Experimental Evidence for the Incorporation of Two Metals at Equivalent Lattice Positions in Mixed‐Metal Metal–Organic Frameworks

Abstract Metal–organic frameworks containing multiple metals distributed over crystallographically equivalent framework positions (mixed‐metal MOFs) represent an interesting class of materials, since the close vicinity of isolated metal centers often gives rise to synergistic effects. However, appropriate characterization techniques for detailed investigations of these mixed‐metal metal–organic framework materials, particularly addressing the distribution of metals within the lattice, are rarely available. The synthesis of mixed‐metal FeCuBTC materials in direct syntheses proved to be difficult and only a thorough characterization using various techniques, like powder X‐ray diffraction, X‐ray absorption spectroscopy and electron paramagnetic resonance spectroscopy, unambiguously evidenced the formation of a mixed‐metal FeCuBTC material with HKUST‐1 structure, which contained bimetallic Fe−Cu paddlewheels as well as monometallic Cu−Cu and Fe−Fe units under optimized synthesis conditions. The in‐depth characterization showed that other synthetic procedures led to impurities, which contained the majority of the applied iron and were impossible or difficult to identify using solely standard characterization techniques. Therefore, this study shows the necessity to characterize mixed‐metal MOFs extensively to unambiguously prove the incorporation of both metals at the desired positions. The controlled positioning of metal centers in mixed‐metal metal–organic framework materials and the thorough characterization thereof is particularly important to derive structure–property or structure–activity correlations.

One of the most-studied metal-organic frameworkm aterials is CuBTC (also knowna sH KUST-1, [5] MOF-199 [6] or Cu 3 BTC 2 , [7] see Figure 1a), which contains dimericc opper units and benzene-1,3,5-tricarboxylate (BTC) as building blocks. The characteristic and well-knownS BU is ap addlewheel, which consists of ad imericc opper unit that is bridged by four carboxylate groupsf rom four BTC molecules (see Figure1b). Usually,t he two axial positions of the paddlewheels are occupied by water or other solvent molecules, but the removal of these molecules from the axial positions by heat and/or vacuum treatment leads to the formation of coordinatively unsaturated sites (CUS) at the copper atoms. [8] These CUS can be reached by guest molecules entering the pores, which enables ad irect metal-guest interaction. Due to the possibility of direct metalguest interactions within the HKUST-1 structure it is particularly interesting to change the type of metals used forthe synthesis. Indeed, severalo ther MBTC materials (M = Cr, [9] Mn, [10] Fe, [11] Co, [10] Ni, [12] Zn, [13] Mo, [14] Ru [15] )w ith HKUST-1 structure have been reported. Theoretical calculations proposed favorable oxygen adsorptionp roperties for materials with HKUST-1 structure containing early 3d transition metals (Sc or Ti), but such materials have not been synthesized so far. [16] The combination of different types of metals in one framework can lead to outstanding new properties. [17] Of speciali nterest are mixed-metal metal-organic framework materials, in which different metal types are distributed over crystallographically equivalent framework positions. [18] By using this approach,t he properties of ak nown and promising framework structure can be extended anda lteredw ithout changing the framework topology.F or the HKUST-1s tructure, some mixedmetal structures with the metal combinations Cu/Ru, [19] Cu/ Zn, [20] Cu/Ni, [21] Cu/Pd, [22] Cu/Ag, [23] Cu/Mn, [24] Cu/Fe [24] and Cu/ Co [24] have already been reported. The first four have been synthesized via direct syntheses, while the latter ones used postsynthetic metal exchange. The preparation of these mixedmetal HKUST-1 structuresi se xpected to result in interesting new properties, since direct metal-metal interactions are possible within the bimetallic paddlewheel units. Theoretical calculations of Zhang et al. have shown that am ixed-metal CuWBTC should have potential for the activation of CO 2 , [25] although the synthesis and experimental verification of this claim has not been reported yet.
The development of synthesis routes for mixed-metalM OFs, especially the generally favored direct synthesis, is often challenging,t hough. The influence of various parameters (e.g. crystallization rates, different preferred coordination geometries, competing structures or solubility issues) might lead to the formation of phase mixtures of different structures or phase mixtures of monometallic HKUST-1s tructures. Moreover,e ven completely new structures containing two types of metals might be obtained as undesired side products. Furthermore, a metal-organic framework structure with metal nanoparticles within the pores, as reported by Zhang et al., [22] is generally possible as well. Hence, ac areful characterization of the synthesized materials is of high importance.F or only af ew of the reportedm ixed-metal HKUST-1 structures, the presenceo fb imetallicp addlewheel structures has been truly verified and the formation of other undesiredp hases has been unambiguously excluded.
Withint his article, the development of ad irect synthesis route for mixed-metal FeCuBTC is reported. Furthermore,w e present ad etailed characterization with special attention on competing structures. While such impurities have been detected for the majority of the synthesized materials, the characterization data for ap hase-pure FeCuBTCm aterial proved that iron was only incorporated within the paddlewheel units. Moreover, bimetallic paddlewheels containing copper and iron were found. The experimentalp rooff or the incorporationo f both metals in defined positions is important for all mixedmetal metal-organic framework materials, but the technical realization is rather difficult, though.

Results and Discussion
As as tarting point fort he synthesis of mixed-metalF eCuBTC materials, variouss ynthetic procedures were chosen that had been reported in the literature for CuBTC. These included solvothermal syntheses, syntheses at ambient pressure andmicrowave assisted syntheses. Furthermore,p ost-synthetic metal exchange as reported by Sava Gallis et al. [24] has been tested to synthesize an iron-and copper-containing mixed-metal HKUST-1m ateriala sr eference sample, but no phase-pure material could be obtained even after slight variations of the synthetic procedure.
All syntheses were performed with an Fe:Cu ratio of 30:70. In the majority of the obtained products, mixtures of different phases have been observed. The undesired sidep roducts could be clearly identified as a-Fe 2 O 3 ,M IL-100(Fe) andB asolite F300 based on the powder X-ray diffractionp atterns. Particularly,s ynthesest hat contained an excess of total metal amount with respect to 1,3,5-benzenetricarboxylic acid showeds ignificant amountso fi ron-richp hases in the diffractionp atterns. However,s everal samples have been obtained that showedn o or only minor reflectionso fu ndesired side products and these were initially assumed to be neglectable. In the following, three of theses amples will be presentede xemplary,w hich are denoted as CuBTC/Fe 2 O 3 ,C uBTC/MIL-100(Fe) and CuBTC/F300. CuBTCa nd MIL-100(Fe) have been synthesized as phase-pure materials and used as references togetherw ith commercially available a-Fe 2 O 3 .U nfortunately,t he initially intended synthesis of aF eBTC reference material with HKUST-1 structure, as reportedb yX ie et al., [11] could not be reproduced. One of the synthesized samples, in the following denoted as FeCuBTC,d id not show any impurities, even after at horough characterization using various techniques (vide infra). Thus, we claim that this material is truly am ixed-metal HKUST-1s ample, which containsi ron and copperd istributed over equivalent metal framework positions, as will be shown in the following.
The powder X-ray diffraction (PXRD)p atterns ( Figure 2) showed am ixture of phasesf or CuBTC/MIL-100(Fe). Although the respective reflections of the MIL-100(Fe) phase were weak compared to the reflections of the HKUST-1 structure, they could be clearly identified. In the case of CuBTC/Fe 2 O 3 ,o nly the high quality of the data up to 2q = 50 8 (cf. Figure S1) and ac areful comparison with other characterization techniques that will be presented in this article, allowed the unambiguous identification of the a-Fe 2 O 3 phase. The corresponding weak reflectionso fa-Fe 2 O 3 could be identified at 2q = 33.4 8 and 35.9 8 and would have been easily missed if the noise level was more pronounced, as maller 2 q range was chosen for the measurement or the datawere not analyzed thoroughly.
The identification of the F300 phase in CuBTC/F300w as also rather difficult and would have been easily missed on af irst glance.T his was due to the rather amorphous character of the F300 material, which didn ot show any sharp reflections in the PXRD pattern. Initially,t he broad reflection centered at 2q = 10.7 8 was first interpreteda sapart of the background signal originating from the sample holder and only furtherd etailed characterization proved the presence of F300. Based on the powderX -ray diffraction data, FeCuBTC showed no other reflections besidet he HKUST-1s tructure, particularly none of the above-mentioned ones. This indicated the formation of a phase-pure material with solely HKUST-1 structure. The Pawley fit of FeCuBTC (see Figure S2 and Ta ble S1) showedt hat the incorporation of iron into the HKUST-1 structure resultedi na slight expansion of the unit cell compared to CuBTC (26.323 vs. 26.313 ).
Nitrogen physisorption measurements showedt hat all materials featured type Ii sotherms, which are typical for microporous solids.T he specific surfacea reas of all mixed-metal materials were in the range from 670 m 2 g À1 (CuBTC/Fe 2 O 3 )t o 1070 m 2 g À1 (CuBTC/MIL-100(Fe)) and, thus, significantly lower than the value for CuBTC (see Figure S3 and Ta ble S3).
The metal ratios wered etermined using ICP-OES analysiso f digested samples (see Table S2). Both FeCuBTC and CuBTC/ MIL-100(Fe) showedt he expected Fe:Cu ratio of 30:70, which had been applied during the synthesis. Although CuBTC/F300 and CuBTC/Fe 2 O 3 were synthesized using the same initial ratio, the obtained product showed as ignificantly higheri ron content (Fe:Cu % 40:60). This might be related to the fact that an excess of the metal precursors was used with respectt o1 ,3,5-benzetricarboxylic acid. At ime-dependent study on the course of the reaction for aC uBTC/F300r elated material showedt hat an iron-richF 300 phasew as formed in the beginning and only after severalm inutes an HKUST-1 structure was observed, which explained the found metal ratios (see Figure S4 and Ta ble S4).
The recorded ATR-IR spectra provedt he absence of residual linker molecules within the pores (carboxylate vibrations expected at 1700 cm À1 )f or all obtained materials (see Figure 3a). Furthermore, the band pattern of all samples showed several similarities within this region and the main differences were variations in band intensities or small shifts of some bands. In particular, the strong similarity of FeCuBTC and CuBTC indicated similar coordination modes of the carboxylate groups within both materials. In addition, the similar spectrum of CuBTC/Fe 2 O 3 might be reasonable, since the formed a-Fe 2 O 3 is not expected to have any significant bands in the measured range of the ATR-IR spectra.T herefore, only the HKUST-1s tructure of CuBTC/Fe 2 O 3 contributed to the observedI Rs pectrum. The presenceo fs mall shifts of FeCuBTC (e.g. from 1638 cm À1 to 1643 cm À1 andf rom 1550 cm À1 to 1559 cm À1 ,s ee Figure 3b) compared to CuBTC indicated that the vibration frequency of the carboxylate vibration has changed compared to monometallic CuBTC. Possible reasonsf or this might be ad ifferent coordination mode of the carboxylate groupso rt he presenceo f different metals, to which the carboxylate groups are coordinated. The latter possibility would be reasonable if iron was present in the paddlewheel units insteado fc opper. The change of electronegativity from copper to iron would also influence the vibration frequency of the coordinated carboxylate groups.M oreover,acloser view on Figure 3s howed that small shifts were visible for CuBTC/Fe 2 O 3 as well, which might indicate that iron has been partially used for the formationo fa-Fe 2 O 3 and partially for the incorporation in the paddlewheels of the HKUST-1 structure.
Density functional theory (DFT) calculations of non-periodic model systems have been performed to evaluate whether the observed shifts of the IR bands are in ar ealistic range for bimetallicp addlewheel units. These calculations were performed for isolated paddlewheel units with four benzoate ligands. Three cases have been considered: (i)aCuÀCu paddlewheel with two axial water molecules coordinated to the copperc enters, (ii)aCuÀFe paddlewheel with one axial water molecule Figure 2. PowderX-ray diffraction patterns of the synthesized Fe-Cu-BTC containingm aterials and CuBTC as reference. All diffraction patterns were normalized to the largest reflection of each material. Reflections, which do not belongtot he HKUST-1 structure, are indicated by + (a-Fe2O3), *( F300) or #(MIL-100(Fe)).For detailedviews see Figure S1. coordinated to the copperc enter and one axial chloride ion coordinated to the iron center and (iii)aFeÀFe paddlewheel with two chloride ions coordinated to the iron centers. All structures weref ully optimizedu sing ah ybrid functional DFT method (including dispersion corrections) until at rue energy minimum was located (see Figure S5). Also, in the case of Fe-containing paddlewheels, only slightly twisted structures were found, allowing the formation of the periodic MOF structure. Subsequently,an ormalm ode analysisw as performed to calculate the IR vibrational frequencies and intensities (see Figure S6). The resultss howed that the positions of carboxylate vibrations were shifted upon the exchange of the metal centers in the same order of magnitude as observed experimentally and that the band positions of vibrationso ft he bimetallic FeÀCu paddlewheel were in between those for the monometallic systems. However,t he calculated direction of the shifts matched only partially with the experimentally observed ones. Ap ossible reason for these differences couldb ea ni nsufficient accuracyo ft he theoretical level fort hese smallc hanges.I na ddition, the approximation of isolated paddlewheel units instead of ac omplete MOF structure might influence the results. Note that the considered modesw ere linker vibrations, which are only indirectly influenced by the type of the metal centers and, thus, the influences of the metal substitution are difficult to capture. MÀOs tretchv ibrations would, therefore, be a better probe but are experimentally not accessible. Nonetheless, the performed theoretical calculations indicated that the metal substitution within the paddlewheel had an influence on the vibration frequency of the carboxylate groups, which,i n combination with experimentally data, suggested that bimetallic paddlewheels were formed for the FeCuBTCm aterial.
In order to get ac loser insighti nto the chemical environment on the atomic scale, X-ray absorption spectroscopy measurements have been performed using synchrotron radia-tion at PETRA III, DESY (Hamburg). Only small differences were visible in the XANESr egion at the Cu K-edge (see Figure 4a) and the position and shape of the edges suggested that all samples contained copper in the + +II oxidation state, which was in accordance with other literature reports. [8,26] The EXAFS and Fourier-transformed EXAFS spectra (see Figure 4c,e) showeds mall differences between the monometallic CuBTC and the mixed-metal samples. All mixed-metal samples looked alike, though. These observations might indicate the incorporation of minor amountso fi ron next to copperw ithin the paddlewheel units for all mixed-metal samples. In the R-space, the largestd ifferences were visible in the region around R = 2.4 , which would be ar easonable distance for aC u ÀFe dimer in a bimetallic paddlewheel unit if ap hase shift correctioni su sed.
The XANESs pectra recordeda tt he Fe K-edge showedc lear differences (see Figure 4b). The spectra of MIL-100(Fe), CuBTC/ MIL-100(Fe) and CuBTC/F300 were nearly identical, but significantly different from those of CuBTC/Fe 2 O 3 andt he a-Fe 2 O 3 reference, which were also similar.O nt he other hand, the XANESs pectrum of FeCuBTC lookedd ifferent from all other samples. The oxidations tate of iron was determined to be + +III for all samples, though. [27] Ac loser look at the EXAFS spectra (see Figure 4d)l ed to similar conclusions. All samples containing at rimeric SBU (e.g. CuBTC/MIL-100(Fe) and CuBTC/F300) lookedi denticala nd also the samples containing a-Fe 2 O 3 (e.g. a-Fe 2 O 3 and CuBTC/Fe 2 O 3 )w ere similar. For the lattero nes, significant differences in the amplitude (k = 7-10 À1 )w ereo bserved. FeCuBTC was different from the other samples concerning the phase behavior as well as the amplitude, which supported the claim of ap hase-pure mixed-metal FeCuBTC material, in which neither a-Fe 2 O 3 nor MIL-100(Fe) or F300 was present.
MIL-100(Fe),C uBTC/MIL-100(Fe) and CuBTC/F300 differed from each other mainly in the intensity of the first shell in the Fourier-transformed EXAFS spectra (see Figure 4f). MIL-100(Fe) showedt he highest intensity,w hileC uBTC/MIL-100(Fe) and CuBTC/F300w erel ess intensive. This observation was in accordancew ith the results at the Cu K-edge, which indicated that minor amounts of iron might be incorporatedi nt he HKUST-1 structure for the latter two materials. In the MIL-100(Fe)s tructure, iron is coordinated by six oxygen atoms, whereas it would be only coordinated by five oxygen atoms in ap addlewheel unit. If iron wasp resent in both types of SBUs, the total intensity of the first shell would be lower compared to the pure MIL-100(Fe). However,t he overall similarity to MIL-100(Fe)s uggested that the majority of iron was present in a trimericS BU similart oM IL-100(Fe) and only minor amounts have been incorporatedi nto the HKUST-1 structure.
Ac loser look on a-Fe 2 O 3 andC uBTC/Fe 2 O 3 showed that they differed in intensity of the higher shells (R = 3-4 ). The reduced intensity for CuBTC/Fe 2 O 3 compared to pure a-Fe 2 O 3 might have two reasons;e ither the formed iron oxide particles were very small or ap hase mixture, with iron being present as iron oxide and in aM OF structure at the same time, was formed. The first possibility could be ruled out, since reflections of the iron oxide phasew ere observed in the PXRD patterns (see Figure 2), which would not be the case for very small particles showing ar educed intensity of the corresponding shells. Therefore, in agreementw ith the above-mentioned data, the presence of iron in a-Fe 2 O 3 and aM OF structure seemed plausible. Duetot he dominant presenceo fl ight backscatters and, comparedt oi ron oxide, as malln umber of total backscatterers aroundt he iron center,asmall intensityf or shells for R > 2 is expected for SBUs of the treated MOF structures. On the other hand, ac omparable large number of iron backscatterersi nt he iron oxide phase results in high intensities in the range R = 3-4.5 (see Figure 4f). The combination of both situations, which will be effectively measured in XAS for phase mixtures, resulted in ar educed intensity for the higher shells as observed for CuBTC/Fe 2 O 3 .T he identification or extraction of the MOF structure based on these EXAFS data would be ac hallenging task and the data quality would not be sufficient for these purposes. Based on these results, CuBTC/Fe 2 O 3 might be comparable to the Pd@[Cu 3Àx Pd x (BTC) 2 ] n materials reported by Zhang et al. [22] in the form of FeCuBTC differed significantly from all the mentioned samples in the R-space. The intensity of the higher shells seemed plausible for aS BU of am etal-organic framework structure. The lower intensity of the first shell compared to MIL-100(Fe) was also plausible for an HKUST-1s tructure, since only five nearest neighbors are expected for ap addlewheel structure instead of six for MIL-100(Fe).
The Fourier-transformed spectra were furthera nalyzed by multi-shell structure fitting. Since the SBU of these MOF materials contain severals hells with light backscatterers, which are closer to the absorber atom than the next metal shell, several restraints had to be applied to the fitting models (in particular the carbona nd oxygen shells)t oa ssure ar easonablen umber of free variables for the appliedn umber of fitted shells.
The analysis of the Fourier-transformed EXAFS spectra showedt hat the obtained spectrum for CuBTC corresponded closely to the reporteds ingle crystal structure (see Figure S7 and Table 1). The first shell corresponded to fiveo xygens in a distance of 1.95 .F our of them originatedf rom coordinated carboxylate groups, the fifth from coordinated water at the axial position of the paddlewheel. Thes econd coppera tom of the copper dimer in the paddlewheel unit was found at 2.65 and four carbon neighbors at 2.86 were needed to obtain a reasonable fit quality.H igher shellsc ould not be resolved with sufficient accuracy.
As imilar fit was performed on the Fourier-transformed EXAFS spectrum of FeCuBTC at the Cu K-edge, but worse results have been obtained, especially at R = 2.4 (see Figure S8 and Ta ble S5). By adding an iron shell in addition to the already fitted copper shell, the results could be improved (see Figure 5a,c,ea nd Table 1). The total number of iron and copper neighbors wasf ixed to one, but the Fe:Cu ratio was used as a free fitting parameter and resulted in 0.57 AE 0.10 coppern eighbors and 0.43 AE 0.10 iron neighbors. These results indicated that both monometallic copper paddlewheels and bimetallic  iron-copper paddlewheels were present in FeCuBTC,w ith nearly twice as much bimetallic paddlewheel units.B ased on the obtained coordination numbersa tt he Cu K-edge, the calculated Fe:Cu ratio of 30:70 is in accordance with the ICP-OES measurements (see Table S2).
For the MIL-100(Fe)r eference sample, the experimental data at the Fe K-edge were in accordance with the fitted data based on the structureo fM IL-100(Fe) obtained from single crystal analysis( see Figure S9 and Ta bleS6). Similar fits were obtained for the samples CuBTC/MIL-100 and CuBTC/F300( see Figure S10 and Ta ble S7), supporting the above presented assumptionst hat iron was present almost exclusively in the trimeric SBU that is typical for the MIL-100(Fe)a nd F300 structures. In the case of a-Fe 2 O 3 ,m ulti-shell fitting still represents a difficult task, because the number of paths in the range R = 3-4.5 ,i st oo large for reliable results. The same situation was present for CuBTC/Fe 2 O 3 ,s ince not only al arge amounto fp arametersw ould be necessary for a-Fe 2 O 3 ,b ut also for the MOF phase. Therefore, onlyt he qualitative analysis presented above was possible.
The multi-shell structure fitting of FeCuBTCa tt he Fe K-edge gave unexpected and interesting insights into the local structure around the iron centers. During the data treatment, an inherentp roblem with the intensity of the first shell occurred, which could only be solved if either the number of nearest oxygen neighbors was increased to more than six or the amplitude reduction factor was increased to more than 1.0. Both options are physically not meaningful. Thus, an element with a higher atomic number than oxygen neededt ob ep resent close to iron. Since the oxidation state of iron was determined to be + +III andt he chloride salt was used durings ynthesis, a chloridei on seemed plausible, particularly due to the negative charge necessary to assurec hargen eutrality of the paddlewheel unit. Indeed, the fitting procedure with chloride at the axial position of the paddlewheel and four oxygens from the carboxylate groups returned the best results( see Figure S11 and Ta ble S8). Furthermore, af it with the first two oxygen shells from a-Fe 2 O 3 did not result in ar easonable fit (see Figure S12 and Ta ble S9), which excluded the presence of iron in a-Fe 2 O 3 nanoparticles.
Further shells (C, O, Fe) were added according to the reported structure of FeBTC. [11] The resultingf it was acceptable, but still major deviations werev isible in the R-space (R = 2.4 )a nd q-space (q = 10.7 À1 ,s ee Figure S13 andT ableS10). In the Rspace, these deviations were in as imilar region like at the Cu K-edge. By adding ac oppers hell in this region, ag ood fit was obtained (see Figure 5b,d,f and Table 1). As already mentioned for the Cu K-edge, the total number of metal neighbors was fixed to one and the Fe:Cu ratio wasu sed as free parameter. The fit resultedi n0 .58 AE 0.31 copper neighbors in ad istance of 2.79 and 0.42 AE 0.31 iron neighbors in ad istance of 3.03 . These results indicated the presence of both bimetallic FeÀCu and monometallic FeÀFe paddlewheel units. However,t he number of the metal neighbors did not fully match with the found metal ratios by ICP-OES. The complexity of the local structure might be ap ossible explanation. Due to am ixture of monometallic and bimetallic paddlewheel units, which differ in their metal-metal bond distances, ad istortion of the local structure is expectedf or the bimetallic paddlewheel units resultingi nn umerousbackscattering paths with slightly different distances for oxygen and carbon atoms. Therefore, this situation is too complex to be solved with high accuracy by EXAFS structure fitting based on the present data. The overlap of the FeÀFe shell with Ca nd Os hells provides additional challenges in the fitting procedures, which leads to large errors in coordination numbers.
The EXAFS analysisa tb oth the Fe K-and the Cu K-edge of FeCuBTC clearly showed that bimetallic paddlewheelu nits were present.F urthermore, it indicated the presence of monometallicp addlewheels consisting of only copper or only iron. At both edges,o nly the assumption of bimetallic paddlewheel units resulted in good fit results.
Electron paramagnetic resonance (EPR) spectroscopy has been successfully used to investigate paddlewheel-type structures in previouss tudies. [20b, 28] Therefore, EPR spectra have been recorded for all materials to further corroborate the results (see Figure 6a nd Table S11). All samples showedc omparable EPR spectra at 15 Kd isplaying the signal of monomeric Cu 2 + species (A)w ith partly resolved Cu hyperfine (hf)s plitting (parameters of A: g zz = 2.340, g xx/yy = 2.05, A zz = 485 MHz). These features might be assigned to defective CuÀCu paddlewheel units or extra-framework cupric ions. Only for FeCuBTC, an additional broad baselines ignal (B)w as observed. The g xx/yy spectral positiona t3 28 mT of signal A was presenti nt he spectra recorded at 100 Kf or all samples and at RT for CuBTC,C uBTC/ MIL-100(Fe), and CuBTC/Fe 2 O 3 .I nC uBTC at RT,a na dditional broad signal (C)w ith al ine width of about8 0mTand an isotropic g-value of g = 2.149 was observed,w hich could be assigned to the excited S = 1s tate of the antiferromagnetically coupled Cu 2 + ions of the CuÀCu paddlewheel units. [28b] Presumably,t he signal observed for CuBTC/MIL-100(Fe) at RT displaying ac omparable line width might be likewise assigned to the CuÀCu paddlewheel although the g-value was somewhat lower.I nt he case of CuBTC/Fe 2 O 3 ,t he line width of this signal was considerably smaller and an assignment to the CuÀCu paddlewheel was questionable although its g-value was close to that of CuBTC.
FeCuBTCe xhibited ad istinct spectrum at RT.H ere, an intense EPR signal (D)w ith as maller g-value (g = 2.023) and linewidth (36 mT) was observed. The total EPR signal intensity at room temperature was higher in comparison to the three other samples (see Ta ble S11). Neithers ignal A of the monomeric Cu 2 + species nor signal C of CuÀCu paddlewheel units were detectedf or FeCuBTC at RT.M oreover,t he resolved fine structure signals of the S = 1s tate of the CuÀCu paddle-wheels, [28b] which are typicallyo bserved at 20 mT and 474 mT at 100 K, were not detected. Based on its g-value, signal D cannotb ea ssigned to either monomeric Cu 2 + specieso rC u 2 + pairs in the paddlewheel unit. Therefore, we suggest that it was related to Cu 2 + ÀFe 3 + speciesi nt hese samples. In addition, ac omparison of the spectra of FeCuBTC recorded at RT and 100 Kr evealedt hat signal D did not disappear with decreasingt emperature but exhibited ap ronounced line broadening.C onsequently,t he broad baseline signal (B)r ecorded at 15 Kf or this sample might likewise be assigned to Cu 2 + ÀFe 3 + species. Note that high-spin Fe 3 + speciesi nh ighly symmetric octahedral or tetrahedral environmentw ould also provide an isotropic signal at g = 2.00. However,t he disappearance of signal features forF eCuBTC, which would indicate CuÀCu paddlewheel units, suggested that signal D was not related to such Fe 3 + species, but would insteadi ndicate coupled Cu 2 + À Fe 3 + moieties. Furthermore, the substantially larger overall EPR signal intensity of FeCuBTC in comparison with CuBTC, CuBTC/ MIL-100(Fe) and CuBTC/Fe 2 O 3 supported this interpretation that an ovel EPR active speciesw as formed in FeCuBTC,w hich was not presentint he other three materials.
Although the exact nature of these Cu 2 + ÀFe 3 + species could not be derived from thesee xperiments,i ts g-value andt emperatured ependence revealed some of its characteristics. The assumption of strongly exchanged coupledC u 2 + ÀFe 3 + pairs and the coupling between the Cu 2 + electron spin (S Cu = 1/2) and the iron electron spin (S Fe = 5/2) resultsi ntwo spin states of the pair, S = 2a nd S = 3. According to Buluggiu, [29] the S = 3 state provides ag -value of g pair = 2.00, which was in reasonable agreement with the measured g-parameters of signal D if typical isotropic g-values of 2.15 and 2.00 are assumed for Cu 2 + and Fe 3 + .T he isotropic nature of the signal Dw as indicative for strong magnetic exchangei nteractions among the Cu 2 + À Fe 3 + species. The strong line broadening with decreasing temperaturew as astonishing and atypical for paramagnetic systems, but such effects are common for magnetic materials. [30] Conclusions In summary,t he presented results provide ad etailed insight into the synthesis of am ixed-metal HKUST-1 structure containing coppera nd iron. Al arge variety of reactionp arameters was screeneda nd the obtained phases and phase mixtures were characterized thoroughly.I tw as found that ac areful indepth characterization of the obtained mixed-metal materials and ac omparison to other iron-and copper-containingm aterials was necessary to undoubtedly confirmt he formation of a truly phase-pure FeCuBTC material. In particular,t he characterization using EXAFS analysisp rovided the necessary insight on the atomic scale to verify the presence of solely paddlewheel units as SBUs. The recorded EPR spectra further supported the hypothesis of bimetallic paddlewheels for FeCuBTC.F urthermore, this studys howed that small amounts of impurities should not be assumed to be neglectable when workingw ith mixed-metal metal-organic framework materials, since the majority of one metal might be present exclusively within this impurity phase. Moreover, we demonstrated that at horough characterization with high-quality data is particularly important to unambiguously verify the formation of phase-pure materials and the absence of any undesired phases in mixed-metal MOF materials.
Synthesis of CuBTC/F300:I nat ypical synthesis 1,3,5-benzenetricarboxylic acid (H 3 BTC, 0.4896 g, 2.33 mmol, 1.00 equiv.) was dissolved in N,N-dimethylformamide (DMF,2 5mL) at 100 8C. Iron(III) chloride hexahydrate (0.3925 g, 1.45 mmol, 0.62 equiv.) and copper(II) nitrate trihydrate (0.8185 g, 3.39 mmol, 1.45 equiv.; 4.84 mmol or 2.08 equiv.t otal metal amount) were dissolved in demineralized water (25 mL) at room temperature. Both solutions were combined and stirred at 100 8Cf or four hours. Afterwards the solid was filtered off using ag lass filter and washed with 3 20 mL of DMF and 1 20 mL of demineralized water.T he filtered product was dried in air at room temperature overnight and for another three days at 130 8Ci na noven. were weighed into at eflon vessel of am icrowave sample tube. DMF (12.5 mL) and water (11.5 mL) were added and the vessel sealed. The reaction mixture was heated to 140 8Cw ithin 5min in am icrowave oven and kept there for two hours. After cooling, the sample was filtered off, washed with DMF (3 20 mL) and water (1 20 mL), dried at room temperature overnight and for three days at 130 8Ci na ir. Powder X-ray diffraction:P owder X-ray diffraction patterns were recorded using either aB ruker D8 Advance or aP ANalytical Empyrean diffractometer both with Bragg-Brentano geometry.T he samples were analyzed in the range of 2 q = 4-50 8 by using Cu Ka radiation. As tep width of 2 q = 0.0164 (Bruker) or 0.0138 (PANalytical) was used. For visualization, ab ackground subtraction with subsequent smoothing of the obtained diffraction patterns was performed. In order to compare measurements of the two different diffractometers, the obtained diffraction patterns were normalized with respect to the highest reflection of each pattern. X-ray absorption spectroscopy:X AS experiments were performed at PETRA III Extension beamline P65 (energy range:4 -44 keV) at DESY (Deutsches Elektronensynchrotron) in Hamburg (Germany). For the measurements at the Cu K-and/or Fe K-edges, aS i(111)Ctype double crystal monochromator was used. The beam current was 100 mA with ar ing energy of 6.08 GeV.A ll samples were prepared as pellets using cellulose as ab inder.A ll spectra were recorded in continuous scan mode in transmission and fluorescence mode at ambient temperature and pressure in the range of À150 eV to 1000 eV around the edge in 180 sec. For the data analysis, transmission data were used. For calibration, the corresponding metal foils were measured as ar eference simultaneously with the samples.

Synthesis of
The data treatment was performed using the Demeter software package. [31] In order to compensate for the oversampling of the continuous scan mode, the data points of the obtained spectra were reduced with the help of the "rebin" function of the Athena software (edge region: À50 to + 50 eV;p re-edge grid:5 eV; XANES grid:0 .5 eV;E XAFS grid:0 .05 À1 ). For data evaluation, a Victoreen-type polynomial was subtracted from the spectrum to remove the background using the Athena software. The first inflection point was taken as energy E 0 .
The EXAFS analysis was performed using the Artemis software. As starting models, cif-files of CuBTC [5] and FeBTC [11] were used to generate the paths. Both, the ka nd Rr anges were selected based on the data quality.P rior to the fitting procedure, the amplitude reduction factor S 0 2 was determined on am etal reference foil at the corresponding edge and used as fixed parameter for the samples, respectively.F or the fitting procedure, several restraints had to be applied to obtain reasonable results. Therefore, most of the coordination numbers were fixed, Debye-Waller factors were assumed to be the same for several neighbors and, in some cases, the variations of distances (DR) were coupled for shells that were close to each other.
EPR measurements:T he EPR spectra were measured on aB RUKER EMXmicro X-band spectrometer using an Oxford instruments ESR900 cryostat. As all spectra were found to be broadened, the modulation amplitude has been maintained as 10 Gf or the EPR measurements. The microwave power has been kept as either 2mWo r0 .2 mW depending on the signal quality.E PR measurements have been recorded at room temperature (RT), 100 Ka nd 15 K.
DFT calculations:The frequency analysis for the three non-periodic model systems (benzoate paddle-wheels with Cu 2 ,F eCu and Fe 2 metal dimers, axial positions coordinated with H 2 Oi nc ase of Cu 2 + and Cl À in case of Fe 3 + ;s ee Figure S5) were performed using the B3LYP + D3 [32] level of theory with cc-PVDZ basis sets for C, H, O, and Cl and cc-pVTZ for Cu and Fe. [33] In all cases, the high-spin ferromagnetic coupled state was used (Cu 2 : S = 1, CuFe: S = 3, Fe 2 : S = 5) as indicated by the EPR results. In case of Cu 2 ,t he triplet state is energetically slightly above the open shell singlet but shows virtually the same normal modes. [34] The optimization without any symmetry constraints and the analytic computation of the Hessian matrix have been performed with the TURBOMOLE (V7.3) suite of programs. [35]