PHYSICAL REVIEW B VOLUME 62, NUMBER 23 15 DECEMBER 2000-I Adsorption energies and bond lengths of adatoms at surfaces simulated by clusters D. Geschke, S. Fritzsche, W.-D. Sepp, and B. Fricke Fachbereich Physik, Universität Kassel, Heinrich-Plett-Strasse 40, D-34132 Kassel, Germany S. Varga Department of Physics, Chalmers University of Technology, S-41296 Göteborg, Sweden J. Anton Department of Atomic Physics, Stockholm University, Fescativägen 24, S-10405 Stockholm, Sweden 共Received 30 December 1999兲 Using a self-consistent relativistic molecular density-functional method we have calculated adsorption energies and bond lengths of adatoms on surfaces simulated by clusters within the full relativistic description. The results indicate that our method is working so well that first big clusters can be handled in a realistic time and second the numerical accuracy is so good that there is justified hope that adsorption sites and energies can be calculated with improved density functionals in the future. I. INTRODUCTION Calculations on the electronic and geometric structure of polyatomic systems are not only important in solid-state and cluster physics, they are also useful in surface physics to get a deeper insight and a better understanding of processes which may occur on surfaces, most notably catalytic reactions and interactions of single atoms or small molecules with surface atoms.1 The most favorite adsorption site determines which processes may occur 共charge transfer, replacements兲 and how surface properties of a clean surface may be influenced. For a long time, calculations of this physical situation have been performed with different approaches.2–5 Band-structure methods raise difficulties as soon as periodic boundary conditions are no longer applicable. On the other hand, large cluster calculations become expensive and unrealistically time consuming with an increasing number of atoms. However, they offer the possibility of treating the local interactions accurately. Difficulties may arise when heavy elements are involved, because the influence of relativistic effects may be significant. Due to several improvements of our four-component relativistic molecular density functional code, we are now able to calculate total energies as well as bond distances of small molecules or highly symmetric systems such as fullerenes with sufficient accuracy. This enables us to predict their ground-state energies and geometries as well as the energetic and structural differences arising from relativistic effects in heavy elements.6,7 Consequently, we have recently started to extend our method to apply it to larger molecules and/or describe reactions of adatoms with atomic clusters towards an ab initio quantum-mechanical description of moleculesurface interactions. In this paper we present results of atom-cluster calculations. Our aim is not to reproduce experimental data of a specific system but to present a relativistic method which can be used in future to treat problems of more physical relevance. The aim is to show first that big clusters which simulate the solid can be handled in a realistic time, and second that the numerical accuracy is so good that there is hope that 0163-1829/2000/62共23兲/15439共4兲/$15.00 PRB 62 adsorption sites and energies can be calculated with improved density functionals in future. Since barium is a system with dominant outer s-electron valence character which already exhibits strong relativistic effects, we have chosen this system, Ba at a Ba共110兲 surface, as the first test to see if this method can actually be applied, although there are no experimental results for this system. Nevertheless, barium and barium compounds are widely used in material science due to its dielectric properties, for example in titanate thin films8 or in superconductors.9,10 There exist also some theoretical investigations of barium fluoride in luminescent materials.11 Rosen et al. reported on calculations of the electronic structure of BaF2共111兲 surfaces for laser-induced emission of electrons and ions.12 The choice of Ba at a Ba共110兲 surface is also a good starting point because it probably will not reconstruct or relax, so we do not have these complications in this very first test system. II. COMPUTATIONAL Our method is based on a fully relativistic implementation of the density-functional discrete variational method 共DVM兲,13 which is explained at some length in Ref. 14. In our calculations we have used the Slater X ␣ approximation setting ␣ ⫽0.7 to account for correlation. We are aware of the fact that the Slater X ␣ approximation is not the best xc potential and there exist better approximations like such as the generalized gradient approximation 共GGA兲. The results for Au2 which are reported in Ref. 14 are much improved with the newer and better density functionals,15 but from a technical point of view they cannot yet be applied to such big cluster calculations. So we still use the Slater X ␣ approximation here in order to get answers to the following questions: Is the description of an adsorption process within this method possible? Do we get reasonable and numerically accurate potential-energy curves and can we distinguish the various adsorption sites? An essential feature to perform calculations of larger molecules and clusters is the frozen core approximation 共FCA兲, 15 439 ©2000 The American Physical Society BRIEF REPORTS 15 440 PRB 62 valence E FCA⫽ 兺v ⫺ 14 ⫺ ⑀ v ⫺ 12 冕 兺 ij ⬘ v 冕 v V Cv 共 v 兲 d 3 r V xc共 c ⫹ v 兲 ⫹ 34 c Z cj 共 Z j ⬘ ⫺ 12 Z j ⬘ 兲 i⫽ j ⬘ R j j⬘ 冕兺 j M ⫹ 兺 m,n m⬎n c xc c 3 v j V 共 ⫹ 兲d r Z nZ m . 兩 Rn ⫺Rm 兩 共3兲 ⑀ v ⫽ 具 v 兩 关 t⫹V N ⫹V C ( c ⫹ v )⫹V xc( c ⫹ v ) 兴 兩 v 典 Here, represent the one-particle energies with the external nuclear potential V N , the Hartree potential V C , and the exchange potential ( ␣ ⫽0.7), FIG. 1. Potential-energy curve of an adsobed Ba atom at the top position of a Ba共110兲 surface simulated by a nine-atom cluster. which we have implemented here in our fully relativistic program. Due to the fact that the binding mainly arises from the valence electrons, one can neglect the SCF part of the core contributions. These may further be justified by the observation that the main part of the core contributions cancels out when examining the binding energy. Here, one has to subtract the total energies of the system adatom plus cluster and its fragments. The FCA is based on the decomposition of the Hilbert space into a core 共c兲 and a valence ( v ) part. Accordingly, the molecular wave functions can be written as 兩 典 ⫽ 兩 c典 ⫹ 兩 v典 . 共1兲 To ensure the constrained Ritz variation of the valence orbitals, a core-valence orthogonalization is explicitly carried out in each iteration. The transformed wave functions, after orthogonalization, can be written as 兩 c⬘ 典 ⫽ 兩 c 典 , 兩 ⬘v 典 ⫽ 兩 c 典 ⫺S v c 兩 c 典 . Here, S v c represents the core-valence orthogonalization matrix. For chemical applications, the atomic-core wave functions c are used to represent the core parts of the molecular wave functions. However, this should be proven in every case. As a simple criterion for an appropriate definition of core and valence separation, one has to ensure that the overlap of the core states of different atoms can be neglected. As a consequence, the ansatz 共1兲 leads to a block structure of the Fock and overlap matrices, H⫽ 冉 H cc H cv H vc H vv 冊 , 共2兲 with the core-core (H cc ), core-valence (H c v ), and valencevalence (H vv ) parts, respectively. In the calculations the core orbitals are not varied among the SCF iterations and cause no direct changes in the molecular 共valence兲 potential. In the FCA, the core-valence matrix elements in the Fock matrix are neglected. This is the real approximation within the FCA. By treating the core densities of different atoms as point charges, one ends up with the expression for the total energy,16 V xc„ 共 r兲 …⫽⫺3 ␣ 冉 3 共 r兲 8 冊 1/3 . 共4兲 Z cj is the core charge of atom j located at the position R j , while cj represents the corresponding electronic core density. For the molecular calculations we used the C 2 v symmetry. The neutral atomic wave functions of the barium atom were used as basis functions. The core was defined by the 1s 1/2-4d 5//2 atomic orbitals while the 5s 1/2 , 5 p 1/2 , 5p 3/2 , 6s 1/2 , 6p 1/2 , and 6p 3/2 orbitals were considered as the active valence states. In order to check this definition of the core and valence space, we have performed calculations of the Ba2 dimer, first with all electrons and second within the FCA. We found a difference of less than 0.02 eV in the binding energy, which indicates that our choice seems to be appropriate. Furthermore, looking at the Mulliken occupation numbers near the equilibrium distance, we observed no significant changes even for the 5s 1/2 orbitals, reflecting the fact that these orbitals as well as the lower-lying atomic states do not change their atomic character. After this we carried out a basis optimization with respect to the total energy and found the best results for slightly ionized atomic wave functions from Ba⫹0.2. Within this core-valence decomposition and the optimized basis set, we have calculated the cluster with the highly accurate three-dimensional numerical multicenter integration scheme by Boerrigter et al.17 III. RESULTS AND DISCUSSION The results of these first relativistic cluster calculations for the adsorption energy and the adsorption site as a function of the cluster size are given in Figs. 1–3. The choice of the system Ba at Ba共110兲 and the approximations of the density functional have been discussed above. Figures 1 and 2 show the results of Ba at Ba共110兲 for the top position and the hollow position, respectively, as functions of the distance to the surface. In both cases the cluster was modeled by nine barium atoms. Relaxations of the cluster atoms were not taken into account in this case. The shaded atoms in both figures have to be considered as a continuation in the next surface layer but were not included in the calculations. We do clearly observe an energetic difference of more than 0.15 eV between the top and hollow position. These very first results indicate that our method seems to be working, both from a technical as well as from a numerical point of view. PRB 62 BRIEF REPORTS FIG. 2. Potential-energy curve of an adsobed Ba atom at the hollow position of a Ba共110兲 surface simulated by a nine-atom cluster. In a second step, we are now starting from the hollow adsorption site which was found to be the most stable position for the nine-atom cluster. Our goal is to study the dependence of the atom-cluster interaction as a function of the cluster size. By successively increasing the number of atoms in the first layer of the simulated surface, we end up with calculations including 14 and 18 atoms 共Fig. 3兲 which corresponds to an 8-atom and a 12-atom first layer, respectively. This is a simple but very obvious extension of the cluster which neglects possible contributions of additional atoms in a second or even a third layer, reflecting the fact that the adsorption is supposed to be local. To prove this assumption, we have tested the inclusion of four additional atoms for the 14-atom cluster in the second layer but we did not find any considerable difference within our accuracy of about 0.02 eV. On the other hand, this result shows the sensitivity of our method. As one can see, the potential-energy curves near the equilibrium distance for the 8-atom and 12-atom layer are very close, which indicates that we have already converged in cluster size. The rapid convergence with respect to the cluster size for this system Ba at Ba共110兲 may be exceptional. Probably the reason is that the bonding is dominantly done by the isotropic outer s electrons only. For any other atomic species where the bond character is different and anisotropic, the convergence with respect to the cluster size may differ very much and depend on the type of the lattice as well.18,19 Although the absolute value of the adsorption energy C. R. Henry, Surf. Sci. Rep. 31, 232 共1998兲. J. L. Whitten and T. A. Pakkanen, Phys. Rev. B 21, 4357 共1980兲. 3 J. L. Whitten, Chem. Phys. 177, 387 共1993兲; J. L. Whitten and H. Yang, Surf. Sci. Rep. 24, 55 共1996兲. 4 T. Bredow, Surf. Sci. 401, 82 共1998兲. 5 M. Casarin, C. Maccato, and A. Vittadini, Appl. Surf. Sci. 142, 196 共1999兲. 6 T. Baştuğ, K. Rashid, W.-D. Sepp, D. Kolb, and B. Fricke, Phys. Rev. A 55, 1760 共1997兲. 7 T. Baştuğ, P. Kürpick, J. Meyer, W.-D. Sepp, B. Fricke, and A. Rosén, Phys. Rev. B 55, 5015 共1997兲. 1 2 15 441 FIG. 3. Potential-energy curves of an adsorbed Ba atom in the hollow position at a Ba共110兲 surface as function of the number of atoms which simulate the surface. probably is not correct due to the simple approximation of the exchange correlation potential, we nevertheless can conclude the following: 共i兲 such a full relativistic calculation is possible within an acceptable computing time; 共ii兲 the numerical accuracy is so high that realistic potential-energy curves can be calculated; 共iii兲 different surface positions can be calculated which allow a choice of the preferred adsorption site; 共iv兲 a frozen-core calculation within the relativistic code can be applied which leads to the same result as a full electron calculation but is much less time consuming; 共v兲 at least for Ba at Ba共110兲, a convergence with respect to the cluster size can be reached. Thus, as a concluding remark we can state that our method is appropriate for the description of atom-cluster interactions. As we already mentioned above, these investigations should be considered as a first step towards a more sophisticated method. Concerning more state of the art potentials, we have quite recently implemented nonlocal gradient corrected exchange-correlation functionals20 which hopefully can be applied for such cluster calculations soon. Furthermore, we will try to embed the 共adatom兲-cluster system in an external potential which should simulate an infinitely extended crystal and ensure a proper connection with the rest of the solid. ACKNOWLEDGMENTS D.G. gratefully acknowledges support by the Deutsche Forschungsgemeinschaft 共DFG兲, S.V. acknowledges support by the DFG and DAAD, and J.A. by GSI and DFG. 8 T. Sonegawa, C. Grigoriu, K. Masugata, K. Yatsui, Y. Shimotori, S. Furuuchi, and H. Yamamoto, Appl. Phys. Lett. 69, 2193 共1996兲. 9 N. P. Bansal and A. L. Sandkuhl, Appl. Phys. Lett. 52, 323 共1988兲. 10 G. Kordas and M. R. Teepe, Appl. Phys. Lett. 57, 1461 共1990兲. 11 J. M. Vail, E. Emberly, T. Lu, M. Gu, and R. Pandey, Phys. Rev. B 57, 764 共1998兲. 12 A. Rosén, E. Westin, E. Matthias, H. B. Nielsen, and J. Reif, Phys. Scr. T23, 194 共1988兲. 13 A. Rosén, Adv. Quantum Chem. 29, 1 共1997兲. 15 442 14 BRIEF REPORTS T. Baştuğ, D. Heinemann, W.-D. Sepp, D. Kolb, and B. Fricke, Chem. Phys. Lett. 211, 119 共1993兲. 15 S. Varga et al., J. Chem. Phys. 共unpublished兲. 16 T. Baştuğ, Ph.D. thesis, University of Kassel, 1994. 17 P. M. Boerrigter, G. Te Velde, and E. J. Baerends, Int. J. Quantum Chem. 33, 87 共1988兲. 18 PRB 62 X. Shangda, G. Changxin, L. Libin, and D. E. Ellis, Phys. Rev. B 35, 7671 共1986兲. 19 B. Song, H. Nakamatsu, R. Sekine, T. Mukoyama, and K. Taniguchi, J. Phys.: Condens. Matter 10, 9443 共1998兲. 20 S. Varga, E. Engel, W.-D. Sepp, and B. Fricke, Phys. Rev. A 59, 4288 共1999兲.
© Copyright 2024 Paperzz