Complementary operando insights into the activation of multicomponent selective propylene oxidation catalysts Journal of Catalysis

Two Bi–Mo–Co–Fe–O catalysts were synthesized by ﬂame spray pyrolysis and tested for their catalytic performance in selective oxidation of propylene to acrolein. Pronounced structural changes during tem-perature–programmed oxidation and reaction were observed by operando X–ray absorption spec- troscopy, X–ray diffraction, Raman spectroscopy, and thermogravimetric analysis. During oxidative treatment, mainly binary oxide phases ( a –Bi 2 Mo 3 O 12 , b –CoMoO 4 , Fe 2 (MoO 4 ) 3 ) were observed, but single (MoO 3 ) or ternary (Bi 3 (FeO 4 )(MoO 4 ) 2 ) oxides also formed depending on the relative elemental catalyst composition. During propylene oxidation, the reduction of Fe 3+ to Fe 2+ led to a strong rise in activity and induced further phase transformations. MoO 3 was found to be unselective towards acrolein but was essential in binding other single oxides. The formation of b –Co 0.7 Fe 0.3 MoO 4 and Bi 3 (FeO 4 )(MoO 4 ) 2 as well as their synergistic interplay with a –Bi 2 Mo 3 O 12 are key factors for high performance. The combi- nation of complementary operando methods was crucial to reveal new structure–activity/selectivity correlations, therefore bridging the knowledge gap between simpliﬁed model systems and complex applied catalysts. (cid:1) 2021 The Authors. Published by Elsevier Inc. ThisisanopenaccessarticleundertheCCBYlicense(http:// creativecommons.org/licenses/by/4.0/).


Introduction
Selective oxidation reactions, which have been a major focus in the work of Jerzy Haber and Robert K. Grasselli, are key steps in chemical industry for the functionalization of hydrocarbons [1][2][3]. Among these, the selective oxidation of propylene to acrolein plays an important role. Global acrolein production sums up to $5Á10 6 t a -1 [4], and optimization of this process has therefore attracted increased research attention within the last years [5]. Bismuth molybdates (Bi-Mo-O) are well-known as highly selective catalysts in the oxidation of propylene to acrolein [6]. The addition of metals such as Fe, Ni, and Co has furthermore led to dramatic increases in performance via the presence of synergistically acting mixtures of individual phases [7][8][9]. In turn, the structural complexity of the original bismuth molybdate systems has continuously increased, making it difficult to obtain a full understanding of the workings of the catalyst, particularly the link between structure, activity and selectivity. Many studies have focused on basic or binary systems of bismuth molybdates (a-Bi 2 Mo 3 O 12 , b-Bi 2 Mo 2 O 9 , c-Bi 2 MoO 6 ), leading to different and occasionally conflicting statements about their activity and selectivity towards acrolein [10][11][12][13][14][15][16]. Even though the redox properties and thus the activity and selectivity depend on the Bi-Mo-O structure [17][18][19], these properties change when bismuth molybdates are intermixed leading to improved performance [19].
Despite their superior performance, relatively few studies have addressed the complex nature of multicomponent catalysts, allowing only a basic understanding of their working principles. Since the catalyst structure is affected by the composition as well as the preparation method [8], to achieve. Nevertheless, in the 1970s different particle models were developed in an attempt to categorize the key functions of multicomponent systems. First, Wolfs and co-workers [20,21] proposed a simplified particle model for the quaternary system Bi 1 Fe 3 -Co 8 Mo 12 O x . It consisted of a core formed mainly by CoMoO 4 (host structure) and islands of Fe 2 (MoO 4 ) 3 /FeMoO 4 (Fe 3+/2+ redox couple improving oxygen transport). The shell was made up by bismuth molybdates, acting as active centers. Subsequently, the model was revised mainly by including MoO 3 [8]. Over time, general trends for such multicomponent catalysts were established as reviewed by Moro-Oka and Ueda [8]. To 3 , and free MoO 3 have been reported [20][21][22][23][24][25][26][27][28][29], highlighting the complex nature. Hence, the mutual influence of the different phases on the catalytic performance of the finally active catalyst is still not fully clear yet.
Since it is generally accepted for Bi-Mo-O and Bi-Mo-Co-Fe-O catalysts that propylene oxidation to acrolein proceeds via the Mars-van Krevelen mechanism [8,22,[30][31][32][33][34], lattice oxygen plays a crucial role within the catalytic cycle. In four-component Bi-Mo-Co-Fe-O, iron molybdates (i.e., Fe 2 (MoO 4 ) 3 , FeMoO 4 ) or mixed cobalt iron molybdates (Co x Fe 1-x MoO 4 ) are ascribed for oxygen activation and migration through the bulk to the active center (Bi-Mo-O) [20,22,23,35]. Thus, oxygen activation and its insertion into propylene probably take place at different sites [7,8,34]. Consequently, a close contact of the different phases is essential to enable their cooperation for increased catalytic performance [7,13,36]. To study the phase cooperation in Bi-Mo-Co-Fe-O catalysts, and to correlate their structure with activity, mainly (mechanical) mixtures of individual phases were used [24][25][26]35,37]. Even though such mixtures enabled a basic understanding of the catalyst working principles, their simplified nature does not fully represent real multicomponent catalysts. On the other hand, studies on multicomponent catalysts [20,23] often fail to fully unveil the numerous structural changes occurring on stream [28,29]. Thus, complementary methods are required to follow the changes induced by calcination, which is typically carried out after synthesis [20,23,27,28,[38][39][40], and under reaction conditions (e.g., reduction, phase formation/transformation/decomposition, morphological changes) [21,23,25,26,28,41,42] and to finally correlate these changes to catalytic function. Ideally, this demands a combination of in situ [23,35] and operando [27] studies in order to draw meaningful conclusions.
In the present work, we studied the structural evolution of two exemplary Bi-Mo-Co-Fe-O catalysts [20,28] during temperatureprogrammed treatment under oxidizing (TPO) and reaction conditions (TPRM) by complementary operando methods. The aim was to better understand phase formation and evolution during potential thermal pretreatment (i.e., calcination) and during activation of the catalysts as well as the influence of specific phases on catalytic performance. Due to the complex nature of the investigated systems, complementary techniques were necessary to ensure a profound understanding of the structural changes observed with respect to elemental composition, gas atmosphere, and temperature. Thermal analysis, synchrotron X-ray absorption spectroscopy (XAS), synchrotron X-ray diffraction (XRD) combined with Rietveld refinement, and Raman spectroscopy (RS) were applied to identify crystalline and amorphous phases, the evolution and transformation of active phases, short and long range structure as well as the oxidation states of the elements present. Hence, we aimed at detailed operando insights into the processes during catalyst activation to revise our understanding of Bi-Mo-Co-Fe-O catalysts and the metal oxide phases present therein.

Catalyst preparation
Two Bi-Mo-Co-Fe-O catalysts, which differ in their relative elemental composition (FSP-1 and FSP-2, see Table 1), were synthesized via flame spray pyrolysis (FSP) in a single step using a setup similar to that reported by Mädler et al. [43] and our earlier studies [19,28]. In general, FSP is an attractive method for the synthesis of multicomponent Bi-Mo-Co-Fe-O catalysts as it leads to a homogenous elemental distribution, small crystallite size, and large specific surface area [28]. The total volume of the precursor solution was 250 mL with a total metal precursor concentration of 0.30 M (see Table 1). First, bis(2,4-pentanedionato)molybde num(VI) dioxide (MoO 2 (acac) 2 , Alfa Aesar) was dissolved in 60 mL methanol (MeOH, VWR Chemicals) and sonicated until a clear solution was obtained. Bismuth(III) acetate (Bi(OAc) 3 , Sigma Aldrich) and 40 mL acetic acid (HOAc, Sigma Aldrich) were added to the solution and sonicated again until no more solids were visible. Subsequently, cobalt(II) nitrate hexahydrate (Co(NO 3 ) 2 Á6H 2 O, Merck) and tris(2,4-pentanedionato)iron(III) (Fe(acac) 3 , Fluka Chemika) were added. A total volume of 250 mL was achieved by adding a 3/2 mixture (v/v) of MeOH/HOAc. The solution was sonicated for additional 30 min.
For spraying, the obtained solution was transferred into a 50 mL syringe (Henke-Sass Wolf). A syringe pump (Legato 210, KD Scientific) was used to achieve a constant solution dosage rate of 5 mL min À1 through the FSP nozzle. The nozzle itself consisted of a holder system and an insertable steel needle (GA22/60 mm/ PST3, Hamilton). The dosed liquid was finely dispersed at the needle tip by 5.0 L min À1 O 2 (N48, Air Liquide) at 3 bar back-pressure. The needle-surrounding flame ring, which ignites the solution, was formed by a mixture of 0.75 L min À1 CH 4 (N25, Air Liquide) and 1.6 L min À1 O 2 . Controlled gas dosage was realized by mass flow controllers (Bronkhorst). The brownish powder obtained was collected on a glass fiber filter (Ø = 24 cm, GF6, GE Healthcare) via a vacuum pump (R5, Busch GmbH). The filter and needle holder system were constantly water cooled. For each 50 mL syringe batch a new filter was used. Last, the obtained batches were scratched off the filters, combined, and sieved through a 600 mm mesh. The resulting powders were used as-prepared unless stated elsewhere.

Catalytic tests and conventional catalyst characterization
Catalytic performance tests in selective propylene oxidation were performed using a fixed-bed testing setup with an on-line gas chromatograph (GC) for product analysis. The testing procedure comprised different temperature points (T = 345-400°C) at which the total gas flow was varied (N 2 /O 2 /C 3 H 6 /H 2 O = 70/14/8/8 vol%, WHSV C3H6 = 1.14-3.42 h À1 ). The catalysts were characterized ex situ before and after catalytic testing by inductively coupled plasma optical emission spectrometry (ICP-OES), nitrogen physisorption, Raman spectroscopy (RS), laboratory powder X-ray diffraction (XRD), X-ray absorption spectroscopy (XAS), and transmission electron microscopy with energy-dispersive Xray spectroscopy (TEM-EDX). Details about testing and ex situ characterization are given in the Supporting Information. Simultaneous thermogravimetry, differential scanning calorimetry and mass spectrometry (TG-DSC-MS) was performed with a STA 449 F3 Jupiter Ò and a QMS 403 Aëolos (NETZSCH) using alumina crucibles for TG-DSC-MS, and quartz glass crucibles in the case of TG-MS studies. Due to catalytic activity of the thermocouples present at the TG-DSC-MS holder, only TG-MS experiments (without DSC) could be performed under propylene containing atmosphere. 25.1 mg of the as-prepared powders after FSP were filled in the glass crucible for TG-MS and $10 mg were filled in the alumina crucible for TG-DSC-MS. An empty second crucible was used as reference during TG-DSC-MS studies. The whole setup was flushed at ambient temperature with the corresponding gas mixture (oxidizing atmosphere: Ar/O 2 = 80/20 vol%; reaction gas mixture: Ar/O 2 /C 3 H 6 = 80/12/8 vol%; 100 mL min À1 total flow) overnight. The dosage of H 2 O in the gas feed for the reaction gas mixture was not possible due to instrumental limitations. The measurement sequence was as follows: isothermal at 30°C for 2 h, heating to 600°C (5°C min À1 ), 10 min isothermal, fast cooling to ambient temperature. A reference measurement performed with an empty crucible was subtracted from the obtained TG data of the samples to account for the buoyancy effect. Mass changes correspond to the first mass datapoint of the isothermal period at 30°C.

Operando setup for spectroscopy and diffraction studies
Further in situ and operando studies were performed by XAS, XRD, and RS using a fixed-bed quartz glass capillary microreactor [44] (sieve fraction 100-200 mm, Ø = 1 mm, 10 mm wall thickness, Hilgenberg GmbH). The same gas mixtures (oxidizing conditions (TPO): He/O 2 = 88/12 vol%; reaction conditions (TPRM): He/O 2 / C 3 H 6 /H 2 O = 72/12/8/8 vol%; 10 mL min À1 total flow) and temperature ranges (100-600°C) were applied in all cases. Prior to the TPO/TPRM experiments, the temperature inside the capillary was calibrated and the individual heating efficiency was considered. Hence, temperatures given for in situ and operando experiments refer to the calibrated value inside the capillary, which was measured by an inserted type K thermocouple, and in the case of XRD by the thermal lattice expansion of a silver reference. Controlled gas dosing was achieved by mass flow controllers (Bronkhorst) with water vapor dosed using a self-built heated steel saturator. Fast switching between different gas atmospheres was realized by a 4-way valve (VICI). Gas lines were heated to 180°C to prevent water and product condensation. An on-line mass spectrometer (OMNI Star GSD 320, Pfeiffer Vacuum) was used to analyze the product gas mixture.

Operando X-ray absorption spectroscopy
Operando XAS was performed sequentially at all four metal edges (Mo K, Bi L 3 , Co K, Fe K) in a single run in transmission mode at the ROCK beamline (SOLEIL, Saint-Aubin, France). The storage ring was operated at 2.75 GeV under top-up mode at 500 mA. The unique infrastructure available at ROCK enables fast edge changing [45,46], and thus the acquisition of all four metal edges during a single run. The Si(2 2 0) monochromator was used for data acquisition at Mo K-edge, while the Si(1 1 1) monochromator was used for acquisition at the Bi L 3 -, Co K-, and Fe K-edges, respectively (see Table S4). The as-prepared samples after FSP synthesis were mixed with a-Al 2 O 3 (1/3 (m/m) catalyst/a-Al 2 O 3 ), ground, pressed, granulated, and sieved to give the desired fraction. About 7 mg of the sieve fraction was placed in a quartz glass capillary, resulting in a bed length of 10 mm. The capillary was heated with a gas blower (FMB Oxford) in the range of 100-600°C (1.5°-C min À1 ). XAS data was acquired in the middle of the catalyst bed at 2 Hz. To acquire data for all metals in a single experiment, a loop in the sequence of Mo K-, Bi L 3 -and Fe K-/Co K-edge (recorded in a single scan) was performed. At constant temperatures, acquisition at Mo K-edge lasted 5 min, whereas at Bi L 3and Fe K-/Co K-edges it lasted 10 min each. For better temperature resolution, the acquisition time during heating was reduced to 2 min (Mo K), 4 min (Bi L 3 ) and 4 min (Fe K/Co K), respectively. One complete acquisition loop during heating lasted $12 min. Spectra of the initial and final state of the catalysts during the operando measurements were recorded under He atmosphere. The spectra of one acquisition period were averaged to produce a single spectrum for each edge. Energy calibration, averaging, background subtraction, and normalization were done by the beamline software [72]. For further data treatment, the software package IFEFFIT [47] was used. More details of XAS data acquisition and analysis are given in the Supporting Information.

Operando synchrotron X-ray diffraction
The operando XRD studies were conducted at the Swiss-Norwegian beamline (SNBL) BM01 at the European Synchrotron Radiation Facility (ESRF, Grenoble, France). The experiments were performed with a monochromatic beam (k = 0.62779 Å, $250x250 mm 2 ). The detector (PILATUS 2 M, Dectris) was calibrated using LaB 6 powder. 2D diffraction patterns were recorded in the middle of the catalyst bed using the PILATUS@SNBL diffractometer [48] with an acquisition time of 30 s. Azimuthal integration of the acquired 2D images was done with Bubble software [48]. About 5 mg of the undiluted sample was filled in the capillary, resulting in a bed length of 5 mm. For heating of the capillary reactor (100-600°C, 5°C min À1 ), a gas blower available at the beamline was used. Sequential Rietveld refinement (2h range = 3-25°) was performed with the software TOPAS [49] (v6, Bruker AXS) using references available in the Inorganic Crystal Structure Database (ICSD, see Table S6). From Rietveld refinement crystalline phase amounts, crystallite sizes and lattice parameters were obtained. More information about sequential Rietveld refinement can be found in the Supporting Information.

Operando Raman spectroscopy
Operando Raman spectroscopy was performed with an inVia Raman spectrometer (Renishaw) equipped with a frequency doubled Nd:YAG laser (532 nm, $100 mW at the source). The laser beam was focused on the capillary via a video and optical fiber probe (Renishaw) with a long working distance objective. For heating of the capillary reactor (100-600°C, 1.5°C min À1 ), a gas blower (LE MINI SENSOR KIT, Leister Technologies) was used. Raman spectra were recorded continuously (10% laser intensity, 120 s acquisition time, 2400 lines mm À1 grating resulting in a spectral range of $50-1300 cm À1 ). The laser intensity at the sample was measured prior to the experiments and amounted to $4.3 mW. To account for localized probing of the heterogeneous multicomponent systems (Ø laser spot = $70 mm), the fiber optic was moved periodically (1 period min À1 ) parallel to the capillary (center ± 1 mm) via an automated xy-stage. This produced averaged spectra repre- senting the whole sample rather than individual heterogenous regions. Data treatment, which included cosmic ray removal, noise filtering, truncation, and baseline subtraction, was performed with WiRE 4.4 (Renishaw). Assignment of Raman bands was carried out based on the metal oxide phases listed in Table S5 and the references listed therein.

Catalytic performance and conventional characterization of FSPmade Bi-Mo-Co-Fe-O catalysts
Two Bi-Mo-Co-Fe-O catalysts with different metal ratios (see Table 1) were prepared by FSP. ICP-OES results (see Table S1) indicated a good match ($0.2%) between the calculated and experimentally determined metal composition. This underlines the precise control over elemental composition for FSP-made multicomponent catalysts. Furthermore, the metal fractions did not change during testing, indicating high stability within the time scale investigated, i.e., no Mo loss due to the formation of volatile Mo compounds in the presence of water [5,50,51]. N 2 physisorption revealed a surface area of 82 and 80 m 2 g À1 for FSP-1 and FSP-2 in the as-prepared state (denoted as -F), respectively. For comparison, classical preparation methods like co-precipitation typically lead to small surface areas of maximum 7 m 2 g À1 [8]. Additionally, the laboratory XRD patterns of the as-prepared samples ( Fig. 1c) exhibited only a few, rather broad reflections, indicating small crystallite sizes and a high amount of amorphous phases. The few reflections present could be assigned to b-CoMoO 4 /b-Co 0.7 Fe 0.3 MoO 4 and CoO. Amorphous phases were detected by RS, including corresponding bands for b-CoMoO 4 /b-Co 0.7 Fe 0.3 MoO 4 (Fig. 1d). Furthermore, TEM-EDX (Table S2 and Table S3) indicated a rather good mixing of the elements.
Subsequently, both as-prepared samples were tested for their catalytic performance in selective oxidation of propylene in a lab-scale fixed-bed reactor (for details see Supporting Information). The testing results for FSP-1 and FSP-2 are given in Fig. 1a and b, respectively. Beside the target product acrolein, the byproducts CO, CO 2 (both are cumulated to CO x ), and acrylic acid were observed. In general, both catalysts exhibited high propylene conversion compared to bismuth molybdates [19] combined with high selectivity towards acrolein. Furthermore, with increasing WHSV the propylene conversion decreased while the acrolein selectivity increased. This trend is similar to binary bismuth molybdates, but the four-component systems showed a far higher activity (propylene conversion increased by a factor of 2-3 for same WHSV values compared to two-component bismuth molybdates [19]). At 345°C, FSP-1 and FSP-2 were already close to their maximal performance, indicating high activity already in the low temperature range. Notably, FSP-2 showed superior propylene conversion compared to FSP-1, combined with higher acrolein selectivity over the whole temperature range when similar propylene conversion levels are compared. For instance, at 400°C oven temperature and a similar propylene conversion of $60% (different WHSVs), FSP-1 exhibited an acrolein selectivity of 83% and an acrylic acid selectivity of 12%, while FSP-2 was highly selective towards acrolein (90% and 5%, respectively). However, it took more than one day until the first activity point (345°C, 1.14 h À1 ) could be acquired due to changes in selectivity as monitored by an online oxygen sensor. Differences in activity for the first measurement points at 345°C might arise from latent changes in catalyst structure/phase composition [5,28,52,53] influencing the performance.
Post-testing characterization revealed a drastic decrease in surface area to 4 (FSP-1-R) and 3 m 2 g À1 (FSP-2-R), respectively. Nevertheless, the catalysts still showed high activity, emphasizing the predominant role of the metal oxide phase composition forming on stream. Additionally, catalytic testing led to increased crystallinity as indicated by sharper XRD reflections and Raman bands (see Fig. 1c and d, respectively). Furthermore, the phase composition changed as the reflections corresponding to CoO disappeared, while for FSP-1 signals corresponding to MoO 3 and a-Bi 2 Mo 3 O 12 and for FSP-2 signals assigned to a-Bi 2 Mo 3 O 12 and Bi 3 FeMo 2 O 12 appeared. The formation and segregation of metal oxide phases as well as an increased particle size were confirmed by TEM-EDX (see Table S2 and Table S3). In the case of FSP-1, regions with an atomic ratio corresponding to Co 0.7 Fe 0.3 MoO 4 , CoMoO 4 , a-Bi 2 Mo 3 -O 12 , and MoO 3 could be revealed after catalytic testing. Notably, the elemental composition at different spots differed drastically, indicating a separation of phases on the nm level. In the case of FSP-2, the variations at the different spots after testing were far less notable compared to FSP-1. Especially the Co/Mo and Fe/Mo ratios were quite constant, indicating the preferential formation of co-located CoMoO 4 and FeMoO 4 , or mixed Co x Fe 1-x MoO 4 . Mo segregation was not observed, pointing towards the absence of MoO 3 in FSP-2. Ex situ XAS revealed the reduction of Fe 3+ to Fe 2+ . However, due to phase changes caused by grinding (see Supporting Information) to prepare pellets of the catalysts after testing, ex situ XAS data are not further discussed as this data was not considered representative of the catalyst state after testing. Note that phase changes induced by grinding were only found to affect the catalysts after testing.
Overall, ex situ characterization highlighted the nano-crystalline nature of both FSP-made catalysts after synthesis. During testing, both catalysts, particularly FSP-2, showed a high activity in propylene oxidation with a high selectivity towards acrolein. However, pronounced crystallization processes and phase transformations took place during time on stream. To evaluate the influence of gas composition and temperature on the formation of phases, operando characterization was performed under oxidizing and reaction conditions, in particular to investigate the role of Fe 2+ /Fe 3+ redox couples, which are known to play an important role in the catalytic cycle [8,34].

Phase formation during TPO
First, the structural evolution of the samples under oxidative treatment was investigated as catalysts are often calcined prior to testing. However, little is known about the influence of such a thermal pretreatment on the metal oxide phase composition of Bi-Mo-Co-Fe-O catalysts. Furthermore, the TPO experiments then also serve as a reference for the changes occurring under reaction gas atmosphere. In this way, it is possible to judge the stability of certain metal oxide phases with respect to gas atmosphere, and thus their relevance for catalysis.

In situ thermal analysis combined with mass spectrometry
The combined TG-DSC-MS experiments enabled simultaneous monitoring of weight changes (TG), exothermic/endothermic processes (DSC), and gas phase composition (MS) during heating under oxidizing conditions. As shown in Fig. 2, both catalysts exhibited similar TG curves characterized by a weight loss of $2% from 100 to 350°C. This weight loss was assigned via MS to the desorption/release of H 2 O and the desorption/formation of CO 2 . Subsequently, the mass was almost constant. However, significant differences in the DSC curves and their derivatives were observed for temperatures above 300°C (see Fig. 2c). While FSP-1 exhibited in the derivative peaks at around 340, 380, 415, 570 (with a broad shoulder towards lower temperatures) and >590°C, FSP-2 showed peaks at around 345, 365, 445 and 565°C. This indicates that different phases were formed during the oxidative treatment. Hence, the initial catalyst composition as well as the temperature influence the phase composition of both samples. To investigate the phase formation processes during TPO, complementary experiments were performed by in situ XAS, XRD, and RS.

In situ X-ray absorption spectroscopy
In situ multi-edge XAS was performed to reveal the changes in individual metal constituents during TPO. The induced changes for all four metal edges in the case of FSP-1 are given in Fig. 3a-d.
Mo K-edge X-ray absorption near edge structure (XANES) data exhibits two significant features A and B. Feature A ($20,007 eV) is attributed to the dipole-forbidden/quadrupole-allowed 1s-4d transition, which is primarily associated to tetrahedral geometry like in Fe 2 (MoO 4 ) 3 , but also present with weak intensity in distorted octahedral geometry. Feature B ($20,027 eV) is assigned to the dipole-allowed 1s-5p transition and is characteristic for (distorted) octahedral geometry such as in MoO 3 [54]. Comparing the Mo K-edge spectra at 100°C before and after TPO up to 600°C revealed an increased tetrahedral nature of Mo coordination (Fig. 3a).
The obtained Mo K-edge XANES spectra during heating (Fig. 3e) were analyzed by linear combination fitting (LCF) using reference compounds for octahedral (i.e., MoO 3 ) and tetrahedral (i.e., Fe 2 (-MoO 4 ) 3 ) Mo coordination to follow the temperature-dependent changes during TPO (see Fig. 3f). LCF initially revealed a mixture of both tetrahedral and octahedral nature. During TPO, first a slight decrease in octahedral nature from 41 to 34% up to 330°C was observed, followed by a short but significant increase up to 38%. This increase indicates the formation of an octahedrally coordinated Mo phase such as MoO 3 . After a plateau until $450°C, another phase formation or transition took place as the tetrahedral amount increased from 65 to finally 86% (e.g., formation of Fe 2 (-MoO 4 ) 3 ). The transformation towards more tetrahedral nature was additionally indicated by Fourier transform extended X-ray absorption fine structure (FT EXAFS) fitting (see Table S11) as the coordination number (CN) of Mo-O increased from initially 2.9 to finally 3.5, which is close to the ideal value of 4. Furthermore, an increased ordering/crystallization of the sample took place since the backscattering from higher shells (Mo-Fe and Mo-Mo) significantly increased (see Table S11 and Figure S12).
For the Bi L 3 -edge (Fig. 3b) and Co K-edge (Fig. 3c), the changes in XANES region during TPO were less pronounced, but still provide important information for phase analysis. The white line of Bi L 3 -edge XANES data shifted towards lower energy and mostly matched the spectrum of a-Bi 2 Mo 3 O 12 in the final state. Feature C, which is characteristic for Bi 3 (FeO 4 )(MoO 4 ) 2 (see Figure S10), was absent in this case. For Co K-edge XANES, a weak feature D was present in all spectra revealing octahedral coordination. Additionally, the absence of feature E, which is characteristic for a- CoMoO 4 , indicates that the sample consisted mostly of b-CoMoO 4 (c.f. ref. [55]). Furthermore, FT EXAFS fitting revealed dramatic changes for Co as a pronounced backscattering from higher shells (Co-Co and Co-Mo) was only observed after heat-up, while the Co-Co path at 3.00 Å, present in CoO model, disappeared (see Table S11, Figure S11, and Figure S12). This indicates a transformation of initially present CoO into b-CoMoO 4 .
For the initial state at Fe K-edge (Fig. 3d), features F ($7,114 eV), G ($7,122 eV), and H ($7,132 eV) are similar to those of the Fe 2 (MoO 4 ) 3 reference. LCF using Fe 2 (MoO 4 ) 3 and Fe 2 O 3 as references revealed a mixture of 53% and 47%, respectively. During TPO, the features F, G, and H were enhanced showing an increased content of Fe 2 (MoO 4 ) 3 (66%) and lower Fe 2 O 3 (34%), in line with the results from Mo K-edge. In the FT EXAFS after TPO the Fe-Fe backscattering shifted towards higher distances and the appearance of new Fe-Mo backscattering (see Table S11 and Figure S12) was observed, thus indicating the formation of Fe 2 (MoO 4 ) 3 phase during TPO.
In the case of FSP-2, almost the same changes were induced by TPO (see Fig. 4, Figure S13, Figure S14 and Table S12) compared to FSP-1. Mo K-edge XANES data (Fig. 4a, e) showed transition of Mo towards tetrahedral coordination. LCF of Mo K-edge XANES data during heating (Fig. 4f) revealed an initial state of 35% MoO 3 (octahedral coordination) and 65% Fe 2 (MoO 4 ) 3 (tetrahedral coordination), which is close to the initial state of FSP-1 (41% and 59%, respectively). During heating, the tetrahedral nature increased with a pronounced change from $70 to 82% in the temperature range 330-400°C. While FSP-1 showed an increase in octahedral nature in this temperature range (Fig. 3f), such a trend was not observed for FSP-2. In comparison, FSP-2 reached tetrahedral nature already at much lower temperatures. Hence, although the final phase composition was similar for both samples, the evolution of the phases involving Mo in sample FSP-2 differed from those in FSP-1.
For Bi L 3 -edge XANES spectra of FSP-2 (Fig. 4b), a shift of the white line towards lower energy was observed as in the case of FSP-1. However, the final spectrum exhibits a feature C ($13,460 eV) similar to that present in the Bi 3 (FeO 4 )(MoO 4 ) 2 reference (see Figure S10) but not observed in a-Bi 2 Mo 3 O 12 . The course of the Bi L 3 -edge XANES spectra during heating (see Figure S13) revealed that the formation of this ternary phase started at $400°C. The formation of Bi 3 (FeO 4 )(MoO 4 ) 2 was further supported by Fe K-edge XANES spectra (Fig. 4d), where LCF using Fe 2 (MoO 4 ) 3 , Fe 2 O 3 , and Bi 3 (FeO 4 )(MoO 4 ) 2 references revealed contribution of these phases to be 43, 42, and 16% at 600°C, respectively. For comparison, the phase contribution for initial state at 100°C was observed to be 47, 47 and 6%, while in the final state after cooling to 100°C it was 55, 36 and 9%. Here also, Co K-edge XANES data (Fig. 4c) confirmed the presence of b-CoMoO 4 , and FT EXAFS points towards the transformation of CoO into b-CoMoO 4 , similar to FSP-1. FT EXAFS fitting for all edges (see Table S12 and Figure S14) revealed an increased backscattering from higher shells, which is linked both to the formation of new phases and increased structural ordering (i.e., crystallization) during TPO.
In This reveals that while both samples showed increased long-range order, indicating an increase in particle size most likely due to crystallization, the precise phase formation processes between them differed significantly. As it is challenging to identify all phases present using XAS alone due to the averaging nature of the method, additional techniques sensitive to crystalline and amorphous phases were applied to further study the processes occurring during TPO.

In situ synchrotron X-ray diffraction and Raman spectroscopy
Synchrotron XRD combined with Rietveld refinement enabled precise crystalline phase analysis compared to laboratory XRD (see Figure S5), and furthermore allowed quantification of the crystalline phase amounts present. Moreover, RS provided additional complementary insights on amorphous phases. The phase analysis results for both samples during TPO are shown in Fig. 5. Note that an overview of metal oxide phases considered during phase analysis is given in Table S5.
Both as-prepared catalysts showed only the presence of crystalline b-CoMoO 4 and CoO by XRD, while the corresponding Raman spectra exhibited strong features originating from b-CoMoO 4 [56][57][58]. During heating of FSP-1, reflections corresponding to a-Bi 2 Mo 3 O 12 , MoO 3 and Fe 2 (MoO 4 ) 3 appeared (Fig. 5a, Figure S21). This was in line with the changes observed by RS (Fig. 5c, Figure S21), where characteristic bands of the corresponding phases (a-Bi 2 Mo 3 O 12 [19,59,60], MoO 3 [28,61], and Fe 2 (MoO 4 ) 3 [56,62]) were observed. Due to the formation of new crystalline phases, the amount of b-CoMoO 4 decreased during TPO from initially 95 to finally 39 wt%, which is mainly due to the low crystallinity (nano-and non-crystalline nature) in the initial state. Additionally, the amount of CoO continuously decreased from initially 5 to $2 wt% at 440°C (subsequently no refinement was possible) indicating that CoO reacted to form other (most likely binary) metal oxide phases.
Since the amount of MoO 3 and Fe 2 (MoO 4 ) 3 were almost constant during isothermal treatment at 600°C, it can be concluded that all Fe was bound within the Fe 2 (MoO 4 ) 3 phase. Furthermore, the formation of crystalline a-Bi 2 Mo 3 O 12 started at 420°C, while the new Raman band appearing at $895 cm À1 could be clearly assigned to a-Bi 2 Mo 3 O 12 . At $530°C a-Bi 2 Mo 3 O 12 reached a stable amount of $23 wt%, confirming the phase assignment from Bi L 3edge XAS. During the isothermal period at 600°C, no changes in crystalline phase composition could be observed. However, pronounced crystallization processes took place as the crystallite size of all phases increased (see Fig. 5a).
During TPO of FSP-2 (Fig. 5b, d), the same trends for Co containing phases were observed compared to FSP-1. However, the formation of phases was distinctly different from FSP-1, in agreement with XAS. For instance, no crystalline MoO 3 could be observed during TPO for FSP-2. Nevertheless, a very small MoO 3 amount was detected by RS as visible by the characteristic band at 995 cm À1 in the temperature range of around 350-460°C (see Figure S23).
As for FSP-1, a-Bi 2 Mo 3 O 12 was formed but its crystallization started at higher temperatures of 460°C, reaching a maximum amount of 17 wt% at $550°C. Subsequently, the amount of a-Bi 2 -Mo 3 O 12 decreased to 2 wt%, while Bi 3 (FeO 4 )(MoO 4 ) 2 was simulta-

Summary on the phase evolution in Bi-Mo-Co-Fe-O catalysts during TPO
Compared to the two-component bismuth molybdate systems, many more phase transformations occured during TPO in the two investigated four-component Bi-Mo-Co-Fe-O systems as unraveled by in situ TG-DSC-MS, XAS, RS, and XRD, with the different sensitivity of each technique providing complementary insights. For instance, MoO 3 formation during TPO of FSP-1 was only indicated by XAS but was clearly revealed by RS and XRD. An overview of the structural evolution of FSP-1 and FSP-2 during TPO is given in Fig. 6. . However, a temperature difference is expected between DSC and XRD analysis due to the low sensitivity of XRD for the initial crystalliza-tion of amorphous phases in these multiphase mixtures. On the other hand, RS is essential to monitor the formation of amorphous phases but might fail to identify some phases due to its limited material sensitivity.   processes. Subsequently, the Fe/Mo ratio influences the formation of binary (Fe 2 (MoO 4 ) 3 ) as well as ternary phases (Bi 3 (FeO 4 ) (MoO 4 ) 2 ), and thus also determines the presence of MoO 3 .

Thermogravimetric analysis combined with mass spectrometry
Combined TG-MS studies were carried out to follow possible reduction processes occurring under reaction conditions (e.g., to check for the (partial) Fe 3+ to Fe 2+ reduction which is known to occur under reaction conditions [33,42]) and to correlate them with the catalytic performance using on-line MS. Due to pronounced catalytic activity of the TG-MS setup itself under reaction conditions at higher temperatures, only datasets up to 500°C are shown in Fig. 7. Further, the empty crucible measurement (see Figure S8b) revealed that the influence of the TG setup started at $350°C. Thus, the influence of the setup itself should be considered when interpreting the catalytic activity results during TG-MS experiments.
Two pronounced weight losses were observed for both catalysts during TPRM (see Fig. 7). A first loss of $2 wt% occurred in the range of 60-200°C and was assigned to H 2 O desorption/release. A plateau was reached at $200°C, followed by product formation (increase of acrolein, CO 2 , H 2 O). Starting at $270°C, a pronounced second weight loss took place, which led to a strong increase in product formation including CO and acrylic acid. Notably, the product formation of both catalysts started at the same temperature, but FSP-2 showed a shift in the MS data inflection points of $10°C to lower temperature (see Figure S9 for direct comparison). For both catalysts, the inflection point for the second mass loss was at $330°C. Hence, the same reduction processes took place but were more pronounced in the case of FSP-2 (1.8 wt% vs. 1.4 wt% for FSP-1). This indicates a correlation between the catalyst reduction and the activity during TG-MS experiments, and thus during catalytic tests. The reduction process lasted to slightly higher temperatures for FSP-2, correlating with the more pronounced weight loss. From about 380-390°C on, the lower rate of weight loss indicates the presence of fewer reduceable species or transport limitations. This went in hand with a lower selectivity of the catalysts due to increased formation of overoxidation products such as acrylic acid, CO 2 , and CO, while the acrolein signal remained constant (FSP-1) or even decreased (FSP-2). However, at the same time the influence of the setup on the MS data has to be taken into account.

Operando X-ray absorption spectroscopy
To investigate the influence of reaction conditions and the associated reduction observed during TG-MS on the phase formation processes, multi-edge operando XAS was performed. XANES spectra at the studied metal edges for FSP-1 under isothermal conditions during TPRM together with selected reference compounds are given in Fig. 8a-d (for course see Figure S15). Similar to TPO, changes in Mo coordination were observed. However, at Fe K-edge, large changes were observed during TPRM compared to TPO. Strong reduction of Fe was uncovered with the final spectrum pointing towards complete reduction of Fe 3+ to Fe 2+ (Fig. 8d). As the final spectrum was close to FeMoO 4 (c.f. ref. [41,64]), most likely reduction of Fe 2 (MoO 4 ) 3 to FeMoO 4 occurred (see Eq. (2)).
LCF of the Mo K-edge spectra (Fig. 8e) was performed using MoO 3 and Fe 2 (MoO 4 ) 3 as references for octahedral and tetrahedral coordination, respectively (see section 3.2.2). With increasing temperature, the amount of MoO 3 slowly decreased from initially 42 to 36% at 315°C. Subsequently, the amount of MoO 3 temporarily increased up to 43% at $370°C, as also observed for oxidizing conditions (see Fig. 3f). Above 480°C, a pronounced transition towards tetrahedral nature was observed, resulting finally in a similar state as for TPO. Furthermore, FT EXAFS fitting at Mo K-edge (see Table S13 and Figure S16) revealed an increase in the CN of Mo-O at 1.76 Å from initially 2.9 to finally 3.6, which confirms stronger tetrahedral coordination. Further, a stronger backscattering from higher shells was observed, hence indicating crystallization of Mo containing phases.
In the case of Fe K-edge XANES spectra, the first and last spectrum during heating were used for LCF (Fig. 8f) due to poor fitting results from available reference spectra. LCF analysis revealed that the main reduction occurred between 310 and 415°C, as the amount of Fe 2+ changed from 18 to 100%. The reduction temperature range matches the interim MoO 3 formation observed at Mo Kedge, thus supporting the reaction according to Eq. (2). Furthermore, FT EXAFS fitting at Fe K-edge confirmed the reaction-induced loss of oxygen, as the CN of the Fe-O 1 path decreased from initially 5.2 to finally 4.5, indicative of tetrahedral b-FeMoO 4 [41]. An increased backscattering from higher shells was also observed, indicating further crystallization. The final spectra obtained at Bi L 3 - (Fig. 8b) and Co K-edge (Fig. 8c) together with FT EXAFS fitting at Co K-edge (see Table S13 and Figure S16) revealed the formation of a-Bi 2 Mo 3 O 12 and b-CoMoO 4 , respectively, which is similar to TPO.
The XANES spectra of FSP-2 at the studied metal edges under isothermal conditions during TPRM are shown in Fig. 9a-d together with selected references (for course see Figure S17). Similar changes were observed for Mo K-, Co K-, and Fe K-edges compared to FSP-1, including stronger tetrahedral nature for Mo, reduction of Fe 3+ to Fe 2+ , and formation of b-CoMoO 4 . However, different results were obtained at the Bi L 3 -edge ( Fig. 9b and Figure S17), where feature C was present after TPRM revealing the formation of Bi 3 (FeO 4 )(MoO 4 ) 2 . Thus, even though Fe reduction occurred, Fe was further incorporated into the a-Bi 2 Mo 3 O 12 structure [67]. However, the amount of Bi 3 (FeO 4 )(MoO 4 ) 2 phase was low as evident from the almost identical nature of Fe K-edge XANES spectra for FSP-1 and FSP-2 (see Figure S19). Fe K-edge FT EXAFS fitting (see Table S14 and Figure S18) including paths from b-Co 0.7 Fe 0.3 MoO 4 revealed decreased oxygen coordination in the first shell together with a higher bond distance for the final state, supporting the XANES results.
Under reaction conditions, LCF revealed differences in Mo coordination behavior (Fig. 9e) compared to FSP-1. Between 315 and 420°C, the amount of MoO 3 decreased from 32 to 25%. As MoO 3 is formed via the reduction of Fe 3+ to Fe 2+ (see Eq. (2)), it could be incorporated in a tetrahedrally coordinated molybdate structure (e.g., b-FeMoO 4 , b-Co 0.7 Fe 0.3 MoO 4 ). As the main changes in oxidation state occurred between 310 and 415°C (from 20 to 93% Fe 2+ , see Fig. 9f), the Fe reduction proceeded similar to FSP-1. Since TG-MS revealed a pronounced influence of the reduction process on the catalytic activity, a detailed comparison of Fe reduction and simultaneously measured catalytic activity during operando XAS measurements is given in Fig. 10. For higher temperatures (T > 500°C), the influence of the dilutant a-Al 2 O 3 (see Figure S20) should be kept in mind when evaluating the MS data during activation.
The Fe reduction rate (Fig. 10a bottom), which was derived from the reduction progress (Fig. 10a top) and corrected by the Fe amount (see Table 1), was highest at $345°C for both samples. However, in the case of FSP-2 the reduction started at slightly lower temperatures and lasted up to higher temperatures, what can be explained by the higher Fe amount present. The correlation between catalyst reduction and activity, which was already observed during TG-MS studies (Fig. 7), could be confirmed and traced back to the reduction of Fe. The earlier reduction of Fe 3+ to Fe 2+ for FSP-2 resulted in earlier product formation, while the higher amount of Fe present in FSP-2 led to higher activity (see Fig. 10b).
In summary, multi-edge XAS revealed several phase transformations during activation of the as-prepared catalysts. In particular, reduction of Fe 3+ to Fe 2+ led to increased activity. Co (mainly present as b-CoMoO 4 /b-Co 0.7 Fe 0.3 MoO 4 ) and Fe (mainly present as b-FeMoO 4 /b-Co 0.7 Fe 0.3 MoO 4 ) formed similar phases, though incorporation of Mo and Bi differed between the samples. Similar to TPO, the formation of ternary phase Bi 3 (FeO 4 )(MoO 4 ) 2 was only observed for FSP-2. Furthermore, FSP-1 showed a unique interim increase in octahedral Mo coordination and thus the (interim) formation of MoO 3 associated with the reduction of Fe, while for FSP-2 the released MoO 3 could be bound within tetrahedrally coordinated Mo phases. Additionally, the comparison of both samples revealed that the activity correlates with Fe amount and Fe 3+ to Fe 2+ reduction behavior. To further analyze the phase behavior, XRD and RS experiments were conducted under reaction conditions.

Operando X-ray diffraction and Raman spectroscopy
Corresponding results for XRD and RS analysis for FSP-1 and FSP-2 are given in Fig. 11. During heating under reaction conditions of FSP-1 (Fig. 11a, c, e), partially different phases were formed compared to TPO. While for oxidizing conditions Fe 2 (-MoO 4 ) 3 was observed by XAS, XRD and RS, this was not the case for reaction conditions. Furthermore, no FeMoO 4 phase could be observed by XRD nor RS and no other Fe containing phases were detected. Hence, it is very likely that b-Co 0.7 Fe 0.3 MoO 4 was formed during TPRM. Thus, b-Co 0.7 Fe 0.3 MoO 4 instead of b-CoMoO 4 was used for Rietveld refinement of the obtained XRD patterns. However, it should be noted that due to the similar structures of both compounds, it was not possible to unambiguously distinguish these by XRD within the complex phase mixture present. Therefore, only one structure was used for the refinement of an entire TPRM cycle, even though the Fe incorporation into the b-CoMoO 4 structure might have occurred at higher temperatures as indicated by an increased phase amount above 550°C (Fig. 11a).
For completeness, Rietveld refinement was also carried out using b-CoMoO 4 but resulted in higher R wp values (especially for higher temperatures) indicating a worse refinement (see However, a decrease in MoO 3 amount for temperatures above 550°C could not be confirmed by RS. This might be affected by different Raman scattering properties of b-CoMoO 4 and b-Co 0.7 Fe 0.3 -MoO 4 , respectively. Since MoO 3 is a strong Raman scatterer, amorphous MoO 3 was already observed at $330°C, which matches well the MoO 3 formation and Fe reduction process revealed by XAS (see Fig. 8e-f). As monitored by MS (Fig. 11e), the release or formation of MoO 3 during the catalyst reduction led to an interim plateau in acrolein formation at 330-345°C while formation of byproducts like CO and CO 2 was further increasing. Notably, especially CO 2 was formed as indicated by the rather high ion current for m/z = 44 (Fig. 11e). In the temperature range 370-410°C, a second, more pronounced effect of MoO 3 formation/crystallization was observed as the intensity ratio of Raman bands associated to MoO 3 compared to b-Co 0.7 Fe 0.3 MoO 4 was changing significantly (see Fig. 11c and Figure S25) and crystalline MoO 3 formed (see Fig. 11a). Notably, propylene conversion did not follow the trend of increasing activity with increasing temperature in this temperature range as it remained almost constant and only increased later on (T > 410°C). This indicates that (crystalline) MoO 3 does not contribute in a beneficial manner to catalytic activity of Bi-Mo-Co-Fe-O systems.
The structural changes during TPRM for FSP-2 are shown in Fig. 11b, d, f. As for FSP-1, the Rietveld refinement results for b-CoMoO 4 as main phase are given in Figure S31. According to Rietveld refinement (Fig. 11b) and RS (Fig. 11d), the phase formation processes for FSP-2 differed significantly compared to TPO conditions and also compared to FSP-1 under TPRM conditions. In addition to the initially present b-CoMoO 4 /b-Co 0. was only present as an interim phase in the case of FSP-2, as the corresponding signals disappeared above 535°C (RS) and 565°C (XRD). The observed temperature difference might originate from different heating rates in both experiments (XRD: 5°C min À1 , RS: 1.5°C min À1 ). Since the crystalline amount of b-Co 0.7 Fe 0.3 MoO 4 increased from $480°C on, it was likely formed according to Eq.
Since the formation of b-Co 0.7 Fe 0.3 MoO 4 occurred at lower temperatures compared to the formation of Bi 3 (FeO 4 )(MoO 4 ) 2 , it can be concluded that not all Fe oxides could be incorporated into the b-CoMoO 4 structure, and thus were still available for the formation of Bi 3 (FeO 4 )(MoO 4 ) 2 . The limiting factor for the Fe incorporation into b-CoMoO 4 was most probably the amount of available MoO 3 (Eq. (3)), which was influenced by the initial catalyst composition.
In the temperature range 330-400°C, FSP-2 exhibited high selectivity towards CO and CO 2 compared to acrolein as shown in Fig. 11f. In this temperature range, RS revealed the most pronounced changes in relative phase composition (Fig. 11d), as the amount of MoO 3 strongly increased, which was confirmed by XRD. Moreover, the disappearance of MoO 3 caused by its incorpo-  )) at $535°C led to a strong increase in performance. Not only propylene conversion increased but also the increase in the ion current of acrolein was more pronounced compared to the increase in by-products like CO and CO 2 . Thus, it can be concluded that MoO 3 acts as unselective phase in propylene oxidation to acrolein, at least in Bi-Mo-Co-Fe-O catalysts. This is also in line with the findings of Udalova et al. [39] for Co-Mo-Bi-Fe-Sb-K catalysts.
Furthermore, our results support the promoting effect of b-Co 0.7 Fe 0.3 MoO 4 , which was already described for mixtures of bismuth molybdates and mixed cobalt iron molybdates. The synergy effect between these metal oxide phases was traced back to facilitated electron and oxygen ion mobility when they are in close contact, and thus the increased efficiency of the redox mechanism according to the Mars-van Krevelen mechanism [25,26,37,68]. The higher amount of b-Co 0.7 Fe 0.3 MoO 4 in FSP-2 is most likely one origin for the better performance compared to FSP-1, which was also observed by MS during the operando Raman spectroscopic studies (see Figure S29). Such a trend is in line with the screening results of Millet et al.  During the course of TPRM, the reduction of Fe 3+ to Fe 2+ was observed for both samples. This reduction led to a strong increase in catalytic performance linked to formation of different phases compared to TPO. As shown in Fig. 12 [27,35]. Notably, we did not observe the formation of Fe 2 (MoO 4 ) 3 under reaction conditions. However, this might be influenced by the elemental catalyst composition as well as by the gas composition as the amount of O 2 in the gas feed has an influence on the Fe 3+ /Fe 2+ redox couple [27]. In general, our results on the activation of Bi-Mo-Co-Fe-O catalysts underline that more metal oxide phases than the typically investigated phases Co x Fe 1-x MoO 4 and a-Bi 2 - [25,26,35,37] need to be considered when studying the phase cooperation of Bi-Mo-Co-Fe-O systems.
As the individual phase composition of both catalysts after TPRM is similar to the state after catalytic testing (see Fig. 1), the operando experiments revealed the various processes taking place during transition from the as-prepared to the active state. In general, preferentially binary and ternary metal oxides were formed. However, the relative catalyst composition (i.e., metal ratios) had a strong influence on phase composition in the active state and thus catalytic performance. In the case of FSP-1, not all MoO 3 , which acts as unselective phase, could react to form binary or ternary metal oxide phases due to the high Mo amount (Mo/(Co + Fe) ratio = 1.76). For FSP-2 (Mo/(Co + Fe) ratio = 1.09), all free MoO 3 was incorporated in b-Co 0.7 Fe 0.3 MoO 4 .
On one hand, MoO 3 is inevitably formed during reduction of the as-prepared catalysts and subsequently binds free oxides that also act as unselective components (e.g., iron oxides [38]). Thus, a certain MoO 3 amount is required to obtain a well performing catalyst. On the other hand, an excess of MoO 3 decreases the catalytic performance as in the case of FSP-1. These findings confirm the results of Udalova et al. [39] for Co-Mo-Bi-Fe-Sb-K catalysts. However, it should be noted that not only MoO 3 binds free oxides forming binary metal molybdates (e.g., CoMoO 4 , FeMoO 4 ), but also binary phases can bind free oxides forming ternary metal oxide phases (e.g., b-Co 0.7 Fe 0.3 MoO 4 , Bi 3 (FeO 4 )(MoO 4 ) 2 ). The formation of these ternary metal oxides and their synergistic interaction with a-Bi 2 -Mo 3 O 12 appear to be responsible for the outstanding performance of FSP-2 compared to FSP-1. Notably, their sole presence is not sufficient if there is no mutual interaction between them, which might be influenced by the preparation technique. For instance, Bi 3 (FeO 4 )(MoO 4 ) 2 alone does not exhibit a higher activity than bismuth molybdates [69,70], but if it is in contact with bismuth molybdates and iron molybdates a pronounced performance increase can be observed [38,71].
The results shown here highlight the need to investigate real multicomponent catalysts, with both complex structure and intermixed phases. This is in contrast to rather simplified systems or phase mixtures which are often studied in literature but may not be representative of a real catalyst at work. Operando characterization is crucial to cover all relevant phases forming on stream and thus their interaction, which can help to rationalize the outstanding performance of such multicomponent catalysts compared to simplified model systems. Furthermore, a combination of complementary characterization tools is essential to compensate for the limitations in any individual method.

Conclusions
The manifold structural changes of two FSP-made Bi-Mo-Co-Fe-O catalysts during TPO and TPRM were monitored by complementary in situ and operando multi-edge XAS, synchrotron XRD combined with Rietveld refinement, RS, and TG-DSC-MS. Both catalysts underwent strong structural changes depending on relative elemental composition, applied temperature and gas atmosphere. Synchrotron-based methods are especially ideal for such multicomponent systems as XAS can monitor local structural changes of each element (Bi, Mo, Co, Fe) independently, while XRD allows to uncover the complex crystalline phase composition. Under oxidizing conditions (TPO) preferentially binary metal oxides (e.g., b-  reaction conditions and their synergistic interplay with a-Bi 2 Mo 3 -O 12 are required for an outstanding performance (FSP-2). This supports the earlier particle models of interacting phases but extends them at the same time by unravelling the phase formation processes during catalyst activation. Moreover, it is demonstrated that more metal oxide phases than suggested in these models need to be considered for Bi-Mo-Co-Fe-O catalysts.
The multimodal approach applied here enabled detailed operando insights into the various phase formation/transition processes in multicomponent Bi-Mo-Co-Fe-O catalysts and their influence on the performance in selective oxidation of propylene to acrolein. The combination of complementary operando techniques was essential for gaining insights into structure-activity/selectivity correlations within such complex catalyst systems. In future, these emerging operando techniques will provide new possibilities for studying multicomponent mixed metal oxide systems and the cooperation of phases therein, which is crucial in understanding the origin of high performance compared to two-component systems.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.