Research ArticleChemistry

Rationally designed indium oxide catalysts for CO2 hydrogenation to methanol with high activity and selectivity

See allHide authors and affiliations

Science Advances  17 Jun 2020:
Vol. 6, no. 25, eaaz2060
DOI: 10.1126/sciadv.aaz2060


Renewable energy-driven methanol synthesis from CO2 and green hydrogen is a viable and key process in both the “methanol economy” and “liquid sunshine” visions. Recently, In2O3-based catalysts have shown great promise in overcoming the disadvantages of traditional Cu-based catalysts. Here, we report a successful case of theory-guided rational design of a much higher performance In2O3 nanocatalyst. Density functional theory calculations of CO2 hydrogenation pathways over stable facets of cubic and hexagonal In2O3 predict the hexagonal In2O3(104) surface to have far superior catalytic performance. This promotes the synthesis and evaluation of In2O3 in pure phases with different morphologies. Confirming our theoretical prediction, a novel hexagonal In2O3 nanomaterial with high proportion of the exposed {104} surface exhibits the highest activity and methanol selectivity with high catalytic stability. The synergy between theory and experiment proves highly effective in the rational design and experimental realization of oxide catalysts for industry-relevant reactions.


The rapid increase in global energy demand is driving up carbon dioxide (CO2) emission. In 2017, global CO2 emission from energy consumption reached 33 gigatons, twice more than could be taken up by land- and ocean-based CO2 sinks (1). Therefore, it is urgent to develop effective strategies to slow down the increase and even reduce the level of atmospheric CO2 concentration by efficient capture and utilization of the CO2 to be emitted. As methanol is a viable clean alternative fuel to gasoline and diesel and an important feedstock to produce commodity chemicals, the catalytic conversion of CO2 to methanol using H2 originated from renewable energy sources is a promising approach to reduce the CO2 emission and our dependence on fossil fuels by carbon recycling and to store the renewable energy (solar, wind, biomass, and so on) as chemical energy (25). In addition, efficient CO2 hydrogenation to methanol is central to the successful development of the recently proposed “methanol economy” (6) and “liquid sunshine” (1).

Although many types of metal-based catalysts are well known for the CO2 hydrogenation to methanol reaction, modified copper (Cu) catalysts remained the most efficient and the most extensively investigated (713). Different from industrial methanol synthesis from syngas (CO/H2), one of the main challenges of CO2 hydrogenation to methanol is the low product selectivity due to the parasitic reverse water–gas shift reaction (RWGS; CO2 + H2 → CO + H2O), for which Cu is among the most active catalysts. The selectivity toward methanol is usually lower than 60% over traditional Cu/ZnO-based catalysts (1416). Recently, extensive density functional theory (DFT) and experimental studies on CO2 hydrogenation over the (110) surface of the cubic phase of In2O3 suggested that methanol formation over the surface oxygen vacancy site was favorable to the RWGS reaction (1721). With the In2O3 catalyst, methanol selectivity reached nearly 100% at a high space velocity (gas hourly space velocity of >16,000 hour−1) in the temperature range of 200° to 300°C, although the low reactivity led to relatively low single-pass conversions and thus limited space-time yield (STY) of methanol (22). Therefore, the design and development of more efficient In2O3-based nanocatalysts are imperative for their industrial adoptions in CO2 hydrogenation to methanol, although it remains a great challenge to vastly improve their catalytic performance.

To rationally design In2O3 materials with favorable methanol synthesis performance, it is necessary to fully understand the reaction mechanisms of CO2 hydrogenation to methanol and the RWGS reaction for CO formation over different surface active sites. Previous studies show that methanol formation follows the formate (HCOO*, where * denotes the active site) route, which also involves the dioxymethylene (H2COO*) and methoxy (H3CO*) intermediates (17). In this catalytic mechanism, CO2 hydrogenation to HCOO* occurs at the oxygen vacancy site, which is thermodynamically and kinetically favorable with the formation of stable HCOO* and H2COO* species, whereas the formation of H2COO* on the Cu(111) surface is unfavorable with thermodynamically unstable formaldehyde (H2CO*) species (19, 23). In addition, Ghuman et al. (24) performed experimental and computational studies on the catalytic mechanism of the RWGS reaction on the In2O3(111) surface and suggested that the active site consisted of a Lewis base hydroxide adjacent to a Lewis acid indium, in addition to the oxygen vacancy.

Previous experimental and theoretical studies on the In2O3 catalysts focused on the stable surfaces of the cubic In2O3 (c–In2O3) phase, especially its (110) surface. Even for this extensively investigated surface, there remains uncertainty on the detailed catalytic mechanism for CO2 hydrogenation to methanol and the RWGS reaction. Furthermore, two additional phases are known for In2O3, the corundum-type hexagonal phase (h–In2O3) and the Rh2O3(II)-type orthorhombic phase (o–In2O3) (25, 26). Among the three phases, c–In2O3 is thermodynamically the most stable, both c–In2O3 and h–In2O3 can form at ambient conditions, whereas o–In2O3 may form at high pressure (>15 GPa) (25, 2729). Since both c–In2O3 and h–In2O3 are stable in the temperature range of the CO2 hydrogenation to methanol reaction (200° to 300°C) (30), both may work as catalysts for this reaction. In addition, previous studies show that the thermodynamically most stable surfaces are the (110) and (111) surfaces for c–In2O3 (27, 3134) and the (104) and (012) surfaces for h–In2O3 (27, 28, 3537).

In this work, we first performed extensive DFT calculations to establish the catalytic mechanisms of the In2O3 catalyst during CO2 hydrogenation to methanol and to CO by identifying the preferred pathways and rate-determining steps (RDSs). We then evaluated the performance of the stable surfaces of c–In2O3 and h–In2O3 as mentioned above based on the theoretical data to search for a superior catalyst. Our calculations predict the h–In2O3(104) surface to have the best catalytic performance in terms of reactivity and selectivity. On the basis of our computational studies, we synthesized nanoparticles in pure c–In2O3 and h–In2O3 phases of different morphologies and evaluated their performance for CO2 hydrogenation to methanol. The results show that the catalytic performance depends on the shape of In2O3 nanoparticles due to their different phases and exposed facets. Furthermore, h–In2O3 nanocrystals with a high proportion of the exposed {104} surface exhibited the best performance with high catalytic stability during CO2 hydrogenation, confirming our DFT prediction. Methanol selectivity reached 92.4% with a single-pass CO2 conversion of more than 17% under the reaction conditions of 300°C, 5.0 MPa, 9000 ml gcat−1 hour−1, and H2/CO2 = 6. Thus, our work serves as an exemplary case toward using computational methods to design new and more efficient catalysts.


We first performed extensive DFT calculations to determine the most favorable reaction pathways and the corresponding RDS for the formations of CH3OH and CO from CO2 hydrogenation on the c–In2O3(110) surface with an oxygen vacancy. Although several groups including ourselves have previously investigated CO2 hydrogenation pathways on this surface, there remain debates in the formation mechanisms of these two major products.

As shown in fig. S1A, model of the perfect c–In2O3(110) surface is similar to that used in our previous study. According to the calculated formation energies of the different surface oxygen vacancies, we chose the oxygen vacancy at the Ov4 site with the highest formation energy as the active site for CO2 hydrogenation, in expectation that higher oxygen affinity of the Ov4 site will lead to stronger CO2 adsorption and reactivity. Thus, model of the defective c–In2O3(110) surface shown in Fig. 1A also resembles that used in our previous work. An important finding in this work is our identification of the distinct roles of two different CO2 adsorption configurations at the Ov4 site, the linear and bent structures denoted as ln–CO2* and bt–CO2*, as shown in Fig. 1B. In addition, because of the presence of some surface O sites, CO2 can also chemisorb at the O site to form the carb–CO2* structure, as shown in fig. S2A. Calculated adsorption energies of these three configurations are −0.09, −0.42, and − 0.85 eV, respectively. In ln–CO2*, CO2 physisorbs above the Ov4 site, whereas in bt–CO2*, it chemisorbs at the Ov4 site by forming the In–C and In–O bonds. In carb–CO2*, CO2 reacts with the remaining surface O site to form a carbonate species. Despite the thermodynamic stability of carb–CO2*, our further calculations suggest it to be a spectator species.

Fig. 1 CO2 hydrogenation to methanol and CO on the defective c–In2O3(110) surface.

(A) Optimized structure of the c–In2O3(110) surface with an oxygen vacancy (O4v). (B) Optimized structures of two CO2 adsorption configurations on this surface. (C) Potential energy profiles of CO2 hydrogenation to methanol (black lines) and CO (red lines) on this surface with transition state structures shown. Colors: In, brown; C, black; H, white; O, red; Ov, blue. Only exposed surface atoms are explicitly shown.

Calculations of the reaction pathways reveal that the ln–CO2* and bt–CO2* structures are relevant to CO2 hydrogenation pathways to different products, as shown in Fig. 1C by the black lines starting from ln–CO2* for CH3OH formation and the red lines starting from bt–CO2* for CO formation on the defective c–In2O3(110) surface. In the former pathway, CO2 in ln–CO2* is first hydrogenated by a hydride adsorbed at the In3 site to yield mono–HCOO* via transition state TS1 (Ea = 0.24 eV). After facile conversion of mono–HCOO* to bi–HCOO* by a simple rotation, it is further hydrogenated by another hydride located at the In3 site to yield the H2COO* structure via TS2 with a very high energy barrier of 1.54 eV. H2COO* dissociates to fill the oxygen vacancy site giving the H2CO* intermediate via TS3 (Ea = 0.69 eV). H2CO* is again hydrogenated by a hydride situated at the In3 site with the assistance of a proton adsorbed at the O4 site to produce the methoxy H3CO* intermediate via TS4 (Ea = 0.11 eV). H3CO* is further protonated by the proton located at the O4 site to generate the H3COH* structure via TS5 (Ea = 0.02 eV). Desorption of the physisorbed CH3OH molecule in H3COH* results in the perfect c–In2O3(110) surface, and the oxygen vacancy site regenerates by its reaction with H2. Thus, from the calculated energy barriers, the RDS for CH3OH formation is hydrogenation of bi–HCOO* to H2COO* with an energy barrier of 1.54 eV. Consistent with our previous calculations, the formate (bi–HCOO*) and methoxy (H3CO*) species are thermodynamically the most stable surface intermediates in the CO2 hydrogenation to CH3OH pathway.

In sharp contrast, the chemisorbed CO2 molecule in the bt–CO2* structure is first protonated by a proton adsorbed at the nearby O3 site to yield the COOH* structure via TS1b with a high energy barrier of 1.26 eV. COOH* then dissociates to fill the oxygen vacancy by the hydroxyl OH* intermediate and to also give the CO* intermediate via TS2b (Ea = 0.71 eV). Desorption of the physisorbed CO molecule results in the protonated c–In2O3(110) surface, and the oxygen vacancy site also regenerates by its reaction with H2. Thus, the RDS for CO formation is protonation of bt–CO2* to COOH* with an energy barrier of 1.26 eV. As the energy barrier of the RDS for CH3OH formation is higher than that for CO formation, CH3OH selectivity should be lower than CO selectivity when catalyzed by the Ov4 site on the defective c–In2O3(110) surface. Furthermore, compared to ln–CO2*, stronger CO2 adsorption in bt–CO2* should result in a higher sticking coefficient for the CO formation pathway, which further favors CO formation over CH3OH formation.

Although the CH3OH formation pathway predicted by our extensive DFT calculations is in general consistent with the earlier study of Ye and co-workers (17), we note that previous studies have not identified the distinct roles of the different CO2 adsorption structures and especially the different H adsorbates, i.e., hydride adsorbed at the In site and proton adsorbed at the O site, which is crucial for the proper determination of the reaction pathways. Distinction between the two H adsorption structures was obscured in earlier works due to simplified treatment of the oxide surface by ignoring its different nature from the metal surface, in that an H adsorbate at the metal or O site likely carries a negative or positive charge, resulting in its different chemical nature and reactivity.

Conclusions drawn from our mechanistic study of CO2 hydrogenation to CH3OH and CO on the defective c–In2O3(110) surface facilitates the rapid search for the catalytically most effective In2O3 surfaces for CH3OH formation. Thus, we further performed DFT calculations on the other stable c–In2O3 and h–In2O3 surfaces by assuming a similar mechanism, and we also calculated CO2 adsorption energies and energy barriers of the RDS to evaluate the performance of these In2O3 surfaces.

As shown in fig. S1, we predicted the stability of the In2O3 surfaces to follow the order of (111) > (110) for the c–In2O3 phase and (012) > (104) for the h–In2O3 phase. As shown in Fig. 2A, for each surface, we chose the oxygen vacancy site with the highest formation energy as the active site for CO2 hydrogenation. Figure 2B displays the transition state structures for hydrogenation of bi–HCOO* to H2COO*, the RDS for CO2 hydrogenation to CH3OH, whereas Fig. 2C shows those for protonation of bt–CO2* to COOH*, the RDS for CO2 hydrogenation to CO. Structures of the different CO2 adsorption configurations are available in fig. S2.

Fig. 2 Performance of different c–In2O3 and h–In2O3 surfaces for CO2 hydrogenation to methanol.

(A) Optimized structures of the c–In2O3(111), h–In2O3(012), and h–In2O3(104) surfaces each with an oxygen vacancy. (B and C) Transition state structures of the RDSs for CO2 hydrogenation to methanol and CO, respectively, on these three defective surfaces. (D) Histogram of the calculated energy barriers of the RDS for CO2 hydrogenation to methanol (red) and CO (black) and the calculated CO2 adsorption energies in the ln–CO2* (blue) and bt–CO2* (green) configurations on the four defective c–In2O3 and h–In2O3 surfaces. Catalytic activity of CH3OH or CO formation can be approximately correlated with sum of the corresponding energy barrier and CO2 adsorption energy.

We compare the calculated energy barriers of the RDS for all four defective In2O3 surfaces in Fig. 2D along with the CO2 adsorption energies at the chosen oxygen vacancy site in ln–CO2* and bt–CO2* to evaluate their performance in catalyzing the CO2 hydrogenation reaction. Our calculations show that the energy barriers of the RDS of CH3OH formation follow the order of c–In2O3(110) (1.54 eV) > h–In2O3(012) (1.27 eV) > h–In2O3(104) (0.88 eV) > c–In2O3(111) (0.64 eV), whereas CO2 adsorption strength in the ln-CO2* structure follows the order of h–In2O3(104) (−0.14 eV) > c–In2O3(110) (−0.09 eV) > c–In2O3(111) (−0.05 eV) ≈ h–In2O3(012) (−0.05 eV), so the defective c–In2O3(111) and h–In2O3(104) surfaces are the most favorable for CO2 hydrogenation to CH3OH in terms of the catalytic activity. On the other hand, the energy barriers for the RDS of CO formation follow the order of c–In2O3(111) (1.39 eV) > h–In2O3(104) (1.29 eV) > c–In2O3(110) (1.26 eV) > h–In2O3(012) (0.69 eV), whereas CO2 adsorption strength in bt–CO2* follows the order of c–In2O3(110) (−0.42 eV) > h–In2O3(012) (0.18 eV) > c–In2O3(111) (0.28 eV) > h–In2O3(104) (0.69 eV), so the defective c–In2O3(110) and h–In2O3(012) surfaces favor CO formation, whereas the defective h–In2O3(104) and c–In2O3(111) surfaces highly disfavors CO formation. We note that CO2 adsorption strength in the carb–CO2* structure follows the order of h–In2O3(012) (−0.98 eV) > h–In2O3(104) (−0.94 eV) > c–In2O3(110) (−0.85 eV) > c–In2O3(111) (−0.67 eV), although we predict that this structure is a spectator species (fig. S2).

Previous computational studies examined only the flat c–In2O3(110) and c–In2O3(111) surfaces for CO2 hydrogenation, and we further examined the possible role of the stepped surfaces. As shown in fig. S3A, the c–In2O3(110) step surface is actually slightly more stable than its flat surface (fig. S1A), whereas the other stepped surfaces (fig. S3, B to D) are notably less stable than their corresponding flat surfaces (fig. S1). Nevertheless, we further predicted the energy barriers of the RDS and CO2 adsorption energies on all these step surfaces (fig. S3E) and compared to the corresponding flat surfaces. For the c–In2O3(110) step surface, the energy barriers of CH3OH and CO formations are lower by 0.41 and 0.27 eV, respectively, whereas CO2 adsorption energies are slightly less negative by 0.05 eV. Thus, although the c–In2O3(110) step surface is catalytically more active, it also favors CO formation over CH3OH formation. For the other step surfaces, the energy barriers significantly increase by 0.2 to 0.8 eV except for CH3OH formation on the c–In2O3(111) step and h–In2O3(012) step surfaces, where they remain essentially the same. Overall, consideration of these step surfaces does not change our above conclusions, not to ignore the fact that only the c–In2O3(110) step surface is sufficiently stable. As illustrated in Fig. 3, we further rationalize that the threefold oxygen vacancy site on the c–In2O3(111) and h–In2O3(104) surfaces favors the linear CO2 physisorption structure and the HCOO pathway, leading to high CH3OH selectivity, whereas the twofold oxygen vacancy site on the c–In2O3(110) and h–In2O3(012) surfaces facilitates the bent CO2 chemisorption structure and COOH pathway, resulting in high CO selectivity. As the h–In2O3(104) surface is thermodynamically less stable than the h–In2O3(012) surface and the h–In2O3 phase is less stable than the c–In2O3 phase, the highly selective In2O3 catalyst for CO2 hydrogenation to CH3OH preferentially exposing the h–In2O3(104) surface must be prepared by controlled synthesis due to its thermal instability. In contrast, a stable In2O3 catalyst mainly exposing the c–In2O3(111) surface may result in suboptimal selectivity for this reaction. To test the validity of our theoretical prediction, we prepared a series of In2O3 catalysts with different crystal phases, exposed facets, and studied their catalytic performance in CO2 hydrogenation to methanol. In particular, we strived to obtain and test In2O3 catalysts in the hexagonal phase mainly exposing the h–In2O3(104) surface.

Fig. 3 Schematic illustration of the most favorable CO2 hydrogenation pathways on different c–In2O3 and h–In2O3 surfaces.

We synthesized bulk In2O3 (sphere) by the conventional precipitation method. We also prepared three additional crystalline In2O3 nanoparticles (plate, lamellar, and rod) using the same hydrothermal method only with different ratios of water and ethanol for the solvents (table S1). X-ray diffraction (XRD), scanning electron microscopy (SEM), and high-resolution transmission electron microscopy (HRTEM) data in Fig. 4 and fig. S4 show that both spherical- and plate-shaped nanoparticles consists of a highly crystalline c–In2O3 structure with a = b = c = 10.118 Å [Joint Committee on Powder Diffraction Standards (JCPDS) 06-0416]. In contrast, for lamellar and rod forms of In2O3 (Fig. 4, E and F), the same analyses indicate that the crystal phase is h–In2O3 with a = b = 5.487 Å and c = 14.510 Å (JCPDS 22-0336) and no other phases are present (Fig. 4, A and B). TEM and HRTEM pictures showed spherical c–In2O3 nanoparticles (c–In2O3-S) of around 8 nm (Fig. 4C1), which form aggregates. Lattice fringes identified at multiple locations, yielded d-spacings of 0.292 and 0.178 nm, corresponding to the (222) and (440) planes of c–In2O3, respectively (Fig. 4C), which are multiples of the (111) and (110) planes. In addition, aberration-corrected scanning transmission electron microscopy (ac-STEM) studies on c–In2O3-S clearly show the {110} surface adjoining the {111} surface (Fig. 4C3). For other samples, the selected area electron diffraction (SAED) patterns (Fig. 4, D2 to F2, and fig. S4D2) demonstrate the single-crystalline structure of the c–In2O3 nanoplates (denoted as c–In2O3-P) and h–In2O3 nanolamellar (denoted as h–In2O3-L) and nanorod (denoted as h–In2O3-R) without the presence of twins or stacking faults. As shown in SEM and TEM images (Fig. 4 and fig. S4), the average thickness of c–In2O3-P is greater than that of h–In2O3-L (~36 nm versus ~20 nm), while the length of c–In2O3-P is smaller than the diameter of h–In2O3-L (~200 nm versus ~280 nm). According to the sixfold rotational symmetry observed in the SAED pattern of c–In2O3-P (Fig. 4D2 and fig. S4D2), the zone axis should be [111]. In addition, as observed from the HRTEM images, the main exposed planes of c–In2O3-P are {111}, which are the only planes normalized by the set of (440) planes with a lattice spacing of 0.178 nm and the set of (211) planes with a lattice spacing of 0.413 nm, and both the interfacial angles are 60° (Fig. 4D and fig. S4, D and F). The preferential exposure of the {111} facet in the c–In2O3 nanocatalysts is consistent with our DFT prediction that this facet is considerably more stable than the other facets (fig. S1), although the c–In2O3-S nanocatalyst also exposes {110} facet. For h–In2O3, we predict the (012) facet to be slightly more stable than the (104) facet (fig. S1). As shown in Fig. 4A, the (012)/(024) XRD peak intensities of h–In2O3-L are much higher than those of h–In2O3-R, and nearly no (012) peak can be observed for h–In2O3-R, while the rod structure exhibits a relatively higher (104) XRD peak intensity, which is the strongest peak of XRD patterns of h–In2O3-R. The HRTEM image of h–In2O3-L viewed along the [421] direction shows the (011¯2) and (112¯) planes with a lattice spacing of 0.397 nm and an angle of 87.2° (Fig. 4E). On the basis of these results and the corresponding SAED image, we identified the flat plane of the h–In2O3-L as {012}. Although the shape of nanolamellar is irregular, the surface area of flat planes is obviously larger than that of side faces, and thus, we can conclude that the major exposed planes of h–In2O3-L are {012} (fig. S4G). Representative SEM and TEM images of h–In2O3-R shown in Fig. 4F1 and fig. S4 revealed that the nanorods with the diameter within 10 to 50 nm aggregated to form cactus-like morphology. On the basis of the HRTEM images and the SAED patterns of h–In2O3-R, we observed that the flat planes (perpendicular to the observation direction) of nanorods are {104}. Combining with the XRD results, we can also speculate that the {104} facets are exposed on the two sides of the nanorod from the analysis of more than 10 randomly chosen particles (Fig. 4F and fig. S4E). Therefore, the h–In2O3-R nanorods are mainly rounded by {104} planes (fig. S4H). We further simulated the three-dimensional morphology of h–In2O3-R by assuming preferentially exposed {104} surfaces, and the simulated morphology (fig. S4I) is consistent with the experimentally derived one. Consequently, h–In2O3-L contains a higher proportion of the exposed {012} facets than h–In2O3-R, whereas h–In2O3-R has a higher fraction of the exposed {104} facets than h–In2O3-L. We can synthesize c–In2O3 and h–In2O3 nanoparticles with different morphologies and dominant exposed planes by the simple preparation method in the absence of any long chain capping ligands, the commonly used agents in the synthesis process to control the nanocrystals’ morphology, which can exclude the additional effects of residual impurities on the catalytic performance. Tuning the ratio of the solvents and hydrothermal time may change the regularity and aspect ratio of the nanorod, which may further increase the proportion of the exposed {104} facets.

Fig. 4 Structural characterization of various In2O3 catalysts.

(A) XRD patterns of In2O3 catalysts. a.u., arbitrary units. (B) Schematic description of cubic and hexagonal In2O3 models. (C1) HRTEM image of c–In2O3-S with insets showing the corresponding TEM image. (C2) SAED pattern of c–In2O3-S. (C3) STEM image of c–In2O3-S. (D1, E1, and F1) TEM images, (D2, E2, and F2) SAED patterns, and (D3, E3, and F3) HRTEM, and fast Fourier transform (FFT) images of (D1 to D3) c–In2O3-P, (E1 to E3) h–In2O3-L, and (F1 to F3) h–In2O3-R. The F3 is an enlarged view of a dotted area on the left in fig. S4E1. Insets in (D3) to (F3) are the FFT patterns of the high-resolution images. -S, -P, -L, and -R denote sphere, plate, lamellar, and rod, respectively.

We conducted the CO2 hydrogenation to methanol reaction over various In2O3 under the standard reaction conditions of 300°C, a total pressure of 5.0 MPa, 9000 ml gcat−1 hour−1, and H2/CO2/N2 = 73/24/3. Compared to c–In2O3-S, c–In2O3-P mainly exposing the {111} surface exhibits higher methanol selectivity and lower CO2 conversion (Fig. 5A). This is consistent with our DFT prediction that compared with c–In2O3(110), c–In2O3(111) is favorable for methanol formation but has much weaker CO2 adsorption strength in the bt–CO2* structure resulting in lower RWGS activity, so CO2 adsorption strength is another important factor in determining the catalytic activity besides the energy barrier of the RDS. For h–In2O3 catalysts, h–In2O3-R has simultaneously better activity and methanol selectivity in comparison to h–In2O3-L. In addition, we obtained a maximum methanol selectivity of 85.4% over h–In2O3-R predominantly exposing {104} surfaces. As shown in table S2, the CO2 conversion over h–In2O3-R is nearly twice higher than other bulk In2O3 and even higher than that over ZrO2 support-promoted In2O3 catalyst under the similar reaction conditions. From our DFT calculations, the defective h–In2O3(104) surface exhibits superior performance for both CO2 adsorption and CO2 hydrogenation to methanol with the lowest RWGS activity, which is consistent with our experimental results. Therefore, the crystal phase and exposed surfaces of In2O3 catalysts do play important roles in determining the catalytic performance for CO2 hydrogenation to methanol.

Fig. 5 Catalytic performance of various In2O3 for CO2 hydrogenation.

(A) CO2 conversion and methanol selectivity over In2O3 with different crystal phases and morphologies. Insert: Normalized activities for c–In2O3-S, c–In2O3-P, h–In2O3-L, and h–In2O3-R by specific surface area after pretreatment in pure Ar at 300°C for 44 hours. (B) Effect of reaction temperature over c–In2O3-S and h–In2O3-R samples. (C) Effect of reaction temperature on methanol yield over c–In2O3-S and h–In2O3-R samples. (D) Effect of H2/CO2 molar ratio over h–In2O3-R. (E) Stability test of h–In2O3-R. Standard reaction conditions: 300°C, 5.0 MPa, 9000 ml gcat−1 hour−1, H2/CO2/N2 = 73/24/3.

As shown in Fig. 5B, when the reaction temperature increases from 240° to 360°C, methanol selectivity decreases significantly from 95.9 to 17.8% over the c–In2O3-S catalyst, although CO2 conversion increases markedly. We observed similar trend of catalytic performance for h–In2O3-R. Nevertheless, methanol selectivity decreases only moderately from 97.9 to 73.4% when the temperature increases from 240° to 360°C, while the CO2 conversion increases sharply from 3.2 to 15.4%. The increase of reaction temperature favors the RWGS reaction, since it is an endothermic reaction. Our DFT calculations suggest that the defective h–In2O3(104) surface highly disfavors CO formation. Therefore, compared with the RWGS reaction, the methanol synthesis reaction over h–In2O3-R is more favorable even at the very high temperature of 360°C. The STY of methanol over c–In2O3-S reaches the maximum at 320°C, whereas the STY of methanol over h–In2O3-R increases linearly with the reaction temperature mainly due to the moderate decrease of methanol selectivity (Fig. 5C). At 360°C, the STY of methanol reached 10.9 mmol gcat−1 hour−1 over h–In2O3-R, twice higher than that over c–In2O3-S. The relative performance of these two catalysts agrees with our DFT predictions, further offering experimental evidence to support the strong promotional effect of the h–In2O3(104) facet on methanol synthesis from CO2 hydrogenation. Moreover, we found higher space velocity and H2/CO2 ratio to favor methanol formation. Both the selectivity and STY of methanol increase significantly with increasing space velocity or H2/CO2 ratio (fig. S5A and Fig. 5D). The STY of methanol reached above 10.4 mmol gcat−1 hour−1 at weight hourly space velocity of >20,000 ml g−1 hour−1, outperforming the other In2O3 (38), In2O3/ZrO2 (22), supported Cu (11), and noble metal (39) catalysts that are already excellent for CO2 hydrogenation to methanol (table S2). In addition, when increasing the ratio of H2/CO2 to 6:1, the STY of methanol reached 9.0 mmol gcat−1 hour−1 over h–In2O3-R with methanol selectivity of 92.4% and CO2 conversion of 17.6% at 300°C.

The h–In2O3-R catalyst displayed a good stability in a 136-hour test. Figure 5E shows that methanol selectivity increased markedly (from 71 to 85%), while CO2 conversion decreased slightly (from 9.3 to 7.5%) during the initial 44 hours of test. Nevertheless, there was no further deterioration in the CO2 conversion rate, and methanol selectivity maintained at 85 to 89% after 44 hours. It is very interesting that the STY of methanol kept at 6.2 mmol gcat−1 hour−1 for 136 hours on stream. As shown in fig. S6A, XRD analysis of the used catalysts showed that the average particle size of spent h–In2O3-R after 44 hours increased substantially from about 7 to 24 nm, whereas the In2O3 crystal size largely remained after the initial period (28 nm after 136 hours on stream). The formation of water vapor is inevitable for CO2 hydrogenation to methanol, which will dissociate and fill the oxygen vacancies and decrease the catalytic activity. However, H2O inhibition only moderately reduces the CO2 conversion and STY of methanol over h–In2O3-R by 20% and 18%, respectively, when the amount of cofeeding water increases from 0 to 3.2 mol percent, and we observed a similar trend over c–In2O3-S (fig. S5). Further increasing the amount of cofeeding water, a much bigger drop in CO2 conversion and STY of methanol occurred over c–In2O3-S than over h–In2O3-R, which is likely due to the lower H2O adsorption strength on defective h–In2O3(104) surfaces and higher thermal stability of h–In2O3-R nanoparticles (fig. S5C).

The Brunauer-Emmett-Teller (BET) surface area of c–In2O3-S nanoparticles is up to 112 m2 g−1, and the other samples exhibit relatively lower specific surface areas (table S3). As thermal treatment can lead to the oxygen vacancy of In2O3, we further pretreated the samples in argon (Ar) at 300°C for 1 hour before the CO2 hydrogenation reaction. The electron paramagnetic resonance (EPR) spectroscopy revealed a symmetrical signal at g = 2.004 over all fresh In2O3 samples, which corresponds to the oxygen vacancies, indicating that the vacancies were already present after the calcination (fig. S6B). After Ar thermal treatment, the increase in the intensity of the EPR signal suggests the greater density of oxygen vacancies. In addition, the BET surface area of c–In2O3-S dropped sharply from 112 to 90 m2 g−1, while we observed only a slight decrease for other In2O3 samples due to the higher thermal stability (table S3). Compared with c–In2O3-P and h–In2O3-L, c–In2O3-S and h–In2O3-R have much higher concentration of surface oxygen defects (Odefect, ~30% versus ~25%) estimated by measuring the number of O atoms next to the defects in O 1-s spectra by in situ near-ambient pressure x-ray photoelectron spectroscopy (NAP-XPS) at 300°C (Fig. 6A). This result is consistent with the higher activity obtained over the c–In2O3-S and h–In2O3-R samples for CO2 hydrogenation (Fig. 5A and fig. S5D).

Fig. 6 Characterization of oxygen vacancy sites and surface species.

(A) In situ NAP-XPS O 1s spectra of various In2O3 exposed to 50 Pa Ar at 300°C after 1 hour. (B) CO2-TPD spectra for the reduced samples (thermal treatment in Ar at 300°C for 1 hour). (C and D) Operando diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) spectra of surface species formed from the CO2 + H2 reaction over (C) c–In2O3-S and (D) h–In2O3-R. The sample first was exposed to Ar at for 1 hour, which was then switched to H2/CO2/N2 = 73/24/3 at 0.1 MPa with a gas flow rate of 20 ml min−1.

We further characterized the pretreated In2O3 oxides using CO2–temperature-programmed desorption (TPD). We observed two types of absorption peaks for all samples (Fig. 6B). The low temperature α peak (100° to 200°C) corresponds to physisorbed CO2, and the high temperature β peak (350° to 450°C) corresponds to the chemisorbed CO2 due to its reaction with thermally induced oxygen vacancy sites (22). Compared with c–In2O3 samples, the CO2 desorption temperature of the β peak for h–In2O3 is higher, indicating that CO2 chemisorption around the oxygen vacancy site of h–In2O3 is stronger than those of c–In2O3 surfaces, which is in line with our DFT results. Signals of the β peak for the h–In2O3 samples are more intense than those for c–In2O3, hinting at a greater number of thermal-induced oxygen vacancies. Moreover, we studied the reducibility of In2O3 by temperature-programmed reduction in hydrogen (H2-TPR). Presence of the reduction peak in hydrogen before the onset of bulk reduction indicates that oxygen vacancies on both c–In2O3 and h–In2O3 catalyst surfaces could also form upon exposure to the reducing agents (fig. S6C). The h–In2O3 materials show much higher H2 consumption in the low temperature region, suggesting that hexagonal In2O3 has more H2-induced oxygen vacancies. Although H2 reduction is favorable for the formation of oxygen vacancies, the BET surface area decreased significantly after the treatment of fresh In2O3 in pure H2 at 300°C for 1 hour (table S3). We found that both c–In2O3-S and h–In2O3-R catalysts pretreated in Ar yielded higher STY of methanol than those activated in H2 (fig. S5E), which may be due to the substantial drop in the surface area. Therefore, we activated the catalysts in Ar at 300°C before the CO2 hydrogenation reaction.

We carried out operando diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) to further understand the mechanism of CO2 activation and hydrogenation over partially reduced c–In2O3 and h–In2O3 samples. As shown in fig. S7A, we observed several infrared (IR) peaks at 900 to 1100 cm−1 after the adsorption of CO2 on the In2O3 sample activated in Ar at 300°C, which was due to the interaction of CO2 with the oxygen vacancies, as predicted by our DFT calculations (fig. S2). The adsorbed CO2 species over h–In2O3-R were relatively stable and could not be removed completely by Ar treatment even at 350°C, whereas nearly no bands between 900 and 1100 cm−1 were present over other In2O3 samples at 300°C (fig. S7A), indicating the stronger CO2 adsorption at the oxygen vacancies of h–In2O3-R, which is consistent with our CO2-TPD and DFT results. The introduction of H2 generates both formate (HCOO*) and methoxy (CH3O*) species over c–In2O3-S and h–In2O3-R even at the relatively low temperature of 240°C (Fig. 6, C and D). In addition, peak intensities of these species change slightly as the temperature increases from 240° to 300°C. These observations confirm the predicted reaction mechanism for the CO2 hydrogenation to methanol reaction on the In2O3 catalyst.

As our DFT calculations showed the HCOO* and CH3O* species to be the key intermediates during CO2 hydrogenation over In2O3 catalysts, we also investigated the interaction between these species and the catalyst surface by in situ DRIFTS of formic acid and methanol adsorption. With the direct introduction of the formic acid vapor, the IR intensities for νas(OCO) and νs(OCO) of HCOO* are much higher over c–In2O3-S than those over h–In2O3-R (fig. S7B). This trend remained under reaction conditions as shown in Fig. 6 (C and D). In addition, we observed the notable IR peaks for CH3O* upon the adsorption of methanol over c–In2O3-S and h–In2O3-R samples (fig. S7C). Previous studies found that the dehydrogenation of CH3O* to HCOO* could occur over Cu/ZrO2 at room temperature (40). In our case, the dehydrogenation of CH3O* to HCOO* obviously occurred over c–In2O3-S at 30°C, but the peaks attributed to HCOO* were much weaker over h–In2O3-R, indicating the stronger stability of CH3O* as opposed to HCOO* over the defective h–In2O3(104) surface (fig. S7C). Therefore, the defective h–In2O3(104) surface can substantially stabilize the key intermediates involved in methanol formation, which gives the higher methanol selectivity in catalytic CO2 hydrogenation compared with cubic In2O3.

Our combined computational and experimental studies have demonstrated the structure sensitivity of the In2O3 catalyst during CO2 hydrogenation in terms of both the phase and the exposed facet. In addition to the development of a promising high-performance In2O3 material for CO2 hydrogenation to methanol, the present work also opens new avenues toward computer-aided rational design of efficient oxide catalysts for processes beyond methanol synthesis.


DFT calculations

Periodic DFT calculations were carried out with the Vienna Ab Initio Simulation Package using the Perdew-Burke-Ernzerhof exchange–correlation functional and the projector–augmented wave potentials. The semicore 4d electrons of In were treated as valence electrons, with an energy cutoff of 400 eV and a Gaussian smearing width of 0.05 eV. The convergences for the electronic energy of the supercell and the force on all unconstrained atoms are 10−4 eV and 0.03 eV Å−1, respectively.

We first optimized the primitive unit cells (PUC) of c–In2O3 and h–In2O3. The optimized h–In2O3 PUC have lattice parameters of a = b = 5.55 Å and c = 14.18 Å, consistent with the experimental data (a = b = 5.44 Å and c = 14.18 Å) (27). The c–In2O3(111) surface model is a p(1 × 1) slab consisting of 48 O atoms and 32 In atoms distributed in two O—In—O trilayers optimized using a Γ-centered (3 × 3 × 1) Monkhorst-Pack k-point mesh. The supercell has a dimension of 14.44 Å by 14.44 Å by 15.04 Å. The model of the c–In2O3(110) surface is a p(1 × √2) slab also with 48 O atoms and 32 In atoms but distributed in four atomic layers optimized using a Γ-centered (4 × 3 × 1) Monkhorst-Pack k-point mesh. The supercell has a dimension of 10.21 Å by 14.44 Å by 15.99 Å. The h–In2O3(012) surface model is a p(2 × 2) slab consisting of 48 O atoms and 32 In atoms distributed in four atomic layers optimized using a Γ-centered (2 × 2 × 1) Monkhorst-Pack k-point mesh. The supercell has a dimension of 11.04 Å by 11.68 Å by 16.39 Å. The h–In2O3(104) surface model is a p(1 × 2) slab consisting of 48 O atoms and 32 In atoms distributed in four atomic layers optimized using a Γ-centered (3 × 2 × 1) Monkhorst-Pack k-point mesh. The supercell has a dimension of 8.04 Å by 11.09 Å by 21.62 Å. The vacuum layer thickness is 10 Å between adjacent slabs.

An oxygen vacancy on the defective In2O3 surface forms upon removing one oxygen atom from the perfect In2O3 surface. The formation energy of an oxygen vacancy is the reaction energy of the thermal desorption of molecular oxygen (In2O3_P → In2O3_D + 1/2O2), where In2O3_P and In2O3_D denote the perfect and defective surfaces. The adsorption energy of an adsorbate A on a slab surface S was defined as Ead(A) = EA/S – (ES + EA), where EA/S, ES, and EA are total energies of the In2O3 slab with the adsorbate, the clean In2O3 slab, and the adsorbate as a free molecule, respectively. Transition states were obtained using the climbing image nudged elastic band method and were confirmed by further frequency calculations showing one and only one imaginary frequency. All structures were built and visualized using Materials Visualizer from Materials Studio.

Catalyst preparation

The spherical In2O3 was prepared by a precipitation method at 25°C. Typically, 16.26 g of In(NO3)3•4.5H2O was dissolved in a mixture of 48 ml of deionized water and 140 ml of ethanol and a mixture of 36 ml of NH4OH (25 weight % in H2O), and 108 ml of ethanol was used as the precipitant. The product was aged at 80°C for 10 min and then filtered and washed with deionized water until to the pH is 7. The filter cakes were dried overnight at 60°C and calcined in air at 300°C for 5 hours to get the oxides named as the c–In2O3-S. The c–In2O3-P, h–In2O3-L, and h–In2O3-R samples were prepared by the hydrothermal method. First, In(NO3)3•4.5H2O was dissolved in a mixture of deionized water or ethanol, followed by addition of urea solution (urea dissolving into deionized water or ethanol) under vigorous stirring, and the mixture was stirred for 2 hours at room temperature. Then, the aqueous solution was transferred into a Teflon-lined autoclave and crystallized at 120°C for 17 hours. After centrifuging and washing by deionized water, the resulting product was dried overnight at 60°C and then calcined in air at 300°C for 5 hours to get the oxides. Changing the molar ratio of the deionized water and ethanol resulted in synthesizing In2O3 with different crystal phases and morphology structures. Synthesis parameters and corresponding nominations are listed in table S1.

Catalyst characterization

Powder XRD was analyzed in the 2θ range 5° to 90° using a Rigatku Ultima 4 x-ray diffractometer with Cu Ka radiation, operating at 40 kV and 40 mA and in the step mode (0.0167°). Raman spectroscopy was performed using the Thermo Scientific DXR Raman microscope comprising a source with He-Ne laser of 532 nm. The textural properties such as surface area (BET) were derived from N2 adsorption-desorption measurements by using a TriStar II 3020 instrument at −196°C, following the evacuation of samples in vacuum at 200°C for 10 hours. EPR measurements of the free radicals were recorded by using 5,5-dimethyl-1-pyrroline (>99.0%) as a probe at a Bruker EMS-plus instrument (Bruker A300). The morphology of the samples was observed by a SUPRRATM 55 SEM with an accelerating voltage of 2.0 kV. The nanostructure of catalysts was investigated by an ac-STEM instrument (JEM-ARM300F) at 300 kV and a FEI Tecnai G2 F20 S-Twin HRTEM, which was operated at 200 kV. CO2-TPD experiments were carried out with an OmniStar GSD320 02 mass spectrometer. First, the catalyst (100 mg) was treated at 300°C for 60 min in a flow of pure Ar (60 ml min−1) and then cooled down to 50°C. After that, the catalyst was saturated in flowing CO2 for 1 hour with 30 ml min−1, followed by flushing in Ar for 3 hours to remove any physisorbed molecules. The CO2-TPD measurement was carried out at 50° to 750°C with heating rate of 10°C min−1 under continuous flow of Ar at 40 ml min−1. The H2-TPR was carried out on a Micromeritics ChemiSorb 2920 with a thermal conductivity detector. Typically, the catalyst sample (54 mg) was placed in a quartz reactor and pretreated in flowing Ar of 60 ml min−1 at 150°C for 1 hour, followed by cooling down to 50°C. Then, the temperature was raised to 700°C at a rate of 5°C min−1 with 5% H2/Ar mixture gas (30 ml min−1). In situ NAP-XPS was carried out on a SPECS Surface Nano Analysis GmbH equipped with two chambers, including an analysis chamber and a quick ample load-lock chamber. The analysis chamber is composed of a PHOIBOS NAP hemispherical electron energy analyzer, a microfocus monochromatized Al Kα x-ray source with a beam size of 300 μm, a SPECS IQE-11A ion gun, and an IR laser heater. The samples were treated in Ar at 573 K for 1 hour, and then the spectra were collected in the analysis chamber. The in situ DRIFTS measurements were recorded on a Nicolet 6700 instrument equipped with a liquid N2-cooled mercury-cadmium-telluride detector including a cell having a cylindrical cavity (5 mm in diameter and 5 mm in vertical length) for the sample placement. About 20 mg of catalyst powder was placed in the cell and pretreated at 300°C for 1 hour under continuous flow of Ar with 30 ml min−1 and then cooled to 50°C. Of course, the background spectrum at specific temperature was collected during the cooling process. After that, the catalyst was saturated in flowing CO2 of 20 ml min−1 for 1 hour and followed by flushing in Ar for 20 min to remove any physisorbed molecules. Then, the temperature was increased from 50° to 350°C with the Ar (20 ml min−1). For the operando DRIFTS experiments under the reactant gas mixture with a H2/CO2/N2 ratio of 73/24/3 at 0.1 MPa, the samples were also pretreated at 300°C for 1 hour under pure Ar with 20 ml min−1, and then the reactant gas was introduced into the cell. The temperature was increased from 240° to 300°C, and every specific temperature was maintained for 30 min. For in situ DRIFTS experiments of formic acid and methanol adsorption, the catalysts were pretreated at 300°C with pure Ar (20 ml min−1) for 1 hour and then cooled down to 30°C. Then, the vapor of methanol or formic acid was introduced by Ar bubbling through the corresponding high pure liquid reagents for 8 min and then changed to introduce pure Ar.

Catalytic evaluation

The catalytic performance was carried out in a continuous-flow, high-pressure, fixed-bed reactor (dint. = 12 mm). In2O3 oxide (1.0 g) (40 to 60 mesh) with quartz sand mixing in equal volume was placed in a stainless-steel tube reactor. Before reaction, the sample was pretreated at 300°C for 1 hour in pure Ar (150 ml min−1), and then reactant gas mixture with a H2/CO2/N2 ratio of 73/24/3 was introduced into the reactor under a pressure of 5.0 MPa. The effluents were analyzed online with a Shimadzu GC-2010C gas chromatograph equipped with thermal conductivity and flame ionization detectors. The CO2 conversion was calculated by an internal normalization method. The catalytic performance after 44 hours of reaction was typically used for discussion. The CO2 conversion denoted as X(CO2), CH3OH selectivity denoted as S(CH3OH) and STY of methanol denoted as STY(CO2) were calculated according to the following equationsX(CO2)=CO2inletCO2outletCO2inlet×100%S(CH3OH)=CH3OHoutletCO2inletCO2outlet×100%STY(CH3OH)=GHSV×0.241000×22.4×X(CO2)S(CH3OH)100×100×1000=GHSV×0.2422.4×X(CO2)S(CH3OH)10000


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank Y. Ma, Y. Zhou, and B. Lv for the discussion on HRTEM and SAED results. We also thank Z. Liu for the assistance with in situ NAP-XPS. Funding: This work was financially supported by the “Transformational Technologies for Clean Energy and Demonstration,” Strategic Priority Research Program of the Chinese Academy of Sciences (XDA21090204 and XDA21090201), the National Natural Science Foundation of China (21773286, U1832162, 21573148, 11227902, and 21802096), Youth Innovation Promotion Association CAS (2018330), Shanghai Rising-Star Program, China (19QA1409900), the Ministry of Science and Technology of China (2018YFB0604700 and 2016YFA0202802), and the Chinese Academy of Sciences (ZDRW-ZS-2018-1-3). Author contributions: S.D., B.Q., S.L., P.G., and Y.S. conceived the project, analyzed the data, and wrote the paper. S.D., B.Q., S.L., and P.G. drafted the manuscript. B.Q. and S.L. performed DFT calculations. S.D. prepared the samples. S.D. performed the catalytic evaluation. S.D., H.W., J.C., and Y.H. characterized the samples. All authors discussed the results and commented on the manuscript. Competing interests: S.D., P.G., and Y.S. submitted a patent application (Chinese patent application filed on 12 September 2019) titled “Indium oxide catalyst and preparation method and application of the catalyst” and application number 2019108669473. All authors declare that they have no other competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article