Introduction

The strong metal–support interaction (SMSI), characterized by the migration of the support over metal nanoparticles (NPs) on a supported-metal catalyst1, offers a powerful strategy in heterogeneous catalysis for engineering new catalytic sites through interfacial metal-support interactions2,3. Classical SMSI typically proceeds via the formation of mobile species on the support surface, interfacial mass transfer, and the development of encapsulating overlayers, which governed by the thermodynamic imperative to reduce the overall surface energy of the catalyst system. However, uncontrolled migration of the support may obstruct the metal active sites, and the high reduction temperature conditions required for thermally induced SMSI (H2, ≥ 500 °C) can also lead to sintering of the metal NPs4.

A distinct variant of SMSI—adsorbate-induced SMSI (A-SMSI) — has recently emerged, wherein reactive intermediates (e.g., alcohols5,6, levulinic acid7, or CO2-derived species8,9,10) drive the migration of the support. In contrast to SMSI, A-SMSI occurs at markedly lower temperatures (~ 250 °C), owing to the reduced energy input required for its formation reactions11. This strategy effectively circumvents catalyst sintering and deactivation commonly associated with SMSI that necessitates high temperatures. Moreover, A-SMSI facilitates the formation of permeable overlayers that retain accessibility to active sites, enabling favorable modulation of reaction pathways without significantly compromising overall catalytic activity8,11.

Inspired by these advances, considerable efforts have been devoted to manipulating A-SMSI overlayer structures and their associated catalytic performances across various systems and applications12,13. For instance, Matsubu et al. demonstrated that the nature of the support plays a critical role: Rh catalysts supported on reducible oxide supports (TiO2, Nb2O5) could facilitate the formation of HCOx-functionalized encapsulation that benefits the tuning of CO2-reduction selectivity8. Furthermore, Li et al. reported that pretreatment temperature affects the concentration of oxygen vacancy in oxide supports, thereby critically influencing the stability of A-SMSI overlayer and the performance of Ru/TiO2 catalysts14. Recently, by using in situ environmental transmission electron microscopy (ETEM), Deng et al. further demonstrated that pretreatment in CO2 and H2 could dynamically and reversibly alter the morphology of Cu@TiOx catalysts, emphasizing that metal–support interfaces are highly sensitive to the chemical environment. These findings highlight the inherently dynamic nature of A-SMSI and the importance of in situ characterization15. Despite these pioneering efforts, many fundamental questions regarding the creation and manipulation of A-SMSI remain unresolved16,17. In particular, it remains unclear whether and how the atomic structure of the exposed support surface affects the formation, structure, and catalytic performance of A-SMSI. This knowledge gap largely stems from the lack of atomic-level, in situ insights into the overlayer dynamics.

Herein, taking the Ni/TiO2 as a model catalyst18,19,20, we carry out a systematic study of the facet-dependent A-SMSI behavior in CO2 hydrogenation. To fairly compare the Ni catalyst supported on different TiO2 facets, three well-defined Ni/TiO2 model catalysts with distinct exposed TiO2 facets are constructed: TiO2 nanorods (TiO2-R, {100}), bipyramidal nanocrystals (TiO2-B, {101}), and nanosheets (TiO2-S, {001}). In situ ETEM directly visualizes the dynamic evolution of A-SMSI overlayers, revealing complete and partial encapsulation of Ni NPs on {100} and {101} facets, respectively, but none on {001}. The exposed facets regulate both CO2 activation into induction intermediates (HCOx*) and TiO2-x formation of at the Ni-TiO2 interface. Complementary in situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) and in situ X-ray absorption spectroscopy (XAS) further unveil an interfacial electron transfer from TiO2-x to Ni, stabilizing COOH* intermediates that dehydroxylate to form CO. This facet-dependent A-SMSI fundamentally reshapes product selectivity: from 88% CH4 on Ni/TiO2-{001} to over 83% CO on Ni/TiO2-{100}. These findings deepen the mechanistic understanding of A-SMSI and offer a rational strategy for tailoring interfacial properties in oxide-supported catalysts.

Results

Characterization of Ni/TiO2 catalysts with different support facets

To investigate the effect of TiO2 crystal facets on A-SMSI, three well-defined Ni/TiO2 model catalysts were prepared by loading Ni NPs onto TiO2 supports with distinct morphologies. TiO2-R predominantly exposes the surface of {100} facet, with diameters of 40-60 nm and lengths of 200–300 nm; TiO2-B primarily exposes the {101} facet, with lengths of 100-120 nm; TiO2-S mainly exposes the {001} facet, with lengths of 60-110 nm and thicknesses of 5-10 nm, as shown in the transmission electron microscopy (TEM) images in Fig. 1a–c. Atomic-resolution high-angle angular dark field scanning TEM (HAADF STEM) images further identify the interplanar spacings of the three TiO2 supports as 3.78 Å (Fig. 1d), 3.51 Å (Fig. 1e), 2.38 Å (Fig. 1f), corresponding to the {100}, {101}, {001} facets, respectively. These results are consistent with the high-resolution TEM (HRTEM) images shown in Supplementary Fig. 1. The Ti columns could also be clearly identified due to the Z contrast21, which well matches the bulk-truncated surfaces of each facet. The thermodynamically stable {100} facet is exposed by 5-fold coordinated Ti atoms (Ti5c), with 2- and 3-fold oxygen atoms (O2c and O3c); the {101} facet is additionally exposed by Ti6c; while the {001} facet is only exposed by Ti5c and O2c22,23. Given the sensitivity of TiO2 to electron beam irradiation, all subsequent experiments were conducted under relatively low-dose conditions, referring to previous studies24,25,26,27. Moreover, control experiments confirmed that the electron beam did not obviously alter the relative stability and structure of different TiO2 facets (Supplementary Figs. 2, 3).

Fig. 1: Structural characterization of Ni/TiO2 catalysts.
Fig. 1: Structural characterization of Ni/TiO2 catalysts.
Full size image

a–c Low-magnification TEM images of the TiO2 supports: TiO2-R (a), TiO2-B (b), TiO2-S (c). d–f High-magnification HAADF STEM images and corresponding atomic structural models of TiO2 {100} (d), {101} (e), {001} (f) facets. g–i EDX elemental maps of the Ni/TiO2 catalysts: Ni/TiO2-R (g), Ni/TiO2-B (h), Ni/TiO2-S (i), showing the distribution of Ni, Ti and O. j–l Violin plots of kernel density-estimated Ni particle size distributions (n = 100) for Ni/TiO2-R (j), Ni/TiO2-B (k), and Ni/TiO2-S (l). Dashed lines denote the median particle size, and individual dots represent the measured particle diameters.

Ni/TiO2-S, Ni/TiO2-B, and Ni/TiO2-R catalysts were prepared by loading Ni onto the corresponding supports, with Ni loadings of 9.2 wt.%, 9.6 wt.%, and 9.4 wt.%, respectively, as determined by inductively coupled plasma-optical emission spectrometer (ICP-OES). HAADF STEM imaging and energy dispersive X-Ray (EDX) spectroscopy elemental mappings confirm the uniform distribution of Ni NPs on TiO2 (Fig. 1g–i). The size distribution violin plots of Ni NPs, as determined by analyzing 100 NPs per sample, are shown in Fig. 1j–l: 14.66 nm on Ni/TiO2-R, 16.33 nm on Ni/TiO2-B, and 12.55 nm on Ni/TiO2-S. The comparable sizes in the three samples provide an ideal platform to study the facet-dependent behavior28.

Evolution of encapsulation overlayer in various Ni/TiO2 catalysts

To explore the facet-dependent behavior, the ETEM experiments were employed to observe the structural evolution of various Ni/TiO2 catalysts under reactive gas environments. Before the performance test, reduction at elevated temperature (typically 400–500 °C) is a general routine for Ni/TiO2 catalyst activation to obtain metallic Ni. While in the ex-situ characterization, this active metallic Ni catalysts are prone to surface oxidation upon ambient air exposure29,30, forming superficial NiO layers, as revealed in Supplementary Fig. 4. Therefore, the ex-situ observation of Ni/TiO2 catalysts could not represent the realistic state of the catalysts in the reaction. To precisely probe the intrinsic physicochemical properties of Ni-based catalysts, the in situ study of the Ni/TiO2 catalysts under the pretreatment and reaction environments are indispensable for constructing a reliable structure-performance relationship. The corresponding in situ ETEM, HAADF STEM, and electron energy loss spectroscopy (EELS) results for Ni/TiO2-R, Ni/TiO2-B, and Ni/TiO2-S are presented in Fig. 2 and Supplementary Fig. 5.

Fig. 2: The tunable Ni/TiO2 interface under different gas environments.
Fig. 2: The tunable Ni/TiO2 interface under different gas environments.
Full size image

In situ ETEM images of Ni/TiO2-R in (a) reducing atmosphere (H2/N2 = 10/90, 5 × 10−2 Pa, 400 °C), and (b–d) CO2 hydrogenation conditions (CO2/H2/N2 = 15/15/70, 5 × 10−2 Pa, 350 °C). e HAADF STEM image of Ni/TiO2-R after reaction, with (f) magnified view of the boxed region in (e). g EELS mappings (h) Ti L2,3-edge EELS line scan from the selected region in (e). i HAADF STEM image of Ni/TiO2-B, and (j) Ti L2,3-edge EELS spectra for the selected regions in (i). k HAADF STEM image of Ni/TiO2-S, and (l) Ti L2,3-edge EELS spectra for the selected regions in (k).

ETEM imaging after the reduction pretreatment confirmed the complete removal of the NiO overlayer for all the catalysts (Fig. 2a and Supplementary Fig. 5a, i), and fast fourier transform (FFT) analysis further validated the metallic state of Ni, as exemplified by Ni/TiO2-R (Supplementary Fig. 6a). The TiO2-R support was imaged along the [010] direction with (100) facet exposure, while the located Ni was oriented along the [011] direction, exposing the (− 200), (1–11), and (− 1–11) facets (Supplementary Fig. 6a).

Under CO2 hydrogenation conditions, the three catalysts exhibited distinct structural evolution behaviors, as observed from the in situ ETEM images in Fig. 2b–d, and Supplementary Fig. 5. In particular, after 60 min of reactions, the Ni NPs in Ni/TiO2-R were completely covered by encapsulation layers (Fig. 2d), while partial encapsulation was observed in Ni/TiO2-B (Supplementary Fig. 5d) and no encapsulation layer formed in Ni/TiO2-S (Supplementary Fig. 5l).

In situ ETEM imaging of the Ni/TiO2-R interface upon 15 min exposure to reaction conditions (Fig. 2b) revealed the formation of a 1.69 nm-thick amorphous overlayer on the Ni surface, consistent with the dynamic interfacial restructuring previously reported by Matsubu et al. for Rh/TiO28. The Ni NP showed slight morphology changes, with the length parallel to the support surface increasing from 13.98 nm to 14.46 nm (Fig. 2a, b), confirming slight flattening of the Ni NP due to overlayer growth. The FFT patterns of the dashed region in Fig. 2b showed no additional diffraction spots compared to the initial state (Supplementary Fig. 6b), indicating the absence of Ni oxidation. After 30 min of exposure (Fig. 2c), the overlayer grew further and induced a slight counterclockwise rotation of the Ni particle, exposing the {111} facet. Complete encapsulation of the particle was achieved after 60 min, as observed from Fig. 2d. The formation of the encapsulating overlayer requires the synergy of CO2 and H2 rather than either gas alone. In contrast to H2, which caused no significant morphological change (Fig. 2a), CO2 alone induced surface amorphization of the Ni nanoparticles but did not lead to encapsulation. This amorphization, evidenced by the appearance of blurred lattice fringes (Supplementary Fig. 7), aligns with established understanding, which attributes it primarily to the dissociative adsorption of CO2 that reconstructs the Ni surface atoms31. As TiO2 is known to be sensitive to electron beam irradiation in TEM32, the total exposure dose on the sample was carefully limited to below 6.14 × 103 e Å−2 to minimize beam damage.

To further identify the composition of the encapsulating species, HAADF STEM imaging of Ni/TiO2-R was conducted after reaction under atmospheric conditions (CO2/H2/N2 = 15/15/70, 350 °C) (Fig. 2e). The enlarged image (Fig. 2f) revealed a ~ 1.7 nm-thick overlayer with brighter contrast relative to the Ni core, consistent with the ~ 1.69 nm amorphous layer observed via in situ ETEM (Fig. 2b). EELS mappings confirmed that this overlayer was composed of titanium species (Fig. 2g). To probe its electronic properties, EELS line scan of the Ti L2,3-edge was collected from bulk TiO2 to vacuum (indicated in Fig. 2e). As shown in Fig. 2h, the L2,3-edge of Ti species at the edge showed a slight shift (~ 0.6 eV) toward lower energy relative to TiO2, indicating the stabilization of Ti3+ species33,34 and providing direct spectroscopic evidence for the formation of SMSI-associated TiO2-x under reaction conditions.

A TiO2-x overlayer was observed over Ni/TiO2-B as well (Supplementary Fig. 5a–d), whereas no analogous interfacial structure emerged on Ni/TiO2-S under identical reaction conditions (Supplementary Fig. 5i–l). Notably, EELS analysis of the Ti L-edge further confirmed partial encapsulation of Ni particles in Ni/TiO2-B (Fig. 2i, j), which were absent in Ni/TiO2-S (Fig. 2k, l).

Collectively, these results suggest that TiO2-x overlayer formation is more favorable on Ni/TiO2-R and Ni/TiO2-B than on Ni/TiO2-S. Consequently, facet-dependent differences in TiO2 lead to distinct encapsulation structural behaviors: full encapsulation on {100}-faceted Ni/TiO2-R, partial encapsulation on {101}-faceted Ni/TiO2-B, and an unencapsulated metallic surface on {001}-faceted Ni/TiO2-S.

Revelation of catalyst adsorbates and the mechanism of encapsulation induction

It has been reported that a high surface coverage of HCOx* absorbates under CO2/H2 atmospheres could promote H2O desorption8,14, which in turn creates oxygen vacancies and forms non-stoichiometric TiO2-x species that encapsulate the metal35. The facet-dependent nature of TiO2 leads to variations in the adsorption of carbonaceous species, due to differences in atomic arrangements and coordination environments across exposed facets36. To acquire solid evidence for surface species and to elucidate the facet-dependent encapsulation mechanism, in situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) was employed on three Ni/TiO2 catalysts under reaction conditions.

As shown in Fig. 3a, Ni/TiO2-R exhibited absorption bands at ~ 1595 and ~ 1380 cm−1, corresponding to a high coverage of adsorbed HCOO*37,38, along with bands at ~ 1428 and ~ 1248 cm−1 assigned to HCO3*8,39. In contrast, Ni/TiO2-B initially formed monodentate carbonate (~ 1530 cm−1 and ~ 1325 cm−1) and HCO3* (~ 1440 cm−1) after ~5 min of CO2 pre-activation, which gradually transformed into HCOO* (~ 1585 cm−1 and ~ 1380 cm−1) and other carbonate species (Fig. 3b)40. These spectral evolutions highlight the pivotal role of HCOO* among the HCOx* species. Compared to the other two catalysts, Ni/TiO2-S showed significantly weaker bands for surface carbonaceous species (Fig. 3c), and HCOO* species disappeared rapidly upon heating (Supplementary Fig. 8). These results indicate that HCOO* accumulates and stabilizes only on Ni/TiO2-R and Ni/TiO2-B. Further comparison under H2 purge (Fig. 3d) revealed that after 16 min, the HCOO* signal on Ni/TiO2-B (~ 1585 cm−1) decreased by ~ 74%, whereas only a ~ 42% reduction was observed for Ni/TiO2-R (~ 1595 cm−1), indicating relatively weaker adsorption on the former. This trend is consistent with CO2 temperature-programmed desorption (CO2-TPD) results (Fig. 3e), which show that CO2 adsorption at moderate basic sites (150–300 °C)—precursors to HCOO* formation41—follows the order Ni/TiO2-R (55.8 μmol g−1) > Ni/TiO2-B (48.3 μmol g−1) > Ni/TiO2-S (36.7 μmol g−1). The greater CO2 uptake capacity of Ni/TiO2-R thus explains its enhanced ability to accumulate HCOO* and thereby contribute to stronger A-SMSI behavior. Furthermore, H2 temperature-programmed reduction (H2-TPR) profiles showed higher temperature shifts in Ni/TiO2-R and Ni/TiO2-B compared to Ni/TiO2-S, indicating stronger metal-support interactions (Fig. 3f)37,42.

Fig. 3: Investigation on the reaction mechanism over Ni/TiO2 catalysts.
Fig. 3: Investigation on the reaction mechanism over Ni/TiO2 catalysts.
Full size image

a–c In situ DRIFTS results of (a) Ni/TiO2-R, (b) Ni/TiO2-B, and (c) Ni/TiO2-S under reaction conditions (CO2/H2/N2 = 15/15/70, 350 °C) for 30 min. d Normalized intensities of the critical surface adsorbate species under reaction conditions and in H2 (H2/N2 = 15/85, 350 °C, 20 min) as a function of reaction time (HCOO*: ~ 1591 cm−1, COOH*: ~ 1544 cm−1 for Ni/TiO2-R; HCOO*: ~ 1585 cm−1 for Ni/TiO2-B). e CO2-TPD profiles of various Ni/TiO2 catalysts. f H2-TPR profiles of various Ni/TiO2 catalysts. g Adsorption energies (Eads) of HCOO* adsorbed at three different sites on the TiO2-x /Ni (111) model. h DFT transition-state calculations for HCOOH* over Ni/TiO2-R.

The density functional theory (DFT) calculations further rationalize these observations using a TiO2-x /Ni (111) heterostructure model (Supplementary Fig. 9)3. The calculations showed that HCOO* can be stably adsorbed on the TiO2-x surface (Fig. 3g), with adsorption energies (Eads) of − 4.60 eV, − 4.70 eV and − 3.72 eV at three representative sites. In all cases, the negative Eads indicate thermodynamically favorable binding, suggesting that HCOO* adsorption on the TiO2-x overlayer of Ni/TiO2-R provides additional stabilization to the reduced surface, which in turn favor the formation of TiO2-x overlayers and promotes encapsulation.

Combined DRIFTS and DFT results indicate that the adsorption strength and coverage of HCOO* species critically affect the extent of A-SMSI. On Ni/TiO2-R, strongly bound HCOO* stabilizes the TiO2-x overlayer, while on Ni/TiO2-B, weaker adsorption only enables partial encapsulation, as evidenced in Fig. 2i, j and Supplementary Fig. 5. In contrast, the absence of stable HCOO* intermediates on Ni/TiO2-S eliminates this stabilizing effect, preventing overlayer formation.

As noted, once TiO2 is reduced to non-stoichiometric TiO2-x species, migration of the reduced oxide can lead to encapsulation of the active metal35. Previous studies have shown that formic acid (HCOOH*) can create oxygen vacancies on TiO2 via H2O desorption, leaving a HCOO-covered reduced surface (A-SMSI)8,43. To probe this mechanism, DFT calculations were performed for the HCOOH* intermediate on Ni/TiO2-R. Although HCOOH* can decompose into HCO* and OH* during CO2 hydrogenation over Ni-based catalysts (with HCO* subsequently hydrogenating to CH4)44, the calculations indicate a more favorable dissociation into HCOO* and H* on Ni/TiO2-R.

Two competing pathways were evaluated: (path a) the conventional hydrogenation sequence and (path b) the vacancy formation route. Dissociation along path a requires ΔE = 0.57 eV, whereas path b proceeds with a much lower barrier of ΔE = 0.19 eV (Fig. 3h). These results clearly demonstrate that the oxygen-vacancy formation pathway is kinetically highly favorable, thereby providing direct support for the proposed mechanism. Moreover, DFT calculations revealed that interfacial oxygen vacancy formation is most favorable on Ni/TiO2-R (2.67 eV) among three catalysts, providing a fundamental basis for TiO2-x overlayer formation (Supplementary Figs. 10–15).

Notably, both gaseous CO ( ~ 2162 cm−1) and COOH* (1544 cm–1) bands45,46 were observed over Ni/TiO2-R (Fig. 3a), aligning with its CO-rich selectivity (discussed later). Moreover, COOH* (61%) decayed faster than HCOO* (42%) after 16 min in H2 (Fig. 3d), suggesting COOH* is the reactive intermediate for CO formation47,48, while HCOO* mainly stabilizes the TiO2-x overlayer. In contrast, no CO-related bands were observed over Ni/TiO2-B and Ni/TiO2-S (Fig. 3b, c). Instead, new peaks at 3064 and 3014 cm−1 corresponding to C-H stretching vibrations of CH4 appeared on Ni/TiO2-B (Fig. 3b)49, while gaseous CH4 over Ni/TiO2-S was confirmed by bands at ~ 3014 and 1304 cm-1(Fig. 3c). These observations are consistent with CO2 hydrogenation results, where CH4 is the dominant product over Ni/TiO2-B and Ni/TiO2-S (details provided later).

Collectively, these results demonstrate that facet-dependent structural configurations dictate the formation and stabilization of HCOO* species, which play a dual role in the induction A-SMSI. Specifically, HCOO* not only facilitates vacancy formation—contributing to the TiO2-x generation and subsequent encapsulation—while also stabilizes the TiO2-x overlayer through adsorption once encapsulation has occurred.

Electronic structure evolution of Ni induced by A-SMSI

The presence of A-SMSI on Ni/TiO2-R has been conclusively demonstrated by ETEM and DRIFTS, with a key feature being electron transfer between the metal and the support8,11. EELS measurements confirm the presence of Ti3+ species within the TiO2-x overlayer (Fig. 2h). To further elucidate the evolution of the electronic structure of Ni NPs, in situ X-ray absorption spectroscopy (XAS) was performed (Fig. 4). The normalized Ni K-edge X-ray absorption near edge structure (XANES) spectra of Ni/TiO2-S and Ni/TiO2-R after in situ hydrogen reduction exhibit slight edge shifts toward higher photon energy relative to Ni foil (Fig. 4a), indicating a comparable Ni oxidation state, with the coexistence of metallic Ni and NiO. These findings imply that the valence state of Ni alone does not govern the onset of A-SMSI.

Fig. 4: In situ XAS analyses of Ni electronic structure evolution.
Fig. 4: In situ XAS analyses of Ni electronic structure evolution.
Full size image

a Normalized Ni K-edge XANES spectra of Ni/TiO2-R and Ni/TiO2-S in H2 (H2/Ar = 10/90) at 400 °C. b Normalized Ni K-edge XANES spectra of Ni/TiO2-R under different conditions (H2/Ar = 10/90; CO2/Ar = 10/90; CO2/H2/Ar = 10/10/80). c Normalized Ni K-edge Fourier-transform EXAFS spectra under corresponding conditions. d–f WT analyses of (d) NiO/TiO2-R, (e) Ni foil, and (f) Ni/TiO2-R under reaction conditions.

To capture the dynamic evolution of the Ni electronic structure, Ni/TiO2-R was sequentially exposed to different gas exposures, each held for ~ 40 min to allow stabilization (Fig. 4b). Under Ar at room temperature, the extended X-ray absorption fine structure (EXAFS) spectra of unreduced NiO/TiO2-R as well as the Fourier transform and wavelet transform (WT) results (Fig. 4c, d and Supplementary Fig. 16) confirm that NiO dominates on the catalyst. Note that the distances observed in the Fourier transformed spectra (Fig. 4c) are systematically shorter than the actual absorber-backscatter distances (Supplementary Table 1), owing to phase shift effects. After H2 reduction at 400 °C (green curve in Fig. 4b), a mixed Ni/NiO phase was obtained, as discussed above. Subsequent introduction of CO2 (CO2 /Ar = 10/90) at 350 °C resulted in a notable edge shift to higher photon energy (blue curve in Fig. 4b), indicative of partial reoxidation of Ni.

Upon exposure to H2/CO2 (CO2/H2/Ar = 10/10/80), the edge clearly shifted back toward lower energy, approaching that of Ni-foil (yellow curve) shown in Fig. 4b, suggesting that Ni gains electrons under reaction conditions. X-ray photoelectron spectroscopy (XPS) analysis further corroborated this conclusion, as evidenced by the shift of the Ni⁰ peak in Ni/TiO2-R after reaction to a lower binding energy (851.9 → 851.5 eV) compared to the reduced sample (Supplementary Fig. 17). This is likely driven by the accumulation of adsorbed HCOO* intermediates, which induce the formation of a TiO2-x encapsulating layer. However, compared to the Ni foil (Fig. 4c, e), the Ni particle size over Ni/TiO2-R is too large to notably contribute to metal-support bonding (Ni-Ti) (Fig. 4c, f)50. Therefore, changes in disorder (σ²) were analyzed using EXAFS, as tabulated in Supplementary Table S1. The highest static disorder was observed under reaction conditions (11.6 × 10⁻2 Å2), implying strong interaction between Ni particles and decorating TiO2-x species present in Ni/TiO2-R50,51. The observed reduction in Ni valence state is consistent with electron transfer from Ti to Ni, facilitated by the modification of A-SMSI.

To quantitatively elucidate this electron transfer mechanism, the reactive intermediate COOH* was introduced onto a pre-adsorbed TiO2-x/Ni (111)–HCOO* structure (representative of A-SMSI), as shown in Fig. 3g. Differential charge density and Bader charge analyses of the resulting TiO2-x/Ni (111)–HCOO*–COOH* configuration revealed a net electron transfer of 0.17 e⁻ from the TiO2-x overlayer to Ni (Supplementary Figs. 18, 19), consistent with the XAS measurements. Furthermore, the calculated Eads of COOH* on TiO2-x/Ni (111)–HCOO* confirmed that the overlayer provides a favorable environment for intermediate stabilization, consistent with the active role of A-SMSI systems proposed by Matsubu et al.8.

Collectively, these XAS and DFT findings elucidate the electron transfer mechanism governing A-SMSI evolution under CO2 hydrogenation conditions. On Ni/TiO2-R, HCOO* induces the formation of a TiO2-x encapsulation layer (EELS Ti L2,3-edge shift: ΔE = − 0.6 eV, Fig. 2h), enabling directional electron transfer through TiO2-x to Ni. The non-stoichiometric TiO2-x thus acts as the functional component for COOH* adsorption, as supported by both DRIFTS and DFT. This process simultaneously decreases the surface coverage of atomic hydrogen and thereby suppressing the further hydrogenation of CO generated via the dehydroxylation step of COOH*8,14. These findings provide mechanistic insight into the evolution of A-SMSI and the induced selectivity shift in CO2 hydrogenation, as elaborated in the following section.

Tuning catalytic performance by A-SMSI

The influence of A-SMSI on the catalytic performance of CO2 hydrogenation (CO2/H2/N2 = 15/15/70) was further evaluated over different Ni/TiO2 catalysts across 200–500 °C. As shown in Fig. 5a–c, the product selectivity is strongly facet-dependent.

Fig. 5: Catalytic performance of CO2 hydrogenation over different Ni/TiO2 catalysts.
Fig. 5: Catalytic performance of CO2 hydrogenation over different Ni/TiO2 catalysts.
Full size image

a–c Boxplots of CO and CH4 selectivity. Box = Interquartile Range (IQR) (25–75%), whiskers = 1.5 × IQR, and points represent mean values averaged over 10 consecutive time intervals (n = 10). d, e Arrhenius plot of CH4 production rate (rCH4) (d) and CO production rate (rCO) (e). f Long-term stability of Ni/TiO2-S at 350 °C. Reaction conditions: catalyst (1000 mg), 1 bar, CO2 /H2 /N2 = 15/15/70; gas hourly space velocity (GHSV) = 12000 mL g cat−1 h−1).

Generally, increasing temperature thermodynamically favors CO formation via the RWGS reaction. Nevertheless, Ni/TiO2-R consistently exhibited high CO selectivity (83% at 350 °C, even 100% at 230 °C (Fig. 5a), along with the highest CO production rate across the temperature range among three catalysts (Supplementary Fig. 20). In contrast, Ni/TiO2-S strongly favored CH4 formation, reaching selectivity of up to 88% at 350 °C and 98% at 230 °C under identical conditions (Fig. 5c), with the highest CH4 production rate (Supplementary Fig. 20). Ni/TiO2-B exhibited an intermediate behavior with a relatively stable CO and CH4 selectivity profile, with CH4 selectivity remaining above ~ 60% throughout the reaction (Fig. 5b). Importantly, the carbon balance was consistently > 98% across all catalytic experiments (Supplementary Table 2), indicating negligible carbon deposition, in agreement with both TEM (Fig. 2) and DRIFTS observations (Fig. 3). Kinetic experiments were carried out to quantify intrinsic activity of different Ni-TiO2 catalysts (Fig. 5d, e). Reaction temperature of below 400 °C was selected to have the reaction rates determined at relatively low CO2 conversions (i.e., < 20%), avoiding the effect of diffusion control51. Arrhenius plots of Ln (r CH4) and Ln (r CO) versus T-1 plotted by production rates presented good linear correlation for all three catalysts. The apparent activation energies (Ea) of CH4 formation followed the order of Ni/TiO2-R (90.0 kJ mol−1) > Ni/TiO2-B (75.4 kJ mol−1) > Ni/TiO2-S (67.4 kJ mol−1), while those for CO formation were: Ni/TiO2-S (100.3 kJ mol−1) > Ni/TiO2-B (80.7 kJ mol−1) > Ni/TiO2-R (68.1 kJ mol−1). These results indicate that the presence of A-SMSI in Ni/TiO2-R effectively reduces the Ea for CO formation, thereby favoring the RWGS pathway.

Interestingly, for Ni/TiO2-S, CH4 yield remained nearly constant between 350 and 400 °C (17.9 mmol·gcat−1·h−1 → 18.1 mmol·gcat−1·h−1), while dropped to 14.3 mmol·gcat−1·h−1 at 450 °C, indicating a thermal optimum (Supplementary Fig. 20). A 100 h stability test confirmed that Ni/TiO2-S maintained > 88% CH4 selectivity and 20% CO2 conversion, approaching the 23.6% thermodynamic limit of the Sabatier reaction under equimolar CO2/H2 (Fig. 5f). The experimental evidence confirms the absence of A-SMSI in Ni/TiO2-S over long-term reaction.

Collectively, these results unequivocally demonstrate that the extent of A-SMSI—governed by the crystallographic facet of the TiO2 support—plays a pivotal role in steering catalytic pathways and product distribution. Ni/TiO2-S, lacking A-SMSI, catalyzes nearly exclusive CH4 formation. In contrast, the fully developed A-SMSI in Ni/ TiO2-R, characterized by complete TiO2-x encapsulation, strongly promotes CO formation. Ni/TiO2-B, with partial encapsulation, exhibits a balanced CO/CH4 output, bridging the two extremes.

Discussion

In summary, this study systematically elucidates the facet-dependent behavior of A-SMSI in Ni/TiO2 catalysts under CO2 hydrogenation conditions. By employing in situ ETEM, we visualize the dynamic formation of TiO2-x overlayers, revealing that the exposed TiO2 facets dictate both the extent of Ni NPs encapsulation and the resultant reaction pathways. This structural evolution induces electron transfer from TiO2-x to Ni, and stabilizes COOH* intermediates on the overlayer, thereby enabling selective CO formation. Complete overlayer formation on Ni/TiO2-R ({100} facet) leads to enhanced CO selectivity (>83%), whereas Ni/TiO2-S ({001} facet) remains A-SMSI-inactive, favoring CH4 formation (>88% selectivity). These findings establish a direct structure–performance correlation between support facets and catalytic performance, offering a powerful paradigm for engineering interfacial phenomena in oxide-supported catalysts.

Methods

Catalyst synthesis

Three Ni/TiO2 model catalysts were prepared using supports with different well‑defined morphologies.

TiO2-R (nanorod): A two-step hydrothermal approach was employed. First, P25 TiO2 (1000 mg) was converted into Na-titanate by stirring in NaOH (10 M 40 mL) for 10 min, then autoclaving at 120 °C for 24 h. The resulting white precipitate was washed (pH 10.5) and centrifuged (10000 r min−1, 10 min). Next, the wet Na-titanates (4000 mg) were dispersed in deionized water (152 mL) with H2O2 (8 mL), stirred for 15 min, and hydrothermally treated at 180 °C for 20 h, yielding anatase TiO2 nanorods. The precipitate was washed with deionized water/ethanol, then vacuum-dried (60 °C, 12 h).

TiO2-B (bipyramidal nanocrystal): Wet Na-titanate (3000 mg) was mixed with LiCl (393.2 mg) and deionized water (120 mL), stirred for 15 min, and hydrothermally treated at 200 °C for 4 h. The precipitate was washed, centrifuged, and vacuum-dried (60 °C, 12 h).

TiO2-S (nanosheet): Tetrabutyl titanate (10 mL) and HF (1.8 mL, 40 wt%) were mixed, stirred for 5 min, and hydrothermally treated at 200 °C for 20 h. The precipitate was washed with deionized water / ethanol, and NaOH (1 M), then vacuum-dried (60 °C, 12 h).

Ni/TiO2-R, Ni/TiO2-B, and Ni/TiO2-S catalysts were prepared by the wet impregnation method. Each TiO2 support was dispersed in Ni(NO3)2·6H2O solution (30 mL), stirred at 50 °C until solvent evaporation, then dried at 60 °C for 12 h. The solids were calcined in air at 500 °C for 2 h to afford fresh NiO/TiO2, and subsequently reduced in 10% H2/N2 (400 °C, 100 min) to yield the Ni/TiO2-R, Ni/TiO2-B, and Ni/TiO2-S catalysts with a nominal Ni loading of 10 wt%.

Catalytic evaluation

CO2 hydrogenation was carried out in a continuous-flow fixed-bed reactor with a quartz tube (inner diameter, 8 mm; length, 650 mm). Prior to reaction, fresh catalysts (NiO/TiO2-S, NiO/TiO2-B, and NiO/TiO2-R) were in situ reduced under 10% H2/N2 (flow rate: 100 mL min−1) at 400 °C for 100 min. The reaction was conducted in 15 % CO2, 15 % H2, and 70 % N2 at a total flow rate of 200 mL min−1 at ambient pressure, corresponding to a GHSV of 12000 mL gcat−1 h−1. Gas flows were regulated by mass flow controllers, and the outlet flow rate was measured by a membrane flow meter. The outlet gas composition (CO2, H2, CO, CH4) was analyzed online using a NDIR gas analyzer (GASBOARD-3100, Wuhan cubic optoelectronic Co., Ltd.).

The CO2 conversion (X(CO2)), CH4 selectivity (S(CH4)), CO selectivity (S(CO)), CH4 production rate (r(CH4)), CO production rate (r(CO)), and carbon balance were calculated using the following equations:

$$X\left({{CO}}_{2}\right)\%=\frac{{F}_{{in}} * {C}_{{in}}\left({{CO}}_{2}\right)-{F}_{{out}} * {C}_{{out}}\left({{CO}}_{2}\right)}{{F}_{{in}} * {C}_{{in}}\left({{CO}}_{2}\right)}$$
(1)
$$S\left({{CH}}_{4}\right)\%=\frac{n\left({{CH}}_{4}\right)}{\sum n\left({products}\right)}$$
(2)
$$S\left({CO}\right)\%=\frac{n\left({CO}\right)}{\sum n\left({products}\right)}$$
(3)
$$r\left(C{H}_{4}\right)\left({{mmol}}_{{{CH}}_{4}}{\cdot }{g}_{{cat}}^{-1}\;{{\cdot }}\;{h}^{-1}\right)=\frac{{F}_{{in}} * {C}_{{in}}\left({{CO}}_{2}\right) * X\left({{CO}}_{2}\right) * S\left({{CH}}_{4}\right) * 60}{22.4 * {m}_{{cat}}}$$
(4)
$$r\left({CO}\right)\left({{mmol}}_{{CO}}{{\cdot }}{g}_{{cat}}^{-1}{{\cdot }}{h}^{-1}\right)=\frac{{F}_{{in}} * {C}_{{in}}\left({{CO}}_{2}\right) * X\left({{CO}}_{2}\right) * S\left({CO}\right) * 60}{22.4 * {m}_{{cat}}}$$
(5)
$${Carbon\; balance} \;\%=\frac{{F}_{{out}} * {C}_{{out}}\left({CO}\right)+{F}_{{out}} * {C}_{{out}}\left(C{H}_{4}\right)+{F}_{{out}} * {C}_{{out}}\left({{CO}}_{2}\right)}{{F}_{{in}} * {C}_{{in}}\left({{CO}}_{2}\right)}$$
(6)

Here, Fin and Fout (mL min−1) denote the inlet and outlet gas flow rates, respectively. C denotes the concentration (%), mcat denotes the weight of catalyst (g); and n denotes the molar quantity of the products.

Catalyst characterizations

ICP-OES

Ni loadings of the catalysts were quantified via ICP-OES (Agilent 720ES). Approximately 45–50 mg of each sample was digested in acid, diluted to 10 mL (V0), and measured. The elemental concentration (C0, mg L−1) obtained from the instrument was used to calculate the actual Ni mass percentage using the following formula:

$$W\left({Ni}\right)\%=\frac{{C}_{0} * {V}_{0} * f}{m * {10}^{9}} * 100\%$$
(7)

Here, f represents the dilution ratio of C0 (set to 50), and m denotes the samples mass (mg).

TEM characterizations

The morphology, structure and composition of the catalysts were examined by TEM using an FEI Tecnai G2 F20 and an FEI Titan G2 60–300 KV Cs-corrected TEM with HAADF and EDX detectors to determine the distribution of Ni.

STEM characterizations

The HAADF STEM and EELS characterizations were performed on an FEI Titan G2 80–200 TEM/STEM equipped with a spherical aberration corrector (operated at 200 kV).

In situ ETEM experiments

In situ ETEM observations were performed using a Hitachi H-9500 microscope operated at 300 KV, equipped with a heating holder (up to 1000 °C) and enabled pressure around the sample reaching 5 × 10−2 Pa. All data were collected with a Oneview (Gatan) CMOS camera, which was processed using Digital Micrograph software. To minimize electron beam–induced damage, especially considering the high sensitivity of TiO2 supports to beam irradiation, strict low-dose conditions were employed, with the electron dose limited to approximately 6.14 × 103 e Å−2 throughout the ETEM experiments. Furthermore, to avoid cumulative exposure, the beam was activated only during image acquisition and the beam valve remained closed at all other times. For Ni/TiO2-R, -B, -S, individual Ni NPs with representative sizes were selected for tracking the structural evolution. In the in situ experiments, the catalysts were firstly reduced under H2 (5 × 10−2 Pa) at 400 °C to eliminate surface oxides. Subsequently, a CO2/H2/N2 gas mixture (15/15/70) was introduced to simulate CO2 hydrogenation conditions (5 × 10−2 Pa, 350 °C).

In situ DRIFTS experiments

The experiments were performed on a Nicolet iS50 spectrometer (Thermo Scientific) equipped with an in situ IR reaction cell, a gas flow system, and a liquid nitrogen-cooled mercury cadmium telluride (MCT) detector. Spectra were collected with 64 scans at a resolution of 4 cm–1. Prior to measurements, the Ni/TiO2 catalysts were pre-reduced under H2 (H2/N2 = 10/90) at 400 °C, to remove the NiO overlayer formed upon air exposure. The catalysts were then stabilized in N2 at 350 °C to ensure complete removal of adsorbed and absorbed hydrogen species. After acquiring a background spectrum under N2, a reactive gas mixture (CO2/H2/N2 = 15/15/75, 100 mL/min) was introduced at 350 °C to initiate catalytic reactions for the measurements. After a 30 min purge in the reactant mixture, the gas composition was switched to H2 (H2/N2 = 15/85), while maintaining the same flow rate to assess the stability of the adsorbed species.

CO2-TPD

CO2-TPD experiments were conducted on a Microtrac BELCat II. Each sample (100 mg) was loaded into a quartz reaction tube and pretreated by heating from room temperature to 400 °C at a rate of 1 °C min−1 under a H2 flow (H2/Ar = 10/90, 30 mL min−1) for 1 h to remove the surface NiO layer. After cooling to 50 °C, the sample was saturated with a CO2/He mixture (CO2/He = 10/90, 30 mL min–1) for 1 h. Subsequently, the system was purged with pure He (30 mL min−1) for another hour to remove weakly physically adsorbed CO2. Finally, the desorption profile was recorded by heating the sample to 500 °C at 10 °C min−1 under He flow, with CO2 desorption monitored using a thermal conductivity detector (TCD).

H2-TPR

H2-TPR experiments were carried out on a Microtrac BELCAT II instrument. Fresh NiO/TiO2-R, NiO/TiO2-B, and NiO/TiO2-S samples (each 50 mg) were loaded into a U-shaped quartz reactor. The samples were first pretreated by heating from room temperature to 300 °C at a ramp rate of 10 °C min−1 under a He flow (30 mL min−1) for 1 h. After cooling to 50 °C, a H2 mixture (H2/Ar = 10/90, 30 mL min−1) was introduced for 30 min to stabilize the baseline. The reduction profiles were then recorded by heating the sample to 600 °C at 10 °C min−1 under the same H2/Ar flow, while monitoring H2 consumption using a thermal conductivity detector (TCD).

In situ XAS experiments

The XAS of Ni K-edge experiments were performed at the BL20U1 Hard branch of energy material beamline (E-line) of the Shanghai Synchrotron Radiation Facility (SSRF) operating at the 3.5 GeV with an injection current of 220 mA. A Si (111) double-crystal monochromator was used for energy selection. Ni foil was used as a reference, and the measurements were conducted in transmission mode. Data processing was performed using the Demeter software package.

XPS

The XPS analysis was implemented on Thermo Scientific K-Alpha using the Al-Kα X-rays source. The treated catalysts were made into small tablets and placed on the sample holder in a glovebox without exposure to air. The sample was then introduced into the ultra-high vacuum chamber for XPS measurements at room temperature.

DFT calculations

All DFT calculations were conducted using the projector augmented wave (PAW) method and the Perdew-Burke-Ernzerhof exchange-correlation functional, as implemented in the Vienna ab initio simulation package (VASP) 5.4.452,53. The long-range van der Waals (vdW) interaction was described by the DFT-D3 method54. The Brillouin-zone integration was performed using Gama k-mesh to ensure computational accuracy55. No Hubbard correction (LDAU) was applied to the Ni/TiO2 system. Full computational parameters are provided in Section Calculation method and Supplementary Figs. 10–15 of SI.