This document is the Accepted Manuscript version of a Published Work that appeared in final form in Journal of Catalysis 361 (2018) 313e321 hFps://doi.org/10.1016/j.jcat.2018.03.014 Journal of Catalysis 361 (2018) 313–321 Contents lists available at ScienceDirect Journal of Catalysis journal homepage: www.elsevier.com/locate/jcat Mechanism and microkinetics of methanol synthesis via CO2 hydrogenation on indium oxide M.S. Frei a, M. Capdevila-Cortada b, R. García-Muelas b, C. Mondelli a, N. López b,⇑, J.A. Stewart c, D. Curulla Ferré c, J. Pérez-Ramírez a,⇑ a b c Institute for Chemical and Bioengineering, Department of Chemistry and Applied Biosciences, ETH Zurich, Vladimir-Prelog-Weg 1, 8093 Zürich, Switzerland Institute of Chemical Research of Catalonia, ICIQ, The Barcelona Institute of Science and Technology, Av. Països Catalans 16, 43007 Tarragona, Spain Total Research & Technology Feluy, Zone Industrielle Feluy C, 7181 Seneffe, Belgium a r t i c l e i n f o Article history: Received 7 February 2018 Accepted 9 March 2018 Available online 29 March 2018 Keywords: CO2 hydrogenation Thermal catalysis Indium oxide Methanol synthesis (Micro)kinetics Mechanism Density Functional Theory a b s t r a c t Indium oxide has emerged as a highly effective catalyst for methanol synthesis by direct CO2 hydrogenation. Aiming at gathering a deeper fundamental understanding, mechanistic and (micro)kinetic aspects of this catalytic system were investigated. Steady-state evaluation at 5 MPa and variable temperature indicated a lower apparent activation energy for CO2 hydrogenation than for the reverse watergas shift reaction (103 versus 117 kJ molÀ1), which is in line with the high methanol selectivity observed. Upon changing the partial pressure of reactants and products, apparent reaction orders of À0.1, 0.5, À0.2, and À0.9 were determined for CO2, H2, methanol, and water, respectively, which highlight a strong inhibition by the latter. Co-feeding of H2O led to catalyst deactivation by sintering for partial pressures exceeding 0.125 MPa, while addition of the byproduct CO to the gas stream could be favorable at a total pressure below 4 MPa but was detrimental at higher pressures. Density Functional Theory simulations conducted on In2O3(1 1 1), which was experimentally and theoretically shown to be the most exposed surface termination, indicated that oxygen vacancies surrounded by three indium atoms enable the activation of CO2 and split hydrogen heterolytically, stabilizing the polarized species formed. The most energetically favored path to methanol comprises three consecutive additions of hydrides and protons and features CH2OOH and CH2(OH)2 as intermediates. Microkinetic modeling based on the DFT results provided values for temperature and concentration-dependent parameters, which are in good agreement with the empirically obtained data. These results are expected to drive further optimization of In2O3-based materials and serve as a solid basis for reactor and process design, thus fostering advances towards a potential large-scale methanol synthesis from CO2. Ó 2018 Elsevier Inc. All rights reserved. 1. Introduction Aiming at mitigating global warming, the utilization of CO2 captured either from the atmosphere or from localized emission points as a feedstock for the production of fuels and chemicals has taken centrer stage in catalysis research in the last decade [1,2]. Among the routes investigated for the conversion of this rather inert molecule, hydrogenation to methanol has attracted particular attention owing to the extreme relevance of this alcohol for the chemical industry and the energy sector [3]. At present, the industrial synthesis of methanol is carried out from syngas, which is predominantly derived from steam reforming of fossil fuels, in the presence of small amounts (3 vol%) of CO2 over Cu-ZnO-Al2O3 ⇑ Corresponding authors. E-mail addresses: nlopez@iciq.es (N. López), jpr@chem.ethz.ch (J. PérezRamírez). https://doi.org/10.1016/j.jcat.2018.03.014 0021-9517/Ó 2018 Elsevier Inc. All rights reserved. catalysts [4]. When CO2 is the only carbon-containing source (CO2 + 3H2 ¢ CH3OH + H2O), these catalysts exhibit limited activity, selectivity, and stability. Particularly, the production of methanol is suppressed by the competitive reverse water-gas shift (RWGS) reaction (CO2 + H2 ¢ CO + H2O) [5]. Therefore, alternative catalytic systems are sought after to enable the development of a CO2-to-methanol technology [6]. Among the materials investigated, a few Cu-based solids exhibited high methanol selectivity but their long-term stability was not demonstrated [7,8]. In contrast, indium oxide, In2O3, emerged as a catalyst combining multiple desirable features [9]. In addition to being highly selective to methanol, it can be easily supported on zirconia, leading to higher methanol formation rates and outstanding durability, and prepared in an equivalently performing technical form [10]. Its potential for a prospective industrial implementation has motivated further practically-oriented studies. Indeed, it has been reported that In2O3 can be used in bifunctional catalytic systems 314 M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 [11,12] to produce olefins from CO2 and that its activity in methanol synthesis can be enhanced through the addition of a noble metal such as Pd, which is claimed to facilitate hydrogen splitting [13,14]. Furthermore, it sparked the idea of combining another reducible oxide, ZnO, with zirconia, attaining a remarkable catalyst [15]. In contrast, fundamental aspects of CO2 hydrogenation over In2O3, such as the reaction mechanism and kinetics, remain insufficiently understood, thus hindering the optimization of this catalytic process. Under the conditions commonly applied (573 K, 5 MPa, molar H2:CO2 = 4), the thermodynamic equilibrium highly favors the endothermic RWGS reaction over the exothermic hydrogenation to methanol [16]. Hence, the reaction must be carried out under kinetic control to attain a high selectivity to methanol. Nevertheless, basic kinetic information such as the temperature dependence of the transformation and the effect of the partial pressure of reactants and products has not been reported. Characterization through thermal and spectroscopic techniques and CO co-feeding experiments over bulk and ZrO2-supported In2O3 strongly suggested surface oxygen vacancies as the sites where CO2 and H2 are activated [10]. This is in line with Density Functional Theory (DFT) studies by Ge et al. [17,18] and Yu et al. [19], which predicted a superior performance for an oxygen defective In2O3(1 1 0) surface compared to the stoichiometric termination and supported the high methanol selectivity on the basis of greater barriers for the RWGS reaction over the same vacancy-containing surface. Still, the orientation of the surface employed in these studies is not that being associated with the lowest energy (i.e., the most exposed). Indeed, the (1 1 1) facet is the thermodynamically most stable termination of In2O3 [20]. In addition, since the electronic structure of this material is rather subtle, the structures of the adsorbates need to be evaluated with some further restrictions to ensure that forbidden energy levels are not artificially populated. Herein, we investigate the kinetic fingerprints of the CO2 hydrogenation over bulk In2O3 by means of a detailed study covering a wide range of temperature (473–673 K), pressure (1–5 MPa), and reactant (CO2, H2) and product (CO, methanol, H2O) concentrations. These data are correlated with extensive DFT simulations describing the energetics of the reaction over the In2O3(1 1 1) surface and a microkinetic model. Based on the information obtained at the experimental and theoretical levels, key conclusions are put forward on the operation of this oxide in CO2 hydrogenation, which set the ground for its more efficient utilization in this relevant reaction. 2. Experimental 2.1. Catalyst preparation In(OH)3 was obtained by precipitation adding 150 cm3 of a solution of aqueous NH4OH (50 cm3, 25 wt%, Sigma-Aldrich) in ethanol (150 cm3) to a solution of In(NO3)3ÁxH2O (30.1 g, Sigma-Aldrich, 99.99%) in deionized water (120 cm3) and ethanol (350 cm3, Merck, 99.8%,) to reach a pH value of 9.2. The resulting slurry was aged for 20 min at 353 K and then separated by high-pressure filtration and washed twice with deionized water (1000 cm3 each time) followed by drying in a vacuum oven (2 kPa, 323 K, 12 h). The so-obtained hydroxide was calcined for 3 h at 573 K (2 K minÀ1) in static air to yield nanocrystalline In2O3. 2.2. Catalyst characterization Powder X-ray diffraction (XRD) was performed using a PANalytical X’Pert PRO-MPD diffractometer with Ni-filtered Cu Ka radiation (k = 0.1541 nm), acquiring data in the 5–70° 2h range with a step size of 0.05° and a counting time of 12 s per step. The average particle size of In2O3 was estimated applying the Scherrer equation. Nitrogen sorption at 77 K was carried out using a Quantacrome Quadrasorb analyzer. Prior to the measurement, the sample was degassed at 573 K under vacuum for 3 h. He pycnometry was conducted at ambient temperature using a Micromeritics AccuPyc II 1340 instrument after pretreatement of the sample in vacuum at 323 K for 12 h. The skeletal density was determined by averaging 200 measurement cycles after equilibration of the system during 30 cycles. High resolution transmission electron microscopy (HRTEM) was performed using a FEI Talos (200 kV) instrument. Samples were prepared by directly depositing the powders on a lacey carbon film supported onto a copper grid. Temperature-programmed desorption of carbon dioxide, water, and methanol (CO2-, H2O-, MeOH-TPD) and reduction with hydrogen (H2-TPR) were conducted using a Micromeritics AutoChem II 2920 analyzer. The specific conditions applied during each measurement are detailed in Table S1 in the Supplementary Material. UV–vis spectroscopy was carried out using an Ocean Optics Maya2000-Pro spectrometer equipped with a deuterium light source. Spectra were collected in the 200–600 nm range, with an integration time of 100 ms and averaging 50 scans. Raman spectroscopy was performed using a WITec CRM200 confocal Raman spectroscopy system comprising a source with an excitation wavelength of 532 nm, a 100Â objective lens with a numerical aperture of 0.9, and a grating spectrometer (2400 lines mmÀ1). Spectra were obtained by averaging 200 scans with an individual acquisition time of 1.5 s. 2.3. Kinetic evaluation Kinetic investigations were performed in a high-pressure continuous-flow fixed-bed reactor with an inner diameter of 2.2 mm housed in an electrically-heated aluminum brass furnace. The experimental setup is described in detail in Fig. S1. The reactor was loaded with 25 or 50 mg of catalyst with a particle size of 100–125 lm, which was held in place by a bed of quartz wool and heated from ambient temperature to 573 K (5 K minÀ1) at 0.5 MPa under a He (PanGas, !99.999%) flow of 20 cm3 minÀ1. STP After 3 h at 573 K, the pressure was raised to 3.5–5.5 MPa in the same stream, which typically took 20 min. Then, the gas flow was switched to the reaction mixture (20 or 40 cm3 minÀ1) featuring STP distinct concentrations of CO2 (40 mol% CO2 in H2, Messer, !99.997% and !99.999%, respectively), H2 (PanGas, !99.999%), methanol (Sigma-Aldrich, 99.9%, anhydrous), H2O (ABCRChemicals, HPLC grade), and CO (Messer, !99.999%), corresponding to a gas-hourly space-velocity (GSHV) of 38,000 hÀ1. The temperature was either kept at 573 K or varied in the range of 473–673 K. The specific experimental conditions applied in the tests are collected in Table 1. The effluent stream was analyzed by gas chromatography after 1 h on stream and then every 20 min. The catalytic data reported for each reaction condition correspond to the average of at least 7 measurements. The absence of intra- and extraparticle diffusion limitations was corroborated by the fulfillment of the Weisz-Prater and Carberry criteria and secured by a test in which a catalyst sample featuring a sieve fraction of <50 lm instead of the typical 100–125 lm was applied that led to identical performance. 2.4. Computational methods DFT calculations were conducted using the Vienna Ab initio Simulation Package (VASP) [21,22], employing the PerdewBurke-Ernzerhof (PBE) exchange-correlation functional [23]. The core electrons were described by projector augmented-wave pseudopotentials (PAW) [24] with a plane-wave cutoff energy of 315 M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 Table 1 Conditions applied in the steady-state kinetic experiments. Experiment Catalyst mass [mg] Initial gas feeda [cm3 minÀ1] STP Determination of activation energies (Fig. 2a) 50 CO2: H2: 8 32 Determination of reaction order with respect to CO2 and H2 (Fig. 2b) 25 CO2: H2: 8 12 Determination of reaction order with respect to MeOH (Fig. 2c) 50 CO2: H2: MeOH: Determination of reaction order with respect to H2O (Fig. 2c) 50 Impact of low CO concentration on the catalyst (Fig. 4a) 25 Assessment of deactivation by CO (Fig. 4b) 25 Assessment of deactivation by H2O (Fig. S6) a b c d 50 Step changes applied to flowb [cm3 minÀ1] STP Final gas feed [cm3 minÀ1] STP Temperature [K] Total pressure [MPa] 8 32 473–673c 5 À0.8 +0.8 0.8 19.2 573 5 8 32 0 À0.1 À0.4 +0.5 7.5 30 2.5 573 5 CO2: H2: H2O 8 32 0 À0.08 À0.32 +0.4 7.6 30.4 2 573 5 CO2: H2: He: CO: 4 16 4 0 573 3.5, 4.5, or 5.5 À1 +1 4 16 0 4 CO2: H2: CO: He: 4 16 0 4 4 16 1, 2.5, or 4d 3, 1.5, or 0d 573 5.5 CO2: H2: H2O 8 32 0 7.5 31.5 1 573 5 +1, 2.5, or 4 À1, 2.5, or 4 À0.5 À0.5 +1 In all experiments a constant flow of 2.5 cm3 minÀ1 of 20 mol% CH4 in He was added to the effluent stream as an internal standard. STP As many steps as needed to reach the final gas feed were applied. The temperature was increased stepwise by 25 K. After 3 h on stream the initial values were restored. 500 eV for the valence electrons (i.e., 4d, 5s, and 5p for In atoms; 2s and 2p for O and C atoms; and 1s for H atoms). The calculated lattice parameter (acalc = 10.294 Å) for the bixbyite structure of indium oxide, optimized using a dense C-centered 5 Â 5 Â 5 k-point mesh and a higher plane-wave cutoff energy of 650 eV, agrees well with the experimental data (aexp = 10.117 Å). The In2O3(1 1 1) surface was modeled as a periodically repeated p(1 Â 1) slab consisting of five O–In–O trilayers, separated by a vacuum space of 16 Å, and was optimized using a C-centered (3 Â 3 Â 1) k-point mesh. In each surface, the three outermost O–In–O trilayers were allowed to relax, whereas the two bottommost layers were fixed in their bulk positions. The climbing image nudged elastic band (CI-NEB) method was employed to locate the transition states [25]. The nature of all reaction minima and transition states was confirmed by means of numerical frequency analyses. All of the relevant structures can be retrieved from the ioChem-BD database [26]. Gibbs free energies were computed using the ideal-gas approximation for gaseous species and considering only vibrational contributions for the adsorbed species. The rate coefficients for each of the elementary steps were evaluated employing the thermodynamic and kinetic parameters and the partition functions obtained from the DFT calculations and according to Transition State Theory [27]. Microkinetic simulations were performed applying a differential reactor model operated under steady-state conditions [28]. The set of equations in the microkinetic model was resolved with Maple . In the simulations, temperature and partial pressures were set to the values employed experimentally and the catalyst surface was initially free of any adsorbate. From the microkinetic model, several parameters were obtained, like the apparent activation energy and the reaction orders for the reactants and products, according to the method described elsewhere [28]. Energy values were probed using (i) direct DFT results and (ii) alternative DFT results obtained in the presence of a water molecule, which was found to facilitate the first two elementary steps of the RWGS reaction, lowering TM the energy of the corresponding transition states by 0.19 and 0.16 eV, respectively. 3. Results and discussion 3.1. Catalyst properties Bulk In2O3 for experimental investigations was prepared by precipitation of In(OH)3 followed by calcination in static air, as reported earlier [10]. N2 sorption evidenced a type IV isotherm, typical of nanocrystalline materials (Fig. S2). A surface area of 123 m2 gÀ1 and a total pore volume of 0.38 cm3 gÀ1 were calculated, which are in agreement with previous data on this catalyst. Powder X-ray diffraction (XRD) confirmed the sole presence of the cubic bixbyite structure of indium oxide (ICDD 01-088-2160, Fig. 1a) and estimation of the particle size by application of the Scherrer equation led to a value of ca. 8 nm, which is only slightly smaller than that published earlier. The profiles obtained by H2-TPR (Fig. 1b) and CO2-TPD (Fig. 1c) are comparable to those collected in our previous study. HRTEM (Fig. 1d) visualized crystallites of ca. 8 nm, in line with the XRD analysis, which form aggregates. Lattice fringes, visible at multiple locations, almost exclusively indicate a d spacing of 0.29 nm, corresponding to the (2 2 2) plane of In2O3, which is a multiple of the (1 1 1) plane. These observations are in good agreement with the energies calculated by DFT for the low Miller index terminations of In2O3 under vacuum (Fig. 1a, inset). The (1 1 1) surface is more stable than the (1 1 0) surface and the reconstructed oxygen-terminated (1 0 0) surface by 0.30 and 0.92 J mÀ2, respectively, in line with previous computations [20]. Hence, the equilibrium Wulff construction results in an octahedral nanoparticle exposing (1 1 1) facets (Fig. 1a, inset). It should be remarked that such a model is solid, since the typical particle size is significantly larger than 3.5–4 nm, below which deviations from the equilibrium geometry might be plausible. 316 M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 temperatures above 623 K was confirmed by calculations (Fig. S3). Thus, the data collected at this and higher temperatures were not considered in the linearization. From the Arrhenius plot presented in Fig. 2a, apparent activation energies of 103 kJ molÀ1 (1.07 eV) and 117 kJ molÀ1 (1.21 eV) were derived for methanol synthesis and the RWGS reaction, respectively. The higher activation energy for the competitive transformation is in line with the observed low selectivity towards CO. The apparent reaction orders of the reactants were determined by performing measurements altering the partial pressures of CO2 and H2 (Fig. 2b) and normalizing the observed reaction rates (ri) by the rate under the standard conditions (r0) (Fig. S4a, b). Fig. 1. (a) XRD, (b) H2-TPR, (c) CO2-TPD, and (d) TEM and corresponding electron diffraction pattern obtained by fast Fourier transform of bulk In2O3. The inset in (a) shows the Wulff construction of In2O3, highlighting its octahedral morphology and the exclusive exposure of (1 1 1) terminations as well as the energies calculated for low Miller-index surfaces. Recent theoretical and experimental studies [29–31] support the formation of nanooctahedra under oxygen-rich atmospheres, such as that applied upon calcination. Although the computations by Chen et al. [30] suggest a truncated-cube as the thermodynamically more stable morphology of the oxide under CO2 hydrogenation conditions, analysis of In2O3 after reaction shows negligible restructuring [10]. This hints that a transition among structures is kinetically hindered. This, along with the outstanding stability of In2O3 in a 1000-h test, opens the door to a reliable kinetic analysis. In addition, it justifies the unprecedented use of the (1 1 1) termination as the most representative model for the catalyst for molecular-level investigations of CO2 hydrogenation by theoretical methods. 3.2. Kinetic analysis The dependence of the reaction rate on the experimental parameters was investigated relating the catalytic data obtained at variable temperature and partial pressures of reactants and (by)products to the values obtained under standard conditions. These comprised a temperature of 573 K, a flow with a rate of 40 cm3 minÀ1 composed of CO2 and H2 in a molar ratio of 1:4, STP and a GHSV of 38,000 hÀ1, which ensured testing under a differential regime. To estimate the apparent activation energies (Ea,app) for CO2 hydrogenation to methanol and the RWGS reaction, catalytic tests were carried out varying the temperature between 273 and 673 K. Upon each 25 K increment of the temperature, a raise in CO2 conversion was observed, accompanied by a drop in methanol selectivity. The suspected onset of thermodynamic equilibrium at Fig. 2. (a) Arrhenius plot for CO2 hydrogenation to methanol and the RWGS reaction over In2O3. Dependence of the rate on the partial pressure of (b) reactants and (c) products. Observed rates (ri) are normalized to that at standard conditions (r0) of P = 5 MPa, T = 573 K, and H2:CO2 = 4. M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 Interestingly, two sets of values were determined for both species. Above the stoichiometric ratio of CO2:H2 = 1:3, the orders were calculated at À0.1 for CO2 and 0.5 for H2. Below the standard conditions, they assumed values of À0.4 and 0.8, respectively. This variation is predominantly due to the fact that the parasitic RWGS reaction is favored in an excess of CO2 (Fig. S5). Consequently, targeting high methanol productivity, the process should be carried out with the highest possible excess of H2, since this will not only enhance CO2 conversion but also be beneficial to the selectivity towards the alcohol. The effect of the products on the reaction kinetics was explored in a similar way to that of the reactants. Fig. 2c evidences that an increase of the methanol concentration has a minor effect on the reaction rate, while the presence of water significantly inhibits both CO2 hydrogenation and the RWGS reaction. The apparent reaction orders with respect to the products were found to be À0.2 for methanol and À0.9 for water (Fig. S4c,d). In line with this, TPD experiments conducted at atmospheric pressure using the two species as adsorbates (Fig. 3) show that the alcohol desorbs at a much lower temperature than water (peak maxima at 492 versus 613 K). It should be mentioned that minor amounts of methylformate were formed at the higher methanol concentrations investigated. The generation of this compound, which is not observed under standard conditions, is rationalized based on the reaction between methanol and the CO byproduct, a transformation that is applied at the industrial scale to produce this chemical [32]. Remarkably, when more than 0.15 MPa of water were present in the gaseous feed significant catalyst deactivation was observed, which limited the range of partial pressures that could be probed to explore the kinetic effect of this compound (Fig. S4d). Indeed, a progressive depletion of the activity over 12 h was observed when feeding a stream containing water in a 5-times higher amount (pH2O = 0.125 MPa) than that generated upon CO2 hydrogenation under standard conditions (Fig. S6). Analysis by XRD and N2 sorption of the used sample (Fig. S7) evidenced a substantial increase of the particle size (from 8 to 18 nm) and drop in the surface area (from 123 to 47 m2 gÀ1), which point to sintering as the main cause of deactivation. Since the presence of water in the feed is unlikely under practical conditions, its impact on the catalyst was not investigated further. On the contrary, co-feeding of CO, the chief byproduct of this process, is highly likely and may for example be attained through a recycling stream. Earlier, it has been shown that the presence of this reducing gas can boost the catalytic activity, forming additional oxygen vacancies in situ, which are identified as the active sites [10]. However, by adding this compound to the feed in the absence of an inert gas, i.e., at a substantially higher partial pressure of hydrogen, a decrease in the overall activity was Fig. 3. H2O- and MeOH-TPD profiles of bulk In2O3. 317 observed. To understand the role of CO further, In2O3 was tested at variable total pressure (P = 3.5, 4.5, or 5.5 MPa) in a stream with a molar CO2:H2:He ratio of 1:4:0.5 adding progressively higher amounts of CO (16–70 times the quantity produced during reaction, pCO = 0.2–0.8 MPa) and correspondingly decreasing the inert carrier content to ensure that the total flow of CO2 and H2 remained the same as for the other kinetic tests (Fig. 4a). The admission of CO to the feed led to the expected increase in the methanol space-time yield (STYMeOH) at the lowest pressure, to a slight drop of this parameter at the intermediate pressure, and to a significant decrease in activity at the highest pressure. This set of measurements suggests that there are partial pressures of reducing (H2, CO) and oxidizing (CO2) agents in the feed stream for which the surface of the catalyst comprises a high abundance of vacancies but is not yet overly reduced. The role of CO as a reagent cannot be accurately assessed without the use of isotopically labelled gases. However, the experiment conducted at the lowest pressure (3.5 MPa) hints that the catalyst is somewhat active for CO hydrogenation to methanol. Indeed, the ratio of COin:COout was constant at 1.1 and the mass balance was closed and no other products other than methanol were detected. The performance in pure CO hydrogenation could not be accurately assessed, as the feeding of a CO:H2 = 1:2 mixture led to quick (<1 h) deactivation of In2O3, as previously reported [10]. Still, the first chromatogram collected after 20 min on stream suggested a CO conversion of 2% to methanol at 573 K and 5 MPa. Clearly, it is not possible to conclude whether CO is hydrogenated to the alcohol directly or transformed into CO2 by reaction with surface O atoms of the material, which is converted into the alcohol in a second step. To gather further insights into the reducing effect of CO, the catalyst was exposed to gradually more elevated concentrations of this species. Specifically, In2O3 was firstly contacted for 5 h with a gas mixture containing CO2 and H2 in the same molar ratio as applied under the standard conditions and an inert gas (He) in an amount equal to that of CO2. Then, 0.2. 0.5, or 0.8 MPa of CO were added to the gas feed for 3 h while correspondingly decreasing the quantity of inert carrier gas. Finally, the initial conditions were restored (Fig. 4b). The admission of CO to the stream led to a decrease in activity in the range of what observed in the measurements depicted in Fig. 4a. Upon restoring the initial feed conditions, the solids exposed to the lowest and intermediate CO pressures regained their original performance, whereas the catalyst contacted with the highest amount of CO reactivated only partially. In addition, the recovery of functionality was not immediate, as in the two other experiments, but was achieved over 4 h on stream. Samples were extracted from the reactor at different stages during the experiments just described, i.e., after 1 h on stream under standard conditions, after exposure to the stream containing an intermediate amount of CO, and after returning to the initial conditions from the feeds with intermediate and high CO concentrations (points A, B, and C in Fig. 4b, respectively), and analyzed by UV–vis and Raman spectroscopy using the fresh material as a reference (Fig. 4c,d). Since indium oxide is a semiconductor, the presence of oxygen vacancies has been indicated to lower the energy gap between valence and conduction bands [33], which can be probed by these techniques, thus gaining a qualitative indication about such defects in its structure. As-prepared In2O3 produced a strong band in the 240–320 nm region of the UV–vis spectrum, which overlaps with a second equally intense band between 320 and 450 nm. The first is specific to the stoichiometric oxide [34], while the second is indicative of oxygen vacancies [21]. Indeed, a smaller band gap leads to a red shift of the signal. After exposure to the reducing reaction environment (sample A), a broad absorbance in the visible range was additionally detected along with a slight increase of the intensity of the band close to the visible 318 M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 region, which evidence a higher concentration of oxygen vacancies, as expected. The band located at 320–450 nm became weaker after contact of the catalyst with the intermediate CO amount (sample B), likely due to the formation of islands of metallic In. The spectrum of the material exposed again to the standard feed (sample C) comprises the same features as in the spectra of samples A and B, but their intensity is midway, in line with the only partial recovery of the initial activity level. Similar trends are evident from the Raman spectra. In this case, the presence of vacancies causes the broadening of the scattering phenomena 306, 496, and 630 cm–1 associated with InO6 octahedra [35]. The feature at 306 cm–1 is particularly sensitive to vacancies, which cause an asymmetric broadening of the Raman band towards the far infrared region accompanied by the appearance of a new signal between 400 and 430 cm–1 [36,37]. These features are clearly evident in samples A and C, whereas they are absent in sample B. XRD analysis (not shown) indicated no significant changes in morphology and particle size in any of the samples. Overall, these data suggest that, unless the oxide has been excessively reduced, the surface reaction InxOy + CO ¢ InxOyÀ1 + CO2 is in fact a reversible process, dependent on the partial pressures of the two carbon oxides at a given H2 concentration. 3.3. Reaction mechanism Under relevant hydrogen pressures, a partial reduction of the In2O3(1 1 1) surface occurs through the homolytic dissociation of molecular hydrogen forming two surface OH groups, followed by concomitant water formation and desorption and the generation of an oxygen vacancy. As recently reported [20], the conditions applied upon hydrogenation reactions favor a surface configuration, in which one oxygen vacancy per unit cell is predominantly exposed, that is also hereon considered. Indeed, the energy required to abstract a second O atom in the same lattice unit was found to be so high that this event is unlikely even under high pressures of H2. H2 is activated heterolytically as Hd+-HdÀ at the In3O5 ensemble that surrounds the vacancy (Fig. 5a). While Hd+ binds to one oxygen atom, the species holding the negative charge Fig. 4. (a) Evolution of the STYMeOH upon replacing He in the feed by progressively higher amounts of CO (a) at different total pressures and (b) at 5.5 MPa with subsequent restoration of initial feed mixture. (c) UV–vis and (d) Raman spectra of the fresh catalyst and of samples extracted from the reactor at specific points corresponding to the labels in (b). The inset in (c) shows the relative amount of reactive and intert gases to which samples A–C were exposed. Dashed lines and blue-highlighted regions in (d) indicate the bands associated with bulk In2O3 and oxygen vacancies, respectively. Fig. 5. (a) Top-view of the In2O3–x(1 1 1) surface, with an oxygen vacancy per (1 Â 1) cell, surrounding the outlined In3O5 ensemble. (b) Top view of the activated co-adsorption of CO2 and H2 on In2O3–x(1 1 1). Color code: In (blue), O (red), C (black), and H (white). The In and O atoms in the outermost layer are represented as solid three-dimensional spheres, while those in the second and third outermost layers as two-dimensional spheres colored in a progressively lighter color. (c) Gibbs energy profile for the activation of reactants. Conditions: P = 5 MPa and T = 573 K. M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 is accommodated at one In atom and its electron density is delocalized among the three In atoms of the ensemble. Thus, indium oxide is capable of selectively activating hydrogen such that the polarity of the Hd+-HdÀ species is preserved, in a similar way to materials containing frustrated Lewis pairs [38,39]. This H2 activation is exergonic (DG = À0.44 eV) and associated with an activation free energy of 0.95 eV. CO2 chemisorbs at the ensemble, adjacent to the OHd+ and InHdÀ groups formed (Fig. 5b), surpassing an activation barrier of 0.41 eV (Fig. 5c). This configuration, i.e., a surface containing one oxygen vacancy per unit cell, chemisorbed CO2, and heterolytically dissociated H2, is the only stable starting situation for CO2 hydrogenation. Indeed, the formation of a CAH bond from a homolytically dissociated H2, as it would happen over the stoichiometric surface, exhibits an activation free energy of 3.37 eV. Therefore, oxygen vacancies are required for the reaction to proceed under practical conditions. CO, when present in the feed, can increase the vacancy formation rate (DG = À0.29 eV). The reaction profile expressed in terms of the Gibbs free energy at 573 K and 5 MPa is depicted in Fig. 6 and the structures of all adsorbates are presented in Fig. S8. For the sake of clarity, the activation of the reactants shown in Fig. 5 is lumped in a single step in this figure. The reaction profile in terms of internal energy at 0 K is plotted in Fig. S9 for comparison. The hydride in the InHdÀ species formed upon H2 activation is transferred to the chemisorbed CO2 giving rise to a formate species, CHOÀ (Fig. 6, red path). This pro2 cess has an activation free energy of 0.44 eV and is exergonic by À1.0 eV. Then, a proton transfer forming chemisorbed formic acid takes place in a more energy-demanding step (DGà = 1.32 eV) characterized by a reaction free energy of 1.17 eV. The alternative transfer of the surface HdÀ to produce COOH and the subsequent proton transfer to form formic acid has a prohibitive activation free energy (DGà = 1.95 eV). Then, a second H2 molecule is heterolytically adsorbed onto an In-O pair adjacent to the chemisorbed formic acid intermediate. Again, the HdÀ is inserted into the intermediate to form CH2OOH in a step with low activation energy (DGà = 0.29 eV) that is exergonic by À0.81 eV. From this intermediate two scenarios arise: (i) the oxygen in CH2OOH bound to the vacancy surface site is protonated, or (ii) its hydroxyl group is protonated, with concomitant dissociation into CH2O and H2O. Following the first route (Fig. 6, maroon path), the formation of CH2(OH)2 requires 1.21 eV and results in a free energy increase of 0.97 eV. After the heterolytic adsorption of a third H2 molecule, a hydroxyl group dissociates from the intermediate, filling the vacancy site. This step, which exhibits an activation free energy 319 of 0.45 eV and is exergonic by À0.09 eV, also encompasses a proton transfer to a surface oxygen leading to physisorbed formaldehyde. The latter readily reacts with the hydride to form methoxide, in view of the negligible activation energy and high exergonicity (DG = À1.30 eV). Finally, the methoxide is protonated and methanol desorbs with an associated desorption free energy of À0.27 eV. The two surface hydroxyl groups can combine generating water, which desorbs in an overall endergonic process (DG = 0.88 eV). Along the second path (Fig. 6, orange path), the CH2OOH intermediate dissociates into CH2O and OHÀ (DGà = 0.89 eV, DG = À0.15 eV), the latter being subsequently protonated to form water (DG = 0.36 eV), which desorbs (DG = À0.14 eV). A third H2 molecule is heterolytically dissociated at an In-O pair next to the chemisorbed formaldehyde. The HdÀ is transferred to CH2O forming methoxide, a step which requires only 0.48 eV and is exergonic by À1.33 eV. The methoxide is finally protonated in a rather demanding step (DGà = 1.37 eV, DG = 1.07 eV) and the resulting methanol desorbs spontaneously (À0.21 eV). The two paths discussed above describe the consecutive addition of three heterolytically activated H2 molecules, which implies the alternate transfer of negatively and positively polarized hydrogen atoms. The co-adsorption of multiple H2 molecules could allow the formation of several CAH or OAH bonds consecutively. In contrast, Ye et al. [17] reported a mechanism on the (1 1 0) surface in which methanol is formed by three consecutive hydride additions followed by a proton transfer and each elementary step is considered independently, adding neutral hydrogen atoms, which might cause their electrons to have an exceedingly high chemical potential. Given the complexity of the In2O3 electronic structure, such approach might lead to artificially low barriers. Thus, the coadsorption of multiple H2 molecules was assessed (Fig. 6, green path). After the initial HdÀ transfer leading to CHO2, another H2 is adsorbed on the surface heterolytically next to the intermediate. The HdÀ is transferred to the intermediate to form CH2O2 (DGà = 1.15 eV, DG = 0.16 eV). The co-adsorption of a third H2 molecule next to the intermediate and the two hydroxyl groups is highly unfavored, as the step is endergonic by 1.10 eV. Moreover, the activation free energy of the HdÀ to form CH3OÀ adds up 1.19 eV, making the whole process prohibitive with a net barrier of 2.29 eV. Conversely, the formation of CH2OOH from CH2O2 and a surface hydroxyl group only requires 0.69 eV. Therefore, the consecutive formation of two CAH bonds followed by an OAH bond to obtain Fig. 6. Gibbs energy profile for the hydrogenation of CO2 through the most representative paths. The hydrogenation of CO2 to CH2OOH is shown in red. The two possible routes to CH3OH from this species are depicted in orange and maroon. The alternative formation of CH2OOH from CHO2 through the co-adsorption of two H2 molecules is shown in green. The RWGS reaction is marked in blue. Conditions: P = 5 MPa and T = 573 K. Same color coding as in Fig. 5. TS = transition state. 320 M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 CH2OOH is favored over the alternative route, but the direct formation of CH3OÀ is highly unlikely. The mechanism from CH2OOH proceeds as described above. Overall, the mechanism of CO2 hydrogenation on In2O3 can be understood as a hydride-proton transfer, hence resembling transfer-hydrogenation reaction schemes. The analogous action mode of indium oxide to oxides containing frustrated Lewis pairs ensures a higher control on the reaction selectivity. Indeed, even if methanol desorption was thermodynamically less favorable, the polarity of the H species present at the surface would be insufficient to drive its further transformation into methane. The competitive formation of CO via the RWGS reaction has also been evaluated in the presence of one additional water molecule (Fig. 6, blue path). Neglecting the effect of H2O present in the environment would lead to an overestimation of the activation energies involved in the route, since this compound facilitates reaching the transition states. After the initial co-adsorption of CO2 and H2, the HdÀ can transfer to an O atom instead to the C atom giving rise to COOH rather than CHO2 (DGà = 0.62 eV, i.e., 0.18 eV higher than to form CHO2, DG = 0.42 eV). The proton transfer to this same oxygen leading to CO and H2O requires 1.49 eV and the process is endergonic by 1.16 eV. Finally, the desorption of the two products are exergonic steps (DG = À0.15 and À0.39 eV, respectively). Catalyst poisoning by the COOH species, the conversion of which is most demanding along this route, is unlikely since the reverse reaction has a barrier (1.1 eV) that can be overcome under the experimental conditions. 3.4. Microkinetic modeling Based on the DFT results, microkinetic simulations were conducted considering a differential reactor operated under the conditions experimentally applied, to assess the dependence of the reaction rate on the temperature and the partial pressures of reactants and products. Apparent activation energies of 1.73 eV (166 kJ molÀ1) and 1.96 eV (189 kJ molÀ1) were calculated for CO2 hydrogenation to methanol and the RWGS reaction, respectively (Fig. S10a). These are higher than the corresponding values obtained from the experimental results (Fig. 2a), since the influence of the thermodynamic equilibrium is not considered in the calculations while it is never fully suppressed in the actual tests. To account for these deviations, a tangent was asymptotically fitted to the empirical data at low temperatures, as far away from the equilibrium-influenced region as permitted (Fig. S10b). In this way, the computed apparent activation energies converge to 1.56 eV (150 kJ molÀ1) for CO2 hydrogenation and 1.85 eV (179 kJ molÀ1) for the RWGS reaction, which are in reasonable agreement with the theoretical values. These results show that even an elusive parameter such as the activation energy was qualitatively well reproduced in the simulations. Most importantly, the difference in the activation energies for the desired and competitive reaction, which controls the selectivity of the process, is 0.23 eV based on the theoretical data, which properly compares with the experimentally determined gap of 0.29 eV. With respect to the reaction orders, that for hydrogen was positively defined (0.33), while the value determined for CO2 is zero. The detrimental effect of water and methanol were properly retrieved with reaction orders of À0.57 and À0.14, respectively (Fig. S11). Therefore, all the computed reaction orders are in good agreement with the experimental data, with errors within 0.15. The exception is water, for which the deviation is 0.33, a difference which is likely caused by microsolvation effects. Hence, the DFT modeling is qualitatively solid and the reason behind the discrepancy lies in second order terms, i.e., the assistance of water molecules to key selectivity-determining steps. 4. Conclusions In this study, the mechanistic and kinetic fingerprints of CO2 hydrogenation on In2O3 were elucidated in detail. The apparent activation energy experimentally determined for CO2-based methanol synthesis was found to be higher than the one for the RWGS reaction, which explains the superior methanol selectivity of this catalyst. The methanol formation rate slightly diminished when applying higher CO2 concentrations, in view of the moderate promotion of the RWGS reaction, while it was significantly boosted at higher H2 contents, which enhanced both CO2 conversion and methanol selectivity. With respect to the impact of the products concentration on the reaction rate, methanol caused a minor inhibition, whereas water was strongly detrimental. These findings were corroborated by temperature-programmed desorption experiments, showing that methanol evolves from the oxide at a 120 K lower temperature than water. It was also observed that water and the CO byproduct led to catalyst deactivation due to sintering and over-reduction when their concentrations were 4 and 20-times higher than the amounts produced by the reaction under standard reaction conditions, respectively. In the case of CO, the activity was partially depleted due to surface reduction also for quantities equal to 10–20 times of the typical amount, but could be restored upon removal of CO from the feed owing to in situ reoxidation by CO2. Structural characterization of the In2O3 catalyst evidenced a dominant exposure of the (1 1 1) termination, which was in line with the Wulff construction calculated for this material. Theoretical modeling of CO2 hydrogenation over this surface evidenced that the oxygen vacancies generated under the reaction conditions can activate CO2 and heterolytically split H2. The Hd+-HdÀ species formed in the latter process are stabilized by the In3O5 ensemble surrounding the vacancy, hence preserving their polarity. Accordingly, the reaction was considered to proceed via the selective and consecutive addition of hydrides and protons, rather than H radicals as previously reported. Two plausible paths to methanol were outlined, which follow identical steps until the second hydride addition. From the CH2OOH intermediate obtained, the route subsequently comprising the formation of methanediol is energetically favored. Microkinetic simulations provided values for the apparent activation energies of CO2 hydrogenation and the RWGS reaction that are in good agreement with the measured data. They also led to orders for reactants and products in methanol synthesis that generally match the values experimentally obtained. This thorough investigation of the kinetic and molecular aspects of CO2 hydrogenation over In2O3 will be highly instrumental to rationally guide further catalyst optimization and will be essential for an eventual reactor and process design. Acknowledgements Total Research & Technology Feluy and CTQ 2015-68770-R are thanked for financial support and BSC-RES for generous computational resources. Dr. S. Mitchell is thanked for the electron microscopy measurements and the Scientific Center for Optical and Electron Microscopy at the ETH Zurich, ScopeM, for the use of their facilities. Prof. R. Spolenak from ETH Zurich is acknowledged for access to Raman spectroscopy. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.jcat.2018.03.014. The computed structures can be retrieved from the ioChem database at: https://doi.org/10.19061/iochem-db-1-63. M.S. Frei et al. / Journal of Catalysis 361 (2018) 313–321 References [1] E.V. Kondratenko, G. Mul, J. Baltrusaitis, G.O. Larrazábal, J. Pérez-Ramírez, Status and perspectives of CO2 conversion into fuels and chemicals by catalytic, photocatalytic and electrocatalytic processes, Energy Environ. Sci. 32 (2013) 3112–3135, https://doi.org/10.1039/c3ee41272e. [2] A. Álvarez, M. Borges, J.J. Corral-Pérez, J.G. Olcina, L. Hu, D. Cornu, R. Huang, D. Stoian, A. Urakawa, CO2 activation over catalytic surfaces, ChemPhysChem 18 (2017) 3135–3141, https://doi.org/10.1002/cphc.201700782. [3] E. Fiedler, G. Grossmann, D.B. Kersebohm, G. Weiss, C. Witte (Eds.), Methanol, Ullmann’s Encyclopedia of Industrial Chemistry, vol. 23, Wiley-VCH, Weinheim, 2012, pp. 25–48, https://doi.org/10.1002/14356007. [4] K.A. Ali, A.Z. Abdullah, A.R. Mohamed, Recent development in catalytic technologies for methanol synthesis from renewable sources: a critical review, Renew. Sustain. Energy Rev. 44 (2015) 508–518, https://doi.org/ 10.1016/j.rser.2015.01.010. [5] F. Studt, M. Behrens, E.L. Kunkes, N. Thomas, S. Zander, A. Tarasov, J. Schumann, E. Frei, J.B. Varley, F. Abild-Pedersen, J.K. Nørskov, R. Schlögl, The mechanism of CO and CO2 hydrogenation to methanol over Cu-based catalysts, ChemCatChem 7 (2015) 1105–1111, https://doi.org/10.1002/cctc.201500123. [6] A. Álvarez, A. Bansode, A. Urakawa, A.V. Bavykina, T.A. Wezendonk, M. Makkee, J. Gascon, F. Kapteijn, Challenges in the greener production of formates/formic acid, methanol, and DME by heterogeneously catalyzed CO2 hydrogenation processes, Chem. Rev. 117 (2017) 9804–9838, https://doi.org/10.1021/acs. chemrev.6b00816. [7] W. Wang, S. Wang, X. Ma, J. Gong, Recent advances in catalytic hydrogenation of carbon dioxide, Chem. Soc. Rev. 40 (2011) 3703–3727, https://doi.org/ 10.1039/c1cs15008a. [8] K. Chen, X. Duan, H. Fang, X. Liang, Y. Yuan, Selective hydrogenation of CO2 to methanol catalyzed by Cu supported on rod-like La2O2CO3, Catal. Sci. Technol. (2018), https://doi.org/10.1039/C7CY01998J. [9] S. Borman, Carbon dioxide hydrogenated to methanol on large scale, Chem. Eng. News 94 (2016) 7. [10] O. Martin, A.J. Martín, C. Mondelli, S. Mitchell, T.F. Segawa, R. Hauert, C. Drouilly, D. Curulla-Ferré, J. Pérez-Ramírez, Indium oxide as a superior catalyst for methanol synthesis by CO2 hydrogenation, Angew. Chem. Int. Ed. 55 (2016) 6261–6265, https://doi.org/10.1002/anie.201600943. [11] P. Gao, S. Li, X. Bu, S. Dang, Z. Liu, H. Wang, L. Zhong, M. Qiu, C. Yang, J. Cai, W. Wie, Y. Sun, Direct conversion of CO2 into liquid fuels with high selectivity over a bifunctional catalyst, Nat. Chem. 9 (2017) 1019–1024, https://doi.org/ 10.1038/nchem.2794. [12] J. Su, D. Wang, Y. Wang, H. Zhou, C. Liu, S. Liu, C. Wang, W. Yang, Z. Xie, M. He, Direct conversion of syngas to light olefins over Zr–In2O3 and SAPO–34 bifunctional catalysts: design of oxide component and construction of reaction network, ChemCatChem (2018), https://doi.org/10.1002/cctc.201702054. [13] N. Ruia, Z. Wang, K. Sun, J. Ye, Q. Ge, C. Liu, CO2 hydrogenation to methanol over Pd/In2O3: effects of Pd and oxygen vacancy, Appl. Catal. B 218 (2017) 488–497, https://doi.org/10.1016/j.apcatb.2017.06.069. [14] A. García-Trenco, A. Regoutz, E.R. White, D.J. Payne, M.S.P. Shaffer, C.K. Williams, PdIn intermetallic nanoparticles for the hydrogenation of CO2 to methanol, Appl. Catal. B 220 (2018) 9–18, https://doi.org/10.1016/j. apcatb.2017.07.069. [15] J. Wang, G. Li, Z. Li, C. Tang, Z. Feng, H. An, H. Liu, T. Liu, C. Li, A highly selective and stable ZnO-ZrO2 solid solution catalyst for CO2 hydrogenation to methanol, Sci. Adv. 3 (2017), https://doi.org/10.1126/sciadv.1701290. [16] J. Skrzypek, M. Lachowska, D. Serafin, Methanol synthesis from CO2 and H2: dependence of equilibrium conversions and exit equilibrium concentrations of components on the main process variables, Chem. Eng. Sci. 45 (1990) 89–96, https://doi.org/10.1016/0009-2509(90)87083-5. [17] J. Ye, C. Liu, D. Mei, Q. Ge, Active oxygen vacancy site for methanol synthesis from CO2 hydrogenation on In2O3(110): a DFT Study, ACS Catal. 3 (2013) 1296–1306, https://doi.org/10.1021/cs400132a. [18] J. Ye, C. Liu, Q. Ge, DFT study of CO2 adsorption and hydrogenation on the In2O3 surface, J. Phys. Chem. C 116 (2012) 7817–7825, https://doi.org/10.1021/ jp3004773. [19] M. Dou, M. Zhang, Y. Chen, Y. Yu, DFT study of In2O3-catalyzed methanol synthesis from CO2 and CO hydrogenation on the defective site, New J. Chem. (2018), https://doi.org/10.1039/c7nj04273f. 321 [20] D. Albani, M. Capdevila-Cortada, G. Vilé, S. Mitchell, O. Martin, N. López, J. Pérez-Ramírez, Semihydrogenation of acetylene on indium oxide: proposed single–ensemble catalysis, Angew. Chem. Int. Ed. 56 (2017) 10755–10760, https://doi.org/10.1002/anie.201704999. [21] G. Kresse, J. Furthmüller, Efficiency of ab initio total energy calculations for metals and semiconductors using a plane–wave basis set, Comput. Mater. Sci. 6 (1996) 15–50, https://doi.org/10.1016/0927-0256(96)00008-0. [22] G. Kresse, J. Hafner, Ab initio molecular dynamics for liquid metals, Phys. Rev. B 47 (1993) 558–561, https://doi.org/10.1103/PhysRevB.47.558. [23] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77 (1996) 3865–3868, https://doi.org/10.1103/ PhysRevLett. 77.3865. [24] P.E. Blöchl, Projector augmented-wave method, Phys. Rev. B 50 (1994) 17953– 17979, https://doi.org/10.1103/PhysRevB.50.17953. [25] G. Henkelman, B.P. Uberuaga, H. Jónsson, A climbing image nudged elastic band method for finding saddle points and minimum energy paths, J. Chem. Phys. 113 (2000) 9901–9904, https://doi.org/10.1063/1.1329672. [26] M. Álvarez-Moreno, C. De Graaf, N. Lopez, F. Maseras, J.M. Poblet, C. Bo, Managing the computational chemistry big data problem: the ioChem–BD Platform, J. Chem. Inf. Mod. 55 (2014) 95–103, https://doi.org/ 10.1021/ci500593j. [27] I. Chorkendorff, J.W. Niemantsverdriet, Reaction Rate Theory, Concepts of Modern Catalysis And Kinetics, vol. 1, Wiley-VCH, Weinheim, 2003, pp. 108– 115, https://doi.org/10.1002/3527602658. [28] Q. Li, R. García-Muelas, N. López, Microkinetics of alcohol reforming for H2 production from a FAIR density functional theory database, Nat. Commun. 9 (2018) 526–533, https://doi.org/10.1038/s41467-018-02884-y. [29] O. Bierwagen, J. Rombach, J.S. Speck, Faceting control by the stoichiometry influence on the surface free energy of low-index bcc-In2O3 surfaces, J. Phys.: Condens. Matter 28 (2016) 224006–2240013, https://doi.org/10.1088/09538984/28/22/224006. [30] M. Zhang, W. Wang, Y. Chen, Insight of DFT and ab initio atomistic thermodynamics on the surface stability and morphology of In2O3, Appl. Surf. Sci. 434 (2018) 1344–1352, https://doi.org/10.1016/j. apsusc.2017.11.258. [31] O. Bierwagen, M.E. White, M.Y. Tsai, J.S. Speck, Plasma-assisted molecular beam epitaxy of high quality In2O3(001) thin films on Y-stabilized ZrO2(001) using In as an auto surfactant, Appl. Phys. Lett. 95 (2009) 262105–262107, https://doi.org/10.1063/1.3276910. [32] P. Zehner, K. Bittins, W. Haarde, U. Eiden, D. Wolff, M. Herr, L. Hupfer, Preparation of methyl formate, US Patent 5 401 873 (1995), to BASF AG. [33] F. Lei, Y. Sun, K. Liu, S. Gao, L. Liang, B. Pan, Y. Xie, Oxygen vacancies confined in ultrathin indium oxide porous sheets for promoted visible-light water splitting, J. Am. Chem. Soc. 136 (2014) 6826–6829, https://doi.org/ 10.1021/ja501866r. [34] F. Jiang, H. Zhao, H. Chen, C. Xu, J. Chen, Enhancement of photocatalytic decomposition of perfluorooctanoic acid on CeO2/In2O3, RSC Adv. 6 (2016) 72015–72021, https://doi.org/10.1039/c6ra09856h. [35] M. Kaur, N. Jain, K. Sharma, S. Bhattacharya, M. Roy, A.K. Tyagi, S.K. Gupta, J.V. Yakhmi, Room-temperature H2S gas sensing at ppb level by single crystal In2O3 whiskers, Sens. Actuat. B Chem 133 (2008) 456–461, https://doi.org/ 10.1016/j.snb.2008.03.003. [36] M. Kumar, V.N. Singh, F. Singh, K.V. Lakshmi, B.R. Mehta, J.P. Singh, On the origin of photoluminescence in indium oxide octahedron structures, Appl. Phys. Lett. 92 (2008) 171907–171909, https://doi.org/10.1063/1.2910501. [37] T. Bielz, H. Lorenz, W. Jochum, R. Kaindl, F. Klauser, B. Klötzer, S. Penner, Hydrogen on In2O3: reducibility, bonding, defect formation, and reactivity, J. Phys. Chem. C 114 (2010) 9022–9029, https://doi.org/10.1021/jp1017423. [38] M. Ghoussoub, S. Yadav, K.K. Ghuman, G.A. Ozin, V.C. Singh, Metadynamicsbiased ab Initio molecular dynamics study of heterogeneous CO2 reduction via surface frustrated lewis pairs, ACS Catal. 6 (2016) 7109–7117, https://doi.org/ 10.1021/acscatal.6b01545. [39] S. Zhang, Z. Huang, Y. Ma, W. Gao, J. Li, F. Cao, L. Li, C. Chang, Y. Qu, Solid frustrated-Lewis-pair catalysts constructed by regulations on surface defects of porous nanorods of CeO2, Nat. Commun. 8 (2017) 15266–15276, https://doi. org/10.1038/ncomms15266.