J. MATER. CHEM., 1991, 1(4), 563-568 27AINuclear Magnetic Resonance Spectroscopy Investigation of Thermal Transformation Sequences of Alumina Hydrates Part 1.-Gibbsite, y-AI(OH), Robert C. T. Slade,*" Jennifer C. Southernb and Ian M. Thompsonb a Department of Chemistry, University of Exeter, Exeter EX4 4QD, UK Alcan Chemicals Ltd, Chalfont Park, Gerrards Cross, Buckinghamshire SL9 OQB, UK High-resolution 27AI magic-angle spinning nuclear magnetic resonance spectroscopy (MAS NMR) (78.15 MHz) spectra are reported for commercial and natural gibbsites and materials occurring in the thermal transformation sequences of gibbsite (boehmite; x, a and p-aluminas; calcines produced at 400, 700, 900 and 1000 "C). Separate peaks assignable to aluminiums in near-regular six- and four-co-ordination in oxygen are observed for the materials in which use of such sites is believed to occur.For p-alumina a third peak (at 27 ppm) is seen due to the presence of Al in five-co-ordination. Ic-Alumina may retain approximately hexagonal-close-packed oxygens (with a defect structure leading to a grossly asymmetric contributory line from four-co-ordinate Al) or may contain some five-co-ordinate Al (AIO,) in addition to AIO, and AIO, environments (and hence have oxygen packing deviating from h.c.p.). Distortions/disorders (when present) result in broader underlying signals and manifest themselves instrumentally for the most disordered systems as wings extending to either side of the central spectrum . Keywords: Gib bsite ; Ca lcina tion ; Deh ydroxyla tion; Alumina hydrate ; Nuclear magnetic resonance spectroscopy Alumina (Al,O,) and its 'hydrates' [the hydroxide y-Al(OH), and the oxy-hydroxide y-AlOOH] are commercially import- ant materials marketed both as dried hydrate and as calcined aluminas produced by heating.Gibbsite [y-Al(OH),, some- times known as hydrargillite] is the product of the Bayer process, and is itself used as a flame retardant/smoke sup- pressant in polymers, as a toothpaste filler, as a paper additive, in the manufacture of synthetic marble, in stabilisation of titania and as feedstock for the manufacture of other alu- minium chemicals including ceramic powders. Dehydroxyl- ation (during calcination) of gibbsite leads first to one or more transition aluminas with partially disordered structures (all based on close-packed oxygen layers with varying interstitial aluminium configurations) and varying residual hydrogen contents.These have their own particular applications, e.g. as high-surface-area sorbents or glass additives. As the calci- nation temperature increases, the structures become more ordered until the final transformation to the extremely stable corundum (a-Al,O,) form, which is itself used in abrasives, refractories, polishing, aluminous porcelain, technical and engineering ceramics and catalyst supports. The crystal structures of gibbsite [y-Al(OH),], boehmite (y-A100H) and corundum (a-Al,O,) are well known. Gibbsite is monoclinic (space group P2Jb :a =8.64 A, b= 5.07 A, c= 9.72 A, fl= 85"26').' The structure contains double layers of have examined the dehydration (strictly dehydroxylation) sequences of aluminium hydroxides during calcination e.g.ref. 4-13. There is still disagreement on the sequence of tran- sition alumina phases, the mechanism of dehydration and the structures of the transitional phases. It is generally accepted14 that in vucuogibbsite transforms via p-alumina (100-400 "C), q-alumina (270-500 "C), and &alumina (870-1 150 "C) In air it follows two paths: (i) via boehmite (60-300 "C), y-alumina (500-850 "C), &alumina (850-1050 "C) and &alumina (1050-1150 "C); (ii) via x-alumina (300-500 "C) and Ic-alumina (800-1150 "C). The relative proportions following these two paths depend on a number of factors (including gibbsite particle size, moisture, alkalinity, pressure, bed depth and heating rate).The various transformation sequences for gibbsite are illustrated in Fig. 1. Characteristic structures of transition aluminas are believed to be as follows: p-Alumina is considerably amorphous, with only a few diffuse lines in its X-ray powder diffraction pattern. in vacuo in air hydroxyl ions (each layer in hexagonal close packing) with aluminiums in octahedral co-ordination inside the layers in a pattern of hexagonal rings. The double layers stack to give an AB BA anion sequence. Boehmite is orthorhombic (space group Cmcm:a=2.87 A, b= 12.20 A, c= 3.69 A).2The struc- ture is isomorphous with lepidocrocite and contains octa- hedrally co-ordinated aluminiums, with the octahedral units linking to form complex layers and with hydrogens present ->+boehmite Y 6 --+ 8 as interlayer hydroxyl groups (i.e.attached to oxygens at top and bottom of the layers). The oxygens are approximately cubic-close-packed. Corundum is rhombohedra1 (space group R3c:a=4.76 A, c= 12.99 A).3 The structure contains oxygen in approximate hexagonal close packing with aluminiums in two-thirds of the octahedral sites. A number of studies utilising a wide variety of techniques I I I I II I 0 200 400 600 800 lo00 1200 TIT Fig. 1 Thermal transformation sequences for gibbsite. Temperature increases from left to right. Transition aluminas are denoted by the appropriate Greek symbol q-Alumina has cubic close packing (c.c.P.) of oxygens and is a defect spinel structure with A1 atoms in both tetrahedral and octahedral site^.'^,'^ &Alumina is monoclinic with approximately cubic close packing of oxygens and may also be related to the spinel structure.Aluminiums are in both octahedral and tetrahedral sites, but the ratio of tetrahedral A1 to octahedral A1 is higher than for q-alumina. y-and 6-aluminas are both spinel-related, with cubic-close-packed oxygens. y may be ordered or disordered with respect to the aluminium arrangement depending on the production con- ditions. The proportion of tetrahedral aluminiums increases through the sequence y, 6, 8 (as the calcination temperature is increased).x and IC aluminas both have hexagonal close packing (h.c.p.) of oxygens. x has a fairly diffuse X-ray powder diffraction pattern and contains a large number of stacking faults. Aluminiums are disordered over octahedral and tetra- hedral sites, with progressive filling of octahedral sites as temperature is increased. The advent of more routine multinuclear high-resolution NMR spectroscopy of solids is well known to have provided a powerful probe of the chemical environments of probe nuclei in condensed matter. We now present a systematic examination of A1 environments in gibbsite and its calcines (both single phase and polyphasic materials) via variations in the high-res~lution~~ A1 NMR absorption spectrum. Experimenta1 Materials A variety of gibbsites and gibbsite calcines were taken for examination using high-resolution NMR techniques.Some materials were commercial products, while others were pro- duced for this study. X-Ray powder diffraction patterns (Ni- filtered Cu-Ka radiation) for all materials were recorded using a computer-controlled Philips PW 1050 goniometer incorpor- ating accumulation of a variable number of scans. Gibbsites from three different origins were examined: two Bayer process gibbsites (BACO SF 7 and BACO UF 35 E) and a natural (Greek) gibbsite. The Bayer process products are highly crystalline, while the natural gibbsite is less crystal- line (with broader X-ray peaks). SF 7 has a medium particle size of ca.0.7 pm, with that for UF 35 E being ca.0.5 pm.UF 35 E has a lower soda content (0.14% Na20) than SF 7 (0.35% Na,O). Other single-phase materials used were as follows: commer- cial samples (BA Chemicals) of boehmite (BACO Cera Hydrate), a-alumina (BACO LS 2, mean particle size 7.5 pm) and X-alumina (BACO AA 101, median particle size 9 pm); p-alumina prepared from gibbsite UF 35 E by calcination in vacua at 200 "C. X-Ray powder diffraction patterns for single- phase materials are given in Fig. 2. Gibbsite SF 7 was calcined in air in a thin static bed to various temperatures with 2 h soak times at temperature and 5 K min-heating and cooling rates. X-Ray powder diffrac- tion patterns (Fig. 3) showed the presence of the following phases (categorised according to soak temperature): 400 "C, boehmite and palumina; 700 "C, y-and X-aluminas; 900 "C, 6-, 8-and Ic-aluminas (predominantly IC);1000 "C, 8-, IC-and a-aluminas.NMR Spectroscopy High-resolution MAS NMR spectra for 27Al (78.15 MHz) were recorded at ambient temperature using a Varian VXR300 spectrometer. A high spin-rate (Doty) probe was used with spin rates of ca.12 kHz. The use of high spin rates is essential in this work: use of a lower spin rate (such as the more usual J. MATER. CHEM., 1991 VOL. 1 I I I I I I (e ) 10 20 30 40 50 60 70 80 2ep Fig. 2 X-Ray powder diffraction patterns for single-phase materials (see text). (a) Gibbsite; (b) boehmite; (c) alpha; (d) chi; (e) rho ca. 3 kHz at the Larmor frequency in this study) would result in spinning sidebands within the shift range discussed below, and consequent difficulty in discussing the form of some of the spectra. A small signal due to A1 in the probe was subtracted from recorded spectra. Relaxation delays (0.5- 2.0 s) were more than sufficient to avoid saturation. Spectra are referenced to Al(H,O):+(aq.) and 46 r.f.pulses (liquid sample) were employed. Results and Discussion The magic-angle spinning removes the featureless internuclear dipolar broadenings characteristic of 'static' spectra. Each J. MATER. CHEM., 1991 VOL. I k---l-----L---’ I I I I I I I 10 20 30 40 50 60 70 80 2elo Fig. 3 X-Ray powder diffraction patterns for materials produced on calcination of gibbsite BACO SF 7 in air to various soak temperatures (see text).T/ “C:(a)400;(b)700; (c) 900; (d) 1000 27Al nucleus (I = 5/2) possesses a quadrupole moment, eQ, which will interact with non-zero electric-field gradients, eq, (determined by the charge distribution around the nucleus); that interaction is characterised by the asymmetry q of the electric-field gradient tensor and by the quadrupole coupling constant e2qQ/h [or the quadrupole frequency vQ : vQ = (3/20)x (eZqQ/h)for the 1/2+3/2 transition if q=O]. Usually, only the central +1/2+ -1/2 transition is observed by NMR spectroscopy. That transition is usually free of quadrupole effects to first order, but is subject to smaller second-order effects (including quadrupole shifts oQS)and the lineshape can be asymmetric in consequence.In this study, the linewidths for resonances in the spectra of calcines assignable to A106 are considerably greater than those of the more ordered materials (gibbsite, boehmite, corundum). This is consequent on the defect structures of the calcines, which lead to electric- field gradient (EFG) distributions at A1 sites. Discussions of 27Al spectra commonly neglect line-broadenings and oQs(the true field-independent chemical shift acs=uCG-uQs, where gCGis the centre of gravity of the observed lineshape). Ranges of chemical shift (acs)for A1 in various co-ordinations to oxygens in aluminosilicates have been given:I7 A104 (‘tetra- hedral’) 50-80, A105 (‘trigonal bipyramidal’) 30-40, A106 (‘octahedral’) -10 to + 20 ppm. Detailed consideration of second-order effects for the systems in this study would be problematic and further discussion of shifts is restricted to the position of observed peaks.The spectra in this study show clear similarities to those obtained from aluminosilicates and will be discussed in a manner entirely analogous to that used in discussion of those ~pectra.’~ Intensities of different lines in 27Al spectra cannot usually be simply related to relative populations in different environ- ments: A1 in near-regular geometries leads to the observation of distinct peaks at characteristic shifts but some lines (arising from A1 in low-symmetry environments) can be broadened (perhaps beyond observation) giving no distinct peak.Further- more, observed intensities can be affected by possible exci- tation of satellite transitions by the r.f. pulse and in the case of overlapping complex absorptions there is no non-arbitrary method for deconvolution of the unknown contributory line- shapes. It follows that information concerning the number and identity of distinct and observable A1 environments of near-regular geometry can be obtained (from the positions of observed peaks), but that in the general case (unknown lineshapes, etc.)relative populations in different environments cannot usually be deduced. It follows from the discussion above that peak positions will be slightly field dependent. Gibbsite Spectra for the two Bayer process products and for the natural sample are given in Fig.4. In order to present the data in the form leading to the maximum amount of derived information each spectrum is presented twice: the right-hand spectrum in each case is an expansion of the central features of the left hand spectrum. In each case the central absorption is asym- metric. Observed shifts of the peaks are 6.0 (SF 7), 4.7 (UF 35 E) and 4.4ppm (natural). These peak positions are fully consistent with six-co-ordinate (‘octahedral’) Al, as known from diffraction studies.’ Examination of the expanded (right-hand) spectra alone could suggest little difference between the two Bayer process products. Examination of the wider-shift-range spectra reveals an extensive side-band pattern in both cases, indicative of the breadth of the static (non-MAS) spectra in each case.In the case of UF 35 E a residual underlying broad (unnarrowed) component is evident, indicative of a proportion of A1 present being in low-symmetry environments (i.e. not in sites in 400 0 -400 100 0 -100 6 (PPW Fig. 4 27Al MAS NMR spectra (78.15 MHz) for commercial gibbsites (BA Chemicals) (a) BACO SF 7 and (b) BACO UF 35 E and (c) a natural (Greek) gibbsite 100 0 -100 I: Fig. 5 "A1 MAS NMR spectra (78.15 MHz) for (a)commercial (BA Chemicals) boehmite (BACO Cera Hydrate) and (b)or-alumina (BACO LS 2) crystalline gibbsite). This is likely to be a consequence of UF 35 E being a milled product, unlike SF 7. The lower crystallinity and lower purity of natural gibbsite manifests itself in considerable smoothing of the shoulder 400 0 -400 J.MATER. CHEM., 1991 VOL. 1 evident to high field (at ca. -13 ppm) of the peak in the Bayer process products. Boehmite The spectrum for synthetic boehmite is presented in Fig. 5. Side-bands extend to higher and lower fields. No underlying broad component is seen (i.e. all A1 are in the same environ- ment). The peak in the nearly symmetrical absorption occurs at 3.4 ppm, fully consistent with near-regular octahedral co- ordination of A1 as known from crystallographic studies.2 a-Alumina The spectrum for a-alumina (corundum) is also presented in Fig. 5. Side-bands extend to higher and lower fields. No underlying broad component is seen (i.e. all A1 are in the same environment).The peak occurs at 11.2 ppm, fully consist- ent with octahedral co-ordination of Al, as known from crystallographic ~tudies.~ The observation of a nearly sym- metrical line again corresponds to a more regular octahedral co-ordination of A1 than in gibbsite. X-Alumina The spectrum of X-alumina is shown in Fig. 6 in wide-shift- range and expanded forms. Peaks observed at 5.7 and 63.8 ppm are simply assigned to six-co-ordinate A1 and four- co-ordinate Al, consistent with the description of this material given in the introduction. There is a low-intensity broader component underlying the side-band structure, this being a result of the stacking faults (and lower local site symmetries) in this material. This component is, however, contained within the wide shift-window shown. This is not the case for the considerably broader underlying components (indicative of massively disordered regions) characteristic of Bayer gibbsite UF 35 E (see above) and of p-alumina and the gibbsite calcines in this study (see below).100 0 -100 l""1""I""I""I"' "1""1""1""1""1""1' 1000 0 -1000 100 0 -100 6 (PPm) Fig. 6 27Al MAS NMR spectra (78.15 MHz) for (a)x (BACO AA-101) and (b)p (200 "Cin uucuocalcination of gibbsite BACO UF 35 E) aluminas J. MATER. CHEM., 1991 VOL. 1 p-Alumina The spectrum of p-alumina is also shown in Fig. 6 in wide- shift-range and expanded forms. The massive disorder charac- teristic of this material is evident in the 'amorphous' X-ray pattern (Fig.1) and also in the high-intensity wings (unnar- rowed signal) in the absorption spectrum, those A1 in very distorted geometries giving rise to very broad spectral contributions. Examination of the expanded (right-hand) spectrum clearly shows three distinct peaks corresponding to A1 sites of near regular geometry. The peaks at 1.9, 27.2 and 54.4 ppm are easily assigned to six-co-ordinate, five-co-ordinate and four- co-ordinate Al, respectively. The slight upfield shift of the peak assigned to A105 relative to the chemical shift (acs) range cited above is a consequence of (a) the material not being an aluminosilicate, (b)the superposition of three absorp- tions (each broad) and (c) a positive (field-dependent) second- order quadrupole shift (aas).Assignment of a peak at ca. 30 ppm to A105 is unambiguous in the case of andalusiteI8 and has also been made in the case of a variety of thermally/ hydrothermally treated alumino~ilicates.~~ The presence of five-co-ordinate aluminium in an alumina has previously been reported for anodically formed amorphous alumina films (peak at 27-29 ppm, very similar to the value in this study)." It should be re-emphasised that the use of high spin rates is essential in this work. Use of a lower spin rate (ca. 3 kHz is more usual at the Larmor frequency in this study) would result in a spinning sideband close to the location of the peak assigned to five-co-ordinate Al, and consequent difficulty in deriving any conclusion as to the origin of that feature.Calcines of Gibbsite SF 7 The spectra for the calcines at 400, 700, 900 and 1000 "C are shown in Fig. 7 in both wide-shift-range and expanded forms. The wide rising wings (unnarrowed signal) evident in the wide-range spectra show that in each calcine some A1 is present in sites of highly irregular geometry. These underlying absorptions are, however, considerably less intense than in the case of p-alumina (above). None of the calcines analysed (X-ray diffraction, see above) as a single-phase material. The spectra therefore have to be discussed in terms of superposition of spectra for the various chemical components present. The spectrum for the 400°C calcine (contained boehmite and X-alumina) has peaks at 5.9 and 62.3 ppm, assigned to six- and four-co-ordinate Al, respectively.The spectra for boehmite and palumina themselves are discussed above. The absorptions for six-co-ordinate A1 in the different materials overlap in the mixture (calcine). The absorption assigned to four-co-ordinate A1 will arise solely from the X-alumina pre- sent. That the calcine differs from a simple admixture is evident on comparison of the wide-shift-range spectrum of the calcine with those of boehmite and X-alumina. The un- narrowed (underlying) component for the calcine is consider- ably broader than for either material, indicative of some A1 being present in more highly distorted environments than characteristic of either material on its own. The spectrum of the 700 "C calcine (containing x-and y-aluminas) has peaks at 6.8 and 62.7 ppm, assignable to six and four-co-ordinate Al, respectively.Both aluminas present are believed to have A1 present in both four- and six-co- ordination (see introduction) and the appropriate resonances for equivalent co-ordination numbers will overlap in the spectrum of the calcine. The spectrum for the 900°C calcine (predominantly IC-alumina) has peaks at 4.0 and 64.6 ppm (assigned to six- and four-co-ordinate Al, respectively) consistent with use of octa- jlI n --m"l--p ,.(..,.,.... ,,.. ,.,. .,,, , 400 h0 -400 100 0 ' -100 /I'I i\ i; L 400 -4 00 100 0 -100 Fig. 7 27Al MAS NMR spectra (78.15 MHz) for calcines of gibbsite BACO SF 7 produced with soak temperatures of (a)400, (b)700,(c) 900 and (d) 1000"C(see text) hedral and terahedral sites in an h.c.p. array of oxygens, as discussed in the introduction.In the area 30-70ppm the spectrum has an asymmetry evident in a shoulder at ca. 45 ppm. This could arise from the defect structure and the associated range of environments and consequent EFG distri- bution for Al0,-type sites, with a consequent marked asym- metry in the contributory lineshape. An alternative explanation is that some A1 in ic-alumina are in A105 environ- ments, leading to overlapping resonances arising from near- regular A106, A105 and A104. No spectral feature unambigu- ously assignable to A105 environments (which could result only from deviation of the oxygen packing from h.c.p.) is evident in the spectrum however.Overlapping is a major interpretational problem due to the widths of the two (reten- tion of h.c.p.) or three (deviation from h.c.p.) contributory lines; these will be broad and complex owing to the EFG distribution at each occupied A1 site type, and there is no realistic basis on which the contributory lineshapes can be modelled. Resolution of this problem is not possible with the instrumentation available in this study; spectra at a much higher field would improve the resolution of contributory lines, with further information also being available from spectra recorded using instruments enabling variable or dynamic angle spinning or double rotation. The spectrum for the 1000 "C calcine (containing 8-,ic-and a-aluminas) has a peak at 6.2ppm that is assigned to six-co- ordinate aluminium.In the shift region 30-70ppm some structure is apparent, with a peak 64.9ppm and another apparent maximum at 45.6 ppm. As discussed in the introduc- tion, 0 (c.c.P. oxygens) and IC (h.c.p. oxygens) are both believed to have A1 in both four- and six-co-ordination, while a has A1 only in six-co-ordination. The contributions from six-co- ordinate A1 in the various phases overlap. The peak at 64.9 pprn is assigned to overlapping contributions from four- co-ordinate A1 in the components of the calcine. The feature at 45.6 ppm is in the region associated with a shoulder in the contributory spectrum of the Ic-alumina component present (see above).The contributory spectrum of 0-alumina is not available. Conclusions High-resolution 27Al MAS NMR spectra reveal information about the sites occupied by A1 atoms in gibbsite and its derived calcines. A peak assigned to six-co-ordinate A1 is always observed. A second peak, assigned to four-co-ordinate Al, is also seen for those materials in which such sites are commonly believed to be used. In the case of the massively disordered and poorly understood p-alumina, a third peak is observed. That peak arises from the presence of some five- co-ordinate Al, as previously reported for anodically formed amorphous alumina. For calcines containing Ic-alumina, a shoulder or maximum at ca.45 ppm (upfield of the region usually associated with A104) may arise from a highly defective structure (and consequent changes in the shape and position of resonances arising from tetrahedral Al), but could indicate that the oxygen packing in that material deviates from h.c.p.by the introduction of five-co-ordinate A105 sites. Disorders resulting in A1 sites of much reduced symmetry manifest themselves in broad (incompletely narrowed) under- lying components in the spectra. For the most massively disordered/distorted systems the broad components manifest themselves instrumentally as wings extending well above and well below the central (narrowed) spectrum. Only those A1 in near-regular geometries give rise to the peaks in the preceding discussion. No information concerning the distorted geo-metries is available from the spectra.These are likely to be grossly distorted four- and six-co-ordinated aluminiums, but J. MATER. CHEM., 1991 VOL. 1 the possibility of similarly distorted five-co-ordinate alu-miniums cannot be ruled out for any of the materials whose spectra feature a broad background. R.C.T.S. thanks SERC for supporting the study of thermal transformation sequences under grant GR/E 8 1999 and for access to the national solid-state NMR service (University of Durham). We thank the staff of that service for recording high-resolution spectra. We thank the Laboratoire des Agre- gats Moleculaires et Materiaux Inorganiques (URA CNRS 79, Montpellier) for access to X-ray equipment. We thank BA Chemicals Ltd, and Alcan Chemicals Ltd, for supplying commerical-grade aluminas and hydrates. References H.D. Megaw, 2. Krist., 1934, A87, 185. P. P. Reichertz and W. J. Yost, J. Chem. Phys., 1946, 14, 495. R. E. Newnham and Y. M. de Haan, Z. Krist., 1962, 117, 235. R. Tertian and D. Papee, J. Chim. Phys., 1958, 55, 341. H. Saalfeld, N. Jb. Miner. Abh., 1960, 95, 1. G. W. Brindley and J. 0.Choe, Am. Mineral., 1961, 46, 771. B. C. Lippens and J. H. de Boer, Acta Crystallogr., 1964, 17, 1312. 8 J. Beretka and M. J. Ridge, J. Chem. SOC.(A), 1967, 2106. 9 S. J. Wilson, Proc. Br. Ceram. SOC., 1979, 28, 281. 10 H. Dexpert, J. F. Larne, I. Mutin, B. Moraweck, Y.Bertaud and A. Renouprez, J. Metal., 1985, 37, 17. 11 T. Sato, J. Thermal Anal., 1987, 32, 61. 12 G. D. Chunkin and Yu. L. Seleznev, Inorg. Muter., 1987,23, 374. 13 V. Jayaram and C. G. Levi, Acta Metal., 1989, 37, 569. 14 W. H. Gitzen, Alumina as a Ceramic Material, American Ceramic Society, Columbus, 1970. 15 E. J. W. Verwey, Z. Krist., 1935, 91, 65. 16 A. J. Leonard, P. N. Semaille and J. J. Fritpiat, Proc. Br. Ceram. SOC., 1969, 13, 103. 17 G. Engelhardt and D. Michel, High Resolution Solid-State NMR of Silicates and Zeolites, Wiley, Chichester, 1987. 18 E. Lipmaa, A. Samoson and M. Magi, J. Am. Chem. SOC., 1986, 108, 1730. 19 J. Gilson, G. C. Edwards, A. W. Peters, K. Rajagopalan, R. F. Wormsbecher, T. G. Roberie and M. P. Shattock. J. Chem. SOC., Chem. Commun., 1987, 91. 20 R. Dupree, I. Farnan, A. J. Forty, S. El-Mashri and L. Bottyan, J. Phys. (Paris), 1985, 46, C8-113. Paper 1/00304F; Received 22nd January, 199 1