Single-atom catalysts (SACs), a new frontier in heterogeneous catalysis, have attracted intensive research interest in the past few years [1-11]. The single-atom dispersion of an active metal component allows maximal utilization efficiency of the metal atom, which is of particular importance for expensive and scarce platinum group metals (PGMs). Furthermore, the uniform structure of the active sites in SACs offers unrivalled selectivity for a variety of chemoselective transformations [12-15], which would be difficult to access using nanoparticle counterparts. With these unique advantages, SACs are expected to find wide application in heterogeneous catalysis, and even to act as a bridge between homogeneous and heterogeneous catalysis [3, 11, 16-19].

One of the challenges in practical application of SACs is the stability of single atoms. Although some SACs (e.g., Pt/CeO₂ [20-21], Pt/Fe₂O₃ [22], etc.) are reported to be highly stable under an oxidizing atmosphere, they barely survive under a reducing atmosphere, even at mild temperatures, due to the severe aggregation of single atoms, which greatly limits the application of SACs in hydrogen-involved reactions, such as hydrogenation, dehydrogenation, and hydrogenolysis. To fabricate stable SACs under hydrogen-involved reaction conditions, a fundamental understanding of the mechanism responsible for the stabilization of single atoms on the support is very important. In fact, for reducible oxide-supported single atoms, both the metal single atoms and the oxide supports are subject to reduction in the presence of hydrogen [12, 23]. In this case, the oxide support forms non-stoichiometric sub-oxides with abundant oxygen vacancies. Therefore, the stability of single atoms on the nonstoichiometric sub-oxides will, to a certain extent, determine the stability of the SACs under a reducing atmosphere.

The activation of hydrogen is another important issue because it is associated with the application of SACs in hydrogen-involved reactions. It has been well established that the homolytic dissociation of H₂ can occur spontaneously to yield metal-H species on the multi-center ensembles of metal surfaces or nanoparticles [24-25]. However, such a process may be limited, due to the absence of contiguous metal atoms in SACs. In fact, when metal single atoms are bonded with the oxygen or nitrogen atoms of the supports, they are significantly positively charged, which provides an alternative pathway for hydrogen dissociation-heterolytic dissociation [26, 27]. For example, when Pd single atoms were anchored to a C₃N₄ support, heterolytic dissociation of H₂ occurred, with one hydrogen atom bound to an N atom, and the other to a Pd atom [26]. In the Pd₁/TiO₂-EG single-atom catalyst, hydrogen was also heterolytically dissociated, leading to much higher hydrogenation activity in the unsaturated polar group than in the nonpolar group [27]. However, for single-atom alloy (SAA) catalysts such as Pd-Cu, hydrogen can be homolytically dissociated from the Pd monomers even at very low temperatures [28]. These results suggest that the activation mode of hydrogen on single atoms depends on the local coordination environment, just like the interplay between ligands and metal centers in homogeneous organometallic chemistry [3, 11, 19].

Pt/WO_x (2 < x < 3) is a typical strong metal support interaction (SMSI) system [29] and has been proven highly active and selective in a variety of hydrogen-involved reactions [23, 30-32]. For the glycerol hydrogenolysis reaction in particular, Pt/WO_x catalysts showed unique chemoselectivity to 1, 3-propanediol (1, 3-PDO) [23, 33, 34]. In our previous experimental work [23], we compared Pt/WO_x and Pt/WO₃ and found that Pt single atoms can be stabilized on WO_x possessing abundant oxygen defects, but shows significant aggregation on well-crystallized WO₃. Moreover, hydrogen dissociation on the Pt/WO_x SAC formed in situ Brønsted acid, which was the key to the selective hydrogenolysis of glycerol to 1, 3-PDO. However, an atomic level understanding of the mechanism is still lacking, such as how and where Pt single atoms are stabilized on the WO_x support, how hydrogen molecules are activated, and how Brønsted acids are formed.

In this contribution, by using density function theory (DFT) calculations we performed a comparative study of WO_x-supported Pt, Pd, and Au to address the above fundamental issues. Our DFT calculation results indicated that the oxygen defects in the WO_x play a vital role in the stabilization of Pt single atoms, and the SMSI on the defective site is favorable to maintaining the metallic nature of Pt single atoms. Consequently, unlike the positively charged single atoms, homolytic dissociation of H₂ is favored on metallic Pt single atoms. Meanwhile, the diffusion of H species from the metal active site to the surface oxygen of the support is also greatly favored, resulting in the formation of Brønsted acids on the WO_x. Consistent with the DFT calculations, we experimentally synthesized WO_x-supported Pt, Pd, and Au catalysts, which showed that only Pt could be atomically dispersed on the WO_x, while Pd formed small nanoparticles and Au formed large particles.

All calculations were performed with the Vienna ab initio simulation package (VASP) [35, 36]. The generalized gradient approximation (GGA) parameterized by a Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional was employed, and core electrons were represented by the projector-augmented wave (PAW) method, with an energy cutoff for the plane waves of 430 eV. The geometric optimization of surfaces was performed using a 2 × 2 × 1 Monkhorst-Pack mesh for the Κ-points sampling. The self-consistent iterations were converged within a criterion of less than 1 x 10^‒5 eV, and ionic optimization steps were converged at less than 0.02 eV/Å. To locate saddle points and calculate the reaction pathways, we used the climbing image nudged elastic band (CI-NEB) method to yield a minimum energy path. Atomic charge analysis was performed by Bader charge analysis [30, 35].

With the optimized unit cell (Table S1), the WO₃ surface was built by cleaving the (001) surface on the O plane from a monoclinic or orthorhombic WO₃ supercell due to its comparable thermodynamic stability in experiments [36]. The O-terminated surface was used as the model of the catalyst surface, on which the oxidation/reduction was first performed by ab initio thermodynamics analysis under different oxygen chemical potentials, i.e., O-poor or O-rich conditions. As such, the WO₃(001) surface was modeled by a four-layer slab, with the two bottom atomic layers frozen in the bulk configuration, and the top two atomic layers allowed to relax. A p (2×2) supercell with a 15 Å vacuum region between adjacent slabs was used to ensure that none of the adsorbate interacted with its periodic image [30, 35].

WO_x was prepared using a modified procedure according to previous reports [37]. In detail, 3 g WCl₆ was dissolved into 100 mL ethanol while the mixture was stirred, and the solution was then transferred to a Teflon autoclave for heating in an oven at 433 K for 36 h. After cooling to room temperature, the dark blue precipitate was collected by filtration, washed with ethanol and water, and dried at 323 K for 6 h in a vacuum to obtain the WO_x support. Pt/WO_x was obtained by impregnation of WO_x with H₂PtCl₆ solution. After it was dried at 323 K for 6 h in a vacuum and reduced in flowing H₂ at 573 K for 1 h, the obtained Pt/WO_x catalyst was passivated using 1% O₂/N₂ for 4 h at room temperature before exposure to air. Pd/WO_x and Au/WO_x catalysts were prepared using a similar procedure.

X-ray powder diffraction (XRD) analysis was carried out on a Panalytical X'pert diffractometer using nickel-filtered Cu K_α radiation with scanning angle (2θ) in the range 10°-80°, operated at 40 kV and 40 mA. High resolution transmission electron microscopy (HRTEM) images were obtained with a Tecnai G² Spirit (FEI) microscope operating at 200 kV. Prior to observations, the powder sample was ultrasonically dispersed in ethanol, and a few droplets of the suspension were put on copper grids covered with a holey carbon film and dried at room temperature. Hydrogen chemisorption was conducted on a Micromeritics AutoChem Ⅱ 2920 chemisorber. Prior to H₂ adsorption at 323 K, the samples were degassed in Ar flow at 583 K for 30 min and reduced in H₂ flow at 573 K for 1 h.

The activity measurements for the hydrogenolysis of glycerol was performed at 413 K and 1 MPa H₂ for 12 h in a 75-mL autoclave with Teflon lining, as we reported earlier [34]. Prior to the reaction, the autoclave was flushed with H₂ several times, and then charged with 0.3 g of the catalyst and 12 g of an aqueous solution of 5 wt% glycerol. The liquid products were collected and analyzed with a GC equipped with an HP-INNO WAX capillary column (30 m × 0.32 mm × 0.5 μm, FID detector) using n-butanol as an internal standard.

To model the possible exposed surface of WO₃ under a hydrogenolysis environment, i.e., the O-poor reaction condition, we first constructed a phase diagram of the W-terminated (001) surfaces (Figure 1a; denoted as WO₃(001)) with different coverages of oxygen adatoms or vacancies [38]. Two types of crystal-phase tungsten oxide, including orthogonal and monoclinic phases, were considered in our research, and these represent the two most common WO₃ phases observed under experimental conditions. Taking the orthogonal WO₃(001) surface as a reference, as seen in Figure 2, the oxidation of WO₃(001) through the formation of a terminal O (O_1t) on the (001) surface is thermodynamically unfavorable (Table S2 and Figure S2). For example, the formation energy of an oxygen adatom is endothermic, with a value as high as 3.94 eV in O-poor conditions. With full coverage of oxygen adatoms on the surface to form an O-terminated WO₃(001) surface, the endothermicity shows a gradual increase, with an exothermic energy of 5.34 eV. On the other hand, the further removal of a bridged oxygen (O_2b) atom on WO₃(001) to form a vacancy on the WO₃ surface is also not facile. The formation energy is endothermic with an energy of 0.60 eV in the O-poor condition, indicating the difficulty of deep reduction through the removal of bridging O atoms on the WO₃ surface. The surface formed by removing bridging O atoms is referred to as WO_x(001) in the text (Figure 1b). This suggests that the terminal oxygen atoms on the O-terminated (001) surface of WO₃ are facile to remove to generate Lewis active sites under O-poor conditions, and the (001) surface of WO₃ without the terminal O is exposed in an O-poor, hydrogenation, or hydrogenolysis environment. In the case of the monoclinic WO₃(001) surface, although it tended to be oxidized under O-rich conditions by forming oxygen adatoms on the surface, a similar preference to expose the W-terminated surface under O-poor conditions remained according to our thermodynamic calculations. Accordingly, the orthogonal WO₃ was used as the model surface in this research to obtain insight into the stabilization mechanism of single-atom active sites and its reactivity toward H₂ activation.

Subsequently, various sites on the WO₃(001) and WO_x(001) surfaces are considered for locating Pt single atoms through optimization with density functional theory (DFT) calculations (Figure S4). The most favorable geometric structures for Pt single atoms on the surface are schematically shown in Figure 3. As we can see, the O_2c site is the most favorable site for Pt location on the WO₃(001) surface. The binding energy of the Pt single atom on the support, calculated as the energy difference between the Pt₁/oxide surface slab and the sum of energies of the free WO₃(001) surface and the isolated Pt atom, is only -2.25 eV with a Pt-W bond length of 2.72 Å (Table 1), implying a relatively weak chemical interaction between the Pt single atom and the WO₃(001) surface. It is noted, however, that the presence of the Pt single atom on the WO₃(001) surface greatly promotes the reduction of the surface, i.e., the formation of a bridged oxygen vacancy on the WO₃(001) surface is exothermic, with an energy of 1.51 eV in the O-poor condition. In this regard, the presence of Pt single atoms on the WO₃ support leads to the reduction of O_2c site surface oxygen atoms, which can be ascribed to the much stronger interaction of Pt single atoms with the WO_x(001) surface than with the WO₃(001) surface. As shown in Table 1, the binding energy of Pt is calculated to be as high as -4.36 eV on the WO_x(001) surface, which is nearly twice that of the WO₃(001) surface. Geometrically, the Pt single atom prefers to be trapped in an oxygen vacancy by a slight deformation, and a great decrease in the Pt-W bond length to 2.45 Å was also observed. It was found that a Pt atom embedded in an oxygen vacancy was thermodynamically more favorable by 0.29 eV than a surface atom, and the diffusion barrier to the surface was 1.45 eV. In other words, a surface Pt atom was also unlikely to migrate to the subsurface (with a barrier of 1.16 eV, Figure S5). In addition, only surface Pt single atoms can be involved in the chemisorption of H₂ by an exothermic chemisorption energy as high as -1.37 eV. Therefore, surface Pt single atoms were thought to be predominant under reaction conditions.

For comparison, Pd and Au single atoms on both the WO_x(001) and WO₃(001) surfaces were also calculated. Similar to the case of Pt single atoms, oxygen vacancies on the WO_x(001) surface showed much stronger binding energies than the O_2c on the WO₃(001) surface towards both Pd and Au single atoms (-2.34 eV vs. -1.32 eV for Pd, -2.38 eV vs. -0.55 eV for Au), with a corresponding decrease in the Pd-W bond length from 2.77 to 2.59 Å, and in the Au-W bond length from 3.33 to 2.70 Å. Clearly, the generation of oxygen vacancies on the WO₃ support surface is crucial for stabilizing metal single atoms. Such a stabilization effect by oxygen vacancies was also reported previously on other supports for SACs, such as FeO_x and MnO_x [39]. However, the binding energy of the three different metal single atoms decreased in the order of Pt > Pd > Au, suggesting that Pt single atoms possess higher stability on the WO_x support than Au or Pd. It was also noted that both Pd and Pt single atoms adopted a similar geometry when they anchored on the WO_x(001) and WO₃(001) surfaces. That is, Pt was located on the O_2c site (a Pt-O distance of 1.96 Å) with neighboring Pt-W bond lengths of 2.72 Å and 3.36 Å on WO₃(001), and preferred to be trapped in the hollow site near the oxygen vacancy on the WO_x(001) surface by short Pt-W distances (2.46 Å) and elongated Pt-O distances (2.38 Å). Such bonding geometry was also found for Pd on the WO₃ surface, but with increased Pd-W and Pd-O bond lengths due to their relatively weak interactions. In contrast, Au single atoms formed a different bonding geometry (Figure 3), by anchoring on top of the O_2c site of WO₃(001) and the bridge site of two neighboring W atoms on the WO_x(001) surface. This difference might have been caused by the different bonding natures of Pt, Pd, and Au single atoms, in which d-electron bonding is involved for Pt and Pd atoms, while s-electron bonding predominates for Au single atoms. In addition, the distinct metal/support interactions of different metals on a support also result in different charge states of metal single atoms, especially for the prominent anchoring of Pt single atoms on the WO_x support. As indicated by Bader charge analysis (Table 1), electron transfer occurs from oxygen vacancy to metal atoms, which is suggested by the enriched charges on Pt, Pd, and Au with the presence of oxygen vacancies on the WO_x support. In particular, due to the strong metal/support interaction, a negative charge as high as 0.32 |e| is found on the Pt atoms, which differs from the Pd and Au atoms on the WO_x surface. The charge difference between Pt and Pd may originate from the different electron affinity of these two metal atoms. According to previous studies, Pt single atoms possesses a much higher electron affinity than Pd [39], which therefore leads to a different electron transfer between metal single atoms and WO_x. Due to the formation of oxygen vacancies on WO_x(001), residual electrons accumulate in the hollow sites neighboring the oxygen vacancies. This was confirmed by the differential electron density analysis on WO_x(001) (Figure 4a). As a result, a charge transfer to the Pt single atoms of Pt₁/WO_x was discovered, consistent with the previous report of single metal atoms on defective oxides [39]. However, the differential charge density analysis indicates that Pt single atoms are chemically bonded by an electron-enriched hollow site (Figure 4b). Meanwhile, the density of states (DOS) of Pt-5d for Pt₁/WO_x and the Pt(111) metal surface are shown in Figure 5. As we can see, the d-band density of Pt single atoms has a distribution similar to that on the Pt(111) surface, indicative of the similar electronic properties of these two different Pt active sites. Moreover, we also calculated the d-band center of the d-state, which was located at -2.01 and -2.37 eV for Pt₁/WO_x and Pt(111), respectively. This negative charge implies that Pt single atoms on the WO_x support can retain the metallic nature of Pt, which is quite different from most supported SACs, in which metal single atoms are usually positively charged [40-46].

Hydrogen activation is a key step in many hydrogenation and hydrogenolysis reactions. Particularly, when the active metal component is atomically dispersed, hydrogen activation may become a rate-determining step due to the absence of contiguous metal atoms [47]. Based on the above calculations, Pt single atoms are stabilized well on the oxygen vacancies of the WO_x(001) surface, and maintain the metallic nature of Pt, whereas Pd and Au are less stable on the same sites. We then studied the dissociation of H₂ on the Pt₁/WO_x(001) surface. Our DFT calculation indicated that as it moves closer to the Pt single atom, the gaseous H₂ molecule dissociates spontaneously into two hydrogen atoms with an exothermic energy of 1.37 eV by forming separated metal-H species on the Pt₁ active sites (Figure 6). Moreover, the two H atoms linked to the Pt metal site possess identical Pt-H bond lengths (1.57 Å) and charges (-0.10 |e|), similar to the H adatom on the Pt(111) surface. Clearly, the hydrogen molecule can proceed with homolytic dissociation on the Pt₁/WO_x(001), just like on the Pt(111) surface (1.56 Å). This result is in contrast to other SACs, for which heterolytic dissociation of hydrogen is preferred [26, 27]. The facile dissociation of hydrogen on the Pt₁/WO_x(001) can be attributed to the negative charge of Pt single atoms due to the electron transfer from oxygen vacancies to the Pt single atoms. Furthermore, the low-coordination metal single atoms also offer the opportunity for multi-coordination toward adsorbates; that is, in the Pt₁/WO_x system, the Pt single atom active site can capture more than two metal-H species. As seen in Figure 7, our DFT calculation indicates that the Pt₁ single atom on WO_x can capture as many as three metal-H species, although the binding energy per metal-H shows a great decrease with the increase in metal-H bonds (from 1.25 eV to 1.37 eV and then to 0.87 eV). In this respect, the single atom catalyst behaves much like a homogeneous catalyst due to its tunable coordination environment on the support.

However, due to the maximized interface between metal single atoms and supports in SACs, the as-formed metal-H species on Pt single atoms is straightforward to transfer to the support via interface diffusion. As seen in Figure 8, the calculated energy barrier for metal-H diffusion to its neighboring O_2c is only 0.82 eV, with a slightly endothermic energy of 0.37 eV. This implies the facile spillover of H during reaction conditions. More importantly, the H atom on the O_2c site of the WO_x support behaves as a Brønsted acid site with a +0.69 |e| charge state (Table 2). As a result, we can conclude that both the metal-H (H^δ-) and Brønsted acid (H⁺) can form concurrently on the Pt₁/WO_x catalyst in the presence of H₂ gas. It is interesting that the Brønsted acid (H⁺) produced in situ by Pt single atoms possesses a stronger acidity than the intrinsic acid on the WO_x surface, as indicated by their different O-H frequency (3540 cm^‒1 vs. 3797 cm^‒1). In other words, the Pt₁/WO_x surface behaves like a heterolytic cleave toward H-H, but it is the result of the homolytic cleavage with the further H spillover process. The H⁺ Brønsted acid is thought to play a key role in the hydrogenolysis reaction of polyols [23, 34].

Unlike Pt₁/WO_x(001), gaseous H₂ cannot dissociate on the Pd₁/WO_x(001) surface to form metal-H species spontaneously. Instead, it can be chemisorbed and activated on Pd single atoms. The binding energy was calculated to be -0.76 eV with the H–H bond distance stretching from 0.75 to 0.92 Å. This is quite different from that on the Pd(111) surface where the dissociation is barrierless with an exothermic energy 0.98 eV. The lower activity of the Pd₁/WO_x(001) for hydrogen dissociation can be ascribed to the less metallic nature of Pd single atoms on the WO_x support; in fact, different from Pt, the Pd single atoms on the WO_x support carry slightly positive charges (0.10 |e|), which leads to the dissociation of H₂ at the interface of Pd and WO_x to form metal-H and H⁺ on the O_2c site with the help of the support [27]. Obviously, such a process on Pd₁/WO_x causes a heterolytic cleavage of the H-H bond. The transition state calculations indicate that this process is very straightforward, with a barrier of 0.67 eV and a slight exothermic energy (0.05 eV). Unlike the Pt and Pd atoms, the Au single atom on the WO_x surface is completely inert toward the chemisorption of H₂. Accordingly, direct dissociation of H₂ cannot occur on the Au₁/WO_x surface. As calculated, the heterolytic cleavage mechanism requires an activation barrier of 1.07 eV with an endothermic energy of 0.39 eV, indicative of the lowest activity of an Au single atom for H₂ dissociation on the WO_x surface, similar to the inert property of Au(111) for hydrogen activation [48].

As predicted by our above DFT calculations, the stability of single atoms on the WO_x support decreases in the order of Pt > Pd > Au, and the hydrogen molecule can dissociate spontaneously on the Pt₁/WO_x, while it can only be heterolytically dissociated with the help of WO_x on the Pd₁/WO_x, and cannot dissociate at all on the Au₁/WO_x. To validate the DFT predictions, we experimentally prepared Pt/WO_x, Pd/WO_x, and Au/WO_x catalysts using incipient wetness impregnation followed by reduction in hydrogen gas. The XRD patterns (Figure 9) of the three samples show the reflections of WO_2.92 (PDF 00-030-1387), indicating the presence of oxygen vacancies. Moreover, there are no reflections of Pt or PtO_x species despite the relatively high Pt loading (2 wt%), indicative of highly dispersed Pt species. Considering that Pd and Au are predicted to be less stable on the WO_x support, we decreased the metal loading to 0.5 wt% for these two samples. In the case of Pd/WO_x, Pd species may still be highly dispersed, as suggested by the absence of any metal reflections in the XRD pattern. However, four typical reflection peaks of metallic Au can clearly be observed in the Au/WO_x sample, suggesting the formation of large particles of Au. In agreement with the XRD results, the TEM images (Figure 10) clearly show large Au particles on the WO_x surface, while no particles can be observed on either the Pt/WO_x or Pd/WO_x sample, assuming that Pt and Pd are atomically dispersed on the WO_x support. Unfortunately, the similarity of the atomic numbers of Pt and W does not allow us to distinguish them either through HAADF-STEM or X-ray absorption spectroscopy. In conclusion, both the XRD and TEM results agree with our DFT prediction about the stability of different metal atoms on the WO_x support, that is, Pt/WO_x > Pd/WO_x > Au/WO_x.

Subsequently, we performed hydrogen chemisorption on the three samples. As predicted by DFT calculations, the hydrogen uptakes followed the order of Pt/WO_x (137 μmol/g-Pt) > Pd/WO_x (43 μmol/g-Pd) >> Au/WO_x (4 μmol/g-Au). However, hydrogen spillover can produce Brønsted acid sites, which enables the Pt/WO_x to function as a bi-functional catalyst. Therefore, we used glycerol hydrogenolysis as a probe reaction to test for bi-functionality. As shown in Table 3, among the three catalysts, only the Pt/WO_x sample was active in glycerol hydrogenolysis, affording a 37.4% glycerol conversion and 35.1% 1, 3-PDO selectivity. Neither Pd nor Au was active on the WO_x. This result is in good agreement with the mode of hydrogen activation; the greater the hydrogen spillover and the more easily it happened, the higher the activity and selectivity. By stabilizing Pt single atoms on the oxygen vacancy of the WO_x support, hydrogen activation could proceed spontaneously and homolytically, producing two separate Pt-H^‒ species, and one of them spilled over to the neighboring oxygen atom to form proton (H⁺). Such a pair of H⁺-H^‒ is required for glycerol hydrogenolysis to form 1, 3-PDO [23, 34].

Using a combination of experimental and theoretical studies, the dispersion and stability of noble metals including Pt, Pd, and Au on tungsten oxides were investigated. Under oxygen-poor conditions, the formation of bridging oxygen vacancies was thermodynamically favored with the promotion of Pt single atoms, and the Pt single atoms preferred to be trapped into hollow sites near the oxygen vacancies on the WO_x(001) surface, resulting in a highly stable and negatively charged Pt site. Such a negatively charged Pt single atom facilitates the dissociation of H₂ into the metal-H species through the homolytic cleavage of the H-H bond, and leads to spillover of H to the WO_x to form Brønsted acid sites. The stability of the single atoms on the WO_x(001) surface decreases in the order of Pt > Pd >> Au, and hydrogen can only be heterolytically dissociated at the interface between Pd and WO_x, while it cannot dissociate at all on Au/WO_x. This accounts for the unique activity and selectivity of Pt₁/WO_x in the hydrogenolysis of glycerol. Our findings in the present study not only address the stabilization mechanism of metal single atoms on oxides under a reducing atmosphere, but also give new insight into the activation mechanism of H₂ on single-atom catalysts.