Reduction‐induced metal/oxide interfacial sites for selective CO2 hydrogenation

The interfacial structures of bimetallic‐derived catalysts play an important role in promoting the activation of reactants such as CO2. In particular, both the physical property (e.g., local bonding environment) and the electronic property (e.g., oxidation state) can evolve from their native states under different environments, such as upon reduction and during the catalytic reaction. Hence, taking the CO2 hydrogenation reaction over Rh‐based catalysts as a case study, the present work compares the interfacial structures in tuning the selectivity toward CH4 or CO. The combination of ex situ and in situ characterization reveals two representative interfacial structures: the Rh/CeOx interface formed over Rh/CeO2 is active and selective to produce CH4 (~95%) by following a formate‐mediated pathway; in comparison, the InOx/Rh interface derived after reduction is active for CO2 activation and enables a redox mechanism for the exclusive formation of CO (~100%). This work provides insights into the environment‐induced structural evolution at the metal−oxide interfaces, as well as the role of distinct interfacial active sites in tuning the selectivity of CO2 hydrogenation.


| INTRODUCTION
The combination of metal and metal oxide (MO) has served as the cornerstone of many industrial catalysts. 1,2 In addition to being used as catalysts directly, metal nanoparticles or clusters are typically dispersed on MOs as supported catalysts. [2][3][4][5] Yet, the MO support is usually more than a large surface area carrier. Instead, it often imposes electronic perturbations on the supported metal ensembles due to their distinct work functions, thereby modifying the binding strength of reactants on the supported metals as well as the affinity of metals to the support. 6 In peculiar, the resulting metal/metal oxide (M/M'O x , with M representing supported metal and M′ O x representing the metal oxide support) interfaces over reducible supports (e.g., CeO 2 and TiO 2 ) may act as additional active sites that exhibit unique electronic properties and enhanced catalytic performance for reactions such as CO oxidation, (reverse) water gasshift [(R)WGS], and hydrocarbon reforming. [6][7][8] In parallel, subject to minimizing the free energy of the metal surface, often under catalytic reaction environments, MO may prefer to wet the metal substrate to form the so-called metal oxide/metal (M'O x /M) inverse interface. [9][10][11] Compared with M/M'O x , the M'O x /M structure can not only attain oxide configurations that are not available in a bulk oxide but also provide more defects or oxygen vacancies and impose confinement effects. Thus, the M'O x /M inverse interface offers additional possibilities for enhancing reactivity, tuning selectivity, and retaining stability. [11][12][13][14] So far, many M'O x /M structures (e.g., ZnO x /Cu, MoO x /Cu, TiO x /Pt, TiO x /Rh, and FeO x /Pt) have been synthesized and investigated for semiconductor devices as well as catalysis typically using model systems, with the metal substrate mostly being a single crystal surface. 9,15,16 Thus, synthesis and studies of powder systems under reaction conditions are critical to bridging the so-called "material gap" and "pressure gap" between model surfaces and practical catalysts.
It is noted that the M'O x moieties (e.g., FeO x , CeO x , NbO x , and TiO x ) in the inverse oxide catalysts are usually those with variable valences that allow for the migration toward metal surfaces via strong metal−oxide interactions. [17][18][19][20][21] However, the main group elements (e.g., Sn, In, and Ga) with such properties are less investigated for the M'O x /M inverse interface in catalysis. The M'O x /M inverse interfaces have been mainly generated by the H 2 -reduction-induced strong metal−support interaction from a parent M/M'O x structure or by the chemical vapor deposition (CVD) of M'O x over the M substrate. 10,18,19,[22][23][24][25] Our previous study reported a bimetallic-derived principle that the formation of the M'O x /M inverse interface can be favorable when the reactive oxygen binding energy is greater than the alloy formation energy, which is the case with the PdM systems (M = Sn, In, and Ga). 11,21,26,27 Different from the predefined M'O x /M structure by the CVD method, the bimetallic-derived method undergoes thermodynamically favorable structural evolution to generate the inverse interface that is induced by reaction conditions.
Despite the distinct and unique properties of M/M'O x and M'O x /M interfaces, they are rarely compared in controlling the product selectivity for catalysis. Thus, herein, we employed CO 2 hydrogenation as a probe reaction, which has been identified as a promising route for mitigating the greenhouse gas CO 2 if H 2 is generated from renewable sources while producing a variety of value-added commodity chemicals (e.g., CO, light alkanes/olefins, alcohols, and aromatics). [28][29][30] The reaction was carried out over Rh-based monometallic (Rh) and bimetallic (RhIn) catalysts. A combination of kinetic studies using steady-state reactors and in situ and ex situ characterization revealed an Rh/CeO x interface over Rh/CeO 2 and a bimetallic-derived InO x /Rh(In) inverse interface over RhIn 3 /CeO 2 . The mechanistic study indicated a facile CO 2 activation over the Rh/CeO x interface, leading to CH 4 production with a high selectivity (~95%) by following a formate-mediated pathway; in contrast, the InO x /Rh(In) inverse interface was nearly 100% selective to CO due to CO 2  Sigma-Aldrich)] were simultaneously dissolved in deionized water (30 mL) to obtain an Rh/In atomic ratio of 1:3. The solution was subject to ultrasonication for 15 min before the addition of support [CeO 2 (35-45 m 2 /g, <25 nm particle size; Sigma-Aldrich) or SiO 2 (68 m 2 /g; Alfa-Aesar)], followed by another 10 min of ultrasonication. The obtained slurry suspension was stirred and dried at 343 K overnight before the calcination in static air at 673 K for 2 h with a heating rate of 1.0 K/min. Likewise, the monometallic catalysts (Rh/CeO 2 , Rh/ In 2 O 3 , and In 3 /CeO 2 ) were synthesized with the same loading amount of the corresponding metal as that in the RhIn bimetallic catalysts.

| Catalyst characterization
The particle size and elemental distributions over the reduced and spent catalysts were analyzed using highangle annular dark field (HAADF)-scanning transmission electron microscopy (STEM) and energy dispersive spectrometry (EDS), respectively. The catalyst was dispersed by ethanol and ultrasonicated for 30 min before being dropped onto a Lacey carbon-supported copper grid. All the HAADF and EDS measurements were conducted using an FEI Talos F200X TEM (operated at 200 kV).
The diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) measurements of CO adsorption on the reduced catalysts were carried out with a Thermo Nicolet 6700 spectrometer. Before the measurement, the catalyst was reduced in 50 vol% H 2 in He (40 mL/min) at 623 K for 1 h and then cooled down to room temperature in He (40 mL/min), after which the background spectrum was collected. The sample was subsequently exposed to the 10 vol% CO in He (CO/He = 2/18 mL/min) for 15 min and He (20 mL/min) for 30 min. For the in situ DRIFTS experiments, the reduced sample was subject to the reaction stream (CO 2 /H 2 /He = 3/9/ 8 mL/min) for 2 h before switching to an H 2 atmosphere (H 2 /He = 9/11 mL/min) for 1 h. The sample spectra of all the DRIFTS experiments were collected continuously with 32 scans for each spectrum at a resolution of 4 cm −1 .
The amount of chemisorbed CO was measured using an AMI-300ip (Altamira) instrument. Approximately, 100 mg of the as-prepared catalyst was pretreated in He (50 mL/min) at 393 K for 30 min and then cooled to 323 K. Afterward, the sample was heated to 623 K (10 K/min) in 10 vol% H 2 in Ar (50 mL/min) and held for 1 h. The catalyst was subsequently cooled to 313 K in He (50 mL/min) before pulsing 10 vol% CO in He (590 μL loop). The number of exposed metal active sites on the catalyst was approximated based on an assumption of the CO/metal ratio of 2:1.
The in situ X-ray absorption fine structure (XAFS) spectra of the Rh K and In K-edges were collected at beamline 7-BM (QAS) of the National Synchrotron Light Source II (NSLS-II) at Brookhaven National Laboratory. For each measurement, an appropriate amount of the catalyst was compressed into a wafer (OD 1.3 cm) before being loaded into a Nashner-Adler reaction cell 31 sealed by Kapton windows, which allows for the simultaneous collection of both the transmission and fluorescence signals. The sample was reduced by 50 vol% H 2 in N 2 (40 mL/min) at 623 K for 1 h before being exposed to the reaction stream of CO 2 and H 2 (CO 2 /H 2 / N 2 = 3/9/8 mL/min) at 623 K for 1 h. The XAFS spectra were collected successively at a speed of 2 scans/min. Rh and In foils as the standard references were adopted to calibrate energy shifts as well as obtain the passive electron reduction factor (S0 2 ) used for the extended X-ray absorption fine structure (EXAFS) fitting. All the XAFS data processing was conducted using the IFEFFIT package. 32

| Catalytic performance evaluation
All the catalytic performance tests were carried out using a quartz flow reactor (OD 6.35 mm, ID 4 mm) at ambient pressure. For each experiment, approximately 100 mg of the catalyst (60-80 mesh) was fixed within the isothermal zone of the quartz tube using two pieces of quartz wool on both sides. The reduction and reaction conditions were the same as that described in the XAFS measurements except for a longer reaction time (~11 h). For the temperature-dependent experiments, the reaction was carried out at 523, 573, 623, 673, and 723 K with a dwelling for 1.5 h at each temperature stage. To avoid any condensation of water vapor, the gas line from the reactor outlet to the gas chromatography (GC) inlet was wrapped with heating tape and maintained at 423 K. The product stream was analyzed by an Agilent 7890B GC (PLOT Q and MOLESEIVE columns) equipped with a thermal conductivity detector and a flame-ionized detector. The carbon balance was within 100% ± 2% for all the experiments. The calculations of conversion (X), selectivity (S), and yield (Y) were described below.
where F i referred to the internal standard (N 2 ) corrected molar flow rates (mol/min) of species i at the reactor outlet.
where the product species i could be CO or CH 4 .

| Catalytic performance
The Rh-based monometallic and bimetallic catalysts were synthesized via the slurry impregnation method and evaluated for CO 2 hydrogenation. As shown in Figure 1A,B, the Rh/CeO 2 catalyst demonstrated a high selectivity (74.3%-96.5%) to methane between 523 and 723 K, while the RhIn 3 /CeO 2 catalyst was 100% selective to the formation of CO. The equilibrium conversion of CO 2 hydrogenation to methane (56.0%) and CO (43.7%) was nearly achieved at 723 K over Rh/CeO 2 and RhIn 3 /CeO 2 , respectively. To avoid equilibrium control but retain appreciable conversion, the catalysts were evaluated at a lower temperature, 623 K, for 11 h. Again, Rh/CeO 2 demonstrated a higher CO 2 conversion (53.8%) than RhIn 3 /CeO 2 (12.5%); yet the turnover frequency of CO 2 was comparable-11.3 min −1 over Rh/CeO 2 and 12.2 min −1 over RhIn 3 /CeO 2 . Given the significant conversion difference, the conversion of Rh/CeO 2 was reduced to 14.1% via diluting the catalyst and increasing the weight hourly space velocity (WHSV). Under a similar CO 2 conversion (i.e., 14.1% vs. 12.5%), Rh/CeO 2 (denoted as "Rh/CeO 2 *" in Figure 1C) still exhibited 70% selectivity to CH 4 , while RhIn 3 /CeO 2 was exclusively selective to CO. It is noted that the CH 4 selectivity decreased with the increase of WHSV (or the decrease of residence time), suggesting that CO was most likely an important intermediate for the formation of CH 4 , as indicated by the in situ DRIFTS measurements in Section 2.3.
Compared with Rh/CeO 2 , RhIn 3 /CeO 2 was more stable ( Figure 1D) within 11-h time on stream. As a reference, monometallic In 3 /CeO 2 was nearly inactive for CO 2 hydrogenation. Thus, the synergistic interaction between Rh and In resulted in a switch in selectivity from CH 4 to CO, as well as enhanced stability. This was further validated by investigating the Rh/In 2 O 3 catalyst prepared by supporting Rh over the In 2 O 3 support, which also exhibited stable and exclusive selectivity to CO. In contrast, a SiO 2 -supported RhIn catalyst was also exclusively selective to CO, but it displayed a lower conversion (5.0%) with a gradual deactivation, highlighting the positive role of the CeO 2 support. Interestingly, negligible CO chemisorption was observed over both the reduced Rh/In 2 O 3 and RhIn 3 /SiO 2 catalysts despite their activity. This suggested that the active site for CO 2 activation most likely resulted from the In-modified interfacial structure.

| Structural characterization
To understand the origin of the selectivity difference between the Rh and RhIn catalysts, the catalyst properties were characterized using both in situ and ex situ techniques. The particle size and elemental distributions over Rh/CeO 2 and RhIn 3 /CeO 2 were probed using STEM. Owing to the low Z-contrast of Rh (Z = 45) and In (Z = 49) relative to the heavy Ce (Z = 58) in the CeO 2 support, typical HAADF imaging ( Figure 2K,O) is usually difficult to show sharp imaging. Thus, as shown in Figure 2A-D, the simultaneous secondary electron (SE) and HAADF techniques were employed. Taking advantage of the depth information offered by the high-resolution SE imaging, a sharp contrast was obtained, which clearly imaged the metal particles over the surface of CeO 2 . The particle size distribution ( Figure 2E-H) indicated a similar average size (1.5-1.6 nm) of metal particles over the Rh/CeO 2 and RhIn 3 /CeO 2 catalysts for reduced (after H 2 treatment) and spent (after CO 2 hydrogenation) samples. Thus, the metal sintering issue should be negligible over both catalysts under the reduction and reaction conditions at 623 K.  The HAADF imaging and EDS mapping ( Figure 2I-L) over the reduced Rh/CeO 2 catalyst confirmed that the Rh metal particles resided on CeO 2 . As for the reduced Rh/CeO 2 catalyst, the lowmagnification imaging and mapping ( Figure 2M,N) indicated a uniform distribution of Rh and In over CeO 2 ; the high-magnification observation ( Figure 2O,P) suggested that, as illustrated by the observable particles (highlighted by the squares in Figure 2P), In not only appeared with Rh inside the particles but also enriched on the surface or surrounding of the particles.
The surface decoration of Rh by In species was validated by the DRIFTS measurements of CO chemisorption and the in situ XAFS measurements. Figure 3A shows one pair of peaks at 2083 and 2017 cm −1 , characteristic of gem-dicarbonyls on Rh atoms (i.e., Rh + (CO) 2 ) over the reduced Rh/CeO 2 catalyst. 33 Moreover, the presence of Rh nanoparticles was indicated by the peaks associated with adsorbed CO in the linear (2061 cm −1 ), bridge (~1940 cm −1 ), and hollow (1815 cm −1 ) configuration. 34 The gemdicarbonyls and linear carbonyl were also observed over the reduced RhIn 3 /CeO 2 catalyst ( Figure 3B), although peaks associated with the bridge and hollow carbonyls were absent. The different types of carbonyl peaks over RhIn 3 /CeO 2 from those over Rh/CeO 2 suggested an electronic modification of Rh by In, as supported by the in situ XAFS studies below. In contrast, both the RhIn 3 /SiO 2 ( Figure 3C) and Rh/In 2 O 3 ( Figure 3D) catalysts displayed no CO adsorption after reduction, indicating that the entire surface of Rh was modified by the In species, which are typically inactive for CO adsorption.
The chemical states of Rh and In and their interaction were investigated using the in situ XAFS measurements. As demonstrated in Figure 3E, the similar X-ray absorption near-edge spectroscopy (XANES) feature of the Rh K-edge with the Rh foil indicated a predominant metallic state of Rh over Rh/CeO 2 under reduction. The XANES results of all the In-containing catalysts ( Figure 3F-H), that is, Rh/In 2 O 3 , RhIn 3 /CeO 2 , and RhIn 3 /SiO 2 , also indicated the reduction of Rh as shown by the diminished white line; yet, compared with the Rh foil, the first resonance peak of XANES shifted to the lower energy along with an enhanced XANES intensity before 23235 eV, typically characteristic of the formation of RhIn alloys involving a charge transfer from In to Rh. 27,35,36 The In K-edge XANES measurements over both RhIn 3 /CeO 2 ( Figure 3I) and RhIn 3 /SiO 2 ( Figure 3J) also showed significantly diminished white lines under reduction, which were closer to the In foil. Thus, In over both catalysts was largely reduced but with a small amount of unreduced InO x species. However, In in RhIn 3 /CeO 2 was less reduced than that in RhIn 3 /SiO 2 as indicated by its higher edge position, thereby highlighting the support effect on the reducibility of In. After exposure to the reaction mixture of CO 2 and H 2 at 623 K, the Rh and In K-edge XANES spectra remained nearly unchanged for all the Rh-based catalysts.
As shown in Figure 4A, the EXAFS results in R space displayed a significant decrease in intensity of the Rh-O peak (1.52 Å, phase-uncorrected) after reduction with the appearance of two prominent peaks at 1.89 and 2.48 Å, of which the relative intensity was different from that of the Rh foil. Accordingly, the EXAFS fitting results (Supporting Information: Figure S1 and Table S1) indicated the appearance of metallic Rh-Rh bond at around 2.68 Å with an average coordination number (CN) of 5.7 ± 0.6. Moreover, a minor Rh-O bond (CN = 2.2 ± 0.4) was observed at around 2.21 Å, but longer than the typical Rh-O bond (~2.00 Å) in Rh 2 O 3 . Given the gem-dicarbonyls on Rh + atoms from the DRIFTS measurements, the longer Rh-O bond most likely correlated with the slightly oxidized Rh atoms attached to the CeO 2 support. As for the RhIn 3 / CeO 2 catalyst, the Rh-O peak was also significantly reduced ( Figure 4B), but the two newly formed peaks were much lower in intensity than the counterparts of the Rh/CeO 2 catalyst. The EXAFS fitting in Supporting Information: Table S2 indicated an Rh-In(Rh) bond at around 2.69 Å (CN =~3.3). Additionally, a longer metallic peak was observed at around 3.34 Å (CN =~1.8), which was not observed in the Rh foil and Rh/CeO 2 but more like the Rh-Rh bond in the RhIn alloy. This is consistent with the formation of the RhIn alloy as revealed by the XANES analysis. Besides that, an Rh-O peak appeared at around 2.20 Å (CN =~1.4), likely characteristic of the Rh−CeO 2 interaction and/or the InO x −Rh interaction. To support this, the EXAFS results in Figure 4C,D were also analyzed for the RhIn 3 /SiO 2 and Rh/In 2 O 3 catalysts. The formation of the RhIn alloy over both catalysts was also indicated in the Supporting Information: Tables S3 and S4 by  RhIn/SiO 2 (J) catalysts under the fresh, reduction, and reaction conditions. Rh and In foils were also measured as standards for comparison. "Red.@623 K": reduction at 623 K; "Rxn.@623 K: reaction at 623 K". DRIFTS, diffuse reflectance infrared Fourier transform spectroscopy; XAFS, X-ray absorption fine structure; XANES, X-ray absorption near-edge spectroscopy.
Rh(In) inverse interface (with metal fully encapsulated) over Rh/In 2 O 3 and RhIn 3 /SiO 2 . In addition, the RhIn alloy formed over all the In-containing Rh catalysts.

| Mechanistic insight
To obtain a mechanistic insight, in situ DRIFTS measurements were carried out in the presence of the reaction mixture of CO 2 and H 2 at 623 K, followed by the exposure to H 2 only at the same temperature. Compared with CeO 2 (Figure 5A), Rh/CeO 2 ( Figure 5B) showed peaks associated with linear (2020 cm −1 ) and bridge (1920 cm −1 ) CO adsorption over the Rh metal sites. Additionally, a broad peak was observed between 1840 and 1660 cm −1 , which has been assigned to the bridge CO adsorption (1752 cm −1 ) over the Rh-CeO x interface (denoted as Rh-CeO x (CO)) as well the threefold hollow adsorption (1800 cm −1 ). 33,37 The presence of the Rh-CeO x interface agreed with the appearance of the Rh-O bond from the EXAFS fitting. Gaseous CH 4 (3016 and 1306 cm −1 ), formate (2940, 2831, 2711 (not shown), 1557, 1390, and 1327 (not labeled) cm −1 ), and carbonates (1578, 1506, 1425, and 1287 cm −1 ) were also detected over Rh/CeO 2 . [37][38][39] After switching to H 2 only, the evolution profile (Supporting Information: Figure S2) of gaseous CH 4 over Rh/CeO 2 overlapped well with that of the formate species, suggesting a formate-mediated pathway for the formation of CH 4 . Given the low CO selectivity (e.g., <4%) and the significant depletion of carbonyls and carbonates by H 2 compared with CeO 2 , the latter two intermediates should be predominantly hydrogenated to formate as well. As a result, the rates of carbonyl hydrogenation and desorption would affect the CO formation, as shown in Figure 1A,C, consistent with the reduced CH 4 selectivity at high temperature (faster desorption) and high WHSV (lower residence time for hydrogenation). As indicated by the DRIFTS and XAFS analyses, the carbonyl species should mainly come from the activation of CO 2 over the Rh/CeO x interface as well as the Rh metal sites.
In contrast, the vibrational peak associated with the Rh-CeO x interface was not observed over the RhIn 3 /CeO 2 catalyst under the reaction condition, whereas a sharp linear CO peak appeared at 1989 cm −1 at 0.2 min and F I G U R E 4 In situ Fourier-transformed extended X-ray absorption fine structure plots of Rh K-edge over different catalysts as well as the Rh foil. (A) Rh/CeO 2 , (B) RhIn 3 /CeO 2 , (C) Rh/In 2 O 3 , and (D) RhIn 3 /SiO 2 . "Red.@623 K": reduction at 623 K; "Rxn.@623 K: reaction at 623 K".
shifted to 2007 cm −1 at 120 min, a lower frequency than that (2021 cm −1 ) over Rh/CeO 2 . Such a red-shift relative to Rh/CeO 2 was consistent with the In→Rh charge transfer in the RhIn alloy structure, thereby enhancing the 4d → 2π* back donation and in turn weakening the C═O bond. In another word, the linearly-bonded CO species adsorbed more strongly over RhIn 3 /CeO 2 than Rh/CeO 2 , consistent with its negligible desorption or consumption after being exposed to H 2 only at 623 K (the lower panel in Figure 5C). Thus, the strongly bound carbonyl species would reduce the surface coverage of H* for further hydrogenation. However, two negative peaks were detected at 1964 and 1894 cm −1 under the reaction condition, which were observed over neither CeO 2 nor Rh/CeO 2 . Recalling the InO x /Rh(In) inverse interface from XAFS analysis, these two peaks were most likely associated with CO adsorption over Rh atoms within the InO x /Rh(In) inverse interface. It is noted that both peaks were nearly recovered after the atmosphere was switched from CO 2 + H 2 to H 2 only, while a control experiment with CO 2 only at 623 K rapidly decreased these two peaks. Thus, it is most likely that CO 2 strongly competed with CO for the interfacial In-O v -Rh sites over RhIn 3 / CeO 2 and in turn promoted the desorption of CO*. The presence of H 2 was essential for the generation of O v even though it could also slowly hydrogenate the formate (2942, 2843, 2711 (not shown), 1548, 1388, and 1327 cm −1 ) and carbonate (1586, 1431, and 1291 cm −1 ). To validate the role of H 2 in CO 2 activation, as shown in Supporting Information: Figure S3, the catalyst was treated in Helium (He) at 623 K for 20 h. After exposure to 15 vol% CO 2 in He at 623 K, only a small peak of adsorbed CO was observed at around 2010 cm −1 , which disappeared rapidly after the He purging. Yet, a noticeable CO peak was recovered with the co-feed of CO 2 and H 2 . Thus, the production of CO over RhIn 3 / CeO 2 should be dictated by a redox mechanism that CO 2 was activated over O v within the InO x /Rh(In) inverse interface, after which the formed CO* species desorbed rapidly, while the O v can be regenerated by H 2 .
No CO adsorption was observed over the reduced RhIn 3 /SiO 2 catalyst at room temperature, indicating the absence of exposed Rh atoms, that is, the complete surface encapsulation of the Rh atoms by the In species. In addition, Supporting Information: Figure S4 shows that no adsorbed intermediates were observed on the surface of the RhIn 3 /SiO 2 catalyst regardless of room temperature or 623 K, again highlighting the dominance of a redox mechanism in CO 2 activation.