Reduced Carbon Monoxide Saturation Coverage on Vicinal Palladium Surfaces: the Importance of the Adsorption Site

Steps at metal surfaces may influence energetics and kinetics of catalytic reactions in unexpected ways. Here, we report a significant reduction of the CO saturation coverage in Pd vicinal surfaces, which in turn is relevant for the light-off of the CO oxidation reaction. The study is based on a systematic investigation of CO adsorption on vicinal Pd(111) surfaces making use of a curved Pd crystal. A combined X-ray Photoelectron Spectroscopy and DFT analysis allows us to demonstrate that an entire row of atomic sites under Pd steps remains free of CO upon saturation at 300 K, leading to a step-density-dependent reduction of CO coverage that correlates with the observed decrease of the light-off temperature during CO oxidation in vicinal Pd surfaces.


Contents
(800 ºC, 5 min) were employed to clean the sample, until no contamination was detected by means of XPS, and a smoothly increasing splitting of the hexagonal LEED pattern away from the (111) plane was obtained. [2][3][4] The split-pattern remains stable upon CO dosing at 300 K, proving the high stability of the monatomic step array. Beam-damage leads to CO cracking and graphitic carbon build-up, as revealed by the characteristic C 1s satellite at 284 eV. 4 The intensity (area under the peak) of graphitic carbon, though, remains below 5% of the total C 1s emission at every point on the curved sample.

S1.2 W -model
The step-size-W model is used to evaluate, out of the experimental photoemission intensity variation across the curved surface, the effective area of the vicinal surface affected by adsorption at atomic steps (W ). The model assumes constant CO layer structure and coverage at terraces and steps at every vicinal angle α, 4 leaving the lateral extension of the step W as a fitting parameter. On these grounds, it is easy to see that terrace (Θ T ) and step (Θ S ) coverages must vary as as a function of α as: where Θ 0 T and Θ 0 S refer to the constant saturation coverage at (111) terraces (0.5 ML at 300 K 5 ) and steps, while d −1 and h stand for the step density and the height of a Pd monoatomic step (2.25 Å). By consistently fitting α-dependent terrace and step peak intensities to Equation 1 and Equation 2, W and the step coverage Θ 0 S are readily determined. Since d −1 = sin(|α|)/h, both equations are expressed in terms of the step density or the vicinal angle.

S2 Supplementary data and analysis
S2.1 0.5 ML CO on Pd(111) LEED simulation LEED patterns for several structures were simulated using the kinematic approximation, to explore the qualitative effects of various CO site occupations and arrangements in comparison to that of a c(4×2) phase involving hollow (f t , h t ) and bridge b t sites. Standard in-plane kinematic structure factors were calculated for each phase using the form factor of a carbon atom and a shape factor corresponding to Gaussian beams of width 0.15 Å −1 . Each pattern shown in Figure S2 is the average from the set of equivalent domains expected based on the substrate p3m1 symmetry. All of them correspond to 0.5 ML coverage with hollow/bridge=1 ratio, as deduced from the XPS fitting (Fig. 2 of the main text). For the ease of comparison, the original c(4×2)-4CO spots are indicated with red circles, while the connecting arcs of the saturation superstructure ( Fig. 1 in the main text) are marked with blue shades. The highest symmetry c(4×2)-4CO superstructure is shown in Figure S2a), showing the expected diffraction spots. If antiphase domains are considered [ Figure S2b)], a significant amount of spots are still observed at lower wavevector q than those of the original structure. Several low-q spots arise when tilted domain boundaries are considered [ Figure S2c)], yet the spots split, resembling our arc-like pattern. If vertical domains are simulated [ Figure S2d)], the low-q spots remain, although the main spots split, although in the opposite direction as ours. Finally, when a random site occupation is simulated [ Figure S2e)], only the main spots of the c(4×2) structure arise, since most of the lower-q reflections no longer appear in the simulation. Therefore, we propose the formation of several disordered domains of c(4×2)-4CO superstructures, which would give rise to the pattern that we observed at Pd(111).

S2.2 Deconvolution of core level spectra
The fitting procedure was performed using the lmfit Python package. 6 We considered Doniac-Šunjić lines 7 convoluted with a Gaussian profile and a Shirley-type 8 background for the model. As reported in the literature, 1,9,10 two vibrational excitations of the adsorbed CO molecules were considered for each of the CO species included in the fitting routine. Such satellites were fixed at +0.2 and +0.4 eV above the major contribution, while their height were fixed at 45% and 9% of that of the main line. There parameters were extracted from the fitting of the (111) surface after CO saturation. Assuming an equal behaviour for all peaks, very similar factors (i.e. asymmetry, width, satellites) were used for all CO species discussed in the manuscript. The area of the satellites was added to that of the major contribution in order to properly calculate the coverage of each of the individual CO adsorption sites.
The fitting routine for the stepped surfaces is not straightforward, since the peaks overlap and are difficult to distinguish. In order to reasonably extract each species, we added a single line (step-CO peak) at the beginning of the CO uptake curves (Fig. 3 of main text). Such contribution was allowed to grow until a step coverage of 0.5 ML was reached (i.e. one CO per two Pd step atoms). After saturation of steps, the parameters of the step peak were held constant (only small variations were allowed), and an additional line (f t ) was included in the fitting procedure. Mimicking the (111) plane uptake shown in Fig. 2 of the main text, a third peak (b t ) was added when the overall spectrum started to shift towards larger binding energy.

S3 Density functional theory calculations
Density function theory (DFT) calculations have been performed with the Vienna Ab-initio Simulation Package (VASP) implementation, 11-13 using the PBE 14 exchange-correlation functional and the projector-augmented wave (PAW) method 15,16 to describe the interaction between the valence electrons and the core. The Brillouin zone was sampled using the Monkhorst-Pack 17 scheme.
The lattice parameter for Pd (3.946 Å) was obtained by optimizing the bulk unit cell with a 800 eV kinetic energy cut-off and a (16,16,16) k-point sampling. A 450-eV kinetic energy cut-off was used for the slab calculations. The calculations were done with a Gaussian smearing using σ = 0.1. The considered surface models are summarized in Table 1 reporting the surface cell, total number of layers (N Layer ), number of frozen layers (N frozen ) and kpoint sampling. Repeated slabs were separated by at least 12 Å vacuum. Structures were optimized until the forces were below 0.01 eV/Å.

S3.1 CO adsorption on Pd
It is experimentally established that the stable low-coverage site for Pd(111) is the fcc/hcp hollow site. In contrast to Pt(111), 18-20 the hierarchy of adsorption energies calculated with PBE for Pd(111) is in agreement with the experimental observation. The issue for Pt (111) has been traced to an underestimation of the HOMO-LUMO (5σ-and 2π * ) separation in CO using conventional approximations to the exchange-correlation functional and the unbalanced 5σ-and 2π * -contributions to the CO-metal bond is present also for Pd(111). One sign that the bond is not properly described is that the bridge position is only a shallow minimum on the potential energy surface. Thus, all structures with CO in bridge position on the Pd surfaces have been obtained by constraining the C and O atoms to the center of the bridge positions. The CO adsorption energy is calculated as: with E CO,metal being the total energy of the adsorbed structure, E CO the molecular gasphase reference and E metal the metal surface reference. Exothermic adsorption implies that the adsorption energy is negative. The corresponding desorption energy is the converse of the adsorption energy and, hence, positive for exothermic adsorption. heats of adsorption were evaluated from the adsorption isotherms in the zero coverage limit.
The description of small gas-phase species is often an issue calculations using semi-local functionals, which causes problems when evaluating adsorption energies as the cancellation of errors is limited. One pragmatic way to reduce the problem is to shift the gas-phase energies so the heat of formations for a set of gas-phase reactions agree with tabulated experimental values. 22 We have adopted this approach and minimized the mean absolute error and maximal error by correcting the gas-phase total energies for 10 gas-phase reactions including CO, H 2 and O 2 . 22 The total energies for gas-phase CO, H 2 and O 2 are in this way shifted by -0.290 eV, +0.153 eV and +0.287 eV, respectively. Only the value for CO is used in this work. Because the calculated zero-point energy for CO adsorbed in bridge and fcc site on Pd(111) is similar, we use an overall shift of +0.346 eV to adsorption energies evaluated without zero-point correction and gas-phase correction (as reported in Table 3) to the values in Figure 1b in the main text.
CO adsorption has been explored for a range of different surfaces and coverages. The resulting adsorption and differential adsorption energies are listed in Table 3. For the stepped surfaces, different adsorption sites across terraces have been considered, as schematically depicted in Fig. S4.

S3.2 CO desorption
To determine the desorption energy for CO at the experimental conditions and compare it to the calculated values (Fig. 1b, main text), we calculated the rate constant for desorption from the adsorption rate constant and the equilibrium constant: where p is the CO pressure, s 0 = 0.9 the sticking coefficient, A site the surface site area, ∆H the corrected enthalpy of adsorption, S surf the entropy of the adsorbate and S gas the entropy of the molecule in that gas-phase. Figure S3 shows the adsorption/desorption rate versa desorption energy for different CO pressures. Lower CO pressures shift the desorption energy to higher values.

S3.3 C 1s core level shift
To identify possible adsorbate configurations on the stepped Pd surfaces, we calculated the C 1s core level shift (CLS) of CO adsorbed in different positions and at different coverages.
To compare the CLS between stepped and flat surfaces, shifts were calculated with respect to a carbon atom placed deep in the bulk of the slab. To reduce the interaction between the carbon atom placed in the bulk and the slab surfaces, the slab thickness was increased to at least 8 layers. CLS's were calculated including screening of the core hole. 23 In this final-state approach, the CLS is calculated as a total energy difference between systems with a core-ionized atom replacing either the atom of interest or the reference atom: E surf C1s is the total energy with the C 1s core-hole in one of the adsorbates on the surface and E ref C1s is the total energy with the C 1s core-hole in the reference carbon atom. A schematic representation of the used structure is given in Figure S5a. Because the same super-cell structure is used to calculate E ref C1s and E surf C1s it is not necessary to calculate the ground state energy of the unperturbed structure. The calculations with core-ionized atoms were done using neutral super-cells, by adding an extra electron to the valence electrons.
The CLS is related to the adsorption energy in the Z+1 picture. 24 In Figure S5b we represent the CLS relative to the low-coverage fcc on Pd(111) against the adsorption energy.
For CO adsorption on (111) transition metal surfaces, a linear trend of the CLS with coverage has been reported previously, 25-27 as well as for ethylene adsorption on Pd clusters. 28 The linear trend is also found for the stepped surfaces in the present work. The data for the coverage dependence of the CLS on Pd(111) is reported in Table 4. For the low coverage on Pd(112) and Pd(221), the data is reported in Table 5. The structure sketched in a) corresponds to a c(2 × 2)-2CO superstructure, while that of b) reflects the same superstructure with alternating anti-phase domains. c) combines tilted domains of the aforementioned structure, whilst d) corresponds to two vertical domains of the original superstructure. Finally, e) shows the simulation considering random occupation of sites (note the extinction of low-q spots).