Role of the Oxide Support on the Structural and Chemical Evolution of Fe Catalysts during the Hydrogenation of CO2

Iron-based catalysts are considered active for the hydrogenation of CO2 toward high-order hydrocarbons. Here, we address the structural and chemical evolution of oxide-supported iron nanoparticles (NPs) during the activation stages and during the CO2 hydrogenation reaction. Fe NPs were deposited onto planar SiO2 and Al2O3 substrates by dip coating with a colloidal NP precursor and by physical vapor deposition of Fe. These model catalysts were studied in situ by near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) in pure O2, then in H2, and finally in the CO2 + H2 (1:3) reaction mixture in the mbar pressure range and at elevated temperatures. The NAP-XPS results revealed the preferential formation of Fe(III)and Fe(II)containing surface oxides under reaction conditions, independently of the initial degree of iron reduction prior to the reaction, suggesting that CO2 behaves as an oxidizing agent even in excess of hydrogen. The formation of the iron carbide phase, often reported for unsupported Fe catalysts in this reaction, was never observed in our systems, even on the samples exposed to industrially relevant pressure and temperature (e.g., 10 bar of CO2 + H2, 300 °C). Moreover, the same behavior is observed for Fe NPs deposited on nanocrystalline silica and alumina powder supports, which were monitored in situ by X-ray absorption spectroscopy (XAS). Our findings are assigned to the nanometer size of the Fe particles, which undergo strong interaction with the oxide support. The combined XPS and XAS results suggest that a core (metal-rich)−shell (oxide-rich) structure is formed within the Fe NPs during the CO2 hydrogenation reaction. The results highlight the important role played by the oxide support in the final structure and composition of nanosized catalysts.


INTRODUCTION
Due to ecological and global climate concerns, there is a growing interest in developing environmentally friendly technologies to utilize CO 2 , especially in applications where it is reacted with H 2 (which can be obtained from renewable sources) to produce value-added chemical products such as high-order hydrocarbons and alcohols. 1,2 In this respect, one approach that has received considerable attention is the socalled "modified" Fischer−Tropsch (FT) synthesis (or CO 2 − FTS), which uses CO 2 as a feedstock instead of CO. The concept is to combine the reverse water−gas shift (r-WGS) reaction (i.e., CO 2 + H 2 → CO + H 2 O) and the subsequent hydrogenation of syngas (CO + H 2 ) via the classical FT process on a single catalyst. Since Fe-based catalysts were found active for both reactions, iron has become the most studied component of the CO 2 −FTS catalysts. To improve both the activity and selectivity toward C 2+ products, iron is usually promoted by other 3d metals like Co, Ni, Cu, 3,4 or alkali metals like K and Na. 5−7 It is not surprising that many ideas about CO 2 hydrogenation catalysts on Fe originate from the more studied FTS systems, where the active phase is commonly associated with Fe carbides, in particular Hagg carbide (χ-Fe 5 C 2 ), probably coexisting with the Fe oxide phase. 8−12 Iron carbides are also frequently considered to be the active phase(s) for CO 2 hydrogenation, with the reduced state of iron playing an essential role for the carbide formation. 13 For example, a comparative study of CO 2 −FTS catalysts prepared from Fe 2 O 3 and CuFeO 2 and activated in H 2 revealed a greater extent of the Fe(III) → Fe(0) reduction in CuFeO 2 , that favored the selective carburization toward the Hagg carbide and improved the selectivity toward higher (C 5+ ) hydrocarbons considerably. 4 Direct conversion of CO 2 to gasoline-range (C 5 −C 11 ) hydrocarbons was demonstrated on a multifunctional catalyst (Na−Fe 3 O 4 /HZSM-5), which cooperatively catalyzed a tandem reaction, presumably on three types of active sites (Fe 3 O 4 , Fe 5 C 2 , and acid sites in zeolite). 14 Recently, the structural evolution of the bulk iron-containing phases in CO 2 hydrogenation was established under reaction conditions using a combination of Raman spectroscopy and X-ray diffraction coupled with online gas chromatography. 15 During the activation in pure hydrogen and the subsequent CO 2 hydrogenation reaction, the iron oxides (α-Fe 2 O 3 and γ-Fe 2 O 3 ) underwent the following transformation: α-Fe 2 O 3 (γ-Fe 2 O 3 ) → α-Fe 3 O 4 (γ-Fe 3 O 4 ) → α-Fe (γ-Fe) → χ-Fe 5 C 2 (θ-Fe 3 C). Both iron carbide phases showed high catalytic activities, with χ-Fe 5 C 2 exhibiting highest selectivity toward lower olefins.
Although the Fe-based CO 2 −FTS catalysts are often used unsupported, it has been documented that the support can influence the catalytic performance as a result of its effect on the reduction and carburization. For example, the activity of Fe/SiO 2 catalysts was found to be substantially lower than Fe/ C catalysts. 16 It was hypothesized that a stronger interaction of the iron oxide with the silica inhibited the reduction to the active metallic iron. Additionally, mixing silica and iron oxide resulted in a nonlinear dependence of the catalytic activity and selectivity in the FT reaction as a function of the Fe/Si ratio. 17 The complexity of NP/support interactions has been addressed in several review articles; see, for example, ref 18. The support may affect the initial dispersion of the Fe nanoparticles (NPs) and the possible agglomeration during the reaction that may, in turn, influence the catalytic performance. 19 Considerable particle size effects have recently been reported for ZrO 2 -supported Fe catalysts, 20 as well as for SiO 2supported Ni NPs, 21 but a systematic study of particle size effects for C 2+ production is still missing.
Despite significant progress in recent years, further work is still needed to elucidate the reaction mechanism. This would allow the rational design of more efficient and selective CO 2 − FTS catalysts. In this respect, fundamental studies applying surface-sensitive techniques to well-defined model systems may substantially contribute to a deeper understanding of the relationship between reactivity and structural/chemical parameters of the real catalysts. Although such "surface science" studies are usually carried out in ultrahigh vacuum (UHV) conditions, modern advances of surface science techniques such as near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) provide the possibility of studying catalytic systems under more realistic conditions. 22−25 In order to shed light on the state of Fe under CO 2 hydrogenation reaction conditions and the role of the support in the structural and chemical evolution of the Fe-based catalysts, in this work, we performed NAP-XPS studies of several model systems prepared on planar SiO 2 and Al 2 O 3 supports. Fe NPs were synthesized using an inverse micelle encapsulation method that allows the fabrication of monodispersed NPs. 26,27 In addition, we used physical vapor deposition (PVD) of Fe to mimic fully reduced, metallic NPs. To model supported powder catalysts, micellar NP precursors were deposited onto nanocrystalline SiO 2 and Al 2 O 3 powder supports and studied in situ by NAP-XPS and Xray absorption spectroscopy (XAS) in different gas atmospheres. The XAS results revealed a high stability of the iron oxide phases under reaction conditions and considerable amounts of metallic iron in the catalysts. The combined XPS and XAS results indicate the formation of a core (metal-rich)− shell (oxide-rich) NP structure. Moreover, in contrast to previously reported studies on Fe-based catalysts, iron carbide phases were not observed in any of our model catalysts studied, revealing the crucial role of the support on the structural evolution of highly dispersed Fe catalysts.

METHODS AND MATERIALS
Iron NPs were synthesized by the inverse micelle encapsulation method. 28 A solution using poly(styrene-b-2-vinylpyridine) diblock copolymers (PS:P2PVP ratio 8200:8300, Polymer Source, Inc.) and toluene (ROTH, 99.5%) was prepared to form micellar cages. Then FeCl 3 (Sigma-Aldrich, 99.99%) was dissolved into the micellar solution and stirred for 48 h. The prepared micelles were deposited by dip coating onto Si(100) wafers having an oxide layer about 285 nm in thickness (provided by Sigert Wafer GmbH) and on an α-Al 2 O 3 (0001) single crystal substrate (SurfaceNet). The polymeric ligand was removed by exposure to O 2 plasma for 20 min (SPI Plasma Prep III Plasma Etcher operated at 20 W in 0.3 mbar of O 2 ). Then the samples were transferred into an UHV chamber through air.
In another set of experiments, Fe was deposited by PVD onto the same substrates from a Fe rod (99.99%, Goodfellow) using an e-beam assisted evaporator (Focus EFM4). Prior to the Fe deposition, the SiO 2 /Si(100) and Al 2 O 3 (0001) substrates were annealed in UHV at 500°C for 20 min and inspected by XPS.
XPS measurements were carried out in an UHV system (base pressure ∼2 × 10 −10 mbar) manufactured by SPECS. The spectra were collected using a monochromatic Al K α X-ray source (hν = 1486.6 eV) and a hemispherical analyzer (Phoibos 150). For energy calibration, we set the Si 2p 3/2 level in SiO 2 at 103.3 eV and the Al 2p 3/2 level in Al 2 O 3 at 74.4 eV. Due to severe sample charging on the alumina single crystal in UHV, a flood gun (SPECS FG 15/40) was used for charge compensation. Spectra were processed using the CasaXPS software.
To study the structural transformations at catalytically relevant conditions, the "as prepared" samples were transferred into a high-pressure cell (SPECS HPC20) attached to the main UHV chamber through a gate valve. The sample was heated at a rate of 10°C/min by the focused light of a halogen lamp through a quartz window. After treatment, the reactor was pumped out, and the sample was transferred back into the XPS chamber without exposure to air.
Powder catalysts were prepared by incipient wetness impregnation of γ-Al 2 O 3 (Inframat, 99.99%) and SiO 2 (Strem Chemicals, 99%) powders with the Fe-containing micellar solution corresponding to an Fe loading of about 10 wt %. The catalysts were calcined in 1 atm of O 2 + Ar (1:1) at 400°C for 6 h. The calcined catalysts were characterized by scanning transmission electron microscopy (STEM, FEI Talos F200X). The specific surface area of the powder catalysts was measured with Autosorb-1-C (Quantachrome Instrum.) using N 2 adsorption. CO 2 hydrogenation reactions on the powdered catalysts were carried out in a fixed-bed mass flow reactor made of stainless steel, with the inside of the reactor tube coated by a glass liner. The samples were first reduced in pure H 2 (15 mL min −1 ) for 2 h at 400°C then cooled down to 300°C, and subsequently the CO 2 + H 2 (1:3) reaction mixture was introduced into the reactor at 10 bar. The gas hourly space velocity (GHSV) was set to 3600 mL min −1 g −1 in all catalytic tests. The reaction was monitored using online gas chromatography coupled with mass spectrometry (GC/MS, Clarus 690GC-SQ8MS from PerkinElmer). The corresponding mass spectra of each GC peak were analyzed using the NIST Mass Search Program (version 2.3).
XAS measurements at the Fe K-edge (7112.0 eV) were carried out at beamline BL 2-2 at Stanford Synchrotron Radiation Lightsource (SSRL). A Si(220) monochromator detuned by 45% was used for energy selection, and XAS data were collected in transmission mode. Ionization chambers, filled with pure N 2 , were used to measure the intensity of the X-rays before and after the sample. The catalyst was diluted with silica in the 1:1 ratio and packed into a quartz capillary (1.0 mm diameter, wall thickness 10 μm) mounted into a Clausen cell. 29 Spectra alignment and normalization were performed using the Athena software. 30 Linear combination fitting of the XANES spectra was performed in the energy range between 7100 and 7230 eV, using spectra for an Fe foil, FeO and FeOOH as references for Fe(0), Fe(II) and Fe(III) states, respectively, and requiring the total concentration of Fe(0), Fe(II) and Fe(III) species to be equal to 100%. Extended X-ray absorption fine structure (EXAFS) data fitting was performed with the FEFFIT code, 30 using phase and amplitude functions obtained from ab initio calculations performed with the FEFF8.5 code for reference materials (bcc-Fe and FeO) to model contributions of Fe−Fe and Fe−O bonds, respectively. Fitting of χ(k)k 2 spectra was carried out in R-space in the range from 0.6 to 3.0 Å. Fourier transforms were carried out in the k range from 2.0 to 9 Å −1 . To model the EXAFS spectra of the "as-prepared" samples, we only included the contribution of the Fe−O bond. Furthermore, the coordination number N, bond-length R, disorder factor σ 2 , third cumulant of the bond-length distribution c (3) , skewness of the distribution, and the correction to the photoelectron reference energy ΔE 0 were treated as fitting parameters. For the "reduced" samples, we additionally included two Fe−Fe paths corresponding to the first two coordination shells in bcc-Fe. To reduce the number of fitting variables for the reduced samples, we have set c (3) to 0 for all paths and σ 2 to be the same for both Fe−Fe paths and also assume that the ratios of the interatomic distances and coordination numbers for these paths are the same as for the ideal bcc-Fe structure. The amplitude reduction factor due to many-electronic processes (S 0 2 factor) was 0.74 ± 0.05 obtained from the fitting of the EXAFS spectrum for the Fe foil.

RESULTS AND DISCUSSION
Parts a and b of Figure 1 display typical AFM images of the micelle-synthesized Fe NPs deposited onto SiO 2 /Si(100) and Al 2 O 3 (0001) substrates, respectively, obtained after dipcoating, and an oxygen plasma treatment was needed to remove the polymeric ligands. Since the lateral size of the particles imaged by AFM is overestimated due to the tip-NP convolution effect, the particle size is determined by analysis of their heights, which are 4.0 ± 1.1 nm and 4.4 ± 1.3 nm for NPs on SiO 2 and Al 2 O 3 supports, respectively (see Figure S1 in the Supporting Information). The differences in the particle density can be attributed to different wetting behavior of the two oxide supports.
The chemical composition of the micellar NPs in different ambient conditions was investigated by NAP-XPS ( Figure  2a,b). The Fe 2p spectra of the "as deposited" samples measured at room temperature in UHV are also shown for comparison. The binding energy (BE) of the Fe 2p 3/2 core level (thereafter only used in the discussion) at 711.4 eV (for Fe/SiO 2 ) and 711.0 eV (for Fe/Al 2 O 3 ) and a weak satellite ("shake up") signal at around 719 eV are characteristic for Fe(III) species, as in hematite Fe 2 O 3 . 31−33 These species are formed during Fe NP synthesis and subsequent O 2 -plasma treatment. Besides the Fe-related signals and those from the bare oxide support, XPS spectra revealed adventitious carbon ( Figures S2 and S3) from the sample transfer into UHV through the air, which was removed by oxidation in 1 mbar of O 2 at 400°C. The latter mimics the calcination procedure commonly used for the preparation of powder catalysts. The oxidation only slightly alters the Fe 2p lines, indicating that Fe in the "as prepared" samples is predominantly in the highest oxidation state 3+.
In the next step, the samples were exposed to 1 mbar of H 2 at 400°C. This treatment is usually considered as an "activation" step for CO 2 hydrogenation catalysts. 2 On the Fe/SiO 2 sample, the Fe 2p 3/2 peak shifts by 1.6 eV toward lower BE (709.8 eV) and a weak satellite signal appears at ∼715 eV, thus indicating the formation of Fe(II) species, as in FeO. In addition, a new state appears as a shoulder at ∼706.6 eV (Figure 2a), which is characteristic of the metallic Fe(0) state. Therefore, Fe NPs on SiO 2 undergo reduction of Fe(III) to Fe(II) and partially to Fe(0), in general agreement with the reduction behavior of iron oxides (see, for instance, refs 34 and 35), commonly described as an Fe 2 O 3 → Fe 3 O 4 → FeO sequential transformation at these moderate temperatures. According to temperature-programmed reduction data, 34 the full reduction to Fe(0) only occurs at temperatures as high as 650°C, although the complete reduction of Fe 2 O 3 up to metallic Fe was reported after exposure to 0.2 mbar H 2 at 450°C . 36 When the sample is exposed to the CO 2 + H 2 reaction mixture (1:3, 1 mbar in total) at 300°C, the intensity of the Fe(0) signal decreases, suggesting relative instability of the metallic state under reaction conditions. Since XPS spectra of metallic Fe and Fe carbides are similar and thus difficult to differentiate, 37,38 we also inspected the C 1s region (Figure S2), which only revealed traces of carbon at BEs around 284 eV, distinctly different from "carbidic" carbon BEs at 283 eV and below. 37 We therefore conclude that the Fe(0) phase, formed during activation in pure hydrogen, becomes oxidized in CO 2 hydrogenation reaction conditions.
The same type of experiments performed on the aluminasupported NPs revealed a different behavior (Figure 2b). When exposed to 1 mbar of H 2 at 400°C, the main peak shifts to a lower BE by ∼0.4 eV only, substantially smaller than the 1.6 eV observed for Fe/SiO 2 . Although the shift points to the partial reduction, Fe(III) species still dominate the surface. In addition, no metallic state is observed, that is again at variance with Fe/SiO 2 . The Fe 2p spectrum remains virtually the same in the CO 2 + H 2 mixture at 300°C. The lack of intensity at 706−707 eV provides clear evidence for the absence of both, metallic and iron carbide phases on the Fe/Al 2 O 3 surface under CO 2 hydrogenation reaction conditions.
In summary, during the activation step the alumina support stabilizes Fe oxide in oxidation state 3+, whereas NPs supported on silica can be readily reduced in pure hydrogen to Fe(II) and partially to the metallic Fe(0) state. However, both catalysts showed neither metallic nor carbide phase under the CO 2 hydrogenation reaction conditions. Post characterization of the samples by AFM (Figure 1c,d) revealed neither particle sintering nor aggregation on both model catalysts.
In order to address whether the results from the NAP-XPS experiments are the result of kinetic limitations, due to being performed at lower pressure to catalytically relevant pressure, we performed XPS measurements on the samples exposed sequentially to 1 bar O 2 at 400°C, then 1 bar H 2 at 400°C, and finally, to 10 bar of the CO 2 + H 2 (1:3) reaction mixture at 300°C using a high-pressure cell. After each treatment, the cell was evacuated, and the sample cooled and transferred into the analysis chamber through vacuum (i.e., without exposure to air). XPS spectra were recorded in UHV at room temperature. The Fe 2p region is displayed in Figures 2c,d, while the corresponding Si 2p, Al 2p, O 1s, and C 1s spectra are shown in Figures S4 and S5.
Following the oxidation step, the Fe 2p spectra for both supports are similar and only revealed Fe(III) species. Note, however, that the Fe 2p 3/2 BEs observed at 1 bar (711.9 and 711.8 eV, for silica and alumina supports, respectively) are higher than those for the samples oxidized at 1 mbar (711.4 and 711.2 eV, respectively).
After the pure hydrogen treatment the Fe(0) component in the SiO 2 -supported NPs increases considerably at 1 bar and shifts to lower BE, i.e., from 706.6 to 706.0 eV, while the Fe(II) peak stays at 709.7 (±0.1) eV (Figure 2c). However, the Fe(0) signal disappears after the CO 2 hydrogenation reaction at 10 bar, and the spectrum becomes similar to that measured in the reaction mixture at 1 mbar at 300°C. A more pronounced H 2 pressure effect is observed for the Al 2 O 3 -supported NPs (Figure 2d): a strong Fe(0) signal shows up at 706.7 eV at 1 bar, which was not present in the NAP-XPS spectra at 1 mbar. The Fe(III) peak decreases and slightly shifts to higher BE (from 711.8 to 712.1 eV). Note also that the integral intensity of the Fe 2p signal is considerably reduced, in contrast to the silica-supported NPs. The loss of intensity can be indicative either of the NP shape change and/ or partial Fe migration into the alumina support. Nonetheless, after the CO 2 hydrogenation reaction, the signal at 706.7 eV decreases, whereas the Fe(III) signal gains in intensity. Since the inspection of the C 1s region revealed no carbon on this sample ( Figure S5), we can safely assign the 706.7 eV signal to the metallic Fe surface and not to iron carbide.
Thus, we conclude that the oxide support does play a role in the chemical transformations of the iron oxide NPs. Alumina stabilizes the Fe(III) oxide even in the hydrogen atmosphere, while silica is rather "passive" with respect to the Fe(III) reduction to Fe(II), which readily occurs on unsupported iron oxides.
The results presented above were obtained on supported NPs prepared by an inverse micelle encapsulation method, which needed exposure to oxygen plasma for ligand removal, thus leading to oxidation of the as-prepared Fe NPs. To examine whether the iron precursor or the sample preparation method alter the chemistry of the supported NPs in the CO 2 hydrogenation reaction, we investigated another planar model system, where Fe NPs were prepared by PVD of Fe onto the same substrates. Note, however, that the PVD preparation results in a much higher NP density on the support surface as compared to that of the samples synthesized from the micellar precursor (see also refs 37 and 39). The typical morphology of the PVD samples observed by AFM is shown in Figure S6. Figure 3 summarizes NAP-XPS results for the Fe 2p region in the PVD-prepared samples (other core levels of interest are presented in Figures S7 and S8). The spectra for "as deposited" Fe NPs on both supports, measured in UHV at room temperature, showed the Fe 2p 3/2 peak at 706.9 eV ( Figure  3a,b), henceforth used as the internal reference for the metallic Fe(0) state, although the value may slightly deviate depending on the particle size due to the so-called "final state" effects. 40 Upon exposure to 1 mbar of O 2 , the particles become fully oxidized giving rise to the Fe 2p 3/2 peak at 711.4 eV and the "shake up" satellite at ∼719 eV, both characteristic of Fe(III). In fact, the oxidation occurs even at room temperature (not shown here), in full agreement with refs 41 and 42.
Upon subsequent exposure to a hydrogen atmosphere (Figure 3a), the Fe 2p spectrum for the Fe/SiO 2 sample shifts to lower BEs by 1.4 eV (710.0 eV), and a weak satellite signal appears at ∼715 eV, both indicating the formation of the Fe(II) oxide. In addition, a new state shows up at 706.5 eV, suggesting the partial reduction to Fe(0). A similar behavior is also observed for Fe/Al 2 O 3 (Figure 3b), although the relative intensity of the Fe(0) signal at 706.9 eV is substantially higher than for Fe/SiO 2 . For both catalysts, the Fe(0) state disappears in the CO 2 + H 2 mixture. The spectra shift to higher BEs (710.6 ± 0.1 eV), and the Fe(II) satellite signal at ∼715 eV attenuates, all indicating the partial oxidation of Fe(II) to Fe(III) in the presence of CO 2 . Therefore, CO 2 acts as an oxidizing agent for metallic Fe even in excess of H 2 in the reaction atmosphere. Actually, the blank experiments using pure CO 2 showed that CO 2 readily oxidizes metallic Fe at room temperature (see also ref 43). Once oxidized, the surface In panels (c,d), the samples were exposed directly to the reaction atmosphere omitting oxidation and reduction treatments.
cannot be reduced to the metallic state by hydrogen in the reaction mixture at 300°C as it cannot be reduced even in pure H 2 at 400°C (see Figure 3a,b). In addition, the water produced via the r-WGS reaction pathway in the CO 2 + H 2 mixture is also expected to facilitate the oxidation of Fe.
The chemical state of Fe under the CO 2 hydrogenation conditions is rather independent of the sample prehistory. Direct exposure of the pure metallic NPs to the reaction atmosphere (Figure 3c,d) results in the same spectra as for the samples subjected to oxidation−reduction treatments prior to the reaction (Figure 3a,b). No "pre-history" effect is expected for the micellar samples as well, since the Fe-oxide phase dominates both the "as prepared" and "oxidized-reduced" NPs. Again, the formation of Fe carbide is not observed under reaction conditions, in full agreement with the NAP-XPS results obtained on the samples prepared using the micellar precursor.
Comparison of the Fe 2p NAP-XPS spectra obtained for the Fe/SiO 2 model catalysts prepared using two methods (see Figures 2a and 3a) revealed no difference in O 2 and H 2 atmospheres, and only a deviation (∼0.7 eV) in the peak position when exposed to CO 2 + H 2 , reflecting different Fe(II):Fe(III) ratios in the latter sample. For Fe/Al 2 O 3 , the most striking difference between the two preparation methods is the degree of reduction observed in pure hydrogen ambient, which is substantial for the PVD-prepared system and close to zero for the micellar samples. (In principle, this finding may be indicative of different NP/Al 2 O 3 interfaces formed upon deposition of metallic Fe by PVD, in the former case, and of polymer-covered Fe-oxide NPs, in the latter.) Nevertheless, the spectra recorded on these two Fe/Al 2 O 3 samples in the CO 2 hydrogenation atmosphere are very similar (710.6 vs 711.0 eV) and do not show Fe(0) species.
Therefore, all our NAP-XPS results, obtained on different oxide supported model catalysts, demonstrate that, independently of the initial state of the iron NPs before the CO 2 hydrogenation reaction, the NPs' surface under the reaction conditions contains a mixture of Fe(II) and Fe(III) species, with no indication of the carbide phase.
To link the results obtained on planar model systems to supported powder catalysts, we prepared catalysts by impregnation of high surface area silica and alumina powders with the same Fe colloidal precursor. The specific surface areas of the calcined catalysts (157 and 129 m 2 g −1 for Fe/Al 2 O 3 and Fe/SiO 2 , respectively) and adsorption isotherms (see Figure   S9) reflect the nonporous, but nanoparticulate structure of the oxide supports used. The average particle size (4.3 ± 1.5 nm) of the Fe NPs supported on SiO 2 determined by STEM (Figure 4a, Figure S10) is very close to that of observed by AFM on the planar samples ( Figure 1).
The electronic state of iron in the powder catalysts was inspected by NAP-XPS on the samples prepared by drop casting onto a flat silica substrate. Figure 4b displays the Fe 2p spectra, which are, in fact, similar to the spectra obtained on the planar systems (Figure 2a,b). No metallic Fe is detected in pure hydrogen at 400°C as well as in the CO 2 + H 2 reaction atmosphere at 300°C. All spectral changes can be explained in terms of the different Fe(II):Fe(III) ratios at the NP surface.
The nanocrystalline powder catalysts were tested in the CO 2 hydrogenation reaction under conditions typical for this reaction. 4 The main results are summarized in Figure 5 and    44 and Fe/ZrO 2 20 catalysts prepared by the incipient wetness impregnation. Our catalysts primarily produce CO and CH 4 , with Fe/Al 2 O 3 showing higher activity for methane as compared to Fe/SiO 2 . However, a major difference in the product distribution is the olefin/paraffin ratio, which is substantially higher on the silica-supported catalyst, showing rather a complex pattern of the hydrocarbon products ( Figure  5b, see also Figures S11−13).
The catalysts were further investigated by XAS in situ. In particular, the X-ray absorption near edge structure (XANES) allows quantitative determination of the Fe oxidation state in situ. 11,16,45,46 XANES spectra of the "as prepared" and subsequently calcined samples (Figure 6c) match well those of goethite FeOOH and hematite Fe 2 O 3 with iron in the 3+ state, i.e., in full agreement with the XPS results ( Figure 2). Considerable changes are observed upon heating to 400°C in the H 2 /He (1:4) atmosphere, with the spectral evolution proceeding differently for the two catalysts (Figure 6a,b). In particular, the shoulder at ∼7115 eV at the onset of the FeKedge, which develops with time and closely resembles that in the metallic Fe foil, is more intense in the Fe/SiO 2 than Fe/ Al 2 O 3 sample. Linear combination fitting analysis of the XANES spectra using an Fe foil, FeO and FeOOH compounds as references (see the representative fits in Figure S14) revealed that the "as-prepared" catalysts contain mostly Fe(III) species, with a small contribution of Fe(0) (Figure 7). During heating in H 2 /He, both samples become partially reduced, and Fe(III) species transformed into Fe(II) and Fe(0). However, the fraction of metallic iron in Fe/Al 2 O 3 is significantly lower than in Fe/SiO 2 (65 and 85% (±4%), respectively), indicating that Fe-oxide NPs on the alumina support are more resistant toward reduction in hydrogen, in agreement with the NAP-XPS results (Figure 2). EXAFS spectra of the Fe K-edge are displayed in Figure 8b. The main characteristics of the Fe coordination sphere derived from the EXAFS spectra are summarized in Table 2 (all fitting parameters are presented in Table S1). EXAFS data for the "as  Table 2) confirmed that the main peak in the FT-EXAFS of the reduced samples is dominated by the two nearest Fe−Fe coordination shells in bcc-like Fe, in addition to Fe−O bonds, suggesting that Fe reduction is not complete, in agreement with the above-presented XANES results.
The Fe−O coordination numbers are higher, and concomitantly the Fe−Fe coordination numbers are lower in the reduced Fe/Al 2 O 3 sample as compared to reduced Fe/SiO 2 , which is in agreement with the higher contribution of oxidized species remained in the alumina supported catalysts upon To monitor the state of the catalysts under reaction conditions, we performed XANES measurements in 10 bar of the CO 2 + H 2 (1:3) reaction mixture at 300°C. In addition, both XANES and EXAFS spectra were measured after reducing the pressure to 1 bar and cooling the samples to room temperature (rt). XANES (Figure 6c and Figure 7) and EXAFS ( Figure 8 and Table 2) data all showed no significant changes of the Fe oxidation state when compared to the reduced NPs. Note, however, the decrease in the Fe−Fe coordination numbers observed by EXAFS on the spent catalysts (1 bar, rt), which can be attributed to changes in the catalyst morphology.
Importantly, XANES spectra of the reduced samples and those in the reaction atmosphere differ significantly from that of a Fe 3 C carbide reference. This is most clearly seen in the pre-edge region (zoomed in the inset in Figure 6c), where the signal intensity in the reduced catalysts is substantially higher than that of the Fe 3 C reference, allowing us to differentiate metallic and carbide phases. 11 Note also that the EXAFS oscillations in k-space (see Figure 8a) for the reduced samples match well those of the metallic Fe foil and are clearly different from iron carbide Fe 3 C.
Therefore, an iron carbide phase was not observed under reaction conditions, thus fully supporting the XPS-based conclusion about the absence of Fe−carbide surface layers in the model catalysts. However, there is a difference between the XAS and the XPS results regarding the state of the catalyst in the CO 2 + H 2 mixture. While XAS revealed no significant changes if compared to the reduced catalysts as far as the oxidation state of Fe in the NPs is concerned, and the chemical composition of the catalyst is described as a mixture of Fe(II) and Fe(0) species, XPS data showed the reoxidation of the catalyst so that the Fe(0) signal disappears, and Fe(II) is partially oxidized to the Fe(III) state. Such a discrepancy cannot be explained solely by the different morphology of the oxide supports investigated by XAS and XPS (nanocrystalline powder vs planar substrates, respectively), since NAP-XPS spectra revealed very similar spectral evolution of the Fe 2p state in the micelles deposited on both flat and powder supports (cf. Figures 2a,b and 4b). However, XPS is more surface sensitive as compared to XAS that probes primarily the bulk phase of the NPs. Indeed, for the Al Kα X-ray source used, the escape depth of the Fe 2p photoelectrons in Fe and Fe oxides is about 1.2 and 1.6 nm, respectively. Therefore, the difference in the phase composition formed under reaction conditions determined by XAS and XPS most likely originate from different phases constituting the surface and the bulk of the NPs. Accordingly, the Fe NPs in CO 2 hydrogenation atmosphere expose an oxide-rich surface, while the particles' core is metal-rich.
Turning back to the CO 2 hydrogenation activity of our model catalysts ( Figure 5), we notice that the difference observed in the olefin/paraffin ratio for the two catalysts correlates with a considerably higher concentration of Fe(II) species at the NP surface in the Fe/SiO 2 catalysts as compared to Fe/Al 2 O 3 (both after reaction at 10 bar, see Figure 2c,d). It appears that Fe(II) leads to a larger amount of olefins in the hydrocarbon products, whereas Fe(III) to paraffins.

SUMMARY AND CONCLUSIONS
In this work, we studied the structure and composition of Febased catalysts, commonly employed for the hydrogenation of CO 2 toward high-order hydrocarbons. Specifically, we monitored the chemical evolution of oxide supported iron nanoparticles during calcination (oxidation) and reduction (activation) pretreatments as well as in the CO 2 hydrogenation reaction using NAP-XPS in the mbar range of pressures. The NAP-XPS results showed the formation of both, Fe(III) and Fe(II) oxides on the NP surface under CO 2 hydrogenation reaction conditions, independently of the iron precursor and the oxide support. It appears, however, that the alumina support stabilizes the higher oxidation state (3+) of iron. In contrast to previous studies performed on unsupported Fe catalysts, an iron carbide phase was never observed in our catalysts, even at catalytically relevant conditions, i.e., 10 bar of CO 2 + H 2 (1:3) at 300°C. The latter experiments, conducted on powder samples prepared by impregnation of nanocrystalline SiO 2 and Al 2 O 3 with Fe-containing micelles, revealed similar findings as those obtained by NAP-XPS. We associate the effects with the nanometer size of the Fe particles, which  Figure 4a,b. The red line shows the temperature profile and referenced to the y-axis on the right. The three separate columns show the results for the sample reduced in H 2 at 400°C and cooled to room temperature, under reaction conditions at 300°C, and after cooling the samples to room temperature. Red horizontal lines indicate the measurement temperature. undergo strong interaction with the oxide support. The combined XPS (which are more surface sensitive) and XAS results (probing both the surface and the bulk) favor the model of Fe NP consisting of a metal-rich core and oxide-rich shell under the CO 2 hydrogenation reaction conditions. The catalytic tests provide solid evidence for hydrocarbon production on the Fe oxide surface, which is considered active primarily for the r-WGS reaction producing CO. Interestingly, the olefin/paraffin product ratio was found to be dependent on the nature of the support, which influences the final oxidation state of the Fe NPs during the reaction as well as their structure. Since many recent FTS and CO 2 −FTS studies point to iron carbide as the key phase for the production of highorder hydrocarbons (C 5+ ), the iron oxide catalysts presented here remain promising for light hydrocarbon production. In this respect, "surface-science" oriented studies in combination with operando XAS measurements provide a playground for the rational design of Fe-based catalysts selective for high valueadded light olefins.
Particle size distribution; XPS spectra of Si 2p, Al 2p, O 1s and C 1s regions; N 2 adsorption isotherms; TEM images of Fe/SiO 2 ; GC/MS results of reaction products; fitting and structure parameters for Fe K-edge XANES and EXAFS spectra (PDF) Figure 8. Fe K-edge EXAFS spectra (a) and their Fourier transforms (b) for micellar Fe NPs supported on SiO 2 and Al 2 O 3 (as prepared, after reduction in hydrogen, and after reaction in CO 2 + H 2 (1:3)) all acquired in the corresponding gas atmosphere after cooling the samples to room temperature. The spectra are offset for clarity. Spectra for reference compounds (Fe foil, Fe 3 C, and FeOOH) are shown for comparison. Results of EXAFS data fitting are shown in (b). The error bar is given in parentheses.