Linear Activation Energy-Reaction Energy Relations for LaBO3 (B = Mn, Fe, Co, Ni) Supported Single-Atom Platinum Group Metal Catalysts for CO Oxidation.

Single-atom catalysts are at the center of attention of the heterogeneous catalysis community because they exhibit unique electronic structures distinct from nanoparticulate forms, resulting in very different catalytic performance combined with increased usage of often costly transition metals. Proper selection of a support that can stably keep the metal in a high dispersion is crucial. Here, we employ spin-polarized density functional theory and microkinetics simulations to identify optimum LaBO3 (B = Mn, Fe, Co, Ni) supported catalysts dispersing platinum group metals as atoms on their surface. We identify a strong correlation between the CO adsorption energy and the d-band center of the doped metal atom. These CO adsorption strength differences are explained in terms of the electronic structure. In general, Pd-doped surfaces exhibit substantially lower activation barriers for CO2 formation than the Rh- and Pt-doped surfaces. Strong Brønsted-Evans-Polanyi correlations are found for CO oxidation on these single-atom catalysts, providing a tool to predict promising compositions. Microkinetics simulations show that Pd-doped LaCoO3 is the most active catalyst for low-temperature CO oxidation. Moderate CO adsorption strength and low reaction barriers explain the high activity of this composition. Our approach provides guidelines for the design of highly active and cost-effective perovskite supported single-atom catalysts.

Introduction

Platinum group metals dispersed on metal oxide supports are quintessential in automotive exhaust control, employing more than half of the annual production of noble metals such as Pt, Pd, and Rh. CO oxidation is a key reaction in this respect and is also regarded as an archetypical model reaction in heterogeneous catalysis.[1−3] The current generation of automotive three-way catalysts used in gasoline cars is based on Rh and Pt or Pd nanoparticles supported on alumina and ceria-zirconia. The cost of platinum group metals, their relatively low performance during cold start and their tendency to sinter at elevated temperature pose challenges in developing improved catalysts to comply with increasingly stringent environmental legislation related to exhaust gas abatement. Catalytic performance depends strongly on the surface morphology,[4] particle size,[5] and the reducibility of the support.[6] Despite their good performance, the metal atom utilization in these catalysts is rather low because only the exposed surface metal atoms are involved in catalyzing CO oxidation.

To achieve good catalytic performance in combination with the highest possible metal atom utilization efficiency, single-atom catalysts (SACs) have been recently proposed.[7−11] A key aspect of SACs is the under-coordinated nature of the single metal atom stabilized on typically metal oxide supports.[12−14] In practice, single atoms anchored to oxide supports are susceptible to sintering.[15−17] Therefore, there is a need to identify metal–support combinations that allow for stabilization of the metal in atomic form and high activity. In their pioneering work, Qiao et al. reported a stable anchored single Pt atom on Fe-oxide with excellent performance of CO oxidation.[14] A very recent study indicated that activation of the surface oxygen species of ceria-supported Pt single atom catalysts can drastically improve the CO conversion rate at low temperature.[18] Many factors need to be considered when designing highly active SACs, such as the type and oxidation state of the metal atom,[19−21] and its location on the surface.[22]

As general in heterogeneous catalysis, it would be useful if catalytic properties of candidate SACs could be estimated from simple activity descriptors by theoretical methods. An early study in this direction reported that the CO oxidation activity on transition metal surfaces and clusters correlates well with the adsorption energy of the adsorbate.[23] A recent study reported a linear relation between the transition-state energies and adsorption energy of CO and O2 reactants.[24] Many computational studies reported the utility of Bronsted–Evans–Polanyi (BEP) relations for various chemical reactions occurring on transition metal surfaces,[25−27] metal clusters,[23,28] as well as transition metal oxide surfaces.[29] However, such correlations have not been explored yet in detail for SACs.

Recently, perovskites have been receiving increased attention in the field of catalysis owing to their remarkable electronic properties, relatively high structural stability and tunability in terms of composition and, therefore, catalytic function.[30−36] Relevant to the present study, perovskite has been mentioned as a potential material to replace platinum group metals for environmental catalysis.[37−40] A recent study emphasized the role of single atoms of Pt stabilized on perovskite for catalytic CO oxidation.[41] Among perovskite materials, La-based perovskites with the general formula ABO3 with A = La exhibit a high oxygen storage capacity comparable to that of ceria,[42,43] suggesting that these compositions may have a potential to act as a support in three-way catalysts. As such, perovskite represents a tunable platform capable of reflecting catalytic properties of oxide systems such as FeO, MnO, and MgAl2O4. To support the design of such catalysts, it is useful to understand the stability and performance of single-atom metal atoms stabilized by a perovskite support.

In the present study, we systematically investigated the catalytic properties with respect to CO oxidation of a series of single metal (M) atoms (M = Rh, Pd, and Pt) supported by LaBO3 (B = Mn, Fe, Co, and Ni) using density functional theory (DFT) and microkinetics simulations. We identify a strong correlation between the CO adsorption energy and the d-band center of the doped metal atom and, for the first time, a linear dependence of the activation barrier for CO2 formation and the reaction enthalpy as a strong BEP relation for SACs. Microkinetics simulations predict that CO oxidation rates are much higher on Pd-doped surfaces than on Rh- or Pt-doped surfaces. In particular, Pd-doped LaCoO3 exhibits a promising activity for low-temperature CO oxidation. These insights provide guidance in the design of cost-effective and highly active perovskite-based catalysts for oxidation reaction.

Computational Methods

DFT Calculations

All spin-polarized DFT calculations reported here were carried out using the Vienna Ab Initio Simulation Package (VASP version 5.3.5).[44] The Perdew–Burke–Ernzerhof (PBE) functional was used to account for exchange and correlation effects.[45] A Hubbard-like term, U, describing the on-site Coulombic interactions, was introduced. We employed effective U values of 4.0, 4.0, 3.3, and 6.4 eV for Mn, Fe, Co, and Ni, respectively. These values were taken from previous computational studies, which provided a good description of the electronic structures of the corresponding LaBO3 perovskites.[46,47] The cutoff energy for the plane-wave basis was set at 400 eV. The Gaussian smearing method with a smearing width of 0.05 eV was adopted to determine partial occupancies. VdW interactions were considered via DFT-D3 corrections in all our calculations.[48]

To optimize the lattice constant of bulk LaBO3, a Pbnm symmetry unit cell was used in the calculations. We first evaluated the influence of different magnetic structures on the total energy and determined that the most stable magnetic structures for LaMnO3 and LaFeO3 are of the antiferromagnetic A- and G-type, respectively. Ferromagnetic structures were the most stable ones for bulk LaCoO3 and LaNiO3. The optimized lattice parameters are listed in Table S1. For the surface calculations, we constructed a BO2-terminated LaBO3(001) slab model, which is known to be a stable and exposed facet of this material in catalytic reactions.[49] Here, we constructed a 2 × 2 unit cell with the following sizes: 11.29 × 11.12 Å for LaMnO3, 11.29 × 11.11 Å for LaFeO3, 10.92 × 10.93 Å for LaCoO3, and 10.76 × 10.92 Å for LaNiO3. A six atom-layer slab separated by a 12 Å vacuum was used for all surface calculations. The top four atomic layers were relaxed, and the bottom two layers were fixed. The LaBO3 (001) slab model contains 24 La atoms, 24 B atoms, and 72 O atoms. A B atom in the surface of these LaBO3 supercells was replaced by a platinum group metal (Rh, Pd, or Pt). As a consequence, the dopant concentration in the perovskite lattice is 4.2% respective to the number of B atoms in the unit cell. This computational setup is adequate to avoid interactions between periodic images that would affect the results of the predicted surface catalytic properties involving the single atom dopant and small adsorbates. For the Brillouin zone integration, a Monkhorst–Pack k-point mesh of 1 × 1 × 1 was used. Geometries optimization was achieved until the residual Hellmann–Feynman forces were smaller than 0.05 eV/Å.

In order to explore the reaction mechanism, we calculated the location and total energy of transition states by the climbing-image nudged elastic band (CI-NEB) method.[50] A frequency analysis was done to make sure that each transition state has a single imaginary frequency in the direction of the reaction coordinate (Table S3). Adsorption energies of CO and O2 are computed byhere Esurf+g, Esurf, and Eg are the total energies of the adsorbed surface, the empty surface, and the corresponding gas phase species, respectively.

The d-band center (εd) was computed as[51]where ρ(ε) is the partial density of states (PDOS) of the d orbital at a particular energy ε. We determined εd by integrating ρ(ε) for the dopant from −20 to 5 eV relative to the Fermi level.

Microkinetics Simulations

Microkinetics simulations were carried out on the basis of the computed energetics for the explored CO oxidation pathways. These were employed to determine the reaction rates and the composition of the adsorbed layer. For surface reactions, the calculated activation barriers were used to estimate the forward and backward rate constant using the Eyring equationwhere k is the reaction rate constant, kb is the Boltzmann constant, and h is the Planck constant. T is the temperature (in K), and Ea is the activation barrier (in J), f* and f are the partition functions of the transition state and the ground state, respectively. Here, all vibrational partition functions were assumed to be a unity. This yields a prefactor for all surface elementary reaction steps of ∼1013 s–1.

For the adsorption process, we use the approximation that the molecule loses one of its translational degrees of freedom with respect to the gas phase. Then, the rate for molecular adsorption is given byHere, P refers to the partial pressure of the reactant in the gas phase, and A represents the surface area of the adsorption site. S and m are the sticking coefficient and the mass of the reactant, respectively. The surface area was set to the area of a square planar doped site, i.e., 7.3 × 10–20 m2. The sticking coefficients were set to unity in all the simulations.

For product desorption from the surface, we assumed that the activated complex has two translational and three rotational degrees of freedom. Accordingly, the rate for molecular desorption can be defined as[52]Here, σ is the symmetry number, θrot is the characteristic temperature for rotation, and Edes represents the desorption energy. The symmetry number of CO is 1. The symmetry numbers of O2 and CO2 are 2. The characteristic temperature of rotation for CO, O2, and CO2 are 2.73, 2.08, and 0.56 K, respectively.[53]

The details of the microkinetics simulations have been described in our previous studies and briefly mention the main aspect of the approach.[54] The rate constants of the elementary steps are used to construct the differential equations for all surface reaction intermediates. For each of the X components involved in the reaction pathway, a single differential equation is defined asHerein k refers to the elementary reaction rate constant, ν to the stoichiometric coefficient of component i in elementary reaction step k, and c indicates the concentration of component k on the reaction surface.

In order to identify the elementary steps that determine the overall reaction rate of the CO oxidation, a degree of rate control (DRC) analysis[55] was implemented. To a specific elementary step i, the DRC coefficient χ is calculated by

In eq , r is the overall reaction rate and k and K are the forward rate and the equilibrium constants for step i, respectively. The first-principles-based microkinetics simulations were performed using the in-house-developed MKMCXX program.[56] The overall CO2 formation rate reaction, steady-state coverage, and product distribution were computed as a function of temperature by integrating the ordinary differential equations with respect to time using the backward differentiation formula method.[57,58]

Results and Discussion

CO and O2 Adsorption

We first investigated the structure and relative stability of M-doped LaBO3 surfaces (LaBO3-M). To estimate their relative stability, we computed the exchange energy associated with replacing a B atom in the stoichiometric LaBO3 surface with an M atom compared to bulk B and M (Figure ). The M-doped LaMnO3, LaFeO3, LaCoO3, and LaNiO3 surfaces are denoted in the following as LaMnO3-M, LaFeO3-M, LaCoO3-M, and LaNiO3-M, respectively. From Table S2, it can be seen that the exchange energies for LaCoO3-M and LaNiO3-M are smaller than those for LaMnO3-M and LaFeO3-M. This means that the incorporation of Rh, Pd, and Pt into the surfaces of LaCoO3 and LaNiO3 is more favorable than into those of LaMnO3 and LaFeO3. Previous computational and experimental works studied the stability of transition metal in different layers of LaFeO3 and LaMnO3 and showed that the B site can be substituted by Rh, Pd, and Pt. Among these, Pd surface doping in the first layer of such perovskites surfaces is preferred.[59−62]

Figure 1

(left) Top view and (right) side view of LaBO3-M surfaces used for theoretical modeling of CO and O2 adsorption and CO oxidation (colors: sky blue, La; blue, B = Mn, Fe, Co, and Ni; red, O; turquoise, M = Rh, Pd, and Pt).

We then investigated the adsorption of CO and O2 on the LaBO3-M surfaces in order to identify structures relevant to CO oxidation. Top coordination of CO on the doped transition metal atoms is preferred in all cases. We also found that CO adsorption on the M atoms at B sites is much stronger than on the other native B sites (Table S4). For Rh, we determined the following CO adsorption energies, i.e., −2.49 eV (LaMnO3-Rh), −2.14 eV (LaFeO3-Rh), −1.87 eV (LaCoO3-Rh), and −1.70 eV (LaNiO3-Rh). Despite the fact that CO binds strongly to Rh, there is a strong influence of the reactivity of the B substituent on the adsorption strength. When the perovskite lattice contains a more reactive 3d transition metal B ion, the CO binding is stronger. Similar trends are observed for the Pt- and Rh-doped LaBO3 surfaces. On the Pd-containing surfaces, the CO binding strengths are weakest. By comparing the CO adsorption energy for the different Pd-containing models, we find that CO binds with comparable strength to LaMnO3-Pd (−1.19 eV) and LaCoO3-Pd (−1.22 eV) as to Pd-doped CeO2(111).

To understand the properties of CO adsorption on LaBO3-M, we analyze the partial density of states (PDOS) of these models. From the PDOS of CO adsorbed on LaBO3-M, we observe that the overlap between CO orbitals and d orbitals of the doped M atoms depend on the M atom (Figure and Figure S1). We find strong correlations between the CO adsorption energy and the integrated crystal orbital Hamiltonian population (ICOHP) between the M atom and the C atom (of CO) for the LaBO3-M models (Figure and Table S6). These correlations indicate that CO binds stronger to Rh and Pt than to Pd. LaMnO3-Rh exhibits the strongest CO adsorption among the LaMnO3-M models because the Rh 4d orbitals are less filled than the Pt 5d orbitals of LaMnO3-Pt, which leads to a different ICOHP. Similar to the ICOHP of LaMnO3-M, Rh doping in LaFeO3, LaCoO3, and LaNiO3 also results in more negative values than Pt doping.

Figure 2

PDOS of doped M atom (M = Pd, Pt, and Rh) and adsorbed CO on (left) LaMnO3-M and (right) LaFeO3-M. The states marked in green, red, and blue are the PDOS of adsorbed CO, the Pd atom of CO adsorbed LaBO3-Pd, and the Pd atom of LaBO3-Pd, respectively.

Figure 3

Correlation between the CO adsorption energy and the average ICOHP value of doped M atom (M = Rh, Pt, and Pd) and C atom in (a) LaMnO3-M, (b) LaFeO3-M, (c) LaCoO3-M, and (b) LaNiO3-M.

In order to understand the effect of the support on CO adsorption, we compare the PDOS as well as the d-band center of Pd for different LaBO3-Pd models. We note that the d-band stems from the electronic interaction between the doped atom and perovskite lattice, and there are similarities between our d-band model and the one from Nørskov et al.[63] To our model, the d-band center reflects the location of the d states of the doped atom relative to the Ef. Thus, it is reasonable to introduce this concept to rationalize the properties of CO adsorption on different models. In fact, the combined analysis of the PDOS and the d-band center is widely accepted to understand the relation between catalytic activity and electronic structure.[64−66] In Figure a, we can clearly observe that the Pd 4d states in LaMnO3-Pd and LaCoO3-Pd are shifted to energy positions closer to the Fermi level compared to the corresponding states in LaFeO3-Pd and LaNiO3-Pd. The location of d states closer to the Fermi level provides more empty d orbitals for CO adsorption. This is also in line with the relatively high d-band center of Pd atoms in LaMnO3 and LaCoO3 compared to LaFeO3 and LaNiO3 (Figure a and Table ). It is worth noting that the d-band center reflects the location of the d states of the doped M atom relative to the Ef. The closer the d-band center of the dopant is to the Ef, the easier electron transfer from d orbitals to CO orbitals will be and the stronger the bond between M and CO. Interestingly, we find a linear correlation between the d-band center and the CO adsorption energy for these Pd-doped LaBO3 models (Figure b). Such a linear relation is also observed for CO adsorption on Rh- and Pt-containing models (Figure a,b). The PDOS for these two models are plotted in Figure c,d. A clear variation in Rh-occupied states in different LaBO3-Rh models is evident, and the computed d-band center trend of LaMnO3-Rh > LaFeO3-Rh > LaCoO3-Rh > LaNiO3-Rh tracks the trend in CO adsorption energy. A similar trend of the PDOS and d-band center is obtained for LaBO3-Pt as follows from Figure b,d. In general, the models with their d-band center shifted to higher energies exhibit a higher CO adsorption strength.

Figure 4

(a) PDOS of adsorbed CO and doped Pd in LaBO3-Pd (B = Mn, Fe, Co, Ni). The states marked in green, red, and blue are PDOS of adsorbed CO, Pd atom of CO adsorbed LaBO3-Pd, and Pd atom of LaBO3-Pd, respectively. The blue line refers to the d-band center of the Pd atom of LaBO3-Pd. (b) A linear relation between the CO adsorption energy and the d-band center of the Pd atom on the corresponding surfaces. (c) Charge density difference between LaBO3-Pd surfaces after and before CO adsorption (charge density isosurface 0.07 e/Å3). Yellow: increase of electron density. Blue: decrease of electron density.

Table 1

Calculated d-Band Center (eV) for the Different Surface Models (M = Rh, Pd, and Pt)

model	LaMnO3-M	LaFeO3-M	LaCoO3-M	LaNiO3-M
Rh	–0.48	–0.82	–2.35	–3.07
Pd	–1.86	–3.56	–2.05	–3.15
Pt	–0.65	–2.58	–2.65	–3.26

Figure 5

Correlation between the CO adsorption energy and the d-band center of the doped atom in (a) LaBO3-Rh and (b) LaBO3-Pt, (c) PDOS of adsorbed CO and doped Rh in LaBO3-Rh (B = Mn, Fe, Co, Ni). (d) PDOS of adsorbed CO and doped Pt in LaBO3-Pt. The states marked in green, red, and blue are PDOS of adsorbed CO, the Rh (Pt) atom of CO adsorbed LaBO3-Rh (LaBO3-Pt) and the Rh (Pt) atom of LaBO3-Rh (LaBO3-Pt), respectively. Blue line refers to the d-band center of Rh (Pt) atom of LaBO3-Rh (LaBO3-Pt).

We also performed a Bader charge analysis on these models (Table ). From the Bader charges for Pd atoms, we determined the amount of electrons transferred from Pd to the support, i.e., 1.41 e (LaMnO3), 1.52 e (LaFeO3), 1.21 e (LaCoO3), and 1.90 e (LaNiO3). Thus, Pd in LaMnO3 and LaCoO3 hold more valence electrons than in LaFeO3 and LaNiO3. It should be noted that CO adsorption on Pd results in a redistribution of the charge. After CO adsorption, the amount of Pd valence electrons are decreased by 0.45, 0.15, 0.64, and 0.05 e in LaMnO3-Pd, LaFeO3-Pd, LaCoO3-Pd, and LaNiO3-Pd, respectively. This means that the Pd atoms in LaMnO3 and LaCoO3 transfer more electrons to the adsorbed CO compared to Pd atoms in LaFeO3 and LaNiO3. This strong back-donation of electrons results in a strong chemical bond between Pd and CO. Moreover, the charge density difference for CO adsorbed on LaBO3-Pd surfaces (Figure c) confirms that the Pd atoms lose more electrons in LaMnO3-Pd and LaCoO3-Pd than in LaFeO3-Pd and LaNiO3-Pd upon CO adsorption. For the charge distribution in Rh- and Pt-containing models (Table ), there are apparently less valence electrons on doped Rh or Pt than on Pd, which is consistent with the higher amount of unoccupied d states of Rh and Pt. The stronger CO adsorption on LaBO3-Rh and LaBO3-Pt can be ascribed to the enhanced donation of CO electron density into the less occupied Rh or Pt d orbitals. The increased electron donation is reflected by the accumulation of electrons between CO and Rh or Pt, as shown in Figure S2.

Table 2

Bader Charge (|e|) of M Atoms in Different Surface Modelsa

model	LaMnO3-M	LaFeO3-M	LaCoO3-M	LaNiO3-M
Rh	1.75 (1.93)	1.78 (1.94)	1.92 (2.01)	1.90 (2.16)
Pd	1.41 (1.86)	1.52 (1.67)	1.21 (1.85)	1.90 (1.95)
Pt	1.83 (2.43)	1.94 (2.42)	2.05 (2.45)	2.44 (2.58)

Values in parentheses are the corresponding charges of the M atoms after CO adsorption.

In the catalytic reaction mechanism to be discussed below, adsorbed CO reacts with a surface lattice O atom to produce CO2. This constitutes a Mars-van Krevelen (M-vK) mechanism. Molecular oxygen will then bind on the formed surface oxygen vacancy. The O2 adsorption energies on the vacancy sites of LaMnO3-Rh, LaMnO3-Pd, and LaMnO3-Pt are −1.56 eV, −1.10 eV, and −0.46 eV, respectively. The strength of O2 adsorption is strongly associated with the surface O vacancy formation energies, as follows from the calculated surface O vacancy formation energies for LaMnO3-Rh (2.02 eV), LaMnO3-Pd (1.67 eV), and LaMnO3-Pt (0.84 eV). For O2 adsorption on LaFeO3-M, we determined the following adsorption energies: −2.09 eV (LaFeO3-Rh), −1.53 eV (LaFeO3-Pd), and −2.17 eV (LaFeO3-Pt). The surface O vacancy formation energies for LaFeO3-Rh, LaFeO3-Pd, and LaFeO3-Pt are 2.42, 2.28, and 2.47 eV, respectively, in line with the trend of O2 adsorption energies. In comparison with LaMnO3-M and LaFeO3-M models, LaCoO3-M and LaNiO3-M models exhibit relatively weak O2 adsorption. This can be ascribed to the lower surface O vacancy formation energies for the latter surfaces.

CO Oxidation

We first established that coadsorption of CO and O2 on the stoichiometric LaBO3-M surfaces is unfavorable. Accordingly, a Langmuir–Hinshelwood reaction mechanism for CO oxidation is not likely for our models. As experimental work has shown that the oxygen species of perovskite participates in CO2 formation,[67] we considered CO oxidation involving a surface lattice O atom. The reaction starts with CO adsorption followed by migration of CO to allow CO2 formation. Subsequently, CO2 desorbs from the surface, resulting in an oxygen vacancy. Molecular oxygen can then adsorb on this vacancy, followed by adsorption of another CO on the M atom. This CO adsorbate migrates to the adsorbed O2 to form CO2 via a transition state in which the oxygen–oxygen bond of adsorbed O2 is further elongated. Afterward, the second CO2 molecule is spontaneously released from the surface.

We systematically examined CO oxidation on Rh-, Pd-, and Pt-doped LaBO3 systems, involving in total 12 reaction cycles for CO oxidation. The reaction pathways for the CO oxidation on these surfaces are displayed in Figure . For the M-doped LaMnO3 surfaces, the barriers of adsorbed CO reaction with a lattice O to form CO2 are 0.88 eV (Rh), 0.52 eV (Pd), and 0.81 eV (Pt). Not surprisingly, this reaction step encounters its lowest barrier on LaMnO3-Pd because of a relatively weak CO adsorption on the Pd atom, which facilitates the migration of CO, as shown in Figure a.

Figure 6

Reaction pathway of CO oxidation on LaBO3-M (B = Mn, Fe, Co, and Ni, M = Rh, Pd, and Pt).

Figure 7

Potential energy diagrams for CO oxidation on (a) LaMnO3-Rh, LaMnO3-Pd, and LaMnO3-Pt, (b) LaFeO3-Rh, LaFeO3-Pd, and LaFeO3-Pt, (c) LaCoO3-Rh, LaCoO3-Pd, and LaCoO3-Pt, and (d) LaNiO3-Rh, LaNiO3-Pd, and LaNiO3-Pt.

For LaMnO3-Rh, CO2 desorption is more difficult than for LaMnO3-Pd and LaMnO3-Pt. This can be ascribed to the relatively higher surface oxygen vacancy formation energy (Evo) for LaMnO3-Rh (2.02 eV) compared to LaMnO3-Pd (1.67 eV) and LaMnO3-Pt (0.84 eV). After CO2 desorption, a surface O vacancy is created, which is filled by O2 adsorption. Here, the most stable configuration for O2 adsorption is the one in which one of the O atoms fills the vacancy and the other is coordinating to the doped M atom, see Figure . The resulting surface O species will readily react with adsorbed CO with low barriers on LaMnO3-Pd (0.47 eV), LaMnO3-Pt (0.67 eV), and LaMnO3-Rh (0.72 eV). Subsequently, CO2 desorbs from the surface and completes the reaction cycle.

For LaFeO3-M surfaces, the adsorbed CO reacts with lattice O to form the first CO2 product molecule in the cycle with relatively high barriers on LaFeO3-Rh (1.38 eV) and LaFeO3-Pt (0.94 eV) in contrast to LaFeO3-Pd (0.48 eV). The potential energy diagram in Figure b shows that CO2 desorption on LaFeO3-Pt and LaFeO3-Rh is more difficult than on LaFeO3-Pd. This is also in line with the lower Evo values for LaFeO3-Pd (2.28 eV) than for LaFeO3-Pt (2.47 eV) and LaFeO3-Rh (2.42 eV). After adsorption of O2 and CO, the following CO2 formation step on LaFeO3-Pd only needs to overcome a barrier of 0.58 eV, substantially lower than the barrier for LaFeO3-Rh (2.33 eV) and LaFeO3-Pt (2.22 eV).

The CO oxidation potential energy diagrams for the LaCoO3-M surfaces indicate that the first CO2 formation step on LaCoO3-Pd is very favorable with an activation barrier of 0.41 eV (Figure c). This barrier is much lower than for the other dopants, i.e., LaCoO3-Rh (1.16 eV) and LaCoO3-Pt (1.01 eV). The energies of CO2 desorption on LaCoO3-Pt (0.85 eV) and LaCoO3-Rh (0.58 eV) are both higher than the value of 0.26 eV for LaCoO3-Pd. This is in line with the fact that Pd doping only slightly increases Evo by 0.05 eV, while it is increased by 0.70 and 0.81 eV upon Rh and Pt doping. The barriers for the second CO2 formation step on the LaCoO3-Pd (0.49 eV) are lower than on LaCoO3-Rh (2.02 eV) and LaCoO3-Pt (1.43 eV). The increasing barriers for Pt- and Rh-doped LaCoO3 surfaces are closely associated with their Evo. In general, a lower Evo favors the migration of O species, resulting in a lower barrier for CO2 formation.

For CO oxidation on LaNiO3-M surfaces, the activation barriers for the first CO2 formation step are 0.61, 0.54, and 0.74 eV for LaNiO3-Rh, LaNiO3-Pd, and LaNiO3-Pt, respectively (Figure d). CO2 desorption is not difficult, which is in line with the relatively low surface O vacancy formation energies for LaNiO3-Rh (1.32 eV), LaNiO3-Pd (1.60 eV), and LaNiO3-Pt (1.1 eV). The second CO2 formation step on the LaNiO3-M surface is favorable because this process is highly exothermic with very low barriers for LaNiO3-Pd (0.24 eV), LaNiO3-Pt (0.34 eV), and LaNiO3-Rh (0.52 eV).

Brønsted–Evans–Polanyi Relations

It has been reported that activation barriers of molecular dissociation on transition metal surfaces are linearly correlated with corresponding reaction energies.[68,69] Such scaling relations are rooted in Hammond’s postulate, which states that the transition state of an (organic) reaction resembles either the reactant or product state.[70] In our study, we further analyzed the relationship between these parameters for CO oxidation on LaBO3-M systems. The strong correlations identified between the activation barriers and reaction energies for CO oxidation on these models show that the BEP principle holds for CO oxidation on perovskite-supported single-atom catalysts. For the fitted dependency in Figure , all the correlation coefficients are high, which emphasizes their utility in accurately predicting activation barriers. In this regard, these relations provide a convenient approach to investigate trends in catalytic reactions on perovskite supported single-atom catalysts.

Figure 8

BEP relation for CO oxidation on LaBO3-M. The activation barrier is linearly related to the reaction energy of CO oxidation. (a) LaMnO3-M, (b) LaFeO3-M, (c) LaCoO3-M, and (d) LaNiO3-M.

We observe that the activation barriers typically decrease for more exothermic reactions (more negative reaction energies). In addition, the reaction energies for the first CO oxidation event are substantially higher than those for the second CO oxidation one on corresponding surfaces. Notably, the trends in the barrier for the first CO oxidation step are very different from those for the second CO oxidation step. Figure a,d shows that the barriers for the first step on the LaMnO3-M and LaNiO3-M surfaces are higher than for the second one. On the contrary, the LaFeO3-M and LaCoO3-M models exhibit much lower activation barriers for the first CO oxidation step than the second one (Figure b,c). The slope in the linear BEP relation is usually indicated by α and is understood as a parameter reflecting the nature of the transition state. The α value close to 1 for the second CO oxidation step on the LaFeO3-M can be interpreted in terms of a late transition state structure, resembling the final state along the reaction coordinate. For the first CO oxidation step on LaMnO3-M and LaNiO3-M, the α value is close to 0, pointing to an early transition state structure, which is akin to the initial state. We also note that there is poor correlation between the barriers and the reaction energies of CO oxidation on all the LaBO3-M surfaces. This is most likely due to the fact that the different LaBO3-M systems exhibit different electronic properties, resulting in a lack of similarity of the geometrical and theoretical structures along the reaction coordinate.

To systematically evaluate the catalytic activities of these transition metal-doped LaBO3 surfaces, we estimated the CO oxidation reaction rates by microkinetics simulations based on the computed potential energy diagrams. To this end, different models namely LaMnO3-M, LaFeO3-M, LaCoO3-M, and LaNiO3-M surfaces were considered for our simulations. The predicted CO oxidation rates are displayed as Arrhenius curves in Figure . Clearly, the LaBO3-Pd surfaces exhibit a much higher CO2 formation rate than the LaBO3-Pt and LaBO3-Rh surfaces at low temperature. This can be ascribed to the low barriers for CO oxidation on the Pd-doped surfaces.

Figure 9

Microkinetics simulations for CO oxidation on LaFeO3-M. CO2 formation rates r (in mol s–1) as a function of temperature on different models (P = 1 atm, CO/O2 ratio = 1). (a) LaMnO3-M, (b) LaFeO3-M, (c) LaCoO3-M and (d) LaNiO3-M.

By comparing the activity of CO oxidation on different Pd-doped supports, we find that the LaMnO3-Pd and LaCoO3-Pd surfaces exhibit higher reaction rates than the LaFeO3-Pd and LaNiO3-Pd at low temperature. It is not surprising that the former two are more active for CO oxidation than the latter two because the LaMnO3-Pd and LaCoO3-Pd surfaces exhibit a relatively strong CO and O2 adsorption and, henceforth, relatively low overall barriers for the surface reaction. In addition, CO oxidation on the LaMnO3-Rh and LaNiO3-Rh surfaces proceeds at higher rates than on LaMnO3-Pd and LaNiO3-Pd surfaces at intermediate temperatures. This is because CO only can favorably adsorb on doped atoms at low temperature. As the temperature increases, the LaMnO3-Rh and LaNiO3-Rh surfaces can still efficiently bind CO molecules which facilitate CO2 formation. Besides, the low overall barriers for CO oxidation on LaMnO3-Rh and LaNiO3-Rh are relevant in this respect. Therefore, these two Rh-doped surfaces exhibit good performance at intermediate temperatures.

From the above analysis, we find a preference for LaBO3-Pd surfaces for CO oxidation. We then plot the steady-state coverages of surface reaction intermediate and rate-controlling steps for the LaBO3-Pd models. Figure a shows these surface coverages as a function of temperature for LaMnO3-Pd. At low temperatures, CO* (* representing Pd adsorption sites) and O# (# representing O vacancy adsorption sites) are the predominant surface states. This observation is supported by a DRC analysis, which indicates that the reaction of adsorbed CO react with a lattice O atom is the rate-controlling step at low temperatures (Figure e). For LaFeO3-Pd, Figure b shows that the surface vacancy site is completely occupied by O2 at low temperatures, which is in line with strong O2 adsorption. The DRC analysis reflects that the overall CO conversion rate is limited by the second CO2 formation step. This is because the activation barrier for the second CO2 formation step is higher than for the first one. The composition of the surface adsorbed layer for LaCoO3-Pd is different from the one of LaFeO3-Pd. CO* remains preferentially on Pd at intermediate temperatures, and O2 covers the surface vacancy sites only at very low temperatures. This can be ascribed to the relatively weak O2 adsorption on LaCoO3-Pd. The DRC analysis shows that the second CO2 formation step controls the overall reaction rate because the activation barrier of the second CO2 formation is higher than that of the first one. For LaNiO3-Pd, the steady-state surface coverages as a function of temperature show that CO* occupies the Pd sites and O species covers the surface O vacancy sites at low temperatures. That is to say, the perovskite is predominantly present in its stoichiometric form. Similar kinetic trends are observed for the Rh- and Pt-containing LaBO3 surfaces (Figures S3 and S4).

Figure 10

(a–d) Calculated steady-state coverages and (e–h) DRC analysis for CO oxidation on LaBO3-Pd. Herein, * and # represent the surface sites on the Pd atom and O vacancy site, respectively (P = 1 atm, CO/O2 ratio = 1).

Conclusions

Activity trends of CO oxidation on La-based LaBO3 perovskites with B = Mn, Fe, Co, and Ni surface into which single atoms of precious group Rh, Pd, and Pt metals were doped were comprehensively investigated by first-principles DFT calculations and microkinetics simulations. CO adsorption on LaBO3-Pd surfaces is weaker than on corresponding LaBO3-Pt and LaBO3-Rh surfaces. This difference in adsorption strength is rationalized by the presence of more empty d-states just above the Fermi level for systems doped with Rh or Pt, resulting in a stronger interaction between CO orbitals and d-orbitals of Pt or Rh atoms. Given the preference for Pd, the influence of the composition of the perovskite support was investigated. A strong correlation is identified between CO adsorption energy and the d-band center of the M atom for the LaBO3-M surfaces. A charge analysis for LaBO3-M indicates that oxidation state of M has a strong influence on the CO adsorption strength. We find that CO oxidation is preferred on Pd-doped surfaces because of weak CO adsorption relative to Pt and Rh. For the first time, a linear BEP-type relation is found between the activation barrier for the surface oxidation reactions and the corresponding reaction energies for these perovskite-supported single-atom catalysts. We expect that these relationships can help identify the promising perovskite catalysts for environmental catalysis. Microkinetics simulations show that LaCoO3-Pd, i.e., Pd doped in LaCoO3, is the most promising system for CO oxidation, due to moderate CO adsorption strength and the lowest activation barrier for CO oxidation. Doping LaBO3 surfaces with transition metal dopants is argued to be a promising way to modify the surface electronic structure and improve catalytic performance toward CO oxidation.