2 h, followed by 1 h in Ar. The background spectrum was obtained at 500 °C, 400 °C, 300 °C, and 200 °C, and then at room temperature (r.t., 30 °C) in an Ar flow. After cooling the catalyst down to r.t., the gas flow was changed to 5% CO₂/Ar (40 mL·min⁻¹, 0.1 MPa), and the spectra were collected simultaneously. Once the adsorption was completed (i.e., the intensity of the CO₂-adsorption bands no longer increased), the temperature was increased from 30 °C to 500 °C in 5% CO₂/Ar (40 mL·min⁻¹, 0.1 MPa), and the spectrum was continuously recorded. The reaction DRIFTS spectra and H-D isotope exchange DRIFT spectra were obtained using a similar procedure. The catalyst was pretreated with H₂ at 500 °C for 2 h, sequentially in Ar for 1 h. The reaction infra-red spectra were continuously recorded during exposure to 5% CO₂, 15% H₂, and 80% Ar (40 mL·min⁻¹, 0.1 MPa). The H-D isotope exchange DRIFT spectra were continuously recorded during exposure to D₂.

DFT calculation

DFT calculation methods are described in the Supporting Information.

Catalytic performance evaluation and kinetic experiments

The CO₂ hydrogenation reaction was carried out at ambient pressure in a quartz tube reactor (d = 8 mm) under a gas atmosphere of 24% CO₂/72% H₂/N₂ (40 mL·min⁻¹) with the GHSV of 24,000 mL·g_cat.⁻¹·h⁻¹. Before the activity test, 100 mg catalysts (160–400 µm) with 1.0 g SiC (400–450 µm) as a diluent were activated by 100% H₂ (40 mL·min⁻¹) at 500 °C for 2 h at a ramp of 5 °C·min⁻¹. After the reduction, the H₂ stream was switched to Ar until the temperature was heated to 600 °C, and the reactant gas was switched to pass through the catalyst bed. The effluent gas was analysed online by an Agilent 7890B gas chromatography (GC) instrument equipped with a thermal conductivity detector (TCD). The long-term stability of the catalyst was also measured at 600 °C. CO₂ conversion, CO selectivity and RWGS rate were calculated by using the following equations:

$$\mathrm{Conversion}\left(\mathrm{C}{\mathrm{O}}_2,\%\right)=\left(1-\frac{F_{C{O}_2,out}}{F_{C{O}_2,in}}\right)\times 100\%$$ 1

$$\mathrm{Selectivity}\left(\mathrm{CO},\%\right)=\frac{F_{CO,out}}{F_{C{O}_2,in}-F_{C{O}_2,out}}\times 100\%$$ 2

$$\mathrm{RWGS\; rate}\left(\mathrm{umol\cdot }{\mathrm{g}}_{\mathrm{cat}.}^{-1}\mathrm{\cdot }{\mathrm{s}}^{-1}\right)=\frac{F_{C{O}_2,in}-F_{C{O}_2,out}}{m_{cat.}}$$ 3

where $F_{C{O}_2,in}$ is the concentration of CO₂ in the reaction stream, and $F_{C{O}_2,out}$, $F_{CO,out}$ are the concentrations of CO₂ or CO (respectively) in the outlet stream. m_cat. (g) represents the mass of the catalyst during the catalytic performance test.

The kinetic experiments were performed in the same reactor. The CO₂ conversion was kept below 15% by adjusting the gas hourly space velocity (GHSV) to eliminate any diffusion resistance. The reaction order with respect to reactants (CO₂ and H₂) was obtained at 500 °C, and changing the flow rate of H₂ between 6–72 mL·min⁻¹, the flow rate of CO₂ between 6–72 mL·min⁻¹, while using N₂ as a balance to maintain a 100 mL·min⁻¹ total flow rate. The apparent activation energy was measured similarly.

For deactivation resistance experiments in the sulfur-containing impurity gases, the catalysts were allowed to reach a steady-state rate at 600 °C in an H₂S-free reactant stream. Subsequently, 5 ppm, 50 ppm, and 300 ppm H₂S were introduced into the reactant stream in sequence at 600 °C for reaction stability testing. Finally, the catalysts were again exposed to the H₂S-free reactant stream at 600 °C to study the stability. The Ar balance gas flow rate must be adjusted to account for the addition of H₂S to maintain the same gas hourly space velocity (GHSV).

Author contributions

Y. Liu and G.C. conceived the ideas and supervised the study. S.L. conducted most of the material synthesis, in-situ/ex-situ characterisations, and carried out all the catalytic tests. Z.G. contributed to the materials synthesis. S.L., T.Q., X.L., and Y. Liu contributed to the measurement and analysis of the HR-TEM/STEM experiments. N.L. and R.Z. performed the DFT calculations and analysis. Y. Lyu, L.C., S.P., and G.C. provided helpful discussions and data analysis. S.L., G.C., and Y. Liu wrote the paper. All authors contributed to the discussion and revision of the paper.

Peer review

Peer review information

Nature Communications thanks Luis F. Bobadilla and Kohsuke Mori for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was financially supported by the NSFC of China (22472168 and 22172161), Natural Science Foundation of Liaoning Province (2024-MSBA-57), Dalian Science and Technology Innovation Fund (2024JJ12RC034), Dalian Institute of Chemical Physics (DICP I202421),  the Energy Revolution S&T Programme of Yulin Innovation Institute of Clean Energy (E411030705), and State Key Laboratory of Clean and Efficient Coal Utilization, Taiyuan University of Technology (MJNYSKL202505). R.Z. acknowledged the financial support from the National Science Foundation of China (U23A20132) and the Science Foundation for Distinguished Young Scholars of Shanxi Province (20210302121005).

Data availability

Data generated and analyzed in this study are included in the manuscript, Supplementary Information. All data are available from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74091-z.