Metal catalysts play an important role in petrochemical, pharmaceutical, and fine chemical synthesis, etc [1]. The challenge is to identify descriptors of catalytic activity and discover how they determine the activity and selectivity of a catalytic material [2]. The Sabatier principle is primarily used to optimize the catalyst with an intermediate binding strength between the catalyst and atoms or molecules at the surface. Similar volcano-shaped curves can be obtained empirically relating the bond energy for a reaction with the position of the transition metal in the periodic table, and with the catalytic activity of the metal [3]. Therefore, determining a descriptor to represent the "bond strength" is important. The heat of formation of the metal oxide [4, 5], the "d-hole" [6, 7], or the number of d-orbital electrons [8, 9] in the early literature has been often used to understand these volcano curves.

In recent years, theoretical chemists have proposed descriptors based on theoretical models in an effort to correlate the electronic structure of the catalyst with the catalytic activity [10, 11]. Nörskov et al. [12, 13] qualitatively described the adsorption energy of adsorbate molecules on the metal surface by using the d-band center of the transition metal surface, because the properties of the chemical bond for adsorption are intrinsically related to the electronic structure of the transition metal surface. The adsorption energy of a simple adsorbed species or the enthalpy of a rate-determining step reflects the binding energy between the adsorbate and the surface of the catalyst, and should be located near the peak position of the volcano curve. The effectiveness of this theoretical descriptor has been verified in several simple reactions, illustrating the possibility of the rational design and optimization of catalysts for the related reactions. For example, on the basis of the Brönsted-Evans-Polanyi relations and adsorption energy correlations, the descriptor for the methanation reaction is the dissociation chemisorption energy of CO [14]. The resulting mode is in good quantitative agreement with experiment, showing a well-behaved volcano curve relating the experimentally measured activity to the dissociation energy, and Ru and Co are the most active transition metals [15]. Some Fe-Ni alloys which show a higher methanation activity, with a lower price of the constituents, is predicted [16].

The limitations for these descriptors are their application in a real and complex catalytic reaction system, and the difficulty in quantitative measuring them with experiment [17]. Therefore, for a fine chemical synthesis system, exploring the nanocatalyst activity descriptors that are convenient for experimental measurement has more important practical significance for the rational design of the catalyst. Very recently, we have reported the experimentally measured surface d charge at Pd sites as a descriptor that quantitatively correlates with the adsorption strength and catalytic activity of a material for the oxidation of benzyl alcohol [18]. The validity of this descriptor needs to be extended.

The catalytic hydrogenation of N-heterocyclic compounds is of fundamental interest in organic synthesis [19]. For example, 1, 2, 3, 4-tetrahydroquinolines (py-THQ) are ubiquitous in numerous biologically active natural products and pharmacologically relevant therapeutic agents [20, 21]. But their hydrogenation remains a scientific and technological challenge due to the potential poisoning of the catalyst by either the substrate or their reduced products [22]. The electronic structure would significantly change the bond strength between metal and adsorbates and adsorption configuration [23, 24]. For example, an amine-containing support with electron-donating properties offers an electron-rich surface of Pd nanoparticles. Catalysts prefer the adsorption of quinoline over py-THQ which in turn enhances the catalytic activity and selectivity [25]. To the best of our knowledge, this relationship has not been measured.

Here an attempt is made to correlate the experimentally measurable surface d charge at Pd sites in the AuPd nanocatalysts to the activation entropy and catalytic activity for the selective hydrogenation of quinoline. A series of AuPd nanoalloys with different Pd concentrations has been synthesized which show a volcano-shape curve for the surface d-charge gain at Pd sites with Pd concentration. The maximum occurs at 33–50 at% Pd. The activation entropy and catalytic activity follow a similar volcano-shape curve. The good qualitative agreement between the surface d charge at Pd sites and the activation entropy and catalytic activity paves the way for an understanding of the trends for these complex reactions using experimentally measurable descriptors, and further optimizing and designing novel catalysts.

Ordered mesoporous carbon-supported AuPd nanoalloys were synthesized by a coordination-assisted method reported earlier using a commercial triblock copolymer (F127) as a structure-directing agent, thiol-containing silane (MPTMS) as a coordination agent, chloroauric acid and palladium chloride as metal sources, phenolic resins as the carbon source, and tetraethylorthosilicate (TEOS) as the silicon source [18]. In a typical synthesis, 5.2 g TEOS (Sigma-Aldrich, > 99 wt%), 4.9 g MPTMS (Acros Chemical Inc, 85 wt%), 5.0 g HCl (0.2 M) and 25.0 g ethanol were mixed for 30 min. Then, 8.0 g F127 (Acros Chemical Inc, EO₁₀₆PO₇₀EO₁₀₆, M_W= 12600 g mol^‒1), 20.5 mL PdCl₂ (palladium concentration in ethanol: 56.4 mmol L^‒1) and 9.0 mL HAuCl₄ (gold concentration in ethanol: 24.3 mmol L^‒1), and 25.0 g phenolic resins, which were prepared according to the established procedure as described in Supporting Information (SI), were added to the solution in that order. The mixed solution was stirred for 2 h at 40 ℃ and was then poured into several dishes. After continuous evaporation of the ethanol in an oven at 40 ℃ for 6 h followed by a further 100 ℃ thermopolymerization in an oven for 24 h, light yellow transparent films were scraped from the dishes and were refluxed in sulfuric acid with mechanical stirring (1.0 g of solid per 100 mL of 50 wt% sulfuric acid) at 90 ℃ for 24 h. After filtration, washing with distilled water, and drying at 80 ℃ under vacuum overnight, yellow solid materials were obtained. These materials were finally calcined under a nitrogen flow with heating rates of 1 ℃ min^-1 from room temperature to 350 ℃ and then 5 ℃ min^‒1 to 600 ℃, where it was kept for 3 h. The obtained materials were denoted Au₅₀Pd₅₀, the subscripts representing the respective Au/(Au + Pd) and Pd/(Au + Pd) values in molar percentages in the mother liquor.

The total amount of metals is ~3.3 wt% for the catalyst and both the amounts and the Au:Pd ratios are close to the theoretical values. The synthesis of catalysts with different Au:Pd ratios was identical except for the use of different amounts of chloroauric acid and palladium chloride. Monometallic Au and Pd catalysts were synthesized by a similar method but using only one of the precursors.

The quinoline hydrogenation reactions were carried out in a 25 mL autoclave equipped with a Teflon tube (Parr). 5 mL of Milli-Q water, 0.5 mmol of quinoline (Admas-beta, 99 wt%) and 20 mg of the Au₅₀Pd₅₀ catalyst were added to the autoclave. The catalysts were treated at 180 ℃ in an oven for 6 h before use. The autoclave was purged at least three times with hydrogen and the H₂ pressure was set to 1.0 MPa. The stirring speed was fixed at 800 rpm to exclude diffusion limitations, and the temperature was kept at 80 ℃. After each reaction, the catalyst was separated by hot filtration. The filtrate was extracted with 30 mL ethyl acetate and analyzed using an Agilent 7890B gas chromatograph (GC) equipped with a DB-1 capillary column and a flame ionization detector. The solid catalyst was recovered for reuse by washing with water and drying at 80 ℃ overnight under vacuum. Each test was repeated at least three times with an experimental error of ± 5%. The catalytic results were given in terms of the conversion of quinoline, yield of py-THQ, 5, 6, 7, 8-tetrahydroquinoline (bz-THQ) and decahydroquinoline (DHQ), initial reaction rate (r₀), turn-over frequency (TOF). The estimated TOF on the basis of exposed Pd-involving sites (TOF_Pd) was taken to eliminate the effect of Au dilution. The calculations for TOF_Pd, activation energy (E_a), and entropy of activation (ΔS^0*) were described in SI. The TOF_Pd value was calculated at a conversion below 20% and was reproducible to within ± 5%.

A soluble Pd trapping experiment using thiol group-modified mesoporous silica (SH-SBA-15) was performed to test the stability of the catalysts. SH-SBA-15 was added to the reaction with the substrate and catalyst (S:Pd molar ratio 35:1). The synthesis details of the SH-SBA-15 are given in the SI.

The recycled hydrogenation reaction was carried out over used catalyst under the reaction conditions described above. After the sixth run the catalyst was referred to as Au₅₀Pd₅₀-R. In order to keep the same amount of catalyst in each run, several parallel reactions were carried out at the same time.

N₂ adsorption-desorption isotherms were measured at 77 K using a Micromeritics TriStar II 3020 analyzer. The material was vacuum dried at 385 K for 8 h. The specific surface area (A_BET) of the material was obtained by the Brunauer-Emmett-Teller (BET) method; the pore volume (V) and pore size (D) of the material were calculated using the Barrett-Joyner-Halenda (BJH) model. X-ray diffraction (XRD) measurements were performed on a Rigaku D max-3C diffractometer using Cu K_α radiation (40 kV, 20 mA, λ = 0.15408 nm). According to the Scherrer formula, the metal (111) interlayer spacing was calculated as d = λ/2sinθ, based on the (111) diffraction peak in the wide-angle XRD pattern. Transmission electron microscopy (TEM) images were recorded on a JEM 2100 microscope operating at 200 kV. A JEM 2100 microscope and a FEI Tecnai G2 F30 microscope operating at 200 kV were used for high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) imaging. The energy-dispersive X-ray spectroscopy (EDX) spectrum was obtained in the STEM mode with a focused electron beam in the sub-nanometer range. A Varian VISTAMPX inductively coupled plasma-atomic emission spectrometry (ICP-AES) was used to determine the metal content. X-ray photoelectron spectroscopy (XPS) was performed on a Perkin-Elmer PHI 5000 CESCA model instrument. The measured XPS data was calibrated with a carbon C 1s binding energy of 284.6 eV. X-ray absorption spectra of the Au L₃-edge (11919 eV) and Pd K-edge (24350 eV) of samples were collected on a BL14W1 of the Shanghai Synchrotron Radiation Facility (SSRF). The data processing was performed using the ATHENA program. The Au and Pd extended X-ray absorption fine structure (EXAFS) data of the sample were fitted by a fast Fourier inverse transform (IFEFFIT) in R space using a software package.

The selective hydrogenation of quinoline to py-THQ was first conducted over pristine palladium and gold catalysts (Fig. 1). The maximum conversion of quinoline was as low as 5% over the gold catalyst, indicating a rather low activity. This is in agreement with both theoretical calculations and experiment. High hydrogen dissociation energies were observed over the Au (111) surface [26]. Hydrogenation reactions catalyzed by Au typically occur under demanding conditions, for example, supported on reducible oxides including TiO₂ and Fe₂O₃ and pre-adsorption of the quinoline, which enhances the dissociation of H₂ [20]. An increased activity was observed for the Pd catalyst, but the selectivity was quite low, especially at high conversions. It has been reported that by using Pd/AC as the catalyst, only 81.7% selectivity to py-THQ was obtained with the byproduct being bz-THQ [27]. With the Au₅₀Pd₅₀ nanoalloy catalyst, significant quinoline conversion was obtained. Remarkably, no byproducts are detectable.

The leaching of dissolved metal ions is first excluded since these ions have been long argued as catalysing hydrogenation in solution, and the stability of the catalysts has been studied. A bare thiol-containing solid has been used to quench quinoline hydrogenation by soluble molecular metallic species. For example, Jones et al. found that catalysis by soluble Pd species for Suzuki and Mizoroki-Heck coupling reactions is quenched by the addition of SH-SiO₂ either at the initial stage or during the reaction [28]. We have previously demonstrated SH-SiO₂ as an effective and versatile poison compared with poly(4-vinylpyridine) and thiourea-based resin QuadraPure^® TU, in quenching any catalysis by leached Pd and gold species [29, 30]. No significant difference in the kinetic profile of AuPd nanoalloys was observed in the reactions in the presence/absence of SH-SBA-15 with an S:Pd molar ratio of 35, confirming that catalysis occurs on the Pd surface rather than in solution (Fig. S1). The stability of the catalyst was studied in more detail. Quinoline conversion in the initial phase of the reaction and the end of the reaction show no substantial changes (Fig. 2) in repeated uses. These results indicate the negligible leaching of metal ions and little aggregation of the Pd particles. In addition, the strong adsorption of the reactants and products and their accumulation on the Pd surface is possibly inhibited. The solution after each run was collected, and the concentration of metal ions was below the detection limit, confirming that the catalyst has not leached [28]. These results guarantee that this is purely an investigation of surface catalysis and surface metal concentrations are almost unchanged after the reaction.

To ensure a kinetic regime, the limitations of diffusion and mass transfer were eliminated. The experiments were carried out at different stirring rates and the results showed that the reaction rate remained constant in the range of 600–1000 rpm. The Madon-Boudart test was also conducted [31]. Au₅₀Pd₅₀ catalysts with different metal loadings, ranging from 1.2 to 2.0 wt%, were synthesized (Figs. S2 and S3). The reaction rate followed a linear relationship with the metal content in the catalyst, demonstrating that there was no mass transfer effect (Fig. 3). The metal nanoparticles were easily accessible by the reactants.

Figure 1(a) shows the variation of quinoline conversion for AuPd with different Pd concentrations. As the Pd content in the catalyst decreased from 100 at% to 80 at%, 67 at% and 50 at%, the quinoline conversion increased from 12.71% to 22.92%, 26.60% and 50.64% for one hour reactions. However, decreasing the Pd concentration to 33 at% did not result in an increase of the quinoline conversion. The quinoline conversion was reduced to approximately 36.75% and 12.50% for Au₆₇Pd₃₃ and Au₈₀Pd₂₀ for one hour reactions, respectively. A similar relationship has been observed between the activity and Pd concentration in the selective oxidation of aromatic alcohols [18]. Byproducts were detected over Au₂₀Pd₈₀ and Pd during the reaction. In contrast, the selectivity to py-THQ was extremely high over the Au₈₀Pd₂₀, Au₆₇Pd_33, Au₅₀Pd₅₀ and Au₃₃Pd₆₇ catalysts with no detectable byproducts. This result is analogous to the N-modified Pd and thermally-reduced Pd supported on carbon nanotubes [32].

It is noted that the conversion plot is nearly linear during the initial period of the reaction over the catalysts studied, similar to the kinetics for the oxidation of benzyl alcohol over AuPd nanoalloys, suggesting pseudo zero-order reaction kinetics [18]. In addition, a change of quinoline concentration in the range of 0.05–0.2 mmol mL^‒1 results in an almost unchanged initial reaction rate over Au₅₀Pd₅₀ (Fig. S4). These phenomena imply that the reaction order for quinoline is pseudo zero order. Since the gold catalyst shows relatively low activity in hydrogenation, we simplify the active sites to Pd [20]. The reaction constant (k_Pd) can be deduced from r₀, and an Arrhenius plot can be plotted from which E_a and pre-exponential factors (A) can be estimated (Fig. S5). As shown in Fig. 4, the activation energies follow a reverse volcano-like curve as a function of percent Pd with the minimum at Au₅₀Pd₅₀. The apparent entropy change (ΔS^0*) reflects the activation step of the chemical reaction in the thermodynamic formulation of the rate constant and would be influenced by changes in the available electronic energy states. It has been reported that the electronic energy states may promote an electron to a localized band for adsorption of a reaction intermediate [33]. ΔS^0* follows a volcano-shaped curve, where the maximum ΔS^0* is also observed for the 50 at% Au catalyst. A linear relationship is observed between E_a and ΔS^0* (Fig. S6), which is called the compensation effect. This effect can perhaps be explained by the change in the freedom of the system, due to the loosened bonds between the surface atoms and the adsorbates as the energy of the system increases [34]. These observations strikingly demonstrate the influence of the electronic energy states of the metal on the adsorption, and in turn on the catalytic properties.

It is most important that the interpretation of catalytic results for metal alloys takes into account the possibility of phase separation [35]. TEM images of bimetallic Au-Pd catalysts supported by a carbonaceous matrix show that the metal nanoparticles are homogeneously dispersed in the matrix with an average size of approximately 3.0 nm, regardless of the metal composition (Figs. 5, 6, S7 and S8). Therefore, the size effect for the bimetallic AuPd nanoparticles on the catalytic behavior can be excluded. Hexagonally, and linearly arranged pores in large domains in the absence of distinct defects confirm the ordered 2D hexagonal mesostructure with p6m symmetry for all catalysts. HAADF-STEM images accompanied by EDX patterns were taken. The Au:Pd ratio is calculated to be close to the value of the mother liquor used for their synthesis (Table S1). Using the EDX spectra obtained with a focused electron beam in the sub-nanometer range in the STEM mode, the distributions of gold and palladium can be seen. Representative line profiles for Au₆₇Pd₃₃ and Au₅₀Pd₅₀ as examples of the different compositions in a single particle are shown in Figs. 5(d) and 6(d). These lines match well from the periphery to the centre, demonstrating that these bimetallic catalysts have uniform compositions throughout the particles with the Au-to-Pd molar ratio of approximately 2 and 1, respectively [36]. The homogeneity of the Pd distribution on the surface instead of a Pd rich phase confirms the uniform structure of the alloy.

Small-angle XRD (SAXRD) patterns show well-resolved diffraction peaks belonging to a 2D hexagonal mesostructure with a lattice constant of approximately 9.0 nm (Fig. 7(a)), regardless of the metal composition, in agreement with the TEM results. Typical type-IV nitrogen sorption isotherms are detected for all the catalysts studied, and are characteristic of mesoporous solids with a uniform pore size (Fig. S9). The asymmetric hysteresis loops are attributed to H2 type, different from the H1 type for a pristine ordered mesoporous silica/carbon composite (OMSC), indicates the non-uniformity of the mesopores originating from the intercalated AuPd nanoparticles. The calculated pore-size distribution curves are narrow with a most probable value of approximately 3.6 nm. The pore volumes and the BET surface areas are close to those for pristine OMSC and are approximately 0.23 cm³ g^‒1 and 340 m² g^‒1 (Table S1). These phenomena indicate that monometallic or bimetallic nanoparticles show no obvious effect on the mesostructure.

Wide-angle XRD (WAXRD) patterns show wide diffraction peaks at approximately 23° which are attributed to amorphous carbon and silica, and the diffuse peak at approximately 39° is attributed to the AuPd alloy with a very small particle size. Higher resolution scans were also collected in the vicinity of the diffraction peak at approximately 39° (Fig. 7(b)) and show that the diffraction peak is symmetric, which is characteristic of a nanoalloy. A clear shift of the characteristic diffraction peak for AuPd alloys compared to monometallic Au and Pd is observed, and the peak is located between them. For example, the Au₅₀Pd₅₀ sample has a cell constant of 0.4017 nm, giving further evidence on the alloy structure. As a result, the formation of a monometallic core-shell structure or macroscopic phase separation is excluded. The elemental compositions, estimated using Vegard's law in WAXRD patterns, are close to the theoretical values for dilute Pd catalysts, but show the distinct deviation for Pd-rich ones, due to an obviously larger lattice constant compared to the alloy having complete solid solubility. This phenomenon implies that an ideal alloy is formed for dilute Pd catalysts, while gold segregation in micro-domains occurs for dilute Au catalysts.

Typical doublets of the Au 4f, and Pd 3d core level bands are detected for monometallic or bimetallic catalysts (Fig. 8). The doublets for Au 4f_7/2 and 4f_5/2 can be extensively deconvoluted, with the fitted dominant peaks at 84.30 and 88.00 eV corresponding to Au⁰. With increased Pd concentration, the Au 4f core levels show a gradual negative shift for AuPd nanoalloys. Similar core-level shifts have been reported for either bulk AuPd alloys or nanoparticles, which is ascribed to the Au-Pd alloying interaction and increased number of Au-Pd bonds [37-40]. A relatively large shift occurs when mixing Pd with Au, and the shift then undergoes an extremely slow change in dilute Au alloys, implying that the Au-Pd coordination number increases according to the Pd concentration in dilute Pd alloys, but may show a slight increase or even be constant in dilute Au alloys. This result is in good agreement with the XRD results, thus confirming the alloy structure.

Since a distinct overlap is observed between the Pd 3d_5/2 and the Au 4d_5/2 components for the bimetallic catalysts, the Au 4d_5/2 intensity was calculated from the well-resolved Au 4f_5/2 intensity, and this value was subtracted from the above overlapped peak to determine the Pd 3d_5/2 intensity. The extensively deconvoluted doublet in the Pd 3d region for the monometallic Pd catalyst shows that the fitted major peak is due to metallic Pd, and the minor one to a Pd–O coordination [30, 41]. It is noted that the Pd 3d core levels for AuPd nanoalloys show a gradual negative shift with decreased Pd loading due to the Au-Pd alloying interaction, and for dilute Pd alloys is relatively large. Considering the atomically dispersed Pd⁰ species in these alloys, which are mainly responsible for the significant XPS positive chemical shift of 1.4 eV for Pd 3d lines due to their smaller electron relaxation energy [42], the negative shifts for the Pd 3d core-level bands in dilute Pd nanoalloys are even more dominant. It is interesting that both the Au 4f and Pd 3d core level bands in bimetallic species show obvious negative shifts compared to the corresponding monometallic species. These phenomena are in accordance with bulk AuPd alloys [43], but in contrast to Cu–Pd bulk alloys [44], where the Pd shifts are positive and those for Cu are negative. Medlin et al. [45] also observed a shift of the Au 4f core level to a lower binding energy in a AuPd/C nanoalloy catalyst compared to monometallic Au/C, and agreed with that charge transfer may not be the correct explanation for the core-level shifts. Therefore, the hybridization changes between Pd and Au are the possible reason that the occupied valence shells redistribute. The estimated Au:Pd ratios in alloys with different Pd concentrations are similar to the theoretical values further confirming the alloy structure, and the surface adopting similar Pd-Au coordination with the bulk (Table S1).

The above results are in good agreement with the previous reports for AuPd nanoalloy catalysts. As a consequence, we used the earlier XAFS spectra for the present catalysts. The ratio of CN_Au–Au to CN_Au–Pd, generated from Au L₃ k³-weighted and Fourier transformed EXAFS fittings (Table S2), is very close to the Au-to-Pd molar ratio for Au-rich nanoalloys (including Au:Pd = 1), indicative of the formation of homogenous alloy structures. A deviation occurs for dilute Au catalysts, for example, the CN_Au–Au to CN_Au–Pd ratios for both Au₃₃Pd₆₇ and Au₂₀Pd₈₀ are much larger than the Au-to-Pd molar ratios. These results coincidently indicate a slightly different structure for dilute Au alloys compared to dilute Pd-alloys. Gold clusters are more likely to be formed in dilute gold alloys. A calculation from the EXAFS spectra of the AuPd nanoparticles near the Pd K-edge gives a reduction of CN_Pd–Pd from 2.8 to 0. The Pd–O contribution is dominant in Pd-rich catalysts, demonstrating that extremely small Pd clusters are formed when a small amount of Au is dissolved in Pd. Considering the surface Pd–O and Pd–Pd coordination, the coordination number for Pd clusters to that for Pd–Au is close to the Pd/Au ratio, implying a random alloy. When the Au-to-Pd ratio exceeds 2, the peaks to Pd–Pd and Pd–O shells almost disappear. The XAFS features show high similarity to AuPd nanoalloy phase [46]. Contiguous Pd clusters disappear with increasing Au loading, isolated Pd ensembles form, and eventually each Pd atom is separated by Au. An Au-rich core and a Pd-rich shell can also be excluded because Pd–Pd coordination is clearly detected for this structure when the Pd concentration is greater than 20 at% [47, 48].

The microstructural discrepancies related to the metal concentrations are caused by surface energy versus surface mixing energy, determining whether the two components mix or form islands. The segregation energy of Au–Pd is 0.15 eV for Au-host and ‒0.14 eV for Pd-host alloys [49]. As a result, Pd thermodynamically prefers a random-distribution, and individual Pd atoms form a complete solid solution with the Au-host. While Au prefers to dissolve in Pd as small Au islands [49, 50]. These results are consistence with the previous report, and suggest the existence of a complete range of stable solid solutions.

Band theory indicates that in a face-centered cubic (fcc) transition metal alloy, the narrow e_g band with a high density of states will be more readily filled than the t_2g state by the introduction of s electrons from the non-transition metal present in the alloy [33]. The complete filling of d-bands at a specified alloy composition results a pronounced change in the available electronic energy states.

XANES spectra of the Au L₃-edge for the Au foil clearly shows the edge that probes the transition from the 2p_3/2 to 5d_5/2 and 5d_3/2 dipole-allowed transitions, resulting in a white line at 11925 eV whose intensity reflects the number of holes in the d-band of the Au atoms (Fig. S10). On one hand, the nanosize effect causes a decrease in the line intensity because of the smaller number of atoms forming a lattice, and therefore a decrease in the d-band width [51]. A comparison of the white line intensity for a 3 nm Au catalyst with that of a Au foil, indicates an increase in the d electron count of the Au atoms in the former. Since dilute Pd alloys have similar Au coordination numbers to a pure Au catalyst, the almost loss of the white line intensity when mixing Pd with Au indicates the alloy structure. The Au L₃-edge resonance is insensitive to the alloy composition, and the involvement of 20% Pd is sufficient to deplete unoccupied Au 5d_5/2 states near the Fermi level. The presence of Pd has a dominant effect on the filling of the Au d-band (an increase of the d-electron density). A combination of the hybridization changes between Pd and Au is indicated by the XPS results, which can be attributed to a change in d-band hybridization such as a decreased band width and spin-orbit splitting and is not due to charge transfer, in agreement with the fact that the extent of charge transfer is small for Pd-Au alloys [52]. It has been reported that the Au-Pd alloying interaction produces a more pronounced change of the Au d electron density than the size effect does [53]. A dominant increase in the intensity of the second feature after the edge is observed with the increase of Pd content. This change is caused by the change in the distance between the two types of atom which causes an intra-atomic redistribution of charge [37, 54]. As a result, the gold is in the metallic state, and the Pd–Au d-d interaction is believed to have an important effect on the electronic structure of the alloy. In particular, with increasing Pd content the Au 5d_5/2 partial density of states is broadened and shifted to a higher binding energy relative to the Fermi level [52].

The Pd XANES spectrum at the K-edge of the Pd foil exhibits a pronounced white line due to the unfilled Pd d-band, and the K-edge of the PdO exhibits an even more dominant white line with a shift of the threshold energy to a higher energy because of the higher oxidation state of Pd in PdO (Fig. S11). The hybridization-mediated 1s → 4d, dp (a) absorption transition in the pre-edge region is unresolved. The near-edge spectrum shows two resonance peaks due to 1s → 5p (p) and 1s → 4f (f)transitions. The shape of the second absorption edge reflects the extent of 4d-5p hybridization [55]. This feature is insignificant for the PdO reference material. Similar features are observed in the spectrum of the monometallic Pd nanocatalyst and in that of the Pd foil, showing a similar coordination symmetry to the Pd metal of fcc structure. However, the post-edge threshold energy shifts to a slightly higher energy for the monometallic Pd catalyst. The possible reasons for this lie in the tiny Pd particles having a Pd metal fcc structure with very short range order or the coordination of Pd with a light O atom [56]. Upon dissolution of 20 at% and 33 at% Au into Pd nanoparticles, three differences can be seen. First, a further shift of the white line energy to a higher energy is observed, in nearly identical fashion to the absorption edge of PdO, in good agreement with the increase of Pd–O coordination. Second, a positive shift in the 1s → 5p (p) transition accompanied by signal broadening is detected, which is possibly due to an electronic perturbation of the Pd atoms going from a Pd-like environment in the monometallic structure to the AuPd alloy phase. The Pd d-band becomes occupied [57]. Third, the 1s → 4f (f) transition shifts to a lower energy and its intensity is significantly reduced. This shift can be interpreted as due to an increase in the atomic separation and the 4d-5p hybridization that has been observed for the local distortion of the Pd lattice by interstitial C [58] and AuPd bimetallic nanoparticles [59]. A further decrease in Pd concentration to 50% (Au₅₀Pd₅₀) leads to no further shift, implying that the 4d band is filled [60]. By comparison, both the white line and p and f transitions in dilute Pd nanoalloys shift towards that for a Pd foil. As a result, the orbital character of the screening charge may change [61].

In the present AuPd nanoalloy, for a greater than 50 at% Pd content the Pd d-band becomes occupied when alloyed with Au. With a further reduction of Pd concentration, it may be conjectured that the 5sp electrons of the surface Pd atoms lose some d-character possibly due to the increased Pd–Pd bond distances and isolated Pd atoms. The d-d separation would narrow the 4d band of surface Pd sites effectively pushing its top below the Fermi level. For Au-rich alloys with a low density of states the screening efficiency will be lower [62]. Due to strong Au–Pd d-d interactions, the orbital character of the unoccupied states at the Fermi energy changes from predominantly d to predominantly sp-type as the Pd concentration in the alloy decreases [61].

Non-d- and d-charge transfer onto atom sites can be estimated based on the XPS and XANES results (Table S3) [63]. The binding energy shift, ΔE_B, between the pure metal and the alloy measured relative to the Fermi level in XPS experiments is related to the change of the Hartree-Fock one-electron energy of the core level (Δϵ(i)), the change in the work function (ΔФ), and the final relaxation state of a core hole (ΔE_γ) [64]. Δϵ(i) is directly related to the charge transfer in the initial ground state, and is caused by the effects of the Coulomb repulsion energy between the core electron and the valence electron which arises from the redistribution of valence electrons, and the change in the number of the valence electrons which is associated with the charge changes on other lattice sites due to the total charge transfer in or out of the parent atomic site. A theoretical calculation on alloy systems based on the pseudopotential linear response method, shows that in the presence of the core hole △E_γ can be as large as 2 eV, which is the same order of magnitude as the initial-state effect Δϵ(i) and therefore cannot be neglected [65]. The estimate is in good agreement with the above discussion. When adding Au to Pd, the alloy has a very small amount of charge transfer (< 0.1 electron). The d charge transfer at the Pd sites is small and the compensation comes from the gaining of 0.1–0.2 sp electrons at Au sites. The decrease of Pd concentration results in a pronounced change in the available electronic energy states, and an increase in the number of d electrons at Pd sites. At approximately 33 at%–50 at% Pd the alloy contains Pd atom pairs or isolated Pd atoms, and the d-electron gain reaches the maximum value of approximately 0.28 e. For a further decrease of Pd content the total charge transfer is large, but the d-electron gain at Pd sites is reduced to some extent. From a kinetic point of view, the measured apparent entropy change is the difference between the entropy of the transition state and the reactant. The magnitude of the entropy value, in general, shows that the activated complex is an immobile molecule attached to the catalyst surface [66, 67]. As a result, the density of states close to the Fermi level of the transition metal has a large effect on ΔS^0* because of its effect on the bound state of the absorbed intermediate [68]. On the other hand, the d-band theory proposes that for a given metal and adsorbent, the center of themetal d states relative to the Fermi level determines the binding energy of the adsorbent to the surface. Inspired by these works [69], we therefore attempted to relate the increased d charge at the Pd sites to the activation entropy, ΔS^0*, and the TOF_Pd value (Fig. 9), and both showed a linear relationship. These results indicate that the Pd charge of the Group VIII-IB nanoalloy is fundamental to the adsorbate binding strength in the surface reaction. In addition, the increase in ΔS^0*, which is related to the adsorption strength, is highly related to the increase in the reaction rate. Accordingly, the d charge at surface Pd sites can serve as a descriptor, which can correlate well with the catalytic activity. It should be mentioned that we simplify the relationship. On one hand, the surface changes from Pd clusters with Pd–Pd coordination, to Pd islands with Pd–Pd and Pd–Au coordination, and finally to isolated Pd with Pd–Au coordination. Comparing the activity of Au₆₇Pd₃₃ to Au₈₀Pd₂₀ both of which possess isolated Pd, the TOF_Pd value is higher for the former than the latter. An overloading of Au which reduces the d-electron gain leads to a decreasing activity. As a result, we simplify the effect by electronic properties instead of structures. On the other hand, the present study is limited to a strong-adsorbate-involved reaction. Quinoline is reported to adapt a di-bridge adsorption configuration on Pd cluster with a high adsorption energy[70]. Both the adsorption strength and configuration are significantly affected by the electronic properties, in particular, d-electron. A linear relationship has been found to correlate the adsorption strength and configuration, here represented by activation entropy, with the d-electron gain at Pd sites. Taking account of the complexity for surface reaction on nanoparticles, other descriptors should also play important roles. The factors are still being investigated in our lab, and deserve to be reported separately.

The absolute thermoelectric power, S [71] of AuPd alloys is found to be highly sensitive to the presence of unoccupied d-band states. A sudden drop of the S value is expected as the d-band is filled. The results on bulk AuPd alloys indicated that the d-band filling point was very close to 50 at% Au. In the present ordered mesoporous carbon-confined AuPd nanocatalysts, the d-band filling tendency is in good agreement with that for bulk AuPd alloy. We therefore attempted to relate the thermoelectric power S to the activation entropy ΔS^0* (Fig. S12). The entropy change indeed follows a similar trend with alloy composition as does the thermoelectric power, which shows a minimum at 50 at% Au. This indicates that the d-band energy position in the Group VIII-IB nanoalloy is fundamental to the activation entropy associated with the adsorbate binding strength in the surface reaction, similar to that for transition metals [72, 73]. An increased reaction rate for the "true" nanoalloy would ordinarily be expected from the decrease in E_a and the compensation effect by the decrease in ΔS^0* which is related to a weaker adsorption strength.

Experimentally measurable surface d charge at Pd sites has been proposed to serve as a descriptor that quantitatively correlates to the adsorption strength and catalytic activity of AuPd alloys for the selective hydrogenation of quinoline. AuPd nanoalloys with different Pd concentrations supported on ordered mesoporous carbon have been synthesized and used as model catalysts. These nanoalloys have tunable Au–Pd, Au–Au and Pd–Pd (Pd–O) coordination numbers, and therefore tunable electronic properties. Hybridization changes between Pd and Au are the possible reason for changing the electronic properties, which in turn have a great influence on the adsorption energy. Non-d- and d-charge transfer onto the Pd atom sites has been estimated. The d-electron gain on the Pd surface has a linear relationship with the activation entropy and catalytic activity. The maximum d-charge gain at the Pd sites occurs at Pd concentrations of 33 at%–50 at%, with the alloy showing an almost 11 times higher activity than a pure Pd catalyst.