CO oxidation is one of the most widely investigated reactions in heterogeneous catalysis because of its significance in both fundamental research [1] and practical applications such as elimination of a trace amount of CO from automotive exhaust [2, 3], or purification of hydrogen stream for polymer electrolyte membrane fuel cells [4, 5]. A few metal oxides or composite oxides are highly active for CO oxidation even at ambient temperature [1, 6, 7] but suffer from the low thermal stability and high susceptibility to poisons (e.g., sulfur and water) [8, 9]. Therefore, supported noble metals, such as Au, Pt and Pd, are generally used as the highly active and/or stable catalysts for CO oxidation [8, 10-13]. However, supported Rh catalysts are less regarded as good candidates for CO oxidation.

Single-atom catalysts (SACs), with isolated individual atoms dispersed on various supports, are emerging as a new frontier in heterogeneous catalysis and have gained increasing interest due to their unique catalytic properties [14-20]. SACs have demonstrated superior performance in many reactions including CO oxidation [21-23], water gas shift reaction [24-26], selective hydrogenation [27, 28] and selective oxidation [29]. Especially, supported Rh catalysts were found to be highly active for CO oxidation when dispersed in the form of subnanometer clusters or single atoms [30-33]. For example, TiO₂ supported subnanometer Rh clusters were proved to be highly active in CO oxidation even at cryogenic temperatures [31]. Moreover, Rh atoms dispersed over CeO₂ [30], phosphotungstic acid [32, 34], and γ-Al₂O₃ [33] were also considered as effective catalysts. These studies suggested that supported Rh SACs may possess the characteristics of a highly active CO oxidation catalyst.

In the previously published works, we successfully developed ZnO nanowire (denoted as ZnO-nw) supported Au, Pt and Rh SACs for various applications [35-37]. Herein, we thoroughly investigated their catalytic performance for CO oxidation and surprisingly discovered that the Rh₁/ZnO SAC possessed the highest activity. It also exhibited good stability under high working temperature and a simulated condition, making the Rh₁/ZnO SAC suitable for CO oxidation under practical conditions. DFT calculations provided the possible reaction mechanism and the nature of the observed high activity for CO oxidation.

ZnO nanowires which mainly expose {10-10} facets were used as the support to minimize the structural effects of the support surfaces. About 0.01 wt% loading of Rh, Au and Pt precursors were separately deposited on the ZnO-nw surfaces to fabricate the ZnO-nw supported SACs, denoted as Rh₁/ZnO-nw, Au₁/ZnO-nw and Pt₁/ZnO-nw, respectively [35, 37]. Increasing the Rh loading to about 0.4 wt% generated the Rh/ZnO-nw nanocluster catalyst (denoted as Rh_NC/ZnO-nw). Moreover, ZnO nanoflakes supported Rh SAC was also synthesized for comparison (denoted as Rh₁/ZnO-nf). The actual metal loadings of the above catalysts are listed in Table S1, and the detailed preparation method is provided in the Supporting Information (SI).

The representative scanning electron microscopy (SEM) images of ZnO-nw and ZnO-nf are shown in Fig. S1. The average diameter of the ZnO-nws is ~80 nm with lengths ranging 5–50 μm, while the average diameter of the ZnO-nfs is ~1.2 μm with lengths in the range of 200–350 nm. The total surface area of the {0001} facets is negligible for the ZnO-nws while it dominates for the ZnO-nfs (Figs. S1e and S1f). High-angle annular dark-field scanning transmission electron microscopy (HADDF-STEM) images of various catalysts are presented in Fig. 1. As shown in Figs. 1a and S2a, Rh clusters with an average diameter of ~1.1 nm were deposited on the surfaces of ZnO-nws (Fig. S2b). When the actual Rh loading was decreased to 0.007 wt%, no Rh clusters were found in the low-magnification STEM images (Fig. S3a) but isolated Rh atoms became observable (Figs. 1b and S3b). The Au₁/ZnO-nw and Pt₁/ZnO-nw SACs were also successfully synthesized with only Au or Pt atomic species anchored onto the surfaces of the ZnO-nw support (Figs. 1c, 1d, S4 and S5), similar to those reported previously [35].

CO oxidation reaction was performed under the CO/O₂ gas flow with a space velocity (SV) of 40 000 mL g_cat⁻¹ h⁻¹. The light-off curves of ZnO-nw and the corresponding SACs are shown in Fig. 2. The CO conversion of ZnO-nws alone was only 40% at 400 ℃, which is much lower than that of iron oxide or other highly reducible oxides [38]. The Au₁/ZnO-nw showed much higher activity than the bare ZnO-nws, suggesting that Au single atoms are active for CO oxidation. The activity of the Au₁/ZnO-nw was much lower than that of other Au SACs [39-41] owing to the low redox capability of the ZnO-nws [42]. Even though the Pt₁/ZnO-nw was more reactive than the Au₁/ZnO-nw, the Rh₁/ZnO-nw exhibited the best performance among these SACs because CO oxidation started at 160 ℃. Moreover, the T₁₀₀ (temperature for total CO conversion) of the Rh₁/ZnO-nw was 210 ℃, much lower than that of the Pt₁/ZnO-nw (300 ℃) and Au₁/ZnO-nw (400 ℃). Although CO was fully eliminated at ~160 ℃ on the Rh nanocluster catalyst (Fig. S6), isolated Rh atoms still exhibited superior activity considering the fact that the Rh usage of the nanocluster catalyst is ~60 times higher than that of the Rh₁ SAC. The CO conversion curve of the Rh₁/ZnO-nf almost coincides with that of the Rh₁/ZnO-nw (Fig. S6), suggesting that the high activity mainly originates from the singly dispersed Rh atoms, regardless of the facets of the ZnO support. In the three-way-catalysts (TWCs), NO_x and CO are eliminated by Rh and Pt/Pd components, respectively [2]. The capability of the Rh₁/ZnO SAC for low temperature CO conversion makes it a promising candidate to replace Pt/Pd in practical TWCs.

The specific rate and turnover frequency (TOF) of the Rh- and Pt-based catalysts in this work were compared with those reported in the literature (Table 1). The TOF of the Rh₁/ZnO-nw at 180 ℃ was 0.63 s^–1, higher than those of Rh_NC/ZnO-nw, Pt/θ-Al₂O₃ SAC [43], and Pt/zeolite SAC [44], demonstrating the good performance of the Rh₁/ZnO-nw SAC. However, the Rh/SBA-15 nanoparticle catalyst with a TOF of 1.69 s^–1 showed higher reactivity [45]. The E_a of the Rh₁/ZnO-nw and Rh_NC/ZnO-nw was calculated to be 85.6 and 80.5 kJ/mol, respectively (Fig. S7), much lower than those of the reported Rh nanocatalysts such as Rh/SiO₂ (104 kJ/mol) [46] and Rh/SBA-15 (124.8 kJ/mol) [45]. The lower activation barrier of the Rh/ZnO-nw catalysts might be the main cause for their better CO oxidation performance. To obtain deeper insights into the CO oxidation reaction mechanisms of the ZnO-nw supported noble metal SACs, we performed the density functional theory (DFT) calculations to understand the underlying factors governing the different activities of ZnO-nw supported Rh₁, Pt₁, and Au₁ SACs.

According to our previous results [35], the single Rh, Pt, and Au atoms were prone to substitute in a surface lattice Zn position on ZnO{10-10} surface since these single metal atoms can be stabilized by the lattice oxygen (o_o^x), resulting in partially positively charged Rh, Pt, and Au. The most stable structures of the intermediates involved in CO oxidation on the Rh₁/ZnO, Pt₁/ZnO, and Au₁/ZnO are shown in Fig. S8 and the corresponding binding energies are listed in Table S2. We found that CO tends to adsorb on top of these single noble metal atoms; the single Au atom exhibits a lower CO binding strength than that on the single Pt and Rh atom. For O₂ adsorption, the two O atoms prefer to respectively adsorb on the single metal atom and the adjacent surface Zn atom. The bridge site, composed of one single metal atom and the neighboring surface Zn atom, is most favorable for atomic O adsorption. It was found that the Rh₁/ZnO shows the strongest binding for O₂ and O while the Au₁/ZnO shows the weakest. Table S2 indicates that these single noble metal atoms exhibit significantly stronger binding for all the intermediates than that of the bare ZnO{10-10}, suggesting that the single noble metal atoms are the important active sites for CO oxidation. Moreover, the stronger binding of CO than O₂ on these single noble metal atoms implies that the Eley-Rideal mechanism is not favorable for CO oxidation on these systems because the active site is prone to be occupied by CO. On the other hand, the high barriers of O₂ dissociation on the Au₁/ZnO (2.49 eV), Pt₁/ZnO (1.49 eV), and Rh₁/ZnO (0.97 eV) make CO oxidation via the Langmuir-Hinshelwood mechanism challenging (Table S3). Thus, the alternative Mars-van Krevelen mechanism was investigated. Such a process involves CO reaction with a surface lattice O to generate a surface O vacancy (V_O) first, followed by the dissociation of O₂ at the V_O site and the subsequent formation of an O adatom on the surface that can be removed by another CO.

The Gibbs free energy diagrams for CO oxidation via the Mars-van Krevelen mechanism are shown in Fig. 3 and the calculated barriers and reaction energies of the elementary steps are listed in Table S3. The entropic contribution of the gas-phase species in the reaction energy at 250 ℃ at standard pressure was considered in this calculation [48]. The energy losses (exothermic process) for the gas-phase CO and O₂ adsorption onto the surface were determined to be 0.88 and 0.89 eV, respectively, while the desorption of the adsorbed CO₂ into the gas phase would gain an energy of 0.91 eV (endothermic process). The calculated barrier to generate surface V_O, via CO (adsorbed onto the noble metal atom) reaction with a surface lattice O (of ZnO {10-10}), follows Au₁/ZnO (0.07 eV) < Pt₁/ZnO (0.55 eV) < Rh₁/ZnO (0.77 eV) and an opposite trend was observed for their exothermicity (Fig. 3a). The factors that determine these trends are mainly affected by the bonding strength between the lattice O and the single noble metal atom, which can be evaluated by the surface V_O formation energy. The calculated surface V_O formation energies with respect to an O of the gas phase O₂ are 0.70, 1.60, and 2.28 eV on Au₁/ZnO, Pt₁/ZnO, and Rh₁/ZnO, respectively, indicating that the Au–O bond exhibits the weakest bonding strength while the Rh–O the strongest. The weakest bonding strength implies the most favorable removal of lattice O by CO, resulting in the highest exothermicity and lowest barrier for this reaction. In addition, the weakest Au–O bonding strength gives rise to the weakest binding energy of –0.58 eV for O₂ adsorption on the surface V_O (Table S2), resulting in the highest barrier (1.03 eV) for O₂ dissociation at this surface V_O site of the Au₁/ZnO (Fig. 3b). Among these systems, Rh₁/ZnO showed the lowest barrier (0.83 eV) and the highest exothermicity (–1.50 eV) for O₂ dissociation at the V_O site owing to the strongest Rh–O bond. Upon dissociative adsorption of the gas phase O₂ at the V_O site, one O atom fills in the oxygen vacancy on the ZnO {10-10} surface while the other O adsorbs as an adatom, readily to react with another CO molecule. The CO reaction with an O adatom is an exothermic process on all these systems and the Au₁/ZnO showed the highest exothermicity and lowest barrier (0.25 eV) because of the weakest binding of the O adatom (Table S2).

In the catalytic cycle for CO oxidation via the Mars-van Krevelen mechanism, we found that the O₂ dissociation on a surface V_O is the rate-limiting step for all the SAC systems we investigated. For this specific step, the calculated barriers were 0.83, 0.93, and 1.03 eV on the Rh₁/ZnO, Pt₁/ZnO, and Au₁/ZnO, respectively, suggesting that the activity of CO oxidation is in the order Rh₁/ZnO > Pt₁/ZnO > Au₁/ZnO, consistent with our experimental results. Furthermore, the barrier of 0.83 eV on the Rh₁/ZnO is comparable with the experimental value (Fig. S7).

The sintering resistance is vital for practical car-exhaust catalysts because they usually experience high temperatures (~750 ℃) [49]. On the other hand, the catalysts can also be deactivated by impurities in the exhausting gas [50-52]. Therefore, as shown in Fig. 4a, the 100-h stability test of the Rh₁/ZnO-nw SAC was carried out at 300 ℃ under standard conditions (1 vol% CO, 1 vol% O₂ and He balance) as well as the simulated car-exhaust gas mixture (1.6 vol% CO, 1 vol% O₂, 0.01 vol% propylene, 0.0087 vol% toluene, 10 vol% H₂O and He balance). The CO conversion was maintained for 100 h without any deactivation, demonstrating the excellent stability of the Rh₁/ZnO-nw SAC even in the presence of water and hydrocarbon contaminants. The high temperature stability was further tested at 800 ℃ (Fig. 4b). However, during the second cycle, the activity of the catalyst obviously decreased, probably due to the aggregation or oxidation of the isolated Rh species. Regeneration of the catalyst by in situ reduction at 200 ℃ in 10 vol% H₂/He, however, almost completely recovered the activity. To confirm the state of the Rh species after the first reaction run, AC-HAADF-STEM imaging and EDS mapping were carried out (Fig. S9). These results show that most of the Rh single atoms were stable and remained isolated although a small amount of small Rh nanoclusters were also observed. When we consider the fact that the activity can almost be restored by a mild reduction treatment, we can deduce that the Rh₁/ZnO-nw SAC was stable and the observed deactivation mainly originated from the oxidation of the Rh atoms during the CO oxidation reaction.

In summary, we have successfully fabricated ZnO-nw supported Rh, Au, and Pt SACs, catalysts among which the Rh₁/ZnO-nw SAC showed the highest activity for CO oxidation. DFT calculations revealed that the reaction probably undergoes via Mars-van Krevelen mechanism and the O₂ dissociation at the surface oxygen vacancy sites is the rate-determining step. Among the three noble metals investigated, the Rh₁/ZnO-nw SAC possesses the lowest barrier for the rate-limiting step and thus provides the highest CO oxidation activity. Furthermore, the resistance of Rh₁/ZnO-nw SAC to poisoning and sintering at high temperatures makes it a potential candidate for practical applications.

This work was supported by the National Natural Science Foundation of China (21606222, 21776270) and Liaoning Revitalization Talents Program (XLYC1807068). J. Xu and J. Liu were supported by the US National Science Foundation under CHE-1465057. B. Qiao thanks the support of DNL Cooperation Fund, CAS (180403).