Surface structure of cubic aluminum oxide

PHYSICAL REVIEW BVOLUME 50, NUMBER 415 JULY 1994-II Surface structure of cubic aluminum oxide Luis Javier Alvarez and Luis Eduardo Leon Cuf dad Universitaria, 04510Mexico, Distrito Federal, Mexico Javier Fernandez Sanz* Departamento de Quimica Fisica, Facultad de Quimica, Universidad de Sevilla, E4101-2 Sevilla, Spain Maria Jose Capitan and Jose Antonio Odriozola Departamento de Quimica Inorganica eInstituto de Ciencia de Materiales, Universidad de Sevilla Co-nsej oSuperior de Investigaciones Cientificas, E41012-Sevilla, Spain (Received 22 February 1994) Molecular-dynamics simulations using aPauling type pairwise potential have been carried out in order to study the structure of the surface of y-A1203. Starting from an ideal (100) face, areconstruction process occurs in which the unbalanced coordination of surface ions tends to fulfill their coordination capabilities. Oxygen ions come to the surface and pore formation with aconcomitant increase of apparent volume takes place. The structure ofthe surface ofthe resulting material corresponds to aporous amorphouslike phase in which only short-range order is present. Analysis of the radial distribution function agrees with experimental x-ray-diffraction data. I. INTRODUCTION Since the pioneering work by Peri' the surface of yA1203 has been the object ofconsiderable interest because of its importance as catalyst and as acatalyst support. Most of the surface models are idealized views of the catalyst surface. In Peri's model the (100) surface of NaC1 type structure is considered when it is well known that y-A1203 has adefective spinel structure. Later models considered the actual structure of the solid but with an excess of aluminum cations to solve the problem of the random occupation oftetrahedral and octahedral sites. ' In all these surface models the coordination of cations at the surface is considered to be tetrahedral and octahedral. However, this assumption disagrees with data reported recently by Chen, Davis, and Fripiat, who from magic angle spinning NMR Al spectroscopy found the existence of five-coordinated aluminum atoms. Recently, we reported molecular-dynamics simulations on y-A1203 in which the actual stoichiometry was taken into account. Also, from cleavage fracture planes of the bulk, idealized faces according to the most usually accepted exposed surfaces were reported. In our work, we showed that five-coordinated aluminum ions occur both in the bulk and at the surface. Nevertheless, none of these efforts to understand the structure of the y-A1203 considered the reconstruction ofthe outer layers ofthe solid. Surface reconstruction of metals is awell-described phenomenon; however, in the case of nonconducting solids such as metal oxides, very little has been achieved due to the experimental limitations regarding the obtainment of LEED patterns, and only reconstruction of few surfaces such as ZnO, Ti02, and a-A1203 have been described. The experimental study ofthe reconstruction of y-Alz03 has the additional limitation of the crystal size, II. COMPUTATIONAL PROCEDURE The interaction potential used in our simulations is the same as the one used in previous works on alumina. This potential was shown to give excellent results concerning both the structure of the bulk, through the calculation of radial distribution functions, and the energetics of the system, predicting heat of formation and vibrational spectra in agreement with experimental data. The potential is of Pauling type and has aCoulombic term plus asteric repulsion given by 2 q;q.e V(rj)= 1— sgn(q;q. ) lj CTI +CTJ where r,.-is the interionic distance, q; are the effective charges, and cr,-are the ionic radii. The exponent nwas taken to be 9according to Adams and McDonald. The effective charges of aluminum and oxygen were calculated from ab initio self-consistent-field — molecular orbital calculations carried out on aseries of small clusters of Al(OH)„using HQNDo-8. 4program. 'In these clusters, xranges from 2to 6, and thus effective charges are the weighted average of the most common coordination poalthough experimental data based on adsorption of probe molecules suggest that asurface reconstruction indeed takes place. The theoretical and experimental limitations mentioned above make the study of the structure of the surface of metal oxides in general and particular of y-Alz03 suitable for numerical simulation. In this work we report molecular-dynamics simulations of the surface layers of y-A1203 in order to study their structure and surface reconstruction. 0163-1829/94/50(4)/2561(5)/$06. 00 50 2561 1994 The American Physical Society 2562 ALVAREZ, LEON, SANZ, CAPITAN, AND ODRIOZOLA 50 lyhedra of cubic alumina. The oxygen radius was that proposed by Shannon and Prewitt, "whereas aluminum radius was chosen in such way as to minimize the difFerences between tetrahedral and octahedral coordinations. ' The system consisted of 1440 particles in an originally cubic configuration of 27 unit cells at adensity of 3.66 gcm .Periodic boundary conditions were taken in the xand ydirections. The long-range Coulombic interaction was handled with the Ewald summation method and atime step of 10 ps was used. The initial configuration was taken from aprevious work, where the system was equilibrated at 300 Kfor 5.5ps in the bulk. After removing the boundary conditions in the zdirection only the upper 720 particles were allowed to move, leaving the system free to evolve for 10 ps rescaling velocities when temperature increased above 300 K. After this equilibration process a5ps production run was performed in which temperature variations were of less than 5' and energy fluctuations were of less than 0.1%. Calculations were performed using SIMULA code running on the Cray YMPl464 supercomputer of the Universidad Nacional Autonoma de Mexico computing center. III. RESULTS AND DISCUSSION The removal of the boundary conditions in the zdirection generates coordination defects of surface ions" which, in turn, give way to uncompensated interaction potential energy. The system tends to eliminate defects and to compensate the interaction potential energy by altering both position and coordination of atoms at the outermost layers. Along with this compensation there is apore formation process which results in an increase of the apparent volume. After the equilibration stage the local structural characteristics such as bond distances and coordination numbers of the particles are preserved, keeping only short-range order. Figure 1shows a snapshot of the moving particles of the system after the 15 ps simulation run where some pores can be observed. A. Porosity In order to give an insight into the existence and distribution of pores in the surface, we have calculated the apparent density profile along the zdirection shown in Fig. 2. The plotted values were obtained calculating the density of aset of horizontal slabs 3Athick. At depths larger than half the original box size (11.8A) the density values lie around the density of the original system in the bulk of 3.66 gcm, and the values at lower depths vary around 2.3gem . The variations of density in this region are due to the presence of large pores in the structure, and the lower oscillations around the mean value with respect to acrystalline structure are due to the nearly homogeneous distribution of pores. The calculated pore volume yields 0.07 cm g', which is within the limits of the expected range of micropore volume reported for y-Alz03 between 0.04 and 1.3cm g'.'This value was obtained on the basis of fitting spheres of radii larger than 2.0Ato avoid accounting for interstitial volume. Assuming slit-shaped pores, as experimentally observed, the minimum distance between the alumina lamellae would be 4.0A. The average width being 5.1Awith a maximum of up to 8.2A. Figure 3shows aview of the surface layers up to adepth of 5Awhere cavities of difFerent sizes can be observed. The largest one near the upper left corner is 10.5Along, 4.0Awide and its depth was estimated to be around 8A. From our simulations it can be stated that the micropore volume is not only due to empty octahedral sites as has been suggested, but also and more important, to the existence of genuine pores in an amorphous superficial phase. In Fig. 3, where alarge elliptical pore is shown, it can ~~~~ !l! (QiPgcp ~pIII 0 — 25 — 20 — 15 — 10 z&A) 10 FIG. 1. Snapshot of the moving particles of the system after 15 ps simUlation time. FIG. 2. Apparent density profile along the zdirection. 50 SURFACE STRUCTURE OF CUBIC ALUMINUM OXIDE 2563 tems. Taking into account the experimental conditions under which the synthesis of cubic aluminas is performed, namely high temperature, they would never expose asurface of the (100)type, and in fact, areconstruction such as that reported in this work would never occur. B. Radial distribution functions 0 FIG. 3. View of the surface layers up to adepth of 5A. Cavities ofdifferent size can be observed. be seen that its main axis is parallel to the (110)crystallographic direction and smaller pores are located along (110}and (110}directions. Figure 4shows aslab of the whole system 3Athick, and parallel to the z-y plane. In the region where particles are allowed to move, there are pores of different size and shape, however, their location and orientation is along planes parallel or perpendicular to the (110)direction. This same feature can be observed at the surface where the uppermost layer has ajigsaw shape also oriented along these planes. In summary, in our simulations, starting from asurface exposing the (100) surface plane there is areconstruction process at the end of which the surfaces exposed both at the gas solid interface and within pore systems are parallel or perpendicular to the (110)plane. This result agrees with experimental work in which the (110) plane has been found to be the more likely exposed. 'One additional question concerns the conditions under which areconstruction of the type described here can occur in real sysThe radial distribution functions, g(r}, of the moving particles have been calculated in order to compare them with the ones calculated from experimental x-raydiffraction data (XRD). Figure 5shows the partial g(r) for the pairs Al-Al, Al-O, and O-O, along with the positions of maxima obtained from the experimental data up to 6A. The shift ofpositions of maxima of the calculated g(r) with respect to the XRD results is at most 0.1A; therefore, using the partial radial distribution functions it is possible to identify their contributions to the total g(r) shown in Fig. 6. The structure of radial distribution functions indicates the existence of short-range order in an amorphouslike phase, long-range order being hardly observable. This is due to the porous structure ofthe system in which although there is certain order, and atendency to expose aplane similar to the (110) plane of the ideal crystal, the long-range order is not present. The xray-diffraction pattern of y-A1203 presents broad and poorly resolved peaks indicating an amorphous or avery Al — Al AI — 0 3 2 0— 0 I R(A) I 10 FIG. 4. Slab ofthe whole system 3Athick, parallel to the z-y plane. FIG. 5. Partial g(r) for the pairs Al-Al, Al-O, and O-O along with the positions ofmaxima obtained from experimental data. 2564 ALVAREZ, LEON, SANZ, CAPITAN, AND ODRIOZOLA 80 0 O150 iI 60 100 i40 50I(Ix /1 Ijt IJ oj R(A) 6810 10 FIG. 8. Total g(r) of the ideal spinel (A) and ofthe resulting solid after eliminating boundary conditions in the zdirection (B). FIG. 6. Contributions of partial radial distribution functions to the total g(r). poorly crystallized material. Figure 7shows experimental XRD patterns of three samples of y-A1203 of different origin' and, for the sake of comparison, the XRD pattern of asample of a-A1203 obtained by calcining at 1473 Ky-A1203 samples in air. As it is well known the intensity ratio and resolution of peaks in the XRD spectra is a function of the preparation method. The poorly resolved diffractograms correspond to samples with higher specific surface areas [(a) 70 mg'; (b) 167 m2g ', (c) 127 mg'; (d) 5mg'], and consequently with larger pore volume. On the basis of the above discussion the structure of catalytically active y-A1203 may be understood as an amorphous material having short-range order in some way resembling the disordered spinel structure previously reported. This idea is illustrated in Fig. 8where the total g(r) of the ideal spinel structure, curve A, and that of the resulting solid after eliminating the boundary condition in the zdirection, curve 8,are shown. Table Ishows the percentages of the observed coordination polyhedra for aluminum and oxygen in the ideal spinel and the simulated surface. The relative proportion of tetrahedral polyhedra of aluminum at the surface increases with respect to the ideal spinel structure at the expense of the octahedrally coordinated ones. First of all, the truncation of the bulk produces adecrease of coordiTABLE I. Observed percentages of coordination polyhedra for aluminum and oxygen in the ideal spinel and the simulated surface. ;&0 (l0 Al Coordination number 3 4 5 6 Ideal spinel 0.0 28.1 0.2 71.7 Surface structure 1.0 56.4 32.4 10.1 20 (deg) FIG. 7. Experimental XRD patterns of yand a-A1,0, obtained under different experimental conditions. 02 3 4 5 0.6 38.1 60.2 1.4 16.7 68.0 13.9 1.4 50 SURFACE STRUCTURE OF CUBIC ALUMINUM OXIDE 2565 nation numbers in general; therefore, threeand fivecoordinated aluminum atoms should be expected. However, three-coordinated aluminum ions are hardly found confirming their absence in experimental observations, and the number of five-coordinated cations is lower than expected if they were only due to the truncation of the bulk. This is due to the above-mentioned reconstruction process in which oxygens tend to emerge to the surface in order to saturate undercoordinated aluminum ions. This reconstruction process is also responsible for the formation of large cavities within the y-A1203 structure. The relative proportion of octahedral plus five-coordinated to tetrahedral coordinated aluminum ions is close to that found in the ideal (110) surface plane because the system tends to expose similar planes both at the surface and within pore systems. IV. CONCLUSIONS Molecular-dynamics simulations of the surface of yAlz03 allows us to draw an alternative interpretation of its structure and to explain the roughness and porosity experimentally found for this material. Reconfiguration of surface ions gives way to pore formation with loss of long-range order. Both at the outer surface and at the pore surfaces, the exposed structures resemble the ideal (110) plane, as has been suggested from experimental work. Considering the surface as an amorphouslike phase, micropore volume and x-ray-difFraction data can be well reproduced. Further work is being carried out in order to describe the reconstruction and cavity formation processes in aquantitative way. ACKNOWLEDGMENTS Financial support for this project was obtained from Comision Interministerial de Ciencia yTecnologfa (PB92-0665 and PB92-0662) of the Spanish Government and partially supported by Cray Research Inc. under supercomputing Grant No. DGAPASC-000192, UNAM (Mexico). L.J.A. thanks the Junta de Andalucia for partially supporting astay in the Universidad de Sevilla (AGCI-6006 and AGCI-1039). 'Author to whom correspondence should be addressed. J.B.Peri, J.Phys. Chem. 69, 220 (1965). ~B.C. Lippens and J.J.Steggerda, in Physical and Chemical Aspects of Adsorbents and Catalysts, edited by B. G. Linsen (Academic, New York, 1970), p. 171;A. Zecchina, S. Coluccia, and C.Morterra, Appl. Spectrosc. Rev. 21, 259 (1985). H. Knozinger and P. Ratnasamy, Catal. Rev. Sci. Eng. 17, 31 (1978). 4F. R. Chen, J. G. Davis, and J. J. Fripiat, J. Catal. 133, 263 (1992). 5L. J. Alvarez, J. Fernandez Sanz, M. J. Capitan, and J. A. Odriozola, Chem. Phys. Lett. 192,463 (1992). 6L. J. Alvarez, J. Fernandez Sanz, M. J. Capitan, M. A. Centeno, and J. A. Odriozola, J.Chem. Soc. Faraday. Trans. 89, 3623 (1993). G. A. Somorjai, Chemistry in Tue Dimensions (Cornell University Press, Ithaca, 1981). H. H. Kung, Stud. Surf. Sci.Catal. 45, 1(1989);P. A. Thiel and T. E. Madey, Surf. Sci. Rep. 7, 211 (1987); S. Varma, G. S. Chottiner, and M. Arbab, J. Vac. Sci. Technol. A10, 2857 (1992); C. C. Chang, J. Appl. Phys. 39, 5570 (1968); T. M. French and G.A. Somorjai, J.Phys. Chem. 74, 2489 (1970). D.J.Adams and I.R.McDonald, Physica B79, 159 (1979). M. Dupuis, A. Farazdel, S.P. Kama, and S.A. Maluendes, in Modern Techniques in Computational Chemistry, edited by E. Clementi (Escom, Leiden, The Netherlands, 1990),p. 277; M. Dupuis, S.Chin, and A. Marquez, in Relativistic and Electron Correlation sects in Molectdes and Solids, edited by G. L. Malli (Plenum, New York, 1994). R. D. Shannon and C. T.Prewitt, Acta Crystallogr. Sec. B25, 925 (1969). L.J.Alvarez, L.E.Leon, and H. Mu5oz (unpublished). sIMULA is amolecular dynamics and visualization software developed by L. J.Alvarez, Universidad Nacional Autonoma de Mexico, 1993. S.Soled, J.Catal. 81,252 (1983). J. P. Beaufils and Y. Barbaux, J.Chim. Phys. 78, 347 (1981); Y.Chen and L.Zhang, Catal. Lett. 12, 51 (1992). y-A1203 samples are obtained from Degussa (Aluminumoxid C) labeled Ain Fig. 7, and home prepared by precipitating with urea at 340 Kan aqueous solution of aluminum nitrate. In this later case the precipitated was calcined at 1073, 1273, or 1473 Kin air for 8hprior to the obtainment of diffractograms 8, C, and D, respectively, in Fig. 6.