Computer modelling of V20,:surface structures, crystal morphology and ethene sorption Dean C. Sayle,"" David H. Gay," Andrew L. Rohl," C. Richard A. Catlow," John H. Harding,b Marc A. Perrid and Patrice Nortierc a Royal Institute of Great Britain, 21 Albemarle Street, London, UK WlX 4BS AEA Technology, B424.4 HarweEl Laboratory, Didcot, Oxfordshire, UK OX11 ORA Centre de Recherche de RhGne-Poulenc, 52 rue de la Haie Coq, 93308 Aubervilliers Cedex, France We report simulations of the surface structures and crystal morphologies of V205. The V205(OOl) surface is calculated to be the most stable, dominating the morphology. The low-energy adsorption sites of ethene on the V205(OOl), (200)and (301) surfaces are identified. The ethene molecule is observed to approach closer to the exposed ions on the (301) surface compared with those on the (001) or (200) surfaces, which is reflected in the binding energy of the ethene on the V,O5(301) surface being 44 kJ mol-' greater than on the V205(OOl) surface and 54 kJ rno1-l greater than on the V205(200) surface.Vanadium oxides have received considerable attention owing to their important role in partial oxidative catalysis.'-3 Substantial efforts are presently being made to understand these material^,^ especially their surface structure and proper- ties.' However, because of the inherent difficulty in characteris- ing mixed oxide/oxide interfaces, little is known of the detailed atomic structure and active sites on these oxides and of the reaction pathways of adsorbed molecules.In this study we employ atomistic simulation methods which are excellent tools with which to study such problems. Indeed, the development of the surface codes MIDAS6 and, more recently, MARVINY7 has led to a clearer understanding of the atomic processes that occur at surfaces and interfaces of inorganic The focus of this study is the investigation of the energetics and relaxed structures of the low index faces of V205. We predict the crystal morphology of the material based on the energies of the various crystallographic faces; our predictions can then be compared directly with experimental observations of the crystal habit. Finally we examine the low-energy sorption sites of ethene on the dominant crystallographic faces.The characterisation of the relaxed surface structures and sorption configurations of small molecules on the unsupported V205 surfaces is the first step in gaining a greater understanding of vanadia catalysts. In view of the activity and selectivity of supported V, O5monolayers as partial oxidative catalysts, we will extend our studies in future publications to consider the effect of supporting the material on Ti02 with respect to the V205 thin film surface structure. V,O, Crystal Structure The structure of V205 was reported by Bystrom et a1.l' with a refinement to the structure proposed in 1986 by Enjalbert and Galy.I2 V20, is orthorhombic with space gr9up symmetry Pmmn with u = 11.512, b = 3.564 and c = 4.368 A. V205 has a layered structure, built up from VO, square pyramids sharing edges and corners, with V205 sheets held together via weak vanadium-oxygen interactions (see Fig.1). The structure of the crystal has been successfully modelled by computational techniques as described in greater detail below. Simulation Methods and Potential Models The MARVIN program is a new code for the investigation of surfaces and interface^.^ The methodology, closely related to the earlier MIDAS code,6 considers the crystal as a stack of planes periodic in two dimensions with ions interacting via specific interatomic potentials. This stack is divided into two regions: I, where the ions are allowed to relax explicitly; and 11, where the ions are held fixed relative to each other.Region I1 is included to ensure the potential of the ions at the bottom of region I is correctly represented. The top of region I is the free surface unless two blocks are placed together, enabling the cohesive energy of the perfect crystal, grain boundary or hetero-interface to be calculated. The surfaces are created by MARVIN by cleaving the crystal at a particular crystallographic plane, which results in the formation of one of the three types of s~rfaces,'~ designated Fig. 1 (a)Diagrammatic representation of the V205 crystal structure illustrating the corner-and edge-sharing V05 square pyramids. (b) Enlargement of three square pyramids indicating the individual vanadium and oxygen species: vanadium ions are represented by filled circles, vanadyl oxygens are hatched and bridging oxygens are open circles.J. Muter. Chem., 1996,6(4), 653-660 653 types I, I1 and I11 All the surfaces in this work are created so as to fall into either type I or type I1 surface categones, where any dipole normal to the surface is quenched uzu an appropnate cleavage of the crystal surface or appropriate arrangement of surface ion species In several cases this is achieved by manipul- ating the charge distribution to form a surface where the dipole has been quenched More details of this procedure are given in the Appendix Potential Models The interionic potentials used in the present study are based on the Born model of the ionic solid, which includes a long- range Coulombic interaction and a short-range term to model the Pauli repulsions and van der Waals attractions between the ions The shell model14 describes the electronic polaris- ability of the component ions The potential parameters for V205 were taken from Dietrich et ul ,I5 whose model was able to reproduce accurately the crystal structure of the material For the ethene molecule, we use the potential parameters from the cvff frc forcefield of BIOSYM Technologies16 and potentials between ethene and V205 are from Vetrivel et all7 The potential parameters are listed in Tables 1-6 Table 1 Short-range potential energy terms between the component ions of V,O, [the analytical function is of the oform V(r)= A exp(-r/p) -0-6 with a short-range cut-off of 10 0 A], O-vanadyl oxygen, O(1)=bndging oxygen species A/eV PIA C/ev A6 v-0 2549 73 0 34115 00 v-O( 1) 5312 99 0 26797 00 0-0 22764 3 0 149 23 0 0-O( 1) 22764 3 0 149 23 0 O(1 )-O( 1 ) 22764 3 0 149 23 0 Table 2 Potential parameters of the analytical form V(r)= ~,{1-e~p[-l,(r-r,,)]}~ species DJeV AlA ro/A cut-off/A V-0 100 230217 1584 50 C-H 39201 20 1 09 14 C-C 7103 20 133 16 Table 3 Potential parameters describing the short-range potential- energy terms between ethene and V20, [the analytical functionJs of the form V(r)=Ar l2 -Br with a short-range cut-off of 100 A] species AIeV B/eV A6 O-H 1557 522 5 574 O(1)-H 1557 522 5 5740-c 15118 161 22 579 O(1)-c 15118 161 22 579 Table 4 Three-body potential parameters of the analytical form V(r)= 0 5K[O -O0l2 species KIeV Bo/degrees cut-off/A C-C-H 14657 121 2 16 13 29 Table 5 Four-body potential parameters of the analytical form V(r)= K[ 1+ Scos(Phase -Phi)] species KIeV S phase cut-off/A H-C-C-H 0 7068 -1 2 13 16 13 Table 6 Ionic charges and shell model parameters of the component ions [the analytical function is of the form V(r)= 0 5Kr2+ Klv4] species charge/e K/eVA K,/eV A ~~ 0 shell -2 717 54 952 00 O(1) shell V C -2 717 50 -02 54 952 rigid ion rigid ion 00 H 01 rigid ion Crystal Morphology For a particular face of a crystal to be important in catalysis, it must be not only active, but also exposed The degree of exposure of the various crystallographic faces in V205 can be found by calculating the crystal morphology Note in this context that surfaces which are unstable with respect to alternative crystallographic planes, although possibly active, will not be observed in the crystal morphology However, these surfaces may be selectively exposed when supported, e g TiO,, or when stabilised by surface defects, as will be investigated in future work Crystal morphologies can be calculated using either attach- ment energies or surface energies, leading to growth and equilibrium morphology predictions, respectively ’The growth morphology will be determined by the relative deposition rates of particles on the various faces of the crystal, with the slowest growing faces of greatest morphological importance These growth rates have commonly been assumed to be dependent on the binding energy of a growth slice to a particular face which crudely approximates the particle binding energy The MARVIN code allows the calculation of such attachment energies, although we should emphasise the simplicity of the approach which ignores factors such as nucleation sites (steps and edge effects), temperature, supersaturation, solvent and impurity concentration The alternative approach is to assume thermodynamic control, I e to calculate the surface energies of the various faces with the most stable surfaces dominating the morphology In order to minimise the surface energy of the crystal Indeed, Gibbs first proposed18 that the equilibrium form of a crystal should, for a given volume, possess a minimum surface free energy, z e the crystal morphology will correspond to the case where y =Z,y,A,is a minimum for constant volume y, and A,are the surface free energy (which we will approximate as the surface energy) and the surface area of the zth crystallo- graphic face Equilibrium crystal morphologies, based on surface energies, also have limitations, as crystal growth is often not an equilib- rium process In this respect we therefore present crystal morphologies based on both methods of calculation Using either approach, the shape of a crystal may be predicted by ensuring that a vector h, normal to the face is proportional to either the surface energy or the attachment energies of the crystallographic face l9 Both methods are discussed in more detail by Gay and Rohl ’ Results V, O5surfaces The first problem in calculating the growth and equilibrium crystal morphologies of V2 0, concerns the crystallographic surfaces which must be considered In practice, the attachment energies of high-index faces are very highly negative and therefore such surfaces will not feature in the crystal mor- phology Unfortunately, this is not the case for equilibrium morphologies For example, if we consider MgO(20 10), we find that the surface energy is very similar to that of MgO( 100) and must be given equal weighting in the morphological prediction However, (20 1 0) is better described as a step [with 654 J Muter Chem, 1996, 6(4),653-660 Miller index (1 lo)] on the (100) surface and therefore most of the surface will be (100) with a small proportion of the (110) surface.We therefore introduce an approximation which assumes that the surface configurations of high-index surfaces are represented, in part, by surfaces of lower Miller index. For many simple materials, the important low-Miller-index crystallographic surfaces, which are expected to feature in the morphology, may be established by inspection. However, for V205the situation is more complicated and therefore the low- index surfaces are obtained by using procedures available in the BIOSYM INSIGHT11 program,16 which reveal all the low- index crystallographic surfaces with interplanar spacing greater than a pre-determined value. It also eliminates the possibility of minimising symmetry-equivalent surfaces which, for complex materials, may not be obvious from inspection of the Miller indices.In this study, the !urfaces of V205 with interplanar spacings greater than 1.7 A were considered. Surface Miller indices quoted in this work represent the irreducible growth slice of the surface, i.e. the smallest repeat unit to facilitate the complete addition of further planes. For example, the V205(100) surface is designated the V205(200) surface, as a (100) plane is reducible to two (200) planes. Care must be exercised when comparing the quoted Miller indices in this work with other works where Bachmann's notation2' may have been used (in which b and c are permuted). For any particular surface, there may be several termination planes resulting in bulk dipole free surfaces, in which case only the termination which results in the surface with the lowest energy is considered.Of particular importance to the V205 system is the vanadyl (V=O) species which is considered to play a significant role in the catalytic behaviour of vanadium oxide compound^.^ Moreover, our results suggest that cleavage of these strong bonds results in severe destabilisation of the surface. Vanadyl bonds are not therefore cleaved on any of the surfaces considered. Table 7 gives the calculated surface energies, attachment energies and interplanar spacings for all the surfaces studied. The relaxed structures of selected low-energy surfaces are displayed diagrammatically in Plate 1 (u)-(h).A common structural feature of the surface is the tendency for the vanadyl oxygens to relax outwards which enables the V=O species to be closer to the surface normal (Table8). This feature may enhance the accessibility of the vanadyl oxygens to any reacting molecule. Furthermore, such relaxations will change the dis- tances between neighbouring vanadyl oxygens, which may be of importance to the activity and specificity of the catalyst. We note that this detailed information on surface geometry will be of great utility in future quantum mechanical studies of reaction mechanisms. The equilibrium crystal morphology, based on calculated surface energies, is given in Fig. 2, while Fig. 3 shows the growth crystal morphology which is based on the calculated attachment energies.It is clear that the (001) face dominates the crystal morphology. The observed V205crystal habit3 is a rectangular prism which exposes predominantly the (001) surface. Oshio et ~1.~'obtained atomic resolution images of a V, O,( 001) single crystal using scanning tunnelling microscopy (STM). The STM image suggests that cleavage of the V205(OOl) surface is along the weak van der Waals V-0 bonds (which hold the layers together) and therefore the surface is terminated by the vanadyl oxygen species. The surface structure is therefore identical to our proposed model of the surface structure [see Plate 1(a)]. These gratifying results sug- gest that we can have confidence in our potential model and simulation methodologies.Furthermore, the surface structure and morphology can only Table 7 Calculated surface energies (E,) and attachment energes (EA)of low-index faces of V,O, with interplanar spacing greater than 1.7 A miller plane E,, unrelaxed/J m-' E,, relaxed/J m-' EA, relaxed/J mol-' x lo6 interplanar spacing/A ~ 3 2.2 -1.3 5.69 3 2.3 -1.4 4.37 1 0.7 -0.1 33 2 26 3.3 1.2 6.5 - -6.1 -0.9 -3.9 -16.4 4.08 3.46 3 1.6 -2.0 3.40 8 1.7 -4.8 2.86 36 2.0 -6.7 26 3 5.4 1.6 -6.4 -2.6 2.76 6 11 2.2 3.2 -1.3 -6.9 2.68 8 7 5 4 2.8 2.1 2.1 1.8 -8.3 -5.8 -1.2 -6.7 2.60 2.48 1.79 Table 8 Surface vanadyl oxygen concentrations and angles (after relaxation) of the low-index faces of V,05 (an angle of 90" represents a V=O bond normal to the surface plane) miller plane V=O surface concentration/pmol m-, V= 0 surfaceldegrees V=O bulkldegrees A8"ldegrees 4.1 67 90 +23 8.2 23 0 -23 3.8 10,20 21 +lo, -1 6.5 6.4 5.4 2.6 15, 44 65, 74 20, 33 56 37 90 49 51 +22, -7 +25, +16 +28, +15 -5 2.5 43 135 +92 9.3 6.7 17, 50, 64, 72 64 90 +26 ~______ ~ "A0=angle through which the surface V=O species have relaxed.J. Muter. Chem., 1996, 6(4),653-660 655 Plate 1 Diagrammatic representations of the various surfaces of V20, after relaxation (u)V20,(OOl) surface, (b)V20,( 200) surface, (c) V20,(020) surface, (d)V,O,( 110) surface, (e) V20,( 101) surface, (f)V,O,(Oll) surface, (8)V205( 111) surface, (h)v20,(301) surface be reproduced after the appropriate ‘manipulation’ (detailed Sorption of ethene on the V, 0, surfaces in the Appendix) of surface ions A simple cleavage of the surface will not give the experimentally observed surface A study of the adsorption of molecules on the relaxed V,O, structure and crystal morphology In this respect, we have surfaces is a first step in understanding the oxidative reaction direct evidence (we believe for the first time) that such a of vanadium oxide catalysts For example, the location of manipulation of the surface structure, to remove the surface such adsorption sites can, as noted, be used as starting dipole, has a physical basis and is not purely a requirement of configurations for quantum mechanical studies of the mechan- our calculations isms in partial oxidative reactions Location of low-energy 656 J Muter Chem, 1996, 6(4), 653-660 Fig.2 Predicted crystal morphology of V205 using the relaxed surface energies (equilibrium morphology). Three projections of the crystal habit are shown. Fig. 3 Predicted crystal morphology of V205 using the calculated attachment energies (growth morphology). Three projections of the crystal habit are shown. sorption sites presents a formidable undertaking for quantum mechanical techniques. The atomistic simulation approach is much less computationally expensive, and is therefore the natural starting point for any study of sorption and reaction. Considering the limitations of this present model (primarily resulting from the formal charges assigned to the individual atoms of V205) the resulting ‘absolute’ adsorption energies calculated in this work must be treated with some caution.However, we are primarily interested in the relative sorption energies of ethene on various V205 surfaces. In this instance the limitations of the potential model will largely cancel when a comparison between different surfaces is made, and the sorption behaviour of the ethene molecule will reliably reflect the essential differences between the potential field environment of each individual V205 surface. To calculate the adsorption sites of ethene on all the surfaces of the V,O, surface would be prohibitive. We therefore have to be selective in the surfaces chosen.Our main criterion must be the surface exposure of the various crystallographic planes and therefore, as the (001) and (200) surfaces feature in both the growth and equilibrium morphologies these surfaces are considered. We also examined the (301) surface, which has a high concentration of V=O species which have been linked to the catalytic activity1*’ of the material. Furthermore, these V=O species appear to be easily accessible to reactant mol- ecules [Plate l (h)]. To establish a low-energy adsorption site, the ethene mol- ecule was placed at various positions on each of the three surfaces and the system relaxed until zero force acted on each of the component ions. Several starting configurations were considered to ensure that the lowest adsorption-energy mini- mum has been located, from which we calculated the adsorp- tion energy (defined as the energy difference between the ethene molecule docked on the V205 surface compared with the perfect V205 surface and the ethene molecule at infinity).The adsorption energies, together with the ethene-V, O5 bond distances for the energy minima, are given in Table 9. Plates 2, 3 and 4 show the relaxed configurations of the ethene adsorbed on the surfaces of the (Ool), (301) and (200) surfaces, respectively. Table9 shows that both the adsorption energies and ethene-V,O, bond distances for ethene adsorbed on the (001) and (200) surfaces are very similar, suggesting that the ethene molecule is in a very similar environment or ‘sorption site’, which is perhaps surprising when one considers the structural differences between the two surfaces.Table9 also shows that the ethene molecule, when adsorbed on the (301) surface, is significantly closer to the surface than for configurations OF the (001) or (200) surfaces: in particu!ar, hydrogen is 0.13 A closer to the vanadyl oxygen and 0.70 A closer to the bridging oxygen for ethene on the (301) surface compared with the nearest distances for ethene on the (001) surface. That the molecule is able to approach closer to the V205 surface is reflected in the adsorption energies which are Plate 2 Ball-and-stick representation of ethene on the V205(OOl ) surface. Three views are displayed; the top left is a view looking down on the V205 surface.Only one ‘layer’ of the V205 surface is included for clarity. Vanadium atoms are yellow, vanadyl oxygens are red, oxygens are orange, carbons are green and hydrogens are white. Table 9 Adsorption energies and hydrogen-oxygen bond distances of ethene on the (200), (001) and (301) surfaces of V205 H-0 bond lengths/A V, 0,(200) surface V205(OOl) surface V2O5(3O1) surface v=o v-0-v v=o v-0-v v=o v-0-v ~ ~ ~ ~ ~~ ~ ~~ ~ ~ ~ ~ ~ ~ ~ ~ 2.65 3.40 2.65 3.40 2.68 3.25 2.68 3.25 2.58 3.14, 3.50 2.60 3.30, 3.34 2.60 3.30, 3.34 2.57 3.15, 3.50 2.44 2.50 2.44 2.50 -2.44 -2.44 Adsorption energy/kJ mol-’ 23 33 77 J. Muter. Chem., 1996, 6(4), 653-660 657 Plate 3 (a)Ball-and-stick representation of ethene on the V205(301) surface.(b) Space-filling representation of ethene on the V20,( 301) surface. (Colours as for Plate 2.) Plate4 Ball-and-stick representation of ethene on the V205(200) surface (colours as for Plate 2) ca. 50 kJ mol-I higher on the (301) compared with the (001) or (200)surfaces. Conclusions We have shown the crystal morphology of V,O, to be com- pletely dominated by the (001) surface. The predicted crystal morphology and surface structure are consistent with exper- imental observations. We have also found that the adsorption behaviour of ethene on the (001) and (200)surfaces are similar with respect to hydrogen-oxygen bond lengths and adsorption energy. However, for ethene adsorbed on the (301) surface substantial differences exist; the ethene molecule is observed to approach closer to the (301) surface compared with the (001)or (200) surface ions, which is reflected in higher binding energies to the former surface.Subsequent studies will explore the supported V, 0,-TiO, catalyst systems and will apply quantum mechanical techniques 658 J. Muter. Chem., 1996, 6(4), 653-660 to investigate the reaction mechanisms of the sorbed configur- ations identified in this study. We thank Rh6ne-Poulenc for financing this project and BIOSYM Technology Inc. for the provision of the INSIGHT11 software used in the molecular graphics representations of the surfaces and crystal morphologies. We would also like to thank Dr. D. J. Ilett for help in constructing the datasets used in this study.Appendix Here we describe the calculation of surface energies for dipolar surfaces. It is well known that the Ewald summation is only conditionally convergent. In standard applications the bound- ary conditions are automatically chosen so that this is not apparent. However, in the two-dimensional summations required for surfaces, the problem is unavoidable. A notorious example occurs when the repeating unit of a stack of planes has a finite dipole moment perpendicular to the surface. In such a case the Ewald summation diverges and the energy of the surface is infinite. This was first pointed out by Kummer and Yao22 for the particular case of the (111) surface of the alkali-metal halides.More extended discussions are given by Tasker13 and Berta~t.,~ The problem arises because, since each addition of a repeat- ing unit to the stack adds to the dipole moment, the moment of the stack increases without limit. The point may be illus- trated by approximating the crystal planes as charged sheets. The electrostatic potential of a plane with charge density 0= QIA at a distance Z is given by VZ) =2nQz/A. If the spacing between the planes is a, the potential of a pair of planes of opposite charge is V(z)= 2naQ/A; this is independent of z. It is still possible to calculate a surface energy for a polar stack provided that the dipole moment is quenched. We take first the simple case considered by Kummer and Yao.,, We consider a stack of planes where the interplanar spacings are all equal and the charge densities are fa on alternative planes.We construct a new stack where the top and bottom planes have charge density a/2 (i.e. we remove half the charge density from the top and put it on the bottom). This gives us a stack as shown in Fig. 4. This new construction has a zero dipole moment. It is, however, necessary to add an extra term to the energy. This is an interaction between the (now charged) outer regions: nQ2uE2, = -2A where is the distance between the innermost edges of the outer blocks. This is the case for the V205(OOl) surface; in order to quench the dipole, half the vanadyl oxygen species are removed from the V205(OOl) surface to the bottom of region 11.The weak van der Waals interactions between the vanadium ions and vanadyl oxygen species that hold the V205 layers together are broken, rather than the strong vanadyl (V=O) bonds. The interactions between ions on the outer regions (region 11) are automatically included in the MARVIN code. The result is the same if the two blocks have different Q/2 -Q Q Q -Q QQ Fig. 4 Schematic representation of the alkali-metal halide (111)surface where the dipole has been quenched by removing half the charge density from the top and placing it on the bottom. 0’ -Q Q -Q Q Q -Q Q -Q Q“ Fig. 5 Schematic representation of the wurtzite surface where the dipole has been quenched via charge manipulation. -V--400-+4 -V-4.0--4 -V--4001 4 -v--4-01-4 -V-400-+4 -V-4-0--4 -V-4001 +44-V--0--2 --v-400-+1 -V--4 -V----ooo-+4-0--v-49-4 -V--000-+4 -V--09-4 -V-400-+4 -V--2 Fig.6 Representation of the V,O,(OOl) surface before (a)and after (b) the dipole has been quenched. To the right of each figure a schematic of the ionic species and net charge of each plane is shown. The filled small circles represent vanadium and the hollow circles represent oxygen. The hatched circles are the vanadyl species which are removed from the top of the V,O,(OOl) surface and replaced at the bottom to quench the surface dipole. interplanar spacings [e.g. if we considered the ( 111) surface of NaCl joined to the (111) surface of KCl]. Matters are more complicated if all the interplanar spacings in a block are not the same.This is a little more unusual than one might think; many common crystal structures correspond to the simple case. An example that does not is the wurtzite structure. This case is shown in Fig. 5. In this case, as shown by Duffy and Ta~ker,~~ the charges on the two ends of the stack are not equal. They are given by Q’ = a2Q/(al + a2) and Q” = Q -Q’ = a, Q/(al + a2).The interaction between the outer blocks is now: 2nQ2alu2E-22 -A(u, + LI~)~ (2) The most complex case one is likely to encounter in practice is the case where two different materials, each with two interplanar spacings, are joined together. Here the charges on each end are: Q’ = rQ and Q” = Q -Q’, where and the interaction energy between the blocks is It is obviously possible to think of more complex cases, the most obvious being where charges on the planes for each block are different.These present no new questions of principle, merely increasing algebraic complexity. We now illustrate this methodology by considering the V20,(001) surface, which is shown graphically in Fig. 6, which also shows a schematic of the ionic species and net charge of each plane. This surface has equal interplanar spacings and charge density +o on alternate planes (if we consider the vanadium and bridging oxygens to comprise one plane and the vanadyl oxygens, the other). Clearly, the surface repeat unit will add to the dipole moment, resulting in an undefined surface energy.This surface is therefore classified as type I11 and must be modified, by quenching the dipole via charge manipulation, to calculate the surface energy. In practice this is achieved by removing half the vanadyl oxygens from the top and replacing them at the bottom (Fig. 6). The new charge density at the top and bottom is now -o/2 and the system has no dipole normal to the surface. As the charge density was modified by the manipulation of ionic species, the extra term in the energy [eqn. (1)] is implicitly calculated in the MARVIN code. This type of procedure is repeated for all the type I11 surfaces considered in this work to quench the dipole. References 1 G. Centi, S. Perathoner and F. Trifiro, Research on Chemical Intermediates, 1991, 15,49.2 A. Vejux and P. Courtine, J. Solid State Chem., 1978,23,93. 3 A. Vejux and P. Courtine, J. Solid State Chem., 1986,63, 179. 4 G. Centi, F. Trifiro, J. R. Ebner and V. M. Franchetti, Chern. Reu., 1988,88, 55. 5 A. Satsuma, A. Furuta, T. Hattori and Y. Murakami, J. Phys. Chem., 1991,95,3248. 6 P. W. Tasker, Harwell Report, AERE-R9130,1978. 7 D. H. Gay and A. L. Rohl, J. Chem. Soc., Faraday Trans, 1995, 91,925. 8 D. M. Duffy, J. Phys. C: Solid State Phys., 1986,17,4393. 9 J. H. Harding, Rep. Prog. Phys., 1990,53, 1403. 10 D. C. Sayle, T. X. T. Sayle, S. C. Parker, C. R. A. Catlow and J. H. Harding, Phys. Rev., 1994,50, 14498. 11 A. Bystrom, K-A. Wilhelmi and 0. Brotzen, Acta Chem. Scand., 1950,4, 1119. 12 R. Enjalbert and J. Galy, Acta Crystallogr., Sect. C, 1986,42, 1467. 13 P. W. Tasker, J. Phys. C: Solid State Phys., 1979,12,4977. 14 B. G. Dick and A. W. Overhauser, Phys. Rev., 1958,112,90. J. Mater. Chem., 1996, 6(4), 653-660 659 15 A Dietrich, C R A Catlow and B Maigret, Molecular Simulation, 1993,11,251 21 1993,7,33 T Oshio, Y Sakai, T Moriya and S Ehara, Scanning Microscopy, 16 17 Biosym/MSI Technologes, 9685 Scranton Road, San Diego, USA R Vetrivel, C R A Catlow and E A Colbourn, J Chem SOC, 23 22 F Bertaut, Compt Rend, 1958,246,3447 J T Kummer and Y Y Yao, Can J Chem , 1967,45421 18 Faraday Trans 2,1989,85,497 J W Gibbs, Collected Works, Longman, New York, 1928 24 D M Duffy and P W Tasker, J Appl Phys, 1984,56,971 19 20 G Wulff, 2,Krzstallogr ,1901,34,449 H G Bachmann, F R Ahmed and W H Barnes, Z Kristallogr , Paper 5/04556H Received 11 th July, 1995 1961,115,110 660 J Muter Chern, 1996, 6(4), 653-660