PdAg/Ag(111) Surface Alloys: A Highly Efficient Catalyst of Oxygen Reduction Reaction

In this article, the behavior of various Pd ensembles on the PdAg(111) surfaces was systematically investigated for oxygen reduction reaction (ORR) intermediates using density functional theory (DFT) simulation. The Pd monomer on the PdAg(111) surface (with a Pd subsurface layer) has the best predicted performance, with a higher limiting potential (0.82 V) than Pt(111) (0.80 V). It could be explained by the subsurface coordination, which was also proven by the analysis of electronic properties. In this case, it is necessary to consider the influence of the near-surface layers when modeling the single-atom alloy (SAA) catalyst processes. Another important advantage of PdAg SAA is that atomic-dispersed Pd as adsorption sites can significantly improve the resistance to CO poisoning. Furthermore, by adjusting the Pd ensembles on the catalyst surface, an exciting ORR catalyst combination with predicted activity and high tolerance to CO poisoning can be designed.


Introduction
The rapid development of human society increases the demand for energy, which would then further exacerbate the environmental problems, such as climate change and air pollution, caused by the consumption of non-renewable fossil fuels. Therefore, there is an urgent need to develop clean, renewable, and high-capacity energy conversion/storage technologies [1,2], such as the use of fuel cells, which is an efficient and promising energy conversion technology with clean reaction products and zero greenhouse gas emissions [3]. However, the broad deployment of fuel cells is still a challenge due to the high cost and sluggish oxygen reduction reactions (ORR) of cathode electrocatalysts [4,5]. Despite that great efforts have been made to search for new ORR catalysts, the most effective catalysts currently used for ORR are still pure platinum (Pt) and its alloys [6,7]. Many noble metalbased catalysts, especially Pt-based electrocatalysts, are susceptible to inactivity due to carbon monoxide (CO) poisoning. CO molecules block the reaction pathway by binding tightly to the active site, thus limiting the overall efficiency of proton exchange membrane (PEM) devices. In this case, the development of highly active, durable, and low-cost non-Pt catalysts resistant to CO has attracted great interest [8][9][10].
To reach a feasible stage of practical and general use of PEM fuel cells, one of the main obstacles is to develop a new catalyst at least as active as Pt-based metals, but with a lower noble metal content. The relative composition and distribution of metallic substances in bimetallic structures change as a result of the simultaneous presence of various competing effects, i.e., electron ligand, strain, and ensemble effects [11]. Non-linear changes in reactivity and selectivity are usually observed. In this case, it becomes important to understand these various effects [12]. A promising strategy is to dope non-reactive host materials with trace amounts of transition metal atoms as active centers, in order to form the monomer of transition metal on the surface, the so-called single-atom alloy (SAA) [13,14]. also produces a second layer of Pd enrichment on the alloy surface. Obviously, the PdAg SAA alloy prepared by the casting and quenching strategy is a potential candidate as a new catalyst of fuel cells, and this unique distribution might play a key role in the catalytic properties.
In this paper, a density functional theory (DFT) study on the reactivity and thermodynamics effects of Pd ensembles on the PdAg(111) surface is presented to assess the effect of different sizes and configurations on the ORR activity. To account for effects introduced by subsurface Pd, we also investigated PdAg(111) surface alloys with a Pd subsurface layer underneath. These different surface structures with ORR intermediates (OOH, O, and OH) were considered using the limiting potential as a metric of activity to study their trends, as influenced by the size of the Pd ensembles. Finally, we used the adsorption energy of CO to assess the extent of CO poisoning on the surface of these alloys compared with Pt(111).

Density Functional Theory Calculations
All the DFT calculations in this study were performed using the Vienna Ab-initio Simulation Package (VASP) [50,51]. The projector-augmented wave (PAW) method was applied to treat electron-ion interactions, and the Perdew-Burke-Ernzerhof (PBE) exchangecorrelation functional within the generalized gradient approximation (GGA) was employed to describe the electron interactions with a cutoff energy of 500 eV [52,53]. All structures were fully relaxed until the energy and force reached the convergence thresholds of 10 −5 eV and 0.01 eV/Å. For the first Brillouin zone integration, the 25 × 25 × 25 and 5 × 5 × 1 Γ-centered k-point meshes were used at structure optimization for the bulk unit cell and slab models, respectively. For the calculation of the electronic structure of the surfaces, a k-point of 10 × 10 × 1 was used. The convergence of the results with respect to all the above parameters was carefully checked.
The simulated bulk lattice parameters obtained for Pd, Pt, and Ag were 3.94 Å, 3.97 Å and 4.15 Å, respectively, which are in good agreement with the experimental values of 3.89 Å (error: 1.3%), 3.92 Å (error: 1.3%), and 4.09 Å (error: 1.4%) [41,54]. The slabs were separated by 20 Å of vacuum in the perpendicular z-direction to avoid interactions between the two periodic units, and a dipole correction was applied. The surfaces were modeled by a 5-layer (4 × 4) supercell. The bottom three layers of each slab were fixed with the bulk lattice constant of the corresponding host metal (Ag, 4.15 Å), while the top two layers and the adsorbed species were fully relaxed. A detailed illustration of the considered structures is shown in Figure 1 for the various configurations. For the various structures studied in this work, we have used the following notation: Pd 1L Ag(111) denotes a surface alloy at the topmost layer and a Pd sublayer underneath. In the case of a surface alloy, a certain number of Ag atoms (1-3 Ag atoms) at the topmost Ag overlayer of a Pd 1L Ag(111) structure were replaced by Pd atoms. Note that, in our model ( Figure 1C), the whole second layer was completely replaced by Pd atoms to simulate the fact that Pd is highly enriched in the second layer [48]. The pure metal surfaces (Ag(111), Pd(111), and Pt(111) in Figure 1A) and modified Ag(111) surfaces that were replaced by Pd atoms only on the first layer ( Figure 1B) were also investigated to compare with the Pd 1L Ag(111) structures. The formation energies of Pd ensembles (monomer (M), dimer (D), and trimer (T)) in these structures were calculated to investigate their relative stability. Here, the ensemble formation energy per Pd atom is given by: where E PdAg , E Ag(111) , E Ag−bulk , and E Pd−bulk represent the total energies of Pd X @Ag(111) or Pd X @Pd 1L Ag(111) (X = M, D, T1, T2, T3), pure Ag(111), bulk Ag (per atom), and bulk Pd (per atom), respectively, and N Pd indicates the number of Pd atoms in the calculated surfaces. where , ( ) , , and represent the total energies of PdX@Ag(111) or PdX@Pd1LAg(111) (X = M, D ,T1, T2, T3), pure Ag(111), bulk Ag (per atom), and bulk Pd (per atom), respectively, and indicates the number of Pd atoms in the calculated surfaces. To identify the energetically most favorable configuration, various possible adsorption sites were considered, including the top (T), bridge (B), and hollow (H) sites for the adsorbates (i.e., O, OH, OH, and CO). All geometric structures were visualized by using the Atomic Simulation Environment (ASE) [55]. The charge density distribution was visualized using VESTA [56]. The Bader charges were calculated using the Bader Charge Analysis Code [57].

Oxygen Reduction Reaction
The ORR pathway usually involves four electron transfers and at least three intermediates [46,58]. The associative reaction pathway is given by: where * denotes the adsorption site for oxygen-containing intermediates. The computational hydrogen electrode (CHE) model was adopted to calculate the adsorption-free energies of the four-electron transfer step for ORR, which defined the chemical potential of a proton-electron pair ( + ) equal to half of the chemical potential of gaseous at pH = 0 in the electrolyte, 1 bar of in the gas at 298.15 K, and 0 (where is the reversible hydrogen electrode) [59]. To avoid the use of electronic energy, which is difficult to determine accurately within standard GGA-DFT [60], we reference the experimental formation energy of (4.92 eV) and DFT-calculated energies of and molecules to deduce the binding energy of each intermediate, which was used as follows: The reaction-free energy of each elementary step was calculated as follows: To identify the energetically most favorable configuration, various possible adsorption sites were considered, including the top (T), bridge (B), and hollow (H) sites for the adsorbates (i.e., O, OH, OH, and CO). All geometric structures were visualized by using the Atomic Simulation Environment (ASE) [55]. The charge density distribution was visualized using VESTA [56]. The Bader charges were calculated using the Bader Charge Analysis Code [57].

Oxygen Reduction Reaction
The ORR pathway usually involves four electron transfers and at least three intermediates [46,58]. The associative reaction pathway is given by: where * denotes the adsorption site for oxygen-containing intermediates. The computational hydrogen electrode (CHE) model was adopted to calculate the adsorption-free energies of the four-electron transfer step for ORR, which defined the chemical potential of a proton-electron pair (H + + e − ) equal to half of the chemical potential of gaseous H 2 at pH = 0 in the electrolyte, 1 bar of H 2 in the gas at 298.15 K, and 0 U RHE . (where RHE is the reversible hydrogen electrode) [59]. To avoid the use of O 2 . electronic energy, which is difficult to determine accurately within standard GGA-DFT [60], we reference the experimental formation energy of H 2 O (4.92 eV) and DFT-calculated energies of H 2 O. and H 2 molecules to deduce the binding energy of each intermediate, which was used as follows: The reaction-free energy of each elementary step was calculated as follows: where ∆E is the difference in DFT-calculated electronic energy, ∆ZPE is the zero-point energy (ZPE) change, ∆ C P dT is the enthalpy change, T is the absolute temperature (here, T = 298.15 K), and ∆S is the entropy change for the reaction. All thermodynamic data were processed with VASPKIT code [61]. The adsorption-free energies were calculated as follows: where G * , G * OOH , G * O , and G * OH are the free energies of the clean substrate and the substrate binding with * OOH, * O, and * OH, respectively. G OOH , G O , and G OH are the free energies of isolated gas molecules, which can be replaced by 2G The catalytic activities of ORR with different structures were estimated by determining the theoretical thermodynamic limiting potential with * OOH, * O, and * OH as reaction intermediates. The reaction-free energy of each elementary reaction step for electrochemical oxygen reduction can be defined as the difference between two adjacent steps, shown as follows: The theoretical thermodynamic ORR limiting potential (U L ) and overpotential (η), which can be the measure of the activity of a catalyst, are then defined from the largest ∆G among reactions (Equations (2)-(5)), as follows: For ORR, the theoretical minimum half-cell potential is 1.23 V. A higher U L value corresponds to a lower overpotential, η (1.23 V-U L ), indicating improved theoretical activity. The theoretical overpotential is able to be benchmarked by comparing it with that of Pt(111), which has an η of 0.43 V [62]. Hence, the activity of a novel catalyst is considered to be improved when it has η < 0.43 V (U L > 0.80 V). Note that η should not be compared directly with the experimentally measured overpotential, which depends on the current density [63]. Details of the adsorption energies, d-band centers, Bader charge, and atomic structures are provided in the Supplementary Materials.

Reaction Profile and Calculated Overpotentials
To clarify the stability of the constructed structures, the formation energies, E f , of Pd ensembles (monomer, dimer, and trimers) were investigated as shown in Table 1. All formation energies were negative, indicating that it is thermodynamically possible to form these ensembles [8]. Similar formation energies of Pd monomers and dimers on Ag(111) (0.01 eV) and Pd 1L Ag(111) (0.02 eV) suggest that the formation of Pd monomers and dimers should be almost equally favorable, which indicates that all kinds of ensembles could be distributed on the PdAg surfaces at high temperatures. Note that according to the number of Pd ensembles studied in previous MC simulations and experiments, the Pd monomers surrounded by Ag atoms at low temperatures are numerically more than the dimers and trimers. Therefore, the ORR overpotentials of Pd M @Ag(111) and Pd M @Pd 1L Ag(111) have a greater influence on the catalytic performance [48].
According to the Sabatier principle, as an efficient catalyst, the binding between the active site and the reactants or the final products should not be too weak or too strong [40]. The reaction processes of the ORR intermediates on different surfaces were investigated and summarized. The efficiency of catalysts is often qualitatively gauged by overpotential (η) based on thermodynamics (lower η corresponds to greater predicted activity) [62]. Table 2 shows the overpotentials of all surfaces calculated using the formula in Equations (11)- (14). The reaction on Pd M @Pd 1L Ag(111) had the lowest η (η = 0.41 V). Some surfaces had similar overpotential, such as Pd D @Ag(111) (η = 0.42 V), Pd M @Ag(111) (η = 0.46 V), and Pd D @Pd 1L Ag(111) (η = 0.47 V), especially on Pd M @Pd 1L Ag(111), where they even performed better. Due to the non-linear relationship between the adsorption energy and overpotential, as shown in Table S1, it is hard to quantitatively point out the influence of the Pd sublayer. However, it is obvious that the existence of the Pd sublayer significantly changes the activity of the surfaces. The lower overpotentials of the modified structures compared to pure Ag (0.62 V) in Table 2 are consistent with the realization that the surface contains Pd atoms surrounded by Ag atoms, as reported in the experimental literature, which maximizes the ability of the heteroatomic sites to amplify the activity of each Pd atom, thus enhancing the ORR activity [24,26]. Table 2. The reaction-free energies of the four e − ORR reactions, as well as the corresponding theoretical limiting potential and overpotential. To understand the reaction on the four best-performing surfaces in more depth, Figure 2 illustrates the ORR reaction profiles at various electrode potentials. It is obvious that all the reaction steps were downhill (negative free energy changes), implying a facile reaction at U = 0 V, and the ORR reaction intermediates can spontaneously adsorb on these structures. With the increasing electrode potential, the change of free energy became less negative. At the voltage U = 1.23 V, both the O and OH hydrogenation reactions became endothermic. Consequently, there is a highest electrode potential under which all reaction steps along the reaction decreased the free energy. This limiting potential defines the working potential of the electrocatalysts. Note that the potential-determining step (PDS) on the four best-performing surfaces lies in the reduction of O-containing species: OOH or OH. The PDS of ORR on the Pd M @Pd 1L Ag(111), Pd D @Pd 1L Ag(111), and Pd D @Ag(111) is located in the fourth elemental step (the reduction of OH to form the final production of H 2 O, Equation (5), and on the Pd M @Ag(111) surface is the reduction of the O 2 to form OOH (Equation (2)).

Active Site Model
optimal ( Δ − Δ = −0.0155 eV for PdM@Ag(111), Δ − Δ = −0.0220 eV for PdD@Ag(111), and Δ −Δ = −0.0002 eV for PdM@Pd1LAg(111)). Generally, increasing for one of the steps will decrease Δ for the other step due to the linear relations between the adsorption energies. Suitable reaction energy barriers lead to lower overpotential values and higher catalytic activity [64]. Interestingly, the adsorption of * being ratelimiting for PdM@Ag(111) reflects the fact that the surface binds the intermediates too weakly relative to the optimal ORR catalyst [40]. However, on the other hand, the lower overpotential of PdM@Pd1LAg(111) than that of PdM@Ag(111) could also be explained by the fact that the introduction of the Pd sublayer enhances the adsorption of the intermediates.   For the surfaces Pd M @Ag(111), Pd D @Ag(111), and Pd M @Pd 1L Ag(111), these two steps determined the same ORR potential, reflecting that both of them have a reactivity close to optimal (∆G 1 − ∆G 4 = −0.0155 eV for Pd M @Ag(111), ∆G 1 −∆G 4 = −0.0220 eV for Pd D @Ag(111), and ∆G 1 −∆G 4 = −0.0002 eV for Pd M @Pd 1L Ag(111)). Generally, increasing ∆G for one of the steps will decrease ∆G for the other step due to the linear relations between the adsorption energies. Suitable reaction energy barriers lead to lower overpotential values and higher catalytic activity [64]. Interestingly, the adsorption of * OOH being rate-limiting for Pd M @Ag(111) reflects the fact that the surface binds the intermediates too weakly relative to the optimal ORR catalyst [40]. However, on the other hand, the lower overpotential of Pd M @Pd 1L Ag(111) than that of Pd M @Ag(111) could also be explained by the fact that the introduction of the Pd sublayer enhances the adsorption of the intermediates. Figure 3 shows details of the most favorable adsorption geometries of the ORR intermediates ( * OOH, * O, and * OH) on the four best-performing surfaces (other surfaces are shown in Figure S2). For each intermediate, we examined its optimized adsorption site at various possible positions and then determined its lowest-energy adsorption conformation on the surface. In Table S1 and Figure S2, the calculated lowest binding energies and site preferences for various chemical species on the modified surfaces are listed. According to previous studies, the surface inhomogeneity and valence rules of adsorption sites of SAA thus exhibit unique catalytic properties, which can create new opportunities for designing catalysts with excellent performance [65]. As can be seen in Figure 3, the adsorption of intermediates mostly occurs at the bridge or hollow sites between the atoms of dissimilar  Figure 3A,C), * OOH binds at the top-bridge site between the host (Ag) and dopant (Pd) atoms. The * OOH adsorbate is tilted and therefore nearly parallel to the surface. * OOH is bound to the surface through the interaction of the non-protonated O atom with the dopant Pd atom. The protonated O atom is bound to only one host atom close to the top site. For Pd dimer surfaces, * OOH prefers to adsorb in bridge sites and is tilted at a different angle than in the case of the Pd monomer surface. To investigate the effect of the electronic structure on the ORR activity of the surface, partial density of states (PDOS) and charge density difference analysis of PdM@Pd1LAg(111) were performed, as shown in Figure 4. The d-projected density of states of Pd atoms on the PdM@Pd1LAg(111) and PdM@Ag(111) surfaces was significantly different from that on the Pd(111) surface, which has a formation of a sharp peak near the Fermi level (lower part of Figure 4A). Compared with the pure Pd surface, an obvious d-DOS shift on bimetal surfaces was observed moving close to the Fermi level, which can contribute more electron density to the metal-adsorbate and thus improve the reactivity. The peak of Pd on PdM@Pd1LAg(111) was closer to the Fermi level than that of PdM@Ag(111). After oxygen adsorption, the sharp peak of Pd on PdM@Pd1LAg(111) disappeared (upper part of Figure 4A). There was a strong hybridization and splitting in the bonding and antibonding states between the * and Pd atoms at about −0.20 and 0.41 eV, which indicates a strong interaction between them. The high reactivity exhibited by the dopant Pd atom changes the site preference of the adsorbates so that they are adsorbed in close proximity to the Pd atom. PDOS with O atom adsorption showed surface-adsorbate interactions with contributions from the Pd d-band. The O atom can draw more electrons from the Pd atom on the first layer of the catalyst surface ( Figure 4B, orange color indicates electron accumulation and blue indicates electron depletion). It can be seen that the Pd atom in the For * O adsorbed on the surface, the only energetically most favorable conformation is at the fcc site. On all surfaces except the two T1-type surfaces, * O is bound at sites sharing the dopant-host atoms. For * OH, it binds to the hollow site. On the Ag(111)based surface, * OH prefers the fcc site with one dopant atom and two host atoms, but on the Pd 1L Ag surface, it prefers the fcc site with two dopant atoms and one host atom. Although we calculated different surface aggregates, it can be concluded that adsorption always occurs on landmark structures, such as monoatomic and dimeric aggregates for the adsorption process. Except for T1-type ensemble surfaces, * OH and * O absorbed on T2-and T3-type surfaces have the same behavior as the dimer and monomer because of similar structural features. The properties of the two T1-type Pd ensembles exhibit a pure Pd(111) behavior, which is also verified again later on the adsorption energy of CO. This decouples the binding strength of these fragments and thus defines a weakly correlated scaling relationship. The binding sites of these adsorbates vary depending on the size of the Pd ensembles, resulting in the decoupling of the adsorption strength.
To investigate the effect of the electronic structure on the ORR activity of the surface, partial density of states (PDOS) and charge density difference analysis of Pd M @Pd 1L Ag(111) were performed, as shown in Figure 4. The d-projected density of states of Pd atoms on the Pd M @Pd 1L Ag(111) and Pd M @Ag(111) surfaces was significantly different from that on the Pd(111) surface, which has a formation of a sharp peak near the Fermi level (lower part of Figure 4A). Compared with the pure Pd surface, an obvious d-DOS shift on bimetal surfaces was observed moving close to the Fermi level, which can contribute more electron density to the metal-adsorbate and thus improve the reactivity. The peak of Pd on Pd M @Pd 1L Ag(111) was closer to the Fermi level than that of Pd M @Ag(111). After oxygen adsorption, the sharp peak of Pd on Pd M @Pd 1L Ag(111) disappeared (upper part of Figure 4A (111)), it contributes more to the electron density of the mixed metal-adsorbent band than the lower d-band, thus improving the reactivity. Bader charge analysis also showed that the Pd atoms of the four surface models with high activity received more electrons from host metal atoms, which also corroborates the better activity.  (PdD@Pd1LAg(111)), respectively. Since the d-band center of these dopant Pd atoms is closer to the Fermi level compared to the same type of substrate surface (Ag(111) or Pd1LAg(111)), it contributes more to the electron density of the mixed metal-adsorbent band than the lower d-band, thus improving the reactivity. Bader charge analysis also showed that the Pd atoms of the four surface models with high activity received more electrons from host metal atoms, which also corroborates the better activity. The ORR activity can be evaluated by the thermodynamic limiting potential, , which depends on the surface binding energy of the reacting oxygen adsorbate. Thus, the relationship between Δ * and the limiting potential, , of each fundamental process of ORR can be described by a volcano diagram, where the system at the peak is the best system. Based on the correlation between the binding energies of surface * and * of different metals and their role in the thermodynamic limiting step of ORR, the orange theoretical volcano shown in Figure 5 was obtained. It can be clearly found that the limiting potential increased with Δ * and then decreased. When Δ * was smaller than 0.86 eV, on the left side of the volcano, the ORR was limited by the strong The ORR activity can be evaluated by the thermodynamic limiting potential, U L , which depends on the surface binding energy of the reacting oxygen adsorbate. Thus, the relationship between ∆G * OH and the limiting potential, U L , of each fundamental process of ORR can be described by a volcano diagram, where the system at the peak is the best system. Based on the correlation between the binding energies of surface * OH and * OOH of different metals and their role in the thermodynamic limiting step of ORR, the orange theoretical volcano shown in Figure 5 was obtained. It can be clearly found that the limiting potential increased with ∆G * OH and then decreased. When ∆G * OH was smaller than 0.86 eV, on the left side of the volcano, the ORR was limited by the strong * OH adsorption at the metal center, which poisons the catalytic surface. When the * OH adsorption was weaker, in turn, the reaction behaved better, as shown in Figure 5. The species to the right of 0.86 eV on the X-axis illustrates the opposite situation, where the H 2 O molecule formed in the last hydrogenation step evolved into the RDS, i.e., the adsorption step was thermodynamically limited. In fact, only when ∆G * OH was in a moderate range was the formation energy of the intermediate species neither too strong nor too weak, thus obtaining a relatively low overpotential [46]. As predicted, Pd M @Pd 1L Ag(111) was shown to be closer to the peak of the volcano curve and can be the ideal active catalyst. In this case, the ORR can reach high catalytic performance. Unlike several existing experimental studies on the ORR reactions of PdAg clusters or thin-film alloys [24,62,66,67], here, we simulated a model of SAA with a special inner-layer structure which has been prepared by heat treatment combined with quenching in our previous experiments. As can be seen from Figure 5, all the structures had higher limiting potentials than pure Ag(111) (U L = 0.61 V), and Pd M @Pd 1L Ag(111) and Pd D @Ag(111) were higher than pure Pd(111) (U L = 0.79 V), indicating a synergistic enhancement between Pd and Ag, which can provide solution ideas for the design of bimetallic catalysts. Note that the modified surface Pd M @Pd 1L Ag(111) possesses limiting potentials beyond that of pure Pt(111) and is an excellent candidate for an ORR catalyst, where the influence of the inner layer plays an important role in the catalytic activity.

Reduced CO Poisoning of the Catalyst
The above calculations indicate that the combination of Pd ensembles on the alloy surfaces designed and prepared in our previous work (especially PdM@Pd1LAg(111)) exhibits a theoretical overpotential comparable to that of Pt(111). Pt-based catalysts are easily poisoned by CO impurity gas even at low concentrations, resulting in decreased performance and a shortened service life. Most noble metal catalysts are susceptible to poisoning by CO molecules due to their inherent strong CO chemisorption, while Pt catalysts are highly susceptible to catalytic poisoning by CO. The adsorption energy of CO on each surface was calculated, as shown in Figure 6. The CO chemisorption on PdM@Ag(111) was

Reduced CO Poisoning of the Catalyst
The above calculations indicate that the combination of Pd ensembles on the alloy surfaces designed and prepared in our previous work (especially Pd M @Pd 1L Ag(111)) exhibits a theoretical overpotential comparable to that of Pt(111). Pt-based catalysts are easily poisoned by CO impurity gas even at low concentrations, resulting in decreased performance and a shortened service life. Most noble metal catalysts are susceptible to poisoning by CO molecules due to their inherent strong CO chemisorption, while Pt catalysts are highly susceptible to catalytic poisoning by CO. The adsorption energy of CO on each surface was calculated, as shown in Figure 6. The CO chemisorption on Pd M @Ag(111) was the weakest compared to pure Pd(111) and Pt(111) surfaces. It can be seen that the adsorption energies of all monomer and dimer structures were smaller than those of the pure Pt(111) surface, indicating that these structures are more tolerant to CO than the pure Pt(111) surface. Figure 7 shows that CO prefers to adsorb at the top site of the monomer and the bridge site of the dimer, while none of the intermediates of the ORR prefer to adsorb at these two sites, which would make the effect of CO weaker. The adsorption energies of the Pd 1L Ag(111) surface were stronger than those of the Pd X @Ag(111) surfaces due to the substitution of the Ag atom in the second layer by Pd. The introduction of the Pd layer made the overall CO adsorption energies stronger by a fixed value, but still weaker than the adsorption energy on pure Pd(111) and pure Pt(111). The Pd trimer is the structure that determined the lowest adsorption energy. The adsorption energy of the Pd T1 @Pd 1L Ag(111) surface was close to that of pure Pd. Once the T1-type Pd trimers are available in the PdAg surface, they become the most favorable adsorption sites (on hollow sites). This ensemble effect is consistent with the experimental observation in [41]. Since the CO adsorption sites on the Pd monomer and dimer are at the top and bridge sites, respectively, the adsorption sites of the ORR intermediates on the monomer and dimer surfaces are * OOH at the top-bridge site (monomer) or bridge site (dimer), with * O and * OH at the hollow site. Therefore, the adsorption sites are not conflicting for the intermediates (except for the * OOH on the Pd dimer surfaces), which will lead to an increase in the catalysts' surface resistance to poisoning.
Nanomaterials 2022, 12, x 12 of 16 sites). This ensemble effect is consistent with the experimental observation in [41]. Since the CO adsorption sites on the Pd monomer and dimer are at the top and bridge sites, respectively, the adsorption sites of the ORR intermediates on the monomer and dimer surfaces are * at the top-bridge site (monomer) or bridge site (dimer), with * and * at the hollow site. Therefore, the adsorption sites are not conflicting for the intermediates (except for the * on the Pd dimer surfaces), which will lead to an increase in the catalysts' surface resistance to poisoning.

Conclusions
In this article, we systematically investigated the behavior of various Pd ensembles on the PdAg(111) surfaces (with/without a Pd subsurface layer) for ORR intermediates. The most favorable adsorption conformations of * , * , and * on each structure were elucidated. A tendency of ORR intermediates to have different adsorption sites at different ensemble sizes was found. Furthermore, using the CHE model, the free energy profiles of whole ORR pathways were estimated and analyzed. The results showed that PdM@Pd1LAg(111) had the best predicted performance, with a higher limiting potential (0.82 V) than Pt(111) (0.80 V), which could be explained by the subsurface coordination. It could also be proven by the analysis of the electronic properties. In this case, it is necessary to consider the influence of the near-surface layers when modeling the SAA catalyst processes. Another important advantage of PdAg SAA is that atomic-dispersed Pd as adsorption sites can significantly improve the resistance to CO poisoning. The best surface in terms of resistance to CO poisoning was PdM@Ag(111) with a single Pd atom. Furthermore, due to the wide variation of adsorption energy on these bimetallic surfaces, it is possible to precisely modulate the adsorption energy by adjusting the Pd surface content, combining activity with high tolerance to CO poisoning, which could constitute an exciting and ideal combination.

Conclusions
In this article, we systematically investigated the behavior of various Pd ensembles on the PdAg(111) surfaces (with/without a Pd subsurface layer) for ORR intermediates. The most favorable adsorption conformations of * OOH, * O, and * OH on each structure were elucidated. A tendency of ORR intermediates to have different adsorption sites at different ensemble sizes was found. Furthermore, using the CHE model, the free energy profiles of whole ORR pathways were estimated and analyzed. The results showed that Pd M @Pd 1L Ag(111) had the best predicted performance, with a higher limiting potential (0.82 V) than Pt(111) (0.80 V), which could be explained by the subsurface coordination. It could also be proven by the analysis of the electronic properties. In this case, it is necessary to consider the influence of the near-surface layers when modeling the SAA catalyst processes. Another important advantage of PdAg SAA is that atomic-dispersed Pd as adsorption sites can significantly improve the resistance to CO poisoning. The best surface in terms of resistance to CO poisoning was Pd M @Ag(111) with a single Pd atom. Furthermore, due to the wide variation of adsorption energy on these bimetallic surfaces, it is possible to precisely modulate the adsorption energy by adjusting the Pd surface content, combining activity with high tolerance to CO poisoning, which could constitute an exciting and ideal combination.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/nano12111802/s1, Table S1: The free energies of the adsorption of ORR intermediates in eV on all active site models; Table S2: d-band center and Bader charge of Pd atom at the first layer of all models; Figure S1: MC snapshots of the atomic structure of PdAg(111) with 50% Pd at 1200 K; Figure S2: Top view of the atomistic structures of all ORR intermediates on all considered surfaces.