Redox Dynamics of Active VOx Sites Promoted by TiOx during Oxidative Dehydrogenation of Ethanol Detected by Operando Quick XAS

Titania-supported vanadia (VOx/TiO2) catalysts exhibit outstanding catalytic in a number of selective oxidation and reduction processes. In spite of numerous investigations, the nature of redox transformations of vanadium and titanium involved in various catalytic processes remains difficult to detect and correlate to the rate of products formation. In this work, we studied the redox dynamics of active sites in a bilayered 5% V2O5/15% TiO2/SiO2 catalyst (consisting of submonolayer VOx species anchored onto a TiOx monolayer, which in turn is supported on SiO2) during the oxidative dehydrogenation of ethanol. The VOx species in 5% V2O5/15% TiO2/SiO2 show high selectivity to acetaldehyde and an ca. 40 times higher acetaldehyde formation rate in comparison to VOx species supported on SiO2 with a similar density. Operando time-resolved V and Ti K-edge X-ray absorption near-edge spectroscopy, coupled with a transient experimental strategy, quantitatively showed that the formation of acetaldehyde over 5% V2O5/15% TiO2/SiO2 is kinetically coupled to the formation of a V4+ intermediate, while the formation of V3+ is delayed and 10–70 times slower. The low-coordinated nature of various redox states of VOx species (V5+, V4+, and V3+) in the 5% V2O5/15% TiO2/SiO2 catalyst is confirmed using the extensive database of V K-edge XANES spectra of standards and specially synthesized molecular crystals. Much weaker redox activity of the Ti4+/Ti3+ couple was also detected; however, it was found to not be kinetically coupled to the rate-determining step of ethanol oxidation. Thus, the promoter effect of TiOx is rather complex. TiOx species might be involved in a fast electron transport between VOx species and might affect the electronic structure of VOx, thereby promoting their reducibility. This study demonstrates the high potential of element-specific operando X-ray absorption spectroscopy for uncovering complex catalytic mechanisms involving the redox kinetics of various metal oxides.


Surface VOx and TiOx density
The surface density of TiOx (SDTiOx) was found in the next way: Where %TiO2 is a TiO2 loading in %, NA -Avogadro's constant, MTiO2 -TiO2 molar mass in g · mol -1 , SAsurface area of the catalyst after TiOx supporting.

In situ diffuse reflectance (DR) UV-Vis spectroscopy
Silica was used as a diluent to ensure DR UV-vis signal linearity, with signal intensity (KMQSS-032G-6, Omega Engineering) was custom fit to directly probe the sample bed through a spare port and sealed with a PTFE ferrule.
For dehydration experiments, the samples were treated in 30 cm 3 /min 10% O2/Ar (Certified Standard, Praxair) at 400 °C for 1 h (ramp rate 10 °C/min) and cooled to 120 °C (ramp rate 10 °C/min) for spectral acquisition. Spectra (200 -800 nm, 1 nm resolution, 0.1 s/nm averaging) were collected in double-beam mode with full slit height, a slit beam width of 4, and a 1.5 absorbance filter in the reference beam. The reflectance spectra were analyzed using the Kubelka-Munk formalism to convert the reflectance into the equivalent absorption coefficient, The direct allowed optical band gap (Eg) was determined from the x-intercept of the linear fit of a plot of ( ( ∞ )ℎ ) 2 versus ℎ with data points selected to yield the largest positive slope, where ℎ is the incident photon energy. The absolute error in the band gap was set to 0.1 eV owing to the uncertainty in the manual fitting process. Spectra were further analyzed by calculating a difference spectrum between the 5% V2O5/15% TiO2/SiO2 spectrum and the 15% TiO2/SiO2 subtrahend spectrum, where spectra were first constant-baseline corrected and a 1.23x multiplication factor was applied to the 15% TiO2/SiO2 spectrum. The direct band gap of this difference spectrum was calculated as mentioned above. Finally, the 5% V2O5/15% TiO2/SiO2 and 15% TiO2/SiO2 spectra were analyzed by least-squares fitting in Fityk version 1.3.1. 3 Spectra constant-baseline corrected and fit simultaneously with Gaussian peaks in the 50 000 -20 000 cm -1 and 50 000 -25 000 cm -1 ranges for 5% V2O5/15% TiO2/SiO2 and 15% TiO2/SiO2, respectively. All peak positions and FWHM values were constrained to be equal to other across the spectra, while peak areas were constrained to be non-negative. Finally, direct band gaps of the Gaussian peaks were calculated as mentioned above.

In situ Raman spectroscopy
The dehydrated Raman spectra of the 15% TiO2/SiO2 and 5% V2O5/15% TiO2/SiO2 samples were acquired with a Horiba LabRAM HR Evolution confocal Raman spectrometer. A He-Cd 5 laser source (Kimmon IK5751I-G) generated the 442 nm laser excitation with a filtered power output of 7.5 mW. A confocal microscope with a 50x objective (Olympus BX-30-LWD) was used to focus the laser onto the sample. The 520.7 cm -1 band of a silicon wafer standard was used to calibrate the spectrometer prior to spectral acquisition. A spectral resolution of <2.7 cm -1 was achieved while averaging 5 scans (120 s/scan) and using a 100 μm hole. The in situ Raman spectra were collected in a high-temperature reaction cell (CCR1000, Linkam Scientific) controlled with a high temperature controller (TMS94, Linkam Scientific). The inlet 10%O2/Ar gas mixture used in the dehydration treatment was supplied using a Brooks mass flow controller (5850E) with a set point of 30 cm 3 /min. Approximately 30 mg of each sample was loaded onto a quartz wool-padded sample cup before sealing the reaction cell. The samples were heated to 450 °C at a rate of 10 °C min -1 , where they were dehydrated for 1h before cooling down to 120 °C at a rate of 10 °C min -1 , and finally acquiring the spectra.

X-ray beam damage tests at the V K absorption-edge
To make sure that there is no influence of the X-ray beam on the state of the measured VOx species, we performed a series of dedicated experiments at relevant temperatures. We have performed oxygen cut-off experiments with a different X-ray beam size (from 300 x 200 μm 2 to 500 x 400 μm 2 ) controlled by the Rh-coated toroidal mirror. In addition, we have done a periodical X-ray beam on/off switching experiments. Prior to the beam-switching experiment, the catalyst was exposed for 30 min to steady-state conditions at 160 °C in 1.6 vol% EtOH, 6vol% O2 in He. We started the data acquisition while exposing the sample to the ethanoloxygen mixture (1.6 vol% EtOH, 6.4 vol% O2 in He) at 160 °C. After 10 min, we switched the beam off by closing a shutter, while continuing the data acquisition, and switched it on again after 10 min. For the analysis, 20 XAS scans were averaged and normalized as described in the main text. The time interval without beam was removed from the data file. The processed spectra from both types of experiments were analyzed using a linear combination fit using the XAS spectra resolved by MCR-ALS (obtained previously from TPR and oxygen switch-off experiments) in the ProQEXAFS software. 4

Time-resolved Ti K-edge XAS experiments and data analysis
We performed two types of transient experiments, named modulation excitation (ME) experiments while monitoring the Ti K-edge XAS. The first type of experiment was similar to the oxygen cut-off experiments performed at the V K-edge: for 5 min the catalyst was exposed to the ethanol-oxygen flow (1.6 vol% EtOH, 6.4 vol% O2 in He) and for the next 5 min in an 6 oxygen-free ethanol containing mixture (1.6 vol% EtOH in He). At each temperature, 10 switching cycles were performed. In the second type of experiment, the oxygen-containing mixture (6.4 vol% O2 in He) was alternated with the ethanol-containing mixture (1.6 vol% EtOH in He). The gases were switched every 5 min and 10 cycles were performed at each temperature.
The four Ti K-edge XAS spectra were averaged, background-subtracted, and normalized to the edge jump of one. In the pre-edge region, the background subtraction was performed using a linear function in the interval of -61.0 --8.7 eV (relative to E0=4966.0 eV). For normalization in the post-edge region, a cubic polynomial function in the 100.0 -492.6 eV interval was used.
A bespoke python script was used for the phase-sensitive detection (PSD) analysis and period averaging. PSD allows the analysis of tiny changes caused by periodic perturbation in the composition of gas mixture. 5 It transforms time-resolved spectra into phase-resolved spectra and can be described by the following equation: where ω is the frequency of the stimulation, k is the demodulation index, φ PSD is the phase angle for demodulation k, I(t) and I(φ PSD ) are the intensities in the time and phase domains, respectively.
Thus, in each ME XAS experiment the initially measured 12 000 spectra (10 cycles of 10 min with 2 scans/s), after processing and period averaging ended up in one averaged gas switching cycle consisting of 300 spectra with a time resolution of 2 s.
To receive an average in the certain feed composition spectra (which were used for differential spectra calculations), we used phase-averaged spectra with serial numbers 14-149 and 164-299 (spectrum 0 corresponds to gas switching); fist 15 spectra under each (new) conditions were not considered to exclude the transitional period from one condition to another ( Figure S3).

X-ray Crystallographic Analysis
For single crystal X-ray diffraction analysis, suitable crystals were placed onto MiTeGen loop pins coated in paratone oil and mounted under a flow of nitrogen at 100 K on a Bruker Smart Apex II or Bruker D8 Venture diffractometer with CCD area detector using Mo Kα irradiation. Using Olex2 6 the structures were solved with the Super flip package 7,8 and refined using SHELXL. 9 All non-hydrogen atoms were refined with anisotropic displacement parameters. Hydrogen atoms were placed in ideal positions and refined as riding atoms.

Determination of the number of components in V K-edge XANES data
To determine the number of spectral components in the time-resolved V K-edge XANES spectra, we analysed the MCR-ALS results, received using different number of components.
SIMPLISMA approach implemented in the MCR-ALS GUI2.0 software 10 helped us to guess the initial spectra. The lack of fit in each solution was quantified according to this equation: where dij is the element of the data matrix D, eij is the element of the residual matrix E (see equation 1 in the main text).

MCR ALS of V K-edge XANES spectra
The final solution was found with the use of initial guesses of V 5+ , V 4+ , and V 3+ component spectra. The first and the last spectra from the EtOH TPR experiment were used as the initial guesses for V 5+ and V 3+ , respectively. As the initial guess for V 4+ , we used the last spectrum measured in ethanol during oxygen cut-off experiments at 210 °C. The sodium metavanadates β-NaVO3 and α-NaVO3, sodium decavanadate Na6V10O28•18H2O and Na1.164V3O8 were prepared following published methods. 11 Ammonium decavanadate ((NH4)6V10O28•6H2O) was prepared using the method described in 12 .

Synthesis of V-containing references
V(=O)(OTBOS)3 complex was synthesized according to the procedure reported in 13 .
V(OTBOS)4. The titled complex was synthesized according to the reported procedure for analogous vanadium(IV) siloxide complexes. 14

Pre-edge analysis of V K-edge XAS
The area under the pre-edge peak, the position of the center of mass (centroid) and the halfedge step position (the energy at 0.5 a.u. adsorption) were calculated for each resolved and reference spectra. For a more accurate calculation of the pre-edge area and the pre-edge centroid position, cumulative distribution function (CDF, eq. S2) was used as a baseline to correct the edge rising ( Figure S4). The edge jump was fitted with CDF with the use of the least square method (implemented on the base of Python). The CDF was fitted in two energy intervals: before the edge and at the edge raise; these intervals were chosen manually for every calculated spectrum, all of them are shown in Figures S5, S6. The resulting peak was integrated.
In the case, if a spectrum contained an additional shoulder on the edge (B-peak in Figure S4 b), the peaks received after subtraction were fitted with multiple pseudo-Voight functions to identify and subtract the contribution of the edge shoulder ( Figure S4 c).
Where x -the energy scale, μ -fitted parameter, defines the x position of fCDF, σfitted parameter, defines the slope of fCDF. Figure S7 shows the IR spectra of ethanol and possible catalytic products. For the quantitative analysis of ethanol and acetaldehyde concentrations during operando oxygen cut-off experiments, specific bands belonging exclusively to these molecules (1294-1181 cm -1 for ethanol, 1853-1658 cm -1 for acetaldehyde) were extracted and analyzed using ALS-MCR. A series of IR spectra corresponding to different concentrations of pure ethanol and acetaldehyde were also included in the experimental data set for the ALS-MCR analysis. 10 The applied constraints were: non-negativity of the spectral components and the sum of concentrations of all components < 1.

Quantitate analysis of the products by IR spectroscopy
For the quantitative analysis of ethylene concentration (observed only in the TPR experiment), the intensity of the band at 948 cm -1 was analyzed. Since ethane does not show any bands, which do not overlap with the bands of the other products, its quantitative analysis was hindered. To follow the ethane formation in the TPR experiment, we used the intensity of the band at 3060 cm -1 . The IR spectra of carbon dioxide and carbon monoxide are also shown in Figure S7, however, their formation was not observed under catalytic conditions.

Quantification of the rates of V 5+ consumption and V 3+ and V 4+ formation upon oxygen switch off
To estimate the initial rates of V 5+ consumption and V 4+ and V 3+ formation after oxygen cutoff, we fitted the concentration profiles of the V 5+ , V 4+ and V 3+ components with the following functions: Linear: where t is the time, t0 is the time when the change in component concentration is observed (t0 = 10 min for V 4+ and V 5+ , when oxygen is switched off; t0=10.4 min for V 3+ , since V 3+ appears with a time delay), y is the fraction of certain vanadium species, Y0, A, A1, A2, τ, τ1, and τ2 are the fitted parameters.
The constant component concentrations before oxygen is switched off (t<t0) in every experiment were fitted as a plateau (y=Y0). The interval when oxygen is switched off was fitted with linear, exponential, or double exponential functions. The fitting was implemented in the Origin software using piecewise PWL2 (for linear fitting), ExpAssosDelay1 (for single exponential function) and ExpAssosDelay2 (for double exponential function) functions.
The choice of the fitting function was done based on the data quality of the resolved concentration profiles. The fitting interval usually was 1-15 min, however, in several experiments, we had to limit it to 12 min to diminish the impact of noise arising from deeper vanadium reduction. Importantly, this limitation does not significantly influence the results, since the complementary concentration profiles (V 5+ ↔V 4+ ) usually were fitted in the whole time interval. The functions chosen to fit each concentration profile are shown in Table S2, the result of the fits are shown on Figures S15-S16. 11 After fitting, the derivative at the time t0 (coincides with the time when oxygen is switched off for V 4+ and V 5+ components and 0.4 min later for V 3+ ) was taken and multiplied on the total amount of vanadium sites in the catalyst bed to receive rate in mol/min units.

Quantification of the rates of V 4+ re-oxidation by molecular oxygen upon oxygen switch on
To estimate the rate of V 4+/3+ re-oxidation, initially we planned to fit V 4+ and V 3+ concentration profiles similarly to ones made upon oxygen switch-off. However, the noise in the data did not allow us to make reasonable fitting and we decided to use a complementary component (V 5+ ), which has a lower noise level. However, we could use only those temperatures where the fraction of produced V 3+ is negligible (T=160-190 °C). Thus, calculated V 5+ formation rates could be attributed exclusively to V 4+ → V 5+ process.   Equation y = y0 + A * ( mu * (2/pi) * (w / (4*(x-xc)^2 + w^2)) + (1 -mu) * (sqrt(4*ln (2)) / (sqrt(pi) * w)) * exp(-(4*ln (2) Table S3.
The in situ DR UV-vis spectra of 15% TiO2/SiO2 and 5% V2O5/15% TiO2/SiO2 are further analyzed via least-squares fitting in Figure S26 a and Figure S26 c, respectively. An additional band at 35 000 cm -1 is present for the 5% V2O5/15% TiO2/SiO2 sample. Tauc plots that identify the direct band gap of each fitted Gaussian peak are reported in Figure  A summary of the results from least-squares fitting of the in situ UV-vis spectra of 5%   20 Fang calculated molar extinction coefficients for hydrated TiOx species supported on mesoporous silica and found that the molar absorption coefficient for TiO4 monomers is 10-10 2 times that for TiOx polymer and TiO2 nanoparticle species, 22 consistent with the finding that the absorption coefficient of cations usually increases by one or two orders of magnitude if the coordination changes from centrosymmetry to non-centrosymmetry. 23,24 The relative quantification of TiOx species for 15% TiO2/SiO2 and 5% V2O5/15% TiO2/SiO2 catalysts are reported in Table S5. It is found that both catalysts overwhelmingly consist of TiO5/TiO6 oligomers, with trace quantities of TiO4 monomers and TiO2 nanoparticles, consistent with literature reports for this Ti surface density on the silica support. 25,26

In situ Raman spectroscopy
The in situ Raman spectra of dehydrated 15% TiO2/SiO2 and 5% V2O5/15% TiO2/SiO2 are presented in Figure S27. For the 15% TiO2/SiO2, the bands at 148, 392, 507, and 630 cm -1 are assigned to the Eg(1), B1g(1), A1g/B1g(2), and Eg(3) vibrational modes of crystalline TiO2 (anatase). 27,28 The blueshift in the peak position of the Eg(1) mode relative to the reported value for bulk TiO2 (anatase) (144 cm -1 ) is attributed to phonon confinement and indicates the presence of crystallites that are ~6 nm in size. 28 The band at 258 cm -1 is tentatively assigned to the δ(Ti-O-Si) bending mode of two-dimensionally dispersed surface TiOx domains on the SiO2 support. 19 The bands at 670 and 779 cm -1 are assigned to the ν(Ti-OH) and ν(Ti-(OH)2) stretching modes of type I and geminal titanols, respectively. 29,30 The type I titanol has been assigned to TiO4 monomeric species. 29,30 The band at ~835 cm -1 is assigned to a ν(Ti-O-Ti) stretching mode of two-dimensionally dispersed oligomeric TiOx domains. 31,32 The band at 1067 cm -1 is assigned to either a symmetric or asymmetric ν(Ti-O-Si) mode, where the discrimination between two types of vibrational modes is not possible owing to their spectral proximity. 19,29,33 The band at 1164 cm -1 is assigned to the asymmetric, longitudinal optical stretching mode of the silica support, νas(Si-O-Si)LO. 34,35 Altogether, the Raman spectrum indicates the presence of TiO2 (anatase) nanoparticles, monomeric surface TiO4 species, and oligomeric surface TiOx species on the silica support. Furthermore, the two-dimensionally dispersed titania species possess hydroxyls. Figure S27. In situ Raman spectra of (a) 15% TiO2/SiO2 and (b) 5% V2O5/15% TiO2/SiO2 at 120 °C in 10% O2/Ar following treatment at 450 °C for 1 h in the same gas mixture.
For the 5% V2O5/15% TiO2/SiO2 sample, the band at 154 cm -1 is assigned to the Eg (1) mode of ~4 nm TiO2 (anatase) nanoparticles. 27,28 The intensity of the Eg(1) mode relative to other spectral features is much weaker when compared to the 15% TiO2/SiO2 sample, but it is noted that Raman scattering with 442 nm laser excitation will be affected by the optical absorption from VOx domains and, thus, a quantitative comparison cannot be made with the 15% TiO2/SiO2 by Raman spectroscopy. An additional band is observed at 465 cm -1 that is suggestive of Eg mode of rutile TiO2 (reported at 447 cm -1 for the bulk). 36 However, the presence of rutile TiO2 is excluded as its B1g mode is reported at 143 cm -1 while the 447 cm -1 Eg mode will redshift to lower wavenumbers with either phonon confinement, 36 vanadium doping, 37 or increasing laser power. 38 The presence of V2O5 nanoparticles is excluded due to the absence of a band at 996 cm -1 assigned the ν(d 1 ) mode of α-V2O5 39,40 as well as the absence of bands at 942 and 1021 cm -1 assigned to the ν(Vb-O1b) and ν(Va-O1a) of β-V2O5, respectively. 40 The band at 249 cm -1 is assigned to the δ(V-O-Ti) bending mode of V-O-Ti bridging bonds. 26 However, the possibility that δ(V-O-V) vibration also contributes to the 249 cm -1 band cannot be ruled out, but such vibrations can couple with the stretching νs(V-O-V) mode and appear at higher wavenumbers. 40 Compared to the 15% TiO2/SiO2 sample, the shift from 258 to 249 cm -1 suggests that Ti-O-Si and Ti-O-Ti bonds are not responsible for the observed 249 cm -1 band. The band at 339 cm -1 is assigned to the δ(V=O) bending mode of the vanadyl moiety. 41 The bands at 465 and 606 cm -1 are assigned to the symmetric and asymmetric bending modes of V-O-V bonds, respectively, 41 confirming that VOx species exist as oligomeric domains. The relatively higher intensity of these modes is tentatively attributed to 43 resonance enhancement from the 442 nm (22 600 cm -1 ) laser excitation, which is within the optical absorption envelope of the VOx domains (see in situ DR UV-Vis discussion). While silica possesses a band at 610 cm -1 due to 3-membered ring defects, 34,35 its signal is weaker than for the so-called "R-band" of 4-,5-, and 6-membered rings in the 200-500 cm -1 region 34,35 that are not observed. Furthermore, such a band (especially when considering its FWHM) is not clearly observed for the TiO2/SiO2 catalyst. It is, therefore, concluded that the 606 cm -1 band cannot be attributed to the silica support. The bands at 800 and 916 cm -1 are assigned to the asymmetric and symmetric stretching mode of V-O-Ti bonds 42 , respectively, confirming that VOx species anchor on TiOx domains. The band at 1034 cm -1 is assigned to the stretching mode of vanadyl V=O moieties. 26,[42][43][44] The band at 1070 cm -1 is assigned to either a symmetric or asymmetric ν(Ti-O-Si) mode, where the discrimination between two types of vibrational modes is not possible owing to their spectral proximity. 19,29,33 It is noted that bands for νs(V-O-Si) and νas(V-O-Si) 33 of VOx anchored to the silica support are also expected at ~920 cm -1 and ~1070 cm -1 , respectively. 33,44,45 However, the vanadyl stretch for such silica-supported VOx domains is observed at 1041 cm -1 with 442 nm laser excitation. 45 Thus, the presence of VOx domains anchored to the silica support is tentatively ruled out. The band at 1148 cm -1 is assigned to the asymmetric, longitudinal optical stretching mode of the silica support, νas(Si-O-Si)LO. 34,35 Altogether, the Raman spectrum indicates the presence of two-dimensionally dispersed surface VOx oligomers anchored to the TiOx domains.

Determining the number of components in V K-edge XANES
The Lack of fit of MCR ALS solution received with the use of 2, 3, 4, and 5 pure components is shown in Figure S28.

X-ray Crystallographic Analysis of tailored molecular Vreferences
The overall molecular structure of V(OTBOS)3(thf)2 was revealed by X-ray diffraction analysis, though the quality of the diffraction data was low because of a disorder of TBOS ligands, as well as a rotation of coordinating THF; only the connectivity of the molecular structure was clarified. Because of this, a molecular structure of analogous complex, V(OSi(OMes)(O t Bu)2)3(thf)2, is shown here instead.

Assignment of MCR-resolved V K-edge spectra
To assign the MCR resolved V K-edge spectra, we measured V K-edge spectra of 21 Vcontaining reference compounds and characterized them by calculation the pre-edge height, the pre-edge surface area, the pre-edge center of mass position, and the edge position. In Figure   S32 and S33 we built correlation plots using the pre-edge peak area or height and the pre-edge centroid position. [47][48][49] The MCR-resolved components are included in Figure S32 and S33. The assignment of the oxidation state with such correlations, however, is less reliable than one made using the pre-edge area and the edge position ( Figure 4) and does not allow to clearly distinguish neither V 4+ and V 5+ nor V 3+ and V 4+ .

V 4+ re-oxidation rate
Under steady-state conditions, the rate of V 5+ reduction and V 4+ re-oxidation are equal, this can be expressed using the next equation: = or Where rred and rox are the rates of V 5+ reduction and V 4+ re-oxidation, respectively; kred is the constant of reduction and kox' is the constant of re-oxidation (includes the oxygen concentration).
Thus, under steady-state conditions, the ratio between V 5+ and V 4+ -species is equal to: consumption (upon oxygen switch-off) and appearance (upon oxygen switch-on) in oxygen cut-off experiments, respectively.
The fits of V 5+ formation upon oxygen switch on are shown in Figure S20.
As we discussed in the main text, due to oxygen leak, acetaldehyde concentration never reached 0. As a result, a fraction of V-species was involved in the MvK cycles also in the "oxygenfree" ethanol-containing phase and was not detected among reduced species. This means that the observed rate of V 4+ production is underestimated by 2*rAcH (rAcHobserved in oxygen switch off phase rate of acetaldehyde production, 2 mol of V 4+ are produced per 1 mol of acetaldehyde). We took it in an account, and in Table S5, we show the expected V 4+ concentration with the correction to the O2-leak through the graphite window.

Activity of titanium during alcohol oxidation
In the first type of operando ME Ti K-edge XAS experiments, the ethanol concentration in the feed was held constant, and the flow of oxygen was periodically switched on and off every 5 min (denoted as EtOH+O2→EtOH cycling). In the second type of the ME Ti K-edge XAS experiments, the oxygen and ethanol flows were periodically (every 5 min) alternated (denoted as O2→EtOH cycling). The latter experiments represent the most extreme conditions of ODH, reminiscent of a chemical looping operation. After recording 10 cycles at each working temperature, the resulting Ti K-edge XAS spectra were normalized and analyzed using the PSD approach in the XANES region. The phase-resolved spectra from the ME XAS experiments over the supported 5% V2O5/15% TiO2/SiO2 catalyst are shown in Figure S37 together with the Ti K-edge XANES spectrum of the same catalyst measured at 350 °C in an oxygen flow. Norm. Abs., a.u. Fig.1a Comparison of the phase-resolved Ti K-edge spectra in the pre-edge region (4968-4970 eV) ( Figure S37) revealed interesting details. When ethanol is constantly present in the feed ( Figure   S37 b, d), no significant changes in the pre-edge peaks A1 and A2 are detected. However, small changes in the peak intensities of A1 and A2 can be observed if ethanol is periodically removed from the feed (during O2 / EtOH experiments, Figure S37 c, e, f). These peaks have mostly local character; changes in A2 intensity indicate changes in the symmetry or the coordination number of titanium. 50 At 350 °C, A1 and A2 peaks are more intense in the presence of ethanol (the bold curve in Figure S37 f representing EtOH to oxygen switch has positive values; also see Figure S21 c) indicating that the titanium coordination number decreases in the reducing environment. This could be an evidence of oxygen-vacancy formation upon partial titanium reduction.
The O2/EtOH ME Ti K-edge XAS experiments performed at lower temperatures (160 and 210 °C, Figure S37 c, e and S21) demonstrate the reverse character of the A1 and A2 changes: the bold curves (O2 to EtOH switch) are negative in the A1 and A2 regions meaning that the A1 and A2 peaks are less intense in ethanol. This suggests that titanium has more neighbors in the first coordination sphere in EtOH and its coordination structure is closer to a perfect octahedron. Presumably, when ethanol is present in the feed, it can adsorb at the 5-fold coordinated titanium sites leading to an increase in the coordination number of Ti. 51 Apparently, at lower temperatures (160 and 210 °C) this effect is greater than oxygen-vacancy formation due to titanium reduction. However, the intensities of these changes are extremely small indicating that only a small fraction of titanium atoms undergo any transformation ( Figures   S21, S23).