Atomic-Scale Insights into Cu-Modified ZrO₂ Catalysts: The Crucial Role of Surface Clusters in Phenol Carboxylation with CO₂

Abstract

The catalytic performance of metal oxide materials is profoundly influenced by both chemical composition and surface morphology, particularly at high dopant loadings where metallic clusters can form. Here, we use density functional theory (DFT) to elucidate how copper incorporation—either as isolated dopants or as surface clusters—modulates the mechanism and activity of ZrO₂ catalysts in the direct carboxylation of phenol to para-hydroxybenzoic acid. Our results reveal that while Cu doping inhibits C–H bond activation, the presence of Cu clusters at the ZrO₂ surface dramatically lowers the barrier for C–C coupling with CO₂, owing to unique interfacial sites that facilitate substrate activation and CO₂ bending. We show that the reaction mechanism shifts from an Eley–Rideal pathway on pure ZrO₂ to a Langmuir–Hinshelwood mechanism on Cu-modified surfaces, with the rate-determining step depending on the Cu morphology. These findings demonstrate that even small amounts of metallic clusters can fundamentally alter catalytic pathways, providing actionable insights for the rational design of heterogeneous catalysts for selective aromatic carboxylation.

1. Introduction

Catalysis plays a vital role in chemistry, accelerating industrial processes and steering them toward specific products [1]. Among the wide variety of catalysts available, supported nanoparticles are often favored due to their recyclability [2]. These catalysts can be composed of various types of material—including metals, metal oxides, metal carbides, or carbon-based systems—depending on the intended application [3,4]. To further tailor their catalytic properties, they are typically doped with specific chemical elements in controlled ratios [5]. While doping is generally expected to produce homogeneous solid solutions at the macroscale, the formation of very small clusters on nanoparticle surfaces is almost unavoidable, resulting in locally biphasic systems at the microscale [6]. Due to their restricted dimensions and inherent instability, such clusters are difficult to detect with conventional characterization techniques, especially when they are constituted of a very limited number of atoms [7,8]. Nevertheless, the most compelling evidence for their presence remains indirect, inferred from system reactivity, as the catalytic properties of clusters can differ significantly from those of monophasic doped systems [8,9].

Clusters are nanometric or sub-nanometric particles, typically consisting of only a few dozen atoms, and they possess geometric properties that differ significantly from conventional nanoparticles. While nanoparticles are usually well-crystallized with defined facet orientations, clusters often have irregular shapes with a high density of corners and edges. This abundance of surface defects is a key factor behind the remarkable activity and unique catalytic properties of clusters [10,11]. However, these defective structures are highly unstable and require extensive dispersion on various supports for stabilization, which also increases their specific surface area compared to larger nanoparticles [7,8,9]. Additionally, such catalytic systems promote optimal synergy between the adsorbed phase and the support during catalysis, as the small size of clusters maximizes the proportion of interfacial sites [12]. Beyond these configurational and geometric advantages, clusters are also notable for their distinct electronic properties. Parameters such as bandgap width, band edge positions, and the location of the d-band center are all strongly influenced by cluster size, which can impact the energetics of key reaction intermediates and overall surface reactivity [13]. Owing to these characteristics, clusters are widely used in various catalytic applications, particularly for activating molecules that are typically considered chemically inert, such as CO₂ or aromatic rings [14,15].

In this manuscript, we aim to demonstrate how surface texture influences catalyst reactivity for a specific chemical composition, considering the direct carboxylation of phenol at Cu enriched ZrO₂ as a model reaction. It is important to note that this catalyst is considered here solely as a theoretical model system for illustrative purposes and has not been tested experimentally. We selected this catalyst because various metal oxides have already been reported to facilitate the carboxylation of aromatic compounds [16], and copper is recognized for its effectiveness in CO₂ activation [17]. Depending on the copper content, XPS characterization reveals different oxidation states of Cu in Cu-doped ZrO₂: Cu²⁺ is predominant at low doping levels (<5 wt%), while Cu⁰ becomes apparent at higher doping levels (5–10 wt%) [18,19]. This suggests the potential coexistence of two distinct phases on the catalyst: a monophasic metal oxide system with Cu dissolved in ZrO₂, and a mixed metal/metal oxide system featuring metallic Cu clusters deposited on the ZrO₂ surface. Building on previous theoretical work using density functional theory (DFT), which established the reaction mechanism for phenol carboxylation on pure ZrO₂ [20], we now investigate the effects of both Cu dopants and Cu clusters on the reaction mechanism and catalytic activity. In this work, we will examine both structural configurations of Zr_(1−x)Cu_xO₂ and evaluate two commonly discussed mechanisms in heterogeneous catalysis: the Langmuir–Hinshelwood (LH) and Eley–Rideal (ER) mechanisms. Finally, we will show the physical origins of the texture effect on surface activity are directly related to local geometric features of the surface, significantly modulating the stability of key reaction intermediates. To the best of our knowledge, this is the first computational and fundamental study comparing, for a specific metal element, the reactivity of metal dopants incorporated into metal oxide catalysts with that of metal clusters deposited on their surfaces, thereby unveiling reaction processes at the atomic scale.

2. Results and Discussion

In this section, we will first examine two major mechanisms commonly discussed in heterogeneous catalysis. We will then investigate the effects of doping and surface texture on surface reactivity.

2.1. The Langmuir–Hinshelwood Mechanism

The LH mechanism is a multistep process in which all reaction intermediates are strongly adsorbed on the catalyst surface [21]. For the direct carboxylation of phenol to p-HBZA, three main steps are identified (see Figure 1): (1) dissociation of the C–H bond at the para position relative to the phenol –OH group, forming an aryl radical; (2) C–C coupling between this intermediate and adsorbed CO₂, yielding the para-hydroxybenzoate (p-HBZ) anion; and (3) hydrogenation of the carboxylate intermediate at the –COO group, resulting in the final carboxylic acid, para-hydroxybenzoic acid p-HBZA.

We evaluated the LH mechanism on both monophasic Cu–ZrO₂ and biphasic Cu@ZrO₂ systems. The optimized ball-and-stick structures are represented in Figure 2 for each elementary step reactant and product, as well as the corresponding reaction energies and activation barriers. For Cu–ZrO₂, phenol is only weakly physisorbed, not strongly chemisorbed. Among the intermediates, the aryl radical binds to the Cu dopant via its carbon atom while the carboxylated products (p-HBZ and p-HBZA) interact with the catalyst through their carboxylic acid group, specifically binding their oxygen atoms to Zr surface atoms. This behavior is attributed to the lower oxidation state of Cu (+2) compared to Zr (+4), reflecting Cu’s lower oxophilicity [22]. Thermodynamic calculations indicate that only the C–C coupling step is exothermic (ΔE = −1.26 eV) and therefore thermodynamically favorable, whereas both C–H dissociation (ΔE = +0.24 eV) and O–H association (ΔE = +1.55 eV) are endothermic and thus not favored. Kinetic analysis supports these trends: the lowest activation barrier is found for the C–C coupling step (Ea = 1.23 eV), whereas the first and third steps have similar, higher barriers (Ea ≈ 1.80 eV). For the Cu@ZrO₂ system, three distinct adsorption sites are available: the Cu cluster, the ZrO₂ surface, and the Cu–ZrO₂ interface. Notably, phenol does not stabilize on the metallic cluster and remains weakly adsorbed on the ZrO₂ surface, allowing it to diffuse easily toward active sites. After C–H dissociation, the resulting aryl radical preferentially binds to the metal phase, while the dissociated hydrogen atom is co-adsorbed elsewhere on the cluster. This configuration is highly advantageous for reactivity, as CO₂ is most stably adsorbed at the base of the cluster, precisely at the Cu–ZrO₂ interface, which facilitates the C–C coupling step. The coexistence of metallic and oxide phases also promotes CO₂ activation by assisting its bending: the carbon atom binds to Cu, while the oxygen atoms attach to two Zr centers. The carboxylated products (p-HBZ and p-HBZA) preferentially adsorb on ZrO₂ near the cluster, which facilitates the final O–H association step, given that hydrogen adatoms are more stable on the metallic phase. In terms of reactivity, the initial and final steps on Cu@ZrO₂ are less favorable, with high reaction energies (ΔE = +0.27 and +1.24 eV) and significant activation barriers (Ea = 1.25 and 1.72 eV). In contrast, the C–C coupling in the middle step is strongly favored, both thermodynamically (ΔE = −0.90 eV) and kinetically (Ea = 0.99 eV).

2.2. The Eley–Rideal Mechanism

The ER mechanism involves a single elementary step, where one reactant is strongly chemisorbed on the catalyst surface and the other is only weakly physisorbed nearby [23]. The reaction proceeds through a concerted transition state, with C–H dissociation, C–C coupling, and O–H association occurring simultaneously. As a result, if the process follows the ER mechanism, phenol and CO₂ are directly converted to p-HBZA without forming any intermediates (see Figure 3).

We evaluated the ER mechanism on both monophasic Cu–ZrO₂ and biphasic Cu@ZrO₂ systems. The optimized ball-and-stick structures are represented in Figure 4 for each elementary step reactant and product, as well as the corresponding reaction energies and activation barriers. For Cu–ZrO₂, the initial state features phenol physisorbed above the catalyst and CO₂ chemisorbed nearby via one oxygen atom bonded to Zr. Initially, phenol lies parallel to the catalyst surface, but in the final state, the p-HBZA product is oriented orthogonally, with its carboxylic group strongly attached to the surface. In the Cu@ZrO₂ system, CO₂ is chemisorbed at the base of the Cu cluster, with phenol physisorbed above at the metal–metal oxide interface; the final p-HBZA product is adsorbed on the ZrO₂ surface near the cluster. DFT calculations show that for Cu–ZrO₂, the reaction energy and activation barrier are −0.16 eV and 1.84 eV, respectively. For Cu@ZrO₂, these values are +0.49 eV and 2.46 eV, indicating a less favorable process in the latter case.

2.3. Assessing Cu Doping Effects and Surface Texture Influence on Catalytic Activity

As a result, our calculations indicate that the ER mechanism is not feasible when ZrO₂ is enriched with Cu, either as a dopant or as a metallic cluster. This outcome differs from our previous study, where the ER mechanism was identified as the most likely pathway on pure ZrO₂. Indeed, on undoped ZrO₂, the process was found to be highly exothermic (ΔE = −0.33 eV) with a moderate activation barrier (Ea = 1.62 eV). However, the presence of Cu significantly raises the activation barriers, reaching up to 2.46 eV for Cu@ZrO₂. Therefore, for the Zr_(1-x)Cu_xO₂ system, only the LH mechanism should be considered. Firstly, it is important to highlight that the third LH reaction step involving O–H association is not significantly influenced by the catalyst’s chemical composition (pure or Cu-enriched ZrO₂) or by the surface texture (Cu as a dopant or as a cluster). Across all surfaces, the activation barriers remain high, between 1.70 and 1.90 eV, and the reaction energies are above +1.20 eV, indicating a strongly endothermic process. Nevertheless, although O–H association appears unfavorable on the Zr_(1-x)Cu_xO₂ surface, it should be noted that this step actually corresponds to the protonation of the carboxylate intermediate (p-HBZ) to form the carboxylic acid product (p-HBZA). This transformation can be easily achieved in aqueous solution at acidic pH [24] and, as such, should not be considered a limiting step. Consequently, the catalytic activity should be evaluated exclusively on the basis of based on C–H dissociation and C–C coupling steps, independently of O–H association. Let us now turn our attention to the first and second LH steps.

Table 1. Reaction energies and activation barriers (ΔE/Ea) in eV, corresponding to the three individual steps of the Langmuir–Hinshelwood mechanism and the single concerted step of the Eley–Rideal mechanism. Similar computational parameters and slab models with identical oxygen vacancy densities were used for both Cu-ZrO₂ and pure ZrO₂.

	Langmuir–Hinshelwood			Eley–Rideal
(ΔE/Ea) in eV	Step 1 (C–H dissociation)	Step 2 (C–C coupling)	Step 3 (O–H association)	Concerted Step
Cu-ZrO2	+0.24/1.74	−1.26/1.23	+1.55/1.88	−0.16/1.84
Cu@ZrO2	+0.27/1.25	−0.90/0.99	+1.24/1.72	+0.49/2.46
Pure ZrO2	−0.18/1.22	+0.12/3.52	+1.40/1.76	−0.33/1.62

reveals that the introduction of Cu hinders C–H bond cleavage in the phenol aromatic ring, increasing the reaction energies by about 0.40 eV compared to pure ZrO₂, and even raising the activation barrier up to 1.74 eV when Cu is incorporated as a dopant in the material structure (vs only 1.22 eV for the undoped catalyst). In contrast, the C–C coupling step shows a markedly different behavior. With the addition of Cu, C–C coupling becomes highly exothermic, and the activation barriers are dramatically reduced. The activation energy falls from 3.52 eV on pure ZrO₂ to 1.23 eV on Cu–ZrO₂, and drops even further to 0.99 eV on Cu@ZrO₂. This demonstrates that enriching ZrO₂ with Cu greatly facilitates C–C coupling, particularly in biphasic systems where metal oxide and metallic phases coexist. For greater clarity, we have also presented in Figure 5 the potential energy surface (PES) diagrams of the reaction on the three different surfaces for both LH and ER mechanisms.

These trends are generally attributed to electronic factors, particularly shifts in the d-band center that strongly depend on Cu morphology (see Section SI in the Supplementary Materials, SM) [25,26,27], thus impacting the adsorption energies of key intermediates differently when Cu is incorporated as a dopant in ZrO₂ or when Cu is deposited as a cluster on the surface. However, in this case, the geometric features of the surface also play a crucial role in determining the system’s reactivity. Because Cu is less oxophilic than Zr—exhibiting an oxidation state of +2 compared to +4 for Zr—it has low solubility in the ZrO₂ matrix, explaining why larger and more frequent metallic Cu clusters are observed experimentally at high doping ratios [18,19]. As a result, the oxygen coordination environment around Cu differs from that of Zr, increasing the bond distance between the Cu center and neighboring oxygen atoms and causing local geometric distortion (see Figure 6(a₁,a₂)). This has important implications for the C–H dissociation step, as the corresponding transition state is significantly destabilized on the Cu-doped surface. Consequently, the activation barrier is higher for Cu–ZrO₂ than for pure ZrO₂, while the phenol initial state remains only weakly physisorbed in both cases and is thus minimally affected by the catalyst. In contrast, the surface texture has a pronounced promoting effect on the C–C coupling step, especially when Cu is present as metallic clusters. In this configuration, the kinetic barrier drops dramatically from 3.52 eV on pure ZrO₂ to 0.99 eV on Cu@ZrO₂. This improvement is a direct consequence of the local biphasic nature of the catalyst. Specifically, in this topography, the CO₂ molecule is adsorbed at the base of the metallic cluster on ZrO₂, while the aryl radical is attached just above on Cu (see Figure 6(b₁,b₂)). Unlike on a flat, pure metal oxide surface where desorption is required, this arrangement allows the aryl radical to couple with CO₂ without desorbing, resulting in a much more stable transition state. Additionally, this surface configuration helps maintain the bent conformation of CO₂ by keeping its two oxygen atoms bonded to the ZrO₂ phase while the carbon atom interacts with the Cu cluster, thereby activating the CO₂ molecule. Overall, these observations highlight that both the chemical composition of the catalyst and the surface texture can alter the reaction mechanism. Specifically, while Cu shifts the preference from the ER mechanism (favored on pure ZrO₂) to the LH mechanism, the rate-determining step is C-H dissociation both at Cu-ZrO₂ and Cu@ZrO₂, assuming the final O–H association step occurs in solution.

3. Materials and Computational Details

3.1. Materials

Our computational models were constructed based on data from our previous work on phenol carboxylation over pure ZrO₂ [20], focusing on the monoclinic phase [28]. In that study, we demonstrated that the pristine material is inactive and that the presence of an oxygen vacancy is necessary to initiate the reaction process [20]. The original slab model, oriented along the (111) plane and containing an oxygen vacancy, consists of four metal oxide atomic layers with unit cell dimensions of 7.32 × 7.44 × 26.92 ų. The Cu-doped ZrO₂ system (Cu–ZrO₂) was constructed by substituting the exposed Zr atom in the slab with a Cu atom, while maintaining the same p(1 × 1) cell and four-layer thickness as the pure material (see Figure 7a). Given the low doping ratio (1:15, Cu:Zr), the cell dimensions were assumed to remain largely unaffected by the dopant. For the Cu cluster supported on ZrO₂ (Cu@ZrO₂), a Cu₂₂ cluster was generated from fcc Cu bulk cleaved along the (111) plane and adsorbed onto the original ZrO₂ (111) surface with one oxygen vacancy (see Figure 7b). Prior to deposition, the cell dimensions were doubled in the x and y directions to prevent interactions between periodic images of the Cu cluster, resulting in a cell size of 14.64 × 14.87 × 26.93 ų. Several adsorption modes and orientations were tested for the cluster, with only the most stable configuration retained. In this model, the ZrO₂ slab thickness was reduced to two metal oxide layers to save computational time. In both Cu–ZrO₂ and Cu@ZrO₂ models, all atoms in the metal oxide layers were allowed to relax during geometry optimizations. While the dimensions of the atomistic model used for Cu-ZrO₂ were already validated in our previous study on pure ZrO₂ [20], we conducted additional tests on slab thickness and cluster size, as described in Section SII.1 of the SM, demonstrating that this geometry allows adsorption energies for Cu@ZrO₂ to be converged to within 0.05 eV.

3.2. Computational Details

Spin-unpolarized calculations were performed using VASP 5.4 within the periodic DFT framework [29,30,31], employing the Perdew–Burke–Ernzerhof (PBE) exchange–correlation functional [32], as we have already demonstrated in previous work that the Hubbard correction and dispersion forces are irrelevant for this system [20]. Core and valence electron interactions were described using the projected augmented wave (PAW) method [33], with a plane-wave energy cutoff of 400 eV. For Brillouin zone sampling, Gamma-centered k-meshes of 3 × 3 × 1 and 1 × 1 × 1 points were used for Cu–ZrO₂ and Cu@ZrO₂, respectively, with both grids generated according to the Monkhorst–Pack scheme [34]. Electronic and ionic relaxations were converged to 10⁻⁶ eV and 0.05 eV/Å, respectively, with the conjugate gradient algorithm employed for optimizing intermediate states (IS). Transition states (TS) were located using the nudged elastic band (NEB) [35,36] and dimer methods [37,38]. The chosen computational parameters ensure that the adsorption energies are converged to within 0.05 eV with respect to the k-mesh density, and in the range of ~0.10 eV with respect to the plane-wave basis set (see Section SII.2 in the Supplementary Materials). Zero-point energy (ZPE) and entropic corrections were omitted, as only adsorbed phases were considered. For each reaction intermediate, extensive configurational pre-screening was performed, but only the most stable adsorption modes were included in the reactivity analysis. The structural coordinates of the various IS and TS involved in the different mechanisms are reported in the Supplementary Materials (Section SIII).

4. Conclusions

In conclusion, this paper demonstrates that both the chemical composition and surface texture of the catalyst are crucial in determining its reactivity. Indeed, in the direct carboxylation of phenol to p-HBZA over Cu-enriched ZrO₂, the way Cu is incorporated plays a decisive role: when Cu is introduced as a dopant, C–H dissociation is inhibited, while the presence of Cu as metallic clusters on the surface significantly promotes C–C coupling with CO₂. The overall reaction mechanism is thus influenced by both catalyst composition and surface topography. While the ER mechanism is favored on undoped ZrO₂, introducing Cu shifts the preference to the LH mechanism, with C–H dissociation becoming the rate-determining step. Notably, only when Cu is dispersed as metallic clusters on the surface—rather than incorporated as a dopant—does the C–H cleavage barrier become lower than the ER barrier observed for the undoped catalyst. These results highlight that the coexistence of a metallic phase with a metal oxide phase is essential—even when only tiny clusters and minimal amounts of Cu are present—to activate both CO₂ and aromatic rings, thereby enabling efficient phenol carboxylation. Although this study focused on a specific model system, we expect that the insights gained here can be extended to other metal oxide surfaces doped with various metallic elements. As such, this work provides valuable knowledge for understanding and rationalizing the catalytic properties of clusters at the atomic scale.