Inverse In₂O_3-x/Ni interfaces via Ni₃InC_0.5 surface reconstruction for efficient CO₂ hydrogenation to methanol

Abstract

Catalyst surface reconstruction under reaction conditions is ubiquitous and crucial for creating unusual active sites, thereby enhancing catalytic performance. Here, we report the surface reconstruction of supported Ni₃InC_0.5 nanoparticles, leading to the formation of defective In₂O_3-x overlayers and inverse In₂O_3-x/Ni interfaces, driven by CO₂-induced selective surface oxidation during CO₂ hydrogenation. The synergy between In₂O_3-x overlayers and inverse In₂O_3-x/Ni interfaces facilitates CO₂ adsorption and activation, as well as the following hydrogenation of HCOO^* and CH_xO^* intermediates, enabling efficient methanol synthesis from CO₂. Accordingly, the optimized LDH-NiInCAl catalyst achieves an impressive CO₂ conversion of 19% with 65% methanol selectivity and 508.4 ${{\rm{mg}}}{{{\rm{g}}}}_{{{\rm{cat}}}}^{-1}{{{\rm{h}}}}^{-1}$ methanol space-time yield at 260 °C, 5 MPa, and 12000 ${{\rm{mL}}}{{{\rm{g}}}}_{{{\rm{cat}}}}^{-1}{{{\rm{h}}}}^{-1}$, outperforming commercial Cu/ZnO/Al₂O₃ catalysts. This work showcases how structural evolution and surface reconstruction enhance catalytic performance, providing new insights into the dynamic structure-activity relationship.

Introduction

Surface reconstruction of catalysts, triggered by structural evolution during pretreatments or reactions, can give rise to new surface structures distinct from their original state¹. This phenomenon is prevalent in redox reactions—such as CO₂ hydrogenation^2,3,4,5, CO oxidation⁶, Fischer-Tropsch synthesis⁷, and methane dry reforming^8,9, where the dynamic redox environment facilitates these transformations. Diverse structural changes can enhance or modify catalytic activity and selectivity^2,3,4,5,6, and in some cases improve stability^8,10. Despite the recognized importance of catalyst structural evolution and surface reconstruction, their precise characterization remains challenging due to the demands of high-resolution and in-situ techniques^11,12. These constraints hinder accurate understandings of the structure-performance relationship and the underlying mechanisms driving catalytic behavior, thereby limiting the effective utilization of surface reconstruction in catalyst design.

In the past decade, CO₂ hydrogenation to methanol has garnered significant attention as a promising and sustainable route for chemical and fuel production, supporting the global pursuit of net-zero industrial emissions¹³. Traditionally, Cu/ZnO/Al₂O₃ (CZA) catalyst has been the benchmark catalyst for industrial methanol synthesis from CO or CO₂, leveraging the interplay between Cu and ZnO to enhance methanol selectivity through the formation of ZnO_x species on Zn-decorated Cu or Cu-Zn alloy nanoparticles (NPs) under CO₂/H₂ atmospheres^14,15. However, the structural sensitivity to reaction atmospheres leads to ongoing catalyst evolution, including sintering and valence change of Cu species with ZnO agglomeration, which compromises its long-term stability¹⁶. This trade-off has driven the search for alternative catalysts for CO₂-to-methanol conversion. Recently, theoretical studies have demonstrated that In₂O₃ is excellent for CO₂ activation and stabilization of key intermediates in methanol synthesis^17,18, exhibiting even higher methanol selectivity and activity than conventional Cu-based catalysts¹⁹. This insight has spurred growing interest in the development of more efficient In₂O₃-based catalysts for CO₂-to-methanol conversion. As a result, significant progress has been made in the design of advanced In₂O₃-based catalytic systems, including In₂O₃-supported metal catalysts^20,21 and zirconia-supported In₂O₃ catalysts^22,23. Typically, introducing metal promoters, such as Pd or Ni compensates for the inherently weak H₂ activation ability of In₂O₃, significantly boosting methanol production rates^20,21. Meanwhile, the formation of In-based intermetallic compounds (IMCs) may provide more active sites for intermediate stabilization^24,25. However, the ultra-high H₂ activation capability of these metals can inadvertently lead to over-reduction of the In₂O₃ support. Although such catalysts often exhibit high initial performance, this over-reduction impairs the CO₂ activation ability of In₂O₃ and ultimately undermines catalyst stability^26,27. So far, achieving stable metal/In₂O₃ interfaces, such as Ni/In₂O₃, remains a key but challenging issue in practice.

In this context, we report the surface reconstruction of supported Ni₃InC_0.5NPs during CO₂ hydrogenation that can induce stable defective In₂O_3-x overlayers on the carbide phase. This unique structure introduces additional active sites for CO₂ adsorption, stabilizes formate intermediates, and establishes robust inverse In₂O_3-x/Ni interfaces featured with highly active asymmetric Ni-In sites, enabling accelerated hydrogenation of formate intermediates and methanol production. On this basis, an alumina-supported NiIn carbide catalyst derived from layered double hydroxide (LDH-NiInCAl) was developed, achieving an impressive CO₂ conversion of 19%, methanol selectivity of 65%, and a methanol space-time yield of 508.4${\mbox{mg}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$ at 260°C, 5MPa, and 12000${\mbox{mL}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$. This performance surpasses that of commercial Cu/ZnO/Al₂O₃ catalysts, ranks among the best state-of-the-art catalysts reported, and demonstrates strong potential for industrial methanol synthesis from CO₂.

Results

Synthesis of supported Ni₃InC_0.5 catalysts

Figure 1a and Supplementary Fig. 1 illustrates the synthetic approach employed to produce supported Ni₃InC_0.5 catalysts for efficient and stable CO₂ hydrogenation to methanol, detailing the crystal structural evolution from Ni₃In to Ni₃InC_0.5. In brief, during thermal treatment under a 25 vol% CO₂/H₂ gas flow at 600 °C, as determined by H₂ temperature-programmed reduction (H₂-TPR) profiles (Supplementary Fig. 2), the supported oxide precursor NiInZrO, comprising NiO, In₂O₃, and an oxide support (monoclinic zirconia as a paradigm), is initially reduced by H₂ to generate supported Ni₃In IMC NPs. These supported IMC NPs then facilitate the reverse water-gas shift (RWGS) reaction, producing CO as the carburizing agent²⁸. Adsorbed CO subsequently reacts with Ni₃In NPs, as described by Eq. (1), wherein carbon atoms diffuse into the Ni₃In lattice and occupy octahedral interstices of Ni¹⁰, leading to lattice expansion and phase transformation to Ni₃InC_0.5.

$$2{{{\rm{Ni}}}}_{3}{{\rm{In}}}({{\rm{s}}})+{{\rm{CO}}}({{\rm{g}}})+{{{\rm{H}}}}_{2}({{\rm{g}}})\to 2{{{\rm{Ni}}}}_{3}{{\rm{In}}}{{{\rm{C}}}}_{0.5}({{\rm{s}}})+{{{\rm{H}}}}_{2}{{\rm{O}}}({{\rm{g}}})$$

(1)

Fig. 1: Synthesis of supported Ni₃InC_0.5 catalysts.

a Schematic illustration of the structure and phase transformation of Ni₃InC_0.5 catalysts. The molecular models represent H₂, CO, and H₂O, respectively. b,c In-situ XRD patterns (b) and contour maps (c) of the supported NiIn precursor under 25 vol% CO₂/H₂ during the heating process. Numbers in parentheses indicate the representative crystalline planes of the specific species. d In-situ HRTEM images of a supported Ni₃In NP. White dotted lines: grain boundary between Ni₃InC_0.5 and Ni₃In. Yellow arrows: direction of the grain boundary moving. Atmosphere: CO/CO₂/H₂ = 10:22.5:67.5, 400 °C, 1 bar.

In addition to the zirconia-supported Ni₃In and Ni₃InC_0.5 catalysts, named NiInZr and NiInCZr, we also designed supported Ni₃Zn, Ni₃Ga, and their corresponding carbide catalysts as reference samples for explorative and extended studies. X-ray diffraction (XRD) analysis identified the supported species in the two NiIn catalysts as hexagonal Ni₃In (D0₁₉-type IMCs) and cubic Ni₃InC_0.5 (derived from L1₂-type IMCs) (Supplementary Fig. 3)^29,30. X-ray fluorescence spectroscopy (XRF) confirmed that the nominal Ni content (33 wt.%) and Ni/In stoichiometric ratio (R_Ni/In = 3.0) in NiInCZr are accurately achieved (Supplementary Table 1). Textural property measurements revealed that both catalysts exhibit similar surface areas and pore volumes (Supplementary Table 2). High-resolution transmission electron microscopy (HRTEM) images showed that both catalysts possess classical supported-catalyst architectures, with IMC/carbide NPs approximately 11 nm in size randomly distributed across the monoclinic ZrO₂ (m-ZrO₂) support surface (Supplementary Figs. 4–6).

The carbide formation mechanism was elucidated by monitoring the carburization at both macro and micro scales. Specifically, in-situ XRD and in-situ thermogravimetric analysis (TGA) identified three distinct transformation stages for the NiInZrO precursor during carburization (Fig. 1b,c and Supplementary Figs. 7,8): (i) formation of IMCs (Ni₃In and Ni₄In), (ii) conversion of IMCs to carbide (Ni₃InC_x, where x < 0.5), and (iii) further carburization to Ni₃InC_0.5, supporting the process of alloying, CO production, then carburization. In-situ HRTEM studies provided nanoscale insights into the evolution mechanism, visualizing the gradual transformation of NPs from Ni₃In to Ni₃InC_0.5 during the RWGS reaction (Fig. 1d and Supplementary Movie 1). The movement of the grain boundary between Ni₃InC_0.5 and Ni₃In phases demonstrates the phase transformation, progressing from one side of NPs to the other under the CO/CO₂/H₂ atmosphere. Insights from theoretical calculations suggest that the underlying mechanism might be easier C formation from CO on the carbide surface (Supplementary Fig. 9).

In₂O_3-x overlayers on Ni₃InC_0.5 NPs via surface reconstruction

Accompanying the phase transformation, a nanoscale overlayer with low crystallinity was observed covering the surface of Ni₃InC_0.5 (Supplementary Fig. 10), as further verified by the quasi-in-situ aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC-HAADF-STEM) analysis of the NiInCZr catalyst transferred under vacuum conditions after carburization (Fig. 2a). The composition and origin of this overlayer were further explored by comparing the X-ray photoelectron spectroscopy (XPS) spectra of the air-passivated and re-activated NiIn carbide catalyst, transferred inertly by a specialist sample holder. Following reactivation of the passivated sample under a 25 vol% CO₂/H₂ flow at 400 °C and 1 bar, the peaks of oxidative Ni disappeared, while those of indium oxides partially remained (Supplementary Fig. 11). Therefore, overlayers on Ni₃InC_0.5 NPs are composed of In₂O₃ derived from surface reconstruction, specifically the selective oxidation of the Ni₃InC_0.5 phase during the RWGS reaction. AC-HAADF-STEM image of the fresh NiInCZr sample further identified the overlayers, approximately 1 nm thick, as cubic In₂O₃ with an exposed (400) facet (lattice spacing of 2.53 Å) (Fig. 2b and Supplementary Fig. 12).

Fig. 2: In-situ characterizations on the formation of In₂O_3-x overlayers.

a Quasi-in-situ AC-HAADF-STEM image of the NiInCZr catalyst transferred using glove boxes and Mel-Build Double tilt Atmos Defend holder. b Enlarged AC-HAADF-STEM image of the fresh NiInCZr catalyst. c In-situ AC-HAADF-STEM image of a supported Ni₃In NP after one and a half hours of reaction. d In-situ Raman spectra of the unsupported Ni₃InC_0.5 sample under 25 vol% CO₂/H₂ atmosphere, 400 °C, 1 bar. e In-situ Ni 3 d, In 3 d, O 1 s, and C 1 s XPS spectra of the NiInZrO precursor treated by H₂ then 25 vol% CO₂/H₂ at 400 °C, 1 bar. BE: binding energy. f,g Proportion of surface In³⁺ species calculated from XPS In 3 d spectra: NiInO, NiInZrO, and NiInCZr sample subjected to a sequential in-situ XPS experiment (f) and NiInCZr catalysts synthesized under different CO/CO₂/H₂ atmosphere (g). h H₂O signals from H₂-TPR coupled with MS corresponding to NiInCZr catalysts synthesized under the same conditions. CO/CO₂/H₂ gas used: CO/H₂ (1:2), CO/CO₂/H₂ (3:3:14), and CO₂/H₂ (1:3).

Notably, this selective surface oxidation occurs exclusively on the Ni₃InC_0.5 phase, as no changes, neither carburization nor surface oxidation, were observed over the Ni₃In phase during the in-situ STEM experiment under the CO₂/H₂ atmosphere without CO co-feeding (Fig. 2c and Supplementary Fig. 13).

More solid evidence of surface reconstruction was obtained through multiple in-situ spectroscopic characterizations, as detailed in Supplementary Table 3 and the Methods section. For in-situ Raman spectroscopy, an unsupported Ni₃InC_0.5 (NiInC) sample was carburized using pure CO in advance to remove the air-passivated oxide layer and avoid interference from the Raman peaks of the ZrO₂ support (Supplementary Fig. 14). Upon introducing 25 vol% CO₂/H₂ gas into the in-situ cell, two new bands at 106 and 305 cm⁻¹ appeared and gradually intensified, corresponding to In-O vibration and δ bending vibration (E_1g) modes in the [InO₆] octahedral units of c-In₂O₃, respectively (Fig. 2d)^31,32,33. Importantly, no peaks assigned to NiO species were detected³⁴, reinforcing the selective feature of surface oxidation.

In-situ XPS also substantiated these findings (Fig. 2e and Supplementary Table 4). In the Ni 2p spectra, peaks at binding energy (BE) of 852.4, 855.6, and 861.4 eV correspond to Ni⁰, Ni²⁺, shake-up peaks in the Ni 2p_3/2 region^9,10,35. Following sequential treatments with H₂ and 25 vol% CO₂/H₂, Ni species transitioned from NiO (Ni²⁺) to Ni₃In (Ni⁰) and further to the Ni₃InC_0.5 phase (Ni⁰). These transformations were also observed by the C 1 s spectra, where the appearance of a Ni-C peak at 283.2 eV followed the vanishing of the C-O peak at 286.0 eV, the decline and later intensification of the C = O peak at 289.0 eV corresponding to the surface carbonate groups^28,36,37. In the In 3 d spectra, peaks with BE of 444.5 eV and 452.1 eV correspond to the oxidative state (In³⁺) of In 3d_5/2 and 3d_3/2, respectively, while peaks for In⁰ species are located at 443.3 and 450.9 eV^28,38. The presence of In³⁺ peaks in the reduced sample may stem from small amounts of unreduced In₂O₃ strongly interacting with the support at a low reduction temperature of 400 °C, or from intercalated In₂O₃ species located between the NiIn IMC and the ZrO₂ support^39,40. More importantly, the increased peak area for In³⁺ after carburization, along with a rise in the O 1 s peak area at 531.2 eV, indicates In oxidation and a concentration increase in oxygen vacancy^9,38, suggesting the formation of a sub-stoichiometric, defective In₂O_3-x overlayer containing abundant oxygen vacancies. The same variation trend in the proportion of surface In³⁺ species was also observed from the unsupported NiO/In₂O₃ (NiInO), NiInZrO, and NiInCZr samples after exposure to CO₂/H₂ (1:3) atmosphere, following sequential H₂ and CO/H₂ (1:2) treatments during in-situ XPS experiments (Fig. 2f and Supplementary 15).

Moreover, the content of In₂O₃ species was found to be tunable by adjusting the CO₂ concentration within the CO/CO₂/H₂ feed gas during the synthesis of NiInCZr catalysts. As revealed by the quasi-in-situ XPS In 3 d spectra, the proportion of surface In³⁺ species increased with rising CO₂ concentration, with the NiInCZr sample synthesized under CO₂/H₂ (1:2) exhibiting the highest amount (Fig. 2g and Supplementary 16). These findings were further supported by H₂-TPR coupled with MS analysis, wherein the NiInCZr-CO₂/H₂ sample showed the most prominent H₂O signal peak at ~500 °C, corresponding to the reduction of surface In₂O₃ species (Fig. 2h and Supplementary 17). Collectively, these results strongly suggest that CO₂ might act as the driving force for selective surface oxidation of Ni₃InC_0.5, leading to the formation of In₂O_3-x overlayers.

CO₂-induced in exsolution and inverse In₂O_3-x/Ni interfaces

Given the critical role of CO₂ in selective surface oxidation, two oxidation experiments were conducted on NiInZr and NiInCZr catalysts, each exposed to one of the oxidative components involved in the RWGS reaction, CO₂ and H₂O, respectively, without H₂ co-feeding.

Upon CO₂ oxidation, the NiInZr catalyst developed a significant amount of crystalline c-In₂O₃ species assembled on the surface of the NiIn IMC NPs, with no detectable NiO phase (Supplementary Figs. 18a and 19a, b). As expected, Ni 2p spectra showed no significant changes, whereas metallic In was fully oxidized to In³⁺ (Supplementary Fig. 19c). In contrast, the NiInCZr catalyst exhibited greater structural stability under identical conditions. However, XPS still detected a moderate increase in surface In³⁺ content. To sum up, CO₂ can preferentially oxidize In when In alloys with Ni to form Ni₃In and Ni₃InC_0.5, consistent with thermodynamic analysis of CO₂ oxidation for Ni and In (Supplementary Fig. 18b).

On the contrary, H₂O exhibits a stronger oxidation ability, simultaneously oxidizing both surface Ni and Integrating these findings with the Ni-In binary phase diagram⁴¹, we propose that In atoms exsolve from the surface of NiIn species (IMC and carbide), leaving behind a Ni-rich layer between the NiIn core and the In₂O_3-x shell, thereby forming an inverse In₂O_3-x/Ni interface. In the NiIn IMC and carbide catalysts (Supplementary Fig. 20).

To further validate the formation of the inverse In₂O₃/Ni structure, quasi-in-situ and in-situ X-ray absorption spectroscopy (XAS) were performed on the NiIn-based catalysts. As shown in Fig. 3a,b and Supplementary Table 5, X-ray absorption near edge structure (XANES), R-space extended X-ray absorption fine structure (EXAFS) spectra, and corresponding first-shell coordination fitting reveal that Ni in all NiIn carbide samples remains in a metallic state, coordinated by Ni-C, Ni-Ni, and Ni-In paths. Compared with Ni foil, the positive shift in absorption edge and the enhanced white line intensity indicate an electron-deficient character of Ni atoms in both fresh and spent NiInCZr samples^42,43, consistent with XPS analysis (Supplementary Fig. 11).

Fig. 3: Inverse In₂O_3-x/Ni interfaces from CO₂-induced selective surface oxidation.

a,b Investigation of the coordination of NiInCZr and referenced samples: Ni K-edge XANES (a) and EXAFS spectra (b). All samples were transferred through a glove box and sealed with tape before test. c In-situ Fourier transformed k³-weighted Ni-EXAFS spectra of the NiInZrO precursor with atmosphere changing from CO₂/H₂ to CO/H₂ to CO₂/H₂. Conditions: 400 °C, 1 bar. d CN ratio of Ni-Ni path to Ni-In path over different Ni₃InC_0.5-based catalyst. NiInC: unsupported Ni₃InC_0.5; Spent: spent NiInCZr; CO/H₂, CO/CO₂/H₂, CO₂/H₂ represent the NiInCZr sample synthesized using corresponding gas. e Computed phase diagram of Ni₃In, Ni₃InC_0.5, Ni + In₂O₃, and NiO + In₂O₃ with each solid line indicating phase equilibrium between two species. f Schematic illustration of the inverse In₂O_3-x/Ni interface structure.

In-situ XAS measurements under sequential gas treatment (H₂, CO₂/H₂, CO/H₂, and again CO₂/H₂) provided direct insight into In exsolution and Ni-rich layer formation (Fig. 3c and Supplementary Fig. 21). After carburization in CO₂/H₂ and CO/H₂, subsequent CO₂/H₂ treatment induced surface oxidation. The observed peak sharpening during these processes indicates enhanced short-range order of the NiInC carbide compared to the NiIn alloy, reflecting higher crystallinity. Moreover, the increasing intensity of the Ni-Ni path, accompanied by a decreasing Ni-In path, supports the formation of an In₂O₃ overlayer via In atoms exsolution from the Ni₃InC_0.5 surface, leaving a Ni-rich layer underneath and yielding an inverse In₂O_3-x/Ni interface.

Further evidence was provided by EXAFS spectra and coordination number (CN) fitting of NiInCZr catalysts synthesized under various CO/CO₂/H₂ atmospheres (Supplementary Figs. 22, 23 and Table 5). Compared to the benchmark unsupported NiInC sample, catalysts prepared with higher CO₂ content exhibited a higher CN ratio (R_CN) of Ni-Ni path to Ni-In path (Fig. 3d). This trend indicates an increase in both the Ni-rich layer and In₂O₃ overlayer content, reaffirming that CO₂ promotes selective surface oxidation and facilitates inverse In₂O₃/Ni interface formation.

Phase diagram of the NiIn-based system, calculated as a function of chemical potentials of carbon (μ_C) and oxygen (μ_O), provide additional theoretical support for the formation of this inverse structure (Fig. 3e). Under the CO₂/CO/H₂ atmosphere, bulk carburization is thermodynamically favored over surface oxidation on Ni₃In, as the reducing power of H₂ outweighs the oxidizing effect of CO₂. According to Eq. (2), CO₂ selectively oxidizes the surface of the Ni₃InC_0.5 phase upon its formation. Ultimately, under the equilibrium conditions, the phase settles in the Ni and In₂O₃ domain on the phase diagram, consistent with the inverse In₂O_3-x/Ni interface structure

$$2{{{\rm{Ni}}}}_{3}{{\rm{In}}}{{{\rm{C}}}}_{0.5}({{\rm{s}}})+4{{{\rm{CO}}}}_{2}({{\rm{g}}})\to 6{{\rm{Ni}}}({{\rm{s}}})+{{{\rm{In}}}}_{2}{{{\rm{O}}}}_{3}({{\rm{s}}})+5{{\rm{CO}}}({{\rm{g}}})$$

(2)

Taken together, the CO₂ oxidation behaviors of NiInZr and NiInCZr elucidate the mechanism of selective oxidation on the Ni₃InC_0.5 surface: during the RWGS reaction, CO₂ initiates the reverse Boudouard reaction (Eq. (2)), releasing oxygen atoms to the surface and promoting selective oxidation of In. With continued In exsolution, the remaining Ni atoms form a Ni-rich layer between the Ni₃InC_0.5 core and the In₂O_3-x overlayer, thus creating an inverse In₂O_3-x/Ni interface (Fig. 3f).

Catalytic performance and structure-activity relationship

To assess the roles of In₂O_3-x overlayers and inverse In₂O_3-x/Ni interfaces on CO₂ hydrogenation to methanol, supported Ni₃InC_0.5 and the other referenced catalysts were evaluated under reaction conditions of 4 MPa, H₂/CO₂/N₂ = 72:24:4, and a weight hourly space velocity (WHSV) of 12,000 ${\mbox{mL}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$, unless otherwise specified.

The NiInCZr catalyst exhibited good activity for CO₂ hydrogenation to methanol, surpassing the selectivity limitations of traditional Ni-based catalysts that favor CH₄^44,45 and CO^2,46 production (Fig. 4a, b and Supplementary Figs. 24–26). Specifically, NiInCZr reached a CO₂ conversion (${X}_{{{\mbox{CO}}}_{2}}$) of ca. 9% at 300 °C, while its methanol selectivity (S_MeOH) showed an inverse trend as the temperature increased. Both parameters were consistently higher across all temperatures than those of NiInZr. Consequently, the methanol space-time yield (STY_MeOH) exhibited a volcano-shaped profile, peaking at 193.6${\mbox{mg}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$ at 280 °C with 7% of ${X}_{{{\mbox{CO}}}_{2}}$ and 70% of S_MeOH. Even in the absence of support effects, unsupported NiInC maintained methanol selectivity above 80% (Supplementary Fig. 27).

Fig. 4: Catalytic performance of CO₂ hydrogenation over the supported Ni₃InC_0.5 catalyst.

a–c Catalytic performance: CO₂ conversion (a), selectivity (b), stability at 280 °C (c). Insets in (c) is the illustration of in-situ carburization and surface oxidation over the Ni₃In IMC phase during the reaction. d Relative content of two active phases in the NiInZr catalyst at different TOS determined by XRD analysis. The (201) of Ni₃In and (111) of Ni₃InC_0.5 chosen as the representative planes were normalized by ($\bar{1}$11) of m-ZrO₂. TOS: time-on-stream. e H₂O signals from H₂-TPR coupled with MS corresponding to spent NiInCZr and NiInZr-10h samples. f Arrhenius plots of the Ni₃InC_0.5 catalyst. InZrO: ZrO₂ supported In₂O₃ catalyst. Standard evaluation conditions: 4 MPa, CO₂/H₂ = 1:3, WHSV of 12,000${\mbox{mL}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$.

During the long-term stability test, the NiInCZr catalyst exhibited a rapid increase in performance initially and remained stable within 250 h in terms of ${X}_{{{\mbox{CO}}}_{2}}$, S_MeOH, and STY_MeOH, surpassing most of the In₂O₃-based catalysts, as well as the commercial Cu-based catalyst for CO₂-to-methanol (Fig. 4c, Supplementary Fig. 28 and Table 6)^16,47,48. As anticipated, the spent NiInCZr catalyst maintained a stable bulk crystal structure, with no observable new peaks emerging or peak shifting in the XRD pattern (Supplementary Fig. 29). Besides, the AC-HAADF-STEM image displayed the well-preserved In₂O_3-x overlayer structure after the long-term reaction (Supplementary Fig. 30). From XPS spectra, Ni remained in the metallic state, while the content of oxidative In increased slightly (Supplementary Fig. 31). These results demonstrate the high stability of the inverse In₂O_3-x/Ni structures on the Ni₃InC_0.5 phase, addressing the inherent shortcomings of the metal-promoted In₂O₃ catalysts^20,21,47,48.

As for the NiInZr catalyst, its STY_MeOH, as well as ${X}_{{{\mbox{CO}}}_{2}}$ and S_MeOH, started with ultra-low values and continued to increase within 10 h on stream, hinting at its persistent structural evolution to form active species (Fig. 4c and Supplementary Fig. 28). According to its XRD patterns at a different time on stream (TOS), a phase transformation from Ni₃In to Ni₃InC_0.5 was found over time (Fig. 4d and Supplementary Fig. 32), which was also observed with the unsupported Ni₃In catalyst during the long-term stability test (Supplementary Fig. 33). Furthermore, the HRTEM image of the NiInZr TOS-2h sample exhibited a half-transformed state of a supported Ni₃In NP after 2 h of the reaction (Supplementary Fig. 32b), characterized by a distinct grain boundary between the Ni₃In (102) and Ni₃InC_0.5 (200) planes. Analysis of the H₂-TPR coupled with MS reveals that the NiInZr−10 h sample exhibits only a minor H₂O signal peak at ~500 °C, corresponding to the reduction of In₂O₃ species (Fig. 4e). This low signal intensity, indicative of limited In₂O₃ content and IMC-like surface, correlates with the incomplete carburization and further highlights the critical roles of In₂O_3-x overlayers and inverse In₂O_3-x/Ni interfaces in CO₂ hydrogenation to methanol, as well as the sequence order of the bulk carburization and surface oxidation. The catalytic performance at the beginning two hours of catalyst evaluation tests over the NiInCZr catalysts synthesized under various CO/CO₂/H₂-containing atmospheres again approve this conclusion (Supplementary Fig. 34).

Kinetic studies give a deeper understanding of the structure-activity relationship. External diffusion effect was eliminated by varying the WHSV applied (Supplementary Fig. 35). The extracted apparent activation energy (E_a,app) values show that NiInCZr has the lowest E_a,app for methanol formation (51.6 kJ mol⁻¹) and the highest for CO (73.8 kJ mol^-1), correlating with the highest S_MeOH and lowest S_CO (Fig. 4f and Supplementary Fig. 36). Reaction orders (n_i) with respect to CO₂ and H₂, which reflects surface reactant coverage^22,49, indicate that NiInCZr has a superior CO₂ adsorption/activation ability (${n}_{{{\mbox{CO}}}_{2}}$ = 0.14), despite its comparatively low H coverage (${n}_{{{\mbox{H}}}_{2}}$ = 1.39) (Supplementary Fig. 37). Besides, the H₂ reaction order for NiInCZr showed an unusual trend: the methanol formation rate (r_MeOH) increased as the H₂/CO₂ ratio (${R}_{{{\mbox{H}}}_{2}/{\mbox{C}}{{\mbox{O}}}_{2}}$) raised from 3 to 3.5, displaying a positive ${{\mbox{n}}}_{{{\mbox{H}}}_{2}}$. However, beyond an ${R}_{{{\mbox{H}}}_{2}/{\mbox{C}}{{\mbox{O}}}_{2}}$ of 3.5, further increases in H₂ partial pressure led to a sharp decline in r_MeOH, resulting in a negative ${n}_{{{\mbox{H}}}_{2}}$, contrary to conventional CO₂ hydrogenation behavior^22,49,50. Considering the surface structure of the Ni₃InC_0.5 NPs, we surmise that the abnormal upheaval in H₂ reaction order should be attributed to the degradation of In₂O_3-x overlayers under elevated H₂ partial pressures⁴⁷.

Similar conclusions could be drawn from the temperature-programmed desorption (TPD) tests (Supplementary Figs. 38 and 39). From CO₂-TPD profiles, NiInCZr displayed more medium and strong basic sites for CO₂ adsorption than NiInZr and the m-ZrO₂ support⁵¹. Due to the presence of In₂O_3-x overlayer, it can provide extra active sites for CO₂ activation on Ni₃InC_0.5 NPs^22,38,47, accounting for the enhanced ${X}_{{{\mbox{CO}}}_{2}}$ over the NiInCZr catalyst. Therefore, even without support, Ni₃InC_0.5 remains more effective for CO₂ hydrogenation to methanol than the NiInZr catalyst (Supplementary Fig. 27). On the other hand, H₂-TPD analysis indicates poorer H₂ activation over Ni₃InC_0.5 than Ni₃In IMC. Both CO₂ and H₂-TPD results are consistent with the low H₂ and high CO₂ coverage inferred from the kinetic studies, owing to the presence of the defective In₂O_3-x overlayers.

To further enhance the catalytic performance of Ni₃InC_0.5-based catalysts for CO₂ to methanol, we optimized the oxide support and identified Al₂O₃ as the most effective option (Supplementary Fig. 40 and Table 7). Given that layered double hydroxides (LDHs) are excellent precursors for achieving high dispersion of active species on Al₂O₃ support^52,53, we subsequently developed an LDH-derived Al₂O₃-supported NiIn carbide catalyst (LDH-NiInCAl). This catalyst achieved an ${X}_{{{\mbox{CO}}}_{2}}$ of 19% and S_MeOH of 65%, with a STY_MeOH of 508.4 ${\mbox{mg}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$ at 260 °C, 5 MPa, and 12,000 ${\mbox{mL}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$, surpassing the commercial CZA catalysts and ranking among the state-of-the-art catalysts reported, including noble metal-based catalysts, showing great promising for industrial application (Supplementary Fig. 41 and Table 8).

Reaction mechanisms and active sites

Fundamental knowledge of the underlying reaction mechanisms and active sites of the Ni₃InC_0.5 phase was investigated by in-situ transient Fourier transform infrared spectroscopy (FTIR) performed at 280 °C, 2 MPa, for the NiInCZr catalyst. Detailed vibration band assignments for surface species are listed in Supplementary Table 9.

During CO₂ adsorption, monodentate carbonate (m-CO₃^*, 1398 and 1495 cm⁻¹)^54,55,56 dominated at 0.1 MPa, with some ionic bicarbonate species (i-HCO₃^*, at 1420 cm⁻¹)⁵⁷ forming as well (Fig. 5a). Since ZrO₂ and In₂O₃ species are the only two candidates for CO₂ adsorption on NiInCZr catalyst, the distinct CO₂ adsorption profile of m-ZrO₂ (b-CO₃^*) demonstrate that the In₂O_3-x overlayer serves as the primary sites for CO₂ activation (Supplementary Fig. 42). This conclusion is consistent with both the kinetic and chemisorption results. At 2 MPa, m-CO₃^* intensified, and small amounts of formate species (HCOO^*, ca. 1585 cm⁻¹) formed via hydrogenation of carbonate by surface hydrides (Ni-H or In-H). The observed bands at 2150–1850 cm⁻¹ and 1100–1000 cm⁻¹, e.g., the signals at 2077, 2055, 1080, and 1050 cm⁻¹, are assigned to gas-phase CO₂, rather than adsorbed CO^* species from CO₂ dissociation^58,59.

Fig. 5: Identifying active sites and reaction mechanisms of Ni₃InC_0.5 carbide catalyst.

a,b High-pressure in-situ FTIR spectra: CO₂ adsorption and subsequent H₂ hydrogenation (a), H₂ flushing post-CO₂ hydrogenation (b). c Corresponding normalized IR intensities of various species versus reaction time during the H₂ flushing process. Reaction conditions: 2 MPa, 280 °C, 30 mL min^-1 of reactant gas (H₂:CO₂ = 3:1). Intensity normalization was calculated based on the highest intensity value of the specific species. Assignments of ball models: blue (Ni), pink (In), gray (C), red (O), and white (H).

Upon switching to H₂, m-CO₃^* bands gradually faded, accompanied by the emergence of bidentate formate (bi-HCOO^*, located between 1600-1300 cm⁻¹), linearly bonded carbonyl (CO^* on exposed Ni sites, at 2022 cm⁻¹)⁶⁰, and methoxy (H₃CO^*, at 1146 and 1075 cm⁻¹) species, along with a decline in the gaseous CO₂ signals. Similar to the behavior of the ZnZrO_x catalyst⁵⁷, NiInCZr exhibited bi-HCOO^* adsorbed on adjacent In and Ni sites at the inverse In₂O_3-x/Ni interfaces (bi-HCOO-Ni-In, at 1595, 1347, and 1318 cm⁻¹), while bands at 1578, 1382, and 1371 cm⁻¹ correspond to bi-HCOO^* adsorbed on neighboring In sites on the In₂O_3-x overlayers (bi-HCOO-In-In)^{55,56,61,62,63,64}. Identical in-situ FTIR spectra of the unsupported NiInC catalyst further support the presence of these two active site configurations (Supplementary Fig. 43).

To assess their relevance to methanol synthesis, formate species consumption was tracked under H₂ flow after saturating the NiInCZr catalyst surface with intermediates in the 25 vol% CO₂/H₂ atmosphere. During CO₂ hydrogenation, the consumption of m-CO₃^* and formation of bi-HCOO^*, H₃CO^*, and CO^* were observed until the system reached a steady state (Supplementary Fig. 44a, b). When the mixed gas (25 vol% CO₂/H₂) was replaced with H₂, formate species, and gaseous CO₂ declined, while CO^* remained stable, indicating that methanol synthesis follows the formate pathway (Supplementary Fig. 44c and Fig. 5b). Notably, the consumption rates of the two formate species differ significantly. According to normalized intensity over time, bi-HCOO-In-In at 1578 cm⁻¹ peaked and then declined, while bi-HCOO-Ni-In at 1348 cm⁻¹ continued to decrease throughout the process (Fig. 5c). This result shows that the hydrogenation rate of bi-HCOO^* at the Ni-In sites is much faster than at In-In sites. The asymmetric Ni-In sites at the inverse In₂O_3-x/Ni interfaces are more active for methanol formation in the NiInCZr catalyst. A similar trend could also be observed from the in-situ FTIR spectra at different reaction temperatures, where elevated temperatures led to a visibly faster decrease in bi-HCOO-Ni-In intensity than that of the bi-HCOO-In-In (Supplementary Fig. 45).

In conclusion, two types of active sites contribute to the high methanol selectivity of the NiInCZr catalyst: (1) the In₂O_3-x overlayers, which facilitate CO₂ activation and perform the hydrogenation resembling conventional In₂O₃-based catalysts; and (2) the inverse In₂O_3-x/Ni interfaces, where Ni promotes the hydrogenation of formate species in cooperation with adjacent In (asymmetric Ni-In site).

Theoretical descriptions of reaction mechanisms

To deepen the understanding of the structure-dependent activity of the Ni₃InC_0.5 phase for CO₂ hydrogenation to methanol, density functional theory (DFT) calculations were conducted. Surface models of Ni₃In (111), Ni₃InC_0.5 (111), cubic In₂O₃ with an exposed (100) plane, and In₂O₃ clusters on Ni (111) (In₂O₃/Ni) were constructed to simulate the Ni₃In IMC, Ni₃InC_0.5, and the two active sites of the selectively oxidized Ni₃InC_0.5 phase: In-In sites on In₂O_3-x overlayers and asymmetric Ni-In sites at the inverse In₂O_3-x/Ni interfaces, respectively (Supplementary Fig. 46).

For CO₂ activation, the adsorption mechanism varied among the models. On the In₂O₃ and In₂O₃/Ni models, CO₂ adsorbs via its C atom binding to the lattice oxygen on the In₂O₃ surface, while one of its O atoms interacts with the In atom (${{\mbox{m}}-{\mbox{CO}}}_{3}^{*}$ configuration, Supplementary Fig. 47). In contrast, physical adsorption occurred on the Ni₃In and Ni₃InC_0.5 models. The calculated adsorption energies were higher on In₂O₃-related structures, particularly c-In₂O₃ (−0.90 eV), align with CO₂-TPD and kinetics studies, validating the enhanced CO₂ activation ability of the selectively oxidized Ni₃InC_0.5 phase. Both In₂O_3-x overlayers and inverse In₂O_3-x/Ni interfaces contribute key active sites for CO₂ activation.

For H₂ activation, heterolytic dissociation occurred on the In₂O₃ model with an activation energy (E_a) of 0.39 eV, forming hydroxyl (O-H) and hydridic H (In-H) species (Supplementary Fig. 48)¹⁷. In contrast, homolytic H₂ dissociation on the Ni (111) surface was nearly barrierless (E_a = 0.01 eV)⁶⁵. These results suggest that the exposed Ni sites at the inverse In₂O_3-x/Ni interfaces, in addition to the conventional active sites on In₂O_3-x overlayers, offer a stronger H₂ activation capability.

The formate mechanism for methanol synthesis was further analyzed to compare the activities of In-In and Ni-In sites (Fig. 6 and Supplementary Table 10–12). The initial step involves the hydrogenation of ${{\mbox{m}}-{\mbox{CO}}}_{3}^{*}$ to bi-HCOO^*. On the oxide overlayer, proton-like H from In₂O₃ drives this hydrogenation with an E_a of 1.50 eV, while at the interface, homolytic H from Ni facilitates the reaction with a lower E_a of 1.14 eV. The reduced E_a at the Ni-In sites is attributed to the different coordination structure, where two oxygen of bi-HCOO^* species bond to Ni and In atoms, respectively, in contrast to bonding two adjacent In atoms at the In-In sites. Calculations on Ni-doped In₂O₃, and In₂O₃/Ni₃InC_0.5 further validate the rationality of the constructed In₂O₃ overlayer and inverse In₂O₃/Ni models (Supplementary Figs. 49–52). For CO formation via the COOH^* pathway, the E_a is substantially higher than the formate pathway on both models (1.55 and 1.92 eV for the In₂O₃ and In₂O₃/Ni models, respectively).

Fig. 6: DFT studies of the reaction mechanisms at the interfacial asymmetric Ni-In sites of the NiInCZr catalyst.

Free-energy diagram of CO₂ hydrogenation to methanol following formate pathways at Ni-In and In-In sites. The insets represent the reverse In₂O_3-x/Ni interfaces and the reaction intermediates. The label (g) means gas phase. Color scheme for ball-and-stick models: blue-Ni, brown-In, gray-C, red-O, and white-H.

Across the elementary steps, the hydrogenation of bi-HCOO^* to H₂COO^* should be more kinetic relevant (Fig. 6). Interestingly, the E_a for H₂COO^* formation is significantly lower on the In₂O₃/Ni model (1.07 eV) than on In₂O₃ (2.26 eV), corroborating the observed acceleration of bi-HCOO^* hydrogenation at the asymmetric Ni-In sites in in-situ FTIR analysis. In addition, the energy barriers of subsequent hydrogenation of CH_xO species are also lower at the Ni-In sites.

Further insights into the accelerated hydrogenation rates at asymmetric Ni-In sites were gained from electronic structure analysis, including electron density difference and electron localized function (ELF) calculations (Supplementary Fig. 53). The analysis reveals electron accumulation around the O atom of bi-HCOO* species, accompanied by electron depletion around the C and In atoms. Importantly, the electron density between the O and C atoms at the Ni-In site is significantly reduced compared to the In-In site on In₂O₃. This weaker covalent bonding at the inverse In₂O_3-x/Ni interface facilitates faster hydrogenation of bi-HCOO* species, consistent with the experimentally observed enhanced catalytic activity.

Discussion

In summary, we presented a highly efficient supported Ni₃InC_0.5 phase for CO₂ hydrogenation to methanol, providing a comprehensive investigation into its active sites, structure-activity relationships, and underlying reaction mechanisms. Through multiple in-situ microscopic and spectroscopic characterizations, combined with DFT calculations, we elucidated the phase transformation of NiIn alloy to carbide and surface reconstruction mechanisms that underpin the catalytic activity of the Ni₃InC_0.5 phase. The formation of defective In₂O_3-x overlayers and stable inverse In₂O_3-x/Ni interfaces under reaction conditions, induced by CO₂-selective surface oxidation, emerged as a pivotal structural feature, enhancing CO₂ adsorption and stabilizing formate intermediates. These In₂O_3-x overlayers, in conjunction with interfacial asymmetric Ni-In sites, accelerate the hydrogenation of formate species and efficiently drive methanol production through the formate pathway. The distinctive interdependence between surface structure and reaction environment endows the Ni₃InC_0.5 phase with exceptional stability, maintaining its structure and catalytic activity over a long-term (250 h) test without any decay. In virtue of the strong CO₂ activation ability of surface-oxidized Ni₃InC_0.5, the LDH-NiInCAl catalyst achieves an impressive 19% CO₂ conversion, 65% methanol selectivity, and 508.4 ${\mbox{mg}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$ STY_MeOH at 260 °C, 5 MPa, and 12,000 ${\mbox{mL}}{{\mbox{g}}}_{{\mbox{cat}}}^{-1}\,{{\mbox{h}}}^{-1}$, outperforming the commercial Cu/ZnO/Al₂O₃ catalyst and ranking among the reported state-of-the-art catalysts. This work not only deepens the fundamental understanding of surface reconstruction and active site generation in CO₂ hydrogenation but also highlights the immense potential of NiIn carbide-based catalysts as a robust platform for developing high-performance catalysts for industrial methanol synthesis.

Methods

Catalyst synthesis

The precursors were prepared using a modified ammonia carbonate co-precipitation (ACP) method⁶⁶. Before synthesis, the crystalline water content of all metal nitrate hydrate materials was precisely calibrated using TGA to accurately control the molar ratio of Ni to a second metal (R_Ni/M2) at 3:1 required for the stoichiometry. Unless otherwise specified, the default Ni loading amount for carbide catalysts was 33 wt.%. In a typical process, a specific amount of Ni(NO₃)₂·xH₂O and In(NO₃)₃·xH₂O, with a molar ratio 3:1, were dissolved in 50 mL of deionized water in a round-bottom flask. Next, 0.5 g of commercial m-ZrO₂ powder was dispersed into the solution with 15 min of stirring, followed by 15 min of ultrasonication. Subsequently, 15 mL of (NH₄)₂CO₃ aqueous solution was added dropwise at 72 °C under vigorous stirring. After continuous stirring for 2 h and aging for another 2 h under reflux, the residue was cooled to room temperature, filtered, washed with DI water (three times, 300 mL each), and dried at 70 °C overnight. The dried cake was then ground and calcined at 350 °C for 2 h in a muffle furnace to obtain the powder oxide precursor. For the synthesis of NiInZr IMC catalyst, the obtained precursor was packed into a continuous-flow glass reactor with an inner diameter of 7 mm and reduced at 600 °C for 3 h under a flow of 30 mL min⁻¹ of H₂ (ramp rate, 5 °C min⁻¹). The synthesis process for the NiInCZr carbide sample is similar, except that a flow of 30 mL min⁻¹ of mixed gas (25 vol% CO₂/H₂) was used. After cooling to room temperature, both samples were flushed with Ar for 30 min before being removed from the glass tube. Details of the preparation for other IMC and carbide samples are provided in the Supplementary Methods. The Ni loading amount and the molar quantity of ammonia carbonate are calculated using the following equations:

$${{\rm{Ni}}}\; {{\rm{loading}}}\; {{\rm{amount}}}=\frac{{m}_{{{\rm{Ni}}}}}{{m}_{{{\rm{Zr}}}{{{\rm{O}}}}_{2}}+{m}_{{{\rm{carbide}}}}}\times 100\%$$

(3)

$$n({{\rm{ammonia\; carbonate}}})={n}_{{{\rm{Ni}}}}+{n}_{{{\rm{In}}}}\times 1.5$$

(4)

Where m represents the mass and n is the molar quantity.

Catalyst characterizations

TGA for metal nitrate hydrate samples was conducted using a thermal analyzer (Q50, TA Instruments) from room temperature to 800 °C at a ramp rate of 10 °C min⁻¹ in air. For in-situ carburization TGA, oxide precursors or Ni-based IMC catalysts were tested by changing the atmosphere in the thermal analyzer to 25 vol% CO₂/H₂. The samples were heated to 600 °C at a ramp rate of 5 °C min⁻¹, held for 30 min, and then allowed to cool.

XRD patterns were collected using a Rigaku MiniFlex 600 X-ray diffractometer with a Cu Kα radiation source (λ = 1.5418 Å). The fresh or spent catalyst powder was scanned in the 2θ range of 10–80° at a rate of 8° min⁻¹ with a step size of 0.02°.

In-situ XRD was performed on the Rigaku SmartLab (8 KW) high-resolution diffractometer to study the reduction and carburization behavior of the oxide precursors under a mixed gas atmosphere (25 vol% CO₂/H₂). To observe structural evolution during heating, the zirconia-supported NiIn precursor was scanned in the 2θ range of 15–60° at temperature intervals of 20 °C, with a rate of 15° min⁻¹ and default step size under 20 mL min⁻¹ of mixed gas flow until reaching 600 °C. For structural evolution during the isothermal process, the precursor was heated to 400 °C under 20 mL min⁻¹ of Ar flow, which was then switched to 20 mL min⁻¹ of mixed gas, and data were recorded using the same program for the following 1 h. The zirconia-supported NiZn and NiGa precursors were tested using a similar process but with the 2θ range adjusted to 30–75°.

The metal loading amount was calculated based on the X-ray fluorescence spectroscopy (XRF) results from oxide precursors using a Rigaku Supermini 200 spectrometer. The specific surface area (S_BET) and pore volume were determined using the Brunauer-Emmett-Teller (BET) method and Barrett-Joyner-Halenda (BJH) desorption branch, respectively, on a Micromeritics ASAP 2460 system by N₂ adsorption-desorption at 77 K. Prior to the tests, fresh catalysts were degassed at 300 °C for 12 h²⁸.

The nanomorphology and lattice structure were observed using transmission electron microscopy (TEM) with JEM-2100 and JEM-F200 instruments. Energy-dispersive X-ray spectroscopy (EDS) elemental mapping and high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were also obtained using the corresponding accessories from the TEM facilities. Aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC-HAADF-STEM) images were acquired on JEM-ARM200F microscope equipped with a dual-energy dispersive X-ray (EDX) spectroscopy system.

Quasi-in-situ/in-situ AC-HAADF-STEM and in-situ HRTEM experiments were conducted on JEOL ARM200F and JEM-F200 microscopes, respectively, both equipped with a Protochips Atmosphere 210 gas cell system. For in-situ experiments, the NiInZr catalyst powder was dispersed in ethanol and drop-casted onto the Protochips (Protochips Inc.) micro-electromechanical systems (MEMS) E-chips, which were subsequently assembled into a closed cell within a Protochips holder. The catalyst was firstly reduced in pure hydrogen (H₂) with a flow rate of 0.1 sccm (~ 0.02 m/s) at 760 Torr and 600 °C (ramp rate of 1 °C/s) for 60 min and then cooled to 400 °C. Then, the sample was exposed to CO/CO₂/H₂ (10:22.5:67.5) or CO₂/H₂ (25:75) at 400 °C to observe structure evolution. For quasi-in-situ tests, check valves at both ends of the glass reactor were closed after synthesis. Then, the fresh NiInCZr catalyst were transferred through a glove box and Mel-Build Double tilt Atmos Defend Holder, which was then loaded into a JEOL ARM200F microscope for HADDF-STEM image taking.

In-situ Raman spectra were obtained using a HORIBA LabRAM Odyssey Deep Ultraviolet Laser Confocal Raman Spectrometer with a 532 nm laser excitation. The unsupported Ni₃InC_0.5 powder was filled into the in-situ cell and ramped up to 400 °C with pure CO for 1 h to avoid interference from the characteristic peaks of zirconia support and passivated oxide overlayer. Afterwards, the sample was treated with 25 vol% CO₂/H₂ for 2 h.

The surface chemical states of representative catalysts were determined by XPS using a Thermo Fisher K-Alpha spectrometer (Al Kα radiation source, E = 1486.6 eV). For quasi-in-situ tests, check valves at both ends of the glass reactor were closed after reduction or carburization. Then, the catalysts were transferred through a glove box to the specialist vacuum sample holder of the K-Alpha facility from the glass reactor.

In-situ XPS experiments were carried out using a Thermo Fisher ESCALAB 250Xi spectrometer (Al Kα). Oxide Precursors were loaded and transferred into the vacuum chamber. After recording the initial states, H₂ was introduced into the system to reduce the samples at 400 °C for 2 h, followed by cooling to room temperature and vacuuming the system for data acquisition. Subsequently, the feed was switched to mixed gas (25 vol% CO₂/H₂) to carburize the reduced IMCs at 400 °C for 2 h before final scanning. All spectra were calibrated to the adventitious carbon C 1 s peak at 284.8 eV and further analyzed using the Avantage software.

Although XAS by transmission mode shows more bulk information of the samples, by in-situ experiment and suitable sample comparison, it can still be used to probe the changes on the surface. In-situ XAS measurements on Ni K-edge were conducted in transmission mode using wafer-like samples (I.D. 10 mm pellets) in a customized reaction cell at the X-ray absorption fine structure for catalysis (XAFCA) beamline of the Singapore Synchrotron Light Source (SSLS). Specifically, the NiInZrO precursor sample was sequentially treated with H₂, 25 vol% CO₂/H₂, CO/H₂, and then again 25 vol.% CO₂/H₂ at a constant temperature of 400 °C, with each gas atmosphere maintained for 1 h. Ni K-edge XAS spectra were collected throughout the process. Quasi-in-situ XAS on the Ni K-edge was measured under the transmission mode at Synchrotron Light Research Institute (Beamline 1.1 W Multiple X-ray Techniques) in Thailand. Like quasi-in-situ XPS tests, the activated catalysts were transferred into a glove box to minimize oxidation. An appropriate amount of sample was spread evenly in a plastic frame and sealed with Kapton tape⁴⁴.The XAS data were processed using Demeter software packages⁶⁷.

Temperature-programmed experiments were performed using a Micromeritics AutoChem Ⅱ 2920 TPR/TPD instrument equipped with a Hiden QIC-20 mass spectrometer (MS). H₂-temperature programmed reduction (H₂-TPR) was used to analyze the reducibility of the oxide precursors. Approximately 100 mg of the precursor was pretreated in an Ar flow (30 mL min⁻¹) at 200 °C for 1 h, then cooled to 60 °C to remove the moisture and other contaminants. The sample was then reduced under a flow of 10 vol% H₂/Ar (30 mL min⁻¹) and heated to 800 °C with a ramp rate of 10 °C min⁻¹. A cooling trap was installed between the sample and the thermal conductivity detector (TCD) to remove the water formed during the reduction.

The same instrument was used for TPD of H₂, CO₂, and CO. Approximately 100 mg of the sample was treated at 400 °C for 1 h by 30 mL min⁻¹ of pre-treatment gas (H₂ for IMC catalysts or mixed gas for carbide catalysts), followed by flushing with He (30 mL min⁻¹) for 30 min to clean its surface. After cooling to 60 °C, the sample was exposed to H₂ flow (30 mL min⁻¹) for 30 min for saturate adsorption of H₂, then purged with carrier gas, Ar (30 mL min⁻¹), for 30 min. Once the TCD signal baseline stabilized, the sample was heated to 800 °C at a rate of 10 °C min⁻¹, recording the TPD profile from TCD. For the TPD of CO₂ and CO, the sample was treated similarly, except that CO₂ or CO replaced H₂ to saturate the surface, and carrier gas was changed to He (30 mL min⁻¹). The MS signals of H₂, H₂O, CO, and CO₂ (m/z  =  2, 18, 28, and 44, respectively) were recorded simultaneously as the sample was heated to record the TPD profile.

H₂-TPR coupled with MS was employed to analyze the surface In₂O₃ species on both fresh and spent catalysts. After synthesis or catalytic evaluation, approximately 0.1 g of the sample was cooled to 60 °C under an Ar flow, with the reactor effluent connected to a Shimadzu GCMS-QP2020 NX instrument. Subsequently, the sample was subjected to reduction under a 10 vol% H₂/Ar (30 mL min⁻¹) while being heated to 700 °C at a ramp rate of 10 °C min⁻¹. The actual temperature of the fixed bed was monitored using a thermocouple.

The adsorption, formation, and consumption of the intermediates during the reaction were monitored by in-situ Fourier transform infrared spectroscopy (in-situ FTIR) in transmission mode using a Thermo Fisher Nicolet iS50 spectrometer equipped with a Specac high-temperature high-pressure (HTHP) cell and an MCT/A detector. Each spectrum was recorded at 4 cm⁻¹ resolutions with 32 scans. To reduce the strong infrared light adsorption by Ni-base IMC and carbide catalysts, 5 mg of the sample was diluted with 30 mg CaF₂, ground, and pressed into a 12 mm diameter wafer. The wafer was activated with 30 mL min⁻¹ of 25 vol% CO₂/H₂ at 400 °C for 1 h, purged with Ar (30 mL min⁻¹) for 30 min to remove the adsorbates, and then cooled to the target temperature for background spectra collection. For the CO₂ adsorption and hydrogenation process, the sample was exposed to CO₂ (30 mL min⁻¹) at 280 °C and ambient pressure until the spectra stabilized. The system was then pressurized to 2 MPa with CO₂, followed by switching the feed to H₂ to monitor changes in the adsorbed CO₂ species for 60 min. For the CO₂ hydrogenation process, 30 mL min⁻¹ of mixed gas (H₂: CO₂ = 3/1) was used to activate the catalyst. After cooling to 280 °C under Ar flow, the activated sample was exposed to a mixed gas (H₂: CO₂ = 3/1, 30 mL min⁻¹) at ambient pressure until the spectra stabilized, then pressurized to 2 MPa.

The influence of temperature on intermediates was studied by monitoring the reaction at different temperatures (220/260/280, and 300 °C). After mixed gas activation, the cell was cooled to 100 °C, and the mixed gas (H₂: CO₂ = 3/1) was introduced at 30 mL min⁻¹ to pressurize to 0.3 MPa. After equilibrium, the sample was heated to 140 °C, 180 °C, 220 °C, 260 °C, 280 °C, and 300 °C at a rate of 10 °C min⁻¹, with steady spectra collected at each temperature.

To study the consumption speed of intermediates during hydrogenation, after saturating the sample’s surface at 280 °C and 2 MPa under mixed gas flow (H₂: CO₂ = 3/1, 30 mL min⁻¹), the sample was purged by 30 mL min⁻¹ of H₂ for 150 min. FTIR spectra were collected every 10 s during all processes.

Catalyst evaluation

CO₂ hydrogenation was conducted in a fixed-bed continuous-flow stainless steel reactor. Before the evaluation, the prepared catalysts were pressed, crushed, and sieved into 40–60 mesh granules to minimize the impact of internal diffusion. Approximately 0.1 g of the sieved catalyst granules was loaded into a reactor tube with an external diameter of 3/8 inch and a thickness of 1 mm. To remove the passivated layer on the nanoparticle surface, Ni-based IMC catalysts were activated using 30 mL min⁻¹ of H₂ at 400 °C for 1 h under ambient pressure. For the Ni-based carbide catalysts, 30 mL min⁻¹ of mixed gas (H₂/CO₂ = 3/1) was used for pretreatment instead. For general catalyst evaluation, after cooling to the target reaction temperature, the reactor was pressurized to 4 MPa with mixed gas (H₂:CO₂:N₂ = 72/24/4, with N₂ as an internal standard). To prevent condensation of the products, H₂O and CH₃OH, the gas line between the reactor and the online gas chromatography (GC, Shimadzu GC-2014C) was maintained at 150 °C. The gas products were directly analyzed by the online GC, equipped with two loops: one connected to two packed columns, porapak N (1 m × 3 mm) and TDX-01 (3 m × 3 mm), with a TCD, and the second connected to a capillary column, SH-Q-BOND (30 m × 0.32 mm), with a flame ionization detector (FID). The calculations of catalytic activity, including CO₂ conversion (${X}_{{{\mbox{CO}}}_{2}}$), methanol selectivity (S_MeOH), and methanol space-time yield (STY_MeOH), are detailed in the Supplementary Information, along with the procedure for the reaction kinetics study.

Computational details

Spin-polarized DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP, version 5.4.1), with the generalized gradient approximation (GGA) of Perdew-Burke-Ernzerhof (PBE) to describe electron exchange and correlation^68,69. The cut-off energy of the plane-wave basis set was set to 400 eV⁷⁰. The convergence tolerances for the energy optimization criterion were set to 10^-5eV, and the atomic forces in the optimized structures were smaller than 0.03 eV Å⁻¹.

We chose the c-In₂O₃ (100) surface termination as it is the most stable surface under CO₂ hydrogenation reaction conditions⁷¹. Meanwhile, according to our AC-STEM observation, the exposed facet of the surface c-In₂O₃ layer on Ni₃InC_0.5 NPs is (100) plane. A 3×3-unit cell was used to construct a c-In₂O₃ (100) slab model consisting of 24 In atoms and 32 O atoms distributed in three In-O-In layers. The atoms in the bottom layer were frozen, while the top two In-O-In layers were relaxed. The Brillouin zone was sampled using a 3 × 3 × 1 k-point grid based on the Monkhorst-Pack method. Considering the surface oxidation mechanism of Ni₃InC_0.5 under reaction conditions and the core-shell structure of the Ni₃InC_0.5-In₂O₃, when metallic In atoms are oxidized, Ni atoms are left and fixed between the bulk Ni₃InC_0.5 and surface In₂O₃ layer. Thus, the model of Ni (111)-supported In₂O₃ clusters was constructed to represent the interface structure. The bottom two layers of Ni atoms were frozen, while one layer of Ni and supported (In₂O₃)₂ clusters were allowed to perturb. A 1 × 1 × 1 k-point grid generated with the Monkhorst-Pack scheme was found to give converged results. More detailed theoretical calculations were provided in Supplementary Information.