Direct observation of chemisorption induced changes in concentration profile in Pd–Au alloy nanosystems via in situ X-ray powder diffraction

Abstract

Pd–Au nanocrystalline systems supported on silica were exposed to different gas environments and studied by in-situ XRD. Subtle changes were detected in the XRD patterns when after heating in dry air and flushing with argon a short pulse of H₂ was injected into the stream of argon. These changes were interpreted in terms of an inversion of concentration profile in Pd–Au particles from Pd segregation induced by oxygen chemisorption to Au segregation for the clean surface. Full atomistic models of the alloy nanoparticles for sizes of up to 5 nm were computed using Sutton–Chen potentials, configurational minimization and molecular dynamics. Thus it was possible to simulate the inversion of the concentration profile and its effect on the XRD pattern. The results of the simulation allowed qualitative analysis of the experimental XRD evolution. A specially designed in situ camera together with a new method of process monitoring and analysis allowed direct structural observation of the dynamics of change of the concentration profile within the nanoparticles. It is demonstrated that XRD, when aided with atomistic simulations, can be suitable to follow in situ evolution of the surface structure of nanopowders.

---

1. Introduction

The surface composition of alloys very often differs from the bulk. This phenomenon usually obeys a few simple rules: (i) on annealing the surface becomes enriched in the element of lower heat of sublimation, (ii) ion bombardment depletes at the surface the element of higher rate of sputtering, (iii) on chemisorption the element more strongly bonding with the adsorbate segregates to the surface.¹

An insight into actual surface composition and morphology of a bimetallic particle in a reaction environment is especially desirable in catalytic studies. To understand the mechanism of the catalytic reaction one has to rely on some assumptions of the atomistic structure of the surface which may be complex and varying during the reaction.

The experimental studies of the phenomenon of surface segregation mostly come from surface science (e.g. Auger electron spectroscopy)^1–5 and EXAFS technique.^6,7 Among these only EXAFS allows studies in the reaction environment. However the results are not easy to interpret and are often obscured by not well determined experimental factors.

The Pd–Au bimetallic system plays an important role in catalysis. It is used as an effective catalyst for naphtha reforming, reduction of pollution from car engines and in chemical manufacturing (e.g. of vinyl acetate from ethylene, acetic acid and oxygen). The system shows no miscibility gap (above 300 K) for the whole range of concentrations, forming continuous solid solutions, which eases interpretation of the experimental data.

An important factor affecting the surface composition of the alloy is its dispersion, expressed commonly as the fraction of atoms exposed at the surface. In alloy nanoparticles the interdiffusion of the constituent metal atoms is quicker and, because of a high contribution of surface atoms, the possible rearrangement of the atoms as a response to changing gaseous atmosphere is more pronounced.

In the bimetalic Pd–Au system gold is characterized by a slightly lower heat of sublimation and surface tension than palladium, thus theory predicts its segregation to the surface.⁸ However the more reactive element is palladium, and in presence of any chemisorbing species (impurities, reactants) in most cases it should dominate at the surface.

This general conclusion is not fully proved experimentally. The Pd surface enrichment due to oxygen chemisorption was reported using Auger electron spectroscopy (AES) for a Pd-rich alloy and to a lesser extent for equal Pd and Au concentration,³ and for an alloy of 20 at.% Pd.¹ For clean surfaces however, neither Au enrichment¹ nor Au segregation⁵ was found. It should be stressed however that in AES studies the samples are usually in bulk form which means that e.g. only 1 ppm of impurities (in volume) when segregated at the surface can cover the whole surface. For nanocrystals the same level of impurities may result in less than one crystallite surface per thousand being affected. Also the kinetic barrier due to the interdiffusion rate of the alloy elements may be much less significant in the case of nanocrystals.

Some light on the surface segregation in Pd–Au alloys in nanocrystalline form was shed by using EXAFS and XANES techniques,^6,7 but the whole picture emerging from these studies is far from being understood. For alloy particles of 1 nm size (Pd/Au atomic ratio 1∶1) supported on silica and unexposed to air, EXAFS data are consistent with the model of a nanoparticle forming a small Au core decorated with Pd atoms so that some Pd atoms are surrounded only by Au nearest neighbors. Oxygen atoms irreversibly chemisorbed on the nanoparticle surface appear to be attached to only these Pd atoms.⁶

In turn, EXAFS studies of Pd–Au particles of 3 nm size supported on silica showed that Pd atoms were segregated to the surface or Pd was homogenously distributed with Au, depending on the sample preparation and treatment.⁷ The observed structural differences between these samples were attributed by the authors to kinetic rather than thermodynamic factors. To this end, Lee at al.⁹ studying bimetallic colloid particles of 6 nm size, suggested that significant intermixing of Pd and Au atoms does not occur at temperatures below 125 °C, and the complete intermixing requires temperatures approaching 300 °C.

Regarding the oxygen chemisorption it is worth to mentioning the study by Weissman-Wenocur and Spicer¹⁰ who noticed that the saturation coverage of O at RT on the (111) surface of a Pd–Au (4∶1) single crystal is reduced by a factor of five in comparison to that on a Pd(111) single crystal. This means that Au atoms at the surface effectively exclude some Pd atoms from chemisorbing oxygen.

The objective of the present work is to record in in-situ conditions at RT and ambient pressure the subtle changes of XRD pattern of Pd–Au alloy nanosystems that occur with changing gaseous atmosphere and to interpret them structurally in terms of changing element concentration profile on the basis of reliable atomistic models. The materials used in this study were formerly used for catalytic investigations and are well characterized in the literature cited below. The current work is thus focused on the data obtained solely from in situ powder diffraction. Prior to the measurements the samples were subjected to thermal and chemical treatment to assure good intermixing of the elements, and the data was recorded over extended periods of time to observe changes driven by thermodynamics of the processes, rather than kinetic factors.

The in situ powder diffraction technique presented in this work and in former studies^11–14 proves to be a unique tool to follow subtle modifications of the surface structure of nanocrystalline metals during chemical processes at normal atmospheric pressure. Some attempts to interpret changes in peak intensity have been undertaken previously^15,16 but without atomistic simulations as a basis to interpret them.

2. Experimental

Two samples were used in this study:

(1) 10 wt% Pd–Au (7∶3)/SiO₂ – a sample of the material described by Juszczyk at al.,¹⁷ XRD particle size 13 nm;

(2) 5 wt% Pd–Au (1∶1)/SiO₂ – a sample kindly provided by Prof. G. C. Bond (Brunel University of West London), prepared by coimpregnation of Davison 70 SiO₂ gel with metal aqueous solutions of respective salts,¹⁸ XRD particle size 8 nm.

The samples had a dominating alloy phase but showed also a residual palladium phase readily transforming into β-palladium hydride in H₂ atmosphere (RT, 1 atm) and into PdO in dry air at 673 K. The particle sizes listed above are for prereduced samples after heating in dry air at 673 K.

The experimental procedure consisted in collecting in situ a large series of powder diffraction patterns over extended times, following the general treatment:

(1) reduction of sample in flow of H₂, RT,

(2) flushing with argon at RT,

(3) heating in flow of dry air from RT to 673 K (3 K min⁻¹ temperature ramp) followed by quick cooling,

(4) evacuating the X-Ray camera and flushing with argon at RT,

(5) injection of a short pulse of H₂ to the stream of argon.

The XRD patterns were recorded with a Siemens 5000 generator with a Cu sealed-off tube using an INEL CPS120 position sensitive detector (PSD), a flat graphite monochromator in the incident beam and a laboratory-made flow camera for in situ studies. Since the measurements require good thermal stability of the angular response of the PSD, suitable tests were done as earlier presented.¹² The camera, as described previously,¹² is working in a flat sample non-focusing geometry, and is covered with a stainless-steel cap equipped with a beryllium vacuum-proof X-ray window. This was especially important for assuring full exchange of gases and providing an oxygen free atmosphere.

Columns filled with MnO/SiO₂ as an oxygen trap were used to supply oxygen-free inert gas (Ar) and to monitor the oxygen content in the outlet gases. One column was specially calibrated to allow detection of traces of oxygen in the outlet gases with an accuracy of up to 10⁻⁸ mol O₂. This accuracy is crucial bearing in mind that the amount of oxygen sufficient for monolayer coverage of the surface of the nanocrystalline metal is of the order of 10⁻⁵ mol. The procedure referred to as “flushing with argon” consisting of pumping out the air from the camera (pressure less than 10⁻⁴ Tr), filling it with argon (passed through MnO/SiO₂ trap) at a rate of 50 ml min⁻¹ and then flushing with argon at a rate of 10 ml min⁻¹, resulted in no measurable traces of oxygen in the outlet stream when monitored for the subsequent 30 min.

The feeding with dry air was achieved by applying a cold trap (solid CO₂) in the gas line. The construction of the XRD camera allows collection of powder diffraction data in the angular range 0–100° (2θ) and heating up to 873 K in controlled flow of reaction gases.

The diffraction patterns were usually collected every 10 min and the whole sequence run for a few days. The set of the diffraction data was then analyzed focusing on one chosen profile in a selected angular range. Every profile from the set was fitted to the gaussian (or lorentzian) curve and its position, peak width and amplitude were reviewed along the history of the sample treatment. The analysis was restricted to the 111 peak, this being the strongest and assuring the best counting statistics. Consequently the atomistic model calculations to be compared with experimental data were also performed for the 111 peak and the analysis presented thereafter refers to this reflection.

The procedure described above was already successfully applied to record the nanocrystalline Pd(111) peak before and after oxygen chemisorption and to interpret its shift as due to the metal surface relaxation.¹³ The measured and simulated via atomistic modeling (for Cu-Kα wavelength) shift was equal to 0.15° for 2 nm Pd nanocrystals and quickly dropped to 0.05° for 4 nm nanocrystals.¹³

Obtaining homogeneous and monophase alloy particles is unlikely. Each of the studied samples in this work had some amount of Pd nanometallic phase. The idea was to transform this phase into PdO, so that in the XRD pattern, the neighborhood of the alloy 111 reflection is free of other contributions. For this purpose each sample was heated in dry air to 673 K (3 K min⁻¹) and quickly cooled down. This treatment resulted in growth of alloy particles and change of their composition. Administering later a short pulse of H₂ at RT was devised to consume the oxygen chemisorbed on the surface of the Pd–Au clusters leaving the PdO “bulk” phase intact. The length of the pulse was adjusted experimentally. In this way, the 111 alloy peak profile could be registered with and without the chemisorbed oxygen, and the data analysis was not obscured by processes affecting the Pd metal phase.

Besides the peak position also the change in peak amplitude and peak width were interpreted structurally using results of atomistic simulations. The peak amplitude was corrected for absorption of X-rays by gas in the camera as described previously.¹²

3. Theoretical calculations

The modeling method exploited in this work is based on semi-empirical N-body Sutton–Chen potentials.¹⁹ For simulations of bimetallic clusters the simple form of Sutton–Chen potential has been extended¹⁴ to the following (eqn. (1)): where r_ij is the distance between atoms i and j. The parameters ε_ij, a_ij, c_ij, n_ij, m_ij are the same for all pairs of atoms of the same kind (Pd or Au); ε_ij is an energy scaling parameter and for a monometallic fcc lattice can be easily determined from cohesive energy of the metal. The a_ij parameter is a distance parameter scaling distances and for a monometallic lattice is assumed to be the equilibrium lattice constant. The n_ij, m_ij parameters are two integers (n_ij > m_ij) which can be chosen for pure elements in order that the product nm is the nearest integer to 18ΩB/E (Ω is the atomic volume in a given structure, B is the bulk modulus and E is the experimental cohesive energy). The c_ij parameter for pure elements can be easily determined from the equilibrium condition. For a monometallic lattice there is not much freedom left in the choice of parameters. One can only pick integer values of n and m that best reproduce elastic constants of the metal.

Surprisingly, the resulting potential works excellently for monoatomic bulk metals and clusters of size D > 2 nm. This corresponds to a number of atoms greater than 300 and the higher is the number of atoms the better is the performance of the potential as the situation approaches an infinite model for which the potentials were parametrized. This performance was extensively tested for palladium¹⁴ and the potential reproduced the available experimental²⁰ values of surface contraction for a few crystallographic faces. Poor reproducibility of surface tension value and, for some metals, of anharmonicity and thermal expansion coefficient are less important for the results presented in this work. Due to the simplicity of eqn. (1), the computational cost of simulations is not much higher than that for Lennard-Jones potentials, and the N-body potential is a natural choice for large scale metal cluster modeling.

For bimetallic clusters, assuming the parameters for pairs of atoms of the same kind to be the same as that for a monometallic lattice, the parameters to be determined are mixing parameters ε_ij, a_ij, c_ij, n_ij, m_ij where i and j are index atoms of different atoms. These parameters can be deduced from the condition of stability of non-ordered alloy with uniform Pd and Au distribution and from the experimental values of the enthalpy of mixing. Fortunately the latter values are available for the whole range of alloy compositions²¹ and the potential parameters can be fitted to them all. Fig. 1 shows the quality of fit and Table 1 lists all the potential parameters used in the present study.

Fig. 1 Enthalpy of mixing for a range of Pd–Au alloy compositions. Literature data after Hultgren et al.²¹ (○), data calculated from Sutton–Chen potential for the parameters used in present work and obtained from the best fit (□).

Table 1 Sutton–Chen potential parameters used in simulations. The values for interaction between atoms of the same type are after Sutton and Chen.¹⁹

Atom pair	ε	a	c	n	m
Au–Au	0.012793	4.08	34.408	10	8
Pd–Pd	0.004179	3.89	108.27	12	7
Pd–Au	0.006696	3.984	81.6518	11	6

---

One of the methods used in simulations was configurational minimization of the model structure with a modified Monte Carlo Metropolis algorithm (MC). In this method the final configuration is determined by a long series of exchanges of randomly picked atoms of different types. The exchange is accepted only when it lowers the energy, but with the probability of exp(ΔE/kT), where ΔE is the change of energy (negative), k is the Boltzmann constant and T absolute temperature. The modification applied in this work consisted in preliminary relaxation of all atom positions (down to a potential gradient of 0.1 eV Å⁻¹) before making the decision of accepting the exchange. If the exchange was rejected, the original structure was restored. This made the choice of a more likely configuration more reliable. The MC calculation lasted sufficiently long to have a number of successful exchanges per 500 exchanges as well as the energy standard deviation stabilizing with time. This usually required few ten thousands of exchanges. This criterion of approaching the minimum of free energy worked well. Tests repeating the calculations with different starting configurations (distribution of palladium and gold) resulted always in practically the same concentration profile and the diffraction pattern peak position was reproduced within 0.005°. The structure relaxation was performed following the Fletcher–Reeves conjugated gradient algorithm. Finally the resulting model was subjected to a molecular dynamics (MD) run at the temperature of the experiments (i.e. RT) using a “leap frog” algorithm. The diffraction pattern for the final structure model was computed by averaging the patterns over the MC run covering a number of thermal oscillations at equilibrium. The computational methods have been described in more detail elsewhere.¹³ The outlined above scheme of calculations was applied to Pd–Au model clusters in the form of closed shell cubooctahedra with the number of shells ranging from 6 to 10. These corresponded to the total number of atoms equalling the magic numbers 923, 1415, 2057, 2869 and 3871, respectively. In each case the cluster composition was 50% Pd, 50% Au. One simulation was done for a 3871 atom cubooctahedron consisting of 75% Pd, 25% Au. The purpose of these simulations was to estimate the diffraction peak shift resulting from the changes of the concentration profile caused by oxygen chemisorption. As oxygen adatoms should favor Pd neighbours the maximum change from the free arrangement should occur when the surface is made up solely of palladium atoms.

This is why the second scheme of calculations was run for the same models as above but with their surface populated only by Pd. This scheme differed from the first one in that the MD run was performed allowing exchanges between all the unlike atoms except those at the surface. However, for the energy comparison before accepting the exchange, all the positions of atoms, including the surface atoms, were relaxed. This procedure was supposed to take account of the boundary conditions imposed by surface chemistry. The application of this surface constraint has a global effect on the morphology of the cluster. Fig. 2 shows the histogram of radial distances between the atoms of each type and the atom at the center for a 10-shell cubooctahedron. The segregating gold in the non-constrained cluster dominates at the surface and about 1 nm below the surface of the cluster whereas palladium seems to occupy intermediate regions compensating energy by increasing number of Pd–Au bonds that are favorable. On imposing the chemical constraint and fixing palladium atoms at the surface the whole situation reverses. The concentration profile turns out to be just opposite to the unconstrained one with gold occupying mostly the cluster center and the region below the surface. The local constraint thus has a global effect. Similar results were obtained for the other closed shell models.

Fig. 2 Histograms of distances around the cluster central atom. Number of Pd atoms around the central atom (black bars), number of Au atoms around the central atom (white bars). The upper histogram corresponds to the free cluster and the lower histogram to the constrained cluster. The histograms are averaged over an MD run after MC configurational constrained minimization.

The comparison between the results of the first and the second scheme of calculations is presented in Fig. 3. The lower part of the figure illustrates the variation of the peak position with cluster size for both free and constrained models for a cluster composition Pd/Au equal to 1∶1. The peak position for the constrained models does not strongly depend on the cluster size whereas the values for the free models quickly rise with decreasing cluster size. The peak shift (upper part of the figure) is much larger and more significant for large clusters than the analogous peak shift due to surface relaxation.¹³ This is why the latter effect has not been taken into account although it may play a minor role since the constrained models should have the effects of surface relaxation supressed by chemisorbing oxygen¹³ thus differing from the free models. The peak shift vs. size of cluster should monotonically approach zero as for large crystals the ratio of the number of surface to bulk atoms approaches zero and the state of the surface becomes increasingly unimportant for the global diffraction intensity. The rate of the decrease of the peak shift with cluster size suggests that it is still measurable even for clusters as large as 10 nm. The variation of the peak position for the constrained models is rather small unlike the variation for monometallic clusters caused solely by the surface relaxation. It seems that constraining palladium atoms to the surface introduces enough strain to the model that the progressive surface relaxation effect with decreasing cluster size does not occur. The results of calculation of the 111 alloy peak position for a 6-closed shell cubooctahedron (Au/Pd = 1∶1) with its surface constrained to consist of equal numbers of Au and Pd atoms distributed randomly, and a 10-closed shell cubooctahedron (Au/Pd = 1∶3) non-constrained and with its surface constrained to be populated by only Pd atoms, are also included in Fig. 3. To illustrate consequences of the considered structural constraints, Fig. 4 shows diffraction patterns calculated for both models. The decrease in peak amplitude while moving from the constrained to the free model (18%) is caused mostly by rearrangement of Au atoms whereas the Pd–Au partial scattering function remains comparable in intensity. Also an apparent particle size, as calculated from the Scherrer formula for this 5 nm cluster, changes from 4.6 nm for the constrained to 4.9 nm for the unconstrained cluster. The total 111 peak intensity falls by 25% on this transition.

Fig. 3 Pd–Au alloy 111 peak position for a range of sizes of cubooctahedral clusters. The data are for 6–10 closed shells (923, 1415, 2057, 2869 and 3871 atoms). Lower diagram: 111 peak position (2θ(Cu-Kα)) for constrained models of 1∶1 composition (●), free models of 1∶1 composition (○). The dotted horizontal line marks the 111 peak position for 1∶1 Pd–Au bulk alloy assuming Vegard’s law. Additional data: 6-shell cubooctahedral Pd–Au (1∶1) cluster with the surface constrained to be randomly populated by equal number of Pd and Au atoms (□), 10-shell cubooctahedral Pd–Au (3∶1) cluster: constrained with Pd atoms at the surface and free model, lower and upper filled square, respectively. The upper diagram displays the 111 peak shift, the difference between the positions of the open and full circles from the lower diagram.

Fig. 4 Comparison of diffraction patterns for two models of 10-shell cubooctahedral Pd–Au (1∶1) alloy cluster. Cluster constrained with Pd at the surface (○), free cluster-open (□). The lower curves are of partial scattering functions being contributions to the full pattern calculated for distances between unlike pairs only (Pd–Au): constrained cluster ([dash dash, graph caption]), free cluster (⋯). The contribution from Pd–Pd pairs is negligible is not shown.

4. Results

4.1 Sample 1

The repeated procedure described in the Experimental section (steps 1–5) resulted in two data sets, each covering a few days measurement run. The fragment the current analysis is focused on is illustrated in Fig. 5 showing evolution of the diffraction pattern for sample 1 during formation and reduction of the tetragonal phase of PdO (peaks at 2θ 33.9, 42, 54.8°) from fcc palladium (peaks at 2θ 40.1, 46.7°). The alloy Pd–Au fcc phase (peaks at 2θ 38.9, 45.2°) initially incorporates some unalloyed palladium during heating to 673 K (stage 3 of the procedure) and thus the peak is finally slightly shifted due to changing alloy composition. This shift was observed during the first transition of Pd to PdO. Further cycles of oxidation at 673 K and reduction in H₂ had much less impact on the alloy peak position. The peak shape after first heating is slightly sharpened and fits well to the gaussian profile.

Fig. 5 Evolution of the Cu-Kα diffraction pattern for sample 1. From back to front: sample with fcc alloy phase and fcc palladium phase, heating to 673 K in dry air and formation of tetragonal PdO phase (A), exposure to argon at RT (B) and injection of hydrogen pulse (C), reduction in hydrogen at RT (Pd transforms into β-hydride phase) (D) and exposure to argon with decomposition of the β-hydride before the next cycle of heating (A).

Evolution of 111 alloy peak parameters for this sample is shown in Fig. 6. The peak position follows a rather complex route that may be elucidated expanding the sample history into a detailed list:

Fig. 6 Evolution of 111 alloy peak position, width and amplitude for sample 1 during the experiment. The letters have the same meaning as in Fig. 5.

Stage A: reduction in hydrogen causing formation of PdH which, together with nearly constant peak position for Pd–Au alloy, shifts the overlaping peaks down to 38.92°,

evacuation and flushing with argon (the first sharp jump in the curve due to the PdH–Pd transition),

gradual change on temperature ramp in dry air, from 39.12° at RT down to 38.77° at 673 K,

quick cooling (the second jump in the curve) with the peak position jumping to 38.9° at RT.

Stage B: the sample evacuated and flushed for an extended time with Ar.

Stage C: injection of a short pulse of H₂ to the stream of argon followed by a long time flow of argon.

Stage D: reduction in H₂, evacuation and flushing with argon.

The last stage is followed by exposure to a flow of dry air for the next cycle of heating to 673 K (A). Analysis of the peak position vs. temperature in the second region A gives the linear expansion coefficient of the studied alloy close to that for bulk palladium (≃12 × 10⁻⁶ K⁻¹).

Alloy average composition can be roughly estimated from the peak position. The assumption of linear dependence of lattice parameter from the alloy composition (Vegard’s law) would give an estimate of the Pd content as 37 at.%. The data in Fig. 3 calculated for an alloy composition of 1∶1 shows that the surface segregation effect causes the peak position to be shifted for bigger clusters down from the value predicted by Vegard’s law (dotted line). This justifies the conclusion that the determined Pd contents are an underestimation and the real value is closer to 50 at.%.

The shift of the peak position between regions B and C is well visible and much larger than the estimated standard deviation of the peak position within both regions. The shift was estimated as 0.024° (2θ) and the standard deviations were 0.012° (B: calculated over 165 points) and 0.008° (C: over 20 points). This shift should be compared to the transition from the lower to the upper curve in Fig. 3, extrapolated to the actual size of crystallites (estimated from the Scherrer formula as 13 nm). The observed peak shift is accompanied by an 8% increase in the peak intensity. This suggests that the gold atoms which have larger scattering factor than Pd atoms (and their distribution affects intensity to greater extent), are now better ordered.

The proposed model of transition from oxygen induced complete surface segregation of palladium to segregation of gold in an inert environment implies a fall of the total peak intensity by 25% (for 5 nm clusters). As the observed experimentally the observed peak shift is evidently due to a change in concentration profile, both the smaller than expected shift and contrary to expected change in intensity, strongly suggest that the chemisorption of oxygen causes only partial segregation of palladium and that the formed structure is havily strained leading to smaller than expected peak intensity. The latter allows for the rise in the peak intensity on reduction of chemisorbed oxygen.

4.2 Sample 2

A similar measurement strategy has been applied to sample 2 and corresponding diffraction patterns before and after transition of Pd to PdO are presented in Fig. 7. The measured evolution of the 111 alloy peak is displayed in Fig. 8. As before, the sample after oxidation of the residual Pd was cooled down to RT and kept in a flow of argon (A). Then a short pulse of hydrogen was injected to the stream of argon (beginning of the region B). The statistics of the measurement are slightly worse than for sample 1 but clearly the peak position shifts up by ∼0.046°. The peak width grows by 6% and the amplitude remains unchanged which results in an overall increase of the peak intensity by 6%. This is similar to the intensity rise observed for sample 1 and fits consistently into the proposed sample scenario of surface disorder evolution.

Fig. 7 Cu-Kα Diffraction pattern of sample 2 before (dashed line) and after transition to PdO (dotted line).

Fig. 8 Evolution of the 111 alloy peak position, width and amplitude for sample 2 during the experiment. Exposure to argon at RT after heating in dry air to 673 K (A), then in argon after injection of an H₂ pulse (B).

Application of Vegard's law gives an estimation of the alloy atomic composition as 70%Pd, 30%Au. Considering non-linearity of the effective lattice parameter vs. cluster composition that may be deduced from Fig. 3 (diamonds), the real composition has to be corrected towards higher Pd contents by about 10% i.e. up to 80% of palladium. This estimation was done on the assumption that the constrained cluster alloy 111 peak position does not depend on the size of the cluster, an assumption partially justified by Fig. 3.

5. Discussion

The experimental results presented above show a well measurable shift in the nanoparticle alloy 111 peak position. It occurs on injection of small portion of hydrogen to the inert gas stream flowing in the chamber after chemisorbing oxygen at the surface of the metal. The shifted peak position is further stable.

To rationalize this observation one has to assume that concentration profile within the Pd–Au alloy particle is quite different for the particle free of adsorbed species and that with oxygen chemisorbed at its surface. These different concentration profiles result in quite different diffraction peak positions reflecting spatial rearrangement of the particle's atoms on removal of the adsobate. The short pulse of hydrogen is supposed to consume the surface oxygen from Pd–Au particles, leaving the PdO bulk phase practically intact as indeed what was observed during the experiment.

The atomistic models considered here show the direction and magnitude of the peak shift between free clusters and that with the outmost alloy layer consisting solely of palladium atoms. These calculations were affordable for clusters of size up to 5 nm. An extrapolation of the upper diagram of Fig. 3 shows that the values obtained experimentally (see Figs. 6 and 8) shift in the same direction as those obtained from the modeling but the model overestimates the magnitude of the shift.

The atomistic model enables calculation of the peak diffraction intensities. The model of a complete adsorption induced segregation of Pd results in the 111 peak intensity being 25% larger than that due to the free model of Au segregation. This fact, together with the experimental observation of the lower peak intensity, also suggests what a realistic model should look like. For example, it is seen from Fig. 3 that modification of the chemical constraint by populating the surface randomly with equal numbers of Pd and Au atoms leads to the peak position (open square) that coresponds to a lower peak shift. The calculated peak amplitude for this model is already comparable to that for the nonconstrained one. This model shows a likely way to more realistic models. The real Pd–Au clusters when exposed to oxygen chemisorption evidently suffer severe surface disorder which decreases the scattered intensity. Such disorder is very likely in the state of incomplete coverage of oxygen. It could not be realistically modelled due to difficulty in including Pd–O interactions into the Sutton–Chen scheme of potentials. This is why the conclusions may have only qualitative character. On the other hand the experiments described here provide data indicative of the real structure of the material in in vivo conditions and such data are sensitive even to subtle surface processes. This creates an unique opportunity to observe and control such subtle processes. The modelling clearly is a tool to understand such data.

The calculated peak shift is not very sensitive to the average nanoparticle composition – the calculated shift for the 3∶1 ratio (Pd/Au) (open diamonds in the diagram) is only slightly smaller than that for 1∶1 composition. This suggests that the real distribution of cluster compositions should not significantly affect the general conclusions. Also the real cluster size distribution is usually an important factor in comparison of experimental and model data. The regular gaussian shape of reflections measured after the initial temperature treatment suggests unimodal size distribution with only a small amount of very small crystallites. This, together with the applied method of peak parameter analysis described earlier means that the parameters obtained relate to the mean crystallite size and the spread of sizes, if not pronounced, does not obscure the results.

It is well known that oxygen under considerable pressure can dissolve into the surface layer of the metal. This effect may have measurable diffraction consequences and in fact some our observations suggest that it has. This is why in the presented experiments the samples after exposure to oxygen (dry air) are kept in the stream of high purity argon. For nanocrystalline Pd–Au in oxygen-free atmosphere, this under surface oxygen should quickly evolve and therefore was not considered.

The observation of a quickly changing segregation profile with chemisorption of oxygen proves that interdiffusion of metal atoms in the alloy cluster proceeds at room temperature relatively quickly. This is why defects and faults in the structure of the clusters would be immediately relaxed and are unlikely.

The studied structural evolution of the samples can show great complexity due to the fact that chemisorption of either, oxygen or hydrogen can cause segregation of palladium, both can be relatively strongly bonded to palladium and both can form subsurface states. These is reflected in the fact that admission of the oxygen pulse after reducing the surface with hydrogen pulse shows no strict repeatability although some qualitative changes in diffraction pattern reappear. This can be also due to the experimental difficulty to precisely adjust the volume of the pulse. The right dose may depend also on the history of the sample. The data presented in the previous section show, however, the maximum peak shift effect that was possible to measure in four experiments each with a fresh sample of the same material for each material (1 and 2).

As the 111 peak position of the constrained models in Fig. 3 depends weakly on size of the cluster, the significant (two-fold) rise in the peak shift observed experimentally with decreasing average cluster size strongly suggests that in argon, after the hydrogen pulse, the surface segregation of gold occurs. On the other hand the magnitude of the experimentally observed peak shift and increase of the peak intensity suggests that clusters with chemisorbed oxygen are not fully covered at their surface with palladium.

6. Conclusions

The presented experimental data show that in the studied Pd–Au alloy clusters there is a measurable experimental change of XRD profile parameters (shift in position, integral intensity) caused by changing atmosphere (reduction, oxidation). The observed phenomena can be rationalized when atomistic modeling is applied. This leads to the conclusion that the surface segregation of gold occurs in argon atmosphere after reduction of pre-chemisorbed oxygen. For the alloy clusters with chemisorbed oxygen the experimental data are not consistent with the model of complete segregation of palladium i.e. the model where the cluster surface consists exclusively of palladium, and suggest surface palladium intermixing with gold.

The observed effect of environment-induced changes of XRD profile parameters caused by changing concentration profile in the alloy particles has, for the first time, been measured in situvia powder diffraction and was measurable for alloy particles as large as 12 nm. It is shown that the in situ XRD method in a well defined, controlled environment can provide data sufficiently accurate to enable verification of a complex molecular scenario. The clearly observed changing diffraction pattern elements cannot be explained by conventional powder diffraction based on the Bragg law and can be interpreted only via atomistic simulations. The proposed technique opens a pathway to direct structural observation of the nanopowder–gas interface during catalytic reactions. The increase in accuracy may provide direct insight into formation of transition states and reaction dynamics.

Acknowledgements

The author thanks Prof. J. Pielaszek for helpful discussions and a critical reading of the manuscript. The work has been financially supported by the Polish State Committee for Scientific Research (KBN) under research grant No 3T09A 024 16.

References

1. G. Maire, L. Hilaire, P. Legare, F. G. Gault and A. J. O'Cinneide, J. Catal., 1976, 44, 293.
2. A. J. O'Cinneide and F. G. Gault, J. Catal., 1975, 37, 311.
3. B. J. Wood and H. Wise, Surf. Sci., 1975, 52, 151.
4. R. Anton, H. Eggers and J. Veletas, Thin Solid Films, 1993, 226, 39.
5. A. Jabłoński, S. H. Overbury and G. A. Somorjai, Surf. Sci., 1977, 65, 578.
6. R. J. Davis and M. Boudart, J. Phys. Chem., 1994, 98, 5471.
7. S. N. Reifsnyder and H. H. Lamb, J. Phys. Chem. B, 1999, 103, 321.
8. F. L. Williams and D. Nason, Surf. Sci., 1974, 45, 377.
9. A. F. Lee, C. J. Baddeley, C. Hardacre, R. M. Ormerod and R. M. Lambert, J. Phys. Chem., 1995, 99, 6096.
10. D. L. Weissman-Wenocur and W. E. Spicer, Surf. Sci., 1983, 133, 499.
11. J. Pielaszek, in Proceedings of the XVI Conference on Applied Crystallography, World Scientific, Singapore, 1995.
12. Z. Kaszkur, J. Appl. Crystallogr., 2000, 33, 1262.
13. Z. Kaszkur, J. Appl. Crystallogr.., 2000, 33, 87.
14. Z. Kaszkur and B. Mierzwa, Philos. Mag. A, 1998, 77, 781.
15. W. Vogel, J. Bradley, O. Vollmer and I. Abraham, J. Phys. Chem. B, 1998, 102, 10 853.
16. V. Gnutzmann and W. Vogel, J. Phys. Chem., 1990, 94, 4991.
17. W. Juszczyk, Z. Karpiński, D. Łomot, J. Pielaszek and J. W. Sobczak, J. Catal., 1995, 151, 67.
18. A. F. Rawle, PhD Thesis, Brunel University, 1979.
19. A. P. Sutton and J. Chen, Philos. Mag. Lett., 1990, 61, 139.
20. C. J. Barnes, M. Q. Ding, M. Lindroos, R. D. Diehl and D. A. King, Surf. Sci., 1985, 162, 59.
21. R. Hultgren, P. A. Desai, D. T. Hawkins, M. Gleiser and K. K. Kelley, Selected Values of the Thermodynamic Properties of Binary Alloys, Metals Park, Ohio, 1973.

---