The A or B positions in the crystal lattice of ABO₃ compounds with perovskite structure can be occupied by almost all metallic elements of the periodic table according to Goldschmidt’s rule (Goldschmidt, 1926). In this way, a wide variety of compounds is obtained. Other materials with special properties are accessible by partial substitution of either A or B leading to AA′BB′O₃ formulas (Kurbakov, 2010; Zhu et al., 2015; Yang et al., 2016; Bartel et al., 2019). Chemical substitution is generally also accompanied by changes in defect structure. For example, the replacement of an A³⁺ with A²⁺ leads to lattice distortion (Kurbakov, 2010) and the formation of oxygen vacancies. Alternatively, partial oxidation of the B element may occur if this is possible (Ciambelli et al., 2000; McFarland and Metiu, 2013; Toniolo and Schmal, 2016). Both phenomena can also occur simultaneously. Defect formation can be achieved not only by the partial or complete substitution of A or B. Thus, reduction of the available amount of A or B also provides access to defect-rich compounds A_1-xB_1-yO₃ that differ significantly from the ideal stoichiometry ABO₃ (Golikov et al., 2005; Neagu et al., 2013). Due to the dense packing of ions in the perovskite structure, the occupation of interstitial sites is very unlikely. Therefore, for defect formation, only vacancy formation or migration of smaller B elements to vacant A positions, which offer more space, can be considered (Wołcyrz et al., 2003; Ignatans et al., 2019).

Defect-rich structures are of particular importance with respect to the application of perovskites in heterogeneous catalysis (Royer et al., 2014; Najjar and Batis, 2016). In addition to the increasing interest in perovskites as catalysts for oxygen reduction and oxygen evolution reactions (Suntivich et al., 2011; Gupta et al., 2016; Zhu et al., 2019), this class of materials is also widely used in photocatalysis (Grabowska, 2016; Tasleem and Tahir, 2020) or total oxidation to remove volatile organic and hazardous compounds (Gil et al., 2004; Liu et al., 2018a; Liu et al., 2018b; Liu et al., 2019). However, perovskites are also suitable catalysts for selective oxidations at low temperatures (Polo-Garzon and Wu, 2018; Kamata, 2019), especially for the oxidative dehydrogenation of propane (Koch et al., 2020). The catalytic properties can be influenced by various factors, such as the surface composition (Fierro and Tejuca, 1987; Koch et al., 2020), the concentration of defects, or the partial reduction of metal ions (Evarestov et al., 2005; Mierwaldt et al., 2014; Ignatans et al., 2019; Koch et al., 2020). Oxygen species at the catalyst surface are responsible for the activation of the propane molecule. X-ray Photoelectron Spectroscopy (XPS) allows the identification of different oxygen species in the surface region of perovskites (Mierwaldt et al., 2014; Stoerzinger et al., 2014; Stoerzinger et al., 2015). Furthermore, the dynamics of these species can be studied as a function of the chemical potential of the gas phase under reaction conditions of the selective oxidation of propane using ambient pressure (AP) XPS (Koch et al., 2020).

Mn-based perovskites are generally catalyzing the total combustion of propane already at low reaction temperatures. However, the addition of water vapor to the feed of propane and oxygen leads to higher selectivity of the desired selective oxidation product propene. The improved selectivity is apparently related to the partial reduction of surface manganese ions and the formation of defects and hydroxides (Koch et al., 2020). To examine these relationships in more detail, a samarium-deficient Sm_0.96MnO₃ perovskite was prepared by combustion synthesis and subjected to various post-treatments at 450°C in inert gas and gases containing H₂O and CO. The compound was chosen because it is structurally stable under relevant reaction conditions (Kamata et al., 1979; Hui et al., 1990; Golikov et al., 2005) and exhibits good catalytic activity at low temperatures as well as significant changes in catalytic performance upon variation of the feed composition. In contrast, LaMnO₃ (Koch et al., 2020), which is often studied in oxidation reactions, is more sensitive to changes in oxygen partial pressure. This can ultimately lead to defect restructuring associated with the transformation of crystal symmetry from rhombohedral to orthorhombic and thus with the change of bulk structure (Miyoshi et al., 2003).

The catalyst was examined before and after the treatments in the different gas phases with respect to its catalytic performance, defect structure and surface composition. AP-XPS experiments and ambient pressure Near Edge X-Ray Absorption Fine Structure (AP-NEXAFS) measurements in the presence of the gas phase were performed to study in detail the changes in the O 1s spectra and the surface oxidation state of the Mn and Sm ions as a function of the pretreatment conditions and the feed composition. Based on the spectroscopic results, possible relationships between the surface structure of the perovskite and its properties as a catalyst in propane oxidation are discussed.

Phase-pure Sm_0.96MnO₃ (Supplementary Figure S2, Table 1) ICSD# 95491 was prepared by the self-combustion synthesis route (Mori et al., 2002). The catalyst exhibited pores and voids of different sizes due to the high amounts of gases formed during ignition synthesis. The pores were retained during calcination. Thermal treatment of the synthesized material in oxygen is generally required to remove remaining organic components and carbonates formed by storage in air (Supplementary Figure S1). During the calcination of the combustion product, small crystallites were formed that adhere to each other and creating cavity-rich macrostructures (Supplementary Figure S3). The predominant size fraction of the nanocrystals was about 50 nm according to SEM, which is in good agreement with the calculated crystallite size from XRD refinement (51 nm, Table 1). However, larger and smaller crystallites are also present, but only in small fractions (Supplementary Figure S3). Due to the irregular macrostructure and the voids, a specific surface area of 7.6 m²⋅g⁻¹ was obtained (Table 2). The desired non-stoichiometry of Sm and Mn in the volume was reached by using a reduced mass of the samarium precursor (Table 2). However, the surface of the catalyst showed a 1:1 stoichiometry of Sm and Mn (Table 2). Analysis of the volume revealed a slight oxygen deficit of the compound (Sm_0.96MnO_2.93, Table 2). Therefore, the segregation of nano-sized binary oxides seems unlikely (Golikov et al., 2005; Neagu et al., 2013). Segregated phases were also not detected by XRD (Supplementary Figure S2) and EDX mapping (Supplementary Figure S4).

The Sm_0.96MnO₃ catalyst is oxidizing propane already at temperatures below 300°C. (Merino et al., 2005; Koch et al., 2020). In the present work, the reaction was performed in either dry feed, i.e. 5% propane and 10% O₂ in nitrogen (5/10/85), or wet feed, i.e. 5% propane, 10% O₂ and 40% H₂O in nitrogen (5/10/45/40) (Koch et al., 2020). The dry feed at 300°C was chosen as a reference point to compare the influence of the different pretreatments on activity and selectivity (Supplementary Figure S5). The selectivity to propene measured at the reference points as a function of propane conversion is shown in Figure 1A for different contact times. Regardless of the different pretreatment conditions, represented by different colors, and regardless of the contact time, represented by full, half-filled and empty symbols, all the measurement points lie on the same S-X curve. Thus, in dry feed, it can be seen that the different pretreatments only affected the integral rates. Compared to the conversion after calcination in air at 450°C, the conversion decreased slightly after treatment in pure N₂ at the same temperature. This loss of activity could not be reversed by subsequent treatment in pure oxygen and further treatment in synthetic air at 450°C. The decreased conversion could be caused by either slight sintering or surface modifications. An increase in the apparent activation energy from 61 ± 6 kJ⋅mol⁻¹ to 76 ± 4 kJ⋅mol⁻¹ after N₂ pretreatment suggests more for the latter (Supplementary Table S1). After the treatments in 10% water vapor and 2% CO, the activity decreases more significantly.

Using wet feed, the selectivity to propene remarkably increased significantly (by about 18–83% over the entire range of measured conversions) at comparable conversion (Figure 1B) indicating a change in the properties of the catalyst probably related to the addition of water vapor. An increase in selectivity in the wet feed has been observed previously for Mn-based perovskites that had a nearly balanced Mn/A element ratio or an excess of Mn at the surface (Koch et al., 2020). In contrast, an Mn deficiency on the surface resulted in reduced selectivity (Koch et al., 2020). This effect was particularly pronounced after the pretreatments in steam and CO at 450°C, and the selectivity improvement was more evident at low propane conversion (Figures 2A,B). Apparently, both steam treatment and reducing conditions during pretreatment promote the formation of a more selective catalyst under reaction conditions, but only when propane oxidation is carried out in the presence of H₂O in the feed. When switched to dry feed, the catalyst returns to its non-selective mode (Figure 1A and Figures 2A,B).

The used catalyst showed the same bulk structural properties as the fresh catalyst (Table 1, Supplementary Figure S2). Only minor changes in lattice parameters were observed, which did not affect the total unit cell volume. Therefore, an extensively increased formation of oxygen vacancies in the bulk could be excluded, because this should have led to lattice expansion and increased cell volume (Miyoshi et al., 2003). Moreover, the excess of Mn in the bulk remained after catalytic testing (data not shown). Recrystallization or crystallite growth did not occur at these low temperatures applied because the crystallite sizes before and after catalytic testing were similar.

In contrast to the stable bulk structure, the surface structure changed significantly when the chemical potential was varied, which was achieved both by changing the feed and varying the pretreatment atmosphere. To understand the significant increase in selectivity to propene in the wet feed after pretreatment with carbon monoxide (Figure 1B and Figures 2C,D), the surface was analyzed under reaction conditions.

Fitting the Mn L_3,2-edges using a linear combination of relevant reference spectra for Mn²⁺, Mn³⁺ and Mn⁴⁺ species in the perovskite structure (Koch et al., 2020) showed that the oxidation state of manganese in the surface region changed under the different conditions (Figure 3). The dominant oxidation state of manganese in perovskites is Mn³⁺, accompanied by smaller proportions of Mn²⁺ and Mn⁴⁺ (Figure 4). (Levasseur and Kaliaguine, 2008; Liu et al., 2018b; Liu et al., 2019) Mn²⁺ species were already present in the fresh Sm_0.96MnO₃ catalyst (Koch et al., 2020) balanced by about 5% Mn⁴⁺ resulting in an average oxidation state of 3.03. After heating in 20% O₂, the average oxidation state decreased to 3.00. Switching to wet feed resulted in a further reduction. All Mn⁴⁺ vanished, and the fraction of Mn²⁺ increased to 4% balanced by Mn³⁺. While treating the sample in CO at 450°C, the Mn²⁺ fraction was growing to 5% leading to an average oxidation state of 2.95. Having switched back to wet feed, the Mn²⁺ fraction decreased to the level reached in the wet feed before the CO treatment at higher temperatures and dropped even further after switching back again to dry feed. Hence, the partial reduction of Mnⁿ⁺ species at higher temperatures was caused by the gas phase composition and is a reversible phenomenon of the Sm_0.96MnO₃ perovskite catalyst (Koch et al., 2020).

Under the more reducing conditions in CO, also Sm³⁺ was slightly reduced indicated by a small peak at 1,072 eV in the Sm 3d_5/2 spectrum (Figure 5) (Krill et al., 1980; Brunckova et al., 2019). This peak was observed after treatment in CO in wet feed and also still in the following dry feed (red spectrum in Figure 5). Its intensity decreased in dry feed compared to wet feed indicating that the process was reversible. But the reoxidation of the partially reduced A, i.e., Sm, was slow (Figure 5, inset) compared to the fast reoxidation of Mn²⁺. However, the reduction of Sm³⁺ was only detected by AP-XPS and not by AP-NEXAFS at the Sm M_5,4-edges (Supplementary Figure S6) showing that only the outermost surface fraction of Sm³⁺ was reduced. Since the electron kinetic energy in the AP-XPS experiments was with 170 eV distinctly smaller compared to the primary kinetic energy of the electrons created in the XAS process at an excitation energy in the NEXAFS experiment (about 800 eV), the resulting depth of information was much smaller in the Sm 3d AP-XPS measurements compared to the AP-NEXAFS measurement at the Sm M-edges. In the NEXAFS spectra at the Sm M-edges, the small surface fraction of reduced Sm²⁺ species was apparently hidden by the larger fraction of Sm³⁺ species in the deeper layers.

The partial reduction of Mn and Sm requires a charge compensation and the redox processes should be reflected in both the O 1s spectra and the O K-edge spectra. First, the O 1s spectra are discussed. In Figure 6 the measured O 1s and the corresponding fitted O 1s spectra are shown for the wet feeds and the CO treatment as surface scans (A), (B) and (C) as well as deep scans (D) and (E). Remarkably, the fraction of each oxygen species, which are introduced and discussed below, changes when comparing the outermost surface region (Figures 6A–C) with the deeper scans (Figures 6D,E). To describe all O 1s spectra measured under different conditions, different types of oxygen species should be considered according to the work published so far (Koch et al., 2020). The assignment and the fit model are summarized in Table 3. O_lattice represents the oxygen positioned in regular coordination in the crystal lattice. O_defect at 530.1 eV is oxygen, which is located in the vicinity of a defect (vacancy, off-centered position, hydroxyl group, etc.) (Fan and Goodenough, 1977; Sunding et al., 2011; Mierwaldt et al., 2014). Here it is noteworthy to mention that the oxygen atoms occupy two different positions in the orthorhombic lattice. Both oxygen atoms have each built up a distorted octahedral polyhedron consisting of two diametrically placed Mn atoms and four additional Sm atoms (Figure 7). These two octahedra differ slightly in their distortion. The distances between each oxygen position and its metal neighbors vary slightly, but the average distances between oxygen atoms and Mn and Sm atoms are similar (Supplementary Table S2). Since the two different oxygen species each have a different local environment, they should be spectroscopically distinguishable. However, it is not clear whether these differences can be resolved experimentally. This requires further experiments and the support of theory. From the available experiments, there is no clear evidence to further distinguish between an O_lattice1 and an O_lattice2, which lay below the signal of O_lattice and could also overlap with the signal of O_defect. Therefore, according to the literature, only the O_lattice and O_defect species are considered to describe the volume of the perovskite in the following discussion. The species O_surface at 530.9 eV is attributed to an electrophilic oxygen species adsorbed on the surface (Merkle and Maier, 2008; Mierwaldt et al., 2014; Stoerzinger et al., 2014). The binding energy 531.2 eV was uniquely assigned to O_carbonate because the peak intensity of this signal correlates with the peak intensity of C_carbonate from C 1s spectra at a binding energy of 288–289 eV (Stoerzinger et al., 2014; Koch et al., 2020). In addition, O_OH species are assigned to the binding energy at 531.8 eV (Nesbitt and Banerjee, 1998; Banger et al., 2011). Adsorbed H₂O and adsorbed oxygenates at binding energies of 532.2 eV and 534–535 eV, respectively, are present at higher temperatures only with a fraction of less than 1%.

Figure 8 and Figure 9 show the evolution of the fraction of oxygen species as a function of the conditions applied. Two photon energies were used for measuring the O 1s spectra. At 700 eV (inelastic mean free path IMFP = 0.6 nm), mainly the outermost surface is analyzed. At 1,300 eV (IMFP = 1.6 nm), deeper surface layers are also detected. In Figure 8 the depth profile for O_lattice and O_defect is shown. In general, the amount of O_lattice was larger in deeper layers than at the outermost surface. With increasing temperature, the fraction of O_defect increased, especially at the outermost surface (Figure 8), which supports the plausibility of the applied fit model.

After heating up in synthetic air to 270°C and subsequently reducing the oxygen content from 20 to 10%, the surface of the catalyst was healed and exhibited the largest fraction of O_lattice. Minor fractions of O_defect, O_surface, and O_OH were also observed indicating complex surface structures. Carbonates were already largely decomposed at the reaction temperature, so their contribution to the O 1s spectrum could be neglected. Remarkably, the increase in O_defect fraction after heating indicates that defect-rich surface structures have already been generated by the increased temperature and/or reduced oxygen partial pressure. This is consistent with model calculations that have shown that defect-rich surfaces of LaMnO₃ are energetically favored (Evarestov et al., 2003; Evarestov et al., 2004). The change in chemical potential by switching to wet feed caused an increase in the concentration of O_defect, O_OH and O_surface at the expense of O_lattice (Figure 8 and Figure 9). Hence, either the H₂O in the feed or the alkane and its derivatives or both actively modified the surface of the catalyst. After treatment with CO at 450°C and decreasing the temperature in CO to 270°C, a further little decrease of O_lattice and a distinct decrease of O_surface and O_OH was observed. At the same time, the fraction of O_defect increased. This is consistent with the partial reduction of Mn³⁺ and Sm³⁺ observed using AP-NEXAFS and AP-XPS, respectively (Figure 4 and Figure 5). The surface oxygen vacancies could have been formed by the oxidation of CO to CO₂ using surface oxygen as an oxidant (Hui et al., 1993). This was confirmed by repeating the CO treatment in a thermogravimetry experiment using a mass spectrometer for gas analysis. Only the formation of CO₂ was observed while heating up to 450°C in 2% CO balanced by Ar (Supplementary Figure S7). Switching to the wet reaction feed caused an increase in O_surface and O_OH and a simultaneous slight decrease in O_defect. The fraction of O_lattice at the outermost surface remained unchanged. Comparing the wet feed before CO treatment with the wet feed after CO treatment, the amounts of O_surface and O_lattice in the surface region were slightly lower after CO treatment, but the O_defect concentration was clearly higher, while the amount of O_OH remained similar (Figure 8 and Figure 9). Thus, the distribution of oxygen species in the surface region resembled the state before CO treatment, except for O_defect species, which was present in higher concentrations. The higher concentration of O_defect is supported by the thermogravimetry in CO-containing atmosphere. The found mass loss of about 0.6% corresponds theoretically to the removal of all oxygen atoms linked to a Mn or a Sm of the outermost surface layer (Supplementary Figure S8, right). This would imply the reduction of the surface to the metallic state by removal of all oxygen atoms of the outermost surface layer (Supplementary Figure S8, left), which is not found spectroscopically. Only 0.2% mass loss is estimated assuming a 001 surface showing a balanced A/B ratio as found for this catalyst (Table 2) and assuming fully saturated with oxygen atoms (Supplementary Figure S8, left). This means also oxygen from the bulk or at least from the near-surface range reacts with CO. Hence, more oxygen vacancies are likely formed and thus the number of O_defect increases, which is oxygen in the vicinity of a defective site. By removing H₂O from the feed, the O_lattice fraction in the deep and surface areas increased slightly again (Figure 8). At the same time, the proportion of O_defect on the surface decreased slightly. This decrease was more evident in the deep region, indicating an ongoing healing process in the bulk. Moreover, the amount of O_surface increased and the amount of O_OH slightly decreased (Figure 9). According to data measured in this study, O_surface appears to be required to start the healing process, reduce the amount of O_defect and increase the amount of O_lattice again.

Surface modifications and the increase of O_defect species were confirmed by near-edge X-ray absorption spectroscopy at the O K-edge (Figures 10A,B). In transition metal oxides containing 3d B metals, the first absorption above the threshold at ∼530 eV corresponds to the excitation from the O 1s state to the hybridized O 2p-B 3d state. The spectrum is difficult to interpret because the intensity and shape of this feature are related in a complex way to a multitude of overlapping and interconnected phenomena, such as oxidation state, spin state, degree of hybridization, strain and defects in the oxide structure. The 3d band of Sm_0.96MnO₃ is split into two peaks at 529.5 and 530.8 eV, which are in the simplified picture of the ligand field theory assigned to Mn e_g and t_2g states, respectively (Mierwaldt et al., 2014; Frati et al., 2020). In the spectra shown here (Figures 10A,B) the positions match well to those reported. The e_g peak is invariant at 529.5 eV while the t_2g peak maximum shifts from 530.5 eV in dry feed to 530.7 eV in wet feed after CO treatment. When the chemical potential of the atmosphere was changed, a change in the onset (during CO treatment at 450°C, grey spectrum) and intensity of the pre-edge peak was clearly observed (Figure 10B left), indicating that charge transfer had occurred. Also, the shape of the peak changed drastically. The overall area of the pre-peak was integrated taking into account a step function for the ionization to the continuum (Figure 10B right). The area below the pre-peak ranges from 2.79 during CO treatment to 3.45 in wet feed after CO treatment. The O K-edge after the CO treatment in wet feed differs strongly from those measured in the wet feed before the CO treatment. Since the pre-edge peak represents holes in the band, the number of holes is consequently reduced when the catalyst is reduced during CO treatment, resulting in a small area of 2.79. Switching back to wet feed yields in a significant increase in the area to 3.45 which is clearly visible from the increased t_2g peak at 530.7 eV (Figures 10A,B arrow). This number is also higher compared to 2.96 measured in the wet feed before CO treatment. Hence, the CO treatment tremendously modified the surface and the sub-surface of the catalysts in the way that the CO reduced partially the surface metals Mn and Sm (Figure 4 and Figure 5) and generated more O_defects (Figure 8), which is supported by thermogravimetry measurement. Finally, these surface modifications yielded in an increased number of holes in the t_2g band when the water of the wet feed or one of the other feed ingredients or reaction intermediates come into play. After switching to the dry feed, the area below the pre-peak is with 3.17 still increased compared to 2.96 in the wet feed before, but it already indicates the slow healing.

The indication of partial reduction of Mn³⁺ to Mn²⁺ in CO at 450°C was complementarily confirmed by Mn L_2,3-edge spectra (Figure 4) and valence band spectroscopy (Figure 10C). In addition, adsorption of propane and propene and the formation of OH groups were evidenced by the appearance of peaks at 8.8 and 10.2 eV (Wernbacher et al., 2019). Thus, the reactants are not only converted on the catalyst surface but also modify the catalyst surface, which results in a reactant-dependent catalytic behavior.

The optimized combustion method is suitable to achieve batches in gram scale of phase pure Sm_0.96MnO₃ powders having low bulk density due to the agglomerated crystallites forming caves, hollow spheres, and large networks (Supplementary Figure S3). The in this way prepared Sm_0.96MnO₃ can be operated as a catalyst in the oxidative dehydrogenation of propane (Koch et al., 2020). The Sm_0.96MnO₃ is characterized by a stable bulk crystal structure under the applied reaction conditions. Neither crystallite growth (Table 1) nor lattice expansion, which were observed for similar LaMnO₃ perovskites (Miyoshi et al., 2003; Koch et al., 2020) occurred. Both show the stability of the lattice and imply that extensive re-organization of the lattice can be neglected.

However, the functionality of the catalyst changes in terms of both activity and selectivity when different pretreatments and wet feed are employed showing that the chemical potential affects the surface of the catalyst and influences, thus, the catalytic properties. The chemical potential reversibly modifies the oxidation state of metals in the surface region. Even heating to 270°C in an oxygen-containing atmosphere and then switching to wet feed leads to full reduction of all initially measured Mn⁴⁺ species to Mn³⁺ and also to an increased formation of Mn²⁺. This illustrates the importance of operando techniques, which can reveal the dynamic redox behavior of the Sm_0.96MnO₃ catalyst under various conditions. The partial reduction of Mnⁿ⁺ in perovskites, e.g. in the oxygen evolution reaction (Mierwaldt et al., 2014) or the oxidative dehydrogenation of propane (Koch et al., 2020) has been related to H₂O in the atmosphere and is feasible when Mnⁿ⁺ occupies an A position (Ignatans et al., 2019; Koch et al., 2020) in the ABO₃ perovskite structure. In the A-deficient Sm_0.96MnO₃ catalyst prepared here, some small Mnⁿ⁺ ions probably sit on spacious A positions. The changes in oxidation state are balanced by types and amount of oxygen in the surface region. When the Mn is partially reduced, also the amount of O_lattice/(Sm + Mn) is reduced showing that the partial reduction is accompanied by loss of oxygen atoms sitting on a perfect lattice position (Figure 11 grey and purple circles). Thus, the increase in O_defect is expected, which supports the applied fit model. In general, more oxygen (O_all is the sum of all species) was found using AP-XPS (1.66 < O_all/(Sm + Mn) < 2.98) than with the chemical analysis of the bulk (O_bulk/(Sm + Mn) = 1.49), showing a highly defective surface. The chemical analysis cannot distinguish between oxygen in different configurations (e.g., O_lattice, O_defect), but gives just the sum O_bulk (oxygen determined by bulk chemical analysis). The O_lattice content determined by surface analysis varies as a function of conditions as follows: 0.72 < O_lattice/(Sm + Mn) < 1.56. It decreases to a minimum in the wet feed after treatment in CO at 450°C and slowly increases in the subsequent dry feed again (Figure 11). Hence, the amount of oxygen in the surface region in relation to A and B does not only depend on the type of B as previously discussed (Fierro and Tejuca, 1987) but also on the feed and the temperature. Interestingly, when only oxygen was present in the atmosphere at 270°C, bulk and surface analysis yielded the most similar figures for O_all/(Sm + Mn), namely 1.49 and 1.66, respectively, with O_lattice having the largest fraction in the surface analysis. This was larger here than in all other conditions. However, the crystal lattice was not perfect, as the O_lattice/(A + B) ratio of 1.18 in the outmost surface layer was lower than the expected ratio of 1.49 for a perfect Sm_0.96MnO₃ lattice (Figure 11, dotted line). This is because there are always unsaturated atoms on the surface. Therefore, only a perfect, defect-free lattice would result likely in a consistent bulk and surface composition. Conversely, it can be deduced that the over-stoichiometry of oxygen in the surface region already indicates additional oxygen species, with the interaction of this oxygen with the metals being lower than the average M-O interaction in the bulk. Noteworthy to mention, the additional oxygen is not coming simply from adsorbed H₂O wetting the surface because the corresponding water signal at 533.2 eV (Table 3) decreases after heating to 270°C in all atmospheres due to high temperature (Figure 8). The oxygen excess on the surface is based on O_OH, O_surface, a little amount of O_carbonate and O_defect. Especially, the O_defect is the major non-lattice oxygen species. The amount of O_defect is increased in wet feed (Mierwaldt et al., 2014; Koch et al., 2020) and further increased by reductive treatment in CO. It slowly decreases in the wet feed again and decreases, even more, when dry feed is employed (Figure 8). Simultaneously, Mn and Sm are partially reduced and reoxidized, respectively. Moreover, the O_defect increases apparently at the expense of O_lattice when the reaction feed streams over the catalyst (Figure 8). The oxygen can undergo hydrogenation through propane and its derivatives creating thus an O_OH, also indicated in the difference plots of the valence bands (Figure 10C), when the oxygen is bound to an Mn (Li et al., 2019). In vicinity to this in situ formed hydroxyl group also O_defects are created causing thus the increase in surface concentration of O_defect species in the feed. These findings suggest that the creation of the O_defect is related to both hydroxylation and oxygen vacancy formation. Moreover, the peak increase in the pre-peak region of the O K-edge NEXAFS spectra at 530.8 eV in wet feed underlines the formation of defective structures. This is also reflected in the valence band measurements, which provide information on the occupied states (Figure 10C).

In conclusion, the assignment of the O 1s signal at 530.1 eV to oxygen, which is part of irregular surface structures (O_defect), is strongly supported by simultaneously observed charge transfer in the O K-edge spectra and the valence band spectra. Hence, O_defect represents the defective state of the surface and this defective perovskite surface catalyzes the oxidative dehydrogenation of propane more to propene compared to a less defective perovskite surface. Once the healing is finished, the initial catalytic properties of Sm_0.96MnO₃ are recovered, which is reflected in the similar performance in the dry feed after the different pretreatments (Figure 1A). Moreover, remembering that Mn³⁺ is partially reduced in wet feed but simultaneously more holes are observed indicating reactant-induced charge distribution in the outermost surface region. Therefore, the healing cannot be easily described as the filling of formed oxygen vacancies. Many processes cause charge transfer and may thus contribute to the catalytic properties of the Sm_0.96MnO₃ catalyst in oxidative dehydrogenation of propane.

Remarkably, switching from dry to wet feed results in an increase in the selectivity to propene. This is more pronounced after additional treatment at higher temperatures and more reducing chemical potentials. Moreover, it is reversible after switching to dry feed. The products formed in propane oxidation are only propene and CO₂. This implies that possibly formed intermediates are fully oxidized under the conditions applied. Changes in selectivity can be caused by either enhanced formation of propene or by reduced formation of CO₂. In order to compare the intrinsic selectivity of the catalyst after various treatments, the initial formation rates of propene and CO₂ are compared (rates extrapolated to 0% conversion). The ratio between the initial propene formation rate and the initial CO₂ formation rate is depicted in Figure 12A. The ratio is almost independent of the pretreatment when the dry feed is set. It slightly increases after steaming and CO treatment (red columns). In contrast to the dry feed, the ratio distinctly increases when the wet feed is employed. In addition, steaming and even more CO treatment cause a further distinct increase of the ratio. Whether the distinct improvements in the wet feed are more related to the increase in propene formation rate or by reduced CO₂ formation rate becomes apparent when the formation rate of each product, i.e., propene or CO₂, after respective treatment are compared to the initially measured formation rate of each product before pretreatments (Figure 12B). The loss in activity after treatments (Figure 1) is clearly mirrored in the loss of the CO₂ formation rate. The corresponding ratio of the initial CO₂ formation rate after each treatment and the initially measured CO₂ formation rate (Figure 12, “initial”) decreases accordingly. The decrease of the CO₂ formation rate appears significantly after treatment in steam and even more in CO. In particular, the CO₂ formation rate strongly decreases in the wet feed. Therefore, treating the catalyst with reducing atmospheres results mainly in the loss of formation rate of the total combustion product CO₂. Presumably, the surface structure responsible for total oxidation is degraded by the treatments, because the feed was the same before and after treatments and thus their effects related to the reactants can be excluded. However, at higher conversions of propane, the performance of the catalyst approaches to those initially measured suggesting that the structures required for total oxidation are slowly recovered. It still cannot be deduced whether the CO₂ which is less formed due to the treatment originates from direct oxidation of propane or oxidation of intermediately formed propene. However, the selectivity changes could indicate an important role of oxygen on perovskite catalysts in propane oxidation. Thus, O_surface species could be responsible for both the high activity and the high total oxidation yield. These oxygen species are present in all conditions applied. Although the selectivity to propene is improved by reductive treatments in wet feed, CO₂ formation cannot be fully suppressed. However, the CO₂ formation rate increases when switching from the wet feed to the dry feed, which is in agreement with the observed decreased selectivity to propene and thus to increased total oxidation. In this discussion, it is assumed that the changes in selectivity are governed by the changes in CO₂ formation rate while the apparent formation rate of propene varies little (Figure 12B). The O_surface species might be more important than O_OH and O_defect in terms of selectivity, which explains the lower selectivity to propene in dry feed, although more O_defect was measured here (Figure 8 and Figure 9). The O_surface species could also be required for the healing of the oxygen vacancies and thus reduction of O_defect species, which have been created by treatments at higher temperatures. This is shown in Figure 13. The selectivity to propene increases after CO treatment with increasing proportion of O_defect, but decreases again after switching to dry feed, because O_defect decreases and O_surface increases. The concentration of surface defects and the concentration of electrophilic oxygen on the surface are interrelated and also have opposite and simultaneous effects on the selectivity to propene, which makes direct correlations difficult.

In summary, it can be concluded that the catalyst is stable in its bulk structure but the surface changes significantly under propane oxidation conditions as a function of the feed applied and thus as a function of the chemical potential. The local geometry of the terminating atoms changes and is limited to the outermost surface of the catalyst. These changes are clearly reflected in the core level and valence band spectra, as well as in the X-ray absorption spectra. However, in total these structural surface transformations are of minor extent. Therefore, the positive effect on the selectivity in propane oxidation is only slight and only significant at low conversion rates. But these results clearly show that for the optimization of a catalyst, not only its solid state and surface chemistry must be known, but also the kinetics of the chemical reactions in the solid state under different process conditions. In the search for more selective catalysts, other methods of catalyst synthesis should be used that more effectively increase the number of defects and thus the dynamic transformation of the catalytically active surface structures.