Immobilizing Metal Nanoparticles on Single Wall Nanotubes. Effect of

Phone: +81-92-802-6743. Cite this:J. Phys. ... Tailoring the nanoparticle–surface support interaction could provide a means to limit nanoparticle mo...
0 downloads 0 Views 9MB Size
Download PDF
Article pubs.acs.org/JPCC

Immobilizing Metal Nanoparticles on Single Wall Nanotubes. Effect of Surface Curvature Aleksandar Staykov,*,† Yuuki Ooishi,‡ and Tatsumi Ishihara†,‡ †

International Institute for Carbon-neutral Energy Research (WPI-I2CNER), Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395 Japan ‡ Department of Applied Chemistry, Faculty of Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395 Japan S Supporting Information *

ABSTRACT: One to several nanometer-size nanoparticles possess supreme catalytic activity for a variety of important synthetic reactions compared to larger particles and bulk surfaces. However, a significant drawback is the catalyst durability as small, active nanoparticles tend to merge to form larger, less active nanocolloids. Tailoring the nanoparticle− surface support interaction could provide a means to limit nanoparticle mobility and thus prevent aggregation. In this study, we demonstrate the stabilization of fine-metal nanoparticles on nanotube surfaces by manipulation of surface curvature. Systematic density functional theory calculations of a large variety of nanoparticle−nanotube complexes revealed that the nanoparticle−nanotube binding interaction depends on, and can be controlled by, the surface curvature. Thus, an effective mechanism is demonstrated for the immobilization of small metal clusters with high catalytic activity on support surfaces. Furthermore, we provide experimental verification of our theory by comparing the aggregation of palladium nanoparticles decorating carbon nanotube and graphene surfaces as a function of time. Our theoretical predictions and experimental observations provide fundamental understanding to the physics of nanoparticle−support interaction and demonstrate how tailoring the support geometry can improve the durability of high-performance nanocatalysts.

■

INTRODUCTION Nanoparticle catalysis has been intensively applied in the past decade in various areas of science and industry. With advances in Li−ion and Li−air batteries, Pd nanoparticles on porous MgO2 supports have been used as catalysts for O2 activation.1 The wide acceptance of fuel cells in automotive and domestic energy applications strongly relies on well-dispersed Pt nanoparticles on transition-metal oxide or porous carbon nanomaterial supports.2−5 A wide area of application has been the direct synthesis of hydrogen peroxide on Pd and Pd/ Au nanoparticles.6−10 The industrial synthesis of H2O2 is an energy-consuming process based on the sequential oxidation and reduction of anthraquinone. The direct synthesis on metal nanoparticles is a promising low-energy, small-scale alternative. Hydrogen peroxide direct synthesis strongly depends on the nanoparticles’ size and composition where the support plays a crucial role. An important field of nanoparticle catalysis is the electrochemical and photochemical splitting of water.11,12 Electrochemical water splitting is nowadays the main source for industrial synthesis of hydrogen together with natural gas dehydrogenation. Photochemical water splitting is the ultimate goal for clean, carbon-neutral energy. In both processes, often the actual water dissociation occurs on Pt, Rh, or Pd nanoparticle cocatalysts situated either on the electrode surfaces or on the light-harvesting materials. Gold nanoparticle catalysis in particular has attracted much recent interest in both © 2014 American Chemical Society

industrial and preparative synthetic chemical aspects as well as from a fundamental physical chemistry standpoint. Since the original discovery of catalytically active gold by Haruta, various reactions on gold nanoparticle surfaces have been investigated and described.13 Among these are the aerobic oxidation of CO, the oxidation of cyclohexane, H2 dissociation, CH4 oxidation, and so forth.14 Of particular note is the work of Xie and coworkers, who mapped experimentally the dependence of gold nanoparticle catalytic activity to the number of gold atoms.15,16 Those works marked out the limits of discrete nanoparticle properties compared to bulk gold and allowed for precise theoretical calculations that could further illuminate the origin of the high nanoparticle catalytic activity.17,18 The enhanced catalytic activity of small nanoparticles was often assumed to be solely related to their larger surface area compared to that of the macroscopic catalyst. This extended surface would also imply a high occurrence of surface defects and irregularities such as vacancies, adatoms, edges, terraces, and so forth that would contribute to catalytic activity. Recent experimental and theoretical reports suggest that sub-2 nm nanoparticles in fact show different physical properties compared to the macroscopic bulk materials attributed to Received: November 1, 2013 Revised: April 5, 2014 Published: April 11, 2014 8907

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

their finite size and discrete electronic states.19−21 Such phenomena were the experimentally reported size dependence of the melting temperature of gold nanoparticles, the theoretically demonstrated rigid core and flexible shell of ∼1 nm gold nanoparticles,22 as well as the theoretically and experimentally estimated core−shell charge separation in ∼1 nm gold and platinum nanoparticles.18,23,24 These deviations from the physical properties of the bulk material originate from the finite size and shape of the nanoparticles and have important contribution to their catalytic activity. For example, the flexible shell structure is capable to adjust its geometry to facilitate different surface reactions22 while the negative surface charge catalyzes the initial step of electrophilic reactions.18 A major challenge for nanoparticle catalysis is the synthesis and stabilization of sub-2 nm nanoparticles. Owing to their supreme reactivity and mobility, sub-2 nm metal clusters easily merge to form larger nanoparticles. Thus, many of the valuable physical properties are lost, and the catalytic activity is suppressed. In colloidal solutions, merging is avoided by attaching protective organic ligands to the metal surface.25 However, in this way, large surface area is occupied by the ligands and only limited active sites are available for catalytic reactions. Another possibility to retain small nanoparticles size is their deposition on surface supports that limit their mobility.15 Tailoring the nanoparticle−surface support interaction could provide a means to limit nanoparticle mobility and thus to prevent aggregation. Carbon nanotubes (CNT) are the preferred support for metal nanoparticles.26,27 For example, stable 1 nm gold clusters and gold oxide nanoparticles have been reported on CNTs.16,28 Recent experiments showed a precise dependence of the adsorbed nanoparticle size on the nanotube surface curvature.29 Furthermore, it was shown experimentally that on CNT surfaces with low curvature, the nanoparticle catalytic activity decreases with time.29 Aside from their catalytic activity, metal nanoparticles deposited on CNTs also often find application as gas sensors.30,31 Theoretical and experimental studies have shown that gold nanoparticles adsorbed on CNTs alter the tubes’ electron-transport properties.31 The fact that metal nanoparticles could alter the conductance of CNTs suggests a binding interaction that involves the mixing of electron levels. Such binding interaction should be responsible for the nanoparticle fixing on the surface and its reduced mobility. Low mobility of the metal clusters would lower the probability for merging and the formation of larger nanoparticles. Thus, the lifetime of the small, ∼1 nm, catalytically active clusters would be significantly increased. In this study, we investigate the nature of interaction between gold or palladium nanoparticles and nanotube surfaces through density functional theory (DFT) and propose an intuitive experiment to verify our theoretical findings. Because small gold or palladium clusters are rarely reported on flat graphene surfaces compared to CNTs, we believe that the surface curvature is crucial for the nanoparticle−support interaction. The curvature effect is investigated by altering the nanotube diameters between 0.5 nm and ∼1.5 nm. The results are compared to the nanoparticle interaction with flat graphene surface.

gold atoms, for example, Au13, to carbon or boron−nitrogen nanotubes. Throughout this study, zigzag metallic CNTs with chirality vectors (n,0) are investigated. The obtained results are verified for armchair and zigzag semiconductor CNTs. The investigated CNTs possess chiral vectors (6,0), (9,0), (12,0), and (18,0) and diameters 0.47, 0.71, 0.94, and 1.41 nm, respectively. Boron−nitrogen nanotubes (BNNTs) are characterized with large band gap and offer different adsorption sites on the surface, that is, nitrogen and boron sites. BNNTs with chiral vectors (6,0), and (18,0) are considered. In addition to nanotubes, the binding energies of Pd13 and Au13 to flat graphene surface are considered. Nanoparticles Pd13 and Au13 are characterized with cube-octahedral symmetry. Periodic, plane wave density functional theory calculations were performed using Vienna Ab-initio Software Package (VASP).32−34 Perdew−Burke−Ernzerhof parametrization of the generalized gradient approximation (GGA-PBE) was employed using projector augmented wave (PAW) pseudopotentials.35 The results were verified with the optB86b-vdW functional, which includes the long-range van der Waals interactions. Both functionals showed similar trends for optimized geometries and binding energies. An electron cutoff energy of 400 eV was employed for all geometry optimizations and density of states (DOS) calculations. An electron cutoff energy of 300 eV was employed for the first-principle molecular dynamics simulations. Owing to the relatively large supercells, 1 × 1 × 1 gamma k-point sampling was used. Density of states and site-projected density of states analysis for the CNTs were performed with 1 × 1 × 10 k-point sampling. Spin polarization was not employed during the calculations. Absolute values of the Fermi energies were determined as relative to the vacuum potential. Throughout this study, we used the graphical visualization package VESTA36 to analyze and visualize the computed DFT electron density distribution. The nanoparticle adsorption was investigated on CNTs and BNNTs with axial length within the supercell of 2.2 nm. Five unit cells were included in one supercell. The width of the unit cells in the directions perpendicular to the nanotubes was determined with the relation [(nanotube diameter) + 2.0 nm]. The DOS spectra were calculated for a single unit cell of the CNTs. The supercell used for particles’ adsorption on flat graphene surface has lattice vectors of (1.8, 1.5, and 2.9 nm). Binding energies of the nanoparticles to the carbon or boron−nitrogen surfaces were estimated with the following relation: E bind = Eall − (Enp + Ent)

(1)

where Ebind denotes the binding energy, Eall is the energy of the nanoparticle−nanotube complex, Enp is the energy of the nanoparticle, and Ent the energy of the nanotube. In addition to the electron density distribution, electron density differences were compared, which were obtained with the following relation: pdiff = pall − (pnp + pnt )

(2)

Here, pdiff represents the electron density difference, pall denotes the electron density of the nanoparticle−nanotube complex, pnp is the electron density of the nanoparticle, and pnt is the electron density of the nanotube. Bader population analysis was performed to allocate electron density to the ions.37 In addition to the geometry optimization, first-principle molecular dynamics calculations were performed for two Pd13

■

METHODS OF INVESTIGATION Theoretical Methods. In our study, we investigate the binding energy of palladium nanoparticle consisting of 13 atoms, for example, Pd13, or gold nanoparticle consisting of 13 8908

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

Figure 1. Adsorption of Pd13 on CNTs and graphene surface. (A) Optimized geometry. All distances are in Å. (B) Electron density difference calculated with eq 2. Blue color denotes positive density, yellow color denotes negative density.

particles adsorbed on flat graphene surface and on CNT(12,0) with diameter of 0.94 nm. First-principle molecular dynamics calculations were performed for two Au13 particles adsorbed on flat graphene surface and on CNT(6,0) and CNT(12,0) with diameters of 0.47 and 0.94 nm, respectively. The supercell used for the first-principle molecular dynamics of flat graphene surface has lattice vectors of (2, 2, and 1.5 nm). The axial length of the CNT used for the first-principle molecular dynamics calculations is 2.2 nm. The dynamics simulations were performed for 1000 steps of 0.5 fs and 500 °C. Experimental Methods. Commercially available single wall CNTs with average diameter of 1−2 nm and axial length of 5−30 μm and graphene sheets were decorated with 1.5 wt % Pd nanoparticles by liquid reduction with hydrazine. The particles’ size distribution was compared with tunneling electron microscopy (TEM) immediately after the loading and after 2 h annealing at 500 °C in N2 flowing at 100 mL/min. The purpose of the experiment was to provide information on the particle size before and after the annealing as a function of the surface curvature. The experimentally investigated tubes are close in diameter to CNT(12,0) which was used for our firstprinciple molecular dynamics simulation. The flat graphene surface is an extreme case of surface with zero curvature.

2.71 Å. The particle surface consists of walls with triangular and square shapes. The particle is placed on the tube surface facing the surface with its triangular base. The optimized geometries of Pd13 on the surfaces of the investigated tubes are shown in Figure 1A. Figure 1B plots the electron density difference calculated with eq 2. In Figure 1 is included the result for Pd13 interaction with graphene as a model of a carbon surface with zero curvature. Figure 1A reveals that the tube−nanoparticle distance varies with the tube diameter. The tube−nanoparticle distance is 2.1 Å for Pd13 on CNT(6,0); 2.0 Å for Pd13 on CNT(9,0), CNT(12,0), and CNT(18,0); and 1.9 Å for Pd13 on CNT(18,0). The particle−surface distance for Pd13 on graphene surface is 2.1 Å. The geometry plotted in Figure 1A reveals that only the Pd−Pd bonds among the Pd atoms connected to the tube’s surface are distorted while the rest of the Pd−Pd bonds remain similar to those of the insulated particle. Pd13 retains its well-ordered close-to-ideal cubeoctahedral symmetry. Better understanding for the interaction between the particle and the tubes can be obtained from the electron density difference plots in Figure 1B. The electron density difference represents the electron density change that results from the particle adsorption on the nanotube surface. Positive electron density is shown in blue, while yellow color indicates negative electron density. The plot for Pd13 on CNT(6,0) suggests good overlap between the pz atomic orbitals of the carbon atoms and the d atomic orbitals of the Pd atoms. Such overlap and electron density distribution point to a covalent type binding interaction. With an increase in the tubes’ diameter, that is, a decrease in the tubes’ curvature, the electron density is significantly reduced and almost disappears for the flat graphene surface. These results indicate that the surface curvature can effectively control the binding interaction between the particle and the tube. Quantitative understanding

■

RESULTS AND DISCUSSION Pd13 Adsorption on Single Wall Carbon Nanotubes. Our calculations begin with a systematic evaluation of the interaction between a small palladium cluster consisting of 13 atoms (Pd13) and 4 CNTs with chiral vectors (6,0), (9,0), (12,0), and (18,0) and diameters 0.47, 0.71, 0.94, and 1.41 nm, respectively. The nanoparticle consists of a central Pd atom surrounded by a shell of 12 Pd atoms. The nanoparticle has cube-octahedral symmetry with bond length on the surface of 8909

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

Figure 2. Density of states and site projected density of states spectra of zigzag CNTs with metallic properties. The spectra are plotted as a function of the electron energy in the interval −5 to 5 eV. The energies are relative to the Fermi energy (EF = 0). The coordinates’ directions are defined locally. (A) DOS spectra; (B) PDOS spectra.

for this interaction is provided by the binding energies calculated through eq 1. The binding energy of Pd13 to CNT(6,0) is −2.84 eV which points to significant chemisorption. With an increase in the tube’s diameter, for example, CNT(9,0), the binding energy is reduced to −2.24 eV. A further increase in the tube’s diameter results in binding energies of −1.71 eV and −1.59 eV for CNT(12,0) and CNT(18,0), respectively. The binding energy of Pd13 to the flat graphene surface is −0.85 eV. These computational results suggest that the particle−tube interaction is affected by the

surface curvature. Stronger bonds are formed between Pd13 and highly curved surfaces while only weak interaction is observed between the particle and surfaces with low curvature. In this study, we provide understanding for the binding interaction of metal nanoparticles with small diameter nanotubes. CNTs and graphene are built by sp2-hybridized carbon atoms. The sp2-hybridization enforces planar geometry and leads to single atomic layer materials such as graphene, a zeroband semiconductor, and boron−nitride, a wide band gap semiconductor. As those materials are wrapped, the sp28910

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

hybridization is perturbed by an additional sp3-hybridization component. The additional sp3-hybridization component is characterized with electron density normal to the tube surface. Furthermore, the sp3-hybridization component implies four valence bonds, while the sp2-hybridization implies three valence bonds. Thus, the larger curvature would increase the sp3hybridization component and would increase the tube’s affinity toward binding interaction with the nanoparticles. To provide quantitative verification to this intuitive explanation, we investigate the tubes’ density of states (DOS) and the site projected density of states (PDOS). The DOS spectra for CNT(6,0), CNT(9,0), CNT(12,0), and CNT(18,0) are plotted in Figure 2A, while Figure 2B plots the PDOS spectra. The DOS spectrum of CNT(18,0) resembles the DOS spectrum of graphene. In the DOS spectrum of CNT(18,0), the peaks below and above the Fermi energy correspond to the in-phase and the out-of-phase overlap of the 2pz atomic orbitals at each carbon atom. This can be verified by the PDOS analysis shown in Figure 2B. With an increase in surface curvature for tube CNT(12,0), a small peak emerges below the Fermi level. The intensity of that peak increases for CNT(9,0), while it becomes a dominant characteristic for the DOS spectrum of CNT(6,0). The peak is clearly related to the surface curvature. It originates from the 2pz AO of the carbon atoms; however, careful analysis of the PDOS spectra of CNT(6,0) and CNT(9,0) shows significant contribution of other 2p AOs, which is evidence for sp3-hybridization component. The new peak overlaps with the d-orbitals of the metal atoms at the nanoparticle surface leading to bonding interaction. To relate the particle−tube binding energy to the particle aggregation, we perform first-principle molecular dynamics simulations of a pair of Pd13 particles on the surface of CNT(12,0) and on the flat graphene surface. The simulations are performed for temperature of 500 °C and 1000 time steps. The starting and ending geometries for the pair of particles on the tube surface and on the graphene surface are shown in Figure 3A and B, respectively. Our intuitive understanding is

molecular dynamics calculations support that understanding. On the tube surface, the particles are slightly displaced, while on the flat surface the particles approach each other. In 300 time steps on the flat graphene surface, the particles make contact with one another and merge into a larger particle, Pd26. From our calculations, we predict that Pd nanoparticles will show significant binding to CNTs with 1−2 nm diameter, which should limit their mobility and thus suppress their aggregation. Our first-principle molecular dynamics calculations suggest that 500 °C would not be sufficient to achieve significant Pd particle mobility on tubes with diameters close to that of CNT(12,0). To provide experimental verification to these theoretical findings, we have designed an intuitive experiment. We have decorated 1−2 nm diameter CNTs and graphene sheets with Pd nanoparticles and have observed the average particle size after the preparation and after annealing for 2 h at 500 °C. We believe that this experimental setup is close to our first-principle molecular dynamics simulations. The tunneling electron microscopy images (TEM) in Figure 4A

Figure 4. Experimental data. (A) TEM images of Pd nanoparticles on graphene surface and on 1−2 nm diameter CNTs’ surfaces. Images are taken after the loading and after sequential annealing for 2 h at 500 °C. (B) HAADF image of Pd nanoparticle on CNT surface.

Figure 3. First principle molecular dynamics simulations of a pair of Pd13 particles on CNT(12,0) and flat graphene surface. Temperature is 500 °C. (A) Starting geometry. (B) Geometry snapshot after 1000 time steps.

show that the particle size immediately after the preparation is 5−20 nm on both surfaces (CNTs and graphene). After 2 h annealing, the particles’ size remains mostly unchanged on the curved CNTs surfaces, between 10−20 nm, while on the flat graphene surface they aggregate to form larger nanocolloids, 50−200 nm in diameter, thus confirming the prediction based on our theoretical study. In Figure 4B is shown an image of Pd nanoparticle on CNT surface obtained by high-angle annular dark-field (HAADF) technique. The experimentally observed

that the particle−surface systems characterized with significant binding energy would show low mobility, while the particles loosely bound to the flat graphene surface would show higher mobility. The reason for the suppressed mobility would be the fact that the chemisorption bonds should be broken, which requires activation energy. The results from the first-principle 8911

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

Figure 5. Adsorption of Au13 on CNTs and graphene surface. (A) Optimized geometry. All distances are in Å. (B) Electron density difference calculated with eq 2. Blue color denotes positive density, yellow color denotes negative density.

the tubes and the particle is reduced. This is visualized by the electron density difference shown in Figure 5B and the values of the binding energy, −1.25 eV and −1.00 eV, respectively. In comparison, the nanoparticle is only loosely adsorbed on the surface of CNT(18,0) with a binding energy of −0.35 eV. The result for Au13 binding energy with the flat graphene surface is 0.03 eV. The binding between Au13 and CNT(6,0) is realized by overlap between the Au dz2 orbitals and the C pz orbitals of the nanotube. As a result, electron density is localized between the particle and the nanotube surface, forming a covalent-type bond. With an increase in the tubes’ diameter, this electron density (and thus the bonding interaction) gradually decreases as seen for Au13 on CNT(18,0) and Au13 on graphene. The electron density difference for Pd13 on CNT(6,0) reveals that one palladium atom binds with two carbon atoms on the nanotube surface, whereas one gold atom in Au13 on CNT(6,0) binds only to one carbon atom. This qualitative comparison of the electron density differences for Pd13 on CNT(6,0) and Au13 on CNT(6,0) shows increased electron density between the palladium particle and the nanotube, suggesting the palladium nanoparticle binds more strongly to the nanotube surface; indeed, the calculated binding energy for Pd13 to CNT(6,0) is −2.84 eV, which is significantly stronger than the binding energy of −1.88 eV for Au13 on CNT(6,0). With an increase in the tubes’ diameter, the electron density between the particles and the tubes’ surfaces is reduced. However, the qualitative comparison of the electron density differences still shows increased electron density between Pd13 and CNT(18,0) compared to Au13 on CNT(18,0). The calculated binding energy for Pd13 to CNT(18,0) is −1.59 eV, which is stronger than the −0.35 eV binding energy of Au13 to CNT(18,0). In the extreme case of the particles adsorbed on the flat graphene surface, Pd13 still shows binding interaction while Au13 shows nonbonding interaction. Thus, we would expect fine palladium nanoparticles on nanotubes with larger diameters.29 To understand the better binding of Pd nanoparticles compared to Au nanoparticles to the CNTs surfaces, we have investigated their DOS spectra plotted in Figure 6. For Pd13,

high crystallinity is in good agreement with the theoretically predicted well-ordered Pd nanoparticle geometries on curved CNTs’ surfaces. We should note the significant difference in particle size in our theoretical and experimental setups. In the theoretical setup, the particle size is ∼0.8 nm in diameter. Larger particles adsorbed on CNTs would significantly increase the demands for computational time and processing power. Classical mechanics simulations would make feasible the study of particles with diameter of ∼5 nm; however, the important information for the electronic interaction on the interface responsible for the binding would be lost. The experimental deposition of ∼1 nm diameter nanoparticles requires complex synthetic techniques. Nevertheless, we believe that our theoretical and experimental observations demonstrate the surface curvature effect on nanoparticles’ aggregation. Au13 Adsorption on Single Wall Carbon Nanotubes. To show that the obtained results are not specific for palladium and that the method can be applied to a larger range of materials, we turn now to 13 atom gold cluster (Au13) adsorbed on CNTs with different diameters. Our further calculations are performed for particles adsorbed on carbon nanotubes CNT(6,0), CNT(9,0), CNT(12,0), and CNT(18,0) with diameters 0.47, 0.71, 0.94, and 1.41 nm, respectively. The insolated Au13 has cube-octahedral geometry with a single Au atom in the center surrounded by a shell of 12 Au atoms with bond length on the surface of 2.8 Å. When this particle is placed on a tube, its geometry alters as a function of the tube diameter. The relaxed geometries of Au13 adsorbed on CNT(6,0), CNT(9,0), CNT(12,0), and CNT(18,0) are shown in Figure 5A, and the electron density difference maps are plotted in Figure 5B. The particle is anchored to the tube through three base atoms. The Au−Au distance for the gold atoms connecting to the tube surfaces increases with the surface curvature, starting from 3.6 Å for Au13 adsorbed on CNT(6,0) to 5.5 Å for Au13 adsorbed on CNT(18,0). The Au−Au distance for the gold atoms connecting to the graphene surfaces is 3.2 Å. The computed binding energy between Au13 and CNT(6,0) is −1.88 eV. With an increase in the tubes’ diameter, for example, CNT(9,0) and CNT(12,0), the binding between 8912

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

Figure 6. Density of states spectra of Au13 and Pd13. The spectra are plotted as a function of the electron energy in the interval −5 to 5 eV. The energies are relative to the Fermi energy (EF = 0).

Figure 7. Site projected density of states spectra for Au, Pd, and C atoms at the nanoparticle−nanotube interface for CNT(6,0) with Au13 and Pd13. The spectra are plotted as a function of the electron energy in the interval −5 to 5 eV. The energies are relative to the Fermi energy (EF = 0). (A) PDOS of Au and C atoms at the CNT(6,0)−Au13 interface. (B) PDOS of Pd and C atoms at the CNT(6,0)−Pd13 interface.

p-band of carbon in the interval of −1.0 to 1.0 eV is significantly weaker because the d-band of the gold atom is shifted to the lower energies (Figure 7A). First-principle molecular dynamics simulations are performed for a pair of Au13 particles on the surface of CNT(6,0) and CNT(12,0) as well as on the flat graphene surface. The simulations are performed for a temperature of 500 °C and 1000 time steps. The starting and ending geometries for the pair of particles on the tubes’ surfaces and on the graphene surface are shown in Figure 8A and B, respectively. On the CNT(12,0) surface, the particles are more strongly displaced compared to the simulation of Pd13 on CNT(12,0). The particles approach each other and merge into a larger nanoparticle, Au26. This result is expected because Au13 binds loosely to the CNT(12,0) surface. To prevent the particles’ merging, one should use nanotubes with significantly smaller diameter. This is demonstrated with the simulation of two Au13 particles on CNT(6,0). Within the 1000 time steps,

the peak in the DOS spectrum that originates from the dorbitals is in the interval 0.0 eV to −1.0 eV. For Au13, the peak in the DOS spectrum that originates from the d-orbitals is in the interval −1.0 eV to −2.0 eV. We have also estimated the values of the Fermi energies relative to the vacuum of the CNT tubes, Pd13, and Au13. For the tubes, the calculated Fermi energy is −4.4 eV; for Pd13, the calculated Fermi energy is −4.5 eV; and for Au13, the calculated Fermi energy is −4.9 eV. Thus, the DOS peaks of Pd13 d-orbitals will have better overlap with the peak below the Fermi energy of small diameter CNTs. In addition to the Fermi energy estimation of Au13 and Pd13, we have analyzed the PDOS spectra of gold and carbon atoms from the CNT(6,0)−Au13 interface as well as the PDOS spectra of palladium and carbon atoms from the CNT(6,0)− Au13 interface. The results are plotted in Figure 7. One can clearly see the good overlap between the d-band of palladium and the p-band of carbon in the interval of −1.0 to 1.0 eV (Figure 7B). The overlap between the d-band of gold and the 8913

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

nanoparticle while only negligible interaction is observed between boron and palladium atoms; the electron donor character of the nitrogen atoms within the nanotube structure allows them to contribute electron density to the nanotube− nanoparticle bond. This may lead to reduced binding interaction between small diameter BNNTs and palladium nanoparticles because the nanoparticle at the adsorption site would always bond to one or several boron atoms. The calculated binding energy for Pd13 adsorbed on BNNT(6,0) is −1.47 eV, while that for Pd13 adsorbed on BNNT(18,0) is −0.68 eV. The calculations for Pd13 adsorption on BNNTs confirmed that surface curvature is important for the nanoparticle−nanotube interaction. Smaller diameter nanotubes bind the nanoparticle more strongly to the tube’s surface independent of the tube’s chemical composition. However, a major difference between BNNTs and CNTs is that on the BNNTs only nitrogen atoms could bind to palladium atoms, which leads to smaller absolute binding energies between the tube and the nanoparticle compared to those for similar diameter CNTs. The relaxed geometries for Au13 on BNNT(6,0) and BNNT(18,0) tubes with diameters of 0.47 and 1.41 nm, respectively, are shown in Figure 9B. Additionally, Figure 9B plots the electron density differences calculated with eq 2. Similar to the results for the Pd13 adsorbed on BNNTs, the electron density difference for Au13 adsorbed on BNNT(6,0) reveals that nitrogen atoms bind to the gold atoms of the nanoparticle while only negligible interaction is observed between boron and gold atoms. This leads to reduced binding interaction between small diameter BNNTs and gold nanoparticles because the nanoparticle at the adsorption site would always bond to one or several boron atoms. The calculated binding energy for Au13 adsorbed on BNNT(6,0) is −0.81 eV, while that for Au13 adsorbed on BNNT(18,0) is −0.36 eV. The calculations for Au13 adsorption on BNNTs confirmed that surface curvature is important for the nanoparticle− nanotube interaction. The results verify that Pd13 nanoparticle

Figure 8. First-principle molecular dynamics simulations of pair Au13 particles on CNT(6,0), CNT(12,0), and flat graphene surface. Temperature is 500 °C. (A) Starting geometry. (B) Geometry snapshot after 1000 time steps.

the particles’ mobility is limited, and they remain separated. On the flat graphene surface, the particles approach one another and in 300 time steps merge into a larger particle, Au26. Pd13 and Au13 Adsorption on Boron−Nitrogen Nanotubes. The calculations on the nanoparticle−CNT systems confirm that the tube’s surface curvature controls the binding interaction. To determine whether this depends on the elemental composition of the tubes or is applicable for a larger range of materials, we performed similar calculations for boron−nitrogen nanotubes (BNNT). Similar to CNTs, BNNTs possess π-electron system of delocalization built by the pz atomic orbitals on boron and nitrogen atoms. Nitrogen atoms are electron donors and provide two electrons to the πelectron system, while boron is an electron acceptor and provides zero electrons to the π-electron system. The relaxed geometries for Pd13 on BNNT(6,0) and BNNT(18,0) tubes with diameters of 0.47 and 1.41 nm, respectively, are shown in Figure 9A. Additionally, Figure 9A plots the electron density differences calculated with eq 2. The electron density difference for Pd13 adsorbed on BNNT(6,0) reveals that nitrogen atoms bind to the palladium atoms of the

Figure 9. Adsorption of Pd13 and Au13 on BNNTs. Optimized geometries and electron density difference calculated with eq 2 are shown. Blue color denotes positive density, yellow color denotes negative density. All distances are in Å. (A) Adsorption of Pd13 on BNNTs. (B) Adsorption of Au13 on BNNTs. 8914

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

■

binds stronger to nanotube surfaces compared to Au13 nanoparticle.

■

ASSOCIATED CONTENT

S Supporting Information *

Optimized geometry and unit cells of all investigated nanoparticles on nanotube surfaces and binding energies calculated with optB86b-vdW functional. This information is available for free via the Internet at http://pubs.acs.org.

■

REFERENCES

(1) Thapa, A. K.; Shin, T. H.; Ida, S.; Sumanasekera, G. U.; Sunkara, M. K.; Ishihara, T. Gold−Palladium Nanoparticles Supported by Mesoporous β-MnO2 Air Electrode for Rechargeable Li-Air Battery. J. Power Sources 2012, 220, 211−216. (2) Chung, H. T.; Won, J. H.; Zelenay, P. Active and Stable Carbon Nanotube/Nanoparticle Composite Electrocatalyst for Oxygen Reduction. Nat. Commun. 2013, 4, 1922. (3) Chi Linh, D.; Thy San, P.; Ngoc Phong, N.; Viet Quan, T. Properties of Pt/C Nanoparticle Catalysts Synthesized by Electroless Deposition for Proton Exchange Membrane Fuel Cell. Adv. Nat. Sci.: Nanosci. Nanotechnol. 2013, 4, 035011. (4) Cui, C.; Gan, L.; Heggen, M.; Rudi, S.; Strasser, P. Compositional Segregation in Shaped Pt Alloy Nanoparticles and Their Structural Behavior During Electrocatalysis. Nat. Mater. 2013, 12, 765−771. (5) Hayashi, A.; Kimijima, K.; Miyamoto, J.; Yagi, I. Direct Observation of Well-Dispersed Pt Nanoparticles Inside the Pores of Mesoporous Carbon through the Cross Section of Pt/Mesoporous Carbon Particles. Chem. Lett. 2009, 38, 346−347. (6) Edwards, J. K.; Solsona, B.; Ntainjua, E.; Carley, A. F.; Herzing, A. A.; Kiely, C. J.; Hutchings, G. Switching Off Hydrogen Peroxide Hydrogenation in the Direct Synthesis Process. Science 2009, 323, 1037−1041. (7) Edwards, J. K.; Solsona, B. E.; Landon, P.; Carley, A. F.; Herzing, A.; Kiely, C. J.; Hutchings, G. Direct Synthesis of Hydrogen Peroxide from H2 and O2 Using TiO2-Supported Au-Pd Catalysts. J. Catal. 2005, 236, 69−79. (8) Staykov, A.; Kamachi, T.; Ishihara, T.; Yoshizawa, K. Theoretical Study of the Direct Synthesis of H2O2 on Pd and Pd/Au Surfaces. J. Phys. Chem. C 2008, 112, 19501−19505. (9) Li, J.; Staykov, A.; Ishihara, T.; Yoshizawa, K. Theoretical Study of the Decomposition and Hydrogenation of H2O2 on Pd and Au@Pd Surfaces: Understanding toward High Selectivity of H2O2 Synthesis. J. Phys. Chem. C 2011, 115, 7392−7398. (10) Nomura, Y.; Ishihara, T.; Hata, Y.; Kitawaki, K.; Kaneko, K.; Matsumoto, H. Nanocolloidal Pd-Au as Catalyst for the Direct Synthesis of Hydrogen Peroxide from H2 and O2. ChemSusChem 2008, 1, 619−621. (11) Maeda, K.; Teramura, K.; Lu, D.; Saito, N.; Inoue, Y.; Domen, K. Roles of Rh/Cr2O3 (Core/Shell) Nanoparticles Photodeposited on Visible-Light-Responsive (Ga1‑xZnx)(N1‑xOx) Solid Solutions in Photocatalytic Overall Water Splitting. J. Phys. Chem. C 2007, 111, 7554−7560. (12) Hagiwara, H.; Inoue, T.; Ida, S.; Ishihara, T. Long-Time Charge Separation in Porphyrin/KTa(Zr)O3 as Water Splitting Photocatalyst. Phys. Chem. Chem. Phys. 2011, 13, 18031−18037. (13) Haruta, M. When Gold Is Not Noble: Catalysis by Nanoparticles. Chem. Rec. 2003, 3, 75−87. (14) Boronat, M.; Leyva-Pérez, A.; Corma, A. Theoretical and Experimental Insights into the Origin of the Catalytic Activity of Subnanometric Gold Clusters: Attempts To Predict Reactivity with Clusters and Nanoparticles of Gold. Acc. Chem. Res. 2014, 47, 834− 844. (15) Liu, Y.; Tsunoyama, H.; Akita, T.; Xie, S.; Tsukuda, T. Aerobic Oxidation of Cyclohexane Catalyzed by Size-Controlled Au Clusters on Hydroxyapatite: Size Effect in the Sub-2 nm Regime. ACS Catal. 2010, 1, 2−6. (16) Xie, S.; Tsunoyama, H.; Kurashige, W.; Negishi, Y.; Tsukuda, T. Enhancement in Aerobic Alcohol Oxidation Catalysis of Au25 Clusters by Single Pd Atom Doping. ACS Catal. 2012, 2, 1519−1523. (17) Shang, C.; Liu, Z.-P. Origin and Activity of Gold Nanoparticles as Aerobic Oxidation Catalysts in Aqueous Solution. J. Am. Chem. Soc. 2011, 133, 9938−9947. (18) Staykov, A.; Nishimi, T.; Yoshizawa, K.; Ishihara, T. Oxygen Activation on Nanometer-Size Gold Nanoparticles. J. Phys. Chem. C 2012, 116, 15992−16000. (19) Roldán, A.; González, S.; Ricart, J. M.; Illas, F. Critical Size for O2 Dissociation by Au Nanoparticles. ChemPhysChem 2009, 10, 348− 351.

CONCLUSIONS In this study, we have provided theoretical understanding, along with experimental verification, for the nature of binding of metal nanoparticles with carbon and boron−nitrogen surfaces. We have demonstrated that of primary importance for the interaction between those nanomaterials is the support surface curvature. Flat surfaces do not show significant binding interaction with the nanoparticles, whereas curved surfaces, that is, nanotubes, show binding properties that depend on the tubes’ diameter. The nature of the binding is electron density transferred from the nanoparticle and the nanotube to the intermediate space, forming a covalent-type bond, with computed binding energies suggesting chemisorption of the nanoparticles. The covalent character of the binding is the result of the electron density shift from the nanotubes and the nanoparticles to the region between the metal and the carbon atoms. For all investigated nanotube−nanoparticle combinations, the binding is strengthened with an increase in surface curvature. We have demonstrated experimentally that Pd nanoparticles on a flat graphene surface aggregate with time at high temperature, while their fine size is preserved on small diameter CNTs. Our theory also provides several practical rules for the design of particle-decorated nanotubes. Larger nanotubes with diameter above 2 nm would bind weak metal nanoparticles compared to the smaller diameter nanotubes. BNNTs bind metal nanoparticles that are weaker compared to CNTs. The reason is that on boron−nitrogen surfaces the metal particles bind only to the electron-donating nitrogen atoms. Palladium particles bind better to CNTs compared to gold nanoparticles because the Fermi level of the palladium particle is higher energetically compared to the gold particle. We believe that our theoretical and experimental findings will guide the preparation of highly active nanocatalysts characterized with improved durability under high temperatures for applications in fuel cells, for hydrogen production through water splitting reaction, and for preparative and synthetic chemistry.

■

Article

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: +81-92-802-6743. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work was supported by World Premier International Research Center Initiative (WPI), Ministry of Education, Culture, Sports, Science, and Technology of Japan (MEXT), Japan. The authors are thankful to Dr. John Druce at I2CNER for reading the manuscript and for the useful comments. 8915

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916

The Journal of Physical Chemistry C

Article

(20) Gao, Y.; Shao, N.; Pei, Y.; Chen, Z.; Zeng, X. C. Catalytic Activities of Subnanometer Gold Clusters (Au16−Au18, Au20, and Au27−Au35) for CO Oxidation. ACS Nano 2011, 5, 7818−7829. (21) Lopez-Sanchez, J. A.; Dimitratos, N.; Hammond, C.; Brett, G. L.; Kesavan, L.; White, S.; Miedziak, P.; Tiruvalam, R.; Jenkins, R. L.; Carley, A. F.; et al. Facile Removal of Stabilizer-Ligands from Supported Gold Nanoparticles. Nat. Chem. 2011, 3, 551−556. (22) Gu, X.; Bulusu, S.; Li, X.; Zeng, X. C.; Li, J.; Gong, X. G.; Wang, L.-S. Au34‑: A Fluxional Core−Shell Cluster. J. Phys. Chem. C 2007, 111, 8228−8232. (23) Yasumatsu, H.; Hayakawa, T.; Kondow, T. Experimental Observation of Two-Dimensional Charge Polarization in Unisized Platinum Cluster Disk Bonded to Silicon (111) Surface. Chem. Phys. Lett. 2010, 487, 279−284. (24) Yasumatsu, H.; Murugan, P.; Kawazoe, Y. Nanospace Charge Polarization of Monatomic-Layered Platinum Cluster Disk Constructed on Silicon(111)-7 × 7 Surface. Phys. Status Solidi B 2012, 249, 1193−1198. (25) Heaven, M. W.; Dass, A.; White, P. S.; Holt, K. M.; Murray, R. W. Crystal Structure of the Gold Nanoparticle [N(C8H17)4][Au25(SCH2CH2Ph)18]. J. Am. Chem. Soc. 2008, 130, 3754−3755. (26) Ozawa, H.; Yi, X.; Fujigaya, T.; Niidome, Y.; Asano, T.; Nakashima, N. Supramolecular Hybrid of Gold Nanoparticles and Semiconducting Single-Walled Carbon Nanotubes Wrapped by a Porphyrin−Fluorene Copolymer. J. Am. Chem. Soc. 2011, 133, 14771− 14777. (27) Fujigaya, T.; Okamoto, M.; Nakashima, N. Design of an Assembly of Pyridine-Containing Polybenzimidazole, Carbon Nanotubes and Pt Nanoparticles for a Fuel Cell Electrocatalyst with a High Electrochemically Active Surface Area. Carbon 2009, 47, 3227−3232. (28) Alves, L.; Ballesteros, B.; Boronat, M.; Cabrero-Antonino, J. R.; Concepción, P.; Corma, A.; Correa-Duarte, M. A.; Mendoza, E. Synthesis and Stabilization of Subnanometric Gold Oxide Nanoparticles on Multiwalled Carbon Nanotubes and Their Catalytic Activity. J. Am. Chem. Soc. 2011, 133, 10251−10261. (29) Cornelio, B.; Rance, G. A.; Laronze-Cochard, M.; Fontana, A.; Sapi, J.; Khlobystov, A. N. Palladium Nanoparticles on Carbon Nanotubes as Catalysts of Cross-Coupling Reactions. J. Mater. Chem. A 2013, 1, 8737−8744. (30) Zanolli, Z.; Charlier, J. C. Single-Molecule Sensing Using Carbon Nanotubes Decorated with Magnetic Clusters. ACS Nano 2012, 6, 10786−10791. (31) Zanolli, Z.; Leghrib, R.; Felten, A.; Pireaux, J.-J.; Llobet, E.; Charlier, J. C. Gas Sensing with Au-Decorated Carbon Nanotubes. ACS Nano 2011, 5, 4592−4599. (32) Kresse, G.; Furthmüller, J. Efficient Iterative Schemes for Ab Initio Total-Energy Calculations Using a Plane-Wave Basis Set. Phys. Rev. B 1996, 54, 11169−11186. (33) Kresse, G.; Furthmüller, J. Efficiency of Ab-Initio Total Energy Calculations for Metals and Semiconductors Using a Plane-Wave Basis Set. Comput. Mater. Sci. 1996, 6, 15−50. (34) Kresse, G.; Hafner, J. Ab Initio Molecular Dynamics for Liquid Metals. Phys. Rev. B 1993, 47, 558−561. (35) Blöchl, P. E. Projector Augmented-Wave Method. Phys. Rev. B 1994, 50, 17953−17979. (36) Momma, K.; Izumi, F. VESTA 3 for Three-Dimensional Visualization of Crystal, Volumetric and Morphology Data. J. Appl. Crystallogr. 2011, 44, 1272−1276. (37) Tang, W.; Sanville, E.; Henkelman, G. A Grid-Based Bader Analysis Algorithm without Lattice Bias. J. Phys.: Condens. Matter 2009, 21, 084204.

8916

dx.doi.org/10.1021/jp410775n | J. Phys. Chem. C 2014, 118, 8907−8916