Operando NRIXS and XAFS Investigation of Segregation Phenomena in Fe‐Cu and Fe‐Ag Nanoparticle Catalysts during CO2 Electroreduction

Abstract Operando nuclear resonant inelastic X‐ray scattering (NRIXS) and X‐ray absorption fine‐structure spectroscopy (XAFS) measurements were used to gain insight into the structure and surface composition of FeCu and FeAg nanoparticles (NPs) during the electrochemical CO2 reduction (CO2RR) and to extract correlations with their catalytic activity and selectivity. The formation of a core–shell structure during CO2RR for FeAg NPs was inferred from the analysis of the operando NRIXS data (phonon density of states, PDOS) and XAFS measurements. Electrochemical analysis of the FeAg NPs revealed a faradaic selectivity of 36 % for CO in 0.1 M KHCO3 at −1.1 V vs. RHE, similar to that of pure Ag NPs. In contrast, a predominant selectivity towards H2 evolution is obtained in the case of the FeCu NPs, analogous to the results obtained for pure Fe NPs, although small Cu NPs have also been shown to favor H2 production.


Introduction
One of the most significant challenges in the field of catalysis is the development and optimization of experimental methods that allow the observation of ac atalyst while at work. [1][2][3] This is especially difficult when nanomaterials are considered under realistic conditions,a sfor example,w hen trying to understand atomic segregation phenomena and other structural/chemicalm odifications at liquid/solid interfaces during an electrochemical process. [4] An umber of established methods are not directly applicable under reaction conditions due to decreased electronic mean free paths, scattering or lack of sufficient spatial resolution or chemical sensitivity.The development of additional techniques suitable for operando implementation in an electrochemical environment is therefore as ubject of high interest. [5,6] Nuclear resonant inelastic X-ray scattering (NRIXS) is as ynchrotron radiation technique that is sensitive to vibrational states of the nuclei of Mçssbauer-active isotopes,most commonly 57 Fe. [7] It can be used to probe vibrational modes of molecular complexes of iron as well as phonons in solid state materials. [8] Because the vibrational modes of acrystal lattice depend heavily on its structure,itispossible to relate NRIXS data to structural and thermodynamic properties of amaterial. [9] Moreover,due to the isotope-specific detection, one can probe only as pecific region of as ample if enriched with the Mçssbauer-sensitive element, while the overall structure can be inferred indirectly from the contribution of the partial 57 Fephonon density of states (PDOS). [10,11] More importantly,this method can be used to investigate aw ide range of materials from bulk materials to thin-films and NPs in experiments under extreme conditions such as high-pressure environments or during thermal or electrochemical catalytic reactions. [12][13][14][15] In this study,t he NRIXS method was used to follow the structural and chemical evolution of small 57 Fe-containing bimetallic NPs in al iquid environment under an applied external potential during the electrochemical reduction of CO 2 (CO 2 RR). X-ray absorption fine-structure spectroscopy (XAFS), another synchrotron technique with high sensitivity to elemental and local structural composition was also applied here under reaction conditions in an electrochemical environment to complement the NRIXS insight, including providing information on the Ag and Cu components of our electrocatalysts. [16] We have used iron-based materials in our study due to their applications as heterogeneous catalysts in the field of sustainable energy conversion. Iron is an abundant metal with little environmental impact, though it is unsuitable for CO 2 RR in its bulk form, since it favors the parasitic hydrogen evolution reaction (HER). [17] Nevertheless,i fp resent in the NP form at the core of nanostructures with aC O 2 RR-active thin shell, it could contribute to ad ecrease in the catalyst price.T od ate,r esearch on iron-based materials for CO 2 RR focuses mainly on molecular complexes,asfor example Fe-N-Cm aterials in porphyrin-like structures. [18,19] Here,w e combine iron with copper and silver within am icellar nanoreactor since these elements are promising for the selective conversion of CO 2 to C 1 -C 3 products (Cu) and CO (Ag). [17] In this work, the structure and composition of the FeCu and FeAg catalysts during the reaction will be extracted from as ynergistic combination of operando NRIXS and XAFS measurements and correlated with the selectivity trends of monometallic Cu, Ag and Fe NPs of similar size.

Results and Discussion
AFM images of the 57 Fe, 57 FeCu and 57 FeAg NPs deposited on SiO 2 /Si(100) are shown in Figure 1. Additional images and particle height histograms can be found in the Supporting Information, Figures S2, S3. Well dispersed, sizeselected NPs with average size (NP height) under 8nmwere obtained with the micellar synthesis.
XPS spectra of the Fe 2p region of the as-prepared bimetallic NPs deposited on aSiO 2 /Si(100) substrate and the corresponding fits displaying the different Fe oxidation states are shown in Figure 2. Additional XPS data from the Cu 2p and Ag 3d regions are shown in the Supporting Information, Figure S5. It is evident that the iron in the NPs after synthesis, N 2 -plasma treatment and subsequent air exposure is almost completely cationic (Fe 2+ or Fe 3+ ;S upporting Information, Table S1). As trong Fe 2p satellite is apparent in the FeCu spectrum, which has previously been reported for oxidized iron in the presence of copper. [20] TheF e:Cu ratio was calculated to be 55:45. TheAg3speak in FeAg overlaps with the Fe 2p region, which we compensated for when we calculated the Fe :Ag ratio of 64:36. Owing to the small size of the nanoparticles,w ee xpect the entire particle volume to be probed by XPS.
Thep roduct selectivities and current densities of our Fe, FeCu, FeAg,A g, and Cu NP samples are shown in Figure 3. While hydrogen is the main product for all samples,t he selectivities for CO of the FeAg and Ag NPs are similar (36 % and 39 %, respectively) at À1.1 Vv s. RHE. This similarity is consistent with ac ore-shell rearrangement of the FeAg NPs under reaction conditions,with Ag at the NP surface and Fe at the core.T he FeCu NPs also produce CO,b ut with al ower selectivity of under 2%.I nc ontrast to FeCu, both CO and formic acid selectivities are higher in the pure Cu NP sample (4.2 %a nd 5.5 %, respectively). This is likely indicative of amore Fe-rich surface in the FeCu NPs.Asexpected, higher hydrocarbons or alcohols were not detected in either of the FeCu or Cu samples,c ontrary to the case of bulk Cu. We attribute this difference to the enhanced number of lowcoordinated surface sites in small NPs that are known to strongly bind hydrogen and favor the HER. [21][22][23] Methane,in turn, is ap roduct for all iron-containing samples here,a lbeit with very low selectivity (< 1.7 %), but not for pure Ag or Cu NPs.M ethane production might be also affected by the NP size, [19] since bulk iron is not known to yield methane. [17] The methane production in the case of the FeAg samples might be explained by the presence of an incomplete (non-uniform) Ag-shell, leaving exposed small areas of the Fe core,w hich might lead to asynergistic interaction between Fe and Ag. We observed only low amounts of C 1 products typical of bulk Cu electrodes (CH 4 ,C O, HCOO À ) [17] for FeCu. Thep ure iron sample produced overwhelmingly hydrogen as its main  product, with only traces of formic acid and methane.Besides CO and hydrogen, the silver NPs were also found to produce formic acid with 4.3 %selectivity,which is thus similar to that of the FeAg NPs (4.4 %). Thecurrent densities normalized to the geometric area of the sample for the Fe,F eCu, FeAg, Ag and Cu NP samples are:0 .71 mA cm À2 ,0 .49 mA cm À2 , 0.35 mA cm À2 ,0 .79 mA cm À2 and 0.99 mA cm À2 respectively. Ap ossible explanation for the lower current density,a sw ell as slightly lower CO selectivity of the FeAg sample as compared to pure Ag, is the compressive strain induced by the smaller iron core on the silver overlayer.Ashift of the d-band center away from the Fermi level, caused by compressive strain, is expected to influence the CO bonding strength negatively,l eading to am ore favorable H 2 production.
NRIXS measurements were carried out before ( 57 Fe, 57 FeCu in air) and operando under CO 2 RR conditions in the electrolyte and under potential control ( 57 Fe, 57 FeCu, and 57 FeAg). Figure 4s hows the corresponding Fe-partial PDOS of these samples.The spectrum from abulk bcc-Fefoil is also shown for reference.T he spectra recorded in air indicate enhanced atomic disorder and oxidic Fe features at 41 meV À43 meV,t hat are at least partially reduced under reaction conditions, Figure 4a.The longitudinal acoustic (LA) phonon peak near 36 meV observed in bulk bcc-Feb ecomes also visible in the NP phonon spectra under reaction conditions. Nevertheless,itappears shifted to lower phonon energies (by 1.3 meV to 2meV; Supporting Information, Table S2) in both of the bimetallic samples due to the presence of the phononically softer Ag and Cu metals in the Fe environment. Furthermore,significant damping is observed in the LA peak of the three NP samples as compared to that of bulk bcc-Fe due to size-effects.
Interestingly,t he PDOS of the 57 FeAg NP sample shows the most clear resemblance to that of bulk bcc-Fe, since for that sample the LA peak is more prominent and we also see the transverse acoustic (TA) phonon modes,enhanced ordering (as indicated by sharper features), and the absence of oxidic species under the reducing conditions available during CO 2 RR. Such observations point towards asignificant reduction of the FeO x species when the potential is applied in  , obtained from raw NRIXS spectra of 57 Fe and 57 FeCu NPs in air and under CO 2 RR. b) The operando CO 2 RR PDOS data of 57 Fe, 57 FeCu, and 57 FeAg NPs together with those for abulk bcc-Fer eference foil for comparison. [12] c) 57 Fe-partial PDOS of 57 FeAg NPs plotted together with thin 57 Fe layers deposited on aAg film (4 nm) reproduced from Ref. [12].R epresentative error bars in different regions of the spectra are also shown in (a) and (b). The spectra have been vertically offset for better visibility and the vertical lines indicate the position of the two TA and the LA peaks of bulk bcc-Fe.
parallel to the segregation of Ag to the NP surface and the formation of aprotective Ag shell and apure bcc-FeNPcore.
Fort he pure 57 Fe NPs we still observe hints of the incomplete FeO x reduction under reaction conditions and enhanced disorder (smeared lines). The 57 FeCu NPs look more disordered and the Fe-partial phonon DOS has alarger deviation with respect to that of pure bcc-Fe ( Figure 4b), indicating some degree of Fe-Cu intermixing.N evertheless, aC u-rich surface is still expected since the PDOS of this sample is still in close agreement with that of pure bcc-Fe aside from the size-dependent phonon damping and enhanced disorder.Such spectra are also seen for thin Fe layers in the proximity of metals with alower phonon cut-off energy, as is the case here for both, Cu and Ag. [14] Overall, the PDOS corresponding to the FeAg and FeCu NP samples acquired under CO 2 RR approaches that of abcc-Festructure,not fcc-Fe as could be expected if Fe would be embedded in the fcc-Cu or fcc-Ag matrixes. [24] It should be noted that according to XPS the relative content of Fe in the as prepared FeCu and FeAg NPs is similar, namely 55 %and 64 %.
TheFe/Ag is alayered system known for alow interfacial intermixing.T he PDOS of the NPs in this sample displays analogous characteristics to that of aF e/Ag multilayer interface which was investigated by Roldan et al. [12] Figure 4c shows the Fe-partial PDOS data of this FeAg NP sample alongside the previously reported data from Fe/Ag multilayers. [12] Apart from the overall shape and correspondence to abcc structure,the positions of the LA peak around 35.5 meV and the TA peaks (20-30 meV) of the FeAg NPs are similar to those of the sandwiched Fe/Ag layers with Fe thicknesses of 2-4 nm. This is in accordance with the average particle size of the FeAg sample of about 4.2 nm (the real size of the iron part is smaller than this owing to the bimetallic composition).
Ac omparison of the PDOS of the 57 FeCu NP sample under working conditions and that of Fe(1.5 nm)/Cu(4 nm) multilayers from Roldan et al. [12] is displayed in Figure S6. While the LA peaks of the multilayer sample and those of the 57 FeCu NPs under working conditions are similarly positioned (34.1 meV vs.3 3.7 meV) and shaped, corresponding to the bcc-Fes tructure,t here is al ack of unambiguous structural features in the TA region, indicating high disorder due to the small NP size and phonon softening caused by the Fe-Cu interaction.
XAFS data were also acquired under CO 2 RR conditions using the same cell design and samples as in the NRIXS experiments in order to extract complementary information about the chemical state and structure of the samples, including their changes under reaction conditions.I nt his case,w ec ould gain access not only to the Fe-component of the bimetallic systems,but also to the Ag and Cu constituents.
It is evident from the XANES spectra in Figure 5a,c that the particles are in an oxidized state from the beginning, as they show typical oxidic features.F or comparison with our oxidized particles,wechose here to show the spectrum for an iron mineral (lepidocrocite) that consists of an aturally occurring hydroxide/oxide mixture,s ince the initial particle composition after air exposure is likely acomplex mixture as well. From Figure 5b,d we can conclude that the first Figure 5. Fe K-edge XANES (a,c) and Fourier-transforms of k 2 -weighted EXAFS data (phase-uncorrected) and the corresponding first-shell fits (dotted lines) (b,d) of 57 FeCu and 57 FeAg NPs as prepared (as-is) and under CO 2 RR conditions at À1.1 Vv s. RHE in 0.1 MKHCO 3 after 3.5 h ( 57 FeCu) and 1.5 h( 57 FeAg). References pectra from lepidocrocite FeO(OH) [37] and abulk iron foil are also shown for comparison.

Angewandte Chemie
Research Articles 22670 www.angewandte.org coordination shell of our oxidized NP samples is similar to that in the reference mineral, but the mean Fe À Odistance of 1.81 AE 0.06 in our samples is smaller than the reference value of 2.00 . [25] Table 1s ummarizes further relevant parameters extracted from the EXAFS measurements,b oth for as-prepared samples and samples under CO 2 RR conditions.
Upon applying ap otential of À1.1 Vv s. RHE, the particles start reducing to am ore metallic state.H owever, we still see as ignificant contribution from oxidized iron species under reaction conditions.S ome of these cationic species might also be assigned to iron compounds dissolved into the electrolyte.
We can show areduction and increase in crystalline order in the iron parts of the NP samples under reaction conditions that are in agreement with our previous NRIXS results.T he operando XAFS measurements were acquired over periods of 1.5-3.5 ho ft he reaction, while the NRIXS data were acquired over periods of 11 h-18 h, which might also explain the higher apparent content of iron oxides in the XAFS data. Thelong acquisition times were needed due to the low count rates in the NRIXS experiment. Figure 6d isplays Cu K-edge (a,b) and Ag K-edge (c,d) XANES and EXAFS data of the 57 FeCu and 57 FeAg NP samples.The presence of oxidic copper compounds similar to CuO was observed, and areduction of copper to the metallic state under operando conditions.T he average interatomic distance for Cu À Oi nt he as-prepared sample is in line with that of bulk CuO (Table 1). Thet ime dependence of the reduction process (Supporting Information, Figure S7) reveals af ast initial CuO reduction, followed by as lower increase over the next 1.5 ho ft he peak in the Fouriertransformed (FT) EXAFS at ca. 2.2 that corresponds to acontribution of the metallic Cu phase.For the as-synthesized samples,b oth XANES and EXAFS data collected at the Ag K-edge and shown in Figure 6c,d, indicate the presence of stable AgCl compounds due to residual Cl from the synthesis procedure.U pon applying the CO 2 RR potential, ac lear increase of the metallic silver contribution (peak in in the FT-EXAFS at ca. 2.7 )isseen, and no other components could be fitted. There is no contribution from Fe À Ag scattering paths,a nd the Fe À Fe and Ag À Ag distances do not hint towards intermixing of Fe and Ag either, since they are in line with those for the corresponding bulk materials.Inthe case of the 57 FeCu sample,t he CuÀCu distance agrees with that in bulk Cu, while the FeÀFe distance of 2.48 AE 0.05 is in line with that of bulk Fe at 2.47 AE 0.01 . [26] After careful evaluation of the operando structural and chemical information extracted from two complementary synchrotron X-ray methods (XAFS and NRIXS), we demonstrate as ignificant reduction of iron oxide to metallic species under reaction conditions,accompanied by astructural transformation from an amorphous or atomically disordered phase to ac rystalline structure.W eo bserved as imilar transformation for the secondary metals in the bimetallic NP samples via XAFS;aclear reduction of CuO to Cu in 57 FeCu, and even more pronounced, AgCl to Ag transformation in 57 FeAg.NRIXS served to distinguish segregated Fe-Ag and Fe-Cu phases (the present case) from alloyed structures, since we only detected ab cc-like Fe structure typical for asegregated Ag shell with bcc-FeNPcore.Asmall degree of Fe-Cu intermixing is obtained for the 57 FeCu sample from the XAFS data analysis.T he bcc-Fephase also predominates the NRIXS signal. Both these findings suggest the predominant segregation of both metals in our samples under CO 2 RR conditions.
It should be noted that the X-ray-based spectroscopic methods employed here for the characterization of our nanosized electrocatalysts (NRIXS and XAS) also have limitations that need to be considered for the interpretation of the data obtained. Fori nstance,b oth methods are bulksensitive ensemble-averaging techniques,t hat is,t he measured signal is an average of contributions from all metal species,residing both at the surface and in the core regions of all NPs within the X-ray irradiated sample area. Therefore, meaningful composition information can only be obtained  (6) Bulk Cu foil [33] 12 2.566 CuO [34] 41 .95(1) Ag foil [35] 12 2.883(7) AgCl [36] 62 .79(1) FeO(OH) [25] 62 .00 Fe foil [26] 8 + 62 .47(1) [a] The data were measured in air (as-is) and operando during CO 2 RR at À1.1 Vvs. RHE in 0.1 MKHCO 3 after 1.5-3.5 h. The uncertainty of the last digit is given in parentheses. Additional parameters and fit results are shown in the Supporting Information,T ables S3 and S4. Asingle FeÀFe scattering path was used to fit the overlapping contributionsfrom the first two coordination shells in bcc Fe owing to the limited resolution in R-space as aresult of the short Fe K-edge spectrum.
when applied to multimetallic samples with ah omogeneous inter-a nd intraparticle composition and chemical state. Furthermore,t he interpretation of structural characteristics extracted from X-ray spectroscopy methods will be further hindered when broad nanoparticle size and shape distributions exist in the as prepared samples. [27,28] This is even more relevant when operando catalysis studies are undertaken, since in that case the coexistence of particles of different size and shape evolving under reaction conditions is common. Therefore,t he interpretation of structural information extracted from XAS and NRIXS in terms of 3D structural motifs for inhomogeneous systems should be carried out with extreme care and must rely on complementary insight from additional techniques such as microscopy methods and theoretical modelling.
In the present study,w ec ombined AFM, TEM, XPS, NRIXS,a nd XAS to gain insight into the evolution of the structure and composition of homogeneously dispersed size and shape-controlled FeCu and FeAg NPs during CO 2 RR. We remark here that for the FeAg NPs,the EDX line profiles for Ag acquired after reaction (Supporting Information, Figure S4) extend further outwards as compared to the Fe profiles,which suggests some degree of Ag segregation to the surface.F or the FeCu NPs,t he line profiles indicate more even mixing of the two elements.T hese STEM-EDX results are,h owever, limited due to the difficulty of obtaining high quality and contaminant-free images in the presence of the Nafion NP binder and carbon support, which strongly interacts with the electron beam, resulting in the need of employing short acquisition times.N evertheless,t he microscopy data are consistent with the X-ray spectroscopy and selectivity data regarding the segregation behavior in the FeCu and FeAg NPs.
According to the respective surface energies [29][30][31] and atomic sizes,t he formation of aC uo rA gs hell and ab cc-Fe core is expected. Forinstance,wehave previously shown the segregation of copper to the surface of CuNi NPs under reducing conditions (CO 2 + CO + H 2 ). [32] This is in good agreement with the electrochemical behavior observed here for the FeAg NPs,which was found to be reminiscent to that of pure silver. Thetrend was not as clear for the FeCu sample, where size effects already demonstrated for pure Cu NPs will also lead to an increase of the H 2 production with decreasing NP size at the expense of CO and hydrocarbons.T herefore, for the Fe-Cu system, we cannot discard the possibility of having Cu-rich and Fe-rich regions coexisting at the NP surface,which would also explain the increased H 2 production (taking place over the exposed Fe component) as compared to the selectivity obtained in similarly sized monometallic Cu NPs.
Although the present nanosized catalysts do not display an outstanding activity or selectivity for CO 2 RR, in line with previous studies for small NPs which also favor H 2 evolution, [21][22][23] our study serves to provide fundamental insight into the dynamic behavior of electrocatalysts under reaction conditions.F urthermore,w ei llustrate ap owerful combination of X-ray based synchrotron techniques for the characterization of the structural and chemical evolution of electro- Figure 6. Cu K-edge (a,b) and Ag K-edge (c,d) XANES (a,c) and Fourier-transforms of k 2 -weighted EXAFS data (phase-uncorrected) (b,d) of 57 FeCu and 57 FeAg NP samples in as-prepared state (as-is) and under CO 2 RR conditions (À1.1 Vvs. RHE in 0.1 MKHCO 3 after 1.5 h( 57 FeCu) and 2.5 h ( 57 FeAg)) Referencespectra from CuO, Cu-foil, AgCl [37] and aAg-foil are shown for comparison.

Angewandte
catalysts that, when available hand in hand with reactivity data, allows one to gain in depth understanding of the structural motifs and chemical compositions responsible for specific selectivity trends.

Conclusion
This study demonstrates the dynamic transformations undergone by FeCu and FeAg NPs during the electrochemical reduction of CO 2 .I np articular, operando NRIXS and XAFS data revealed the formation of an Ag shell surrounding ab cc-Fec ore.O nt he other hand, more intermixing was observed for the FeCu NPs,and the presence of separated Fe and Cu regions on the NP surface under reaction conditions could not be excluded. Interestingly,asimilar CO production was observed for the thin Ag-shell in the FeAg NPs as compared to pure Ag NPs,which indicates the optimization of the use of the noble metal. Overall, and due to the small NP size,the production of H 2 was,however,favored, especially in the Fe and FeCu samples.
Finally,o ur work emphasizes that operando experiments are av ery valuable tool to link catalytic properties to structure and composition of electrocatalysts under realistic working conditions.