Inverse iron oxide/metal catalysts from galvanic replacement

Key chemical transformations require metal and redox sites in proximity at interfaces; however, in traditional oxide-supported materials, this requirement is met only at the perimeters of metal nanoparticles. We report that galvanic replacement can produce inverse FeOx/metal nanostructures in which the concentration of oxide species adjoining metal domains is maximal. The synthesis involves reductive deposition of rhodium or platinum and oxidation of Fe2+ from magnetite (Fe3O4). We discovered a parallel dissolution and adsorption of Fe2+ onto the metal, yielding inverse FeOx-coated metal nanoparticles. This nanostructure exhibits the intrinsic activity in selective CO2 reduction that simple metal nanoparticles have only at interfaces with the support. By enabling a simple way to control the surface functionality of metal particles, our approach is not only scalable but also enables a versatile palette for catalyst design.

M etal particles play a key role in chemical transformations that require activation of H 2 or hydrogenation/dehydrogenation of substrates. In many cases, the metal particles provide only one step in the catalytic cycle. For instance, metals have low activity in CO 2 reduction because of weak CO 2 adsorption, whereas the polar surface of oxides readily adsorbs CO 2 but suffers from low activity for H 2 activation 1-3 . Thus, metal-oxide interfaces are much more effective because both the redox sites required to activate CO 2 and the metals providing active H 2 are in proximity. Challenges for maximizing such interfaces are stabilizing small metal particles on oxide supports [4][5][6][7] or forcing migration of oxides onto metal particles while avoiding harsh synthesis conditions [8][9][10][11][12] .
Inverse catalysts-oxides supported on metals-offer an attractive alternative to overcome the constraints of typical supported metal catalysts because reactants can bind to sites in the oxide overlayer, onto the metal domains, or at their interface. Typically, surface science research selects only well-defined inverse catalysts to provide a basic understanding of their adsorption and catalytic properties; however, advancing from this approach into the more complex conditions relevant to technical applications is essential [13][14][15][16] . In this regard, a major obstacle is encountered because typical surface science approaches for preparing inverse catalysts, such as reduction at high temperature 12 , deposition in ultrahigh vacuum 1,13 , and deposition at atomic layers 17 , are challenging to scale beyond certain models.
We report here a simple galvanic replacement approach for generating inverse FeO x /metal nanostructures. During galvanic replacement, one metal dissolves as a sacrificial template while a different metal ion in solution is reductively deposited onto the template. This process is driven by the differences of reduction potentials of the redox pairs, allowing a single, simple, and lowtemperature step for synthesis of nanostructures [18][19][20][21][22][23] . Following this, research has focused on preparing metals 18 , metal alloys 19,24 , oxides 21 , and metal-oxides 25,26 with controllable shapes. In our case, the solid support undergoing oxidation-hyperstoichiometric and sometimes referred to as cation-excess or partially reduced magnetite (Fe 3 O 3.7 ) 27 -supplies electron equivalents in the form of Fe 2+ enriched at the oxide surface, which reduce Rh 3+ or Pt 4+ , thereby depositing metal nanostructures (Eqs. (1)-(3)).
Rh 3þ þ 3e À ! Rh: ð1Þ We discovered that in addition to acting as sacrificial species, Fe 2+ dissolves, and adsorbs onto the as-formed metal particles as Fe (II)-oxyhydroxide. The surface property of the metal is thus greatly changed by the FeO x overlayer, endowing the nanostructure with the high density of active sites for CO 2 reduction that well-dispersed Rh particles have only at the interface with Fe 3 O 4 . This yields activity and selectivity for CO production significantly higher than well-dispersed Rh particles without FeO x overlayers. Our method demonstrates that the surface of metal nanoparticles can be manipulated by the sacrificial species during galvanic replacement, whereas galvanic replacement was previously thought to control only nanostructure morphologies.

Results and discussion
Identification of FeO x overlayer on Rh. We performed the synthesis by simply suspending Fe 3 O 3.7 ("Methods" section and Supplementary Fig. 1 for synthesis) in aqueous RhCl 3 solution (Fig. 1a) (Supplementary Fig. 3). This indicates that the Fe on Rh nanostructures have a lower average oxidation state (i.e., +2) than that in Fe 3 O 4 (+8/3). Rh K-edge X-ray absorption near edge structure (XANES, Supplementary Fig. 3) showed that the white-line of FeO x /Rh/ Fe 3 O 4 -fresh is similar to that of Rh foil. Linear combination fitting indicated that 77 mol.% of Rh is metallic ( Supplementary  Fig. 3 and Supplementary Table 1). This agrees well with the Rh extended X-ray absorption fine-structure (EXAFS) fitting showing that the Rh species have high Rh-Rh coordination (Supplementary Fig. 6 and Supplementary Table 2). The fitting of the spectra required a Rh-O path with a coordination number of 1.9 ± 0.5. Thus, 23 mol.% of Rh remains oxidized, probably because of its interaction with the FeO x species (Supplementary Table 1). The results indicated that inverse FeO x /Rh nanostructures were formed where Rh was reductively deposited while Fe(II)-oxyhydroxide species bind onto the Rh.
Mechanism for the formation of FeO x overlayer. We used the ferrozine method 28 to monitor changes of Fe 3+ and Fe 2+ concentrations during synthesis of FeO x /Rh/Fe 3 O 4 -fresh (Fig. 1c). Fe 2+ was released immediately after Fe 3 O 3.7 was dispersed in the solution of Rh 3+ , which is consistent with the acidic Fe 3 O 4 oxidation chemistry (Eq. (4)) 27 because Fe 2+ is much more soluble than Fe 3+ , and the solution is initially free of aqueous Fe 2+ upon first contact the Rh 3+ solution. The pH was observed to initially drift upwards from~4 to 5, which is consistent with consumption of protons during the release (see section of methods for pH changes).
The particle surface will be enriched in Fe 2+ during the Fe 2+ release into solution 29 , thereby maintaining a dynamic equilibrium 27 (Fig. 1c), we attribute the Fe(II)-oxyhydroxide coating on Rh particles to the dynamic equilibrium of the Fe 2+ release process (i.e., the reverse of Eq. (4)) 27,29 . To confirm the selective interaction of Fe 2+ with Rh, we contacted pre-formed Rh nanoparticles with solutions containing either Fe 2+ or Fe 3+ cations and analyzed the recovered particles (Fig. 1d). The syntheses of these reference materials are described in the section of methods. The EELS images showed that Fe 2+ species adsorb on the Rh surface to form a core-shell-like nanostructure (Fig. 1e), whereas Fe 3+ species precipitate as a segregated phase with only weak association with Rh (Fig. 1f).
Overall, the charge transfer of galvanic replacement consumes Fe 2+ supplied by Fe 3 O 4 for Rh 3+ reduction yielding Rh particles (Fig. 2a). In parallel, Fe 2+ released from the solid (Fig. 2b) adsorbs selectively on Rh (Fig. 2c). In order to verify the generality of our methodology to prepare inverse nanostructures, we also performed the galvanic replacement between the Pt 4+ cations and Fe 3 O 3.7 . The HAADF-STEM-EELS showed that FeO x species coat the Pt nanoparticles (Supplementary Fig. 7). Hence, during the synthesis of FeO x /metal nanostructures, the Fe 2+ is not only a sacrificial species as one expects from the galvanic replacement alone, but a key constituent for tuning the surface of the metal nanoparticles. The method offers many possibilities to tune the properties and structures of the final materials by controlling the rates of the individual processes taking place during the synthesis. Further  Catalytic improvement by the FeO x overlayer. We targeted CO 2 hydrogenation to test the activity of our inverse catalysts. Thus, we measured isotherms for CO 2 adsorption (Fig. 4c) (Fig. 4d). We used this normalization to reflect the surface of the catalysts that is potentially active, i.e., Rh with or without interactions with the support (note however, these trends are the same per mass of catalyst and mass of Rh). The selectivity to CO and the corresponding CO production rates are also higher on FeO x /Rh/Fe 3 O 4 than on Rh/Fe 3 O 4 (Fig. 4e, f). This highlights  the higher activity of the FeO x -coated particles than simple supported Rh particles. We analyzed the intrinsic activity of the materials not by normalizing rates to the fraction of exposed Rh (as determined from H 2 chemisorption) nor to the fraction of Rh covered by oxide species (Supplementary Table 8 and Supplementary Note). Instead, we considered that the uptake of CO 2 serves as titration of adsorption sites that can potentially produce CO (see the supporting information for more details). The rates of CO production normalized to the concentration of sites that chemisorb CO 2 were, e.g., 1657 and 1222 h −1 on FeO x /Rh/Fe 3 O 4 and Rh/Fe 3 O 4 , respectively, at 250°C. The similarity of these values, and of the activation energies for CO production (Fig. 4f), allows us concluding that the highly active and selective sites in both systems are similar. These sites, in view of the negligible activity of SiO 2 -supported Rh and pure Fe 3 O 4 (Supplementary Table 8) are undoubtedly identified as Rh-Fe 3 O 4 interfaces [30][31][32][33] . We also tested the FeO x /Rh nanoparticles (Fig. 1e) and the parent Rh nanoparticles in CO 2 reduction (Supplementary Table 9). The FeO x /Rh nanoparticles were one order of magnitude more active than the bare Rh nanoparticles. The bare Rh nanoparticles produced both CO and methane in equimolar concentrations, while FeO x /Rh nanoparticles selectively yielded CO. These observations further support our claim that the FeO x adlayers increase the activity for CO 2 conversion and the selectivity to CO. The FeO x /Rh nanoparticles, however, led to 1-2 orders of magnitude lower rate for CO 2 reduction than the inverse FeO x /Rh/Fe 3 O 4 catalyst, which highlights the role of the Fe 3 O 4 support, which maintains the FeO x /Rh nanoparticles separated.
The FeO x /Rh/Fe 3 O 4 inverse catalyst is also more productive than typical supported noble-metal nanoparticles and atomically dispersed Rh (Supplementary Table 10). Thus, leaving some exposed Rh on the surface of FeO x /Rh/Fe 3 O 4 does not lead to low activity because the surface behaves like Rh-Fe 3 O 4 interfaces.
In summary, this galvanic replacement approach to prepare inverse FeO x /metal nanostructures not only yields particularly compelling catalytic reactivity under real conditions but is versatile and easily scalable 13,17,34 . The ability to control the surface functionality of metal nanoparticles enables a palette for catalyst design via galvanic replacement. The presence of the oxide overlayer makes the metal much more efficient for activating CO 2 while maintaining its hydrogenation ability. That is, the whole surface of the metal particle functions as metal/oxide interface with redox sites for adsorbing CO 2 near metal domains that dissociate H 2 but with limited capacity to produce methane.   room temperature. The Fe 3 O 3.7 was then added to the solution and stirred for 7 h. The resulting material was separated, washed with water, and dried at 80°C overnight. The as-prepared material was calcined in air at 450°C with a ramping rate of 2°C/min. The inductively coupled plasma (ICP) analysis showed that the effective Rh loading in the final material was 0.37 wt%. Prior to the catalytic test, the sample was treated at 200°C in H 2 . The purpose of heat treatments is to remove the possible surface ligands and surface-oxidized Rh species that remained during the synthesis. A material containing Pt (FeO x /Pt/Fe 3 O 4 -fresh) was prepared by the same method with aqueous solution of H 2 PtCl 6 as the precursor and a preset Pt loading of 0.5 wt%.
The dynamic changes of the Fe 2+ and Fe 3+ concentrations in the aqueous fraction during the galvanic replacement synthesis (for FeO x /Rh/Fe 3 O 4 -fresh) were analyzed by the ferrozine method 28 . The suspension was centrifuged to isolate the aqueous fraction during the galvanic replacement (0, 0.1, 0.5, 1, 2, 3, 4, 5, 6, and 7 h). The pH values for the aqueous solutions increased slightly at first (from 4.0 to 5.0), and then decreased to~2.5. The resulting aqueous solutions were diluted in a 10 −2 M HCl solution and used for the analysis. The Fe 2+ can react with ferrozine to form a stable magenta complex which gives a maximum absorbance at 562 nm on an ultraviolet-vis spectrophotometer. The Fe 3+ fraction can be detected by reducing with hydroxylamine hydrochloride solution, stabilized in a buffer, and followed by complexing with ferrozine.
Synthesis of Rh/Fe 3 O 4 . The Rh/Fe 3 O 4 with a Rh loading of 0.5 wt% was prepared by a urea hydrolysis assisted deposition method. In a typical procedure, 9.95 g of Fe 3 O 4 were dispersed in 100 mL deionized water. Then, a 10 mL aqueous solution of RhCl 3 at a concentration of 5 mg Rh mL −1 was added into the suspension and rigorously stirred for 12 h at room temperature. An excess of urea (urea/[Rh] molar ratio = 60) was added to the suspension for deposition of Rh 3+ . The Rh 3+ can be deposited homogeneously and slowly with the help of urea hydrolysis in a hydrothermal condition (90°C) for 6 h. The resulting material was separated, washed with water, and dried at 80°C overnight. The as-prepared material was treated in air at 450°C with a ramping rate of 2°C min −1 . The ICP results suggested that the Rh loading was 0.37 wt%. Prior to the catalytic test, the sample was treated at 200°C in H 2 . A reference Rh/SiO 2 with the Rh loading of 0.5 wt% was also prepared by the urea hydrolysis deposition method, followed by the same treatments before catalytic test.
Synthesis of Rh nanoparticles (PVP method). In a typical procedure, Rh nanoparticles were synthesized following a polyol-based method. Rh nitrate (Rh amount 100 mg) was dispersed in 60 mL of ethylene glycol in the presence of a stabilizer (PVP) and heated under reflux for 6 h. The Rh nanoparticles then were washed with acetone and water eight times before used for model synthesis experiments.
Mixing of Rh 3+ and Fe 2+ cations in solution in the absence of solid. In a typical procedure, 0.25 mL RhCl 3 aqueous solution (5 mg [Rh] mL −1 ), 1 mL FeCl 2 aqueous solution (2 mg [Fe] mL −1 ), and 4 mL deionized water were mixed at room temperature and stirred for 7 h. This procedures were perfomed in a N 2 glove box.
Reaction of Rh 0 nanoparticle and Fe 2+ cations. After washing three times with deionized water, 1.25 mg Rh 0 nanoparticles were dispersed in 4 mL deionized water and mixed with 1 mL FeCl 2 aqueous solution (2 mg [Fe] mL −1 ) at room temperature. The resulting suspension then was stirred for 7 h. The Rh 0 nanoparticles immersed in Fe 2+ solution and parent Rh 0 nanoparticles were also diluted in SiO 2 as the reference samples (Rh loading of 0.5%) for handling and catalytic testing.
Reaction of Rh 0 nanoparticles and Fe 3+ cations. After washing three times with deionized, 1.25 mg Rh 0 nanoparticles were dispersed in 4 mL deionized water and mixed with 1 mL FeCl 3 aqueous solution (2 mg [Fe] mL −1 ) at room temperature. The resulting suspension then was stirred for 7 h.
Characterization. HAADF-STEM measurements were conducted with an aberration-corrected FEI Titan 80-300 STEM operated at 300 kV. EELS mapping and analysis were performed with aberration-corrected JEOL-ARM200F instrument operated at 200 kV. The instrument (Quantum 965) is capable of performing dual EELS experiment. The EELS mapping was performed in the STEM mode in the range of -50 to 500 eV for the zero-loss peak, 300 to 800 eV for the iron signal, and 2500 to 3500 eV for rhodium the signal maps. The zero-loss peak for zero-loss calibration was acquired in low loss spectrum images and aligned at 0 eV. The images and EELS data were analyzed and processed using Gatan Digital Micrograph software. The EELS maps were constructed by analyzing the Fe L 2,3 (~708 eV), Rh L 2,3 (~3004 eV), and Pt M 4,5 (~2122 eV) edge peaks after the background subtraction.
X-ray absorption spectroscopy measurements were conducted in sector 20 of the Advanced Photon Source operated by Argonne National Laboratory. A rejection mirror was used to reduce the effects of harmonics. The metal foil was placed downstream of the sample cell, as a reference to calibrate the photon energy of each spectrum. The EXAFS spectra were analyzed with the ATHENA (χ(k) oscillation background removal), FEFF9 (theoretical model calculation), and ARTEMIS software packages. The fits to the Rh K-edge EXAFS χ(k) data were weighted by k 2 and windowed between 1.5 Å −1 < k < 15.0 Å −1 using a Hanning window with dk = 1.0 Å −1 .
H 2 and CO 2 chemisorption experiments were conducted with a Micromeritics 2020 instrument. In a typical procedure, 100 to 200 mg of the sample was degassed at 100°C for 30 min, followed by in situ treatment at 200°C in H 2 and evacuation at 200°C for 30 min. Then, the temperature was decreased to 35°C under vacuum. Prior to the chemisorption experiments, the sample was further evacuated for 40 min. The adsorbates (H 2 or CO 2 ) were introduced into the system for the measurements of chemisorption isotherms. The first chemisorption isotherm was measured in the pressure range of 0-40 kPa at 35°C. The sample was evacuated after the first adsorption cycle and a second chemisorption isotherm was recorded. The CO 2 uptake on the parent Fe 3 O 4 has been subtracted for plotting and derivation of adsorption parameters. N 2 physisorption experiments at −196°C were performed on a Micromeritics 2020 instrument. The samples were degassed in vacuum at 200°C before the measurements.
TOF-SIMS was applied with a TOF-SIMS V spectrometer (IONTOF GmbH, Münster, Germany) equipped with a 25 keV bismuth cluster ion source, a 20 keV Ar n + , and a 2 keV Cs + /O 2 + sputtering ion sources. Prior to the TOF-SIMS experiments, the samples were deposited on an Au(111) substrate and exposed to ultrahigh vacuum overnight.
X-ray diffraction experiments were performed in a Philips X′pert Multi-Purpose Diffractometer equipped with a Cu anode (50 kV and 40 mA).
The elemental composition of samples was measured by ICP optical emission spectroscopy (Perkin Elmer 7300DV). Prior to the ICP experiments, the samples were digested in a mixture of HNO 3 /HCl/HF/H 2 O followed by H 3 BO 3 addition for extra HF treatment.
Reaction tests. The CO 2 reduction was performed in a flow reactor equipped with an online gas chromatograph (Agilent 7890B). In a typical procedure, prior to the catalytic test, 12 mg of 30-80 mesh catalyst (diluted with 50 mg SiC) was loaded into the reactor and treated at 200°C in 20 vol.% H 2 with a ramping rate of 2°C min −1 . After the reactor reached the target reaction temperature, a mixture of CO 2 , H 2 , and He with a total flow rate of 140 mL min −1 was fed into the reactor (CO 2 : H 2 :He = 7:28:105).
Correlations of coordination number and metal dispersion. The correlation between coordination number and metal dispersion was derived from the data in the reference (Supplementary Fig. 12) 35 . The relationship between the coordination number of metal-metal shell and the metal dispersion was derived based on two different shapes of metal particles (spherical and raft-like shapes).
Calculation of adsorption constant and monolayer coverage from isotherms. The adsorption constant and monolayer coverage were derived from the chemisorption isotherms where chemisorption is treated as a chemical reaction between the gas-phase molecule (A) and the site (*) for adsorption (Eq. (5)).
The adsorption parameters can be obtained from the linear form of Eq. (2) (Eq. (7)). In Eq. (7), θ A is the fractional coverage of the adsorption sites, P is the partial pressure of the adsorbate, V m is the volume of the monolayer, and K is the equilibrium adsorption constant.

Data availability
The source data underlying Figs. 1-4 are provided as a Source Data file. The other relevant data that support the findings of this study are available from the corresponding author upon request. Source Data are provided with this paper.