Mechanistic Study of Carbon Dioxide Hydrogenation over Pd/ZnO‐Based Catalysts: The Role of Palladium–Zinc Alloy in Selective Methanol Synthesis

Abstract Pd/ZnO catalysts show good activity and high selectivity to methanol during catalytic CO2 hydrogenation. The Pd‐Zn alloy phase has usually been considered as the active phase, though mechanistic studies under operando conditions have not been conducted to verify this. Here, we report a mechanistic study under realistic conditions of methanol synthesis, using in situ and operando X‐ray absorption spectroscopy, X‐ray powder diffraction, and time‐resolved isotope labeling experiments coupled with FTIR spectroscopy and mass spectrometry. Pd‐Zn alloy‐based catalysts, prepared through reduction of a heterobimetallic PdIIZnII acetate bridge complex, and which do not contain zinc oxide or any PdZn/ZnO interface, produce mostly CO. The Pd‐Zn phase is associated with the formation of CO, and does not provide the active sites required to produce methanol from the direct hydrogenation of carbon dioxide. The presence of a ZnO phase, in contact with a Pd‐Zn phase, is essential for efficient methanol production.

with the XAS from the sample, was used for internal energy calibration. Transient responses of the catalyst, upon switching the reaction mixture, were monitored by means of an Omnistar GSD 300 O2 (Pfeiffer Vacuum) mass spectrometer.
Further in situ XAS measurements were then also made, at the Zn K-edge and Pd K-edges of PdZn/SiO2, PdZn/ Al2O3 and PdZn/ZnO/ SiO2 catalysts at the SuperXAS beamline, Swiss Light Source of the Paul Scherrer Institute (Villigen, Switzerland). After catalytic carbon dioxide hydrogenation, or after pre-treatment in hydrogen (see Figure captions for details), the aforementioned samples were transferred to a glovebox without exposure to air. Samples were then packed into either 2 mm (Pd K-edge) or 1 mm (Zn K-edge) internal diameter quartz capillaries and sealed using epoxy-glue. XAS data were collected in transmission using fast, gridded ion chambers, and a quick scanning channel-cut Si (111) monochromator (oscillation frequency of 1 Hz). [41] Either zinc or palladium foil standards were collected simultaneously for energy calibration. Initial analysis and energy calibration were performed using ProXAS v.2.34 software. [42] Zn K-and Pd K-edge XANES and EXAFS data were background subtracted and normalized using either, PAXAS, [43] Athena, [44] or Prestopronto. [45] Principal component analysis (PCA) was made using the ITFA software developed by Rossberg et al. [46] Fitting of the EXAFS was made using EXCURV (v. 9.3). [31] Operando steady-state 12 CO2/ 13 CO2 isotope transient experiment coupled with infrared spectroscopy Time-resolved isotope transient experiments coupled with FTIR were performed using a standard sandwich flow IR cell reactor configuration. [47] Before the measurements, the sample was diluted with pure silica (1:3 by weight), pressed in a self-supporting disk, and activated in helium for 2 hours at 260 ° C. (heating rate 10 °C./min). The sample was subsequently reduced in hydrogen (25 ml/min) for 1 hour at 260 °C under ambient pressure. The gas flow was then switched from hydrogen to a CO2/H2/Ar mixture (24 vol. %, 72 vol. % and 4 vol. % correspondingly, Messer).
Finally, the total pressure was increased to 15 bar (regulated by a Bronkhorst back-pressure regulator, EL-Press series). After reaching steady state carbon dioxide conversion (ca. 1 h on stream), the isotope switch was carried out using a two-position valve (VICI), from unlabeled 12 CO2/H2 (Messer; 99.5%) to a 13 CO2/H2 labeled mixture (Cambridge Isotopes Laboratories, Inc.; 99% 13 C). An inert tracer (Argon, 4 vol. % of the total flow rate) was used to correct for the gasphase hold-up in the reactor. Isotope labeling experiments were performed at a constant total gas flow of 25 ml/min. The 12 C/ 13 C switch was achieved without perturbing the steady-state of the reaction by maintaining the reaction temperature at 260 °C, the total system pressure at 15 bar, and at carbon dioxide conversion level of ≈2%. On-line MS analysis was performed using a quadrupole mass analyzer (Omnistar GSD 300 O2). The surface species on the catalyst were followed using IR spectroscopy (Thermo Nicolet iS50 equipped with MCT detector). The spectral resolution was 4 cm -1 , and spectra were acquired every 6 secs (by averaging of 4 scans per spectrum).
In the MS, the ion current for m/z = 40 (Ar tracer), 44, 45, 28, 29, 31 and 33 were continuously monitored to determine the isotope content of the original gas sample. Transient responses were normalized by the difference between the initial and final ion signals. The argon decay curve was used to determine the gas-phase holdup of the reactor system, since we assumed that the inert gas did not adsorb on the surface of the catalyst.

Microscopy
For the (scanning) transmission electron microscopy ((S)TEM) investigations, the material was dispersed in ethanol and a few drops of the suspension were deposited onto a perforated carbon foil supported on a Cu TEM grid. After evaporation of the ethanol, the grid was mounted on the single tilt holder of the microscope. TEM and STEM combined with energy-dispersive X-ray spectroscopy (EDXS) was performed on a Talos F200X microscope (ThermoFisher) with a high brightness field emission gun operated at an acceleration potential of 200 kV. The EDX system of this microscope consists of 4 silicon drift detectors (SDD), which enables one to record EDXS maps with good a signal-to-noise ratio in a relatively short collection time (10-20 min).

Catalytic experiments
Catalytic methanol synthesis over palladium-zinc based catalysts was also investigated using a fixed-bed stainless-steel reactor. In a typical experiment, 25 or 50 mg of the catalyst (fraction 50-100 μm), diluted (1:3 by mass) with silicon carbide (Sigma-Aldrich), was loaded between two quartz wool beads and positioned inside a stainless-steel tube reactor (6 mm outer diameter and 4 mm inner diameter). The reactor was mounted inside a single-zone furnace (Carbolite). The temperature was controlled with a Eurotherm 3508 controller using a K-type thermocouple positioned inside the catalyst-bed. The catalyst was activated firstly, under ambient pressure in a flow of argon (5.0 quality, 50 mL/min) at 260 °C (heating rate 5 °C/min) for 2 h, and secondly, by a reduction in hydrogen (50 mL/min) at 260 °C and ambient pressure for 1 hour. Finally, the total pressure was increased to 30 or 50 bar using a back-pressure regulator (Bronkhorst, EL-PRESS).
Carbon dioxide hydrogenation catalysis was initiated by switching the feedstock from pure hydrogen to a feed gas mixture containing 24 vol. % of carbon dioxide, 72 vol. % of hydrogen, and 4 vol. % of argon (all gases used were of 5.0 quality, Messer) at 260 °C (see details in Table   1). After 24 hours on stream, the catalytic properties of the material were further investigated under different temperatures (250, 240 and 220 °C). After any temperature changes, the catalyst was equilibrated for ca. 1 hour until a steady-state methanol yield was achieved. Catalytic data were then acquired for a further 1 hour. Analysis of outlet gases was performed using a 3000 Micro GC gas analyzer (Inficon) equipped with a 10 m Molsieve and an 8 m PlotU columns and TCD detectors.    Figure   S10b then shows the equivalent data rendered in Fourier transform representation of the k 3weighted EXAFS. In both cases, the black lines pertain to the experimental data, the red lines to the fits to that data derived from analysis using EXCURV. [31] Table S3 then gives the structural and statistical information arrived at analytically to yield the fits. In each case the black lines concern the experimental data, the red, fits to that data derived from analysis in EXCURV. [31] (k)) 2 x 100% Where i e and i t are the experimental and theoretical EXAFS respectively, and k is the photoelectron wave vector (Å -1 ). i is the uncertainty in the data, with 1/i = ki n / j N ki n (i e (j)) 2 .. A switch to a hydrogen flow precipitates a complete reduction of the Pd present, along with some of the ZnO, to yield a highly dispersed PdZn alloy phase consistent with an L1o structure (RPd−Pd = 2.9 Å, RPd−Zn = 2.6 Å) [22] . A further scattering interaction may be fitted to either Pd (3.51 Å) or Zn (3.69 Å) though, as can be seen from Table S3, it is not formally, by either change in R factor or Chi 2 , significant. The highly dispersed nature of this phase is indicated by both the reduced levels of coordination in both Pd-Zn and Pd-Pd shells compared to the bulk case (NPdZn = 8, NPdPd = 4 [22] ) and the lack of significant higher shell structure. In this sense, the EXAFS is therefore consistent with the XRD (Figure 2e), which reveals only very broad and weak reflections that may be associated with this phase.
A switch to the catalytic reaction feed does not elicit in any significant changes in the EXAFS and, overall, the k 3 -weighted EXAFS envelope remains rather similar to that observed under pure hydrogen. As such, once again, the alloy nature of the supported phase does not change radically as a result in the change of feed, save for in this case, there is no evidence in the FT for the presence of a third shell at around 3.5 Å.
A switch to a carbon dioxide flow, however, elicits a substantial change in the k 3 -weighted EXAFS, which principally manifests itself in a significant reduction in amplitudes. In Table   S3 this is reflected in a net decrease in the apparent coordination numbers associated with the Pd-Zn (2.57 Å) and Pd-Pd ((2.95 Å) interactions. As before, the third shell may also be fitted to either Pd or Zn but, again, with the recognition that, in terms of the goodness of fit obtained to the k 3 -weighted EXAFS through the addition of this shell, it cannot be regarded as significant.
Exposure to carbon dioxide at 15 bar and 260 °C does not seem to have affected the PdZn alloy nature of the sample, but has elicited some other sort of change, such as a decreased in average particle size, change in morphology, or overall levels of static disorder. However, whilst the results, especially for the Pd-Pd interaction, might indicate a substantial change in this respect, the correlation that exists between coordination numbers in EXAFS and the Debye-Waller (DW) term, does not permit us to specify with any certitude the source of this change.
Examination of the correlations that exists between the Pd-Pd coordination number and the associated DW term, show that the former may exist anywhere in the range 0.5 -2 and the latter, between 0.01 and 0.028 for virtually no change to the R factor or Chi 2 parameter. As a result, therefore, whilst the EXAFS confirms the retention of the PdZn alloy motif, it does not permit us to specify the effect of the carbon dioxide on the nanoparticles present in any more precise way than to state that the presence of this gas leads to a significant (static) disordering of this phase.
Overall therefore, in this case, Pd K-edge EXAFS shows that, once reduced in hydrogen, the nanoparticulate PdZn phase is retained under the conditions applied, irrespective of the gaseous environment it experiences. Whilst some evidence is forthcoming to indicate that this supported phase is, in some way subject to heightened levels of static disorder in the presence of pure carbon dioxide, any further specification as to how these increased levels of disorder manifest themselves on the nanoscale is not possible on the basis of this data.

Figure S11. STEM micrographs (a-b), TEM (c-d) as well as STEM (e)combined with energy dispersive X-ray elemental mapping (f-h) of
PdZn/SiO2 catalyst after catalytic carbon dioxide hydrogenation at 260 °C and 30 bar pressure.

Figure S12. STEM micrographs (a-b), TEM (c-d) as well as STEM (e)combined with energy dispersive X-ray elemental mapping (f-h) of
PdZn/ZnO/SiO2 catalyst after catalytic carbon dioxide hydrogenation at 260 °C and 30 bar pressure. Figure S13.

(a) k 3 -weighted Pd K-edge EXAFS derived from (as indicated) PdZn/SiO2 and PdZn/Al2O3 catalysts post reduction
in hydrogen; (b) the corresponding Fourier transforms of the k 3 -weighted data. In each case, the black lines refer to the experimental data, the red, to fits to that data resulting from analysis using EXCURV [31] . Figure S13a shows the k 3 -weighted Pd K-edge EXAFS derived from (as indicated) PdZn/SiO2 and PdZn/Al2O3 catalysts post reduction in hydrogen. Figure S13b shows the corresponding Fourier transforms of the k 3 -weighted data. In each case, the black lines refer to the experimental data, the red, to fits to that data resulting from analysis using EXCURV. [31] Table S4 then gives the structural and statistical data derived from the analysis.
Both of these systems are characterized by a Pd-Zn scattering shell at ca. 2.57 -2.6 Å. As such, and as previously, the formation of a nano-sized PdZn alloy is indicated. In contrast to the 2Pd-ZnO-np cases previously shown ( Figure S10), however, no significant PdPd shell ca. 2.95 Å is evidenced. Such a shell can be included and fitted to the data, but the addition of such a shell makes no difference to the goodness of fit and the DW factors associated with co-ordinations of only unity or less are unusually high ( ≈ 0.035). As such, if present at all in these materials this second shell of coordination is highly disordered.   (k)) 2 x 100% Where i e and i t are the experimental and theoretical EXAFS respectively, and k is the photoelectron wave vector (Å -1 ). si is the uncertainty in the data, with 1/i = ki n / j N ki n (i e (kj)) 2 .
f Chi 2 = statistical measure of goodness of fit (x 10 -6 ) However, there are also significant differences between the two systems. Firstly, whilst the amplitudes of the EXAFS are similar in each case, a measurable phase shift between these two cases can be observed across the range of the k 3 -weighted EXAFS, with the EXAFS due to the PdZn/Al2O3 system being shifted to higher k(Å -1 ). This is indicative of a contraction of the average Pd-Zn bond distance in this sample compared to the PdZn/SiO2 case. Analysis shows that the magnitude of the contraction is small (2.57 versus 2.6 Å i.e. 0.03 Å) but it is, nonetheless, indicated to be present. Secondly, as evidenced from the Fourier transforms of the EXAFS ( Figure S13b), there is a pronounced difference between the two cases in terms of higher shell structure, with the PdZn/SiO2 yielding evidence above 3 Å in the FT of a developed higher shell environment whereas this is, to all intents, absent in the PdZn/Al2O3 case. Lastly, in the PdZn/SiO2 case, a low level of O coordination (2.16 Å) can also be fitted, whereas it is absent in the PdZn/Al2O3 system. In the former case, the addition of this shell decreases the R factor and Chi 2 by ca. 10% and therefore can be considered as significant, whereas in the latter the addition of such a shell marginal at best in terms of the quality of the fit achieved.
As such, the Pd K-edge EXAFS in both of these systems confirms for the formation of PdZn alloys but, at the same time, shows that in a number of ways these phases are subtly different from each other according to the oxide upon which they are supported. To a first approximation, the observance of a defined higher shell structure and longer Pd-Zn first shell bond distances in the PdZn/SiO2 case would indicate that the alloys formed in this case are significantly more ordered, and possibly larger, than in PdZn/Al2O3. The optimally refined values for the first Pd-Zn scattering interaction (4 versus 3) would also be consistent with this proposition. However, and as with the previous Pd K-edge EXAFS analysis for the 2Pd-ZnOnp case, the correlations that exist between the coordination number and the DW factors mean that we cannot be certain of this difference. Figure S14. ( The overall Zn K-edge EXAFS scattering envelope ( Figure S14a) for both PdZn/SiO2, PdZn/Al2O3 samples appears very similar, whereas that derived from the PdZn/ZnO/SiO2 is radically different beyond ca. k = 5 Å -1 . The source of this difference is shown graphically by the FT representation of the k 3 -weighted EXAFS ( Figure S14b). For PdZn/SiO2, PdZn/Al2O3 the phase-corrected FTs are dominated by the scattering shell at ca. 2.6 Å that we may associate (see also Table S5) with the significant presence of the PdZn alloy phase. In these two cases, this is accompanied by a very broad feature in the FT, indicative of low Z (modelled as O) coordination to some fraction of the Zn. The breadth of this feature might indicate that a high level of disorder is associated with this interaction, and that most likely a number of different Zn-O bonding motifs are present post catalysis. Given that the XANES, from these two cases ( Figure 3d), points squarely to a predominance of zinc being present in a reduced form, and such low Z coordination is not evidenced at all to any significant degree in the Pd K-edge spectra (Figure 3c), this rather significant level of coordination might suggest a significant segregation of the zinc toward the surface of the alloy nanoparticles particles where they may be in contact with either reactant species or the underlying support. Furthermore, whilst very similar to each other, in these senses, the FTs and subsequent analysis, show that in others these two systems are rather different in terms of the longer range levels of order that are indicated to be present within them.
In the PdZn/SiO2 case, whilst there is some evidence for scattering shells beyond ca.
3.5 Å, this structure, is relatively weak and not so significant in terms of the total FT envelope.
By contrast, in PdZn/Al2O3 this higher shell structure is far more developed, and specifically for the scattering feature at ca. 5 Å. As shown in the fits, and in Table S5, this feature can be well-fitted assuming a further Zn-Pd scattering interaction @ ca 4.9 Å. Moreover, the addition of this shell is highly significant in terms of the effects it has on both the R factor and Chi 2 , which drop from 25 to 19 and 0.76 to 0.43 (x 10 -6 ) concurrent with its addition. However, this feature may also be accounted for to some degree by multiple scattering effects that arise from scattering interaction across a two-dimensional (d2h) arrangement of 4 Pd atoms around a central Zn atom. Overall, therefore, the Zn K-edge EXAFS demonstrates that whilst the PdZn phases present in these two samples are nominally similar, this phase, when formed on Al2O3, is significantly more ordered, in at least two dimensions, that its counterpart supported upon SiO2.
Different from both of these cases is the PdZn/ZnO/SiO2. Consistent with the reduced intensity of the low binding energy pre-edge feature observed in XANES, the EXAFS is dominated by a broad low Z (O) scattering shell (Figure 3f). Above ca. 3 Å there is also precious little evidence of any well-defined higher shells of coordination. In between these extremes, however, we find the shell that is indicative of the presence of the PdZn alloy, but at a much reduced level in terms of its contribution to the overall EXAFS envelope. Together with the XANES (Figure 3d) this sample is therefore indicated to be comprised of both PdZn alloy and zinc oxide phases with a much heavier weighting to the latter than the former. In contrast, in the PdZn/SiO2 and PdZn/Al2O3 cases, it is evident from the XANES and the EXAFS that considerably more of the zinc in the system is present in a metallic form within supported PdZn nanoparticles.
Lastly we note that (vide infra) the Zn-Zn scattering shell found to be present in all of these samples post reaction at ca. 2.9 -3 Å could, hypothetically, also be the result of the Zn-Pd interaction expected upon the basis of the PdZn alloy bulk structure (r = 2.95 Å). Attempts to fit this interaction using a Pd scatterer, rather than Zn, fail in the sense that they return unrealistically short bond distances coupled to excessively large DW factors. As such, this Zn-Zn scattering interaction we associate with some form of disordered ZnO phase, the levels of which vary greatly between the three cases but which are greatest in the PdZn/ZnO/SiO2 sample. Once again, we further note that, in this case in particular, the correlations between N and DW factor mean that little precision can be attached to the values of N, which in each case could realistically range between 1 and 4. *Single scattering analysis only: this scattering interaction can also be described to an extent solely upon the basis of multiple scattering occurring across a ZnPd4O2 unit of d2h symmetry. R and Chi 2 values in brackets refer to the goodness of fit obtained in the absence of the Pd shell @ 4.89 Å, or the use of full curved wave multiple scattering calculations. a N = coordination number. b R = bond distance in Angstrom. c DW = Debye-Waller (disorder) factor (2 2 ) where  2 = mean squared displacement of the atom pair with respect to each other. d EF = Fermi energy (eV) e R% = i N 1/i (i e (k) -i t (k)) 2 x 100% Where i e and i t are the experimental and theoretical EXAFS. respectively, and k is the photoelectron wave vector (Å -1 ). si is the uncertainty in the data, with 1/i = ki n / j N ki n (i e (kj)) 2 . f Chi 2 = statistical measure of goodness of fit (x 10 -6 )