Harnessing strong metal–support interactions via a reverse route

Engineering strong metal–support interactions (SMSI) is an effective strategy for tuning structures and performances of supported metal catalysts but induces poor exposure of active sites. Here, we demonstrate a strong metal–support interaction via a reverse route (SMSIR) by starting from the final morphology of SMSI (fully-encapsulated core–shell structure) to obtain the intermediate state with desirable exposure of metal sites. Using core–shell nanoparticles (NPs) as a building block, the Pd–FeOx NPs are transformed into a porous yolk–shell structure along with the formation of SMSIR upon treatment under a reductive atmosphere. The final structure, denoted as Pd–Fe3O4–H, exhibits excellent catalytic performance in semi-hydrogenation of acetylene with 100% conversion and 85.1% selectivity to ethylene at 80 °C. Detailed electron microscopic and spectroscopic experiments coupled with computational modeling demonstrate that the compelling performance stems from the SMSIR, favoring the formation of surface hydrogen on Pd instead of hydride. Strong metal–support interactions (SMSI) are effective in tuning the structures and catalytic performances of catalysts but limited by the poor exposure of active sites. Here, the authors develop a strategy to engineer SMSI via a reverse route, which is in favor of metal site exposure while embracing the SMSI.

S upported metal catalysts have long been recognized as the most important group of heterogeneous catalysts for fundamental investigations and modern chemical industries [1][2][3][4] . Conventionally, these catalysts are synthesized by anchoring the active metal nanoparticles (NPs) onto certain high-surface-area supports to increase the dispersion of catalytically active sites and stabilize the metal against leaching [5][6][7] . Subsequently, the metal-support interface is constructed. Such an interface provides synergistic properties to regulate catalysis by modifying the electronic (charge transfer between the metal sites and the support) and/or geometric (decoration or coverage of metal sites by the support) parameters, and also by modulating the reaction pathways, e.g., lattice oxygen in oxide supports may directly participate in catalytic reactions 7 ; multicomponent interfaces can enable tandem reaction pathways that do not exist on singlecomponent active sites 8,9 .
As a classic prototype in metal-support interactions, the strong metal-support interaction (SMSI) has been defined as the encapsulation of NPs, usually group VIII metals, by partially reduced oxide supports during high-temperature hydrogen (H 2 ) treatment 10,11 . Since the very first discovery of SMSI by Tauster et al. [12][13][14] , SMSI has been widely exploited to tune catalytic performances of group VIII NPs by engineering geometric and/or electronic structures of these metal sites. For example, the adsorption of H 2 or CO on Pd was extremely suppressed upon the formation of SMSI (refs. 13,15 ), suggesting that the active metal sites were largely covered by support, which altered the geometric ensembles and improved the thermal stability of Pd catalysts. Meanwhile, because the reducible oxide support, e.g., TiO 2 , Co 3 O 4 , CeO 2 , and Nb 2 O 5 , is partially reduced to the structure with a nonstoichiometric oxygen concentration during the reductive annealing, electron transfer between metal NPs and oxide supports was detected [16][17][18][19] . Under extreme conditions, the formation of intermetallic structure of the supported metal and metal cations in the supporting oxide was observed 20,21 .
Despite these fascinating interfacial properties in SMSI, the formation of SMSI is restricted to specific combinations of elements, i.e., group VIII metals with high surface energy and transition metal oxides with low surface energy. Consequently, it is extremely challenging for some metals, e.g., Au, to manifest SMSI due to their low work function and surface energy 15,17,22 .
Efforts have been devoted in hope of expanding upon the conventional SMSI. One critical element in this pursuit is switching the high-temperature treatment in H 2 into other conditions and thereby changes the mechanistic pathways for the formation of SMSI. For example, Wang et al. reported SMSI formation between Au NPs and TiO 2 induced by melamine under an oxidative atmosphere. With the formation of SMSI, the Au NPs were encapsulated by a permeable TiO x thin layer, making the Au NPs ultrastable at 800°C (ref. 23 ). Xiao et al. reported a wet chemistry approach to construct SMSI in aqueous solution at room temperature, which was realized by engineering redox interactions between metals and supports. This strategy was applicable to Au, Pt, Pd, and Rh (ref. 15 ). Christopher et al. developed a strongly bounded-adsorbate-mediated strategy to construct SMSI between Rh and TiO 2 through high-temperature treatment in the mixture of CO 2 and H 2 (ref. 24 ). Zhang et al. engineered the SMSI between Au NPs and hydroxyapatite by treating the Au NP-hydroxyapatite composite in the air at high temperatures 17 . Although progress has been made in expanding the boundaries of SMSI, one inevitable issue associated with the conventional SMSI is that upon high-temperature treatment the encapsulation process immediately and uncontrollably takes place, resulting in limited exposure of active sites 25 . In the ideal scenario, the oxide coverage on the metal surface needs to be thin and permeable to small molecules, while still fully encapsulating metal NPs to prevent the dissolution, disintegration, and aggregation of active sites during catalysis.
We recently reported that voids and cavity space can be developed in metal-metal oxide core-shell NPs in response to H 2 treatment at 200°C (ref. 26 ). This observation combined with the current issues in conventional SMSI motivated us to develop alternative routes to metal-support interactions. Here, we denote this type of structural rearrangement as the strong metal-support interaction via a reverse route (SMSIR). Specifically, we start from the final morphology of SMSI (full encapsulation) and end in the intermediate state with partial exposure of metal sites (Fig. 1). As a proof of concept, we demonstrate that the core-shell Pd-FeO x NPs can be restructured into a porous yolk-shell structure after optimized reductive annealing (Pd-Fe 3 Fig. 4), characteristic peaks at 2θ = 40.1°with very low intensities were detected, which can be assigned to the (111) peak of face-centered cubic (fcc) Pd. No additional peaks in the XRD patterns can be found, indicating the amorphous nature of iron oxide shell in both pristine Pd-FeO x and Pd-Fe 3 O 4 -A. On the contrary, in the XRD pattern of Pd-Fe 3 O 4 -H, the intensity of Pd (111) peak increases remarkably, and a series of characteristic peaks at 2θ = 30.6°, 35.9°, 43.5°, 53.9°, 57.3°, 63.0°, and 74.3°are clearly observed, which can be assigned to (220), (311), (400), (422), (511), (440), and (533) lattices of γ-Fe 3 O 4 . The XRD characterization indicates that annealing in the reductive atmosphere may facilitate the spatial redistribution of grains in the oxide shell, and promotes the crystallization of Pd and iron oxides, consistent with our previous report 26 .
The aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images of the Pd-Fe 3 O 4 -H in Fig. 2a show that the core-shell structure of pristine Pd-FeO x NPs evolved into a unique porous yolk-shell structure after reductive annealing at 300°C.  Supplementary Fig. 6). It is known that the reducible metal oxides can be partially reduced after high-temperature treatment in H 2 (ref. 28 ). H 2 reacts with these oxides to produce water and generate oxygen vacancies in the oxide matrix. This process can be further facilitated by platinum-group metal NPs supported on those oxides through a H 2 spillover process 29,30 . In the meantime, the crystallization of oxide shell could promote the rearrangement of  Because Pd-Pd and Pd-Fe bond lengths are similar, it is hard to visualize the difference between these two bonds with Fourier transform results. In this regard, the wavelet transform (WT) EXAFS as a powerful technique was employed to distinguish these two bonds in our samples. It can be clearly seen from  (Fig. 4d, detailed optimizing process see Supplementary Fig. 11). This result may shed some light on the formation mechanism of this unique porous yolk-shell structure. Evidently, the reaction between H 2 molecules and O atoms in the Fe 3 O 4 could generate oxygen vacancies in the structure upon evaporating the produced water. Meanwhile, the crystallization of the oxide shell and the new Fe-Pd bond formation could promote the rearrangement of oxide lattice and the mobility of Pd atoms, expanding the atom vacancies and developing cavity space in the structure.
XPS and DRIFTS investigations. To investigate the charge transfer with the formation of SMSIR, X-ray photoelectron  In the high-resolution Pd 3d XPS of Pd-FeO x NPs, only a Pd 3d 5/2 peak at 335.4 eV, being assigned to metallic Pd, was found 31,32 . When the Pd-FeO x NPs were treated in air at 300°C, an additional Pd 3d 5/2 peak at 336.8 eV assigned to PdO, emerged in Supplementary Fig. 12. This observation is consistent with our XAFS results that the Pd in Pd-Fe 3 O 4 -A possesses an oxidized feature to some extent. Furthermore, a Pd 3d 5/2 shoulder peak at 336.2 eV in the high-resolution Pd 3d XPS of Pd-Fe 3 O 4 -H was detected and assigned to the positively charged Pd (Pd δ+ ) 33 , which is originated from intercalation of Pd into Fe 3 O 4 matrix, leading to the strong interactions between Pd and Fe to form Pd-Fe bond 34  The CO DRIFTS was further carried out. During the test, we found that the CO adsorption peak was weak. We, therefore, subtracted the pure gas-phase signal from each data set. As shown in Supplementary Fig. 14, a peak at~2153 cm −1 was detected both in the CO DRIFTS of Pd-Fe 3 O 4 -H and Pd-Fe 3 O 4 -A, which is assigned to Fe 3+ -CO (ref. 35 ). Due to the core-shell morphology of Pd-Fe 3 O 4 -A where Pd is fully encapsulated by Fe 3 O 4 , no obvious peak was detected in the CO DRIFTS of Pd-Fe 3 O 4 -A. In contrast, a very weak peak at 2102 cm −1 can be seen in the CO DRIFTS of Pd-Fe 3 O 4 -H, which is assigned to the linear CO adsorption on metallic Pd (ref. 36 ). More interestingly, an additional peak at 2134 cm −1 can be found. Compared with the linear CO adsorption on metallic Pd, this blueshifted peak is assigned to linear CO adsorption on positively charged Pd (CO-Pd δ+ ) 37 .  Supplementary Fig. 17 shows that the intensity of CO-Pd δ+ became weaker than that in the CO DRIFTS of Pd-Fe 3 O 4 -H in Supplementary Fig. 14 Catalytic performance. The semi-hydrogenation of C 2 H 2 to C 2 H 4 is an important reaction in industrial purification of the C 2 H 4 stream produced from naphtha cracking. Pd-based catalysts are mostly used for this reaction with a consensus that the selectivity is sensitive to the structure of the catalyst 38 . H 2 molecules that are weakly adsorbed onto the Pd surface to form surface-H and C 2 H 2 molecules that are strongly adsorbed can lead to the production of C 2 H 4 , while the formation of hydride   (Fig. 5b) demonstrate that the conversion of C 2 H 2 increases with the increment of reaction temperature, while the selectivity toward C 2 H 4 shows the opposite trend. To comprehensively compare the catalytic performance between the Pd-Fe 3 O 4 -H catalyst and previously reported values, the turnover frequency (TOF) was calculated based on the dispersion of Pd (obtained from H 2 -pulse chemisorption in Supplementary Table 8). The TOF of Pd-Fe 3 O 4 -H was 6.46 s −1 ,~100-fold higher than those of a series of state-of-the-art single-atom catalysts at 80°C (Supplementary Fig. 18), indicating that the Pd-Fe 3 O 4 -H demonstrated compelling catalytic performance for semi-hydrogenation of C 2 H 2 . Stability tests of the Pd-Fe 3 O 4 -H catalyst were further carried out under both the high and low conversion rates (Fig. 5c, Supplementary Fig. 19). Both results show that the Pd-Fe 3 O 4 -H catalyst was remarkably stable in the semi-hydrogenation of C 2 H 2 to C 2 H 4 , which could be originated from the SMSIR between Pd and Fe 3 O 4 .
The formation of hydride in Pd-based catalysts is temperaturesensitive, and it dominates the total hydrogenation of C 2 H 2 (refs. 38,39 ). Hence, the dispersion of Pd is determined by H 2pulse chemisorption (Supplementary Table 8) at various temperatures to examine the formation of hydride. For the reference Pd NPs (commercial 5 wt.% Pd/Al 2 O 3 ), the dispersion was determined to be 7.6% at −130°C (cold bath of isopentane and liquid N 2 ). The corresponding particle size was calculated to be 14.8 nm. However, H 2 uptake on Pd NPs increased significantly at 35°C, and the estimated particle size decreased to 1.6 nm. This discrepancy can be attributed to the substantial formation of hydride on Pd NPs at higher temperature that interferes with the estimation of particle size 40 . In contrast, the Pd-Fe 3 O 4 -H sample demonstrated a dispersion of 26.7 and 24.4% at −130 and 35°C, respectively. The corresponding particle sizes were calculated to be 4.2 and 4.6 nm, in agreement with the Pd core size from STEM investigations (Fig. 2). These observations indicate that the formation of hydride may be effectively inhibited in our Pd-Fe 3 O 4 -H catalyst with SMSIR, leading to a superior selectivity toward semi-hydrogenated products in the catalytic investigations.
Control experiments. A series of control samples obtained at different treating temperatures (T200, T300, and T400) were prepared (see "Methods" section) to determine the optimized condition for the formation of SMSIR. Here, the T300 stands for the Pd-  Fig. 30). The corresponding structures are summarized here: in the pristine core-shell Pd-FeO x sample, the core and shell were metallic Pd 0 and amorphous Fe 3 O 4 . When the sample was treated in the air at high temperature (Pd-Fe 3 O 4 -A), the core-shell structures maintained with no obvious formation of voids. When the Pd-FeO x NPs sample was treated in H 2 at different temperatures, the core and shell crystallized into Pd 0 and Fe 3 O 4 , respectively. As a result,  To highlight the role of SMSIR in tuning the conversion and selectivity of C 2 H 2 semi-hydrogenation, both sets of control samples were employed in the C 2 H 2 semi-hydrogenation reaction. As shown in Supplementary Figs. 37 and 38, the Pd-Fe 3 O 4 -H, i.e., T300 and ST2, demonstrates the best catalytic performance. This result further reveals that the optimized ST and treating condition are essential to the formation of SMSIR for the promoted semihydrogenation of C 2 H 2 to C 2 H 4 . Based on the structures of the catalysts, the different catalytic outcomes can be attributed to the following factors: (1) regarding the effect of annealing temperatures, all samples possess a similar Pd size, indicating that the difference of catalytic performance is not originated from the difference of particle sizes (Supplementary Fig. 39). In the T200sample, there are fewer voids in the oxide shell and the T200 remains to be a core-shell structure, resulting in poor exposure of Pd active sites with limited activity. In the case of T400 sample, the core-shell structure is completely destroyed, and therefore the formation of hydrides turns to be favorable because of the loss of core-shell structural confinement; (2) for the effect of ST, a thicker shell in ST3 makes the Pd active sites less exposed. However, when the shell becomes too thin as the case in ST1, the structure cannot maintain a fully-encapsulated state, but rather more like a heterostructure with some iron oxide islands on Pd NPs. Consequently, Pd domains tend to form hydrides due to the lack of core-shell structural confinement effect.
Reaction mechanism. The reaction kinetics were further explored to understand the underlying mechanisms. As shown in Supplementary Fig. 40, the reaction order over C 2 H 2 is calculated to be −1 (up to 2.5% atm partial pressure), roughly in agreement with Monnier's work with~−0.7 reaction order 41 , indicating the strong adsorption of C 2 H 2 on the surface of Pd in the Pd-Fe 3 O 4 -H catalyst. There exists a debate regarding the H 2 reaction order. In general, the reaction order varies from~0. 5 (refs. 42,43 ),~1 (refs. 44,45 ), and up to~1.6 (ref. 46 ). In our work, we found the reaction order of H 2 to be~2 (up to 10% atm partial pressure). Such a positive dependence on H 2 partial pressure indicates a much weaker H 2 adsorption than previous studies. The temperature dependence of the Pd-Fe 3 O 4 -H sample was investigated at 1.2%/6% atm partial pressure of C 2 H 2 /H 2 ( Supplementary  Fig. 41). The apparent activation energy was found to be~52.7 kJ mol −1 , in good agreement with Monnier's 12.1 kcal mol −1 (ref. 41 ) and Zhang's 52 kJ mol −1 (ref. 45 ).
The inelastic neutron scattering (INS) spectra of H 2 adsorption on Pd-Fe 3 O 4 -H and bulk Pd were presented in Fig. 6. The signal of H 2 -sorption behavior in Pd-Fe 3 O 4 -H is totally different from that in bulk PdH x . In bulk PdH x sample, an evident signal of hydride was detected, while in Pd-Fe 3 O 4 -H, the signal of hydride was very weak 47,48 . The profile of the peak at 500 cm −1 reflects the status of the hydride. Specifically, the sharp peak followed by a shoulder as seen in bulk PdH x is due to certain dispersion relation of optical phonons in 3D space, which results in this particular distribution of phonon states. When hydride is only formed at or near the surface, the 3D network is lacking, leading to the broad bump in the spectrum of our Pd-Fe 3 O 4 -H sample. The result indicates that only surface-H formed during the reaction process, consistent with our H 2 -chemisorption results.

Discussion
In this work, we reported a strategy to engineer the SMSI between Pd and Fe 3 O 4 by using core-shell NPs as a building block through a reverse process of the formation of conventional SMSI, denoted as SMSIR. With the formation of SMSIR, the core-shell Pd-FeO x NPs was restructured into a unique porous yolk-shell structured Pd-Fe 3 O 4 -H, in favor of the exposure of Pd active sites. The Pd-Fe 3 O 4 -H with SMSIR demonstrated excellent catalytic performance in semi-hydrogenation of C 2 H 2 to C 2 H 4 with 100% conversion, 85.1% selectivity, and a high TOF of 6.46 s −1 at the reaction temperature as low as 80°C. XAFS investigations along with DFT simulations verified that the Pd atoms intercalate into the Fe 3 O 4 matrix and form strong interactions. The electron transfer was probed by CO DRIFTS and XPS, suggesting that with the formation of SMSIR, electrons partially transfers from Pd to Fe 3 O 4 shell. The optimized ST of Pd-FeO x NPs and annealing temperature were found to be essential to the formation of SMSIR. Detailed mechanistic investigations indicated that the SMSIR in Pd-Fe 3 O 4 -H alleviates the strong chemisorption of H 2 on Pd sites, prevents the formation of hydride, and consequently leads to a superior selectivity toward C 2 H 4 . This work not only develops a high-performance catalyst for semi-hydrogenation of C 2 H 2 but also provides an approach for the construction of effective catalytic structures based on unconventional SMSI.
Preparation of 4 nm Pd NPs. The Pd NPs were prepared by a modified method from previous work as following 27 : 70 mg of Pd(acac) 2 was mixed with 15 mL of OAM in a 100 mL of four-neck flask under stirring. The mixture was then heated to 80°C at a ramping rate of 5°C min −1 and kept for 1 h under the protection of N 2 . A total of 0.5 mL of TOP was added to the solution. The mixture was further heated to 250°C at a ramping rate of 5.6°C min −1 , and kept at this temperature for another 1 h before cooling down to room temperature. Subsequently, the mixture was transferred to a 50 mL of centrifuge tube, and 30 mL of ethanol was added. The Pd NPs were separated by centrifugation at 4656 × g for 10 min. Then, the Pd NPs were redispersed in 10 mL of hexane, and precipitated and washed by adding 30 mL of ethanol for two times. Finally, the Pd NPs were dispersed in 10 mL of hexane for further use.
Preparation of Pd-FeO x NPs. A total of 110 mg of Fe(acac) 3 and 20 mL of OAM were added to a 100 mL four-neck flask. The mixture was heated to 90°C at a ramping rate of 5°C min −1 in N 2 . Subsequently, 12.5 mg of Pd NPs were added, followed by heating to 250°C and kept there for 30 min. Afterward, the reaction temperature was raised to 300°C and kept there for another 30 min before naturally cooling to room temperature. Then, 30 mL of ethanol was added to precipitate the Pd-FeO x NPs, and then centrifuged at 4656 × g for 10 min. The Pd-FeO x NPs was redispersed in 10 mL of hexane and washed by 30 mL of ethanol for two times. Characterization. The powder X-ray diffraction (XRD) patterns were collected on a PANalytical X'Pert Pro MPD diffractometer using an X'Celerator RTMS detector. HAADF-STEM and HR-STEM were performed on a Nion Ultra STEM 100 (operated at 100 kV). EELS spectra were collected on a high-resolution Gatan-Enfina ER with a probe size of 1.3 Å. TEM and high-angle annular brightfield scanning transmission electron microscopy (HAABF-STEM) were obtained on a Hitachi HD-200 with bright-field STEM detector operating at 200 kV.
The dispersion of the Pd was evaluated via pulse H 2 -Chemisorption with an Altamira Instruments (AMI-300) system. Before the measurements,~100 mg catalyst was pretreated at 550°C for 3 h under 50 sccm of Ar, followed by cooling down to desired temperature (i.e., −130 and 35°C) under the same flow. Then pulses of 4% H 2 /Ar from a sample loop with a defined volume (~0.5 cc) were injected by switching a six-way valve until the eluted peak area of consecutive pulses was constant. The dispersion of Pd was calculated from the volume of H 2 .
INS experiments were performed at the VISION beamline of the Spallation Neutron Source, Oak Ridge National Laboratory. The Pd-Fe 3 O 4 -H sample was first treated under vacuum at 600°C for 12 h. It was then loaded in an aluminum sample holder in a helium glovebox. The sample holder was attached to a gasloading sample stick connected to a gas panel. The blank sample was first measured at −268°C for 3 h to collect baseline spectrum. H 2 gas was then introduced in situ at −238°C, followed by heating of the sample to −98°C for reaction. The system was then cooled back to −268°C to measure the reacted spectrum. The difference spectrum (reacted minus baseline) shows the signal associated with the hydride species formed during the reaction. The CO DRIFTS results were obtained on a Nicolet 670 Fourier Transform Infrared Spectrometer with an MCT detector by the following process: each sample (~15 mg) was loaded and then pretreated at 200°C under Ar for 30 min. Afterward, the sample was cooled down to −120°C to conduct CO adsorption. When the temperature reached −120°C, the background was measured and then CO adsorption was conducted for 30 min as followed by desorption with Ar for 10 min (CO desorbed within 1 min after flow Ar). XPS characterization was performed on a PHI VersaProbe III scanning XPS microscope using a monochromatic Al K-alpha X-ray source (1486.6 eV). XPS spectra were acquired with 200 µm/50 W/15 kV X-ray settings and dual-beam charge neutralization. All binding energies were referenced to Al 2p peak at 74.8 eV.
Catalytic performance tests. The hydrogenation of C 2 H 2 was carried out in a tubular quartz reactor with a 0.25-inch diameter. In a typical run,~15 mg of catalyst was mixed with 150 mg of 60-80 mesh quartz sand and placed in the center of the reactor. The catalyst bed was held by quartz wool at both ends and the reactor was loaded in a vertical furnace (Carbolite Gero). The catalyst was purged with He for 30 min at a flow rate of 20 sccm prior to the reaction under room temperature. Then, the reactor was heated to the desired temperature (i.e., 30-80°C ), followed by feeding the gas mixture (i.e., 0.6 sccm C 2 H 2 , 3 sccm H 2 balanced with He) at a total flow rate of 50 sccm. The exit gas mixture was analyzed on-line by a ThermoStar Mass Spectrometry (Pfeiffer).
The conversion and selectivity were calculated by using Eqs. (1) and (2): Selectivity % ð Þ ¼ X C 2 H 4 ;out whereas in/out refers to the concentration measured in the inlet/outlet port. Reaction orders with respect to H 2 and C 2 H 2 were calculated by the differential method. The corresponding conversion is maintained below 20% to ensure a true kinetic regime. Apparent activation energy is calculated by the Arrhenius equation.
DFT calculation. The density functional theory calculations were performed with the Vienna Ab Initio Simulation Package (VASP) 49,50 . The on-site Coulomb interaction was included with the DFT + U method by Dudarev et al. 51 in VASP using a Hubbard parameter U = 3.8 eV for the Fe atom. The Perdew-Burke-Ernzerhof 52 functional form of generalized-gradient approximation was used for electron exchange and correlation energies. The projector augmented-wave method was used to describe the electron-core interaction 49,53 . A kinetic energy cutoff of 450 eV was used for the plane waves. A 3 × 2 × 1 sampling of Brillouin zone using a Monkhorst-Pack scheme was used 54 . A vacuum layer of 15 Å was added for the surface slabs along the z-direction; the slab contains a total of four layers, with the bottom two layers fixed in their bulk positions.
XAFS data collection and processing. Approximately 20 mg of sample was enclosed in a nylon washer of 4.953 nm inner diameter and sealed on one side with transparent "Scotch" tape. The sample was pressed by hand to form a uniform pallet, then sealed on the open side with a tape. XAFS investigation were performed ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16674-y at beamline 10ID-B of the Advanced Photon Source at Argonne National laboratory 55 . Spectra were collected at the iron K-edge (7112 eV) and palladium Kedge (24,350 eV) in transmission mode, with an iron and palladium foil as a reference for energy calibration, respectively. All spectra were collected at room temperature and ten scans were collected for each sample. All data were processed and analyzed using the Athena and Artemis program of the IFFEFFIT package 56 based on FEFF 6.0. Reference foil data were aligned to the first zero-crossing of the second derivative of the normalized μ(E) data, which was subsequently calibrated to the literature E 0 for each Fe K-edge and Pd K-edge. The background was removed, and the data were assigned a Rbkg value of 1.0 prior to normalizing to obtain a unit edge step. All data were initially fit with k-weighting of 1, 2, and 3 then finalized with k 3 -weighting in R-space. A fit of the Pd foil and Fe foil was used to determine S 0 2 for each sample. Structure models used to fit the data sets were obtained from crystal structure of iron oxide and DFT calculation. Structure parameters that were determined by the fits include the degeneracy of the scattering path (N degen ), the change in Reff, the mean square relative displacement of the scattering element(σ 2 i ), and the energy shift of the photoelectron(ΔE 0 ). k 3weighting in R-space. Initial fitting was conducted using crystal structure from crystal database. The simulated models were obtained from DFT calculation and scattering paths of selected scattered atom (Fe, Pd) were generated through FEFF calculation. The WT method was adapted for a quantitative analysis of the backscattering atom in the higher coordination shells with EvAX code 57 .

Data availability
The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.