Molecular Dynamics Simulations of Metal Clusters

3522

J. Phys. Chem. C 2010, 114, 3522–3530

Molecular Dynamics Simulations of Metal Clusters Supported on Fishbone Carbon Nanofibers Carlos F. Sanz-Navarro,*,† Per-Olof Åstrand,‡ De Chen,§ Magnus Rønning,§ Adri C. T. van Duin,| and William A. Goddard III# Centre d’InVestigacio´ en Nanocie`ncia i Nanotecnologia (CIN2) CSIC-ICN, Campus UAB, 08193 Bellaterra, Spain, Department of Chemistry, Norwegian UniVersity of Science and Technology (NTNU), 7491 Trondheim, Norway, Department of Chemical Engineering, Norwegian UniVersity of Science and Technology (NTNU), 7491 Trondheim, Norway, Department of Mechanical and Nuclear Engineering, PennsylVania State UniVersity, Philadelphia, PennsylVania 16802, and Materials and Process Simulation Center, DiVision of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, California 91125 ReceiVed: July 30, 2009; ReVised Manuscript ReceiVed: December 14, 2009

The effect of the curvature of the carbon support material has been investigated by molecular dynamics simulations of a Pt100 and a Ni100 cluster adsorbed on fishbone-like carbon nanofibers using the ReaxFF reactive force field. Carbon nanocones both with and without hydrogen termination have been considered. Without hydrogen termination, significant differences are found between adsorbed Pt and Ni clusters for the dependence of the bond strain on the curvature. For instance, the support curvature does not seem to have an appreciable effect on the interatomic distances in adsorbed Pt clusters, while increased bond strain is observed for Ni clusters with increasing curvature. Since the bond length is related to a d-band shift as well as to a change in the bond order, it is concluded that the catalytic performance of Ni clusters can be enhanced by optimizing the curvature of the support material. In general, hydrogen termination attenuates the degree of metal-metal bond strain. 1. Introduction The catalytic properties of a metal cluster are modified substantially by anchoring it to support surfaces. For example, gold clusters supported on cerium oxide have been found to have a high activity and stability for the water-gas shift reaction,1 while gold clusters on titanium oxide exhibit an extremely high activity and selectivity for both low temperature CO oxidation and propylene hydrogenation.2,3 The findings of such high activity of supported gold clusters are somewhat surprising since bulk gold has always been regarded to be chemically inert. Furthermore, these observations have opened new possibilities of using selected supports to tailor metal catalysts with high chemical efficiency and selectivity.4 Current polymer-exchange fuel cells use carbon to support catalyst metal clusters of up to 5 Å.5,6 In particular, carbon nanofibers and nanotubes are often employed as supports because of high surface area, inertness in basic and acidic solution (this is not true for alumina or silica supports),7 and high conductivity.8,9 Using different carbon supports, experimental studies have observed changes in the chemical activity for clusters of mainly Pt10–15 and PtRu12,13,16,17 but also for Ni,18 Pd,19 Rh,20 and Co.21 For a review on the subject, see ref 22. Density functional theory (DFT) calculations have been performed on systems comprised of metal clusters and carbon supports such as palladium clusters adsorbed on carbonaceous * To whom correspondence should be addressed. † Centre d’Investigacio´ en Nanociencia i Nanotecnologia. ‡ Department of Chemistry, Norwegian University of Science and Technology. § Department of Chemical Engineering, Norwegian University of Science and Technology. | Pennsylvania State University. # California Institute of Technology.

supports,23 Pt13 and Au13 clusters adsorbed on graphene sheets and flakes,24 the interaction between a Pt atom and one to two graphite planes,25 and the stabilization of Ptn (n ) 1, 6) by borondoped carbon structures.26 These studies are limited to singlepoint energy calculations or to very short simulations. A simulation over the nanosecond time scale is currently computationally too expensive for ab initio methods. Moreover, DFT studies are often limited to cluster sizes of not more than a dozen atoms. In contrast, classical molecular dynamics (MD) simulations can deal with time scales and system sizes many orders of magnitude larger. For instance, MD simulations have been carried out to study the structure and dynamics of 500-atom Pd-based bimetallic nanoclusters adsorbed on graphite,27 a Pt249 cluster adsorbed on carbon nanotube bundles,28 and oxygen adsorption on Pt clusters supported on graphite surfaces.29 In these simulations, a relatively simple Lennard-Jones potential was employed to model the metal-carbon interaction. Although this functional form of the potential energy can be sufficiently good for metal clusters physisorbed on the basal plane of graphite or on the surface of a nanotube, it is not feasible for simulating clusters chemisorbed on graphite platelet nanofibers (p-CNF) or on fishbone nanofibers (f-CNF). In these cases, reactive force fields, often based on the bond-order approach, are preferred. Several schemes have been proposed to include the bond order into interatomic interactions such as the TersoffBrenner force field,30–37 Pettifor’s bond-order potentials (BOP),38–43 and the ReaxFF force field. ReaxFF is based on a bond-order model in conjunction with a charge-equilibration scheme,44 and it was first developed for hydrocarbons.45 Subsequently, the ReaxFF has been extended to many other elements and compounds such as Si/SiO2,46 Al/AlO,47 Ni/Cu/Co/C,48 Mg/ MgH,49 Li/LiC,50 and BiMoOx.51

10.1021/jp907308f  2010 American Chemical Society Published on Web 02/04/2010

Molecular Dynamics Simulations of Metal Clusters Strain in the bond length and the d-band shift are two fundamental concepts related to each other and to reactivity.18,52,53 Therefore, strain in the interatomic distances in metal clusters, such as those resulting from the interaction of a cluster with a carbon surface, can lead to important changes in their catalytic performance. In addition, an increase in the bond length implies a decrease in the bond order resulting in more reactive atoms. To optimize the catalytic activity of a metal cluster, it becomes essential to understand how a surface can induce a strain in the interatomic distances. In previous studies, we studied Pt10054 and Ni10055 clusters adsorbed on different edges of a p-CNF. Significant differences were found between the adsorption of Pt100 and Ni100 clusters. For instance, it was shown that Pt atoms prefer to be adsorbed in four-fold coordinated interlayer sites on both the armchair and the zigzag edges of a p-CNF. In contrast, it was found that Ni atoms prefer to be bound to a single graphene sheet. Considering the platelet edge, nickel atoms are adsorbed on two-fold coordinated sites on the armchair edge and on one-fold coordinated sites on the zigzag structure. As a result, the effect on the bond length in Pt100 clusters is different than in Ni100 clusters. The mean metal-metal bond length increased after adsorption of Pt clusters, while it hardly changed in Ni clusters. In the latter case, the interatomic distance decreased in the regions close to the surface and increased in regions further away in such a way that these opposite effects canceled each other out when the mean bond strain was calculated. An issue we have not addressed before by simulations is the effect of the radius of curvature (ROC) of the carbon support on the support-cluster interaction. Therefore, in this study, we consider two f-CNFs with different ROCs, and we study how the interatomic distances in the metal cluster change as a result of the adsorption on the support. The data from the new simulations are also compared with previous data for p-CNFs since a p-CNF can be regarded as a limiting case of a f-CNF with infinite ROC. 2. Methodology 2.1. Force Field. ReaxFF force fields for the C/Pt/H and C/Ni/H systems have been used in this study.54,55 Briefly, the ReaxFF contains the following energy terms: energy contributions for bond (Ebond), valence angle (Eval), dihedral (Etors), Coulomb (ECoulomb), van der Waals (EvdWaals), three-body conjugation (Ecoa), four-body conjugation (Econj), and lone-pairs (Elp) and penalty terms for stabilization of systems with valence angles between double bonds (Epen) as well as undercoordinated (Eunder) and overcoordinated (Eover) atoms. Thus, the total ReaxFF energy of an atomistic system is given as

Esystem ) Ebond + Eval + Etors + ECoulomb + EvdWaals + Ecoa + Econj + Elp + Epen + Eunder + Eover (1) The bonded interactions have an explicit dependence on a bond-order term that includes contributions from σ and π bonds as well as corrections for undercoordination and overcoordination. Regarding the Coulomb interaction, the atomic charges are calculated at each step of the simulation by a chargeequilibration scheme.44 Thus, the atomic charges are allowed to change as a result of the formation or breaking of covalent bonds. Reference 45 and the supporting information of ref 48 provide the specific analytical expression for each of the energy contributions in eq 1 along with a detailed explanation of the physical meaning. The Pt/C/H and Ni/C/H ReaxFF parameters were taken from refs 54 and 55. These parameters were obtained

J. Phys. Chem. C, Vol. 114, No. 8, 2010 3523 by optimization against DFT calculations performed with the SeqQuest code56,57 using Gaussian basis set, the Perdew-BurkeErnzerhof (PBE) exchange-correlation functional,58 and normconserving pseudopotentials.59 The traning set contained both periodic systems (graphite, diamond, and various bulk metal structures as well as surfaces) and finite systems such as small hydrocarbons adsorbed on face-centered cubic (fcc) slabs with a few dozen fixed metal atoms. Further details on the training set can be found in refs 54 and 55. 2.2. Simulation Details. The MD simulations have been performed with the GRASP code including the ReaxFF force field.60,61 A few trial simulations in the microcanonical ensemble showed that even a time step size of 0.5 fs leads to a drift in the total energy of the system. However, a time step of 0.25 fs was sufficiently small to yield conservation of the energy. Therefore, we took a time step size of 0.25 fs for all the MD simulations. The relatively short time step is caused by the charge-equilibration scheme. For each considered system, once the clusters were placed close to the carbon surface, the energy of the system was minimized. Then, a 15 × 103 fs preproduction stage dragged the system to thermal equilibration at 600 K using a Nose´-Hoover chain thermostat.62 During the production stage, the metal atoms oscillated around their equilibrium positions for an additional time period of 250 × 103 fs in the canonical ensemble. Metal clusters of 100 atoms have been selected to mimic structural changes in a real cluster since this size is approximately halfway between two very stable clusters of 55 and 14763 and, therefore, such a 100-atom cluster is sufficiently susceptible to deformation so that the effects on the cluster structure after adsorption can be clearly observed. Furthermore, the diameter of a 100-atom metal cluster is around 20 Å, which is within the cluster size investigated experimentally for fuel cell catalysis applications (i.e., 10-50 Å).64–66 The initial structure of the cluster was found by an energy minimization on the basis of a genetic algorithm.67 The optimized structures present no symmetries with both (100)- and (111)-fcc facets and some steps on the surface. The carbon nanocones were built using the Nanotube modeler software.68 The bond distance was chosen as 1.42 Å, which is the distance between covalently bonded carbon atoms in graphite.69 There is no unique experimental interlayer spacing among the f-CNFs; it depends on the preparation procedure.70–72 However, all the experimental values reported are quite close to the interlayer distance of graphite of 3.348 Å.73 Therefore, the interlayer distance in graphite was taken as the initial value for the interlayer spacing of the CNFs in this work. Among the five posible closed cone structures,74 a disclination angle of 60° was considered. Two different f-CNFs were formed by piling up nine carbon nanocones with an ROC of 16 and 27 Å, respectively. These ROCs are much smaller than in typical experiments, which are around 100 Å or are even up to a few hundred nanometers.75,76 However, we have considered relatively small ROCs to reduce the computational cost of the simulations considerably as well as to enhance any effect due to the curvature of the carbon support. Periodic boundary conditions were only applied along the direction parallel to the axis of the f-CNFs. Depending on the CNF prepation method, metal clusters can be directly bound to CNFs or can be bound to CNFs terminated with hydrogen from the residual gas in the sputtering chamber. Furthermore, even on hydrogern-terminated CNF surfaces, the hydrogen atoms may be replaced by metal atoms as observed in previous simulations.54 Therefore, we considered both bare

3524

J. Phys. Chem. C, Vol. 114, No. 8, 2010

Sanz-Navarro et al. However, the error bars are only plotted for every five average points to display the trend of the curve clearer.

Figure 1. Top view showing the initial positions of the three Pt100 clusters with respect to a hydrogen-terminated fishbone-like carbon nanofiber with a radius of curvature of 27 Å.

and hydrogen-terminated f-CNFs. In a first set of simulations, a bare f-CNF with an ROC of 27 Å was employed to study the carbon-metal interaction. In a second set of simulations, f-CNFs with hydrogen termination of ROC of 16 and 27 Å were studied. For comparison with our previous work,54,55 data for adsorption of a Pt100 and Ni100 cluster on a p-CNF have been included in some figures. In each simulation, three clusters were adsorbed on an f-CNF as illustrated in Figure 1. Since the carbon support contains both armchair and zigzag segments along the edge, each of the three clusters was adsorbed on a structurely different edge of the nanocones. The first cluster (denoted by Pt100(I) and Ni100(I)) was adsorbed on a zigzag segment, while the second cluster (denoted by Pt100(II) and Ni100(II)) was placed close to an armchair segment, and finally, the third cluster (denoted by Pt100(III) and Ni100(III)) was placed to coincide with the junction between an armchair and a zigzag segment. Furthermore, the found Pt100 and Ni100 cluster structures have no spherical symmetry but have different vertices, edges, and flat faces in such a way that the cluster can be oriented in different ways with respect to the adsorption site of the carbon support. Thus, at the beginning of the simulation, the cluster-I had a face parallel to the carbon surface where it was adsorbed, while cluster-II and cluster-III were initially in contact with the carbon surface through an edge of the cluster. All these combinations are meant to improve the statistics of our results. The degree of strain in the metal-metal bonds after adsorption is investigated by the representation of the distribution of the probability of finding a given metal-metal first-neighbor distance. For the sake of comparison, all the probability distributions are normalized over the represented range [x1, x2] in such a way that ∫xx21P(x)dx ) 1 where x1 ) 2.4 Å and x2 ) 3.2 Å for the Pt100 cluster and x1 ) 2.2 Å and x2 ) 3.1 Å for the Ni100 cluster. The upper limits were chosen to avoid inclusion of second nearest neighbors. During the production stage, the atom positions were recorded at a frequency of 25 fs. To avoid correlated data, we grouped the data within a range of 500 fs, and the statistical error was calculated by the root-mean-square deviation in the resulting sample. The size of the distribution bins is taken as small as 0.0125 Å to get a smooth curve.

3. Results and Discussion 3.1. Pt Clusters Adsorbed to Carbon Nanofibers. For carbon nanocones without hydrogen termination, the three Pt100 clusters adsorbed to an f-CNF underwent substantial restructuring as noted by comparing Figures 1 and 2. The Pt atoms have a preference for a four-fold site between graphite layers and, therefore, several Pt atoms detached from the cluster surface and moved across the carbon substrate leaving some gaps in the cluster-support interface, which were then filled by new Pt atoms. As a result, there is an increase in the number of Pt atoms adsorbed on the carbon support over time. The same restructuring mechanism has been observed for a Pt100 cluster adsorbed on p-CNFs.54 As shown in Figure 3, perceptible differences between the bond-length distribution for Pt100(I) and the other two arrangements are found, while there is much more similarity between the distribution for Pt100(II) and that for Pt100(III). These differences are much more pronounced than those found in a previous work54 where the orientation of the Pt100 cluster with respect to the carbon support was always the same, but the cluster was adsorbed on different edges of the platelet. Therefore, the orientation of the cluster with respect to the support has a noticeable effect on the degree of strain on the metal-metal bond distance. This is explained by the fact that the Pt100 cluster needs to accommodate itself in the carbon support and, since the cluster is not spherically symmetric, the degree of deformation to fit in the carbon structure will depend on the initial shape of the cluster in contact with the carbon support. This result also implies that the degree in the Pt-Pt bond strain is different for other Pt cluster sizes not only because of the differences in the cohesive energy but also because the shape is a function of the number of cluster atoms.77,78 In contrast, Figure 4 shows that almost no difference is found between the bond-length distribution of a Pt100 cluster for adsorption on an f-CNF and that of a p-CNF using the same cluster orientation with respect to the support as in a previous work.54 This seems to suggest that the curvature of the radius has hardly any effect on the bond-length distribution. This can be understood by the fact that the cluster-support contact area extends to a diameter of around 2 Å, which is very small compared to the nanocone diameter. Furthermore, visualization of the carbonaceous structure at the cluster-support interface region showed that the graphitic-like C-C connectivity remained unaltered, although the graphite layers bent slighlty to accommodate the carbon atoms to the metal cluster structure. The same local structure has been observed when carbon platelets are employed as cluster support.54,55 Thus, apart from the disclination angle, locally the platinum cluster does not see a different situation whether it is adsorbed on a p-CNF or on an f-CNF. Although differences in the bond-length distributions may be observed for adsorption on carbon supports with much smaller ROCs, typically CNFs with radii of several tens of nanometers and Pt clusters of average size of 2-3 Å are employed in experiments. Therefore, according to our simulations, we would not expect to observe experimentally any appreciable difference in the reactivity of a Pt cluster when adsorbed on f-CNFs with different ROCs provided that the f-CNFs have been prepared following the same procedure, for example, no residual impurities such as oxygen are left after CNF preparation.79,80 When the edge of the nanocones are hydrogen-terminated, the strain effect is reduced as noticed by comparison of Figure

Molecular Dynamics Simulations of Metal Clusters

J. Phys. Chem. C, Vol. 114, No. 8, 2010 3525

Figure 2. Final snapshots of the simulation of the adsorption of the three Pt100 clusters adsorbed directly on a hydrogen-terminated fishbone-like carbon nanofiber with a radius of curvature of 27 Å.

Figure 3. Distribution of the probability for Pt-Pt first-neighbor distance. The bond-length distribution of an isolated Pt100 cluster is also depicted to highlight that the bond length changes considerably after adsorption on the carbon support.

Figure 4. Comparison of bond-length distributions for two simulations with the same cluster-support orientation but with different curvature of the carbon support. One of the Pt100 clusters was adsorbed on a zigzag edge of the fishbone carbon nanofiber, while the other cluster was adsorbed on the armchair edge of a platelet.

3 and Figures 5 and 6. In particular, in Figure 5, the distribution for the adsorbed Pt100(I) cluster on the hydrogen-terminated 16 Å ROC f-CNF hardly differs from the distribution of an isolated Pt100 cluster (Figure 3), and the other two distributions are still far from that correspondent to a cluster adsorbed directly on the fishbone-type support (Figure 3). For adsorption on a hydrogen-terminated 27 Å ROC f-CNF (Figure 6), the hydrogen

Figure 5. Comparison of the bond-length distributions for the three Pt100 clusters adsorbed on a fishbone-like carbon nanofiber with a radius of curvature of 16 Å.

Figure 6. Comparison of the bond-length distributions for the three Pt100 clusters adsorbed on a hydrogen-terminated fishbone-like carbon nanofiber with a radius of curvature of 27 Å.

passivation of the carbon support is even more evident since the three distributions are very similar to the one for the isolated cluster (Figure 3). In all simulations with hydrogen termination, a few hydrogen atoms were desorbed from the carbon substrate and were replaced by platinum atoms. A H-Pt substitution was also favorable in previous simulations of adsorption on hydrogenterminated p-CNFs,54 and as already discussed, a time scale of a nanosecond is not sufficient to reach thermodynamic equilibrium for this event. For example, after longer simulation

3526


Sanz-Navarro et al.

Figure 7. Comparison of the bond-length distributions for Pt100 clusters adsorbed on hydrogen-terminated carbon nanofibers with three different radii of curvature (16 Å, 27 Å, and ∞). The data represented have been obtained as an average over the different clusters simulated. For the data referred to adsorption on nanocones, data have been calculated by the average over the bond-length distributions for the three adsorbed clusters represented in Figure 5 and 6. In the case of adsoprtion on a carbon platelet, an average over the bond-length distributions of a Pt100 adsorbed on both the armchair and zigzag edges is considered.

TABLE 1: Number of Adsorbed Pt and Ni Atoms at the End of the Preproduction Stage for Thermal Equilibration at the Temperature of 600 K

M ) Pt M ) Ni

M100(I) @ cone (27 Å)

M100(II) @ cone (27 Å)

M100(III) @ cone (27 Å)

30 51

21 41

27 38

times, a larger quantity of hydrogen atoms may be detached from the surface allowing new Pt atoms to be adsorbed on the support and, consequently, leading to an increase in the Pt-Pt first-neighbor distances. However, visualization of the supportcluster interface reveals that hydrogen termination prevents the support-cluster contact area from expanding. Such an expansion has been proposed as an explanation for the considerable change in the bond-length distribution.54,55 Therefore, the passivation of the carbon support is expected to be appreciable over experimental time scales. In terms of averages (see Figure 7), the close similarity between the bond-length distribution for a hydrogen-terminated f-CNF with the lowest diameter (i.e., 16 Å) and that of a hydrogen-terminated p-CNF (i.e., infinite ROC) provides new evidence of a small effect of the ROC on the bond-length distribution. In fact, differences of these two distribution with that of the clusters adsorbed on a 27 Å ROC f-CNF are more likely to be due to the limited simulation time scales rather than to an effect of the ROC. 3.2. Ni Clusters Adsorbed to Carbon Nanofibers. Unlike the Pt atoms, each adsorbed Ni atom binds to a single graphite layer. Ni clusters adsorbed directly onto the carbon substrate and underwent much more restructuring than adsorbed Pt clusters resulting in a substantial wetting across the support. As indicated by comparison of the number of adsorbed atoms at the end of the preproduction stage (see Table 1), many more Ni atoms than Pt atoms are adsorbed on the support. In addition, a common neighbor analysis (CNA),81 reported in Figure 8, reflects that most of the atoms have lost the fcc-like structure after adsorption thus increasing the disorder of the internal structure substantially. This greater deformation of the structure

Figure 8. CNA signatures for Ni100 clusters adsorbed on a fishbone carbon nanofiber with a radius of curvature of 27 Å. A full explanation of the notation employed to represent local structure by CNA signatures can be found in ref 81. Briefly, “?” denotes a local structure different than any of the following ones: a, fcc bulk; b, fcc (100) surface; c, fcc (111) surface; d, fcc (111)-(100) edge; e, fcc(111)-(111) edge; f, truncated octahedron; g, icosahedral internal twinning plane; h, icosahedral spine; i, icosahedral surface edge; j, icosahedral central atom; k, icosahedral surface vertex or decahedral notch vertex; l, truncated icosahedral vertex or decahedral notch vertex; and m, decahedral notch edge.

of a Ni cluster adsorbed on an f-CNF can be understood by taking into account that the binding energy per atom is considerably lower in a Ni100 cluster (ReaxFF energy per atom ) 89 kcal/mol) than in a Pt100 cluster (ReaxFF energy per atom ) 113 kcal/mol). Moreover, the Ni100 cluster acquires the local curvature of the support (Figure 9). In particular, its shape is like a clam-shell droplet adsorbed on a fiber82,83 although interestingly the cluster mainly elongates along the armchair segment and the direction perpendicular to the zigzag segment. In contrast to an incomprensible liquid, the Ni-Ni interatomic distances vary as compared to the isolated cluster. As displayed in Figure 10, the differences between the bond-length distribution for the three adsorbed clusters are minimal. Similar to the case of adsorption on a p-CNF, Figure 11, the adsorption of the cluster on an f-CNF results in a widening of the bond-length


J. Phys. Chem. C, Vol. 114, No. 8, 2010 3527

Figure 9. Final snapshots of the simulation of the adsorption of the three Ni100 clusters adsorbed directly on the fishbone carbon nanofiber with a radius of curvature of 27 Å.

Figure 10. Comparison of the bond-length distributions for the three Ni100 clusters adsorbed on a fishbone carbon nanofiber with a radius of curvature of 27 Å.

Figure 11. Comparison of the bond-length distributions for Ni100 clusters adsorbed. The data represented have been obtained as an average over the different clusters simulated. For the data referred to as adsorption on a nanocone, data have been calculated by the average over the bond-length distributions for the three adsorbed clusters represented in Figure 10. In the case of a platelet, the averages over bond-length distributions for Pt100 adsorbed on both the armchair and zigzag edges are considered.55

distribution. Interestingly, when the CNA and bond-length distributions are contrasted for each individual simulation, it is found that the widening of the bond-length distribution increases with the degree of disorder measured by the number of atoms with unclassified CNA signature (i.e., none of the typical local structures in fcc metals) as illustrated in Figure 12. This figure also shows that the broadening of the bond-length distribution

Figure 12. Evolution of σBL for a Ni100 cluster as a function of the number of Ni atoms with unclassified structure (i.e., CNA signature denoted by ? in Figure 8). Circles represent data from simulations of adsorption on a fishbone carbon nanofiber with a radius of curvature of 27 Å, and squares represent data from simulations of adsorption of a Ni100 on a carbon platelet.55

and the degree of disorder increases with the ROC. Compared with adsorption on a p-CNF, Figure 11, the probability for longer Ni-Ni bonds is enhanced by the curvature of the nanocones. Therefore, the MD results thus suggest that the catalytic activity of a Ni100 cluster can be enhanced by supporting the cluster directly on a nanofiber of high curvature. Regarding hydrogen-terminated f-CNFs and in contrast to Pt, the Ni atoms are not able to substitute the hydrogen atoms bound to the carbon nanocones in such a way that there are not any covalent bonds established between C and Ni atoms. This was observed already for a Ni100 cluster adsorbed on a p-CNF,55 and it is related to the fact that the C-Ni bond is not stronger than the H-C bond: at the DFT level, the C-H dissociation energy for both a zigzag and an armchair configuration was calculated to be 107 kcal/mol,84 while the C-Ni dissociation energy was estimated to be 80 kcal/mol by a curve-fitting technique using a Lippincott potential function.85 There is still a controversy86 about the actual strength of the Ni-C bond, but the experimental and calculated data up to date shows that the Ni-C bond is not stronger than the C-H bond. As for the Ni-Ni distance, both Figures 13 and 14 indicate that the Ni-Ni bond-length distribution depends somewhat on cluster orientation as well as on binding site. By comparison of the bondlength distribution for direct adsorption on an f-CNF and that of adsorption on a hydrogen-terminated f-CNF (Figure 15), it is once more confirmed that the strain effect is attenuated by

3528


Sanz-Navarro et al.

Figure 13. Comparison of the bond-length distributions for the three Ni100 clusters adsorbed on a hydrogen-terminated fishbone carbon nanofiber with a radius of curvature of 16 Å.

Figure 14. Comparison of the bond-length distributions for the three Ni100 clusters adsorbed on a hydrogen-terminated fishbone carbon nanofiber with a radius of curvature of 27 Å.

Figure 15. Comparison of the bond-length distributions for Ni100 clusters adsorbed on carbon nanofibers with three different radii of curvature (16 Å, 27 Å, and ∞). The data represented have been obtained as an average over data for different simulations. Data for an isolated Ni100 have also been added to visualize the strain effect on the bond length after adsorption.

the presence of hydrogen at least at the time scale reached in our MD simulations. Interestingly, for hydrogen-terminated supports (Figure 15), the strain effect decreases as the ROC decreases. This is the reverse effect to what is observed for direct binding to the carbon support (cf. Figure 11). Visualization of several MD snapshots revealed that the nickel cluster does not wet the hydrogen-terminated support surface, but the cluster retains the initial overall shape, and then there is a minimal

Figure 16. CNA signatures for Ni100 clusters adsorbed on a hydrogenterminated fishbone carbon nanofiber with a radius of curvature of 16 Å.

support-cluster contact. Since the contact area decreases with the ROC, less Ni atoms need to be accommodated in stable positions and hence a less significant deformation of the cluster can be expected for lower ROC. In turn, this is translated into a decrease in the degree of Ni-Ni strain. The importance of the contact area also becomes plausible when the number of Ni atoms with unclassified structure is compared between simulations for two different ROCs (cf. Figures 16 and 17). The clusters adsorbed on a hydrogen-terminated support with ROC of 16 Å have undergone a major loss of the initial structure compared to those supported on a hydrogen-terminated f-CNF of ROC of 27 Å. Once more, there is a clear relation between the amorphitization of the cluster and the broadening of the standard deviation in the bond-length distribution, σBL. For instance, for the 16 Å ROC support, the adsorbed Pt100(II) presents the broadest bondlength distribution (Figure 16) and, at the same time, the highest number of atoms with unclassified structure (Figure 16). In contrast, the adsorbed Pt100(I) presents both the narrowest bondlength distribution and the lowest number of unclassified CNA atoms (Figure 16). Similar correspondence between σBL and the number of atoms with unclassified structure is found for the 27 Å ROC support, Figure 17. Indeed, the representation of the CNA distribution (Figures 16 and 17) makes apparent a correlation with the structural disorder. Figure 18 shows that the trend in the bond-length standard deviation as a function of


J. Phys. Chem. C, Vol. 114, No. 8, 2010 3529 4. Conclusions This work demonstrates a very different effect of the curvature of the nanofiber on the bond strain induced in adsorbed Pt100 and Ni100 clusters, which can be explained by the difference in the strength of the metal-metal binding energy inside the cluster and the difference in the preferential binding sites: Pt atoms move to graphite interlayer sites, binding to two different graphite layers, whereas each Ni atom binds to a single graphite layer. The curvature of the carbon support has no appreciable effect on the bond-length distribution for the Pt clusters. In contrast, the degree of bond strain depends on the curvature of the carbon support for Ni clusters. This is the case for both direct binding and adsorption on hydrogen-terminated support surfaces. Interestently, the degree of bond strain in the Ni cluster increases with curvature for direct adsorption on a carbon nanofiber, while the reverse effect is found for adsorption on hydrogen-terminated carbon nanofibers. Because bond length in the cluster is related to a shift in the d-band and, in turn, to the catalytic performance, our work indicates that while the reactivity of platinum clusters is not expected to be improved by using support surfaces of different radii of curvature, the turnover frequency of reactions catalyzed by nickel clusters can be enhanced using supports with different curvatures. Acknowledgment. C.F.S.N., P.-O. Å., D.C., and M.R. have received support from the Norwegian Research Council through a Nanomat program “FUNMAT: Materials for Hydrogen Technology”, project number 158516/S10. C.F.S.N. and P.-O. Å. have received a grant of computer time from the Norwegian Research Council and NTNU. C.F.S.N. also acknowledges the Ministerio de Ciencia y Tecnologıá for a Ramoń y Cajal contract and additional funding (grant no. FIS2008-02187).

Figure 17. CNA signatures for Ni100 clusters adsorbed on a hydrogenterminated fishbone carbon nanofiber with a radius of curvature of 27 Å.

Figure 18. σBL for a Ni100 cluster adsorbed on hydrogen-terminated supports as a function of the number of Ni atoms with unclassified structure (i.e., CNA signature denoted by ? in Figures 16 and 17). Circles represent data from simulations of adsorption on a fishbone carbon nanofiber with a radius of curvature of 16 Å, and crosses represent data from simulations of adsorption on a fishbone carbon nanofiber with a radius of curvature of 27 Å. In addition, data for an isolated Ni100 cluster is represented by a diamond.

the number of atoms with unclassified CNA signature can be fitted to a straight line fairly well.

References and Notes (1) Fu, Q.; Weber, A.; Flytzani-Stephanopoulos, M. Chem. Lett. 2001, 77, 87–95. (2) Valden, M.; Lai, X.; Goodman, D. W. Science 1998, 281, 1647– 1650. (3) Chou, J.; Franklin, N.; Baeck, S.-H.; Jaramillo, T.; McFarland, E. Chem. Lett. 2004, 95, 107–111. (4) Ertl, G.; Freund, H.-J. Phys. Today 1999, 52, 32–38. (5) Larminie, J.; Dicks, A. Fuel Cell Systems Explained; John Wiley & Sons, Ltd.: Chichester, U.K., 2002. (6) Shao, Y.; Yin, G.; Wang, Z.; Gao, Y. J. Power Sources 2007, 167, 235–242. (7) Stiles, A. B. Catalysts Supports and Supported Catalysts; Butterworths: Boston, MA, 1987. (8) Melechko, A. V.; Merkulov, V. I.; McKnight, T. E.; Guillorn, M. A.; Klein, K. L.; Lowndes, D. H.; Simpson, M. L. J. Appl. Phys. 2005, 97, 041301. (9) Auer, E.; Freund, A.; Pietsch, J.; Tacke, T. Appl. Catal., A 1998, 173, 259–271. (10) Kim, H.; Lee, W.; Yoo, D. Electrochim. Acta 2007, 52, 2620– 2624. (11) Bessel, C.; Laubernds, K.; Rodriguez, N. M.; Baker, R. T. K. J. Phys. Chem. B 2001, 105, 1115–1118. (12) Liu, Z. L.; XY, X. Y. L.; Guo, B.; Hong, L.; Lee, J. Y. J. Power Sources 2007, 167, 272–280. (13) Selvaraj, V.; Alagar, M. Electrochem. Commun. 2007, 9, 1145– 1153. (14) Kim, S.; Park, S. J. J. Solid State Electrochem. 2007, 11, 821– 828. (15) Maruyama, J.; Abe, I. Electrochemistry 2007, 75, 119–121. (16) Park, I. S.; Park, K. W.; Choi, J. H.; Park, C. R.; Sung, Y. E. Carbon 2007, 45, 28–33. (17) Qi, J.; Gao, Y.; Tang, S. H.; Jiang, L. H.; Yan, S. Y.; Guo, J. S.; Q, Q. X.; Sun, G. Q. Chin. J. Catal. 2006, 27, 708–712. (18) Ochoa-Fernańdez, E.; Chen, D.; Yu, Z.; Tøtdal, B.; Rønning, M.; Holmen, A. Surf. Sci. 2004, 554, L107–L112. (19) Zhenga, J.-S.; Zhang, X.-S.; Lia, P.; Zhua, J.; Zhoua, X.-G.; Yuana, W.-K. Electrochem. Commun. 2007, 9, 895–900.

3530


(20) Giordano, R.; Serp, P.; Kalck, P.; Kihn, Y.; Schreiber, J.; Marhic, C.; Duvail, J.-L. Eur. J. Inorg. Chem. 2003, 610–617. (21) Liu, Z. J.; Xu, Z.; Yuan, Z. Y.; Lu, D. Y.; Chen, W. X.; Zhou, W. Z. Chem. Lett. 2001, 72, 203–206. (22) Serp, P.; Corrias, M.; Kalck, P. Appl. Catal., A 2003, 253, 337– 358. (23) Duca, D.; Ferrante, F.; Manna, G. L. J. Phys. Chem. C 2007, 111, 5402–5408. (24) Okamoto, Y. Chem. Phys. Lett. 2006, 420, 382–386. (25) Kong, K.-J.; Choi, Y.; Ryu, B.-H.; Lee, J.-O.; Chang, H. Mater. Sci. Eng., C 2006, 26, 1207–1210. (26) Acharya, C. K.; Turner, C. H. J. Phys. Chem. B 2006, 110, 17706– 17710. (27) Sankaranarayanan, S. K. R. S.; Bhethanabotla, V. R.; Joseph, B. Phys. ReV. B 2005, 72, 195405. (28) Morrow, B. H.; Striolo, A. J. Phys. Chem. C 2007, 111, 17905– 17913. (29) Wu, G.-W.; Chan, K.-Y. J. Electroanal. Chem. 1998, 450, 225– 231. (30) Brenner, D. W. Phys. ReV. B 1990, 42, 9458–9471. (31) Beardmore, K.; Smith, R. Philos. Mag. A 1996, 74, 1439–1466. (32) Ni, B.; Lee, K.-H.; Sinnott, S. B. J. Phys.: Condens. Matter 2004, 16, 7261–7275. (33) Erhart, P.; Juslin, N.; Goy, O.; Nordlund, K.; Mu¨ller, R.; Albe, K. J. Phys.: Condens. Matter 2006, 18, 6585–6605. (34) Juslin, N.; Erhart, P.; Tra¨skelin, P.; Nord, J.; Henriksson, K. O. E.; Nordlund, K.; Salonen, E.; Albe, K. J. Appl. Phys. 2005, 98, 123520. (35) Erhart, P.; Albe, K. Phys. ReV. B 2005, 71, 035211. (36) Nord, J.; Albe, K.; Erhart, P.; Nordlund, K. J. Phys.: Condens. Matter 2003, 15, 5649–5662. (37) Albe, K.; Nordlund, K.; Averback, R. S. Phys. ReV. B 2002, 65, 195124. (38) Pettifor, D. G. Phys. ReV. Lett. 1989, 63, 2480–2483. (39) Drautz, R.; Murdick, D. A.; Nguyen-Manh, D.; Zhou, X. W.; Wadley, H. N. G.; Pettifor, D. G. Phys. ReV. B 2005, 72, 144105. (40) Drautz, R.; Pettifor, D. G. Phys. ReV. B 2006, 74, 174117. (41) Aoki, M.; Nguyen-Manh, D.; Pettifor, D. G.; Vitek, V. Prog. Mater. Sci. 2007, 52, 154–195. (42) Drautz, R.; Zhou, X. W.; Murdick, D. A.; Gillespie, B.; Wadley, H. N. G.; Pettifor, D. G. Prog. Mater. Sci. 2007, 52, 196–229. (43) Gillespie, B. A.; Zhou, X. W.; Murdick, D. A.; Wadley, H. N. G.; Drautz, R.; Pettifor, D. G. Phys. ReV. B 2007, 75, 155207. (44) Rappe´, A. K.; Goddard, W. A., III. J. Phys. Chem. 1991, 95, 3358– 3363. (45) van Duin, A. C. T.; Dasgupta, S.; Lorant, F.; Goddard, W. A., III. J. Phys. Chem. A 2001, 105, 9396–9409. (46) van Duin, A. C. T.; Strachan, A.; Stewman, S.; Zhang, Q.; Xu, X.; Goddard, W. A., III. J. Phys. Chem. A 2003, 107, 3803–3811. (47) Zhang, Q.; Çan, T.; van Duin, A.; Goddard, W. A., III; Qi, Y.; Hector, L. G., Jr. Phys. ReV. B 2004, 69, 045423. (48) Nielson, K. D.; van Duin, A. C. T.; Oxgaard, J.; Deng, W.-Q.; Goddard, W. A., III. J. Phys. Chem. A 2005, 109, 493–499. (49) Cheung, S.; Deng, W.-Q.; van Duin, A. C. T.; Goddard, W. A., III. J. Phys. Chem. A 2005, 109, 851–859. (50) Han, S. S.; van Duin, A. C. T.; Goddard, W. A., III; Lee, H. M. J. Phys. Chem. A 2005, 109, 4575–4582. (51) Goddard, W. A., III; van Duin, A. C. T.; Chenoweth, K.; Cheng, M.-J.; Pudar, S.; Oxgaard, J.; Merinov, B.; Jang, Y. H.; Persson, P. Top. Catal. 2006, 38, 93–103. (52) Sakong, S.; Groβ, A. Surf. Sci. 2003, 525, 107–118. (53) Hammer, B.; Nørskov, J. K. AdV. Catal. 2000, 45, 71–129. (54) Sanz-Navarro, C. F.; Åstrand, P.-O.; Chen, D.; Rønning, M.; van Duin, A. C. T.; Jacob, T.; Goddard, W. A., III. J. Phys. Chem. A 2008, 112, 1392–1402.

Sanz-Navarro et al. (55) Sanz-Navarro, C. F.; Åstrand, P.-O.; Chen, D.; Rønning, M.; van Duin, A. C. T.; Mueller, J. E.; Goddard, W. A., III. J. Phys. Chem. C 2008, 112, 12663–12668. (56) Schultz, P. Unpublished work. A description of the method is in Feibelman, P. J. Phys. ReV. B 1987, 35, 2626. (57) Verdozzi, C.; Schultz, P. A.; Wu, R.; Edwards, A. H.; Kioussis, N. Phys. ReV. B 2002, 66, 125408. (58) Perdew, J. P.; Burke, K.; Ernzerhof, M. C. Phys. ReV. Lett. 1996, 77, 3865–3868. (59) Hamann, D. R. Phys. ReV. B 1989, 40, 2980–2987. (60) Nakano, A.; Kalia, R. K.; Nomura, K.-I.; Sharma, A.; Vashishta, P.; Shimojo, F.; van Duin, A. C. T.; Goddard, W. A., III; Biswas, R.; Srivastava, D. Comput. Mater. Sci. 2007, 38, 642–652. (61) Nomura, K.-I.; Kalia, R. K.; Nakano, A.; Vashishta, P. Comput. Phys. Commun. 2008, 178, 73–87. (62) Martyna, G. J.; Klein, M. L.; Tuckerman, M. J. Chem. Phys. 1992, 97, 2635–2643. (63) Paulus, P. M.; Goossens, A.; Thiel, R. C.; van der Kraan, A. M.; Schmid, G.; de Jongh, L. J. Phys. ReV. B 2001, 64, 205418. (64) Lee, K.; Zhang, J.; Wang, H.; Wilkinson, D. P. J. Appl. Electrochem. 2006, 36, 507–522. (65) Tang, J. M.; Jensen, K.; Waje, M.; Li, W.; Larsen, P.; Pauley, K.; Chen, Z.; Ramesh, P.; Itkis, M. E.; Yan, Y.; Haddon, R. C. J. Phys. Chem. C 2007, 111, 17901–17904. (66) Kvande, I.; Briskeby, S. T.; Tsypkin, M.; Rønning, M.; Sunde, S.; Tunold, R.; Chen, D. Top. Catal. 2007, 45, 81–85. (67) Hobday, S.; Smith, R. J. Chem. Soc., Faraday Trans. 1997, 93, 3919–3926. (68) Webber, S. JCrystalSoft; 2006. (69) Brown, T. L.; LeMay, J. H. E.; Bursten, B. E.; Burdge, J. B. Chemistry: The Central Science, 10th ed.; Prentice Hall: Upper Saddle River, NY, 2005. (70) Tribolet, P.; Kiwi-Minsker, L. Catal. Today 2005, 102-103, 15– 22. (71) Gao, Y.; He, P.; Lian, J.; Schulz, M. J.; Zhao, J.; Wang, W.; Wang, X.; Zhang, J.; Zhou, X.; Shi, D. J. Appl. Polym. Sci. 2006, 103, 3792– 3797. (72) Ren, W. C.; Cheng, H. M. Carbon 2003, 41, 1657–1660. (73) Wyckoff, R. G. Crystal Structures, 2nd ed.; Krieger Publishing Company: FL, 1982; Vol. 1. (74) Eks¸iogˇlu, B.; Nadarajah, A. Carbon 2006, 44, 360–373. (75) Yuan, F. L.; Yu, H. K.; Ryu, H. J. Electrochim. Acta 2007, 50, 685–691. (76) Hacker, V.; Wallno¨fer, E.; Baumgartner, W.; Schaffer, T.; Besenhard, J.; Schro¨ttner, H.; Schmied, M. Electrochem. Commun. 2005, 7, 377– 382. (77) Nie, A. H.; Wu, J. P.; Zhou, C. G.; Yao, S. J.; Forrey, R. C.; Cheng, H. S. Int. J. Quantum Chem. 2007, 107, 219–224. (78) Xiao, L.; Wang, L. C. Int. J. Quantum Chem. 2004, 108, 8605– 8614. (79) Zaragoza-Martıń, F.; Sopena-Escario, D.; Moralloń, E.; de Lecea, C. S.-M. J. Power Sources 2007, 171, 302–309. (80) Kvande, I.; Chen, D.; Zhao, T.-J.; Skoe, I. M.; Walmsley, J. C.; Rønning, M. Top. Catal. 2009, 52, 664–674. (81) Cleveland, C. L.; Luedtke, W. D.; Landman, U. Phys. ReV. B 1999, 60, 5065–5077. (82) McHalle, G.; Newton, M. Colloids Surf. 2002, 206, 79–86. (83) Berim, G. O.; Ruckenstein, E. J. Phys. Chem. B 2005, 109, 12515– 12524. (84) May, K.; Dapprich, S.; Furche, F.; Unterreiner, B. V.; Ahlrichs, R. Phys. Chem. Chem. Phys. 2000, 2, 5084–5088. (85) Rao, P. S.; Reddyb, R.; Potukuchic, D. J. Quant. Spectrosc. Radiat. Transfer 2006, 98, 81–84. (86) Tzeli, D.; Mavridisa, A. J. Chem. Phys. 2007, 126, 194304.

JP907308F

Molecular Dynamics Simulations of Metal Clusters

Recommend Documents