CO₂ emissions are a major environmental problem requiring immediate technological solutions [1, 2]. One option, as a practical solution for substantial CO₂ conversion, is the catalytic hydrogenation of CO₂ as a preferential pathway for CO₂ illumination [3], where conversion of CO₂ to CO is an initial key step. Fischer-Tropsch (FT) synthesis and oxo-synthesis reactions can convert CO₂ into a variety of valuable hydrocarbons [4, 5]. However, the continuous conversion of CO₂ to yield a single final product (CO) still remains a great challenge. There is a wealth of scientific research currently focusing on CO₂ hydrogenation via thermal and electrochemical catalysis [6, 7], by designing highly selective, active and stable catalysts. Numerous efforts have been devoted to further understand the CO₂ hydrogenation reaction; however, the elucidation of detailed reaction mechanisms is still under debate [8-12].

Supported metal catalysts, especially mixed-oxide metal catalysts, are a highly desirable research field in heterogeneous thermal catalysis owing to their outstanding selectivity to specific products [13-15]. The activation of CO₂ and H₂ is a prerequisite for CO₂ hydrogenation. Researchers have observed that H₂ usually dissociates on metal sites while CO₂ is activated on the oxide support [16]. Only catalysts with such dual "reactive activating" functionalities can effectively catalyze the hydrogenation of CO₂ to CO [17]. A great number of mechanisms have been proposed for the CO₂ hydrogenation reaction, which can be divided into two categories depending on whether or not the CO intermediate is produced: (1) CO₂ dissociates into CO with/without the presence of formate intermediates followed by a CO methanation reaction [18], and (2) CO₂ is directly hydrogenated to bicarbonates, formates, methoxides, and finally methane in the absence of CO intermediates [19]. However, the mechanism is greatly influenced by the nature of the catalyst.

Li et al. [15] recently studied the influence of Ir/CeO₂ particle size on the CO₂ hydrogenation reaction. They observed that smaller Ir particles induced a stronger metal-support interaction, which significantly impacted the adsorption and selectivity of CO. Qiao et al. [20] detected through Bader charge analysis that single Pt atoms on Pt/FeO_x catalysts carried considerable positive charges, which resulted in a weak CO adsorption, and concomitantly, improvements to anti-poising performance for the CO oxidation reaction. All of the above works are devoted to developing well-designed catalysts to explore the adsorption behavior of the CO reactant/intermediate, which can elucidate, to some degree, the reaction path and final selectivity. Previous studies have reported that Ru-based noble metal catalysts are typically more likely to promote the formation of CH₄ in the CO₂ reduction reaction [21, 22]. Xu et al. [23] explored the influence of pretreatment temperature on Ru/TiO₂ catalysts in CO₂ methanation and suggested that the dissociation of adsorbed CO (CO_ads) proceeds via a process—assisted by adsorbed H (H_ads) species—that results in CO_ads to be further reduced to methane. At the same time, Wang et al. [24] observed the same CO₂ methanation mechanism in the presence of Ru/Al₂O₃ catalysts, and furthermore, found that the reaction between H_ads and CO_ads, rather than other steps, is the rate-determining step during the CO₂ methanation reaction. Regardless, controlling the production of CO from CO₂, in the absence of CH₄, is a prerequisite for FT reactions. The catalyst design rationale, by simply tuning the synthesis parameters, is essential when developing efficient catalysts. Although it is well known that catalytic properties are well influenced by metal loading, insights into how the loadings affect the active metal site properties, and importantly the reaction process, are lacking.

In this work, highly dispersed Ru composites were deposited onto FeO_x supports. Analysis of various characterization methods has elucidated the unique structure-property relationship of the Ru/FeO_x catalysts, which obtain 100% CO selectivity. The crystallinity, morphology, as well as the chemical states and properties were studied by X-ray diffraction (XRD), X-ray photoelectron spectroscopy (XPS), transmission electron microscopy (TEM), and temperature-programmed experiments. The interaction between the Ru species and the support, the different H₂ activation and dissociation/desorption behaviors, and how these factors influence product selectivity were studied in detail as a function of Ru surface loading.

All chemicals used in this experiment were of analytical grade and used without further purification. RuCl₃·xH₂O was purchased from Aladdin. Other chemicals including NaOH and Fe(NO₃)₃·9H₂O were purchased from Chengdu Kelong Chemical Co., Ltd., China, and deionized water was used in all the experimental processes. All catalysts were prepared via co-precipitation methods [25]. A mixed solution of RuCl₃·xH₂O and Fe(NO₃)₃·9H₂O in desired proportions was added dropwise into 0.4 mol L^-1 NaOH at 80 ℃ with vigorous stirring. After the co-precipitation, the slurry was maintained at pH = 8.0 by NaOH. The slurry was stirred continuously at 80 ℃ for 3 h, and then aged at the same temperature for 1 h. The precipitate was washed with deionized water before drying at 60 ℃ for 15 h without any further heat treatment. The catalyst was denoted as m%Ru/FeO_x with m% as the Ru loading. At the same time, a reference support, denoted as FeO_x, was synthesized by the same method only in the absence of Ru.

XRD diffractograms were recorded on a DX-2700 diffractometer with Cu K_α radiation (40 kV, 30 mA). The 2θ scanning range was 10°-90° with a scan step of 0.03° s^-1. Specific surface area, total pore volume and average pore diameter of the samples were measured from N₂ adsorption-desorption isotherms at -196 ℃ (Quantachrome Nova 1000e apparatus). The specific surface area was calculated by the Brunauer-Emmett-Teller (BET) equation, and the pore size distribution and average pore diameter were determined from the desorption branch of the isotherms using the Barrett-Joyner-Halenda (BJH) model. The m%Ru/FeO_x samples were outgassed to 0.1 Pa at 110 ℃ to ensure structural integrity of the support. TEM was performed using an FEI Tecnai G2 microscope equipped with high angle annular dark field (HAADF) and energy-dispersive X-ray (EDX) analysis detectors at 200 keV. Temperature-programmed reduction with hydrogen (H₂-TPR) was performed using a Tian Jin TP-5080 catalyst characterization system. First, 50 mg of catalyst was loaded into a fixed-bed reactor and purged with N₂ at 120 ℃ for 2 h to remove absorbed carbonates and hydrates. After cooling to 30 ℃, the purge gas was switched to a 10 vol% H₂/N₂ reduction gas composition at a flow rate of 30 mL min^-1. The temperature of the catalyst was raised from 30 to 900 ℃ at a rate of 10 ℃ min^-1. Hydrogen temperature-programmed desorption (H₂-TPD) was performed on a Tian Jin TP-5080 catalyst characterization system. 200 mg of catalyst was first pretreated at 200 ℃ under a hydrogen flow (20 mL min^-1) for 45 min at a heating rate of 10 ℃ min^-1. After cooling to 30 ℃, the sample was purged with N₂ for 90 min to remove weakly adsorbed H₂. Thereafter, the sample was heated from 30 to 500 ℃ at a ramp rate of 10 ℃ min^-1 with the desorbed H₂ detected on-line. In situ temperature-programmed surface reaction (TPSR) experiments for the CO₂ hydrogenation reaction were analyzed on-line by a British Hiden QIC-20 mass spectrometer. Prior to performing the TPSR experiments, the catalyst was placed in a fixed-bed quartz reactor having an internal diameter of 6 mm. The reactor was heated within a tube furnace equipped with a temperature controller, and all gases were monitored by calibrated mass flow controllers. Prior to the experiment test, 200 mg of the catalyst was first pretreated under a 20 mL min^-1 hydrogen flow for 2 h at 200 ℃ at a heating rate of 5 ℃ min^-1. Upon cooling to 100 ℃ under flowing Ar, a mixture of H₂, CO₂ and Ar, having a molar composition of: 3 H₂ : 3 CO₂ : 40 Ar was introduced into the reactor having a total gas flow of 46 mL min^-1. The gas hourly space velocity (GHSV) was maintained at 1800 mL g^-1 h^-1 under atmospheric pressure. After the baseline was stable, the temperature of the reactor was raised to 500 ℃ at a uniform rate of 10 ℃ min^-1. The gas composition of the reverse water gas shift (RWGS) reaction was analyzed on-line by a QIC-20 mass spectrometer during the heating process.

Conventional catalytic evaluation of the RWGS reaction was performed under atmospheric pressure on a micro fixed-bed tubular quartz reactor having an internal diameter of 6 mm. Typically, 300 mg of catalyst was loaded and reduced in situ with pure H₂ at 200 ℃ for 1 h at a gas flow of 30 mL min^-1. The catalyst was subjected to an operating temperature of 240 ℃ and purged with the reaction gas having a molar composition of: 1 H₂ : 1 CO₂ with a GHSV of 1200 mL g^-1 h^-1. The effluent gases from the reactor were analyzed on-line using a SP-7890 model gas chromatograph equipped with a TDX-01 column and a thermal conductivity detector (TCD), using Ar as the carrier gas. Activity was evaluated every 10 ℃ over the temperature range of 240-360 ℃. Catalytic performance was evaluated by the conversion of CO₂ based on the concentration difference between the inlet and outlet, which is defined as:

X(CO₂) = (F(CO₂, in) -F(CO₂, out))/F(CO₂, in) × 100%

CO and CH₄ selectivity is defined as follows:

S(CO) = F(CO, out)/(F(CO₂, in) -F(CO₂, out)) × 100%

        S(CH₄) = (1 -S(CO)) × 100%

where F is the flow rate of each component in the feed or effluent.

Phase identification of the catalysts was analyzed by XRD, as shown in Fig. 1. It can be observed that the Fe₂O₃ structure is the major phase for the 0.01%Ru/FeO_x and 2.50%Ru/FeO_x samples.

The freshly prepared Ru/FeO_x samples do not show any diffraction peaks corresponding to crystalline Ru phases in the 0.01% or 0.25%Ru/FeO_x samples, indicating the presence of highly dispersed Ru species or very low Ru content beyond the detection limit of the diffractometer. However, there is also no observable diffraction peaks related to Ru in the 2.50%Ru/FeO_x sample, which suggests finely dispersed Ru species in this sample. Conversely, XRD analysis of the FeO_x sample shows a phase transition into Fe₃O₄ by thermal treatment under a N₂ atmosphere at 100 ℃, which is consistent with the research of Lin et al. [12], based on Ir/FeO_x catalysts.

The textural properties of the catalysts are summarized in Table 1. The pore structure parameters are generally maintained after Ru addition, although the surface area, surface pore volume and the average diameter slightly decrease as a function of incremental Ru loading. It is suggested that the integrity of the mesoporous structures across the different samples are maintained after Ru loading.

TEM and EDX elemental mapping of the 0.25%Ru/FeO_x and 2.50%Ru/FeO_x catalysts after reduction (at 200 ℃ by H₂) are illustrated in Fig. 2. Even at the higher loading of 2.50% Ru, TEM observations show no visible sign of Ru nanoparticles in the 2.50%Ru/FeO_x sample. This also suggests that Ru exists as highly dispersed species across all of the FeO_x supported samples, which is consistent with the XRD results. EDX elemental mappings of the Ru-supported samples also suggest highly dispersed Ru species supported on the surface of the iron oxide regardless of the Ru loading. However, STEM-EDX analysis shows no discernible Ru peaks for the 0.01%Ru/FeO_x sample as a result of the low Ru concentration.

For CO₂ hydrogenation reactions that are catalyzed by oxide-supported metals, it is generally accepted that H₂ molecules are dissociated on the metallic component (i.e. Ru), while CO₂ can be activated on the oxygen vacancies of the oxides (i.e. FeO_x) [26, 27]. Herein, only catalysts possessing dual functionality are able to enhance the CO and CH₄ selectivity. As the dissociation of the CO₂ step proceeds with a higher activation barrier [28], and more oxygen vacancies become available, the more favorable the overall CO₂ hydrogenation reaction rate [29]. In this regard, the capacity of the catalyst to generate oxygen vacancies are further elucidated by H₂-temperature-programmed experiments.

As shown in Fig. 3(a), it can be observed that the pure FeO_x support displays one small peak at ~330 ℃ with a corresponding shoulder attributed to the reduction of various surface iron oxide species. The overlapped reduction peaks at higher temperatures (500-900 ℃) are ascribed to the sequential reduction of Fe₂O₃ to metallic Fe via FeO in the bulk phase of the catalyst. Compared with the pure FeO_x support, the addition of Ru leads to a considerable and progressive reduction of the temperature at which chemical reduction occurs, as a function of increased Ru loading. The addition of Ru species results in the ascribed reduction peaks to shift to lower temperatures of 100-200 ℃, except for 0.01%Ru/FeO_x. Additionally, the reduction peaks attributed to the surface iron oxide also shift to lower temperatures. Two overlapped reduction peaks (190 and 227 ℃) with a corresponding shoulder are observed in 0.10%Ru/FeO_x. While the 0.25% and 2.50%Ru/FeO_x catalysts display two separate reduction peaks in the range of 100-300 ℃. It is suggested that the first reduction peak in the range of 100-200 ℃ is related to the reduction of positively charged Ru species to Ru⁰, in addition to surface iron oxide in the vicinity of the Ru species, which can be easily reduced because of the Ru-FeO_x synergistic interaction. Such reduction peaks shift to lower temperatures with increased Ru loading. Conversely, the reduction peaks at higher temperatures also shift from 870 ℃ on bare FeO_x to 755 ℃ on the 2.50%Ru/FeO_x catalyst. Furthermore, the shifting of both the Ru and FeO_x reduction peaks to lower temperatures across all Ru/FeO_x samples suggest the existence of a synergistic interaction between Ru and iron oxide, which benefits the co-reduction of the different species in addition to enhancing the redox properties. The uncorrelated H₂ consumption ratio between 0.10% of RuO_x and 99.9% of FeO_x is suggested to be attributed to the fact that the FeO_x support was unable to be fully reduced during the reduction process, since bulk FeO_x existed as the main phase. Similarly, only the reduction of the positively charged Ru species can yield the corresponding H₂ consumption peaks. The relatively high Ru loading amount does not equate to the observed higher levels of effective reducible components that are actually involved in the hydrogen consumption, that in turn would proportionally increase the numbers of active reaction sites. The low temperature peaks present in 2.50%Ru/FeO_x and 0.25%Ru/FeO_x are quite similar, signifying no significant increase in either the effective reactive sites or the reductive components in the 2.50%Ru/FeO_x catalyst.

H₂-TPD experiments of selected catalysts were also performed to elucidate the activation and adsorption behaviors of H₂ as shown in Fig. 3(b). The FeO_x support displays very broad H₂ desorption peaks having a maxima at 280 ℃, reflecting the certain ability of the support to adsorb and activate H₂. The addition of only 0.01% Ru species displays apparent H₂ desorption peaks at 112 and 408 ℃, fully demonstrating the significant contribution of even only a small Ru loading to activate H₂. Such phenomenon may relate to the H₂ spillover effect [30, 31] on noble metals (such as Ru), where H₂ is activated on Ru and dissociated into atomic hydrogen having a stronger reductive ability, which can further diffuse through the metal-support interface to the FeO_x support. Since the reduction ability of atomic hydrogen is much stronger than that of molecular hydrogen, the reduction of iron oxide will be greatly facilitated at significantly lower temperatures in the presence of any hydrogen spillover [25, 32]. With increasing Ru loading, the lower temperature desorption peak associated with the 2.50%Ru/FeO_x catalyst does not shift significantly (106 ℃). However, the peak area (at 106 ℃) is significantly less than that of the 0.01%Ru/FeO_x catalyst, despite the significantly increased Ru loading. It can be observed that both the 0.01%Ru/FeO_x and 2.50% Ru/FeO_x catalysts are highly efficient in H₂ activation at low temperature. It is also remarkable that the 0.01%Ru/FeO_x sample has more hydrogen spillover sites on the FeO_x surface than the 2.50%Ru/FeO_x catalyst based on its easier desorption at 106 ℃ [33].

Conversely, the H₂ desorption peak associated with the FeO_x support shifts from 408 ℃ in 0.01%Ru/FeO_x to a much lower temperature range of 298 ℃ in the 2.50%Ru/FeO_x sample, which is thought to be attributed to the strong interaction between Ru and H_ads [34, 35]. Such strong interactions will prevent the activated H_ads from facile removal. Hence, the reaction energy barrier is greatly enhanced [36] and the reaction between H_ads and CO_ads will be inhibited, which is further evidenced by both the TPSR data and the catalytic test data. Conversely, the 2.50%Ru/FeO_x catalyst with moderate hydrogen desorption behavior is relatively more conducive in promoting further hydrogenation of CO_ads to CH₄.

The H₂-TPD results clearly demonstrate that the addition of Ru to the catalysts significantly promotes H₂ activation spillover. The 0.01%Ru/FeO_x catalyst having an extremely low Ru loading moderately activates H₂. However, the distinctive Ru-support interactions arise as a function of Ru loading, which results in the observed variation to effectively activate H_ads, and its subsequent reaction with CO_ads, thus promoting alternative reaction paths that yield different products (CO or CH₄).

The XPS spectra in the Ru 3d_5/2 region and Fe 2p_1/2 region of the catalysts are presented in Fig. 4 [37]. The XPS data clearly shows differences between the spectra as a function of Ru loading. There is no observable peak related to any Ru-based species when the Ru loading is only 0.01%. Further increasing the Ru loading to 0.25%, a weak signal attributed to Ru 3d_5/2, begins to appear, whereas, a clearly observable peak is present when increasing the Ru loading to 2.50%.

Two overlapping peaks with an associated satellite peak can be observed in the XPS spectra of the Fe 2p_1/2 region [38]. The Fe2p_1/2 region can be fitted by several deconvoluted peaks. The symmetry peaks at 710.2 and 713.0 eV are related to Fe²⁺ and Fe³⁺, respectively [39]. The Fe²⁺ and Fe³⁺ species exist as the major compositions across all the Ru/FeO_x samples. The surface atomic ratio between Fe²⁺ and Fe³⁺ is ~1.4 for all samples, and the same degree of variation is largely dependent on the specific Ru loading in each sample. The surface atomic ratios are: 1.39, 1.45, 1.51, and 1.54 with increased Ru loading, respectively. It can be observed that degree of Ru loading leads to a slight variation of the iron oxide surface state. As previously reported, the electron transfer originates from the precious metal to the oxide support [15, 20], and the variation of the surface iron species may result from the electron transfer from Ru to FeO_x.

A one-step TPSR experiment of CO₂ hydrogenation (Fig. 5) was designed to elucidate the reaction mechanism in detail. A CO₂/H₂/Ar (46 mL min^-1, 3/3/40 v/v/v) flow was introduced into the reactor at 100 ℃ prior to heating to 500 ℃ at a rate of 10 ℃ min^-1 before maintaining for 30 min. There are no significant CO₂ consumption peaks evident across all samples (Fig. 5(a)). The consumption peaks attributed to the H₂ reactant are given at ~360 and 110 ℃ for FeO_x and the 2.50%Ru/FeO_x catalyst, respectively, which show a gradual shift to lower temperatures with increased Ru loading (Fig. 5(b)). It is worth noting that the generation of the CO signal across all samples lags behind the H₂ consumption peak. The H₂-TPR and H₂-TPD results infer that H₂ is activated on the Ru species and dissociated into atomic H, which effectively promotes the reduction of the FeO_x support, resulting in rich oxygen vacancies. Thereafter, CO₂ is activated on the oxygen vacancy of the FeO_x support to form the CO_ads species [40]. Varying the Ru loading modifies the surface coverage of Ru on the support, leading to differences in the adsorption strength between H_ads and the Ru species. This directly impacts H_ads adsorption, and thus the reaction with CO_ads, causing the production of CO or CH₄ via excessive reduction. As a result, a clear CH₄ peak at ~350 ℃ (Fig. 5(c)) over the 2.50%Ru/FeO_x sample is observed, which is absent in the samples with lower Ru content.

Prior to catalyst testing, the samples were reduced with pure H₂ at 200 ℃ for 2 h. The catalytic activities of the Ru/FeO_x catalysts for CO₂ reduction are shown in Fig. 6(a). All catalysts show increased CO₂ conversion as a function of increased reaction temperature (260-360 ℃). The pure FeO_x support also displays initial catalytic activity. The CO₂ conversion of 0.10%Ru/FeO_x is 14.5% at 360 ℃, which is similar to that of the 0.25% and 2.50%Ru/FeO_x catalysts (12.9% and 15.6%). However, when comparing the normalized specific rate of 0.01%Ru/FeO_x (7.71 mol_CO2 mol_Ru^-1 min^-1) with the other catalysts (Table 2), the specific rate is ~154 times that of 2.50%Ru/FeO_x (0.05 mol_CO2 mol_Ru^-1 min^-1).

Conversely, the 2.50%Ru/FeO_x catalyst displays significantly higher selectivity toward CH₄ (82%). For the 0.10%Ru/FeO_x and 0.25%Ru/FeO_x samples, the capacity to selectively produce CO is enhanced, while the selectivity toward CH₄ is reduced to 20%. It is demonstrated that the hydrogenation ability of the catalysts is influenced by reducing the Ru loading, which results in the facile formation of CO—the primary hydrogenation product. Further reducing the Ru loading to 0.01% results in 100% selectivity toward CO. The results indicate that CO₂ hydrogenation selectivity can be tuned by varying the Ru loading. There is no obvious variation or decrease in CO selectivity over the 0.01%Ru/FeO_x catalyst as a function of reaction temperature. The other catalysts display a gradual increase in CO selectivity up to 320 ℃ before finally plateauing. The CO selectivity of the 2.50%Ru/FeO_x sample increases from 21.4% at 260 ℃ to 41.0% at 360 ℃, while the CO selectivity of the other samples is > 92% (Fig. 6 and Table 2).

Summarizing the results shows that the main factors influencing CO₂ conversion and CO production result from the Ru loading, which is related to the Ru coverage over the FeO_x surface, as shown in Fig. 6(d1) and (d2). The catalyst samples display varying degrees of electron transfer from active Ru species to the FeO_x support and the chemical properties, including the Ru-FeO_x synergistic interaction, the Ru-H_ads interaction and the redox properties, depend specifically on the Ru surface coverage. For 2.50%Ru/FeO_x with a higher Ru loading, the interaction between Ru and H_ads is proved to be weak, which results in facile hydrogen desorption and excessive reduction by further reacting with CO or CO₂ to yield CH₄ as the major product. While 0.01%Ru/FeO_x with a lower Ru content displays a stronger Ru-H_ads interaction and hinders H_ads from being easily desorbed. Here, CO selectivity is 100% and the only product formed as the formation of CH₄ is inhibited.

Finely dispersed Ru species were synthesized on FeO_x supports as a function of Ru loading and evaluated for the selective hydrogenation of CO₂. The results suggested that both CO₂ conversion and CO selectively were largely promoted by small Ru loading amounts compared with the pure FeO_x support. While the CO selectivity can be further controlled by adjusting the Ru loading. Regardless of the ability for the catalysts to facilitate H₂ activation, it is the variation in Ru loading that results in differences in the Ru-H_ads interactions, which in turn accounted for the distinctive H_ads desorption behavior. Although CO₂ conversion displays a volcanic trend as a function of Ru loading, with a maxima at 0.1% Ru loading, the 0.01%Ru/FeO_x catalyst displayed 100% selectivity toward CO because of the hard desorption of H_ads with excessive reduction of CO₂ to CH₄ being hindered. The normalized specific rate of 0.01%Ru/FeO_x (7.71 mol_CO2 mol_Ru^-1 min^-1) is ~154 times that of the 2.50%Ru/FeO_x catalyst.

Acknowledgments

The authors gratefully acknowledge Yanan Zhou, Luming Li, Xiaoping Gao, Xionghua Fan and Shuxiong Sun at Sichuan University for useful discussions.