JOURNAL OF CHEMICAL PHYSICS VOLUME 118, NUMBER 7 15 FEBRUARY 2003 Density functional calculations of hydrogen adsorption on palladium–silver alloy surfaces O. M. Løvvika) Department of Physics, University of Oslo, P.O. Box 1048 Blindern, N-0316 Oslo, Norway R. A. Olsen Theoretische Chemie, Vrije Universiteit, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands 共Received 26 August 2002; accepted 19 November 2002兲 Palladium–silver alloy surfaces with and without adsorbed hydrogen have been studied through density functional theory within the generalized gradient approximations employing a slab representation of the surface. Our calculated lattice constants are in good agreement with experimental data, but we find a substantially lower surface energy for Ag共111兲 and Pd共111兲 than experiments. We have calculated adsorption energies of hydrogen on several sites on various alloy surfaces, and found that threefold hollow sites with as many palladium neighbors as possible are preferred. The difference in adsorption energy is so large that we expect trapping of hydrogen around palladium atoms in the surface, possibly resulting in a lower diffusion constant of hydrogen at low coverage on alloy surfaces than on the pure Pd and Ag surfaces. Assuming that the adsorption energy has contributions from geometric 共‘‘ensemble’’兲 and electronic 共‘‘ligand’’兲 effects, we found the geometric contribution to dominate. For the geometric contribution it is seen that the binding strength increases as the d-band center moves toward the Fermi level, a result also found by a number of other theoretical studies. However, for the electronic contribution we found that the variation of the adsorption energy as a function of the d-band center was opposite that reported by others: We saw that hydrogen binds less strongly to the surface as the d-band center moves toward the Fermi level. This could possibly be explained by a large variation of the interaction between the metal sp band and hydrogen. © 2003 American Institute of Physics. 关DOI: 10.1063/1.1536955兴 I. INTRODUCTION the surface. We have, however, previously shown that it is possible to describe quite accurately many of the bulk properties of palladium–silver alloys with and without hydrogen by using periodicity and small unit cells.7 We expect the same to be the case at the surface, since varying the size and composition of the unit cell gives a number of different adsorption sites and electronic surroundings. This should put us in a position to give reasonable estimates of what kind of sites hydrogen prefers, and how surface diffusion most probably takes place on the alloy surface. This means that a periodic study like the present one should be well justified, at least as long as the excess of silver atoms at the surface is taken into account. There has lately been published a number of theoretical studies on alloy surfaces based on density functional theory 共DFT兲 共see, for instance, Refs. 8 –20兲. No studies have so far to our knowledge presented results on palladium–silver alloy surfaces, but Liu and Nørskov have studied the adsorption of carbon monoxide, oxygen, and nitrogen on various gold– palladium alloy surfaces.17 They were able to separate the adsorption properties into geometric and electronic effects by varying the gold content of the surface, and found that both effects contributed to stronger binding of the adsorbate as the center of the d band moved toward the Fermi level. We shall see that we find the same trend for the geometric contribution, while the electronic contribution is the opposite for hydrogen adsorption on palladium–silver alloy surfaces. The interest in hydrogen at alloy surfaces is large and increasing, both because of the interesting properties of such systems from a fundamental point of view,1 and also because of important applications within fields such as heterogeneous catalysis, hydrogen storage in metal hydrides, and hydrogen diffusion membranes.2 Palladium–silver alloys are of particular interest for membranes—a typical industrial metal based hydrogen selective membrane consists of Pd77Ag23 . 3,4 Such membranes are thus relatively well studied, but the understanding of the microscopic properties is still quite crude. One intention of the present study is to increase our detailed knowledge of hydrogen interacting with the surface of palladium–silver alloys, a system providing both a rich experimental basis and a challenge to thorough theoretical studies. There has not been measured any ordered phases in PdAg alloys, but a recent theoretical study has predicted three different ordered structures.5 Experiments on the surface of such alloys show that it is dominated by silver. 共The 共111兲 surface of Pd67Ag33 contains, for instance, 5.2% palladium.6兲 To be truly realistic, a model describing such alloys should thus include both the possibility of random distribution of atoms in the bulk and excess of silver atoms at a兲 Electronic mail: o.m.lovvik@fys.uio.no 0021-9606/2003/118(7)/3268/9/$20.00 3268 © 2003 American Institute of Physics Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 Hydrogen adsorption on palladium–silver alloys 3269 TABLE I. The bases that were used to calculate the adsorption energies 共1兲, the cohesive properties 共2兲, and a basis that was used as a reference 共3兲. The numbers refer to the exponent of a STO, while NAO designates a numerical atomic orbital. All bases have orbitals up to 3d frozen. Basis 4s 4p 4d 5s 5p 4f 1 2 3 NAO NAO, 3.9 NAO, 2.1, 6.2 NAO NAO, 2.7 NAO, 1.35, 3.9 NAO, 1.5 NAO, 1.5 NAO, 1.45, 4.9 NAO, 1.8 NAO, 1.8 NAO, 0.9, 2.6 1.8 1.0, 2.0 1.5 1.5 II. METHOD Our calculations are performed using ADF-BAND,21,22 employing the generalized gradient approximation 共GGA兲 due to Becke23 and Perdew.24 The one-electron basis sets representing the electron density consist of both Herman– Skillman numerical atomic orbitals 共NAOs兲 and Slater-type orbitals 共STOs兲, with a frozen core. Scalar relativistic corrections have been included through the zeroth-order regular approximation.25 Our slab calculations use two-dimensional translational symmetry. Table I shows the basis sets that were used in our calculations, together with a reference basis set for the convergence tests. The convergence of our calculations has been checked by varying all important parameters, summarized in Table II. Relative energies, like adsorption energies, are less sensitive to basis set effects than cohesive properties, for instance the cohesive energy 共as seen from Table II兲. We have therefore used different basis sets to calculate adsorption energies and cohesive properties, respectively. The overall convergence is well within 0.1 eV in all our results, with the GGA and the unit cell size 共including the number of layers兲 as the only remaining approximations. We use five different models to investigate a representative set of surface adsorption sites at the fcc 共111兲 alloy surface: pure silver and palladium slabs, a palladium slab coated with a monolayer of silver 共Ag/Pd兲, a palladium slab coated with a monolayer of 75% silver and 25% palladium (Ag3 Pd/Pd), and a slab containing 75% palladium and 25% silver (AgPd3 ). The alloy slabs are shown in Fig. 1. In the calculation of adsorption energies we have used three fcc stacked atom layers, and a 2⫻2 surface unit cell for all the models. Several previous studies have shown that three layers are enough to describe adsorption reasonably well 共see for instance Refs. 14, 26, 27兲, and we have also checked that this is the case for the slabs with the surface layer being different from the lower layers. A 2⫻2 surface unit cell gives a hydrogen surface coverage of 25%, which should be a reasonable approximation for the low-coverage region. Surface energies were calculated using a 1⫻1 surface unit cell for the pure metal slabs and for Ag/Pd. The relaxed slabs were found by two different methods: either by varying the lattice constant and the top layer relaxation independently, or simultaneously; by creating a twodimensional potential energy surface from the two coordinates. In the former case, the minimum was found using two successive harmonic fits, while in the latter case it was found from a two-dimensional spline fit. We performed this test for both Ag, Pd, and Ag/Pd, using three metal layers. The largest difference between the two methods appeared for ⌬d on Ag, where the independent and simultaneous variation gave 1.4% and 1.0% outwards relaxation of the upper metal layer, respectively. ⌬d on Ag/Pd also varied slightly, the two methods giving 1.2% and 1.6% relaxation, respectively. The other results gave differences well within the numerical errors. We found this satisfactory, and used the independent variation method to find a latt and ⌬d for the last two models. The results in Table III show that our calculated lattice constants are consistently 1%–2% above the experimental ones, a familiar result for the GGA. All the results in Table III are for three layer slabs. We have also increased the number of layers to 15, and found that the relaxed lattice constants of the metal slabs converge when the number of layers reaches about eight. The converged values are about 0.02 Å above those for three layers, but at the same time consistently about 0.01 Å lower than the bulk relaxed lattice constant.7 The calculated top layer relaxation compares favorably with the experimental value for Pd, but the Ag result is far off; we predict an outward relaxation, while experiment shows inward relaxation.28 Other DFT studies have also shown inwards relaxation,29,30 and we conclude that three layers are not sufficient to give reliable results for the surface relaxation. B. Surface energy A common way to calculate the surface energy is to subtract the bulk cohesive energy E bulk from the slab cohesive energy E slab : TABLE II. The most important error sources of the cohesive energy E coh and the adsorption energy E ads in meV. III. RESULTS A. Cohesive properties We have calculated the optimized lattice constant a latt and the top layer relaxation ⌬d for our models, and the results are shown in Table III. All optimizations have been done manually, since no forces are calculated by ADF-BAND. Error 共meV兲 Test E coh E ads Ag Pd H/Ag H/Pd KSPACE ACCINT 7→9 15 9 20 10 4.5→5.0 1 1 8 5 Basis 1→2 345 470 66 29 2→3 52 43 31 62 Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp 3270 J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 O. M. Løvvik and R. A. Olsen ⫽ 12 共 E slab⫺N layersE bulk兲 , 共1兲 where N layers is the number of layers of the slab. This leads, however, to a linear divergence in the surface energy as the number of layers increases.31,32 In order to avoid this, the bulk energy in the above-mentioned formula should be replaced by the increase in cohesive energy when increasing the slab thickness by one layer.31 We have used a linear fit to this energy as the slab thickness is varied: E bulk⫽⌬Ē slab . FIG. 1. Three of the models used in our calculations. We used three layers on each slab when calculating the adsorption energy. The surface unit cell is outlined by a solid line. The adsorption energy of hydrogen along the straight line is plotted in Fig. 2, while the adsorption energy along the suggested diffusion paths is plotted in Fig. 8. 共2兲 We have done this for silver slabs up to 15 layers, and found that the surface energy is converged to about 0.03 eV already after four layers. This was regardless of relaxing the lattice constants or not. We used at least five layers to calculate the other surface energies. The surface energy of the silver-coated palladium slab was calculated by comparing the cohesive energies of slabs with a monolayer of silver on each side of the slab, and an increasing number of palladium layers between the silver layers 共up to five palladium layers between兲. From this cohesive energy we subtracted 2⌬Ē slab taken from the pure silver slab calculations, in addition to the value found for the palladium atoms between the silver layers. The results are shown in Table III. The ‘‘experimental’’ surface energies are 558 and 840 meV for Ag and Pd, respectively.33,34 They are derived from the surface tension of liquid metals, and have unknown uncertainties. Other DFT studies have also calculated the surface energy of Ag共111兲 and Pd共111兲, using a full-potential linear-muffin-tin-orbital 共LMTO兲 method,29 a full charge density LMTO method,35 and a total energy pseudopotential method.30 The former two studies both obtained significantly higher values than the present study; respectively, 0.55 and 0.553 eV/atom for Ag, and 0.68 and 0.824 eV/atom for Pd. The latter study reached almost the same result as ours; 0.357 eV/atom for Ag. All the mentioned studies compared slab calculations to the bulk cohesive energy, which is known to yield linear divergence. It is, however, unclear whether the number of layers used in their calculations and possible problems with linear divergences are sufficient to explain the large differences. To check whether our results could be an artifact due to numerical problems, we performed extensive extra tests. We TABLE III. The lattice constant a latt , the outwards relaxation of the upper metal layer ⌬d 共in % of the lattice constant兲, and the surface energy for our five models. All our results are calculated using the BP GGA, except the WIEN2K calculation, where the PBE GGA has been used 共see the text兲. Slab a latt 共Å兲 ⌬d 共%兲 共meV/atom兲 This study Expta This study Exptb This study WIEN2K Exptc Ag Ag/Pd Ag3 Pd/Pd AgPd3 Pd 4.12 4.08 1.0 ⫺2.5⫾0.5 325 373 558 4.00 3.99 3.98 1.6 1.5 1.0 3.93 3.88 1.1 2.0⫾2.0 560 625 840 303 516 a Reference 49. Ag: Ref. 28, Pd: Ref. 50. c Reference 33. b Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 Hydrogen adsorption on palladium–silver alloys 3271 TABLE IV. Adsorption energy in electron volts at the high coordination adsorption sites on the various surfaces. Slab Ag Ag/Pd Top Bridge fcc hcp 0.64 0.27 0.20 0.23 0.69 0.33 0.25 0.28 Ag3 Pd/Pd ⫺0.02 ⫺0.06 ⫺0.05 0.00 0.79 0.40 0.36 0.39 AgPd3 0.04 ⫺0.33 ⫺0.49 ⫺0.38 E ads⫽E coh共 PdH兲 ⫺E coh共 Pd兲 ⫺ 21 E H2 ⫺E H , FIG. 2. The adsorption energy 共as defined in the text兲 along the straight lines depicted in Fig. 1. The lines between the calculated points are only a guide to the eye. calculated the surface energy using the largest basis set in Table I, which should give results very close to the basis set limit, giving no significant difference. The same results were also obtained when including spin–orbit interactions. We checked three different GGAs 关Becke23 and Perdew,24 Perdew and Wang,36 and Perdew, Burke, and Ernzerhof 共PBE兲兴,37 which all gave the same results within 0.03 eV. To check whether it could be a result of the atomic orbital basises used in the ADF-BAND program, we also calculated the same surface energies using the WIEN2K program, which is a full-potential linear augmented plane wave code.38 This does not rely on pseudo-potentials, and should be able to give results quite close to the GGA limit, just like ADFBAND. The k-space integration parameter was set to give results converged to within 1 meV, and we employed the PBE GGA. The resulting surface energies are shown in Table III, as we see in good correspondence with our results using ADF-BAND. Further investigations are clearly needed to resolve the discrepancy between our results and those presented by Vitos et al.,35 but that is beyond the scope of this paper. The surface energy of silver is in either case much lower than that of palladium, while we have calculated that the surface energy of Ag/Pd is slightly lower than that of silver. We have also calculated the surface energy of a palladium coated silver slab 共not shown in Table III兲, and found that ⫽566 meV for this slab; slightly higher than that of a pure palladium slab. We also note that the surface energy of the AgPd3 alloy is considerably higher, almost just as high as that of Pd. All these findings are consistent with the excess of silver atoms that is found on the surface of palladium–silver alloys; the silver atom is more stable at the surface than a palladium atom on these surfaces. C. Site preference The four high-symmetry sites on the fcc共111兲 surface lie on the straight lines depicted in Fig. 1. We have calculated the adsorption energy of a single hydrogen atom at these sites, and the results are shown in Fig. 2 and Table IV. The adsorption energy is defined as 0.84 ⫺0.27 ⫺0.22 ⫺0.22 Pd ⫺0.07 ⫺0.37 ⫺0.49 ⫺0.45 共3兲 where E coh is the cohesive energy of the slabs, while E H2 ⫽⫺4.80 eV and E H⫽⫺0.93 eV are the calculated GGA energy of the hydrogen molecule and the energy difference between a spin-restricted and spin-unrestricted hydrogen atom, respectively. This means that a negative adsorption energy is stable compared to free hydrogen molecules. The coordinates of the hydrogen atoms at the various sites are taken as the equilibrium coordinates of hydrogen on the pure palladium and silver slabs 共they are practically the same兲. We have not tried to find local minima outside the high coordination site. We first note the similarities between hydrogen adsorption on Pd and Ag, the main difference being that adsorption is about 0.7 eV more stable on the Pd surface. The threefold hollow sites are most stable on both surfaces, with the fcc site being slightly more stable than the hcp site. We further see that sites with the same nearest surroundings may have significantly different adsorption energies; as an example, top sites above a silver atom vary from 0.64 eV 共on Ag兲 to 0.84 eV 共on AgPd3 ). Since the type of atom to which H bonds is unchanged for these surfaces, this effect is purely electronic, and we are thus able to extract the electronic contribution to the adsorption energy from the data in Fig. 2. It has become common to relate this kind of contribution to the center of mass of the electronic density of states projected on the atomic d orbitals of the metal before adsorption 共the d-band center兲, see for instance Refs. 17, 39– 41. We have calculated the d-band center from the density of states of these surfaces, and the result is shown in Fig. 3. Here the adsorption energy is given as a function of the distance of the d-band center from the Fermi level for all the high coordination sites with only one kind of neighbor. There is a clear correlation between the adsorption energy and the d-band center at all the sites, and we see that all the sites show the same behavior: as the center of the d band moves toward the Fermi level, the adsorption gets less stable. But this is the opposite trend that has been reported from previous theoretical studies, which all have seen that adsorption gets more stable when the center of the d band moves toward the Fermi level. It is not surprising that the center of the d band moves toward the Fermi level as we mix more Pd together with the Ag atoms. But why is this in our case followed by a decrease in reactivity, rather than an increase, as in other systems? In our previous study on hydrogen in bulk palladium–silver alloys, we found that an increased lattice constant was important to explain more stable absorption of hydrogen when Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp 3272 J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 O. M. Løvvik and R. A. Olsen FIG. 3. The adsorption energy at sites with only one kind of metal neighbors as a function of the center of the silver d band, illustrating the contribution from electronic effects to the adsorption energy. silver was added to the alloy.7 Mavrikakis et al. also found that the lattice constant was important to the reactivity of various metal surfaces.40 We have checked the influence of the lattice constant on some sites, and found only small changes. As an example, we calculated the adsorption energy at top sites above a silver atom, and found that it changed by less than 20 meV when we changed the lattice constant from the alloy optimized lattice constant to the silver lattice con- stant. The difference in adsorption energy between these sites remained, or even increased. We thus have to look more closely at the details of the electron density, and have calculated the local density of states 共LDOS兲 and the crystal orbital overlap population 共COOP兲 of the relevant sites 共see, e.g., Ref. 42兲. We have plotted the LDOS both with and without hydrogen at the hcp site in Fig. 4 共the fcc site looks quite similar兲. We see that the FIG. 4. LDOS of a top layer silver atom for pure and hydrogen-covered slabs with hydrogen adsorbed at the hcp site. The upper and lower panels show the DOS projected on the silver sp and d bands, respectively. The dotted and solid curves show the LDOS of the silver atom before and after adsorption of a hydrogen atom, respectively. The energy is measured relative to the Fermi level. Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 Hydrogen adsorption on palladium–silver alloys 3273 FIG. 5. COOP 共solid curves兲 and integrated COOP 共dotted curves兲 between a top layer silver atom and an adsorbed hydrogen atom located at the hcp site. The COOP is between silver sp bands and hydrogen in the upper panels, and between silver d bands and hydrogen in the lower panels. The energy is measured relative to the Fermi level. 1s split-off state at about 7 eV below the Fermi level is clearly larger on Ag and Ag/Pd than on Ag3 Pd/Pd, being consistent with a smaller reactivity between Ag and H on Ag3 Pd/Pd (E ads⫽0.39 eV) than on Ag and Ag/Pd (E ads ⫽0.23 and 0.28 eV, respectively兲. The same trend can also be seen in a plot of the COOP between hydrogen and silver at the same site, shown in Fig. 5. We can here clearly see how the overlap between the silver 4d and the hydrogen 1s band is much smaller at Ag3 Pd/Pd than at the other alloys. In addition, we see that the overlap between the Ag sp band and hydrogen is smallest at Ag3 Pd/Pd. The integrated COOP up to the Fermi level is also shown in each plot, giving support to the above-given description. This means that inclusion of more Pd not only affects the d electrons, but also the sp electrons of Ag. This is different from the common assumptions of previous studies, that the interaction with sp electrons remains constant.17,40 Geometric 共‘‘ensemble’’兲 contributions to the adsorption energy have in other systems been found to depend on the center of the d band in the same way as electronic 共‘‘ligand’’兲 contributions to the adsorption energy, only stronger.17,40 That is, the adsorption gets stronger when the d-band center moves toward the Fermi level. We are not able to separate the electronic and geometric effects fully in this study, since we have not included enough different surfaces to calculate all possible electronic surroundings for each geometric configuration. But we are able to find the total effect (geometric⫹electronic), since we have calculated all possible geometric configurations. This is shown in Fig. 6, where the adsorption energy is plotted as a function of the fraction of silver atoms at the different sites. It is evident that the total effect is working the same way here as in other systems, with the adsorption strength decreasing as the fraction of silver atoms at the site increases, and thus as the d-band center moves away from the Fermi level. Since the ligand effect works the opposite way 共see above兲, we conclude that the ensemble effect is the largest, and responsible for the direction of the total effect. It is also quite straightforward to understand the direction of the ensemble effect by studying the COOP between hydrogen and the metal d electrons when hydrogen is adsorbed on a hollow site at the pure metal slabs, as shown in Fig. 7. We see there that the effect FIG. 6. The adsorption energy at fcc sites as a function of the fraction of silver atoms at the site. The symbols refer to the same surfaces as in Fig. 2. Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp 3274 J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 O. M. Løvvik and R. A. Olsen FIG. 8. The adsorption energy along possible surface diffusion paths as lined out in Fig. 1. FIG. 7. The crystal orbital overlap population between the hydrogen s electrons and metal d electrons when hydrogen is adsorbed at the fcc site on the pure silver and palladium surfaces. The energy is measured relative to the Fermi level. of moving from the Pd to the Ag site is twofold: the large bonding peak at the bottom of the overlap is moved away from the Fermi level, while a broad antibonding peak is moved from above the Fermi level to below. This means that, as we include more silver atoms at a site, more antibonding orbitals are occupied, and the reactivity decreases. The correlation between the presence of silver and the adsorption energy in Fig. 6 means that, at the crudest level, the adsorption energy at a particular site can be estimated as the interpolation between the ‘‘pure’’ sites, only taking the nearest neighbors into account. Deviations from this approximation are due to the combination of geometric and electronic effects, and must be calculated directly in each case. D. Diffusion paths There are numerous possible surface diffusion paths on real alloy surfaces, and in order to model some typical possible paths, we have chosen a representative diffusion path on each slab, shown in Fig. 1. The resulting adsorption energy is plotted in Fig. 8. We see that the adsorption energy of hydrogen is so much lower at sites with at least one palladium atom, that palladium atoms at the surface would act as hydrogen traps. At low palladium concentrations in the surface this should not be important, since after the first hydrogen atoms have been trapped by palladium, new hydrogen atoms should be able to diffuse between the pure silver sites. The opposite is the case at high palladium concentrations; then hydrogen should be able to diffuse between sites containing palladium. At intermediate concentrations, however, we should expect that surface diffusion is effectively inhibited by the palladium atoms acting as hydrogen traps. This is in line with the results of Ref. 43, where the diffusion constant through bulk palladium silver alloys was found as a function of the silver content. They found that the diffusion constant was constant from 0 to 25% silver content, then dropped by three orders of magnitude from 25% to 60%, and after that increased back to the diffusion constant of silver again, which is about the same as that of palladium. We are not aware of any studies measuring the surface diffusion constant of hydrogen on a PdAg共111兲 surface, but note that the untreated 共111兲 and 共100兲 surfaces of Pd67Ag33 contain only 5.2% and less than 0.05% palladium, respectively.6 This is clearly not enough to inhibit surface diffusion, but we would expect that hydrogen sticks to the palladium sites when possible. This is exactly what has been observed in a STM study of oxygen on a PdAg surface, where oxygen is found to occupy palladium sites only, and oxygen diffusion takes place between the Pd sites.44 On real PdAg membranes containing around 25% silver, the hydrogen transport is very effective,4,45,46 even with a pure silver surface added.47 This is apparently in conflict with our results, where all sites with only silver atoms have E ads⭓0.20 eV. This could be because of real membrane surfaces differing from the perfect 共111兲 surface, possibly with different surface plane orientations, polycrystallinity, or defects. It could also be that we obtain too high adsorption energies in this study, because of high hydrogen coverage 共25%兲 and uncomplete relaxation. The relatively high hydrogen coverage could possibly be a problem, since the solubility of hydrogen in silver is very low. We have calculated the adsorption energy of hydrogen on a pure silver 共111兲 surface, and found that it decreases from 0.36 to 0.20 eV when the coverage decreases from 1 to 0.25. This is much more than on Pd, where E ads decreases from ⫺0.39 to ⫺0.44 eV with the same coverage variation.48 We thus expect that calculated adsorption energies of hydrogen on silver at very low coverage should be lower than the present results. IV. CONCLUSIONS We have calculated properties of various pure and hydrogen covered 共111兲 surfaces of palladium silver alloys through DFT within the GGAs employing a slab representation of the surface. Our calculated lattice constants are in good agreement with known experimental data, but we find a Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 substantially lower surface energy for Ag共111兲 and Pd共111兲 than experiments33,34 and some previous theoretical studies.29,35 Our results are, however, in line with the results of a total energy pseudopotential study,30 and we also get the same result when employing two different accurate DFT schemes. The potential energy surface of hydrogen on the pure Ag共111兲 and Pd共111兲 surfaces are quite similar 共with the threefold hollow sites being the most stable兲, except that adsorption is about 0.7 eV less stable on the Ag surface. On the alloyed surfaces we found that threefold hollow sites with as many palladium neighbors as possible are preferred. We assumed that the adsorption energy consists of geometric 共‘‘ensemble’’兲 and electronic 共‘‘ligand’’兲 contributions, and found the geometric contribution 共the type of atoms to which H bonds兲 to dominate. The adsorption on the alloy surfaces can therefore in the crudest approximation be found by interpolation between the ‘‘pure’’ adsorption sites, where the fraction of Ag atoms at the adsorption site is the most important parameter in determining the adsorption energy. For the geometric contribution it is seen that the binding strength increases as the d-band center moves toward the Fermi level, a result also found by a number of other theoretical studies. Considering adsorption sites with the same geometric constitution 共the type of atoms to which H bonds兲, but with different surroundings, we could isolate the electronic 共‘‘ligand’’兲 contributions to the adsorption energy 共as also done by others兲. We find that hydrogen adsorption gets less stable when the d-band center moves toward the Fermi level. This is contrary to previous studies, where the opposite trend has been found. We suggest that this can partly be attributed to a large variation of the overlap between the hydrogen 1s state and the metal sp band, which is assumed to be constant in other studies. We have also found that hydrogen binds much more strongly to palladium than to silver, so that palladium atoms work as an effective hydrogen trap. We expect that this should make surface diffusion difficult when the fraction of palladium atoms at the surface is too small to make diffusion between palladium atoms possible, and large enough to inhibit diffusion at silver sites only. At real membrane surfaces, however, the palladium content is probably too low for this to become a problem. ACKNOWLEDGMENTS Financial support and computer time from the Norwegian Research Council 共OML兲 is acknowledged. R.A.O. acknowledges financial support from the Dutch National Research Council-Chemical sciences 共NWO-CW兲 and the National Research School Combination ‘‘Catalysis Controlled by Chemical Design’’ 共NRSC-Catalysis兲. 1 Hydrogen in Intermetallic Compounds I. Electronic, Thermodynamic, and Crystallographic Properties, Preparation, edited by L. Schlapbach 共Springer, Berlin, 1988兲. 2 Hydrogen in Intermetallic Compounds II. Surface and Dynamic Properties, Applications, edited by L. Schlapbach 共Springer, Berlin, 1992兲. 3 J. Shu, B. P. A. Grandjean, A. Vanneste, and S. Kaliaguine, Can. J. Chem. Eng. 69, 1036 共1991兲. Hydrogen adsorption on palladium–silver alloys 3275 S. Uemiya, Sep. Purif. Methods 28, 51 共1999兲. S. Müller and A. Zunger, Phys. Rev. Lett. 87, 165502 共2001兲. 6 P. T. Wouda, M. Schmid, B. E. Nieuwenhuys, and P. Varga, Surf. Sci. 417, 292 共1998兲. 7 O. M. Løvvik and R. A. Olsen, J. Alloys Compd. 330–332, 332 共2002兲. 8 B. Hammer and J. K. Nørskov, Surf. Sci. 343, 211 共1995兲. 9 B. Hammer and M. Scheffler, Phys. Rev. Lett. 74, 3487 共1995兲. 10 B. Hammer, Y. Morikawa, and J. K. Nørskov, Phys. Rev. Lett. 76, 2141 共1996兲. 11 F. Delbecq and P. Sautet, Chem. Phys. Lett. 302, 91 共1999兲. 12 F. Delbecq and P. Sautet, Phys. Rev. B 59, 5142 共1999兲. 13 F. Delbecq and P. Sautet, Surf. Sci. 442, 338 共1999兲. 14 V. Pallassana, M. Neurock, L. B. Hansen, B. Hammer, and J. K. Nørskov, Phys. Rev. B 60, 6146 共1999兲. 15 V. Pallassana, M. Neurock, L. B. Hansen, and J. K. Nørskov, J. Chem. Phys. 112, 5435 共2000兲. 16 S. Y. Choi, Y. S. Kwon, T. H. Rho, S. C. Hong, and J. I. Lee, J. Korean Phys. Soc. 37, 104 共2000兲. 17 P. Liu and J. K. Nørskov, Phys. Chem. Chem. Phys. 3, 3814 共2001兲. 18 R. Hirschl, Y. Jeanvoine, G. Kresse, and J. Hafner, Surf. Sci. 482–485, 712 共2001兲. 19 D. Loffreda, F. Delbecq, D. Simon, and P. Sautet, J. Chem. Phys. 115, 8101 共2001兲. 20 D. Loffreda, F. Delbecq, D. Simon, and P. Sautet, J. Phys. Chem. B 105, 3027 共2001兲. 21 G. te Velde and E. J. Baerends, Phys. Rev. B 44, 7888 共1991兲. 22 G. te Velde and E. J. Baerends, J. Comput. Phys. 99, 84 共1992兲. 23 A. D. Becke, Phys. Rev. A 38, 3098 共1988兲. 24 J. P. Perdew, Phys. Rev. B 33, 8822 共1986兲. 25 E. van Lenthe, E. J. Baerends, and J. G. Snijders, J. Chem. Phys. 101, 9783 共1994兲. 26 J.-F. Paul and P. Sautet, Surf. Sci. 356, L403 共1996兲. 27 R. A. Olsen, G. J. Kroes, and E. J. Baerends, J. Chem. Phys. 111, 11155 共1999兲. 28 P. Statiris, H. C. Lu, and T. Gustafsson, Phys. Rev. Lett. 72, 3574 共1994兲. 29 M. Methfessel, D. Hennig, and M. Scheffler, Phys. Rev. B 46, 4816 共1992兲. 30 Y. Wang, W. Wang, K.-N. Fan, and J. Deng, Surf. Sci. 490, 125 共2001兲. 31 V. Fiorentini and M. Methfessel, J. Phys.: Condens. Matter 8, 6525 共1996兲. 32 J. C. Boettger, J. R. Smith, U. Birkenheuer, N. Rosch, S. B. Trickey, J. R. Sabin, and S. P. Apell, J. Phys.: Condens. Matter 10, 893 共1998兲. 33 W. R. Tyson and W. A. Miller, Surf. Sci. 62, 267 共1977兲. 34 F. R. de Boer et al., Cohesion in Metals: Transition Metal Alloys 共NorthHolland, Amsterdam, 1988兲. 35 L. Vitos, A. V. Ruban, H. L. Skriver, and J. Kollár, Surf. Sci. 411, 186 共1998兲. 36 J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J. Singh, and C. Fiolhais, Phys. Rev. B 46, 6671 共1992兲. 37 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 共1996兲. 38 P. Blaha, K. Schwarz, G. K. H. Madsen, D. Kvasnicka, and J. Luitz, WIEN2k, An Augmented Plane Wave⫹Local Orbitals Program for Calculating Crystal Properties 共Karlheinz Schwarz, Techn. Universität Wien, Austria, 2001兲. 39 Y. L. Lam, J. Criado, and M. Boudart, Nouv. J. Chim. 1, 461 共1997兲. 40 M. Mavrikakis, B. Hammer, and J. K. Nørskov, Phys. Rev. Lett. 81, 2819 共1998兲. 41 Y. Gauthier, M. Schmid, S. Padovani, E. Lundgren, V. Bus, G. Kresse, J. Redinger, and P. Varga, Phys. Rev. Lett. 87, 036103 共2001兲. 42 R. Hoffmann, Solids and Surfaces: A Chemist’s View on Bonding in Extended Structures 共VCH Verlagsgesellschaft, Weinheim, 1988兲. 43 H. Barlag, L. Opara, and H. Züchner, J. Alloys Compd. 330–332, 434 共2002兲. 4 5 Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp 3276 44 J. Chem. Phys., Vol. 118, No. 7, 15 February 2003 P. T. Wouda, M. Schmid, B. E. Nieuwenhuys, and P. Varga, Surf. Sci. 423, L229 共1999兲. 45 E. Wicke and H. Brodowsky, in Hydrogen in Metals II. Applicationoriented Properties, edited by G. Alefeld and J. Völkl 共Springer, Berlin, 1978兲. 46 J. K. Ali, E. J. Newson, and D. W. T. Rippin, J. Membr. Sci. 89, 171 共1994兲. O. M. Løvvik and R. A. Olsen 47 H. Amandusson, L.-G. Ekedahl, and H. Dannetun, J. Membr. Sci. 193, 35 共2001兲. 48 O. M. Løvvik and R. A. Olsen, Phys. Rev. B 58, 10890 共1998兲. 49 C. Kittel, Introduction to Solid State Physics, 7th ed. 共Wiley, New York, 1996兲. 50 H. Ohtani, M. A. Van Hove, and G. A. Somorjai, Surf. Sci. 187, 372 共1987兲. Downloaded 26 Feb 2003 to 129.240.85.53. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp