Pd Au bimetallic catalysts: understanding alloy effects from planar models and (supported) nanoparticlesw

Transcription

1 Chem Soc Rev Dynamic Article Links Cite this: Chem. Soc. Rev., 2012, 41, TUTORIAL REVIEW Pd Au bimetallic catalysts: understanding alloy effects from planar models and (supported) nanoparticlesw Feng Gao* a and D. Wayne Goodmanz b Received 28th April 2012 DOI: /c2cs35160a Pd Au bimetallic catalysts often display enhanced catalytic activities and selectivities compared with Pd-alone catalysts. This enhancement is often caused by two alloy effects, i.e., ensemble and ligand effects. The ensemble effect is a dilution of surface Pd by Au. With increasing surface Au coverage, contiguous Pd ensembles disappear and isolated Pd ensembles form. For certain reactions, for example vinyl acetate synthesis, this effect is responsible for reaction rate enhancement via the formation of highly active surface sites, e.g., isolated Pd pairs. The disappearance of contiguous Pd ensembles also switches off side reactions catalyzed by these sites. This explains the selectivity increase of certain reactions, for example direct H 2 O 2 synthesis. The ligand effects are electronic perturbation of Pd by Au. Via direct charge transfer or by affecting bond lengths, the ligand effects cause the Pd d band to be more filled, moving the d-band center away from the Fermi level. Both changes make Pd more atomic like therefore binding reactants and products more weakly. For certain reactions, this eliminates a so-called self-poisoning effect and enhances activity/selectivity. 1. Introduction Alloys are a class of important heterogeneous catalysts as they frequently exhibit much enhanced catalytic stabilities, activities and selectivities, as compared with their single-metal constituents. 1 The Ponec and Bond definition of alloy is the following: alloy is most conveniently defined as a metallic system containing two or more components, irrespective of their intimacy of mixing or, precise manner of mixing. 2,3 When alloys contain two metallic components, for example Pd Au, they are sometimes also called bimetallics. In this article, we intend to call a Pd Au catalyst alloy when Pd and Au are intimately mixed otherwise bimetallics when Pd and Au are segregated. Among alloy catalysts, Pd Au has received a great deal of attention because of its superior activity in a number of reactions. Pd Au catalysts are used in the industrial synthesis of vinyl acetate (VA). In the United States alone, 4.8 million tons of VA are produced over this catalyst annually. 4 Pd Au catalysts also catalyze low-temperature CO oxidation, 5 7 direct H 2 O 2 synthesis from H 2 and O 2, 8 11 hydrodechlorination of Cl-containing pollutants in underground water, 12 hydrodesulfurization of S-containing organics, 13 hydrogenation a Institute for Integrated Catalysis, Pacific Northwest National Laboratory, P.O. Box 999, Richland, WA 99352, USA. feng.gao@pnnl.gov; Fax: ; Tel: b Department of Chemistry, Texas A&M University, P.O. Box 30012, College Station, TX , USA w Part of the bimetallic nanocatalysts themed issue. z Dr Goodman deceased on February 27, of hydrocarbons, acetylene trimerization, and many other reactions. Alloying induces multiple changes in the physical and chemical properties of the metallic components. Where catalytic properties are concerned, two alloy effects are significant: (1) ensemble effects, i.e., a finite number of atoms in a particular geometric orientation that are required for facilitating a particular catalytic process; and (2) ligand effects, i.e., electronic modifications resulting from hetero-nuclear metal metal bond formation. The latter could involve charge transfer between the metals or orbital rehybridization of one or both metallic components. 1 It has to be noted that one cannot vary the composition of the catalyst surface without affecting both the distribution of the ensembles and changing the electronic structure of the individual constituent atoms in the surface. 21 Still, some suggest that ensemble effects play a more dominant role than ligand effects. 3,21 For the Pd Au system in particular, an ensemble effect is mainly a diluting effect where the catalytically more active component (Pd) is diluted by the less active component (Au). As the surface ratio of Au Pd increases, sizes of contiguous Pd ensembles decrease and eventually all Pd atoms are separated by Au as isolated Pd monomers. 1,5,6,22 With regard to the possible ligand effects in this catalyst, one would intuitively expect that charge is transferred from Pd to Au since Au has higher electronegativity. This statement may be true but is oversimplified. Among metals, Au has one of the largest electron affinities. In bulk gold-containing intermetallic compounds, Au usually gains s, p electrons, but this gain of charge This journal is c The Royal Society of Chemistry 2012 Chem. Soc. Rev., 2012, 41,

2 is partially compensated by a depletion of Au 5d electrons. 23 This situation also applies for the Pd Au system. Indeed, there appears to be general agreement that upon alloying, Au gains s, p electrons and loses d electrons whereas Pd loses s, p electrons but gains d electrons. 1,13,14,23 Charge transfer between Pd and Au also helps to explain why Au is able to fully isolate Pd: notably, there exists some Coulomb Pd Pd repulsion in bulk Pd whereas Pd Au attraction is realized as a result of net charge transfer from Pd to Au. For late-transition metals like Pd and Au, the d-character is much more important than s, p-character in defining their chemisorption and catalytic properties. For Pd, gaining d electrons shifts the d band center away from the Fermi level, which leads to weaker interaction between adsorbates and surface Pd atoms. 21 Indeed, recent theoretical calculations demonstrate that the Pd d band for Pd monomers surrounded by Au is much lower in energy than that for Pd monolayer or bulk Pd surfaces. 24 There is another reason that causes the Pd d band to narrow: a lattice mismatch between Pd and Au, with Pd having a lattice constant B5% smaller than Au. 1 Upon alloying, however, Pd has been found, in certain cases, to adopt the lattice constant of Au. 14,17 In this case, Pd Pd bond length increases causing the Fermi level within the Pd d band to rise. This also enhances the atomic-like character of Pd atoms and correspondingly, weakens binding toward reactants. From both charge transfer and bond length arguments, we understand that Au is able to weaken the binding strength of Pd by perturbing its d band. However, this does not necessarily mean that Au weakens the catalytic activity of Pd. In contrast, enhanced activity of Pd within Pd Au alloys is frequently found as compared to pure Pd. Besides other factors, diminishing the so-called self-poisoning by reactants/products is one reason that accounts for the activity enhancement. Some examples are shown in the following sections. Pd Au catalysts are divided into two categories in this article: (1) planar models used by surface scientists, including single crystals and thin films; (2) nanoparticles for practical applications. The latter include traditional high-surface-area carrier supported metallic particles where the sizes of the particles typically fall in the nano-dimensional range; and, in the past two decades, the rapidly developing unsupported nanoparticles. Each category has its advantages and disadvantages. Planar model catalysts are generally referred to as well-defined catalysts as these are synthesized under wellcontrolled conditions and characterized with a wide array of surface-sensitive techniques such that they are often understood at an atomic level. 25,26 However these model catalysts are typically used under non-practical ultrahigh vacuum (UHV) conditions. To overcome the pressure gap, some researchers use coupled high-pressure reactor/uhv systems to study catalysis on planar models at elevated pressures. 27 On the other hand, nano-particle catalysts are more practical and can be used under various realistic reaction conditions. However, these catalysts are generally too complex to understand at an atomic level. Indeed, as stated by Crooks and co-workers, 28 unambiguous structure determination for particles in the size range of o2 nm remains a major analytical challenge. In this article we will cover both categories of catalysts. However to address the alloy effects more explicitly, more attention is paid to planar model catalysts since both ensemble and ligand effects can be probed at an atomic level in this case. Feng Gao received undergraduate and graduate education at Tianjin University, China, in the 1990s in Chemical Engineering. He joined the University of Wisconsin-Milwaukee in 2000 as a graduate student and received a PhD in Physical Chemistry in 2004 under Prof. Wilfred T. Tysoe. From 2007 to 2009, he was a postdoc at Texas A&M University under Prof. D. Wayne Goodman. He Feng Gao had a brief stay at Washington State University as a research faculty member and is currently a staff scientist at Pacific Northwest National Laboratory (PNNL), conducting research in basic and environmental heterogeneous catalysis. He is a coauthor of 60 publications. Wayne Goodman joined the faculty of the Chemistry Department at Texas A&M in 1988 where he is currently a Distinguished Professor and the Robert A. Welch Chair. Previously he was the Head of the Surface Science Division at Sandia National Laboratories. He was the recipient of the Ipatieff Award of the American Chemical Society in 1983, the Colloid and Surface Chemistry Award D. Wayne Goodman of the American Chemical Society in 1993, the Yarwood Medal of the British Vacuum Society in 1994, a Humboldt Research Award in 1995, a Distinguished Research Award of Texas A&M University in 1997, the Giuseppe Parravano Award in 2001, the Adamson Award for Distinguished Service in the Advancement of Surface Chemistry of the American Chemical Society in 2002, the Gabor A. Somorjai Award of the American Chemical Society in 2005 and elected Fellow of the American Chemical Society in He is the author of over 540 publications/book chapters and an active member/officer of a number of professional societies. He has served as an Associate Editor for the Journal of Catalysis, and served on the Advisory Boards of Surface Science, Langmuir, Catalysis Letters, and The Journal of Physics: Condensed Matter Chem. Soc. Rev., 2012, 41, This journal is c The Royal Society of Chemistry 2012

3 2. Planar model catalysts 2.1. Model catalyst formation and characterization Three types of planar model Pd Au alloy catalysts have been developed: (1) Pd Au single crystals, e.g. AuPd(100) 5,6,22 and Au 3 Pd(100) alloys; 29 (2) Pd Au thin films generated by depositing Au on Pd single crystals, 19,30 34 or depositing Pd on Au single crystals, 18,35,36 or co-depositing Au and Pd on Au 37 or refractory metal substrates; 1,38 and (3) Pd Au clusters generated by co-depositing Au and Pd on planar oxide thin films. 39,40 For the latter cases, annealing of the thin films after deposition is required to facilitate alloying. Since the reactions occur on surfaces of heterogeneous catalysts, knowledge of the surface composition and structure, preferably at an atomic level, is of vital importance. Three important surface science techniques used to characterize Pd Au alloy surfaces are briefly described below. First, since Pd and Au have different electronic density of states (DOS) near the Fermi level and alloying does not eliminate this difference, scanning tunneling microscopy (STM) can be used to image a Pd Au alloy surface which gives a contrast between surface Pd and Au atoms. Fig. 1 presents an example of this for a clean AuPd(100) surface. 22 In this case, as shown in Fig. 1(A), some surface atoms appear to be brighter than the others. Following a simple data treatment to enhance the contrast (shown in Fig. 1(B)), coupled with techniques capable of chemical identification, one is able to construct the surface structure shown in Fig. 1(C) (Pd atoms are displayed as darker spots) where all Pd ensembles (isolated monomers, isolated monomer pairs, dimers, trimers, etc.) can be identified. A second technique that deserves mentioning is low energy ion scattering spectroscopy (LEISS). In this technique, a beam of noble gas ions with energy between 0.1 to 10 kev impinges on a solid surface and are scattered. The energy of outgoing ions is determined by the laws of energy and momentum conservation; therefore, surface atoms with different atomic masses are distinguished. Note that this technique is essentially exclusively sensitive to the topmost layer of a surface since more than 99% of the ions penetrating the first layer will be neutralized and won t be detected by an ion energy analyzer. Fig. 2(A) presents a LEISS example of thin Pd Au alloy films deposited on Mo(110) in ultrahigh vacuum. In this experiment, 5 monolayers (ML) of Au are first deposited on Mo(110) followed by 5 ML of Pd. The thin film is then annealed to various temperatures to allow different levels of intermixing. A stable Pd Au alloy, with apparent Au surface segregation, is achieved between 600 and 1000 K. At higher temperatures, Au first sublimes followed by Pd. 1 Using the scattering intensities of Pd and Au, corrected with a sensitivity factor (i.e., the signal intensity ratio of pure Pd and Au), the surface coverage of Pd and Au for any Pd Au alloy surface can be precisely determined. By varying the bulk Pd Au ratio, the composition of the stable surface layer varies accordingly and a phase diagram is thus constructed as shown in Fig. 2(B). Clearly, strong Au segregation to the surface occurs for all bulk Pd Au ratios in UHV. This is rationalized by the fact that Au has smaller surface free energy than Pd. 1 In general, the surface segregation of an alloy component depends on the enthalpy of mixing, the atomic sizes of the metals and the surface free energies (which are proportional to the heats of sublimation). 23 As will be shown below, segregation can also be adsorption/reaction induced. Both STM and LEISS, and most other surface-sensitive techniques require ultrahigh vacuum to operate; therefore, in situ applications at elevated temperature and pressure are exceedingly difficult. Fortunately, infrared reflection absorption spectroscopy (IRAS), coupled with CO titration, can be used as a powerful tool to probe surface Pd and Au ensembles under such conditions. Especially, when a photoelastic modulator is added to the IR beam to perform polarization modulation (PM) to eliminate gas-phase signals, the so-called PM-IRAS technique is very useful in probing ensembles on planar Pd Au alloy surfaces at elevated temperature and pressures. 5,6,40 Fig. 3 presents temperature-dependent PM-IRAS spectra of Torr (A) and 10 Torr (B) of CO exposed to an AuPd(100) surface initially enriched with Au (B90%). CO vibrational features are assigned as follows: n CO at >2100 cm 1 corresponds to atop CO on Au sites; n CO at cm 1, to atop CO on isolated Pd sites; and n CO between 1900 and 2000 cm 1, to bridging CO on contiguous Pd sites. As is clearly displayed in Fig. 3(A), due to the strong interaction between Pd and CO, subsurface Pd atoms are pulled out to the surface at temperatures higher than B240 K. However the Pd segregation is insufficient to form contiguous Pd sites at this CO pressure. Higher CO pressure is needed to segregate a sufficient amount of Pd to the surface to Fig. 1 (A) STM image of an AuPd(100) bulk alloy (10 nm 10 nm, V s = 15 mv, I t = 6.3 na). The large white features are impurities thought to be carbon. (B) The same STM image as that in (A) excluding all data points below the cutoff height, which is set to 5 pm below the highest point of the image. The color bar scale spans from 0 to 5 pm. The red circles denote the features designated to be Pd atoms. These red circles are set to have an area of B0.15 nm 2. (C) Schematic representation of (A) for clarity. Figure adapted with permission from ref. 22. Copyright (2007) by American Chemical Society. This journal is c The Royal Society of Chemistry 2012 Chem. Soc. Rev., 2012, 41,

4 Fig. 3 (A) Temperature-dependent PM-IRAS spectra of Torr of CO on an AuPd(100) surface well-annealed at 800 K for 30 min. (B) Temperature-dependent PM-IRAS spectra of 10 Torr of CO. Figure adapted with permission from ref. 5. Copyright (2009) by American Chemical Society. Fig. 2 (A) LEISS spectra of 5 ML Pd/5 ML Au/Mo(110) as a function of annealing temperature. LEISS spectra were collected at 300 K after the sample was annealed to the specified temperature. (B) Surface concentration of various Pd Au alloys on Mo(110) measured by LEISS compared to the corresponding bulk concentration. The sample was annealed at 800 K for 20 min. Figure adapted with permission from ref. 1. Copyright (2005) by American Chemical Society. form contiguous Pd sites (Fig. 3(B)). This chemisorption induced segregation appears to be easy to understand: as stated by Haire et al., on thermodynamic grounds, it is to be expected that the surface composition will adjust so that the surface becomes enriched in the element which interacts more strongly with the adsorbate. 41 Importantly, it is found from Fig. 3(B) that at temperatures above B475 K, the bridging CO band disappears while the atop CO band is still present up to B650 K. It is expected, however, that the binding energy of bridging CO species is higher than atop CO. This is best rationalized by the fact that Au starts to diffuse back to the surface to isolate Pd as the temperature rises. This is because the sticking probability of CO decreases at high temperatures and the pulling force added to surface Pd therefore weakens. This simple chemisorption experiment, nevertheless, represents a good example of dynamic catalyst restructuring at elevated temperature and pressures. Spectroscopic methods applicable at such conditions, for example PM-IRAS and sum frequency generation (SFG), are highly desirable in these cases to capture the dynamic changes. Ligand effects have also been found to influence CO adsorption on Pd Au alloy surfaces. It has been well-documented that on pure Pd(111), CO occupies threefold hollow sites at low coverage. 42 Recent XPS measurements on an Au/Pd(111) system has revealed that even 10% surface Au is sufficient to switch off binding at threefold hollow sites. 43 Clearly, 10% surface Au is insufficient to remove all Pd threefold hollow sites (ensemble effect) but apparently does sufficiently destabilize them (ligand effect). This example also demonstrates a synergy between ensembles and ligand effects in affecting chemisorption Examples of catalytic reactions over planar models Acetylene trimerization. Acetylene trimerization (3C 2 H 2 - C 6 H 6 ) is an interesting model reaction as it occurs both in UHV on Pd single crystal surfaces and at elevated pressures on supported Pd catalysts. Surface science studies have revealed that this reaction proceeds via ac 4 H 4 metallacycle intermediate; the coupling of this intermediate with another acetylene molecule gives rise to benzene. The rate limiting step has been found to be benzene desorption that occurs from two states of adsorbed benzene: tilted and flat-lying. The former state occurs at much lower temperatures than the latter; the latter state also accompanies certain levels of product dissociation. 44 The study of acetylene trimerization over Pd Au alloys is interesting since (1) this reaction is sensitive to both the structure and composition of the metal surface, and (2) pure Pd is very active, however, pure Au is totally inert. 17 Lambert and co-workers systematically studied acetylene trimerization over Pd Au thin films formed on Au(111) 18 and Pd(111). 19 Due presumably to the mobility difference of Au in 8012 Chem. Soc. Rev., 2012, 41, This journal is c The Royal Society of Chemistry 2012

5 Pd and Pd in Au, a (O3 O3)R301 (composition Pd 2 Au) surface alloy forms following Pd deposition onto Au(111) and annealing. 18 However, after Au deposition onto Pd(111), Pd and Au distribute quasi-randomly. 19 In this case, higher annealing temperatures induce more Au diffusion into the Pd bulk; therefore, simply by varying the annealing temperatures, surface alloys with different Pd Au ratios are formed. 19,33 In terms of reactivity, benzene forms and desorbs at much lower temperatures over the alloy surfaces than Pd(111) or pure Pd overlayers on Au(111). Lambert and co-workers gave two reasons why the alloy is more effective than pure Pd. First, the degree of rehybridization of the adsorbed acetylene decreases on the alloy surface. This results in more efficient conversion of acetylene to benzene than acetylene decomposition. Second, the strength of adsorption of the benzene product is also decreased on the alloy surface. Again, the weaker interaction with the alloy surface increases the probability of product desorption without decomposition. 18 Both ensemble and ligand effects seem to give satisfactory qualitative explanations for the reduced binding of both the reactant and product with the alloy surfaces. On the Au/Pd(111) alloy surfaces, an AuPd 6 ensemble was found to give rise to the highest benzene yield. This suggests that a 7-atom ensemble is needed for benzene formation. 19 However the high activity of AuPd 6 is obtained under highly idealized conditions during temperature-programmed desorption following adsorption of a saturation coverage of acetylene in UHV. This does not necessarily mean that an AuPd 6 ensemble is highly active, or even needed under other reaction conditions. The (O3 O3)R301 (composition Pd 2 Au) surface alloy formed on Au(111) is also very active yet AuPd 6 ensembles should not be present at high concentrations on this surface. Nevertheless, the planar model catalysts can provide important information for the acetylene trimerization reaction over supported Pd Au catalysts under realistic conditions. For example, supported Pd Au alloy catalysts show higher activity and selectivity than supported Pd-shell Au-core bimetallic catalysts, 17 fully consistent with the surface science studies Vinyl acetate synthesis. Acetoxylation of ethylene on silica-supported bimetallic Pd Au catalysts promoted with potassium acetate is a well-established commercial route for synthesis of vinyl acetate (VA), given by the following reaction: CH 3 COOH + C 2 H O 2 - CH 3 COOCHQCH 2 + H 2 O. 4,35,45,46 Compared with pure Pd, the addition of Au has been shown to significantly improve reaction rates whilst also moderately improving the reaction selectivity. 46 The side reactions are combustion of the reactants acetic acid and ethylene, and the product VA; ethylene combustion has been found to be the dominant one under typical reaction conditions. 4 Using temperature-programmed desorption, the much weaker binding of ethylene with Pd Au alloys (as compared to pure Pd) was confirmed. 45 Since ethylene combustion must involve C C and C H cleavage of chemisorbed ethylene, the fact that ethylene binds much weaker on Pd Au alloys makes it easy to understand why VA selectivity is higher on a Pd Au alloy catalyst. Using silica-supported Pd (1.0 wt%) and Pd Au alloy (1.0 wt% Pd and 0.4 wt% Au) catalysts at reaction conditions mimicking those used in the industrial process, VA formation rates were found to be more than 10 times faster on Pd Au catalysts, consistent with findings for the industrial process. 46 This promotion was initially assigned to a ligand effect where the weaker binding between reactants with the alloy surfaces (for example weakly bound monodentate acetate rather than strongly bound bidentate acetate, weakly bound p-ethylene rather than di-s-ethylene) is expected to yield faster coupling; and the weaker binding between the product VA with the alloy surfaces facilitates its desorption. 46 While it is clear that a ligand effect must play a role, studies by Chen et al. on Pd/Au(100) and Pd/Au(111) model catalysts revealed that ensemble effects play a more significant role in affecting VA formation rates. 35 Since VA formation requires coupling between chemisorbed acetate and ethylene, a correlated pair of Pd sites is needed. Considering the bond lengths of adsorbed ethylene and acetate species, the optimized distance between two active sites is 3.3 A. Separation between pairs of Pd monomers on Au(100) will be 4.08 A, while on Au(111) this distance will be 4.99 A, prohibitively long for coupling of these two reactive intermediates. The reaction rate on Pd/Au(111) is indeed much lower than that on Pd/ Au(100) as evidenced in Fig. 4(A). The bonding and relative distances involved between reacting species are shown schematically in Fig. 4(B). The pair of isolated Pd sites, while aiding in the formation of VA by providing the optimum required spacing for coupling of the surface acetate and ethylene species, was also proposed to suppress the formation of reaction by-products, such as CO and CO 2,thusimproving the overall selectivity CO oxidation. Both planar models and high surface area supported Pd are excellent catalysts for CO oxidation (CO + 0.5O 2 - CO 2 ). Similarity in rates on both these catalyst systems is generally regarded as strong evidence for the structure insensitivity of this reaction over late transition metals. 26 The situation is more complex for Au. While Au nanoparticles supported on certain oxides (especially the reducible ones, e.g., TiO 2 ) are specifically active for lowtemperature CO oxidation, bulk gold is inert. The reason appears easy to understand: as a Langmuir Hinshelwood type of reaction, chemisorbed CO must react with a chemisorbed active oxygen species to form CO 2. In most cases this active oxygen is atomic oxygen. The inertness of bulk gold is due to its inability to activate di-oxygen. Indeed, if atomic oxygen is pre-adsorbed onto planar Au, low-temperature CO oxidation does occur facilely. 47 It is therefore quite interesting to investigate CO oxidation over bulk Pd Au alloys. Presumably by adding Pd (which is capable of activating di-oxygen) to Au, one is able to form an active catalyst. This hypothesis was tested on AuPd(100), 5,6 Pd Au alloy thin films, and supported Pd Au particles. 40 Fig. 5(A) shows reaction data at 10 7 Torr of CO pressure over planar Pd and Pd Au nanoparticles grown on a thin TiO 2 layer deposited on Mo(110). 40 Clearly, while Pd alone is active, adding Au into Pd greatly inhibits CO oxidation in vacuum. When the Au Pd ratio reaches 1, the alloy is totally inert. An AuPd(100) single crystal also shows no activity under identical conditions. 5,6 However, when kinetic measurements were carried out at elevated CO pressures, the reaction data shown in Fig. 5(B) This journal is c The Royal Society of Chemistry 2012 Chem. Soc. Rev., 2012, 41,

6 Fig. 4 (A) Vinyl acetate (VA) formation rates (turnover frequencies, TOFs) as a function of Pd coverage on Au(100) and Au(111). The VA synthesis was carried out at 453 K, with acetic acid, ethylene, and O 2 pressures of 4, 8, and 2 Torr, respectively. The total reaction time was 3 hours. The error bars are standard deviations, based on background rate data. The two insets show Pd monomers and monomer pairs on the Au(100) and Au(111) surfaces. (B) Schematic for VA synthesis from acetic acid and ethylene. The optimized distance between the two active centers for the coupling of surface ethylenic and acetate species to form VA is estimated to be 3.3 Å. With lateral displacement, coupling of an ethylenic and acetate species on a Pd monomer pair is possible on Au(100) but implausible on Au(111). Figure adapted with permission from ref. 35. Copyright (2005) by the American Association for the Advancement of Science. reveal that AuPd(100) becomes orders of magnitude more active than pure Pd at temperatures below B500 K. The reaction kinetics shown in Fig. 5 raise three questions: (1) what causes CO oxidation reaction to turn on at elevated CO pressures on Pd Au? (2) why is the alloy surface much more active than pure Pd at elevated CO pressures and relatively low temperatures? (3) why does CO 2 formation roll over at temperatures higher than B450 K over AuPd(100)? The answer to the first question is twofold. (i) O 2 does not dissociate on isolated Pd sites. This is easily proven by an O 2 temperature-programmed desorption experiment: on a Pd Au alloy surface with only isolated Pd surface sites, O 2 desorption due to recombination of chemisorbed atomic oxygen does not occur suggesting that O 2 does not dissociate during adsorption Fig. 5 (A) CO conversion as a function of reaction temperature over TiO 2 /Mo(110)-supported Pd and Pd Au alloy particles. Reaction was carried out at steady-state using a stoichiometric CO O 2 mixture at P CO = Torr. Note that kinetic data with different surfaces are shown with different symbols. Figure adapted with permission from ref. 40. Copyright (2010) by American Chemical Society. (B) Arrhenius plots of the CO 2 formation rate (in TOF) over AuPd(100) and Pd(110) with 16 Torr CO and 8 Torr O 2. Figure adapted with permission from ref. 6. Copyright (2009) by American Chemical Society. to this surface. 48 For the Pd Au alloy model catalysts used to acquire reaction data shown in Fig. 5(A), the lack of O 2 dissociation precludes the subsequent CO 2 formation reaction. (ii) At near-atmospheric CO pressures, a sufficient amount of Pd segregates to the surface and generates contiguous Pd sites which, in contrast to isolated Pd sites, are capable of dissociating O 2 to allow the CO 2 formation reaction to proceed (Fig. 5(B)). The driving force for this chemisorption induced segregation is the stronger binding of CO with Pd than Au. This phenomenon has been shown clearly by the PM-IRAS data displayed in Fig. 3. 5, Chem. Soc. Rev., 2012, 41, This journal is c The Royal Society of Chemistry 2012

7 Some knowledge of the CO oxidation reaction mechanism over Pt-group metals at near-atmospheric pressures is needed to answer the second question. Briefly, for the majority of late transition metals at relatively low temperatures, the metal surfaces are covered with a full layer of chemisorbed CO under typical reaction conditions. As such, the rate-limiting step is CO desorption which creates empty surface sites for O 2 chemisorption and dissociation. This reaction mechanism is fully supported by (1) reaction rates displaying +1 order dependence in O 2 pressures and 1 order dependence in CO pressures; and (2) the measured reaction activation energies close to the CO desorption activation energies. 25,26 The reaction kinetic data shown in Fig. 5(B) indicate much weaker binding energies for CO on a Pd Au alloy catalyst as compared to pure Pd. This is indeed the case proved by calculating CO binding energies on Pd and Au sites on AuPd(100) using the Clausius Clapeyron relation. At zero CO coverage, the CO heat of adsorption was found to be 69 kj mol 1 on Au sites and 84 kj mol 1 on Pd sites. The CO heat of adsorption on Au single crystal surfaces is typically very low, i.e., B50 kj mol 1 on step sites and less than 40 kj mol 1 on terraces. In contrast, the CO binding energy on pure Pd is as high as 150 kj mol 1. 6 These CO binding energies not only are consistent with reaction kinetics shown in Fig. 5(B), but also provide strong evidence for charge transfer from Pd to Au. Specifically, the CO binding energy of 69 kj mol 1 on Au sites on the AuPd(100) surface is substantially larger than CO binding energies on any pure Au surfaces. This is best explained as a ligand effect where charge transfer from Pd to Au enhances back-donation of electrons from Au to CO, thus increasing CO binding energies. Finally, we note that the surface composition of AuPd(100) surfaces vary dynamically under reaction conditions where higher temperatures cause more Au segregation to the surface (Fig. 3). At the same time, the low binding energies of CO on the alloy surfaces cause decreased CO residence times on the surface at high temperatures. The combination of these two factors explains the rollover of CO 2 formation over AuPd(100) at elevated temperatures. 3. (Supported) Nanoparticle catalysts 3.1. Synthesis Supported Pd Au catalysts can be easily synthesized using traditional impregnation or deposition precipitation of Pd and Au salts, either concurrently or sequentially, followed by calcination and reduction. 4,7 11,13,49,50 The major drawback is the lack of homogeneity of the formed catalysts in terms of particle size, composition and shape. For example, it is not uncommon to find Au alone particles, Pd alone particles and Pd Au alloy particles with different compositions on the same support. 7 11,13 Such catalysts may be useful for various applications but these are not ideal for fundamental catalytic chemistry and catalyst design type of studies. In the latter cases, metal nanoparticles with uniform composition and monodispersion are highly desirable. 51 Fortunately, during the past two decades or so, much progress has been made in nanomaterial synthesis such that the formation of size, shape and orientation specific monometallic nanoparticles for use in catalysis has become rather mature. 51 The most commonly used method is a colloid technique where nanoparticles are synthesized by the reduction of a metal precursor with a reducing agent in the presence of a protective agent to prevent aggregation. Unlike the monometallic case, the formation of bimetallic nanoparticles is more complex. By varying the formation methods, either core shell bimetallics or quasihomogeneous alloys are formed. Different techniques have been developed in the past for bimetallic nanoparticle synthesis including alcohol reduction, citrate reduction, polyol process, solvent extraction reduction, sonochemical method, photolytic reduction, decomposition of organometallic precursors, and electrolysis of bulk metal. 52 These nanoparticles can be used directly as catalysts for certain reactions on condition that the capping layer is porous and stable; or they can be grafted onto high-surface-area supports, followed by a capping layer removal step, as regular heterogeneous catalysts. Two points are addressed for this process: (1) colloid Pd Au bimetallic nanoparticles, as these are usually synthesized from Pd and Au chloride precursors, are negatively charged due to residual Cl. Adjusting the solution ph to make the support surfaces positively charged helps the grafting process. 12 (2) Removal of the capping layer while maintaining the nanoparticle dispersion, composition and size is a great challenge. Decomposing the capping layer at the lowest possible temperature appears to greatly inhibit sintering. 53 In the following, examples are given on size and structure control of Pd Au nanoparticles. Generally, for the most commonly used colloid techniques, particle sizes are affected by multiple factors including type and concentration of the metal precursors, reducing agents and protective agents (soluble polymers, surfactants, and organic ligands), formation temperature, etc. Two interesting size control methods are briefly introduced: (1) Pd Au can be synthesized as dendrimer-encapsulated nanoparticles (DENs). In this method, Pd and Au ions are first extracted from solution into the dendrimer interior via complexation with internal tertiary amines. Second, the metal ions are reduced with BH 4, and the resultant atoms subsequently coalesce to form zero-valent nanoparticles within the dendrimer templates. In this case, the dendrimer framework not only controls the sizes of the nanoparticles but also stabilizes them. 28,54 (2) Another interesting size managing approach is to synthesize Pd Au particles in reverse micelles. First, nanosized water droplets are dispersed in a continuous oil phase. Second, metal precursors and reductants are introduced into these water droplets to react. The sizes of the alloy particles formed are confined by the sizes of the water droplets. 52 As to composition/structure control, the two approaches typically used are simultaneous or sequential reduction of appropriate precursors. In the former, weaker reducing agents, for example polyol, lead to the formation of Au-core Pd-shell structures. This is because Au reduces more easily and provides a seed for the reduction of the Pd shell. The size and thickness of the core shell can be controlled by the ratio of Pd Au in the precursor solutions. Strong reducing agents (e.g., BH 4 ) result in the formation of a quasi-random distribution of Pd Au alloys. In sequential reduction, a monometallic core is synthesized This journal is c The Royal Society of Chemistry 2012 Chem. Soc. Rev., 2012, 41,

8 first and in the second step, the second metal is reduced onto the core surface. This method is used to generate Au-core Pd-shell and Pd-core Au-shell structures where the former are more common. However, if synthesis is carried out at higher temperatures, the enhanced mobility of atoms causes alloy formation instead. 54 Note that Au-core Pd-shell structures are also most often formed for supported Pd Au bimetallics. Besides the easier formation of Au 0 as nucleation seeds, the calcination reduction treatments, often used during the synthesis of such catalysts, also promote formation of a Pd-shell. 55 This is due to Pd diffusion to the surface during calcination since it is more readily oxidized than Au. Upon reduction, Pd can still remain enriched in the shell. It is to be noted, however, that any stable Au-core Pd-shell structure should be considered as kinetically stabilized because of (1) the lower surface free energy of Au and (2) the complete miscibility of the two metals Characterization Numerous techniques are used to characterize supported Pd Au bimetallic catalysts. General characterization methods, e.g., specific surface area, pore size and distribution, metal loading and dispersion, etc., are not discussed here. Instead, some of those used to acquire detailed structural information are briefly described in the following X-Ray diffraction (XRD). XRD is a routinely available technique that can be used to study Pd Au alloys. As fcc metals, Pd and Au have strong diffractions along the (111) and (200) directions between 2y of 30 to 501. The diffractions of Pd x Au y alloy phases fall in between the corresponding diffractions of the pure metals. This offers straightforward identification of alloy formation. Quantitative analysis of the XRD patterns allows the composition of the alloy phases to be determined. This can be done using Vegard s law, 13,15 or more accurately, the Rietveld refinement method. 7 Fig. 6 presents XRD patterns of SiO 2 supported pure Pd, Au and bimetallic Pd Au catalysts at various Pd Au ratios prepared by impregnation of the silica support with an alcohol solution of colloidal dispersion of the two metal particles, followed by Fig. 6 X-Ray diffractograms of monometallic and bimetallic catalysts after calcination at 673 K. The numbers in the sample notation refer to the relative weight percentages of the metals. Figure adapted with permission from ref. 13. Copyright (2003) by Elsevier. air calcination. 13 Several important points are worth mentioning: (1) the heterogeneity of the nanoparticles formed during Pd Au bimetallic synthesis including pure Pd phases (oxidized to PdO during calcination), pure Au phases and alloy phases. This is a good example demonstrating that the formation of uniform alloy nanoparticles remains a major challenge for supported catalysts. (2) The Pd x Au y alloy phases are more resistant to oxidation than pure Pd. This is evidenced by the fact that during calcination in air, pure Pd oxidizes to PdO while the alloy phases maintain metallic. This indicates that, for catalytic reactions in heavily oxidizing environments, Pd Au alloys might be a good option for preventing catalyst deactivation due to Pd oxidation. (3) The precursor Pd Au ratio clearly has profound effects on alloy formation where substantially more alloy forms for Au 75 Pd 25 compared to Au 25 Pd 75 precursor weight ratios X-Ray photoelectron spectroscopy (XPS). The surfacesensitive nature of XPS allows for determination of the near-surface composition of supported Pd Au bimetallic catalysts. 15,7 13,49 This includes (1) near-surface Pd Au ratios and (2) oxidation states of Pd and Au. Also XPS analysis of catalysts before and after catalytic reactions often yields rich indirect information regarding changes of the catalysts during reactions. Note that in ratioing Pd Au XPS signals (Pd 3d 3/2 and Au 4f 7/2 core-level features are generally used), atomic sensitivity factors of both elements must be included. 7 For oxide supported catalysts, charging is always a significant issue affecting accurate oxidation state determination. To circumvent this problem, generally binding energies (BE) are calibrated using internal standards (e.g., adventitious carbon C 1s at B285.0 ev). 7 In principle, the core-level BE change (as compared to BE of pure metals) can be used to verify alloy formation since, as discussed earlier, charge transfer does occur upon Pd Au alloy formation. However, for metal particles in the nano size ranges, surface metal atoms are substantially more undercoordinated than bulk atoms. In this case, final state effects (e.g. screening) can be more pronounced than initial state effects in determining binding energies in core level spectroscopy. Note that even when valence band spectroscopy fails to reveal any charge transfer between the alloy components, one can see changes in the core level BE. 3 Therefore one must be extremely careful in interpreting Pd Au alloy formation and charge transfer using core level XPS analysis. Finally, we note that traditional XPS is measured at pressures lower than B10 7 Torr; therefore, in situ applications at elevated pressures are not possible. In recent years an ambient pressure XPS technique (AP-XPS) has been developed. Although the highest pressure allowable at present is only B1 Torr, this technique has already shown the ability to monitor in situ dynamic changes of Pd Au alloy catalysts during CO oxidation X-Ray absorption spectroscopy (XAS). XAS requires synchrotron X-ray beamlines; therefore, widespread application of the technique is limited. Still, the technique offers unparalleled advantages for in situ applications under realistic high temperature and pressure reaction conditions due to the high energy and flux of synchrotron beams. XAS has two 8016 Chem. Soc. Rev., 2012, 41, This journal is c The Royal Society of Chemistry 2012

9 major uses: X-ray absorption near edge structure (XANES) is used to determine local coordination geometries and metal oxidation states; and extended X-ray absorption fine structure (EXAFS) is used to identify neighboring atoms, interatomic distances, and coordination numbers. Lambert and co-workers 17 were among the first to use EXAFS and XANES to study colloidal Pd Au nanoparticles. More work has been done more recently by other researchers. 53,54,57,58 Perhaps the best use of EXAFS in Pd Au nanoparticles characterization is the measurement of partial nearest-neighbor coordination numbers (CN) n PdPd, n PdAu, n AuPd and n AuAu obtained concurrently for data collected at Pd and Au absorbing atom edges. The total CN for Pd is given by n PdM = n PdPd + n PdAu, whereas the total CN for Au is n AuM = n AuPd + n AuAu. These values are important in distinguishing core shell or random alloy structures. For example, if Pd segregates to the surface while Au stays in the core of the nanoparticles, n PdM will be smaller than n AuM. This is because atoms at the surface have fewer neighbors than those in the core. On the other hand, if n PdM is very close to n AuM, a quasirandom alloy structure is likely. One might expect that EXAFS could be a powerful technique for in situ monitoring of dynamic composition changes of Pd Au particles by measuring the CNs described above, although not much work has been done so far in this regard. Note also that the total coordination numbers also provide a good estimate of particle sizes. The other significant use of EXAFS is the determination of Pd Pd, Pd Au and Au Au bond lengths. This also yields important information regarding the short-range environments around Pd and Au atoms. Pd Pd bond lengths for bulk Pd are Å and Au Au bond lengths for bulk Au are Å. For monometallic Pd and Au nanoparticles, these bond lengths are slightly smaller. In Pd Au alloys, the Pd Au bond lengths are rather close to Au Au bond lengths and elongation of Pd Pd bond lengths also occurs. As has been discussed earlier, this enhances the atomic-like character of Pd atoms and weakens binding with reactants. The situation is more complex for core shell structures. For Pd-core Au-shell structures, Pd Pd bond lengths stay rather close to pure Pd 53 while for Au-core Pd-shell structures, the Pd shell adopts the bond length of the Au core up to a shell thickness of B1 nm. 14,54 Summarily, Pd Pd bond lengths are more informative than Pd Au and Au Au bond lengths for structure determinations. Generally speaking, Pd Pd bond lengths close to pure Pd suggest Pd core structures or a high percentage of Pd-only nanoparticles, whereas Pd Pd bond lengths close to pure Au suggest alloy or Pd-shell bimetallic structures. In XANES studies, even simple comparisons of the spectra for the nanoparticles with those for pure Pd and Au can yield useful qualitative information regarding alloying and charge transfer. For example, normalized derivatives of Pd K-edge XANES spectra give rise to a broad maximum at the edge energy, E 0, attributed to weakly allowed 1s 4d transitions that also reflect the density of unoccupied states in the Pd d band. Comparing the spectra between pure Pd and Pd Au colloid samples led Lambert and co-workers to conclude that Pd atoms go from a Pd-like environment in the core shell structure to an alloy phase upon annealing. 17 Scott and co-workers compared XANES spectra at the Au L III -edge between pure Au and Pd Au nanoparticles and observed a reduction of the white line (the first feature after the edge jump) with increasing Pd content. Since the intensity of the white line depends on the number of d-holes in the Au atoms, this finding can be rationalized by the enhanced filling of the Au d-band in the PdAu alloys via charge transfer from the Pd s-band (and perhaps p-band) to the Au d-band Transmission electron microscopy (TEM). TEM is a powerful and mature technique for studying bare and supported metallic nanoparticles. The most common low magnification mode (bright field) is used to sample many particles simultaneously to obtain particle size distributions. In the high magnification mode (i.e., high-resolution TEM, HRTEM), single particles can be analyzed at an atomic level to obtain information such as faceting and lattice spacing of the nanoparticles. Commonly, TEM instruments are equipped with an energy dispersive spectrometer (EDS) detector for qualitative elemental analysis. TEM has been used extensively for studies of Pd Au nanoparticles. 4,7 11,14,15,46,53 56,60 Two examples are given below. Fig. 7 presents scanning TEM-EDS of large particles of Au Pd (2.5 wt% Au 2.5 wt% Pd) catalysts supported on carbon, TiO 2 and Al 2 O 3 calcined at 400 1C. 10,55 Interestingly for Au Pd/C, Au and Pd maps cover identical areas and the RGB reconstructed map is homogeneous indicating formation of a homogeneous Pd Au alloy. In contrast, for Au Pd/TiO 2 and Au Pd/Al 2 O 3, Pd maps occupy larger areas than Au maps and the RGB maps are apparently non-uniform indicating formation of Pd-shell Au-core structures. In another study, Ferrer et al. used energy-filtered TEM and scanning TEM-EDS line-scanning techniques to image three-layer core shell structures in Pd Au nanoparticles. 59 It is emphasized that such structural details cannot be obtained using the other techniques described above. Other commonly used techniques to characterize Pd Au nanoparticles are UV-Vis spectroscopy 12,14,28,52,54,59 and Fig. 7 Montage of high-angle dark-field (HAADF) imaging (column 1), Au map (column 2), Pd map (column 3) and RGB reconstructed overlap map (column 4) [Au blue: Pd green] for calcined AuPd/C (row 1), calcined AuPd/TiO 2 (row 2) and calcined AuPd/Al 2 O 3 (row 3). Note that the calcined AuPd particles on TiO 2 and Al 2 O 3 supports show a Au richcore Pd-rich shell morphology, whereas calcined AuPd particles on activated C are homogeneous alloys. Figure adapted with permission from ref. 10. Copyright (2008) by the Royal Society of Chemistry. This journal is c The Royal Society of Chemistry 2012 Chem. Soc. Rev., 2012, 41,

10 FTIR (especially DRIFTS). 13,16,46,49 These, however, are not covered in this article. 3.3 Examples of catalytic reactions Direct H 2 O 2 synthesis from H 2 and O 2. Hydrogen peroxide is an important green oxidant that is used in many large scale processes such as bleaching and as a disinfectant. Current industrial synthesis of H 2 O 2 utilizes a sequential hydrogenation oxidation route of an alkyl anthraquinone. The process is only economical at large scale and at high product concentrations. However when used, H 2 O 2 is often required on a much smaller scale and at lower concentrations. Direct small scale production of H 2 O 2 at the site where it is used is highly desirable in these areas. Due to its superior activity in hydrogenation, Pd was the catalyst of choice for many initial investigations. In recent years, Hutchings and co-workers have made major contributions in this area by discovering that supported Pd Au catalysts have higher activity and selectivity for this reaction. 8 11,55,60 In the following the atomic origin of this promoting effect is addressed. First of all, one has to realize that direct H 2 O 2 formation from H 2 and O 2 is not an oxidation reaction, but rather a reduction reaction. This is understood by the fact that an H H bond breaks while an O O bond is maintained during the reaction. The fundamental difficulty of the direct route is that H 2 O 2 is unstable with respect to both hydrogenation and decomposition while the non-selective combustion product, H 2 O, is thermodynamically much more stable. This complexity is shown schematically in Fig Clearly, switching off the combustion path (that is, preventing O O bond cleavage) is of vital importance in enhancing H 2 O 2 selectivity. As is wellknown, Pd is an excellent hydrogenation catalyst; unfortunately for this reaction, it is also an excellent oxidation catalyst. As such, a Pd-alone catalyst cannot achieve very high H 2 O 2 selectivity. The situation is drastically different for Pd Au catalysts. As has been discussed in detail in Section 2, adding Au to Pd dilutes surface Pd concentrations such that contiguous Pd sites disappear and only isolated Pd sites exist at sufficiently high Au coverage. Significantly, isolated Pd sites do not catalyze O 2 dissociation. 5,6,48 On the other hand, the structure insensitivity for hydrogenation reactions over Pd-based catalysts indicates that isolated Pd sites are still able to activate H 2. 2,3 Therefore, by taking advantage of this ensemble effect, one is able to tune the catalytic properties of Pd Au catalysts to dramatically enhance H 2 O 2 selectivity. The reaction results obtained by the Hutchings group are fully consistent with this ensemble effect argument. For Pd Au catalysts supported on Al 2 O 3,TiO 2 and carbon, the Pd Au/C catalyst shows much higher selectivity than the other two. This is because homogeneous alloys form on C while Pd-shell Au-core bimetallics form on Al 2 O 3 and TiO 2 (Fig. 7). 10,55 Clearly, on the alloy surface Pd atoms are better isolated than those on Pd-shell Au-core surfaces. By using acids to treat the Pd Au/C catalysts, these authors found enhanced gold dispersion by generating smaller Pd Au nanoparticles. Again, better Pd isolation is fully expected upon enhanced Au dispersion. Indeed, side reactions can be almost completely switched off following acid treatments Hydrodesulfurization (HDS) reaction. Noble metal catalysts are widely used in hydrodesulfurization of petroleum feed stocks. Pd is a good HDS catalyst although it tends to be poisoned by the sulfur present in the feed. Venezia et al. studied HDS of a model compound, dibenzothiophene (DBT), over Pd/SiO 2 and AuPd/SiO 2 catalysts. 13 The reaction results shown in Fig. 9(A) clearly demonstrate the beneficial effects of alloying, with Pd Au bimetallics showing higher Fig. 8 Schematic of the reactions and the corresponding reaction heats during the direct synthesis of H 2 O 2. Note that H 2 O 2 is unstable with respect to both hydrogenation and decomposition, and the nonselective combustion of hydrogen is a facile competing reaction. Figure adapted with permission from ref. 55. Copyright (2008) by the Royal Society of Chemistry. Fig. 9 (A) Total DBT conversion obtained at 593 K as a function of gold content. (B) X-ray diffractograms of monometallic Pd and bimetallic Au 50 Pd 50 catalyst: (a) after calcination at 673 K; and (b) after HDS of thiophene. Figure adapted with permission from ref. 13. Copyright (2003) by Elsevier Chem. Soc. Rev., 2012, 41, This journal is c The Royal Society of Chemistry 2012