Nanocatalysis 1588830624

Citation preview

Encyclopedia of Nanoscience and Nanotechnology

www.aspbs.com/enn

Nanocatalysis S. Abbet, U. Heiz University of Ulm, Ulm, Germany

CONTENTS 1. Introduction 2. Chemical Reactions on Point Defects of Oxide Surfaces 3. Chemical Reactions and Catalytic Processes on Free and Supported Clusters 4. Chemical Reactions Induced by Confined Electrons 5. Summary Glossary References

1. INTRODUCTION In the Middle Ages, Jabir Ibn Haiyan, known by the name of the alchemist Geber, paved the way to modern catalysis, and in the beginning of the 19th century, Berzelius defined for the first time the term catalysis [1]. Fritz Haber introduced the important catalytic formation of ammonia at the end of the 19th century [2]. This process allowed the synthesis of fertilizer and explosives on a large scale, and accelerated industrialization in the western world. Today, the introduction of fertilizers still leaves traces in nature, as manifested in the sudden increase of annual rings in old trees. Most catalytically active materials are obtained by trial and error. Since 1960, however, a molecular understanding of catalytic processes has been emerging. The characterization of the active parts in a catalyst is mainly influenced by the resolution of the analytical instruments in use. Today, nanoprobes are capable of seeing single atoms, and thus, examples are emerging where very small clusters are responsible for the catalytic action. Nellist and Pennycook reported that clusters of two and three Pt atoms exist in industrial, naphthareforming catalysts [3]. Thus, catalysts in the nanoscale may have been present since the time of Geber, but could not have been detected with the analytical schemes of those days. Here, we define nanocatalysis when two important conditions are fulfilled. First, the valence electrons of the active part of a nanocatalyst are highly confined, leading to physical and chemical properties nonscalable from bulk properties. ISBN: 1-58883-062-4/$35.00 Copyright © 2004 by American Scientific Publishers All rights of reproduction in any form reserved.

This condition is true for clusters/particles in the nanometer length scale or smaller. Second, nanocatalysts are designed in a controlled manner. The exploration of material properties in the nonscalable regime has profound consequences. Atomic clusters, for example, exhibit unique size-dependent electronic [4, 5], magnetic [6, 7], and chemical properties [8] that differ from those of bulk materials. In contrast to supported particles of larger size or extended solid surfaces, small clusters adsorbed at specific sites of a support material change their intrinsic properties to a large extent, in particular when charging of the cluster occurs. Nanocatalysts exhibit remarkable quantum size effects and structural fluxionality. This opens new avenues for the atom-by-atom design of nanocatalysts whose chemical activity, specificity, and selectivity can be tuned by controlling the cluster size, through the incorporation of impurity atoms, and via manipulation of the strength of the cluster–support interaction, the degree of charging of the cluster, and by changing their magnetic properties. In a completely different approach, catalytic reactions can be induced by nanodevices, where the energy of the localized electrons inducing the reaction can be tuned by changing the potential of the device. An example is the use of an STM tip in close contact with the reactants. In this review, we present some studies on model systems for nanocatalysts. With such systems, the complexity of working nanocatalysts can be reduced, and in many cases, simple reaction steps can be identified. It has to be pointed out that, in this review, we use the expression “catalysis” when a chemical reaction is catalyzed by a nanoscale system. We are aware, however, that none of the systems/methods presented here was tested to withstand the harsh conditions of real catalysis. We feel, though, that such studies are extremely important for developing guiding principles in nanocatalysis.

2. CHEMICAL REACTIONS ON POINT DEFECTS OF OXIDE SURFACES In a recent review, Pacchioni presented a detailed classification of point defects on the surfaces of oxide materials [9]. He described at least four majors kinds of irregularities: lowcoordinated sites, divacancies, impurity atoms, and surface

Encyclopedia of Nanoscience and Nanotechnology Edited by H. S. Nalwa Volume 6: Pages (161–177)

162 vacancies. The last category includes cation vacancies, usually called V centers, as well as oxygen vacancies, which are called color centers or F centers (from the German word for “color”—“Farbe”). In a broader sense, an F center is defined as an electron trapped in an anion vacancy, and was first introduced to explain chemical- and radiation-induced coloration of alkali halides. In bivalent oxides, like MgO and CaO, two types of F centers exist. The vacancy containing two electrons is called the F center, and the one containing only one electron is labeled the F+ center. In both vacancies, the electrons are confined by the Madelung potential of the crystal. Thus, in a first approach, the F/F+ centers can be described as two/one electrons moving in a quantum box of nanometer length scale, that is, F/ F+ centers are systems with highly confined electrons, and with our definition, they are examples of nanocatalysts. The unique chemical activity of F centers has been demonstrated recently. In a combined theoretical and experimental study, it was shown that the heterolytic breaking of methanol on MgO thin films is catalyzed by surface oxygen vacancies [10]. In these films, the oxygen vacancies are generated by changing the preparation method [11], for example, the Mg evaporation rate and the oxygen background pressure. In this way, two kinds of films are prepared: defect-poor (Mg evaporation rate: 0.3–0.5 ML/min, O2 background: 5 · 10−7 torr) and defect-rich (Mg evaporation rate: 2–5 ML/min, O2 background: 10−6 torr) films. Both films are annealed to 1200 and 840 K during 10 min, respectively. Auger electron spectroscopy (AES) measurements show a one-to-one stoichiometry for magnesium and oxygen and the absence of any impurity [12]. Typical thicknesses are about ten monolayers, as determined by AES peak intensities [12] and by X-ray photoemission (XPS), using the intensity attenuation of the Mo 3d core level with increasing film coverage [13]. Both films have also been studied by electron energy loss spectroscopy [13]. In contrast to defect-poor films, which are characterized by a loss at about 6 eV in the EEL spectra (Fig. 1a, in good agreement with previous studies on MgO(100) single crystals [14]), the EEL spectra of defect-rich films exhibit characteristic losses between 1 and 4 eV, lying within the MgO bandgap (Fig. 1a). Similar loss structures have been observed before [15], and according to first-principle calculations using large cluster models, they have been attributed to transitions, characteristic of neutral surface F centers in various coordinations on flat terraces, at steps, and at kinks [16]. The density of these oxygen vacancies is estimated to be larger than 5 × 1013 /cm2 . The interaction of methanol with the defect-poor and defect-rich films was studied using thermal desorption spectroscopy (TDS) (Fig. 1c). For both films, the desorption of physisorbed methanol at around 180 K is most dominant. On the defect-poor films, small amounts of chemisorbed methanol desorb up to around 350 K. On defect-rich films, the desorption of chemisorbed methanol evolves in three distinct peaks at 200, 260, and 340K. A small reproducible feature is observed at around 500 K. Most important, H2 desorbs at 580 K only on defect-rich films. The corresponding infrared spectra taken at 90 K (insets of Fig. 1c) confirm the presence of mainly physisorbed CH3 OH with the typical vibrational band for the OH group at 3285 cm−1 , bands of the symmetric C–H stretch (2930 cm−1 /2828 cm−1 ) and

Nanocatalysis

Figure 1. (a) EEL spectra of thin defect-poor and defect-rich MgO(100) films grown on Mo(100) at different experimental conditions (see text). A–D are losses, which are attributed theoretically to transitions characteristic of neutral F centers on MgO. (b) Model of an oxygen vacancy at a terrace of a MgO(100) surface with chemisorbed CH3 –OH (CH3 O− –H+ . (c) Thermal desorption spectra of CH3 OH and H2 on defect-poor and defect-rich MgO(100) films. Note the desorption of H2 at 580 K for defect-rich films. The insets show FTIR spectra recorded at 90 K for adsorbed CH3 OH on both defect-poor and defectrich films.

the C–H bending (1475 cm−1 ) modes of the CH3 group, as well as the C–O vibrational mode at 1080 cm−1 /1050 cm−1 . For defect-poor films, most of the intensity of the vibrational bands disappears between 180 and 200 K, consistent with the desorption observed at 180 K in the TDS. Up to temperatures of about 400 K, small bands are observed for the CH3 – and C–O bands. Their intensity is decreasing with temperature. These bands are attributed to chemisorbed methanol. As for the defect-rich films, the evolution of the IR spectra with temperature is consistent with the corresponding TDS. Physisorbed methanol desorbs before 200 K (disappearance of the OH band), and the more intense vibrational features of chemisorbed methanol are unambiguously detected up to 360 K. At higher temperatures, a clear peak is observed at 1070–1085 cm−1 [10]. It is interesting to note that the disappearance of this band correlates with the desorption of H2 from the surface at around 580 K, and therefore this peak is attributed, in accordance with theoretical studies, to a proton trapped in the cavity of an F center. The calculations show that F centers easily dissociate methanol, giving

Nanocatalysis

a CH3 O group and an H+ ion adsorbed into the F center. This adsorbed proton is strongly bound as the dissociation of a neutral H atom, Fs /H+ → Fs+ + H• , costs 4.1 eV. This represents a crude estimate of the barrier to overcome in order to observe H2 desorption from the surface. Once the H atom is detached from the Fs center, it will rapidly diffuse on the surface. In fact, the binding of an H atom to an O5c ion on a terrace is about 0.5 eV. Diffusion will eventually lead to recombination with a second H to form H2 ; H2 is weakly bound to the MgO terrace sites, and at 580 K will immediately desorb. The adsorbed hydrogen gives rise to vibrations of strong intensity at 830–950 cm−1 when isolated, and to an intense band at 1030 cm−1 in the chemisorbed complex shown in Figure 1b. Therefore, the stable species is assigned to H atoms incorporated into oxygen vacancies. In conclusion, this combined experimental and theoretical study of methanol adsorbed on MgO films with different defect density allows better identification of the surface sites responsible for the MgO reactivity. On the inert terrace sites, only physisorption is observed. Molecular chemisorption, activation, and heterolytic dissociation occur on irregular sites. The low-coordinated Mg–O pairs of ions located at edges and steps can lead to strongly activated, and even dissociated, methanol molecules. Adsorption of CH3 O and H+ fragments seems to be preferred over dissociation into − CH+ 3 and OH units. All of these species are stable on the surface for temperatures up to 350 K, and account for the TDS spectra of the defect-poor films. On defect-rich films (F centers), the O–H bond is selectively dissociated, resulting in the observed desorption of H2 at high temperature. Thus, these oxygen vacancy centers, called F centers, act as nanocatalysts.

3. CHEMICAL REACTIONS AND CATALYTIC PROCESSES ON FREE AND SUPPORTED CLUSTERS 3.1. Catalytic Processes on Free Metal Clusters Free clusters are ideal model systems to probe the influence of their intrinsic, size-dependent properties on the catalytic activity due to the lack of any support interactions. Free clusters are prepared by cluster sources [17], and only very low densities are obtained. They are highly unstable at normal conditions and, even under UHV conditions, exothermal catalytic reactions may lead to fragmentation without the presence of a buffer gas. Thus, free clusters may not become relevant for industrial applications. Nevertheless, they are important vehicles to gain a fundamental understanding of nanocatalysis. The following principle is often used for producing free clusters. The material of interest is evaporated by means of intense laser pulses [18], electrical discharges [19], or highly energetic inert ions [20]. A buffer gas thermalizes the produced ultrahot plasma or supersaturated atomic gas. The mixture expands through a nozzle into the vacuum upon supersonic expansion. By this process, the formed clusters are cooled to cryogenic temperatures, and generate welldefined molecular beams of neutral and charged clusters.

163 The reactivity of the clusters can then be studied by various experimental techniques, including fast flow reactor kinetics in the postvaporization expansion region of a laser evaporation source [21, 22], ion flow tube reactor kinetics of ionic clusters [23, 24], ion cyclotron resonance [25, 26], guidedion-beam [27], and ion-trap experiments [28–30]. Which of these techniques is applied depends on the charge state of reactants (neutral, cationic, anionic), on whether the clusters are size-selected before the reaction zone, on single or multiple collisions of the clusters with the reactants, on the pressure of a buffer gas if present, and on the temperature and collision energy of the reactant molecules. A representative experimental setup may described in the following way [28]. The cluster anions are produced by a sputter source, and are extracted into a helium-filled quadrupole, where they are cooled down to room temperature and collimated to the axis of the ion optics. A second quadrupole selects a single cluster size. The cluster ions are then transferred with a quadrupole ion guide into an octopole ion trap, which can be continuously cooled from TT = 350 to 20 K. The potentials on the entrance and exit lenses can be switched in order to fill the trap, store the ions in the trap, and finally extract product ions for mass analysis. Through collisions with a buffer gas (helium or argon) in the trap, the clusters are already thermalized within a few milliseconds. Different reactants can be added to the ion trap under known partial pressures pR . Typically, the number of reactant molecules is orders of magnitude higher than the number of clusters, and thus it can be taken as constant. After reaction time tR , the charged products are extracted from the trap, and are analyzed by a quadrupole mass spectrometer as a function of tR , pR , and TT . With such experimental techniques, the interaction of reactants, for example, CO and O2 molecules, or even complete catalytic cycles were observed on small Au− n anions [30, 31]. By varying the reaction temperatures and measuring the kinetics of the processes, the energetics of complete reaction mechanisms could be obtained and compared with ab initio simulations [29, 32]. The reactivity of Au− 2 toward O2 or CO, as well as the coadsorption of both molecules, revealed an interesting and unexpected selectivity. O2 only reacts with even-n Au− n clusters, while odd-n clusters and Au− 16 are inert. In addition, the reactive clusters adsorb just one O2 molecule. The relative reactivities correlate well with measured electron affinities of neutral clusters, suggesting a mechanism involving the adsorption of O2 as a one-electron acceptor [33, 34]. The reactivity of gold cluster anions with CO also shows a distinct size selectivity, although less pronounced than the one with O2 [35]. In this reaction, the most reactive cluster sizes are Au11 , Au15 , and Au19 . Most reactive cluster sizes readily adsorb several CO molecules in the saturation limit. Full geometric coverage, for example, one CO per surface gold atom, is, however, never reached. On real catalysts and single crystals, competitive adsorption is usually observed, meaning that, in many examples, CO poisons the active catalysts. This is not true for small Au− n clusters. On the contrary, for distinct cluster sizes, cooperative adsorption is observed, where the presence of preadsorbates enhances the adsorption probability of subsequent adsorbates. As examples, for Au− 3 , preadsorbed CO enables the

Nanocatalysis

164 adsorption probability of O2 ; Au− 2 is inert for the adsorption of CO at around room temperature; the presence of preadsorbed O2 , however, makes the adsorption possible at this temperature [30]. A similar effect is observed for Au− 10 , where O2 enhances the adsorption of CO by more than a factor of 5. Finally, Au− 6 reveals even an enhanced cooperative adsorption with increasing numbers of adsorbate molecules [31]. Let us now turn our attention to Au− 2 , where a full catalytic cycle could unambiguously be observed by measuring the kinetics of the process at different temperatures [29]. At room temperature, oxygen reacts by a straightforward association reaction mechanism with the dimer anion, as determined from the measured product ion concentration as a function of the reaction time, and the negative dependence of the reaction rate with temperature: k

Au2 + O2 −→ Au2 O2

(1)

As mentioned above, no adsorption of CO is observed at around room temperature. When both reactants O2 and CO are introduced into the ion trap, the reaction kinetics of Au− 2 changes drastically, as seen by the offset in the Au− 2 signal. This offset increases when the partial pressure of CO is augmented. In addition, at temperatures below 200 K, the intermediate with the stoichiometry Au2 (CO)O− 2 could be isolated (Fig. 2A and B). The ion stoichiometry clearly shows that CO and O2 are able to coadsorb onto an Au− 2 dimer. − From the kinetics of all observable ions, Au− 2 , Au2 O2 , and − Au2 (CO)O2 , measured under a multitude of different reaction conditions, the catalytic conversion of CO to CO2 could unambiguously be detected. The reaction mechanism that fulfills all of the prerequisites and fits all kinetic data measured under all of the different reaction conditions could be described by the reaction equations below: k1

− Au− 2 + O2 −−−−→ Au2 O2 k2

− −−−−→ Au2 O− 2 + CO ←−−−− Au2 COO2

(2a) (2b)

k3

k4

− Au2 COO− 2 + CO −−−−→ Au2 + 2CO2

(2c)

By studying these kinetics as a function of concentration and temperature (Fig. 2B), some important findings can be obtained. 1. Undoubtedly, O2 adsorption precedes CO adsorption in the catalytic reaction. This is further supported by the reaction kinetics of Au− 2 when solely O2 or CO are present in the trap: The adsorption of O2 is by at least a factor of 10 faster than the adsorption of CO molecules [36]. 2. These experiments revealed that varying the oxygen partial pressure inside the trap mainly affects the rate constant k1 in such a way that k1 rises linearly with increasing p(O2 ), and to a smaller extent, the rate constant k4 , which was attributed to the higher abundance of Au2 (CO)O− 2.

Figure 2. (A) Mass spectra of product ion distributions analyzed after trapping Au−2 for 500 ms inside the octopole ion trap filled with 0.02 Pa O2 , 0.05 Pa CO, and 1.23 Pa He. (a) At a reaction temperature of 300 K, only Au−2 and Au2 O−2 are detected. No further ion signals are observed at temperatures above 200 K. Cooling further down reveals an additional ion signal appearing at the mass of Au2 (CO)O−2 . Mass spectrum (b) shows the ion distribution at 100 K. (B) Product ion concentrations as a function of reaction time for three different reaction temperatures. (a) TT = 190 K; pO2  = 0 06 Pa; pCO = 0 07 Pa; pHe = 1 0 Pa. (b) TT = 150 K; pO2  = 0 04 Pa; pCO = 0 04 Pa; pHe = 1 0 Pa. (c) TT = 100 K; pO2  = 0 02 Pa; pCO = 0 03 Pa; pHe = 1 0 Pa. Open symbols represent the normalized experimental data. The solid lines are obtained by fitting the integrated rate equations of the catalytic reaction cycle (see text) to the experimental data. (C) Five optimized structures (a)–(e) of Au2 CO−3 , with bond lengths in angstroms. The relative stability of these structures is shown in Table 1. (a), (b), (c), and (e) are planar, and the two carbonate species (d) and (e) have C2v symmetry.

3. An increase in the carbon monoxide partial pressure results at room temperature in an increasing back reaction rate k2 in Eq. (2b), that is, the Au− 2 offset is increasing. Resolving the kinetics at lower temperatures at 190 K reveals a clearly positive dependence of k4 [in Eq. 2(c)] on p(CO), also correlated to the Au− 2 offset. 4. Finally, with decreasing temperature, the final equilibrium concentration of Au2 (CO)O− 2 increases, whereas the Au2 O− 2 concentration decreases (Fig. 2B). This is consistent with the decreasing rate constants k1 , k2 , and k3 with increasing temperature, and this negative temperature dependence is indicative of barrierless

Nanocatalysis

165

reaction steps, explained in the framework of the Lindemann theory [36, 37]. Most importantly, however, k4 increases with increasing temperature. Hence, an activation barrier is involved in the corresponding reaction [Eq. 2(c)]. Further details of the reaction mechanism were obtained by complementary theoretical simulations [32], revealing that the bonding of O2 to Au− 2 is characterized by partial electron transfer (0.4e) from the metal cluster to the antibonding -orbital of the oxygen molecule, making the oxygen molecule a superoxo-like species. The binding energy is 1.39 eV. The molecular adsorption is energetically clearly favored over dissociative adsorption. In agreement with the experimental findings, the molecular binding is a nonactivated process, that is, a barrierless reaction channel exists for the approaching oxygen molecule, leading to the described configuration. The binding energy of CO to Au− 2 is 0.91 eV. The stronger binding of O2 to Au− 2 explains the observation that high CO partial pressures are required before CO adsorption can compete with O2 adsorption as the first reaction step. Five structures corresponding to the mass of the complex Au2 (CO)O− 2 theoretically predicted, and the pertinent structural and energetic information are given in Figure 2C and Table 1. Structures (a) and (b) correspond to molecular coadsorption of CO and O2 to Au− 2 . From the two molecularly coadsorbed species, CO can readily (without barrier) bind to the end of the Au–Au axis [structure (a)], whereas a barrier of 0.2 eV was found for CO association from the gas phase to the Au–Au bridging site of structure (b). The barrier for forming (b) from (a) via diffusion of CO from the end of the complex to the Au–Au bridge is high, on the order of 0.9 eV. In both structures (a) and (b), the O–O bond is activated to a value typical to a superoxospecies (about 1.35 Å). Structure (c) is close to the stability of structure (a). It contains a reacted O–O–C–O group bound to the Au2 axis via the carbon atom. The O–O bond is activated to a superoxo state, and this species bears some resemblance to the gold–peroxyformate complex identified in the early experiments of gold atoms in cryogenic CO/O2 matrices [38]. The formation of this species was investigated via two routes: (1) a Langmuir–Hinshelwood (LH) mechanism that involves diffusion of CO from the bridging position in structure (b), and (2) an Eley–Rideal (ER) mechanism where CO(g) approaches the preformed Au2 O− 2 . The LH mechanism has a high activation barrier of 1.1 eV, and since the barrier is on the order of the activation energy for CO desorption from structure (b) (1 eV), formation of (c) Table 1. Relative stability, vertical electron detachment energy, binding energy of CO (with O2 preadsorbed) and binding energy of O2 (with CO preadsorbed), for the five structures (a)–(e) of Au2 CO−3 shown in Figure 2C. Structure (a) (b) (c) (d) (e)

E (eV)

VDE (eV)

BE(CO) (eV)

BE(O2 ) (eV)

2.80 2.94 2.82 0 1.04

2.82 3.32 3.82 4.67 3.38

0.93 0.78 0.91 3.72 2.69

1.34 1.20 1.32 4.14 3.10

from (b) is unlikely. Interestingly, mechanism (2) does not involve an activation barrier. By far the most stable structures corresponding to the mass of Au2 (CO)O− 2 are the two carbonate species (d) and (e). Structure (e) requires a preformed Au2 O− 2 where the molecular axes of Au2 and O2 lie parallel to each other. Since this structure of Au2 O− 2 is 1 eV less stable than the ground state discussed above (the binding energy of oxygen in this configuration is 0.39 eV versus the optimal binding energy of 1.39 eV), it is unlikely to be formed, and consequently, structure (e) does not play a relevant role in the catalytic cycle. On the other hand, formation of the most stable structure (d) by an ER mechanism with CO(g) inserted in the middle of the O–O bond in Au2 O− 2 requires a barrier of only 0.3 eV, which is easily overcome under the experimental conditions. The absence of the complex Au2 C2 O− 4 in the experimental mass spectrum, as well as the calculated high barriers for CO diffusion along the Au–Au axis, justify the considerations of only ER mechanisms for the final reaction step [Eq. 2(c)]. It is also worth noting that further intermediates such as Au2 O− and Au2 CO− 2 are not observed experimentally. This leads to only two possible scenarios for the final reaction step 2c: ER reactions of CO(g) with structures (c) or (d), and the reaction proceeds to completion, releasing 2CO2 molecules. The formation of CO2 from the reaction between CO(g) and structure (c) (Fig. 2C) involves a low barrier of 0.3 eV, resulting in the formation of a metastable Au2 CO− 2 complex, where CO2 is bound to Au2 via the carbon atom. However, the large heat of reaction (4.75 eV) from the formation of the first CO2 molecule is much more than that needed to overcome the binding energy (0.52 eV) of the remaining CO2 to Au− 2, readily facilitating its desorption from the metal cluster. The second scenario involves two branches [32]. First, thermal dissociation of CO2 from the carbonate (d) is endothermic by 1.12 eV, but produces highly reactive Au2 O− which reacts without a barrier with CO(g) to produce CO2 . Second, the ER reaction of CO(g) with (d) to produce CO2 involves a modest barrier of 0.5 eV, but readily releases two CO2 molecules since the remaining Au2 CO− 2 configuration, where CO2 is bound to Au2 via one of the oxygen atoms, is unstable. This is the more likely branch to occur in the experiments (100–300 K) described above. Thus, these studies showed that both experimental and theoretical results prove the existence of the full catalytic cycle. In another example [39, 40], it was demonstrated that gasphase platinum cluster anions, Pt− n (n = 3–7), efficiently catalyze the oxidation of CO to CO2 in the presence of N2 O or O2 near room temperature in a full thermal catalytic reaction cycle. At the end of the process, the intact cluster is regenerated, and each step is exothermic, and occurs rapidly at thermal energies. In these experiments, the produced Pt− n clusters are first thermalized by collisions with a buffer gas. When either O2 or N2 O is introduced downstream in the flow tube, Ptn O− and Ptn O− 2 ions are formed in rapid exothermic reactions [41]. With the introduction of CO into the gas cell, the Pt− n clusters reappear, as observed with mass spectrometry. From these experiments, it was concluded that neutral CO2 is stoichiometrically formed on the clusters, and that negligible fragmentation of the metal cluster for n ≥ 4 occurs. Further experiments revealed that, at low energies,

166 the cross section approaches the calculated collision limit, meaning that the reactions are quite efficient. When Ptn O− 2 ions are selected initially, sequential loss of oxygen atoms is observed to form two CO2 products, as shown by CO pressure dependence studies. This observation implies that O2 is dissociatively adsorbed on the metal cluster as oxygen atoms, rather than chemisorbed or physisorbed as molecular O2 . Two other observations support this conclusion: (1) Ptn O− 2 ions produced by reaction of the bare cluster with either O2 or N2 O show the same reactivity; and (2) collision-induced dissociation of Ptn O− 2 shows a loss of oxygen atoms with no O2 loss. The regeneration of Pt− n ions at low energies proves that a full catalytic oxidation cycle can be completed at near room temperature, in either a single-step or a two-step process. Reaction efficiencies for the oxidation of the first CO molecule were measured for Ptn O− or Ptn O− 2 (n = 3–6). The reaction efficiencies for a single collision are greater than 40% for n ≥ 4, so only a few collisions would be required for complete conversion. From this, it was concluded that, at near room temperatures, the gas-phase metal cluster anions are better catalysts than the supported catalysts used in current technology for automotive catalytic converters. These need to be heated to high temperatures, on platinum surfaces; temperatures of 400 ± 500 K are typically required for the oxidation of CO [42]. Both examples revealed unique and size-dependent catalytic properties of very small metal clusters. As mentioned in the Introduction, gas-phase clusters will probably never be important in real catalysis; however, such small clusters can be stabilized on support materials, and they may become relevant for future applications.

3.2. Chemical Reactions and Catalytic Cycles on Supported Clusters 3.2.1. Single Atoms on Oxide Surfaces F centers can act as nanocatalysts; however, few reactions are catalyzed by oxygen vacancies because only two confined electrons at the same energy level are present. In order to change these two parameters (number of confined electrons and energy levels) without modifying the size of the quantum box, the F center can be decorated with metal atoms to produce a third kind of nanocatalyst: supported atoms on oxide surfaces. The metal atoms are produced by a recently developed high-frequency laser evaporation source [12]. The positively charged ions are guided by home-built ion optics through differentially pumped vacuum chambers, and are size-selected by a quadrupole mass spectrometer (Extrel Merlin System; mass limits: 1000, 4000, 9000 amu). In these experiments, it is important to deposit only 0.5–0.8% of a monolayer atom (1 ML = 2 × 1015 clusters/cm2 ) at 90 K with low kinetic energy in order to land them isolated on the surface and to prevent agglomeration on the defect-rich MgO films. The presence of isolated atoms is confirmed experimentally and theoretically. Experimentally, nickel atoms, dimers, and trimers are used as it is well known that they form stable metalcarbonyls. This carbonyl formation of small deposited Nin (n = 1–3) was studied by exposing the deposited clusters to carbon monoxide. Mass

Nanocatalysis

spectrometry experiments showed that the nuclearity of the formed Nin carbonyls (n = 1–3) is not changed. The absence of, for example, Ni2 (CO)x and Ni3 (CO)x after deposition of Ni atoms directly excludes agglomeration, whereas the absence of Ni(CO)4 after depositing Ni2 excludes fragmentation of the cluster, relevant for the experiments described in the next section [43]. Second, Monte Carlo simulations revealed that, under such experimental conditions, for example, cluster flux (∼109 cm−1 ), cluster density (∼1013 cm−1 ), and defect density (∼5 · 1013 cm−1 ) on the MgO(100) films, less than 10% of the atoms coalesce when migrating to the trapping centers [44]. To characterize the model catalysts in detail, it is, however, necessary to identify the adsorption sites of the clusters. One possibility is studying the adsorption of probe molecules on the deposited clusters by means of TDS and Fourier transform infrared (FTIR) spectroscopies. As an example [45], CO desorbs from the adsorbed Pd atoms at a temperature of about 250 K, which corresponds to a binding energy Eb of about 0.7 ± 0.1 eV. FTIR spectra suggest that, at saturation, two different sites for CO adsorption exist on a single Pd atom. The vibrational frequency of the most stable, singly adsorbed CO molecule is 2055 cm−1 . Density functional cluster model calculations have been used to model possible defect sites at the MgO surface where the Pd atoms are likely to be adsorbed. CO/Pd complexes located at regular or low-coordinated O anions of the surface exhibit considerably stronger binding energies, Eb = 2–2.5 eV, and larger vibrational shifts than were observed in the experiment. CO/Pd complexes located at oxygen vacancies (F or F+ centers) are characterized by much smaller binding energies, Eb = 0 5 ± 0 2 or 0 7 ± 0 2 eV, which are in agreement with the experimental value. Such comparisons therefore identify adsorption sites of the clusters on the MgO surface; particularly in the examples shown below, these trapping centers are shown to be oxygen vacancies. After characterizing this type of nanocatalyst consisting of metal atoms trapped on oxygen vacancies, simple reactions like acetylene polymerization or CO oxidation can be studied by means of temperature-programmed reaction studies and infrared spectroscopy. For cyclotrimerization on Ag, Pd, and Rh atoms and CO oxidation on Pd atoms, the nanocatalysts were exposed at 90 K, using a calibrated molecular beam doser, to about 1 Langmuir (L) of acetylene or 1L of 18 O2 and 13 CO, respectively. In a temperature-programmed reaction (TPR) study, catalytically formed benzene (C6 H6 ), butadiene (C4 H6 ), and butene (C4 H8 ) or 13 C18 O16 O molecules were detected by a mass spectrometer, and were monitored as a function of temperature. It is interesting to note that the reactants (13 CO, 18 O2 , C2 H2 ) are only physisorbed on the MgO, and desorb at temperatures lower than 150 K in all three cases, for example, before the reaction takes place. In addition, the interaction of the product molecules with the model catalysts is weak, as described in [45, 46]. The thermal stabilities of the deposited Pd and Rh atoms were investigated by using Fourier transform infrared spectroscopy and the probe molecule CO. It was found that Pd starts to migrate at around 300 K [45], whereas Rh is stable up to 450 K [47]. Surprisingly, and in contrast to single-crystal studies [48], a single Pd atom already catalyzes the cyclotrimerization

Nanocatalysis

reaction, and the benzene molecule (C6 H6 ) is desorbing at 300 K (Fig. 3a). No other product molecules of the polymerization reaction (e.g., C4 H6 , C4 H8 ) are observed. Thus, this model catalyst is highly selective for the polymerization reaction. We note that, on a clean MgO(100) surface,

Figure 3. (a) Temperature-programmed reaction (TPR) spectra of C6 H6 formed on Ag, Pd, and Rh atoms deposited on defect-rich MgO thin films grown on Mo(100) surfaces. As a comparison, the same experiment was performed on a clean defect-rich MgO film. Also shown is the calculated (C4 H4 (C2 H2 /Pd1 /F5c intermediate of the cyclotrimerization reaction on Pd atoms adsorbed on an F center of the MgO(100) surface. (b) TPR spectra of CO2 from a defect-rich MgO thin film and deposited Pd atoms after exposure to O2 and CO (lower and upper spectrum), as well as from deposited Pd atoms after exposure to CO and O2 . Calculated precursors of the two observed mechanisms (I) and (II) are also depicted.

167 none of the products is formed (Fig. 3a). This surprising result can be rationalized by theoretically studying a palladium atom adsorbed on different MgO sites represented by a cluster of ions embedded in an array of point charges [49]. First, a Pd atom was adsorbed on a five-coordinated oxygen ion on the MgO(001) terrace O5c ; the binding energy is about 1 eV. It was found that, indeed, the Pd(C4 H4 ) complex, is formed, but the third acetylene molecule is not bound to the complex, and therefore this configuration is catalytically inactive. Other bonding sites for the adsorption of Pd atoms are considered. On four-coordinated step or three-coordinated corner oxygen sites, O4c and O3c , respectively, the Pd atom binds slightly stronger, with an energy of 1.2–1.5 eV; in addition, the atom is more reactive. However, on both O3c and O4c sites, the third C2 H2 molecule is only weakly bound or even unbound to the Pd(C4 H4 ) surface complex, with the binding energy smaller than the activation energy of the formation of C6 H6 . Thus, Pd atoms adsorbed on Onc sites cannot explain the observed activity. This is consistent with the results of the study on the adsorption properties of CO on Pd1 /MgO [45], as the experimental results have been rationalized in terms of Pd atoms, which are stabilized on oxygen vacancies (F centers) in neutral or charged states, Fs or F+ s , respectively. The interaction of a Pd atom with the F center is much stronger, 3.4 eV, which makes these centers good candidates for Pd binding. On F+ s centers, binding energies of about 2 eV have been computed [45]. The presence of trapped electrons at the defect site results in a more efficient activation of the supported Pd atom. In fact, the complex (C4 H4 )(C2 H2 )/Pd1 /F5c , Figure 3a, shows a large distortion and a strong interaction of the third C2 H2 molecule. These results indicate that F and F+ centers can act as basic sites on the MgO surface, and turn the inactive Pd atom into an active catalyst. Notice that the supported Pd atoms on defect sites not only activate the cyclization reaction, but also favor benzene desorption, as shown by the very small (C6 H6 )/Pd1 /F5c adsorption energy. The complete reaction path for this specific nanocatalyst has been calculated [49]. The first barrier of the reaction path is the one for the formation of the intermediate Pd(C4 H4 ), and it is 0.48 eV only. The formation of the C4 H4 intermediate is thermodynamically favorable by 0.82 eV. On (C4 H4 )/Pd1 /F5c , the addition of the third acetylene molecule is exothermic by 1.17 eV, leading to a very stable (C4 H4 )(C2 H2 )/Pd1 /F5c intermediate (Fig. 3a). To transform this intermediate into benzene, one has to overcome a barrier of 0.98 eV. The corresponding energy gain is very large, 3.99 eV, and mainly related to the aromaticity of the benzene ring. Once formed, C6 H6 is so weakly bound to the supported Pd atom that it immediately desorbs. Thus, the reaction on Pd/F5c is rate-limited in the last step, the conversion of (C4 H4 )(C2 H2 ) into C6 H6 . This is different from the Pd(111) surface where the rate-determining step for the reaction is benzene desorption. The calculations are consistent with the experimental data. In fact, on Pd1 /F5c , the computed barrier of 0.98 eV corresponds to a desorption temperature of about 300 K, as experimentally observed, Figure 3a. On Pd (111) surfaces, the bonding of benzene is estimated to be ≈1.9 eV. This binding is consistent with a desorption temperature of 500 K, as observed for a low coverage of C6 H6 on Pd(111) [50].

Nanocatalysis

168 The electronic structure of palladium atoms, 4d 10 5s 0 , is unique, and may be responsible for this specific catalytic property for acetylene cyclotrimerization. This raises the question of whether other transition metal atoms are also reactive for this reaction. Results are shown for deposited Rh (4d 8 5s 1 ) and Ag (4d 10 5s 1 ) atoms. Ag atoms are almost unreactive (Fig. 3a); on supported Rh atoms, however, benzene is formed, and desorbs at around 430 K (Fig. 3a). For CO oxidation, it was first verified that the clean MgO(100) thin films are inert; for example, no 13 C16 O8 O was formed in a one-heating-cycle experiment after adsorbing 18 O2 and 13 C16 O (Fig. 3b) or vice versa [51]. When Pd atoms are trapped on the F5c , preadsorption of oxygen and subsequent saturation of CO leads to the formation of carbon dioxide, with desorption peaks at 260 K and at around 500 K (Fig. 3b). The existence of two desorption peaks in the TPR spectrum (Fig. 3b) suggests the presence of two different reaction mechanisms. Note that preadsorption of 13 CO suppresses the catalytic process as it poisons the nanocatalyst (Fig. 3b). Ab initio calculations showed that a single Pd atom strongly binds to the oxygen vacancy (binding energy of 3.31 eV), with a small amount of charge (0.15 e) transferred to the adsorbed atom. In comparison, the binding energy of Pd atoms to terrace oxygen sites is only 1.16 eV. The enhanced binding to the F5c is also reflected in the corresponding bonding lengths of 1.65 and 2.17 Å for MgO(F5c )–Pd and MgO–Pd, respectively. Binding of two CO molecules saturates the MgO(F5c )–Pd system; occupying the MgO(F5c )–Pd system with three CO molecules leads to spontaneous (barrierless) desorption of one of the molecules. In the most stable configuration, the two CO molecules are not equivalent; one CO binds on top, and the second adsorbs on the side of the Pd atom (this top-side geometry is similar to that shown in Fig. 3bI, but without the O2 ), and the total binding energy of the two CO molecules is 1.62 eV. To study the oxidation mechanisms of CO on MgO(F5c )–Pd, the system was optimized first with coadsorbed O2 and two CO molecules. Two stable geometric arrangements were found, with the most stable one shown in Figure 3bI where the CO molecules bind in a top-side configuration, and the O2 is adsorbed parallel to the surface on the other side of the Pd atom. The adsorbed O2 molecular bond is stretched and activated (1.46 Å compared to the calculated gas-phase value of 1.25 Å). In addition, a stable carbonate complex Pd(CO3 )(CO) was found (Fig. 3bII), whose binding energy is 4.08 eV larger than the aforementioned Pd(CO)2 (O2 ) complex. These complexes were identified by comparison of calculated and spectroscopically measured CO vibrational frequencies [52]. Two reaction mechanisms are proposed corresponding to the two CO2 peaks observed experimentally (Fig. 3b). At low temperatures, the two relevant precursors are shown in Figure 3bI and 3bII. Corresponding to the CO2 desorption peak at 260 K, the following reaction mechanism is proposed. In a competitive process, CO desorbs or is oxidized upon heating. The theoretically estimated activation energies of the two processes are 0.89 eV for desorption and 0.84 for oxidation. The formation of CO2 at higher temperatures (corresponding to desorption at around 500 K, Fig. 3b) involves decomposition of the Pd(CO3 )(CO) carbonate complex (Fig. 3bII).

This mechanism is observed in molecular dynamics simulations where the temperature is controlled to 500 K by Langevin dynamics.

3.2.2. Size-Selected Clusters on Oxide Surfaces By decorating F centers with atoms, it is possible to increase the number of confined electrons, but modifying the size of the quantum box only slightly. By using size-selected clusters instead of atoms, the confinement of the valence electrons is subtler, and the examples presented below reveal the possibility of tuning the efficiency and selectivity of chemical reactions atom by atom by simply changing the cluster size or doping the cluster with impurity atoms. Furthermore, a comparison of the experimental results with first-principles theoretical simulations provides insights into the physical factors and microscopic mechanisms that govern nanocatalysis. The metal clusters are produced by the same highfrequency laser evaporation source used to produce metal atoms. In this source, a cold He pulse (40 K) thermalizes the laser-produced plasma [12]. Subsequent supersonic expansion of the helium–metal vapor leads to cold clusters with a narrow kinetic energy distribution (Ekin ≤ 0 2 eV/atom). The positively charged clusters guided by ion optics through differentially pumped vacuum chambers and size-selected by a quadrupole mass spectrometer (Extrel Merlin System; mass limit: 1000, 4000, 9000 amu) are then deposited with low kinetic energy (