Interplay between CO Disproportionation and Oxidation: On the Origin of the CO Reaction Onset on Atomic Layer Deposition-Grown Pt/ZrO2 Model Catalysts

Pt/ZrO2 model catalysts were prepared by atomic layer deposition (ALD) and examined at mbar pressure by operando sum frequency generation (SFG) spectroscopy and near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) combined with differentially pumped mass spectrometry (MS). ALD enables creating model systems ranging from Pt nanoparticles to bulk-like thin films. Polarization-dependent SFG of CO adsorption reveals both the adsorption configuration and the Pt particle morphology. By combining experimental data with ab initio density functional theory (DFT) calculations, we show that the CO reaction onset is determined by a delicate balance between CO disproportionation (Boudouard reaction) and oxidation. CO disproportionation occurs on low-coordinated Pt sites, but only at high CO coverages and when the remaining C atom is stabilized by a favorable coordination. Thus, under the current conditions, initial CO oxidation is found to be strongly influenced by the removal of carbon deposits formed through disproportionation mechanisms rather than being determined by the CO and oxygen inherent activity. Accordingly, at variance with the general expectation, rough Pt nanoparticles are seemingly less active than smoother Pt films. The applied approach enables bridging both the “materials and pressure gaps”.

T he adsorption and catalytic oxidation of CO on Pt are among the most frequently examined surface processes due to their environmental and industrial relevance. Pt exhibits superior catalytic properties for various applications, such as (preferential) CO oxidation for emission control or cleaning of hydrogen streams for fuel cells. 1−5 Despite efforts to replace expensive Pt by cheaper materials, its activity can typically not be matched. Thus, the focus is rather on reducing the Pt amount, e.g., by using Pt atoms, clusters, and small nanoparticles 6−13 (or alloys and core−shell structures 14,15 ) on suitable support materials. It is, however, still challenging to obtain detailed knowledge about increasingly smaller nanoparticles, especially about their inherent activity and metal/ support interaction. 16 −21 In recent years, significant advances have been made in model catalysis, enabling surface characterization at (near) atmospheric pressure, overcoming the "pressure gap", 22−31 but bridging the "materials gap" is evenly important. Previous single-crystal studies have provided fundamental insight, but they cannot fully mimic nanoparticles 25,32,33 (with support effects being apparently inaccessible), which is why more realistic model systems are required, such as oxide-supported nanoparticles/islands 11,16,25,34 or inverse systems. 35−37 In this contribution, we present Pt/ZrO 2 model catalysts prepared by atomic layer deposition (ALD) that were examined at mbar pressures by operando sum frequency generation (SFG) spectroscopy and near-ambient pressure Xray photoelectron spectroscopy (NAP-XPS), with simultaneous mass spectrometry (MS) product analysis, and complemented by density functional theory (DFT) calculations.
ALD has been widely used in industrial manufacturing, 38,39 especially for dielectrics and microelectronics, and is receiving increasing attention for (upscalable) catalyst preparation. 40 The current model catalysts consist of a zirconia film, ALDgrown (400 cycles) on a Si (100) wafer, and Pt deposits prepared by different numbers of ALD cycles (10−250; see transmission electron microscopy (TEM) images in Figure 1a and Figure S3). Whereas the zirconia ALD generated a uniform 42 nm-thick oxide support, using a few (10,50) Pt cycles produced small Pt particles up to 8 nm in size ( Figure  1a, right panel). Upon applying 125 or 250 deposition cycles, the Pt particles coalesced into islands, finally forming a homogeneous Pt film of uniform ∼10 nm thickness (Figure 1a, left panel). These nucleation and growth processes 41 allow for the preparation of different well-defined Pt/ZrO 2 model catalysts ranging from isolated Pt nanoparticles to bulk-like thin films. Herein, the 50 and 250 cycle samples are the most informative ones (other samples are described in the Supporting Information). The ALD approach to catalyst synthesis is not new, 42,43 but the combination with operando surface spectroscopy (SFG and NAP-XPS carried out in highpressure cells with simultaneous MS gas-phase analysis) and DFT calculations provides a complementary picture.
The standard cleaning used for single crystals in ultrahigh vacuum (UHV), i.e., sputtering/annealing, could not be applied as it would have destroyed the ALD samples. Inspired by the (re-)activation of technological catalysts, all samples were thus cleaned from carbonaceous residues by heating in 10 mbar O 2 to 400°C and in 20 mbar CO/O 2 (1:1) to 300°C.
The Pt morphology was then addressed by polarizationdependent SFG of CO adsorption (10 mbar CO at 150°C; Figure 1b; see the Supporting Information for SFG theory and fit values). The spectra show the on-top CO resonance region, as no other binding geometries were observed. Two polarization combinations were employed: ppp (has its maximum intensity for CO bonds parallel to the macroscopic surface normal; black in Figure 1b) and ssp (has its maximum intensity for CO bonds inclined with respect to the macroscopic surface normal, i.e., around 30−40°depending on molecular polarizability; red in Figure 1b). 44,45 Due to the angular dependence, the resulting intensity ratio I ppp /I ssp for CO is expected to decrease with increasing bond inclination of the molecules. 46,47 Because the Pt film of the 250 cycle sample consisted of planar islands with a uniform height of about 10 nm, adsorbed CO was mostly perpendicular to the ZrO 2 / Si(100) surface so that the CO peak intensity was high in ppp and very low in ssp (ratio of 17.4). In light of our previous study of CO/Pt(111), 47 assuming an identical optical interface model, this would correspond to an average CO tilt angle of ∼5°(relative to the macroscopic surface normal), although this value is just meant to show a trend. In contrast, the 50 ALD cycle Pt film consisted of small particles (about 8 nm) with multiple facets, many of which are no longer parallel to the substrate. On these inclined facets, CO still adsorbs  The surface roughness increases from left to right, as indicated by the decreasing intensity and redshift of resonance positions. The indicated values were obtained from data fits (solid lines). For rough surfaces, spectra at 275°C showed a diminishing on-top CO, which was irreversible upon cooldown ( Figure S10). This cannot be explained by a decrease in CO coverage, as for Pt(111), the spectrum remained almost unchanged.

ACS Catalysis pubs.acs.org/acscatalysis
Research Article perpendicularly, but the CO bonds are inclined with respect to the macroscopic surface normal. Accordingly, the SFG intensity is lower in ppp, resulting in a much lower I ppp /I ssp ratio of 2.3. Following the same assumptions as above, the average CO tilt angle would be ∼40°(note that this again is just to show the trend). This rough estimate agrees with the facet inclination and ratio from TEM images. Apart from the intensity, the peak position and peak symmetry/asymmetry are noteworthy. The peak position depends on the coordination of the Pt adsorption site and the CO surface coverage (inducing chemical and dipole− dipole interactions). 27,48,49 As the coverage should be (nearly) the same under identical pressure and temperature conditions, the 3 cm −1 difference points to slightly rougher surfaces for the nanoparticle sample. The (a)symmetry of an SFG signal depends on the amplitudes A r or A nr and phase difference ϕ between resonant (adsorbed CO) and nonresonant signal contributions 50−52 (see the Supporting Information). The nonresonant background A nr may originate from Pt surface defects (changing the electron localization at the surface) and electronic contributions of the ZrO 2 /Si(100) substrate. Comparing the spectra of the 250 and 50 cycle samples, the intensities of the resonant and nonresonant contributions are much more similar in the case of the particulate film (smaller particle size, more defects, and more metal/oxide interface), while the resonance phase relative to the nonresonant background, as obtained by means of the quantitative deconvolution of the data, is found to be similar for both CO-Pt systems. This leads to a more asymmetric line shape for the 50 cycle Pt, directly reflecting its surface morphology.
In order to further characterize the ALD-prepared model catalysts, they were compared to Pt single crystals at different temperatures. Figure 2 shows the SFG ppp spectra of smooth (UHV annealed to 800°C) Pt(111), 250 and 50 ALD cycle Pt/ZrO 2 , and sputtered Pt(111) in 10 mbar CO. Respective fitting values are given in the Supporting Information (our system has a spectral accuracy of 2 cm −1 ). At 175°C, the characteristic on-top CO on Pt(111) was observed at 2092 cm −1 , matching the saturation coverage. 49,53 The 250 cycle sample exhibited a continuous Pt surface (Figure 1 and Figures  S2 and S3), but was still rougher than the annealed Pt(111), as indicated by the redshifted wavenumber (2089 cm −1 ) and increased peak asymmetry. Adsorbed CO on the 50 cycle Pt/ ZrO 2 sample, consisting of 8 nm (connected) particles, exhibited a similar wavenumber, indicating identical coordination, but lower intensity and higher asymmetry due to the inclined facets. The higher number of low-coordinated (step/ kink) sites 54,55 on the ALD samples was confirmed by comparison with the CO spectra of sputtered Pt(111) (2081 cm −1 ) and Pt(110) (2075 cm −1 ) ( Figure S9), showing even lower wavenumbers. 53,56 Apparently, the ALD Pt catalysts exhibit roughness intermediate between the annealed and sputtered Pt(111). Now, turning to the CO adsorption at higher temperatures (225/275°C), the decreased CO coverage induced a redshift. 24,48,49,53 For Pt(111), the on-top CO signal redshifted, but the intensity was similar to that at 175°C. Analogously, the CO signals of the ALD samples and sputtered Pt(111) exhibited small redshifts at 225°C (and a minor intensity loss). However, at 275°C, the rougher surfaces showed a pronounced intensity loss, peak shift, and phase alteration. Previously, a similar observation on polycrystalline Pt foil 57 was explained by CO desorption, but CO is more strongly bonded to steps/defects than terraces. 58,59 Furthermore, the spectral changes were irreversible upon cooldown in CO (see Figure S10) and ϕ changed significantly, ruling out simple adsorption/desorption and rather suggesting a permanent modification/blocking of the adsorption sites. It has been reported that stepped Pt surfaces or small Pt particles/clusters may cause CO dissociation, forming a carbon overlayer, 60 in line with the current observation. Indeed, the CO spectrum of the 250 cycle Pt at 10 mbar/225°C agrees quite well with the reported values of Pt(557) at 30 mbar/250°C, which showed CO dissociation at 275°C. The dissociation hypothesis is supported by the fact that adding O 2 to CO and heating reversed the spectral change by reoxidizing carbon to CO 2 ( Figure S11). Nevertheless, CO dissociation on Pt has been controversially discussed for a long time.
This motivated the DFT calculations of CO dissociation on smooth and rough Pt surfaces (Figure 3). CO dissociation on Pt(111) and Pt(211) is strongly endothermic and barriers >3 eV have been reported in the low coverage limit, 61 making this process improbable. For a more facile dissociation, the adsorbed state of CO should be destabilized, whereas the final state needs to be stabilized. A destabilization of the adsorbed state is achieved by increasing the coverage (gas pressure), whereas the final state is stabilized by CO 2 formation. In particular, direct CO 2 formation according to the Boudouard reaction (CO* + CO* → CO 2 + C*) hinders the backward C−O association reaction. Similar contributions (local coverage and final state) to favoring the Boudouard reaction have been reported for PtSn 61 and Cu. 62,63 For

ACS Catalysis pubs.acs.org/acscatalysis
Research Article efficient CO−CO coupling and reaction, the C weight of the 2π* orbital on one reacting molecule should overlap with the O weight on the other molecule. This may be accomplished on stepped and kinked surfaces, so we considered two model structures: a Pt adatom coordinated with two CO molecules on Pt(100) and Pt(410). The barriers were evaluated at coverages obtained from a thermodynamic analysis ( Figures  S14 and S15). The barriers for the reactions are 1.8 and 2.1 eV for dissociation at the adatom and Pt(410), respectively. For the reaction at the adatom (Figure 3b−d), one of the CO molecules on the Pt adatom is reacting with a CO on the (100) facet. The transition state is a bent O−C−O structure (Figure 3c), whereas the final state (Figure 3d) is gas-phase CO 2 and the remaining carbon atom is in a highly coordinated position. A fourfold coordinated C on Pt(100) is (per carbon atom) as stable as graphite, thus stabilizing the final state. 64 The reaction path on Pt(410) is different as both reacting CO molecules are below the step (Figure 3e−g). The 2π*−2π* match is, in this case, enabled by an initial bending of the CO close to the step, and the final state again has a fourfold coordinated C atom. Accordingly, on the rougher Pt surfaces, CO dissociation is facilitated by a barrier that is lower than 1 eV as compared to the smooth surfaces. Turning to CO oxidation on the Pt/ZrO 2 model catalysts, Figure 4a,b shows the SFG ppp spectra acquired in a reaction mixture of 10 mbar CO and 20 mbar O 2 from 150 to 500°C. At 150°C, the Pt particles were on-top CO-covered (poisoned) and thus inactive, with 2091 cm −1 indicating high-coordination sites for the 250 cycle sample. The 50 cycle sample showed a 10 cm −1 redshift, indicating lower coordination/higher roughness, with lower intensity and more asymmetry, due to the nanoparticle morphology. Upon temperature increase, the peak of adsorbed CO decreased and redshifted (due to decreasing CO coverage) and finally disappeared when the Pt surfaces were fully oxygen-covered and thus active: at 400°C for the Pt thin film, but at 450°C for the Pt nanoparticles. The temperature-dependent shifts were 2091−2080 cm −1 for the smoother Pt film and 2081− 2074 cm −1 for the rougher Pt nanoparticles, the latter wavenumbers indicating stronger bonding on the rougher surfaces. The ignition temperatures were corroborated by the simultaneously acquired CO 2 -MS traces in Figure 4c,d and are in accordance with values reported in the literature 31,65 (note that MS measurements without simultaneous SFG, "laser-off", ruled out any laser-induced effects; Figure S16). As long as adsorbed CO is present, the oxidation reaction is largely inhibited, as the Pt surface can only be in a single stable state, either CO-poisoned (inactive) or O-covered (active). 66 Accordingly, both SFG and MS indicated that about 50°C higher temperature was required for ignition on the rougher (50 cycle) Pt nanoparticles. This is unexpected, because rougher Pt surfaces are generally considered to be more active in CO oxidation, 9,66 as their low-coordinated sites (steps, kinks, and edges) bind both oxygen and CO stronger than terraces. The O-covered active state can be more easily established on rougher surfaces (indicated by the lower ignition temperature and higher CO tolerance) despite the higher CO oxidation barriers 67 (according to the Brønsted− Evans−Polanyi relation 68 ). However, in the current case, the inherent activity is not the only important factor, as shown in the following.
To further examine the reaction onset, NAP-XPS was applied as the second operando technique (which again rules out laser-induced effects). Due to technical limitations, the pressure was limited to 1 mbar CO and 2 mbar O 2 . The C 1s spectra acquired during CO oxidation ( Figure S17) detected graphitic carbon (284.7 eV), 69 adsorbed CO (286.2 eV), 59,70 a weak shoulder (around 288 eV; likely carbonate on zirconia), and gas-phase CO (∼291 eV). Figure 4e,f displays the fitted peak areas vs (increasing) temperature for adsorbed CO and carbon. In analogy to SFG, for smoother (250 cycles) Pt films, CO fully disappears at a temperature 50°C lower than for rougher (50 cycles) Pt nanoparticles (the absolute temperatures are lower due to the 10-fold lower pressure).
However, NAP-XPS showed that much more carbon was present during the reaction on the 50 cycle (rough) Pt nanoparticles, which even increased during the first two temperature steps, clearly indicating CO disproportionation (Figure 4f). Atomic carbon apparently poisons the (lowcoordinated) active sites for oxygen activation until it is removed by oxygen at higher temperatures. This effect explains the higher reaction onset temperature of the Pt nanoparticles despite their presumably more active surface. In contrast, the smoother 250 cycle Pt film was much less affected by C poisoning, yielding a lower reaction onset temperature. After

ACS Catalysis pubs.acs.org/acscatalysis
Research Article ignition, both Pt surfaces were O-covered and showed the expected hysteresis upon lowering the temperature ( Figure  S19). CO readsorbed at a comparably lower temperature paralleled by less coking (due to cooling in an oxygencontaining atmosphere). Throughout all experiments, the operando Pt 4f spectra revealed that Pt remained metallic ( Figure S17). In summary, combining the ALD model catalyst preparation, operando SFG/NAP-XPS/MS spectroscopy, and DFT calculations enabled us to build another bridge across the "materials and pressure gap". Few ALD Pt deposition cycles produced Pt nanoparticles with multiple inclined facets, whereas more deposition cycles (≥125) led to more uniform Pt films. The polarization-dependent SFG revealed the molecular orientation of CO (relative to the macroscopic surface normal) and thus both the morphology and roughness of different ALDgrown Pt model catalysts. Upon CO adsorption at mbar pressure around 275°C, Pt(111) did not show CO disproportionation, whereas rougher Pt particles/films and sputtered Pt(111) did. According to the DFT calculations, direct CO dissociation is unfeasible even at stepped Pt surfaces. Dissociation instead occurs at high coverages via a disproportionation reaction at low-coordinated sites that structurally promote CO−CO coupling and stabilize the remaining C atom. The effect of surface roughness on the CO oxidation was monitored at mbar pressure and elevated temperature by correlating the operando SFG and NAP-XPS spectra with the MS reactivity data. Different from the general expectation, the reaction onset temperature was higher for the smaller/rougher Pt nanoparticles than for the smooth Pt surfaces. The rougher surfaces were poisoned by carbon coking, detected by NAP-XPS, and explained by DFT calculations via CO disproportionation on favorable sites. Only after the removal of the carbon deposits did the rough Pt surfaces become active, but at 50°C higher temperature than for smooth Pt. Upon cooldown, the smooth Pt films exhibited a wider hysteresis window and were hardly affected by CO disproportionation. Future studies of ALD Pt particles and films on different support materials should reveal whether reducible supports facilitate activation at lower temperature and reduce/suppress the initial carbon poisoning.