Dynamic structural evolution of iron catalysts involving competitive oxidation and carburization during CO2 hydrogenation.

Identifying the dynamic structure of heterogeneous catalysts is crucial for the rational design of new ones. In this contribution, the structural evolution of Fe(0) catalysts during CO2 hydrogenation to hydrocarbons has been investigated by using several (quasi) in situ techniques. Upon initial reduction, Fe species are carburized to Fe3C and then to Fe5C2. The by-product of CO2 hydrogenation, H2O, oxidizes the iron carbide to Fe3O4. The formation of Fe3O4@(Fe5C2+Fe3O4) core-shell structure was observed at steady state, and the surface composition depends on the balance of oxidation and carburization, where water plays a key role in the oxidation. The performance of CO2 hydrogenation was also correlated with the dynamic surface structure. Theoretical calculations and controll experiments reveal the interdependence between the phase transition and reactive environment. We also suggest a practical way to tune the competitive reactions to maintain an Fe5C2-rich surface for a desired C2+ productivity.

INTRODUCTION

The structures of heterogeneous catalysts are usually dynamic during redox reactions such as CO2 hydrogenation, where reaction-driven phase transformation or surface reconstruction is often observed (–). Recently, iron-based catalysts have emerged as an active and selective alternative for CO2 hydrogenation into value-added hydrocarbons, and they have been shown to undergo (pre-) reduction, carburization, and oxidation as well (–). Identifying the dynamic structure and understanding the structure-performance relationship during CO2 hydrogenation are highly desirable for clarification of the reaction mechanism and rational design of new catalysts (–). To date, although much attention has been paid to the steady-state performance and the static structure of the catalysts, the dynamic evolution of the structure and performance have rarely been studied.

Nanoscale metallic iron and iron carbide are easily oxidized with exposure to air, which brings uncertainty to the characterizations. The phase transformation and surface reconstruction of iron carbide occur during CO hydrogenation [Fischer-Tropsch synthesis (FTS)] (, ), which possibly also occur during CO2 hydrogenation, given that the two reactions share similar conditions. The observation on spent catalysts is more likely a reflection of an accumulated result over time on stream (TOS), unless time-resolved characterizations are used. Moreover, because of the limitation of characterization techniques, it is still hard to distinguish the diverse iron species (Fe/FeO/FeC) and characterize the surface structure, which is of more significance for catalysis.

The structural evolution of iron-based catalysts has been investigated more widely in FTS (–). By using in situ x-ray diffraction (XRD) and Raman, Paalanen et al. (, ) monitored the evolution of α-Fe/FeO/FeC phases during FTS, where the addition of Na and S promotes the reduction of Fe3O4 to iron carbides. Wang et al. () found the formation of core-shell structures under long-term FTS conditions over various iron catalysts (Fe2O3, χ-Fe5C2, θ-Fe3C, and α-Fe). In addition, note that carburization is affected by the complex surface microenvironment, where the competitive oxygen removal, carbon permeation, hydrogenation, and carbon deposition reactions occur. By interfering them, Tang et al. () yielded high-purity Fe5C2 as an effective FTS catalyst.

The stronger oxidation capability of CO2 compared with CO makes the structural evolution more complicated in CO2 hydrogenation. In general, Fe3O4 and Fe5C2 are the main phases on spent catalysts (, ). Skrypnik et al. () investigated the steady-state composition of FeOC catalysts (mixture of α-Fe, iron carbides, and oxides) along the catalyst bed, which suggests that Fe3O4, FeO, and Fe are converted into iron carbides under reaction conditions. Zhao et al. () developed an Fe2N@C catalyst for selective CO2 hydrogenation to hydrocarbons and found that iron nitrides are in situ converted to highly active iron carbides. Similarly, Zhang et al. () tracked the transformation of the mixture of Fe and Fe3O4 by using in situ XRD and Raman and observed the carburization of α-Fe to χ-Fe5C2 and γ-Fe to θ-Fe3C. Besides carburization, they also showed that the CO-activated Fe5C2 is oxidized to Fe3O4 in CO2 hydrogenation (). Such a transformation strongly depends on the precursors of the catalysts, their microstructure, reaction, and treatment conditions (, , ). After the pre-reduction, Fe(0) is commonly used for CO2 hydrogenation and the performance changes with TOS (, ). However, detailed experimental observations and in-depth understandings on the progressive structural evolution of Fe(0) are still lacking. Likewise, questions concerning the driving force of the structural evolution, the stability of catalysts, and dynamic structure-performance relationship remain to be established.

Here, we prepared Fe(0) nanoparticles (NPs) and tracked the bulk and surface structure with TOS in CO2 hydrogenation. Multiple (quasi) in situ structural characterizations were carried out on the spent catalysts without exposure to air, including in situ XRD, quasi–in situ x-ray photoelectron spectroscopy (XPS), Mössbauer spectroscopy, and high-resolution transmission electron microscopy (HRTEM). Meanwhile, by analyzing the activity and selectivity over pure-phase model catalysts (Fe3O4, Fe3C, and Fe5C2), we inferred the atomic-level surface structural changes. Theoretic calculations and controlled experiments were carried out to further explore the intrinsic driving force of the structural evolution. Moreover, we suggest a practical way to regulate the transformation of the active iron species and the catalytic performance.

RESULTS AND DISCUSSION

Catalytic performance

The monodispersed and crystalline Fe(0) catalysts can be prepared by the pre-reduction of Fe2O3 precursor (figs. S1 and S2). The diffusion limitation has been ruled out (fig. S3); the catalytic performance at 320°C is shown in Fig. 1A and table S1. The data of catalytic performance can be divided into three stages with a linear fitting for each stage (fig. S4). Stage I is an induction period. In the first 2 hours of TOS, a rapid increase of CO2 conversion from 18 to 39% and C2+ selectivity from 20 to 57% and decrease of CH4 selectivity from 56 to 35% and CO selectivity from 24 to 7% were observed. From the second to the 15th hour, the conversion gradually decreases to 28%, and the selectivity to C2+ hydrocarbons decreases to 46%; this deactivation period is termed as stage II. According to the slope of the linearly fitted line in fig. S4A, the deactivation rate in stage II is 0.48%/hour. We defined stage III as the steady state with deactivation rate lower than 0.05%/hour. In this stage, the conversion and selectivity remain almost constant at least for the next 20 hours. Similarly, the chain growth factor α calculated from Anderson-Schulz-Flory model increases during the first 2 to 3 hours, followed by a decreasing period and a steady state (Fig. 1C). The highest α is 0.47, and the highest C2+ and C5+ selectivity (excluding CO) are 61 and 14%, respectively, at the TOS of 3 hours. These three stages can be retarded and more clearly observed at 300°C (Fig. 1B), and both the activation and deactivation rates are lower than that at 320°C (fig. S4B). The induction period is longer (~18 hours) at 280°C, and it takes less time (~7 hours) to go into the steady stage at 340°C (fig. S5). Most previous studies focused on the performance at stage III, usually after 5 to 10 hours on stream, while this work pays more attention to the dynamic stages I and II, that is, the activation and deactivation, where the structures are evolving.

Fig. 1.

CO2 hydrogenation performance versus TOS.

Conversion and selectivity at (A) 320° and (B) 300°C. (C) Chain growth factor and hydrocarbon distribution at 320°C. (D) Specific surface area of spent catalysts and CO2 conversion rate at 320°C. Reaction conditions: P = 3 MPa, gas hour space velocity (GHSV) = 18,000 ml hour−1 gcat−1, and CO2/H2 = 1:3.

Sintering usually leads to the decreased conversion for less active sites exposed. However, the loss of the surface area of the Fe catalyst mainly occurs during the first 3 hours (Fig. 1D), inconsistent with the increasing CO2 conversion during this period. The reaction rate normalized by the surface area continuously increases from 0 to 0.57 mmolCO2 min−1 m−2 during the first 3 hours and then decreases to 0.50 after 10 hours on stream, which further proves that the deactivation should mainly not result from sintering.

Structural characterizations

The iron species in the spent catalysts after 1, 3, 10, and 40 hours on stream at 320°C (denoted as Fe-xh, where x stands for TOS) were characterized by the 57Fe Mössbauer spectroscopy, and the results are shown in fig. S6 and table S2. From the quantitative analysis in Fig. 2C, there is 57% α-Fe in Fe-1h, and only 7 and 3% remains in Fe-3h and Fe-10h, respectively. There is 26% FeC in Fe-1h, and the content of carbides increases to 31% (11% Fe3C and 20% Fe5C2) in Fe-3h and then marginally changes (34%) in Fe-10h. It indicates that the carburization of Fe mainly occurs during the first hour. However, a delay for oxidation was observed, where Fe3O4 presents with a fraction of 6% in Fe-1h and 61% in Fe-3h. Meanwhile, there is 11% Fe2O3 besides 52% Fe3O4 in Fe-10h, and Fe-40h contains mainly 20% Fe5C2 and 80% Fe3O4.

Fig. 2.

Analysis of the bulk composition of spent catalysts.

(A) Overview and (B) the enlarged regions of XRD patterns. (C) Quantitative results of XRD and Mӧssbauer (MES) characterizations. *The low-content and indistinguishable Fe3C and Fe5C2 in Fe-1h are collectively noted as FeC. (D) Heatmap and (E) in situ XRD patterns for Fe(0) during CO2 hydrogenation. Test conditions: T = 340°C, P = 0.8 MPa, and CO2/H2 = 1:3. (F) The intensity change of diffraction peak of Fe5C2 at 44.1° in (D). The standard powder diffraction file cards are Fe (87-0721), Fe3C (65-2413), Fe5C2 (89-7272), and Fe3O4 (89-3854). a.u., arbitrary units.

From the XRD patterns in Fig. 2 (A and B), only metallic Fe was observed in the reduced catalyst. The diffractions of Fe3O4 can be clearly observed in Fe-3h and are more intense in Fe-10h and Fe-40h. The quantitative analysis (Fig. 2C and table S3) shows that there is 83% α-Fe in Fe-1h, and the content of carbide (10%) is lower than that given by Mössbauer (27%). This might be due to the presence of surface amorphous carbides that are not detected by XRD. For Fe-3h, 65% crystalline Fe3O4 and 27% FeC were observed, consistent with the rapid carburization and oxidation suggested by the Mössbauer spectra. With the reactions going on, more crystalline Fe3O4 (75%) was produced at the TOS of 10 hours, and the crystalline compositions remain almost unchanged during stage III, where 78% Fe3O4 and 22% Fe5C2 were observed for Fe-40h, consistent with the Mössbauer results.

It has been shown that Fe or FeC could be oxidized in CO2 hydrogenation (). Liang et al. () found more than 90% Fe3O4 on the unpromoted Fe catalyst after 10 hours on stream, and 33% Fe3O4 was observed by Wei et al. () with the addition of 0.7% Na. Here, the results show that the surface carburization of Fe NPs takes place first to form amorphous or ill crystalline iron carbide during the first 1 hour. Then, a large amount of iron oxides are formed rapidly in the next 2 hours, accompanied by the further crystallization of iron carbide. Then, the oxidation continues until a balanced composition of iron oxides and carbides is reached.

In addition, in the enlarged XRD region in Fig. 2B, with the increasing TOS, the diffractions at 37.7° and 39.7° disappear; those at 42.9°, 43.6°, and 45.9° (rhombus symbols) become weaker, while the diffractions at 39.4°, 44.2°, 47.3°, and 50.2° (heart-shaped symbols) get more intense; these results suggest that Fe3C can transform into Fe5C2. The carburization process was next tracked by in situ XRD at 320° and 340°C, shown in fig. S7 and Fig. 2 (D to F), respectively. In the heatmap (Fig. 2D), the diffractions of FeC appear after 20-min TOS. With the TOS increasing to 250 min, the typical diffractions of Fe5C2 at 43.5° and 44.1° gradually become more intense (green to light yellow and then to red). This can also be observed in the in situ XRD patterns in Fig. 2E and signal changes in Fig. 2F. Besides, the diffractions at 47.3° and 50.0° become stronger; those at 35.2°, 37.5°, and 39.6° shift to lower angles. The results convincingly indicate that metallic Fe can transform into FeC within a short time (20 min) from the onset of CO2 hydrogenation, and the carbon-deficient Fe3C can transform into carbon-rich Fe5C2, which is thermodynamically more stable under these conditions according to carbon potential theory (). Fe3C has been observed to evolve into Fe5C2 under typical FTS conditions (~300°C) (), but it was not yet established in CO2 hydrogenation.

Carbonaceous species were also determined by temperature-programmed hydrogenation (TPH) tests, and the deconvoluted profiles are shown in Fig. 3A. For Fe-1h, a peak of CH4 signal locates at a relatively low temperature of 328°C and a main peak locates at 348°C, corresponding to the hydrogenation of atomic carbon and polymethylene, respectively (, ). There is also a small shoulder peak at 382°C ascribed to surface or amorphous carbide, which is consistent with its weak diffraction. For Fe-3h and Fe-10h, a small shoulder peak of amorphous carbide at 382°C was observed, but the main peaks locate at 407° to 428°C, corresponding to the crystalline carbide (). The largest peak for Fe-10h locates at 428°C, 4°C higher than that for Fe-3h. The ratio of surface to crystalline carbide calculated from the peak area (table S4) for Fe-10h is 0.12, lower than that for Fe-3h (0.28), suggesting the increased crystallinity of FeC.

Fig. 3.

Characterizations on carbon and iron species.

(A) TPH profiles with the mass signal of CH4 and (B) UV Raman spectra of spent catalysts using 257-nm radiation.

In addition, the main peak at 261°C for Fe-1h at the mass signal of H2O corresponds to surface iron oxides (fig. S8). The main peak for Fe-3h and Fe-10h locating at 395°C is ascribed to bulk crystalline Fe3O4, consistent with the XRD and Mössbauer results. Meanwhile, 257-nm ultraviolet (UV) Raman was used to verify the surface (ca. 5 to 10 nm) structure (Fig. 3B). For all the 1-, 3-, and 10-hour spent catalysts, a Raman band at 670 cm−1 and a broad band at 1500 to 1700 cm−1 are observed, which are ascribed to A1g band of Fe3O4 and carbonaceous species, respectively (, ).

The morphological information of the spent catalysts is provided by HRTEM images (Fig. 4). For Fe-1h, an amorphous layer can be observed, and the inner material is confirmed as α-Fe on the basis of the lattice spacing of 1.99 Å, showing a core-shell structure (Fig. 4E). The electron energy loss spectra (EELS) elemental distribution (Fig. 5, A and D) shows that there are both carbon and oxygen on the surface for Fe-1h, and there is almost only Fe observed in the inner part. The TPH, XRD, and Mössbauer results suggest the formation of surface amorphous iron oxide and iron carbide, and the mapping image gives a direct view of the Fe@FeCO structure. The core-shell structure can be more clearly observed for the other samples. For Fe-3h, the core is mainly Fe3O4 and there are nanocrystals of both Fe5C2 and Fe3O4 on the surface (Fig. 4F), and the EELS mapping shows that the carbon distributes on the surface, while the oxygen signals appear at the inner region (Fig. 5E). The core-shell structures of Fe-10h and Fe-40h are similar, where the inner Fe3O4 appears highly crystalline with ordered lattice (Fig. 4, G and H). The Fe3O4 core and the (Fe3O4+Fe5C2) shell are also identified by EELS spectra (Fig. 5F and fig. S9). The thickness of the mixture shell is 8 to 12 nm (fig. S10). The mapping images and the formed crystalline Fe3O4 core show that oxygen migrates from the surface to the bulk, while iron carbides are only formed on the surface (Fig. 5G).

Fig. 4.

HRTEM images of the spent catalysts.

(A and E) Fe-1h, (B and F) Fe-3h, (C and G) Fe-10h, and (D and H) Fe-40h.

Fig. 5.

Microstructure of Fe-xh catalysts.

High-angle annular dark-field–scanning transmission electron microscopy (HAADF-STEM) images of (A) Fe-1h, (B) Fe-3h, and (C) Fe-10h. EELS elemental distribution of (D) Fe-1h, (E) Fe-3h, and (F) Fe-10h. Scale bars, 10 nm. (G) Schematic diagrams of the evolved microstructure. FeCO represents the amorphous and coexisted iron oxide and carbide.

Note that there is no remarkable difference on the bulk structure between Fe-3h and Fe-10h, although they have distinct catalytic performances. To investigate the chemical state of the surface iron species, the binding energy of Fe core-level electron was measured by quasi–in situ XPS. In Fig. 6A, a peak at 706.5 eV of Fe-1h is assigned to the 2p3/2 peak of metallic Fe (). It shifts to 707 to 708 eV for Fe-3h, indicating the carburization of Fe(0) to iron carbide (, ). This can be verified by the C 1s spectra in fig. S11, where a shoulder peak at 283.5 eV assigned to iron carbide () appears for Fe-3h. For Fe-10h, the peak at ~707 eV becomes smaller while the peak at ~710 eV assigned to FeO grows (, ), suggesting the oxidation of surface iron species. In addition, the FeO peak shifts from 710.1 to 710.6 eV along with TOS, corresponding to the increasing oxidation degree. The quantitative analysis is based on the deconvolution of Fe 2p XPS spectra, shown in Fig. 6 (B to D) and table S5. For Fe-1h, besides the 2p3/2 peak at 706.5 eV assigned to Fe(0), signals of Fe5C2 at 707.0 eV, Fe3C at 708.0 eV, and FeO at 709.8 and 711.4 eV were also observed (–). The XPS results suggest that both surface carburization and oxidation occur during the first 1 hour on stream. The peak of Fe(0) can hardly be observed on Fe-3h, and the content of Fe5C2 increases from 8.4 to 20.9% (table S5). In addition, that of Fe3C decreases from 13.2 to 8.0%, indicating the continuous carburization of Fe from carbon-deficient Fe3C to carbon-rich Fe5C2 during 1 to 3 hours. For Fe-10h, the ratio of Fe(C) to FeO decreases from ~40 to 13%, corresponding to the surface oxidation in this period.

Fig. 6.

Quasi–in situ XPS spectra of Fe-xh catalysts.

(A) Fe 2p spectra. Deconvoluted spectra of (B) Fe-1h, (C) Fe-3h, and (D) Fe-10h.

Surface structure

Even so, characterizing the (e.g., sub–2 nm) surface of Fe catalyst is challenging, especially when diverse iron species exist. The catalytic performance can be a sensitive probe for analyzing the surface composition. For this purpose, the relation between iron species and catalytic performance was established on the pure model catalysts—Fe3O4, Fe3C, and Fe5C2—after the diffusion limitation has been ruled out (fig. S12). Under identical conditions (Fig. 7A), CO2 conversion rate (normalized by the surface area) on Fe5C2 is 5.36 mmol hour−1 m−2, higher than that on Fe3C (1.91 mmol hour−1 m−2) and 10 times higher than that on Fe3O4 (0.51 mmol hour−1 m−2). The conversion rate over FeC is more than 20 times higher than that on Fe3O4 at a similar conversion level (table S6). The product distribution at different CO2 conversion over Fe3C is similar to Fe5C2 (tables S6 and S7), which can catalyze the production of C2+ hydrocarbons either directly from CO2 or going through CO intermediate (, , ). Conversely, CO is the major product (more than 70%) on Fe3O4. The C2+ formation rate over Fe5C2 and Fe3C are 2.14 and 0.73 mmol hour−1 m−2, respectively, 30 to 100 times higher than that on Fe3O4. That is, the ability to catalyze C2+ production ranks as Fe5C2 > Fe3C >> Fe3O4. Density functional theory (DFT) investigations showed that CO2 activation is more facile on Fe5C2 and Fe3C than on Fe3O4 (, ), and Han and colleagues (, ) found that the transition of Fe5C2 to Fe3O4 causes the deactivation in CO2 hydrogenation.

Fig. 7.

Comparison of CO2 hydrogenation performance.

(A) CO2 conversion rate and C2+ formation rate over different model catalysts. Reaction conditions: P = 3 MPa, T = 320°C, CO2/H2 = 1:3, TOS = 20 min, and GHSV = 72,000 ml hour−1 gcat−1. (B) Product selectivity and CO2 conversion over model catalysts (on the left of the dashed line) and real Fe(0) catalysts with different TOS (on the right); GHSV = 18,000 ml hour−1 gcat−1. *GHSV = 72,000 ml hour−1 gcat−1.

From the comparison on the performance over model catalysts and Fe NPs with different TOS (Fig. 7B), Fe-1.7h exhibits a CO2 conversion (~38%) and product distribution similar to pure Fe5C2 under the same conditions, indicating that it has an Fe5C2-rich surface. With the limited contribution of Fe3O4 as mentioned, the decreased CO2 conversion (~28%) over Fe-10h should be attributed to the loss of Fe5C2 sites. This is verified by the similar product distribution on Fe-10h to that on pure Fe5C2 at a conversion of 28% [gas hour space velocity (GHSV) increasing from 18,000 to 72,000 ml hour−1 gcat−1]. On the basis of the quantitative analysis by quasi–in situ XPS (Fig. 6 and table S5), the content of iron carbide is highest on Fe-3h and it also exhibits the highest CO2 conversion compared with Fe-1h and Fe-10h (fig. S13). The content of Fe5C2 correlates well with C2+ formation rate, which proves that Fe5C2 is the main active species for C2+ production. Combing the characterization and catalytic performance, we can conclude that the low conversion on Fe-1h results from fewer FeC sites, and the improved C2+ production on Fe-1.7h is caused by the further carburization and the transformation of surface Fe3C to Fe5C2. The deactivation in stage II is due to the surface oxidation of iron carbides.

Consequently, we can conclude the evolution of surface and bulk structure of Fe(0) catalyst (Fig. 8A). First, the dissociated carbon species react with the surface of Fe NPs to form an amorphous layer of iron carbides during the first 1 hour on steam. As the reactions get started, iron oxides were also formed. The oxygen can migrate into the subsurface, while carbon mainly carburizes the surface layers (Fe to Fe3C then to the carbon-rich Fe5C2) during TOS of 2 to 3 hours. This Fe5C2-rich surface gives the highest CO2 conversion rate, C2+ selectivity, and chain growth factor. Once the crystalline Fe3O4 core is formed, excess oxygen will oxidize the surface iron (carbide), leading to deactivation. Last, the surface oxidation, H2 reduction, and carburization reach a dynamic balance, leading to a stable Fe3O4@(Fe5C2+Fe3O4) core-shell structure and a steady catalytic performance.

Fig. 8.

Analysis of the structural evolution of Fe(0).

(A) Schematic diagram of the evolution of iron species during CO2 hydrogenation. The red and gray spheres represent the oxygenous and carbonous species, respectively. (B) The computed phase diagram of iron, iron carbides, and oxides. Each solid line is the boundary between two domains, where two species are in equilibrium. The species with the lowest free energy is marked for each domain. The free energies of reactants and equilibrated conditions are calculated on the basis of the compositions shown in table S10. The red-blue arrow is the reactive environment shift. The black arrows are the phase transition of the catalyst.

Driving force of the structural evolution

To understand the evolution of Fe NPs, we applied DFT and ab initio thermodynamics (). Perdew-Burke-Ernzerhof (PBE)+U (, ) calculations were performed to construct the phase diagram to locate the state of the system at reaction conditions. The bulk energy of Fe, Fe5C2, Fe3C, FeO, Fe3O4, and Fe2O3 were computed, where a U correction was applied for oxides to account for the localized d state electrons with a value of 4 eV (). To make the energies from generalized gradient approximation (GGA) (metal and carbides) and PBE+U (oxides) comparable, we derived a correction term (ΔEM) to account for the arbitrary reference shift caused by the U correction. We derived a ΔEM value of 1.67 eV using the formalism described by Li et al. () and Xu et al. (). The details for the derivation of ΔEM are included in table S9. The computed phase diagram for Fe species as a function of chemical potential of carbon (μC) and oxygen (μO) is shown in Fig. 8B. The calculated free energies of reactants (CO2/H2/N2 = 21:63:16) is located at the red dot in the figure, which falls in the Fe5C2 domain. When metallic Fe is exposed to the reactants, there is a substantial thermodynamic driving force leading to a phase transition from Fe to Fe3C and then to Fe5C2, which agrees well with the experimental observations at 0 to 1 hour. However, once the active Fe5C2 is formed on the surface, hydrogenation of CO2 is promoted, releasing oxygen into the gas phase (mostly in the form of H2O) and leading to the increase of μO, which, in turn, causes oxidation of the catalyst. Carburization is more difficult for Fe3O4 than for Fe (, ), which is evident in the in situ XRD data (fig. S14B), and once oxidation begins, it is more difficult for carbon to permeate and carburize the Fe3O4 interlayer. The increase in μC also causes a drop of μO. Thus, the formation of surface Fe5C2 leads to a shift in the reactive environment, as shown by the red-blue arrow in Fig. 8B. The product gas at equilibrium (TOS = 10 hours; compositions are in table S10) is located at the blue dot in the Fe3O4 domain. As a result, the catalyst loses the thermodynamic driving force for further carburization and the bulk transforms to Fe3O4, which corresponds to the 1- to 10-hour segment in Fig. 8A.

The final core-shell structure of the catalyst is the result of the interdependence between the phase formation on the catalyst and the reaction conditions. The formed FeC species are active for the CO2 hydrogenation, which results in higher μO and lower μC and shifts the reaction conditions toward the Fe3O4 domain on the phase diagram. The changed microenvironment, in turn, leads to oxide formation and prohibits carburization. Thus, the formation of carbides exhibits a negative feedback loop, as underscored with a bidirectional arrow in Fig. 8B. If the microenvironment shifts to the red side slightly, then more carbide will be formed on the surface, which promotes oxygen release and pushes the reaction condition back to the equilibrium position and vice versa. The equilibrium reaction condition is located in the Fe3O4 domain on the phase diagram, so the core bulk structure of the catalyst will be Fe3O4 after a long reaction time. However, full oxidation of the catalyst causes deactivation for CO2 reduction. In such a case, the reaction conditions would correspond to those of the initial reaction conditions with high μC and low μO, which will promote the surface carburization. As a result, the catalyst lastly transforms into the core-shell structure with bulk Fe3O4 and surface FeC.

It can be seen that the oxidation occurs after CO2 hydrogenation starts. To further explore this dynamic process, we carried out several control experiments. After stage I (induction), the catalyst was treated with 20% CO2 for 3 hours. When the feed gas was switched back to the mixture of CO2 and H2, a decreased CO2 conversion of 26% and C2+ selectivity of 42% were observed (Fig. 9A). In addition, the catalyst was treated with 10% water vapor, similar to its partial pressure under reactions (6 to 12% at a CO2 conversion of 30%). As a result, CO2 conversion decreased markedly from 37 to 7%, and CO selectivity increased from 12 to 51% (Fig. 9B). These results reveal that the by-product water notably oxidizes the activated catalyst to form an Fe3O4-rich surface. The temperature-programmed oxidation–mass spectrometry (TPO-MS) experiments show that both CO2 and H2O can oxidize Fe(0) or FeC (fig. S15). The H2O-TPO profile exhibits an oxidation peak at a lower temperature (416°C) compared to the CO2-TPO profile (548°C), suggesting that the oxidation by H2O occurs more easily. After the treatment by CO2 or H2O, the catalytic performance recovered gradually, where CO2 conversion and C2+ selectivity increased. This “self-regeneration” verified again that the surface composition and the catalytic performance are determined by the dynamic balance of oxidation and carburization.

Fig. 9.

Dynamic behaviors of CO2 hydrogenation over Fe catalysts.

(A) Treating with 20% CO2 and (B) treating with 10% H2O. (C) CO2 conversion and (D) C2+ selectivity with co-feeding 1, 5, and 10% H2O. (E) CO2 conversion. (F) CO and C2+ selectivity at different H/C. (G) Regulating GHSV and treating with H2. Reaction conditions: P = 3 MPa, T = 320°C, GHSV = 18,000 ml hour−1 gcat−1 except (G), and H/C = 3 except (E) and (F).

In addition, we co-fed water vapor in CO2 hydrogenation, and the performance also shows three stages (Fig. 9, C and D, and fig. S16). However, with the increased concentration of H2O from 1 to 10%, lower CO2 conversion, C2+ selectivity, and higher CO selectivity were observed at the steady state. We estimated the gas components at each condition, as shown in table S10. Note that the partial pressure of CO2 at steady state remains almost unchanged at ~5 bar. However, when we co-fed 10% H2O, the partial pressure of H2O is 4.5 bar at steady state (including the H2O formed during the reaction), only 0.8 bar higher than that without co-feeding H2O (3.7 bar). The ratio of H2O in total gas increased from 12% (3.7 bar in 30 bar) to 15% (4.5 bar in 30 bar), while that of CO2 remains constant at 17%. As a result, this 3% increase of H2O leads to a much lower CO2 conversion (~10%), and CO replaces C2+ to become the main product. In addition, the quasi–in situ XPS reveals that the surface was almost fully oxidized (fig. S17 and table S11) with co-feeding 5% H2O. Therefore, H2O plays a more important role than CO2 in oxidation during the reaction-induced structural evolution. Moreover, the diffractions of Fe3O4 were observed in the in situ XRD patterns in fig. S18, but they disappeared, and the carburization was accelerated in Fig. 2 because of an on-purpose faster removal of H2O (details in the Supplementary Materials).

After stage III (steady state), the flow rate of reactant gas was decreased from 3600 to 720 ml/hour (Fig. 9G). CO2 conversion increases from 29 to 34% instantly because of the increased contact time, accompanied by the increased C2+ selectivity from 52 to 56%. Nevertheless, the CO2 conversion continually decreases to 24% in the next 8 hours, even lower than that at the steady state with 3600 ml/hour. Afterwards, the flow rate was set back. However, the CO2 conversion and C2+ selectivity significantly decreased instead of going back to the original level. The mass transfer limitation is proved to be small (fig. S19), and the dynamic catalytic performance results from the evolution of surface structure. We assumed that more produced water at the high conversion and low GHSV promotes the oxidation, breaking the balance of oxidation and carburization until a new one is established. This indicates that the surface composition is highly sensitive to the microenvironment, which is easily affected by the operating conditions such as GHSV, and we emphasize that more attention should be paid to the real surface structure, especially when these conditions are changed (). With the increasing TOS, the conversion and C2+ selectivity increase gradually but are still much lower than the original level within 6 hours. That is, the self-regeneration is too slow, likely because of the slower H2 dissociation on the FeO-rich surface than that on FeC, evidenced by H/D exchange experiments (). The following 4-hour H2 reduction can completely regenerate the catalyst, which confirms that the excessive oxidation by H2O caused the deactivation.

Therefore, protecting iron carbides from oxidation is an effective way to slow down the deactivation (, ). The calculations in Fig. 8B predict that increasing μC and decreasing μO can promote the carburization. Then, we increased the H/C ratio (feed H2/CO2) from 3 to 6 (Fig. 9, E and F, and fig. S20), and the induction period becomes longer under the carbon-deficient atmosphere. The decreased magnitude of CO2 conversion in stage II becomes lower, and the decrease of C2+ selectivity and the increase of CO selectivity could hardly be observed at the H/C ratio of 5 and 6. The deactivation is hindered when more H2 is fed. The quasi–in situ XPS spectra show that more FeC is present with higher H/C ratio (5 versus 3; fig. S17 and table S11), indicating that increasing the partial pressure of H2 can effectively increase the content of iron carbide and maintain the desired catalytic performance. However, too much hydrogen also causes the termination of chain growth. The deactivation is significantly inhibited, and the highest C2+ selectivity of 50% (CO2 conversion of ~40%) at the steady state is achieved at an optimized H/C ratio of 5. Regulating the ratio of H2O to H2 can be a practical strategy to tune the surface competitive carburization and oxidation.

In summary, the structural evolution of Fe(0) catalysts during CO2 hydrogenation has been investigated. Quasi–in situ analysis on the model catalysts (Fe3O4, Fe3C, and Fe5C2) provides a benchmark for correlating the dynamic structure with the catalytic performance. During the induction period, reduced Fe is carburized to Fe3C and then to Fe5C2, which is driven by the thermodynamics. The side product of CO2 hydrogenation, H2O, oxidizes the catalysts, resulting in the bulk transition to Fe3O4. Theoretical calculations provide fundamental insight into the interdependence between phase transition and reactive environment, which is the key to the formation of the Fe3O4@(Fe5C2+Fe3O4) core-shell structure. The surface composition depends on the balance of oxidation and carburization, and it is highly sensitive to the operating conditions (e.g., GHSV and H/C), where the by-product water plays a key role. The balance of surface competitive reactions can be tuned by regulating the reaction conditions, providing an opportunity to maintain an Fe5C2-rich surface, desired high CO2 conversion, and C2+ selectivity as well. These findings highlight the importance of real-time analysis on the surface structure and the sensitivity of iron (carbide) catalysts to microenvironment during CO2 hydrogenation to hydrocarbons.

MATERIALS AND METHODS

Preparation of catalysts

Preparation of Fe2O3 NPs

Bulk Fe2O3 was prepared by a hydrothermal method. Briefly, 2.5 g of Fe(NO3)3·9H2O and 1.5 g of urea (analytic grade; Damao Chemical Reagent Factory) were dissolved in 75 ml of deionized water with continuous stirring. The solution was heated to 120°C and kept for 5 hours in a stainless steel Teflon-lined autoclave. Afterwards, the product was centrifuged and washed with deionized water and ethanol three times, dried at 100°C for 12 hours, and then calcined at 500°C in air for 4 hours.

In situ reduction and carburization of Fe2O3 NPs

Before the catalytic testing, the as-prepared Fe2O3 NPs were reduced or carburized in a fixed-bed reactor. The conditions of the pretreatment were determined on the basis of the in situ XRD results in figs. S2 and S14. Treated with pure H2 (30 ml min−1) at 3 MPa, Fe2O3 can be reduced to Fe3O4 at 300°C for 15 min and to metallic Fe at 400°C for 4 hours. Treated with pure CO (30 ml min−1) at ambient pressure, Fe2O3 can be carburized to Fe5C2 at 350°C for 3 hours and to Fe3C at 450°C for 30 min.

Catalytic testing

Typically, 0.2 g of the Fe2O3 powder catalyst was mixed with quartz sand to 1 ml and then loaded in a quartz tube of 8 mm inner diameter in a continuous-flow fixed-bed reactor. After the pretreatments mentioned above (400°C at 3 MPa in H2 unless otherwise noted), the reactor was purged with N2 and set to 320°C. Reactant gas (CO2/H2/N2 = 21:63:16) was then fed in, and the system was pressurized gradually to 3 MPa. The water vapor was co-fed using a steel gas wash bottle, which is placed before the tube reactor and can be pressurized to 3 MPa. The content of water vapor is controlled by regulating the temperature of the bottle based on the saturated vapor pressure. Products were analyzed by an online chromatography (Agilent 7890B), where CO, CH4, CO2, and N2 were detected on a thermal conducted detector and C2-C4 hydrocarbons were detected on a flame ionization detector. CO2 conversion, product selectivity, CO2 conversion rate, and formation rate of product (normalized by surface area) were calculated using the following equationswhere CO2,in and CO2,out represent the molar fraction of CO2 at the inlet and outlet, respectively, C represents the molar fraction of C at the outlet, i represents the number of carbon atoms in C; QCO represents the molar flow rate of inlet CO2, m represents the weight of the catalyst used (typically 0.2 g), and SBET represents the specific surface area of the catalyst shown in Fig. 1D and table S8.

The calculation of chain growth factor α is based on the Anderson-Schulz-Flory distribution model, as shown belowwhere n represents the carbon number of the hydrocarbon products and W represents the mass fraction of the product among all hydrocarbons. ln(W/n) is plotted with n, and the slope of the linearly fitted line is lnα (fig. S21).

Catalyst characterization

All the reduced and spent catalysts were transferred along with the tube reactor and stored into a glove box filled with Ar without exposure to air. In addition, most of the preparation or treatments for characterizations were carried out in the glove box. For XRD measurements, the samples were sealed with Kapton tape, and the patterns were acquired on a Rigaku SmartLab 9-kW diffractometer with Cu Kα radiation (λ = 1.5406 Å) at a voltage of 45 kV and a current of 200 mA.

In situ XRD experiments were carried out in an XRK900 reactor chamber (Anton-Paar Corporation) equipped in the diffractometer. The solid powder was pressed into a pellet and then packed into a ceramic sample stage. Typically, pure H2 or CO was fed at a flow rate of 30 ml min−1 for the pretreatment, and the temperature was controlled by a TCU750 temperature control unit. Then, the feed gas was switched to the reactant gas (CO2/H2/N2 = 21:63:16). The XRD sample chamber can be pressurized to 0.8 MPa. The patterns were collected at a speed of 5° min−1, a voltage of 45 kV, and a current of 200 mA using a one-dimensional D/teX detector. The quantitative analysis on XRD data is based on the Reference Intensity Ratio (RIR) method using the PDXL2 software.

For 57Fe Mössbauer measurements, the samples were sealed with film, and the spectra were recorded on a Topologic 500A spectrometer and a proportional counter at room temperature. 57Co (Rh) moving in a constant acceleration mode was used as a radioactive source, and the analysis on spectra was conducted assuming a Lorentzian line shape for computer folding and fitting.

For TPH experiments, the samples were loaded in a U-type tube, sealed with valves, and then taken out from the glove box. About 0.1 g of the sample was loaded and pretreated with N2 for 1 hour, and then, pure H2 was fed in. The temperature was then increased to 800°C at a ramp of 5°C/min. CH4 was monitored using mass signals of 15 (CH3 fragments of CH4) instead of 16 to avoid the interference of water vapor and CO2 cracking (), and H2O was monitored using mass signals of 18 in a Pfeiffer OmniStar mass spectrometer. For TPO-MS experiments, the sample was treated with 20% CO2/Ar or ~10%H2O/Ar. The temperature was then increased to 700°C at a ramp of 5°C/min. CO and H2 were monitored using mass signals of 28 and 2, respectively.

UV Raman spectra were recorded using a homemade triple stage UV Raman spectrometer with a spectral resolution of 2 cm−1. UV laser line at 257 nm was from a double-frequency 514-nm laser (model NPL-N-257; Changchun New Industries Optoelectronics Technology Co. Ltd.).

For the quasi–in situ XPS measurements, the sample after reaction was transferred to a glove box through an airtight reaction tube and then transferred to the XPS analysis chamber with a mobile transfer chamber. The sample was kept in reaction atmosphere/inert atmosphere/vacuum during the entire transfer process without exposure to air. The quasi–in situ XPS measurements were carried out on a spectrometer equipped with an Al Kα x-ray source operated at 300 W. Binding energies were calibrated using the C 1s peak at 284.8 eV.

The surface area was determined by N2 adsorption-desorption on a Quantachrome AUTO-SORB-1-MP sorption analyzer at −196°C and calculated using the Brunauer-Emmett-Teller model. Before the measurements, the samples were degassed at 150°C for 4 hours. The HRTEM images were taken using a FEI Tecnai F30 instrument, and the ethanol suspension of samples was dropped on copper grids for observation.

The high-angle annular dark-field-scanning transmission electron miscroscopy (HAADF-STEM) images and EELS spectrum images were collected on Themis ETEM (Thermo Scientific) located in the Electron Microscopy Center at the Dalian Institute of Chemical Physics, Chinese Academy of Sciences, which was used at 300 kV in dual EELS mode with an energy resolution better than 1.3 eV at a collection angle of 100 mrad. The FeOC catalysts were loaded onto the Si3N4 chip of a microelectromechanical system–based heating holder (Wildfire, DENSsolutions). To rule out the effect of hydrocarbon contamination, the samples were kept at 150° to 200°C in high vacuum (P < 5 × 10−8 mbar).

Computational methods

Computation settings

All DFT computations are performed with the Vienna Ab initio Simulation Package (VASP 5.4.4) (). The PBE functional was used for the exchange-correlation energy (). The projector augmented wave theory was applied to treat frozen core electrons (). The valence electrons of each element type were taken as Fe (4s23d6), C (2s22p2), and O (2s22p4). All calculations were spin-polarized. The plane wave basis was truncated at 500 eV for gas molecules. For optimization of bulk structures, a 650-eV cutoff was applied to mitigate the effect from Pulay stress. Gaussian smearing with 0.05-eV width was applied to oxides and all molecules. First-order Methfessel-Paxton smearing was applied to Fe and carbides (). Brillouin zones were treated with a Monkhorst-Pack k-point mesh (). The k-points, geometries, and lattice parameters for each bulk structure are summarized in table S12. The energies of all gas molecules are calculated in a cell of 13 Å by 14 Å by 15 Å with a 1 × 1 × 1 k-point mesh. The DFT-D3 method for van der Waals interactions was applied for all calculations (). A U correction was applied to the calculation of oxides to account for the localized d states ().

Ab initio thermodynamics

Ab initio thermodynamics was applied to construct the bulk phase diagram and locate certain reaction conditions. The formation energy was defined as followswhere ΔEform is the formation energy; EFe is the DFT energy of a bulk structure with x Fe atoms, y O atoms, and z C atoms; EFe is the DFT energy of metallic Fe; mO is the chemical potential of oxygen; and mC is the chemical potential of carbon. With the assumption of equilibrium between the gas species under reaction conditions, the chemical potential of H, O, and C were derived by free energies of H2, CO2, and H2O, as they are the dominant species present in the gas phasewhere μH is the chemical potential of hydrogen, GH is the free energy of H2, GH is the free energy of H2O, and GCO is the free energy of CO2. The free energies of molecules were calculated aswhere is the DFT energy of gas molecule, ZPVEgas is the zero-point vibrational energy of the gas molecule, is the enthalpy change of the gas molecule from 0 to 593.15 K, T is the temperature (593.15 K), is the entropy of the gas molecule at 593.15 K, R is the gas constant, Pgas is the partial pressure of the gas, and Pref is the reference pressure (1 bar).