Architecture of industrial Bi-Mo-Co-Fe-K-O propene oxidation catalysts

Abstract

Industrial heterogeneous catalysts show high performance coupled with high material complexity. Deconvoluting this complexity into simplified models eases mechanistic studies. However, this approach dilutes the relevance because models are often less performing. We present a holistic approach to reveal the origin of high performance without losing the relevance by pivoting the system at an industrial benchmark. Combining kinetic and structural analyses, we show how the performance of Bi-Mo-Co-Fe-K-O industrial acrolein catalysts occurs. The surface BiMoO ensembles decorated with K supported on β-Co_1−xFe_xMoO₄ perform the propene oxidation, while the K-doped iron molybdate pools electrons to activate dioxygen. The nanostructured vacancy-rich and self-doped bulk phases ensure the charge transport between the two active sites. The features particular to the real system enable the high performance.

The material complexity of industrial catalysts is linked to the functional complexity that enables their high performance.

INTRODUCTION

Heterogeneous catalysis plays essential roles in supplying chemicals and fuels to society. Optimized industrial catalysts are mostly complex multicomponent systems promoted by one or more elements in addition to a base metal element that can perform the desired catalysis alone albeit with modest performance (1). Prominent examples of promoters include K and Al in Fe-based ammonia synthesis catalyst (2) and Zn and Al in Cu-based methanol synthesis catalyst (3, 4). Increased complexity added by promoters makes mechanistic understanding highly challenging and threatens the relevance of model studies dealing with simplified but less performing systems, often leading to controversial mechanistic views. For example, the origin of the high performance of the Cu/ZnO/Al₂O₃ methanol synthesis catalysts, one of the most studied systems, has remained a matter of intensive debate (5–7) despite decades of industrial use. A comprehensive mechanistic understanding of industrial catalysts has thus rarely been achieved.

Multicomponent molybdate (MCM) propene (amm)oxidation catalysts are used in the industrial production of acrylic acid (via acrolein intermediate) and acrylonitrile, two important commodity monomers with large annual production volumes of >15 metric tons in sum (8–10). The MCM catalysts represent a prominent multipromoted system with high material complexity and practical importance. While pure binary bismuth molybdates can form acrolein in propene oxidation with high selectivity, industrial MCM catalysts with far higher activity (factor of >20) contain at least three more promoter metal elements: divalent cations (Co²⁺ and/or Ni²⁺), Fe³⁺, and K⁺, forming a (minimum) five-component system (9–11). Omitting any of them drastically reduces the performance.

Typical elemental compositions of industrial acrolein catalysts, e.g., Mo₁₂Bi_0.6Co₇Fe₃K_0.08Si_1.6O_x (12) and Mo₁₂Bi_1.8Co_5.2Ni_2.8Fe_1.8K_0.1O_x (13), feature a high content of divalent Co²⁺ and/or Ni²⁺ and a Bi/Mo ratio remarkably lower than that of binary bismuth molybdates (e.g., α-Bi₂Mo₃O₁₂). The system contains diverse molybdate phases: cobalt molybdates (β- and α-CoMoO₄), iron molybdate Fe₂Mo₃O₁₂, and alpha bismuth molybdate α-Bi₂Mo₃O₁₂ (9–11). The α-Bi₂Mo₃O₁₂ in a scheelite-type structure is generally considered the relevant active component (14), whereas Fe-doped beta cobalt molybdate β-Co_1−xFe_xMoO₄ (8, 15, 16) and Fe₂Mo₃O₁₂ (17) are considered to act as a redox-mediating support that activates O₂ to replenish the lattice oxygen at the reduced α-Bi₂Mo₃O₁₂. The postulated reaction mechanisms involve two spatially separated types of sites, performing propene oxidation to acrolein and O₂ activation, respectively, while being connected by transport of oxygen atoms through the bulk (8–10).

These mechanistic insights have been largely derived by model studies using simplified materials, whose relevance remains to be proven. Moreover, little is known about the mechanistic roles of the K-promoter (18, 19). Here, we reveal the structure of industrial MCM catalysts and provide mechanistic explanation about how the complex structural features are linked to the performance. To establish structure-function relationships in a real system, we prepared sets of catalysts with different compositions consisting of up to six components (Bi-Mo-Co-Fe-K-Si) by an industrial recipe via coprecipitation, drying, and calcination (12).

RESULTS

Benchmarking of industrial MCM catalysts

First, we examined a composition Bi_0.6Mo₁₂Co₇Fe₃K_0.1O_x (denoted as K0.1) as a representative of industrial MCM with high performance and a typical industrial composition. Structural analysis by x-ray diffraction (XRD) and scanning transmission electron microscopy with energy-dispersive x-ray spectroscopy (STEM-EDXS) revealed particle aggregates consisting of β- and α-Co_1−xFe_xMoO₄, Fe₂Mo₃O₁₂, and α-Bi₂Mo₃O₁₂ [Fig. 1A (b and e) and figs. S1 to S3] with crystallite sizes of 20 to 80 nm (fig. S4) forming a macroporous structure with a low Brunauer-Emmett-Teller (BET) surface area of 4.7 m² g⁻¹ (table S1). Bismuth iron molybdate scheelite phase Bi₃FeMo₃O₁₂, a potentially relevant active phase (14), was not detected. In propene oxidation, K0.1 showed a propene conversion rate constant of 0.62 s⁻¹ m⁻² ml at 380°C with ~98% selectivity (#2 in table S2). The rate constant was ~90 times higher than that of α-Bi₂Mo₃O₁₂ (0.007 s⁻¹ m⁻² ml at 380°C; #6 in table S2) and at a similar level to the model catalyst representing α-Bi₂Mo₃O₁₂ deposited on β-Co_1−xFe_xMoO₄ (0.53 s⁻¹ m⁻² ml at 380°C; #5 table S2), suggesting possible relevance of the model catalyst. In addition, no detectable deactivation occurred over 340 hours on stream (fig. S5).

Fig. 1. Structural and functional analysis of MCM catalysts.

(A) Characterization of K0.1 (Bi_0.6Mo₁₂Co₇Fe₃K_0.1O_x). (a) Low-magnification high-angle annular dark-field (HAADF)–STEM image. (b) EDXS overlay mapping showing Mo (red), Co (green), Fe (blue), and Bi (cyan). Spectrally mixed colors indicate the copresence of multiple metals [CoMo (yellow) and FeMo (purple)]. EDXS mapping for Bi (c) and K (d). (e) Phase composition by XRD. (f) Medium magnification HAADF-STEM image with EDXS overlay color mapping of Mo and Bi. (g and h) High-resolution HAADF-STEM images. Enrichment of Bi at grain boundaries and at inner surface of embedded voids was observed (yellow arrows), while little Bi enrichment at the external surface was seen (green arrows). (i) Metal composition at the surface by x-ray photoelectron spectroscopy (XPS) compared to the bulk composition. (B) Propene oxidation performance of reference catalyst Bi_0.6Mo₁₂Co₇Fe₃K_0.08Si_1.6O_x (Ref. Bi0.6), Mo₁₂Co₇Fe₃K_0.08Si_1.6O_x (W/o Bi), and Bi_0.06Mo₁₂Co₇Fe₃K_0.08Si_1.6O_x (Imp. Bi0.06) prepared by post-impregnation of Bi on a Bi-free catalyst. The reaction temperature (T) is shown in the plot area. (C) Propene oxidation performance of catalysts with compositions varied within Bi_0.6Mo_11–13Co_5.5–8.5Fe_2–3.5K_0.08Si_1.6O_x. The plot shows T at X = 0.95 versus the metal ratio Mo/(Co + 1.5Fe + 1.5Bi). Test conditions for (B) and (C): One hundred–gram ring catalyst, P = 0.12 MPa, t = 3.6 s g ml⁻¹, feed gas C₃H₆/O₂/N₂ = 5/9.5/85.5 mol %, and time on stream = 3 day. (D) Fe K-edge x-ray absorption near-edge structure (XANES) spectrum of a spent MCM catalyst after several years of use compared with the fresh state.

Identification of the relevant BiMoO species for propene oxidation

Bi, as well as Mo, is unequivocally a part of the active site (9) because any catalysts without Bi show poor performance. The content of α-Bi₂Mo₃O₁₂ in the benchmark catalyst K0.1 was only 8% (Fig. 1A, e). STEM-EDXS analysis showed the presence of highly dispersed Bi species besides the bulk α-Bi₂Mo₃O₁₂ [Fig. 1A (c and f to h) and figs. S2 and S3]. The dispersed Bi did not form extensive overlayer or nanoparticles larger than ~1 nm at the particle surface. Surface analysis by x-ray photoelectron spectroscopy (XPS) indicated no enrichment of Bi and Mo versus the bulk (Fig. 1A, i) contrary to the often-postulated particle model featuring a surface overlayer of bismuth molybdate (9, 20, 21). Interestingly, concentrated dispersed Bi species were often found at the inner surface of the embedded voids (Fig. 1A, h; see text S1 and movie S1 for details) and at grain boundaries (Fig. 1A, f) that were enclosed and not accessible from the gas phase and thus not directly relevant for catalysis.

To study the relevance of the highly dispersed surface Bi species, we prepared a catalyst by impregnation of Bi on a Bi-free analog Mo₁₂Co₇Fe₃K_0.08Si_1.6O_x. The Bi-free support material showed low conversion even at temperature of 350°C, whereas the catalyst impregnated with Bi showed the same activity as the reference, although it contained 90% less Bi (Fig. 1B). The result strongly indicates that the highly dispersed surface Bi species formed the relevant BiMoO sites and, in turn, suggests a minor role of α-Bi₂Mo₃O₁₂, which did not exist in the impregnated sample. The structure of the active surface BiMoO ensemble may be described by structural motifs of α-Bi₂Mo₃O₁₂ and other bismuth molybdates (β-Bi₂Mo₂O₉ and γ-Bi₂MoO₆) that are able to perform the selective propene oxidation (11). However, detailed structural assignment remains elusive. We note that the presence of silica does not essentially change the catalytic functionality (see table S4).

We suggest that Bi segregates at closed peripheries in the course of phase crystallization during calcination because of its large ionic radius (eight-coordinated Bi³⁺, 117 pm) that does not fit in β-Co_1−xFe_xMoO₄ and Fe₂Mo₃O₁₂ that are composed of much smaller cations (octahedral Co²⁺, 75 pm; octahedral Fe³⁺, 65 pm; and tetrahedral Mo⁶⁺, 41 pm).

Engineering of ion vacancies by atomic composition

β-Co_1−xFe_xMoO₄ is the most likely phase working as the support for surface BiMoO ensemble active sites because (i) it was the most abundant component (58% in K0.1; Fig. 1A, e) representing a large fraction of the surface area, (ii) α-Bi₂Mo₃O₁₂ deposited on β-Co_1−xFe_xMoO₄ (15) and physical mixture of the two phases (16) showed remarkably enhanced activity versus single phases, (iii) a Co-free MCM catalyst representing Bi-incorporated Fe₂Mo₃O₁₂ showed little activity (table S3, #1 versus #2), and (iv) β-Co_1−xFe_xMoO₄ could allow epitaxial growth of surface BiMoO ensemble in the α-Bi₂Mo₃O₁₂ structure, which has active site motifs (22) due to the good structural fitting of α-Bi₂Mo₃O₁₂ (111) plane and β-Co_1−xFe_xMoO₄ (001) plane (8, 10, 23). The past studies (9) showed that the incorporation of Fe in the β-CoMoO₄ structure was essential for the activity enhancement, which was interpreted that cation vacancies created by the substitution of Co²⁺ by Fe³⁺ ease the exchange of electrons and oxygen between β-Co_1−xFe_xMoO₄ and active α-Bi₂Mo₃O₁₂ phase.

It is considered that the stoichiometric constraint imposed by the overall atomic composition causes ion vacancies in addition to the substitution of Co by Fe. To study the effect of the ion vacancies in the industrial MCM system, we prepared a series of catalysts varying Mo, Co, and Fe contents within the range of Bi_0.6Mo_11–13Co_5.5–8.5Fe_2–3.5K_0.08Si_1.6O_x (table S5) with the aim of manipulating the concentrations of ion vacancies. Assuming that Co²⁺, Fe³⁺, and Bi³⁺ cations form exclusively metal molybdates with MoO₄²⁻ anion within the studied composition range, the ratio R (Eq. 1)

$$R=\mathrm{Mo}/(\mathrm{Co}+1.5\mathrm{Fe}+1.5\mathrm{Bi})$$ (1)

indicates the stoichiometric balance. We found that the catalytic activity, assessed by the reaction temperature needed to reach 95% propene conversion, clearly correlated with the stoichiometric balance R (Fig. 1C; R range was 0.89 to 1.10; see table S5 for details). Highly active catalysts are characterized by a Mo-deficient composition (R < 1), while excess of Mo (R > 1) leads to a sharp decrease of the activity. The ion vacancies formed in Mo-deficient compositions should comprise both cation (Mo⁶⁺) and anion (O²⁻) vacancies to ensure charge neutrality (text S2). Oxygen vacancies likely ease the oxygen transport and facilitate the O₂ reduction by accommodating its product O²⁻. We attribute the high activity of catalysts with Mo-deficient formulations to the presence of both cation and anion vacancies that enhances the conductivity of electrons and oxygen.

Role of iron molybdate as sites for reoxidation

The strong impact of ion vacancies manifests the vital role of the charge transport through the bulk that bridges the two spatially separated types of the sites for the activation of propene and dioxygen (9). Finding reduced metal species formed by the catalysis helps identify the O₂ activation sites because the four-electron reduction of O₂ to two lattice oxygens 2O²⁻ requires metal sites with dispensable electrons. To study the changes in the oxidation states and the structure caused by catalytic cycles, we analyzed a spent industrial MCM catalyst with a composition of Bi_0.6Mo₁₂Co₇Fe₃K_0.08Si_1.6O_x, an analog catalyst to K0.1, taken from a technical reactor after several years of use.

The spent catalyst showed propene oxidation performance similar to the fresh catalyst (table S6), indicating an excellent stability of the catalyst. Consistently, the BET surface area (table S6) and the particle morphology (fig. S7A versus Fig 1A) were similar to those of the fresh catalysts. The x-ray absorption near-edge structure (XANES) spectra at the energy ranges for Mo, Bi, and Co (fig. S6 and text S3) were almost identical to the fresh status, indicating very minor changes in oxidation state and coordination geometry. Particularly, the Mo centers that are considered to oxidize propene (24) remained as a nonreduced Mo⁶⁺ state.

The Fe components, in contrast to the other metals, showed remarkable changes. The XANES spectra at the Fe K-edge displayed a downward shift of the absorption edge energy (7131 → 7129 eV), indicating that a majority of Fe³⁺ was reduced to Fe²⁺ after the use (Fig. 1D) (25). The formation of Fe²⁺ with a larger ionic radius of 78 pm (versus 65 pm for Fe³⁺) induced transformation of Fe₂Mo₃O₁₂ to β-FeMoO₄ (26) that was indicated by XRD analysis (table S6) showing a large decrease of Fe₂Mo₃O₁₂ from 29 to 7% accompanied by an increase of β-Co_0.7Fe_0.3MoO₄ component from 53 to 71%. We note that β-FeMoO₄ is formally assigned to β-Co_0.7Fe_0.3MoO₄ in the XRD analysis because the two phases are isostructural with similar cationic radii (Fe²⁺, 78 pm and Co²⁺, 75 pm; both octahedral). Consistently, STEM-EDXS analysis of the spent catalyst (fig. S7 and table S7) revealed (i) an increase of Fe content in CoMo-rich domains (table S7B versus table S7A), (ii) the formation of FeCoMo-rich domains (fig. S7A and table S7B), and (iii) the occasional presence of FeMo-rich domains at the surface of β-Co_1−xFe_xMoO₄ particles (fig. S7C), all indicating an epitaxial growth of β-FeMoO₄ on the isostructural β-Co_1−xFe_xMoO₄ and a redox-driven diffusion of Fe and Co within the β-Co_1−xFe_xMoO₄ domains.

Reoxidation of the spent catalyst by calcining at 500°C in air led to a recovery of the Fe₂Mo₃O₁₂ phase at the expense of β-Co_0.7Fe_0.3MoO₄ (table S6), indicating the reversible phase transformation triggered by the Fe²⁺/Fe³⁺ redox.

The substantial reduction of Fe³⁺ to Fe²⁺, which also agrees with early studies (26, 27), together with the unreduced Mo⁶⁺ centers (fig. S6) indicates that the electrons obtained at BiMoO sites move through the bulk to accumulate at Fe²⁺ centers in the Fe₂Mo₃O₁₂/β-FeMoO₄ domains. We assign the activation of O₂ to the surface Fe²⁺ sites because no other metals have dispensable electrons needed for the reduction of O₂. We propose that the electron pooling at Fe²⁺ centers in the bulk is necessary to create a sufficient density of surface Fe²⁺ sites needed to accomplish the four-electron reduction of O₂ at a rate similar to the catalyst reduction (i.e., propene oxidation). The active Fe²⁺ sites are located either at the surface of β-FeMoO₄ domains or at reduced oxygen-defective Fe₂Mo₃O₁₂ formally described as Fe²⁺₂Mo₃O₁₁.

Because the sites for propene oxidation (i.e., BiMoO on β-Co_1−xFe_xMoO₄) and reoxidation (i.e., Fe₂Mo₃O₁₂/β-FeMoO₄) are spatially separated, the charge transfer processes are kinetically relevant. The path length and the conductivity principally determine the efficiency of the charge transport. The nanostructured mixed particle agglomerate (Fig. 1A, b) seems to ensure short paths. As discussed already, ion vacancies enhance the conductivity in general. Moreover, we suggest that the Mo and O vacancies in β-Co_1−xFe_xMoO₄ facilitate the transformation of Fe₂Mo₃O₁₂ to β-FeMoO₄ upon the electron transfer because the phase conversion formally eliminates a MoO₄²⁻ unit according to Eq. 2, wherein the Mo and O vacancies could accommodate it.

$${{{\mathrm{Fe}}^3}^{+}}_2{\mathrm{Mo}}_3{\mathrm{O}}_{12}+4e^{-}\to 2{\mathrm{Fe}}^{2+}{\mathrm{Mo}\mathrm{O}}_4+{{\mathrm{Mo}\mathrm{O}}_4}^{2-}$$ (2)

This effect is considered another reason for why Mo-deficient formulations give high activity (Fig. 1C).

The high Fe²⁺ fraction after the catalysis also suggests that the catalyst reoxidation is the kinetically limited step. In fact, the kinetic behavior that will be discussed in the next section (Fig. 2A) indicated a reoxidation limited kinetics at temperature below 360°C. Moreover, a catalyst containing only one-third of Fe while with the same R value showed very low activity (table S3, #1 versus #3), indicating a kinetic bottleneck of O₂ activation posed by the abundance of surface Fe²⁺ sites. We propose that highly active MCM catalysts contain both Fe₂Mo₃O₁₂ and β-Co_1−xFe_xMoO₄ with a ratio that balances the reduction and the reoxidation of the catalyst, as in the case of the benchmark K0.1 (Fig. 1A, e).

Fig. 2. Effect of K-doping on propene oxidation catalysis and material character.

(A) Propene oxidation performance of K0-K0.3 (Bi_0.6Mo₁₂Co₇Fe₃K_aO_x, a = 0 to 0.3). (a) Selectivity to acrolein plus acrylic acid and (b) propene conversion X as a function of T. (c) Arrhenius plots for propene consumption rate constant. See fig. S11 for the X versus S plot. Test conditions: Ten-gram granulate catalyst with 135 g of inert ring dilution, t = 0.36 s g ml⁻¹, feed gas C₃H₆/O₂/N₂ = 5/9.5/85.5 mol %, P = 0.16 MPa, and time on stream = 200 to 350 hours. (B) Unit cell volume estimated by whole pattern fitting of XRD data. (C) NH₃-TPD (temperature-programmed desorption of ammonia) profiles. (D) Propene-TPR (temperature-programmed reaction with propene) profiles of K0.1 and K0. The mass spectrometer signals for mass/charge ratio (m/z) of 45, 56, 57, and 58 are scaled by magnifications indicated in the label. a.u., arbitrary units.

Role of K-promoter

Next, we studied the effects of the K-promoter. We prepared catalysts with varied K contents Bi_0.6Mo₁₂Co₇Fe₃K_aO_x (a = 0 to 0.3; denoted as K0, K0.05, K0.1, K0.2, and K0.3). The BET surface area, pore structure (table S1 and fig. S8), crystalline phase composition by XRD (fig. S4), microstructure by STEM-EDXS (fig. S9 versus Fig. 1A and figs. S1 to S3; table S7A versus table S7C), and chemical nature of the elements at surface probed by XPS fingerprint (fig. S10) showed only minor changes by varying the K-content.

In contrast, strong effects were seen on both activity and selectivity. The K-doping changed the temperature dependency of the propene conversion (Fig. 2A, b). The Arrhenius plots (Fig. 2A, c) show a distinct transition of the slope at 360°C, which is due to the change of the rate-determining step from the catalyst reoxidation (<360°C) to the catalyst reduction (>360°C) (28). The K-doping substantially reduced the apparent activation energy Ea_300–360 (table S8) at the reoxidation-limited low-temperature regime, suggesting that K eases the reoxidation. In the reduction-limited high-temperature regime (360° to 400°C), the activity reduced along with the K concentration, while the apparent activation energy Ea_360–400 (table S8) remained similar within 44 to 49 kJ/mol, suggesting that K-doping decreased the quantity of the active site for propene oxidation without altering its quality. Notably, the Ea_360–400 values (44 to 49 kJ/mol) were remarkably lower than that of α-Bi₂Mo₃O₁₂ (80 kJ/mol) (9), supporting the hypothesis that not α-Bi₂Mo₃O₁₂ but dispersed BiMoO ensemble sites are relevant.

The selectivity to the target products (=acrolein + acrylic acid) remarkably increased at a minimum K-doping (Fig. 2A, a). The response of the selectivity by variation of the propene conversion measured at constant temperatures (fig. S12) provides insights on reaction pathways. The CO_x selectivity at zero propene conversion relates to the direct combustion of propene, while the increase of the CO_x selectivity with increasing propene conversion indicates CO_x formation by consecutive combustion of acrolein. The data (fig. S12) show that both combustion pathways were suppressed by the K-doping.

To find the mechanistic origin of the effects of the K-doping, we analyzed the bulk and surface properties. The elemental mapping by STEM-EDXS (Fig. 1A, d) revealed the presence of K over all bulk phases. The incorporation of large K⁺ cation (six-coordinated K⁺, 138 pm) caused an expansion and distortion of crystal lattices, which is evidenced by (i) the increased lattice volumes estimated by XRD analysis (see Fig. 2B and fig. S13 for details), (ii) the red shift of Mo─O (~930 cm⁻¹) and Fe─O (~775 cm⁻¹) stretching vibrations (29) in Raman spectra (fig. S14), (iii) the increase of absorption at 510 to 580 nm (fig. S15A) assigned to the symmetry-forbidden d-d transition of octahedral Co²⁺ in β-Co_1−xFe_xMoO₄ (30) indicating less centrosymmetric (i.e., more transition-allowed) coordination geometries of the Co²⁺ centers, and (iv) the decrease of the absorption edge energy due to the ligand-to-metal charge transfer of O²⁻─Mo⁶⁺ centers at ~350 nm (fig. S15B) (31). Moreover, the incorporation of K⁺ in the crystal phases should also create ion vacancies to keep the charge neutrality. We attribute the reduction of Ea_300–360 by K-doping to the accelerated reoxidation due to enhanced conductivity associated with lattice distortion and ion vacancies (9, 32). In addition to the bulk doping effects, surface K⁺ located in the vicinity of Fe²⁺ may accelerate the O₂ activation because K species can show the ability to activate O₂ (33, 34).

Rapid reoxidation seems to reduce unselective combustion. It is generally postulated that partially reduced O₂ (e.g., O₂⁻ and O⁻) with electrophilic characters and reduce naked (i.e., oxygen defective) metal sites acts as entry points for unselective oxidation (35). A potential reaction pathway is the addition of electrophilic oxygen to the electron-rich olefinic bond of propene to form a propylene oxide (surface-) intermediate that undergoes combustion (fig. S16). Because the reoxidation rate has a higher temperature sensitivity than the oxidation rate [Fig. 2A (c) and table S8], the participation of such unselective pathways associated with slow reoxidation increases with decreasing the temperature. In fact, K0 showed higher CO_x selectivity at low temperatures due to an increased propene direct combustion [Fig. 2A (a) and figs. S11 and S12; 330°C versus 360°C]. We suggest that accelerated reoxidation by K-doping reduces the participation of unselective species associated with reduced sites and thus improves the selectivity particularly in the reoxidation-limited low-temperature regime.

K⁺ cation modifies not only the bulk but also the surface. The surface metal ratio by XPS evidenced an enrichment of K⁺ (Fig. 1A, i). The segregation of K⁺ cation is driven by its large ion size (six-coordinated K⁺, 138 pm) compared to Co²⁺, Fe³⁺, and Mo⁶⁺ (the same driving force as discussed for the segregation of Bi³⁺). Consistently, K tended to accumulate at Bi-rich domains [Fig. 1A (c and d) and table S7] because of the similar ionic radii, which could explain the depletion of Bi (Fig. 1A, i).

Because K⁺ is strongly ionic, it likely modifies the surface acid-base properties. To study the surface acid-base properties, we performed temperature-programmed desorption of ammonia (NH₃-TPD) and pyridine adsorption monitored by infrared spectroscopy (pyridine-IR). The NH₃-TPD profiles (Fig. 2C) showed single desorption peak at ~160°C with a tailing decay to ~350°C. We assign the NH₃ adsorption sites to surface tetrahedral (─O─)₂Mo⁶⁺(=O)₂ motifs because (i) the calculated surface NH₃ coverage ranged from 51 to 71% of BET surface areas (table S1) that indicates a major surface site and (ii) the NH₃-TPD fingerprint was very similar to that of surface tetrahedral MoO₄ species on silica comprising the dioxo (─O─)₂Mo⁶⁺(=O)₂ units (Mo⁶⁺ as Lewis acid site) and a minor fraction of ─OH groups in the vicinity of the dioxo MoO₄ units (Brønsted acid site) (36). Consistently, the pyridine-IR data indicated that Lewis acid sites represented the majority (~90%), with a minor fraction of Brønsted acid sites (fig. S17 and table S1) (37).

The NH₃-TPD profile of K0 exhibited distinctly higher tailing component, indicating the presence of an additional broad component at 160° to 350°C that represents another type of acid sites (Fig. 2C). All the other K-doped catalysts showed similar NH₃-TPD fingerprints and similarly improved selectivity in propene oxidation, indicating that the acid sites represented by the additional broad tailing component in the NH₃-TPD were responsible for the high CO_x selectivity of K0. We suggest that the acid sites eliminated by the K-doping are also tetrahedral MoO₄ moieties, however, in irregular defective situations, e.g., frustrated coordination geometry in terms of bond length or angle of Mo─O bonds, because such sites are generally more reactive (38).

To gain insights into the impact of K-doping on the reaction network, we performed temperature-programmed reaction with propene monitored by mass spectrometry (propene-TPR) (Fig. 2D). The propene-TPR profiles >350°C showed massive formation of acrolein and other products without O₂ in the gas phase, which confirms the oxidation reactions by catalyst lattice oxygen atoms. Whereas, product formation at low temperature provides information on the primary products in the reaction network (39). The major feature of K0.1 was the concurrent formation of acrolein [mass/charge ratio (m/z), 56] and its precursor allyl alcohol (m/z, 57) (39–41) at 220° and 310°C. K0 displayed, in addition, the formation of acetone and/or propanal (m/z, 58) at 130°, 220°, 310°C and isopropanol (m/z, 45) at 130°C. These three products are considered to occur by acid catalysis and ultimately end up with CO_x (39–41). Propanal can form via acid-catalyzed isomerization of allyl alcohol or of propylene oxide intermediates (39, 42, 43) or via surface 1-propoxide intermediate formed at Brønsted acid sites (40). Similarly, acetone forms via isopropanol intermediate by acid-catalyzed hydration (44) or via surface 2-propoxide intermediate formed at Brønsted acid sites (40). The propene-TPR confirms that unselective reactions (fig. S16) are hindered by blocking one type of acid sites by K.

Factors enabling high performance

Combining all the results, a model for industrial MCM catalysts for the selective oxidation of propene to acrolein emerges. We propose that high performance arises from the four functions enabled by various structural features (Fig. 3).

Fig. 3. Proposed model of industrial Bi-Mo-Co-Fe-K-O catalysts for propene oxidation to acrolein.

High performance arises from the four functional features (blue text) associated with the various structural features (gray text).

The first function is selective propene oxidation to acrolein. The BiMoO ensemble supported on β-Co_1−xFe_xMoO₄ constitutes the sites for this reaction. The reaction leaves an oxygen vacancy, two protons, and two electrons on the catalyst. The reduced site is repaired by O²⁻ supplied from the support phase to regenerate the active BiMoO ensemble. The small size of BiMoO ensembles is favorable for the charge transport because bulk bismuth molybdates are worse conductors than the support phase (45, 46). Epitaxial deposition of BiMoO ensembles on β-CoMoO₄ structure (8) may help in building BiMoO sites with geometric configurations similar to that of α-Bi₂Mo₃O₁₂, which has active site motifs (22).

Second, surface Fe²⁺ sites for the O₂ reduction are required. The surface Fe²⁺ sites are either on β-FeMoO₄ or reduced oxygen-defective Fe₂Mo₃O₁₂ (i.e., Fe²⁺₂Mo₃O₁₁). The iron molybdate domains pool electrons in the bulk to create a sufficient density of surface Fe²⁺ sites needed to kinetically balance with the catalyst reduction and the charge transport. The K-doping kinetically promotes the catalyst reoxidation.

Third, the transport and the storage of the charge carriers need to be ensured to effectively bridge the two types of active sites. The conductivity and the path length are the major factors determining the overall charge transport efficiency. Short path lengths are realized by the nanostructured particle agglomerates and the small-sized BiMoO domains. A high conductivity is enabled by the combination of anion and cation vacancies created by the stoichiometric constraints and self-doping, the iron redox couple, the lattice distortion caused by K-doping, and the structural fitting of the interfaces between the BiMoO ensemble/β-Co_1−xFe_xMoO₄/β-FeMoO₄.

Last, unselective sites are to be disabled. The K-doping reduces unselective sites in two ways: (i) by masking one type of surface acid sites that promote combustion and (ii) by promoting the O₂ reduction that reduces unselective O species associated with slow catalyst reoxidation.

All the functions coupled with the structural features are relevant to the high performance of the industrial catalysts. The features are simultaneously fulfilled by a proper composition and preparation. The controlled segregation during the calcination directed by crystal structures and ionic radii gives the well-mixed particle aggregates comprising self-doped and surface-decorated crystalline phases.

Minor contents of α-Bi₂Mo₃O₁₂ and α-Co_1−xFe_xMoO₄ in industrial MCM appear to be side products formed during the segregation process. However, α-Bi₂Mo₃O₁₂ may stabilize the system by dispensing BiMoO to regenerate the active surface ensemble sites when they collapse under catalytic cycles because Bi₂Mo₃O₁₂ likely displays mobility of its building blocks at a reaction temperature (300° to 430°C) that is well above its Tamman temperature of 195°C (47). This self-repair function could be important for stability because Mo sublimates from the surface in the long term (48).

DISCUSSION

We showed that the material complexity of the industrial catalysts is linked to the functional complexity that enables their high performance. To understand real industrial catalysts, it is crucial to benchmark the real system carefully in performance and material properties because these characteristics make the deviation from the model systems [“material- and complexity-gap” (49)] measurable and thus allow a proper design of studies addressing the structure-function relationships. Our approach to pivot the system at an industrial benchmark offers a general and pragmatic way to tackle the gap between industrial and model catalysts.

MATERIALS AND METHODS

Catalyst preparation

General procedure

The catalysts with different compositions consisting of up to six components (Bi-Mo-Co-Fe-K-Si) have been prepared by adopting the method described in example 3 of U.S. patent 6881702B2 (12) that involves coprecipitation, drying, tableting, and calcination in air. Briefly, an aqueous solution A made by mixing aqueous solutions of cobalt (II) nitrate, iron (III) nitrate, and bismuth (III) nitrate at 60°C was added to an aqueous solution B containing ammonium heptamolybdate and potassium hydroxide at 60°C under vigorous stirring to form a suspension. Silica sol LUDOX TM-50 (47% silica content; Grace GmbH) was added to the suspension in cases of Si-containing formulations. After aging at 60°C for ca. 30 min, the suspension was dried either by a spray-drying in air flow (inlet/outlet temperature, 350°C/140°C) or by a rotary evaporator (oil bath temperature of 110°C and pressure at 12 kPa) with subsequent drying in air flow at atmospheric pressure and at 120°C for 16 hours. The obtained dried precursor was mixed with 2 to 3% of graphite Asbury 3160 (Asbury carbons) then tableted to form shaped body in ring [outer diameter (OD), 5 mm; height (H), 3 mm; inner hole diameter (ID), 2 mm; and tablet weight, 0.12 g] or cylinder (diameter, 20 mm; height, 3 mm; and tablet weight, 1.5 g) shapes. The precursor shaped body was calcined in an air circulation oven with a continuous feed of fresh air at atmospheric pressure. In the standard calcination protocol, the temperature was increased with heating rate less than 2°C/min stepwise with seven isothermal phases at 130°C for 72 min, 190°C for 72 min, 220°C for 72 min, 265°C for 72 min, 380°C for 187 min, 430°C for 187 min, and lastly 500°C for 467 min. The final highest temperature and the holding time were varied in some experiments.

Post-impregnation of Bi

First, a Bi-free catalyst Mo₁₂Co₇Fe₃K_0.08Si_1.6O_x was prepared in ring shape (OD × H × ID = 5 × 3 × 2), where the final calcination temperature of 450°C was applied. Bi was introduced by an incipient wetness impregnation of an aqueous solution of bismuth (III) nitrate stabilized with an extra nitric acid containing the desired mass of Bi. The impregnation was done in a rotary evaporator under subambient pressure (~20 kPa) and at room temperature. After drying under air flow at 50°C, the impregnated sample was calcined with the final temperature of 490°C for 390 min.

Characterization

Nitrogen adsorption

Nitrogen adsorption was carried out at 77 K on a Micromeritics ASAP 2420. Before the measurement, the samples were outgassed in vacuum at 200°C for 15 hours. The specific surface area A_s was calculated according to the multipoint BET in the pressure range P/P₀ = 0.05 to 0.20, assuming a N₂ cross-sectional area of 16.2 Ų (50). The pore volume Vp_N₂ representing pores with diameters smaller than ~50 nm was estimated by the nitrogen uptake at relative pressure of P/P₀ = 0.95.

Mercury intrusion porosimetry

Mercury intrusion porosimetry was performed at room temperature with a pressure range of 0.0034 to 420 MPa on a Micromeritics AutoPore V 9600 (software: MicroActive 1.03.01). The pore size distribution was calculated from the intrusion volume as a function of intrusion pressure, assuming a contact angle of 140° and a surface tension of 485 mN/m. The macro pore volume Vp_Hg was calculated for the pore diameter range of 0.04 to 10 μm.

Electron microscopy

High-angle annular dark-field (HAADF)–STEM combined with EDXS was performed on a Thermo Fisher Scientific Tecnai Osiris and a Thermo Fisher Scientific Themis Z 3.1 with four integrated EDXS detectors [SuperX G1 (Osiris) or G2 (Themis)] at an acceleration voltage of 200 and 300 kV. Samples were prepared as a dispersion in ethanol, ground between two glass plates, and put on a copper grid with a thin carbon film. For EDXS data analysis, the Bruker ESPRIT 2 and the Thermo Fisher Scientific Velox 2 and 3 software were used.

Powder XRD

XRD measurements were performed on a Bruker D8 ADVANCE Series 2 diffractometer with Cu K_α radiation in the 2θ range from 5° to 70° with a step size of 0.02°. The data were analyzed by whole-pattern Rietveld fitting using Bruker TOPAS software. The reflections of graphite were excluded from the quantification of the phase composition. The following crystal structure data were used for the phase analysis: β-Co_0.7Fe_0.3MoO₄ [Inorganic Crystal Structure Database (ICSD) no. 280035], Fe₂Mo₃O₁₂ (ICSD no. 16402), α-CoMoO₄ (ICSD no. 23808), α-Bi₂Mo₃O₁₂ (ICSD no. 2650), Bi₃FeMo₂O₁₂ (ICSD no. 45), and MoO₃ (ICSD no. 35076).

Raman spectroscopy

Confocal Raman spectra were collected at room temperature on a WITec alpha300 R Raman imaging microscope with a green laser excitation (λ = 532 nm/E = 2.33 eV) and a ×20 objective. The average of three spectra measured at different sample spots was used for evaluation to minimize the local effect of sample inhomogeneity due to the multiphase nature of the materials.

Ultraviolet-visible spectroscopy

Diffuse reflectance ultraviolet-visible (UV-vis) spectra were taken at room temperature on a PerkinElmer Lambda 950 instrument equipped with an Ulbricht sphere (d = 150 mm). Spectralon from Labsphere was used as white standard. The spectra expressed in Kubelka-Munk function F(R) were normalized at the highest intensity at ~360 nm because of the ligand-to-metal charge transfer of Mo⁶⁺─O²⁻ centers.

X-ray photoelectron spectroscopy

XPS measurements were performed on a PHI VersaProbe 5000 equipped with an Al K_α (1487 eV and 32 W) radiation source. The samples were prepared as thin powder films (particle size, <0.5 mm) on indium foil. CasaXPS software version 2.3.25 was used for data processing. Binding energies were referenced to the Mo⁶⁺ 3d_5/2 signal at 232.6 eV. With this referencing method, the peak position of the C 1s was found at binding energy of 284.7 eV.

X-ray absorption spectroscopy

X-ray absorption spectroscopy (XAS) measurements at Mo K-edge (20,000 eV), Fe K-edge (7112 eV), Co K-edge (7709 eV), and Bi L3-edge (13419 eV) were performed in transmission mode at the BM26 beamline at the European Synchrotron Radiation Facility. The sample was mixed and ground with cellulose and pressed to form a pellet for measurement. Monochromator calibration at each absorption edge was performed using a reference foil, which was also used to determine the amplitude reduction factor at each edge. The spectra were collected at ambient condition. Data were processed and analyzed using the Demeter XAS software (51).

Pyridine adsorption monitored by Fourier transform infrared spectroscopy

The Fourier transform infrared (FTIR) measurements were carried out in transmission mode using a Thermo Scientific Nicolet 6700 FTIR spectrometer at a spectral resolution of 4 cm⁻¹. The sample was pressed into a self-supporting wafer with an areal density of ca. 18 mg cm⁻², which was placed in an in situ cell equipped with gas dosing and heating systems. Before pyridine adsorption, the sample was treated in air at 350°C under atmospheric pressure for 1 hour and subsequently under vacuum (10⁻³ Pa) at 350°C for 1 hour and then cooled to 80°C. After exposure of pyridine at the pressure up to 3 hPa at 80°C, the cell was evacuated for 70 min at 80°C, and then, the temperature was increased to 100°C. In each experiment, the spectrum taken before pyridine dosing was used as background. The abundance of Lewis and Brønsted types of acid sites was estimated by the peak area of the absorption bands at 1446 cm⁻¹ for Lewis acid sites and at 1540 cm⁻¹ for Brønsted acid normalized by the wafer thickness.

Temperature-programmed desorption of ammonia

Temperature-programmed experiments were measured on an AutoChem II 2920 instrument combined with a quadrupole mass spectrometer (OmniStar, Pfeiffer). A total of 350 mg of material was heated to 400°C with a ramp rate of 10°C/min, kept for 30 min, and then cooled to 100°C under helium (He) stream (30 ml/min). Adsorption of NH₃ was done at 100°C by feeding 10% NH₃ in He (70 ml/min) for 0.5 hours. After flushing the cell with He (75 ml/min) at 100°C for 0.5 hours, the temperature was raised to 600°C with a heating rate of 10°C/min under He flow (40 ml/min). Desorption of NH₃ was monitored by using the mass spectrometric signal of m/z = 16. The helium signal (m/z = 4) was used as internal standard. The signal intensity was normalized by the mass and the BET surface area to quantify the area-specific density of the NH₃ adsorption site. The surface coverage by NH₃ was calculated assuming a NH₃ cross-sectional area of 14.0 Ų (50).

Temperature-programmed desorption of propene

Propene-TPR was performed on the same setup as the NH₃-TPD experiments. A total of 250 mg of material was heated to 400°C with a ramp rate of 10°C/min, kept for 30 min, and then cooled to 100°C under 10% O₂ in He (30 ml/min). After flushing the cell with He (75 ml/min) at 100°C for 0.5 hours, the flow was switched to 5% propene in He (75 ml/min). The sample temperature was raised to 500°C with a heating rate of 10°C/min. Evolved products were monitored by the mass spectrometer. The helium signal (m/z = 4) was used as internal standard, and the intensity was normalized by the sample mass. The following assignments of mass signals were used: acrolein (m/z = 56), allyl alcohol (m/z = 57), propanal (m/z = 58), acetone (m/z = 58), and isopropanol (m/z = 45).

Catalytic test

Catalytic performance in propene oxidation was measured in a fixed bed tubular reactor (stainless steel; inner diameter, 15 mm; catalyst bed length ca. 80 cm; and immersed in molten salt as heat exchange medium) at pressure (P) of 0.12 to 0.16 MPa in the temperature (T) range of 300° to 400°C. T was the temperature of the heat exchange medium, while the catalyst bed temperature was not measured. The catalyst bed was placed in the reactor tube either as (i) 100 g of pure catalyst in ring shape (OD × H × ID = 5 mm × 3 mm × 2 mm), or (ii) the mixture consists of 10 g of crushed and sieved granulate catalyst (2 to 4 mm) and 135 g of inert ceramic ring (OD × H × ID = 5 mm × 5 mm × 2 mm). The gas feed is composed of 5 mol % of propene, 9.5 mol % of oxygen, and 85.5 mol % of nitrogen. The gas feed flow rate was varied in the range of 40 to 200 liter/hour. The gas composition at the inlet and the outlet of the reactor was analyzed by an online gas chromatograph Agilent 6890N equipped with a flame ionization detector and a thermal conductivity detector. A column system consisting of type Restek Rt-Q-Bond, Varian PoraPLOT Q, and Varian Molecular Sieve was used to separate ethene, propane, propene, acetaldehyde, acrolein, acetic acid, acrylic acid, CO₂, O₂, N₂, and CO. The sampling and analytic lines were kept heated at 200°C to avoid condensation of liquid components. The contact time t (in s g ml⁻¹) is calculated by catalyst mass (in g) excluding the inert dilution and the volumetric total gas feed rate (liter hour⁻¹) wherein the volume is at the standard state (298.15 K and 101.325 kPa).

Conversion of propene X was calculated by comparing the propene molar flow at the reactor inlet and the outlet. Selectivity of individual oxidation product S was calculated on the basis of the number of carbon atoms and on the propene conversion. The performance was measured in the range of 2 to 16 days of time on stream. The rate constant k (in s⁻¹ g⁻¹ ml) for the propene consumption rate r (in mol s⁻¹ ml⁻¹) was determined from the propene conversion X and the contact time t (in s g ml⁻¹), assuming a first-order plug-flow reactor kinetics with respect to propene described by Eq. 3 that was derived from the first-order rate Eq. 4 representing the mass-specific rate r/d (in mol s⁻¹ g⁻¹) wherein d is the catalyst mass filling density in the catalyst bed (in g ml⁻¹) and C is the reactant molar concentration per volume (in mol ml⁻¹)

$$k=-\frac{\ln (1-X)}{t}$$ (3)

$$\frac{r}{d}=\frac{1}{d}\frac{dC}{dt}=-kC$$ (4)

The obtained rate constant k was used in Arrhenius plots to estimate the apparent activation energy E_a. To compare the propene conversion rates with the literature data, the surface area–specific reaction rate r′ (in mol s⁻¹ m⁻²) and the surface area–specific rate constant k′ (in s⁻¹ m⁻² ml) were calculated by dividing the r (in mol s⁻¹ ml⁻¹) and k (in s⁻¹ g⁻¹ ml) by BET surface area A_s (in m² g⁻¹). The k′ values of the literature data were calculated from the reported propene conversion rate r and the inlet propene concentration C values using the Eq. 4 that assumes a first-order kinetics with respect to propene.

We note the kinetics change from an oxygen-dependent regime at T < 360°C to a propene-dependent regime at T > 360°C. The assumption of the first-order kinetics with regard to propene is not valid in the oxygen-dependent regime. However, the formal analysis by Eq. 3 using the propene conversion X still represents the relevant information (i.e., oxygen conversion in the oxygen-dependent regime) because the main reaction (acrolein formation; Eq. 5) occurring with high selectivity (mostly >90 mol %) consumes an equal amount of propene and oxygen

$${\mathrm{C}}_3{\mathrm{H}}_6+{\mathrm{O}}_2\to {\mathrm{C}}_3{\mathrm{H}}_4\mathrm{O}(\mathrm{acrolein})+{\mathrm{H}}_2\mathrm{O}$$ (5)

The effect of intraparticle mass transport was studied by changing the mean granulate size in the range of 0.75 to 3 mm and the tablet body density before calcination in the range of 1.6 to 2.4 g cm⁻³. The results showed <5% of variation of the propene conversion rate and negligible (<0.3 mol %) influence on the selectivity, indicating minor impact of the intraparticle mass transport.

The temperature difference between the catalyst particle and the gas phase estimated by the approach by Mears (52) was at most 2°C. The temperature difference between the heat exchange medium and the peak temperature of the catalyst bed was estimated to be <5°C in the case of the diluted catalyst bed (10 g of catalyst mixed with 135 g of inert) and <20°C in the case of the 100-g undiluted catalyst operated at propene conversion of 95%.

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S17

Tables S1 to S8

Legend for movie S1

References

Other Supplementary Material for this manuscript includes the following:

Movie S1